Skip to content

LLM inference in C/C++

C++ ggml

metal: per-op source split + parallel compile (#26561)

YiChen Lvb615f5b4bd99
diff --git a/ggml/CMakeLists.txt b/ggml/CMakeLists.txt
index 11ed9e8bd7..5f4742a38c 100644
--- a/ggml/CMakeLists.txt
+++ b/ggml/CMakeLists.txt
@@ -342,9 +342,6 @@ set(GGML_PUBLIC_HEADERS
include/gguf.h)
set_target_properties(ggml PROPERTIES PUBLIC_HEADER "${GGML_PUBLIC_HEADERS}")
-#if (GGML_METAL)
-# set_target_properties(ggml PROPERTIES RESOURCE "${CMAKE_CURRENT_SOURCE_DIR}/src/ggml-metal.metal")
-#endif()
install(TARGETS ggml LIBRARY PUBLIC_HEADER)
install(TARGETS ggml-base LIBRARY)
diff --git a/ggml/src/ggml-metal/CMakeLists.txt b/ggml/src/ggml-metal/CMakeLists.txt
index 42054d841a..f02654a2e7 100644
--- a/ggml/src/ggml-metal/CMakeLists.txt
+++ b/ggml/src/ggml-metal/CMakeLists.txt
@@ -24,62 +24,119 @@ if (GGML_METAL_NDEBUG)
endif()
set(METALLIB_COMMON "${CMAKE_CURRENT_SOURCE_DIR}/../ggml-common.h")
+set(METALLIB_KERNELS_COMMON "${CMAKE_CURRENT_SOURCE_DIR}/kernels/common.h")
+set(METALLIB_KERNELS_DEQUANTIZE "${CMAKE_CURRENT_SOURCE_DIR}/kernels/dequantize.h")
+set(METALLIB_KERNELS_QUANTIZE "${CMAKE_CURRENT_SOURCE_DIR}/kernels/quantize.h")
+
+set(METALLIB_KERNEL_SOURCES
+ kernels/fa.metal
+ kernels/mul_mv.metal
+ kernels/mul_mm.metal
+ kernels/quantize.metal
+ kernels/softmax.metal
+ kernels/norm.metal
+ kernels/unary.metal
+ kernels/binbcast.metal
+ kernels/reduce.metal
+ kernels/tri.metal
+ kernels/ssm.metal
+ kernels/wkv.metal
+ kernels/gated_delta_net.metal
+ kernels/solve_tri.metal
+ kernels/rope.metal
+ kernels/conv.metal
+ kernels/upscale.metal
+ kernels/argsort.metal
+ kernels/pool.metal
+ kernels/misc.metal
+)
+
if (GGML_METAL_EMBED_LIBRARY)
enable_language(ASM)
add_compile_definitions(GGML_METAL_EMBED_LIBRARY)
- set(METALLIB_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal.metal")
- set(METALLIB_IMPL "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal-impl.h")
+ set(METALLIB_IMPL "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal-impl.h")
file(MAKE_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/autogenerated")
- # merge ggml-common.h and ggml-metal.metal into a single file
- set(METALLIB_EMBED_ASM "${CMAKE_CURRENT_BINARY_DIR}/autogenerated/ggml-metal-embed.s")
- set(METALLIB_SOURCE_EMBED "${CMAKE_CURRENT_BINARY_DIR}/autogenerated/ggml-metal-embed.metal")
- set(METALLIB_SOURCE_EMBED_TMP "${CMAKE_CURRENT_BINARY_DIR}/autogenerated/ggml-metal-embed.metal.tmp")
+ set(METALLIB_EMBED_ASM_FILES "")
+ foreach(src ${METALLIB_KERNEL_SOURCES})
+ get_filename_component(kind ${src} NAME_WE)
+ # symbol names must be valid C identifiers ('-' is not allowed)
+ string(REPLACE "-" "_" kind_sym ${kind})
+
+ set(SRC "${CMAKE_CURRENT_SOURCE_DIR}/kernels/${kind}.metal")
+ set(EMBED "${CMAKE_CURRENT_BINARY_DIR}/autogenerated/ggml-metal-embed-${kind}.metal")
+ set(ASM "${CMAKE_CURRENT_BINARY_DIR}/autogenerated/ggml-metal-embed-${kind}.s")
+
+ # only prepend headers that this source actually includes
+ set(HEADERS_FOR_SRC ${METALLIB_KERNELS_COMMON})
+ file(STRINGS ${SRC} _has_dequantize REGEX "#include \"dequantize\\.h\"")
+ file(STRINGS ${SRC} _has_quantize REGEX "#include \"quantize\\.h\"")
+ if(_has_dequantize)
+ list(APPEND HEADERS_FOR_SRC ${METALLIB_KERNELS_DEQUANTIZE})
+ endif()
+ if(_has_quantize)
+ list(APPEND HEADERS_FOR_SRC ${METALLIB_KERNELS_QUANTIZE})
+ endif()
+
+ add_custom_command(
+ OUTPUT "${ASM}"
+ # Step 1: concatenate shared headers + this kernel source
+ COMMAND cat ${HEADERS_FOR_SRC} ${SRC} > "${EMBED}.tmp1"
+ # Step 2: remove internal #include and #pragma once
+ COMMAND sed -e "/\#include \"common.h\"/d" -e "/\#include \"dequantize.h\"/d" -e "/\#include \"quantize.h\"/d" -e "/\#pragma once/d" < "${EMBED}.tmp1" > "${EMBED}.tmp2"
+ # Step 3: inline ggml-common.h (replacing __embed_ggml-common.h__ sentinel)
+ COMMAND sed -e "/__embed_ggml-common.h__/r ${METALLIB_COMMON}" -e "/__embed_ggml-common.h__/d" < "${EMBED}.tmp2" > "${EMBED}.tmp3"
+ # Step 4: inline ggml-metal-impl.h
+ COMMAND sed -e "/\#include \"ggml-metal-impl.h\"/r ${METALLIB_IMPL}" -e "/\#include \"ggml-metal-impl.h\"/d" < "${EMBED}.tmp3" > "${EMBED}"
+ # Step 5: emit an asm chunk with kind-specific start/end symbols
+ # note: '-' is illegal in C symbols, so we use kind_sym; the macOS
+ # section name is limited to 16 chars so we keep it shared
+ # across kinds (__ggml_metallib) and only vary the global symbols.
+ COMMAND echo ".section __DATA,__ggml_metallib" > "${ASM}"
+ COMMAND echo ".globl _ggml_metallib_${kind_sym}_start" >> "${ASM}"
+ COMMAND echo "_ggml_metallib_${kind_sym}_start:" >> "${ASM}"
+ COMMAND echo .incbin "\"${EMBED}\"" >> "${ASM}"
+ COMMAND echo ".globl _ggml_metallib_${kind_sym}_end" >> "${ASM}"
+ COMMAND echo "_ggml_metallib_${kind_sym}_end:" >> "${ASM}"
+ DEPENDS ../ggml-common.h ggml-metal-impl.h
+ kernels/common.h kernels/dequantize.h kernels/quantize.h
+ kernels/${kind}.metal
+ COMMENT "Generate embedded Metal library for ${kind}"
+ VERBATIM
+ )
- add_custom_command(
- OUTPUT "${METALLIB_EMBED_ASM}"
- COMMAND echo "Embedding Metal library"
- COMMAND sed -e "/__embed_ggml-common.h__/r ${METALLIB_COMMON}" -e "/__embed_ggml-common.h__/d" < "${METALLIB_SOURCE}" > "${METALLIB_SOURCE_EMBED_TMP}"
- COMMAND sed -e "/\#include \"ggml-metal-impl.h\"/r ${METALLIB_IMPL}" -e "/\#include \"ggml-metal-impl.h\"/d" < "${METALLIB_SOURCE_EMBED_TMP}" > "${METALLIB_SOURCE_EMBED}"
- COMMAND echo ".section __DATA,__ggml_metallib" > "${METALLIB_EMBED_ASM}"
- COMMAND echo ".globl _ggml_metallib_start" >> "${METALLIB_EMBED_ASM}"
- COMMAND echo "_ggml_metallib_start:" >> "${METALLIB_EMBED_ASM}"
- COMMAND echo .incbin "\"${METALLIB_SOURCE_EMBED}\"" >> "${METALLIB_EMBED_ASM}"
- COMMAND echo ".globl _ggml_metallib_end" >> "${METALLIB_EMBED_ASM}"
- COMMAND echo "_ggml_metallib_end:" >> "${METALLIB_EMBED_ASM}"
- DEPENDS ../ggml-common.h ggml-metal.metal ggml-metal-impl.h
- COMMENT "Generate assembly for embedded Metal library"
- VERBATIM
- )
+ list(APPEND METALLIB_EMBED_ASM_FILES "${ASM}")
+ endforeach()
- target_sources(ggml-metal PRIVATE "${METALLIB_EMBED_ASM}")
+ target_sources(ggml-metal PRIVATE ${METALLIB_EMBED_ASM_FILES})
else()
- # copy metal files to bin directory
+ # copy header files to bin directory
configure_file(../ggml-common.h ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h COPYONLY)
- configure_file(ggml-metal.metal ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal COPYONLY)
configure_file(ggml-metal-impl.h ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal-impl.h COPYONLY)
+ file(MAKE_DIRECTORY "${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels")
+ configure_file(kernels/common.h ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels/common.h COPYONLY)
+ configure_file(kernels/dequantize.h ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels/dequantize.h COPYONLY)
+ configure_file(kernels/quantize.h ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels/quantize.h COPYONLY)
+
+ foreach(src ${METALLIB_KERNEL_SOURCES})
+ configure_file(${src} ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${src} COPYONLY)
+ endforeach()
+
if (GGML_METAL_SHADER_DEBUG)
- # custom command to do the following:
- # xcrun -sdk macosx metal -fno-fast-math -c ggml-metal.metal -o ggml-metal.air
- # xcrun -sdk macosx metallib ggml-metal.air -o default.metallib
- #
- # note: this is the only way I found to disable fast-math in Metal. it's ugly, but at least it works
- # disabling fast math is needed in order to pass tests/test-backend-ops
+ # note: disabling fast math is needed in order to pass tests/test-backend-ops
# note: adding -fno-inline fixes the tests when using MTL_SHADER_VALIDATION=1
# note: unfortunately, we have to call it default.metallib instead of ggml.metallib
# ref: https://github.com/ggml-org/whisper.cpp/issues/1720
# note: adding -g causes segmentation fault during compile
- #set(XC_FLAGS -fno-fast-math -fno-inline -g)
set(XC_FLAGS -fno-fast-math -fno-inline)
else()
set(XC_FLAGS -O3)
endif()
- # Append macOS metal versioning flags
if (GGML_METAL_MACOSX_VERSION_MIN)
message(STATUS "Adding -mmacosx-version-min=${GGML_METAL_MACOSX_VERSION_MIN} flag to metal compilation")
list (APPEND XC_FLAGS -mmacosx-version-min=${GGML_METAL_MACOSX_VERSION_MIN})
@@ -90,35 +147,46 @@ else()
list (APPEND XC_FLAGS -std=${GGML_METAL_STD})
endif()
+ # Compile each kernel source to .air, then link into default.metallib
+ set(AIR_FILES "")
+ foreach(src ${METALLIB_KERNEL_SOURCES})
+ get_filename_component(name ${src} NAME_WE)
+ set(AIR "${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${name}.air")
+ list(APPEND AIR_FILES ${AIR})
+ add_custom_command(
+ OUTPUT ${AIR}
+ COMMAND xcrun -sdk macosx metal ${XC_FLAGS} -I ${CMAKE_RUNTIME_OUTPUT_DIRECTORY} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${src} -o ${AIR}
+ DEPENDS ${src} kernels/common.h kernels/dequantize.h kernels/quantize.h ${METALLIB_COMMON} ggml-metal-impl.h
+ COMMENT "Compiling ${src}"
+ VERBATIM
+ )
+ endforeach()
+
add_custom_command(
OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
- COMMAND xcrun -sdk macosx metal ${XC_FLAGS} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal -o - |
- xcrun -sdk macosx metallib - -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
+ COMMAND xcrun -sdk macosx metallib ${AIR_FILES} -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h
- COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal
- DEPENDS ggml-metal.metal ${METALLIB_COMMON}
- COMMENT "Compiling Metal kernels"
- )
+ COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal-impl.h
+ COMMAND rm -rf ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/kernels
+ DEPENDS ${AIR_FILES}
+ COMMENT "Linking Metal kernels into default.metallib"
+ )
- # FIXME: only add to the ggml-metal target?
add_custom_target(
ggml-metal-lib ALL
DEPENDS ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
- )
+ )
endif() # GGML_METAL_EMBED_LIBRARY
if (NOT GGML_METAL_EMBED_LIBRARY)
install(
- FILES src/ggml-metal/ggml-metal.metal
- PERMISSIONS
- OWNER_READ
- OWNER_WRITE
- GROUP_READ
- WORLD_READ
- DESTINATION ${CMAKE_INSTALL_BINDIR})
-
- install(
- FILES ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
- DESTINATION ${CMAKE_INSTALL_BINDIR}
- )
+ DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/kernels/
+ DESTINATION ${CMAKE_INSTALL_BINDIR}/kernels
+ FILES_MATCHING PATTERN "*.metal" PATTERN "*.h"
+ )
+
+ install(
+ FILES ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
+ DESTINATION ${CMAKE_INSTALL_BINDIR}
+ )
endif()
diff --git a/ggml/src/ggml-metal/ggml-metal-device.m b/ggml/src/ggml-metal/ggml-metal-device.m
index 263016df7d..6647cacd61 100644
--- a/ggml/src/ggml-metal/ggml-metal-device.m
+++ b/ggml/src/ggml-metal/ggml-metal-device.m
@@ -95,8 +95,63 @@ int ggml_metal_pipeline_max_theads_per_threadgroup(struct ggml_metal_pipeline_wi
return pipeline.pipeline->obj.maxTotalThreadsPerThreadgroup;
}
+//
+// MTLLibrary collection (one library per op-source, compiled separately)
+//
+
+// Single source of truth for the per-kind metal libraries. The order here
+// defines the enum values and every per-kind table below, so adding a library
+// is a one-line change here (plus adding its source to CMakeLists.txt).
+// X(suffix, name): name is both the kernels/<name>.metal basename and the
+// ggml_metallib_<name>_{start,end} embed-symbol stem.
+#define GGML_METAL_LIBS \
+ X(FA, fa) \
+ X(MUL_MV, mul_mv) \
+ X(MUL_MM, mul_mm) \
+ X(QUANTIZE, quantize) \
+ X(SOFTMAX, softmax) \
+ X(NORM, norm) \
+ X(UNARY, unary) \
+ X(BINBCAST, binbcast) \
+ X(REDUCE, reduce) \
+ X(TRI, tri) \
+ X(SSM, ssm) \
+ X(WKV, wkv) \
+ X(GATED_DELTA_NET, gated_delta_net)\
+ X(SOLVE_TRI, solve_tri) \
+ X(ROPE, rope) \
+ X(CONV, conv) \
+ X(UPSCALE, upscale) \
+ X(ARGSORT, argsort) \
+ X(POOL, pool) \
+ X(MISC, misc)
+
+enum ggml_metal_lib_kind {
+#define X(e, s) GGML_METAL_LIB_##e,
+ GGML_METAL_LIBS
+#undef X
+ GGML_METAL_LIB_COUNT,
+};
+
+static const char * const k_lib_names[GGML_METAL_LIB_COUNT] = {
+#define X(e, s) [GGML_METAL_LIB_##e] = #s,
+ GGML_METAL_LIBS
+#undef X
+};
+
struct ggml_metal_library {
- id<MTLLibrary> obj;
+ // Per-kind compiled libraries. When single_library is true, the whole library
+ // (e.g. a pre-compiled default.metallib or a from-source build) lives at
+ // objs[0] and the remaining slots are nil.
+ id<MTLLibrary> objs[GGML_METAL_LIB_COUNT];
+ bool single_library; // true: combined library at objs[0]; false: per-kind libs in objs[*]
+
+ // Routing table: kernel function name -> objs[] index, populated from each
+ // compiled library's -[MTLLibrary functionNames]. The actual compiled
+ // libraries are the single source of truth for which library owns a kernel,
+ // so adding kernels later requires no manual routing maintenance.
+ // nil in single_library mode (everything resolves to objs[0]).
+ NSMutableDictionary<NSString *, NSNumber *> * fn_to_lib;
ggml_metal_device_t dev;
ggml_metal_pipelines_t pipelines; // cache of compiled pipelines
@@ -104,160 +159,376 @@ struct ggml_metal_library {
NSLock * lock;
};
-ggml_metal_library_t ggml_metal_library_init(ggml_metal_device_t dev) {
- id<MTLLibrary> library = nil;
- id<MTLDevice> device = ggml_metal_device_get_obj(dev);
+// Build the fn_to_lib routing table by querying each compiled library's public
+// function names. Call once after all per-kind libraries have been compiled.
+static void ggml_metal_library_build_index(ggml_metal_library_t lib) {
+ @autoreleasepool {
+ NSMutableDictionary<NSString *, NSNumber *> * index = [[NSMutableDictionary alloc] init];
+ for (int kind = 0; kind < GGML_METAL_LIB_COUNT; ++kind) {
+ for (NSString * fname in [lib->objs[kind] functionNames]) {
+ index[fname] = @(kind);
+ }
+ }
+ lib->fn_to_lib = index;
+ }
+}
- // load library
- //
- // - first check if the library is embedded
- // - then check if the library is in the bundle
- // - if not found, load the source and compile it
- // - if that fails, return NULL
- //
- // TODO: move to a function
- {
- const int64_t t_start = ggml_time_us();
+// Parse a `#include "name"` line. Returns the quoted name in *include_name on
+// success. Whitespace-tolerant; ignores `#include <...>` (system headers).
+static bool ggml_metal_library_parse_quoted_include(NSString * line, NSString ** include_name) {
+ NSScanner * scanner = [NSScanner scannerWithString:line];
+ scanner.charactersToBeSkipped = [NSCharacterSet whitespaceCharacterSet];
- NSError * error = nil;
- NSString * src = nil;
+ if (![scanner scanString:@"#" intoString:NULL] ||
+ ![scanner scanString:@"include" intoString:NULL] ||
+ ![scanner scanString:@"\"" intoString:NULL]) {
+ return false;
+ }
-#if GGML_METAL_EMBED_LIBRARY
- GGML_LOG_INFO("%s: using embedded metal library\n", __func__);
+ NSString * name = nil;
+ if (![scanner scanUpToString:@"\"" intoString:&name]) {
+ return false;
+ }
- extern const char ggml_metallib_start[];
- extern const char ggml_metallib_end[];
+ if (include_name) {
+ *include_name = name;
+ }
+ return true;
+}
- src = [[NSString alloc] initWithBytes:ggml_metallib_start length:(ggml_metallib_end-ggml_metallib_start) encoding:NSUTF8StringEncoding];
-#else
+// Recursively inline `#include "name"` directives. System includes (<...>),
+// `#if/#else/#endif`, and other preprocessor lines are passed through to the
+// Metal compiler unchanged. `#pragma once` is dropped since `seen` already
+// guards against double-inclusion.
+static bool ggml_metal_library_flatten_file(NSMutableString * dst, NSString * path,
+ NSArray<NSString *> * search_paths,
+ NSMutableSet<NSString *> * seen, NSError ** error) {
+ NSString * key = [path stringByStandardizingPath];
+ if ([seen containsObject:key]) {
+ return true;
+ }
+ [seen addObject:key];
-#ifdef SWIFT_PACKAGE
- NSBundle * bundle = SWIFTPM_MODULE_BUNDLE;
-#else
- NSBundle * bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
-#endif
+ NSString * src = [NSString stringWithContentsOfFile:path encoding:NSUTF8StringEncoding error:error];
+ if (!src) {
+ return false;
+ }
- NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
- if (path_lib == nil) {
- // Try to find the resource in the directory where the current binary located.
- NSString * bin_cur = [[NSProcessInfo processInfo] arguments][0];
- NSString * bin_dir = [bin_cur stringByDeletingLastPathComponent];
-
- NSString * path_lib_default = [NSString pathWithComponents:@[bin_dir, @"default.metallib"]];
- if ([[NSFileManager defaultManager] isReadableFileAtPath:path_lib_default]) {
- GGML_LOG_INFO("%s: found '%s'\n", __func__, [path_lib_default UTF8String]);
-
- NSDictionary * atts = [[NSFileManager defaultManager] attributesOfItemAtPath:path_lib_default error:&error];
- if (atts && atts[NSFileType] == NSFileTypeSymbolicLink) {
- // Optionally, if this is a symlink, try to resolve it.
- path_lib_default = [[NSFileManager defaultManager] destinationOfSymbolicLinkAtPath:path_lib_default error:&error];
- if (path_lib_default && [path_lib_default length] > 0 && ![[path_lib_default substringToIndex:1] isEqualToString:@"/"]) {
- // It is a relative path, adding the binary directory as directory prefix.
- path_lib_default = [NSString pathWithComponents:@[bin_dir, path_lib_default]];
- }
- if (!path_lib_default || ![[NSFileManager defaultManager] isReadableFileAtPath:path_lib_default]) {
- // Link to the resource could not be resolved.
- path_lib_default = nil;
- } else {
- GGML_LOG_INFO("%s: symlink resolved '%s'\n", __func__, [path_lib_default UTF8String]);
- }
+ NSFileManager * fm = [NSFileManager defaultManager];
+ for (NSString * line in [src componentsSeparatedByString:@"\n"]) {
+ NSString * trimmed = [line stringByTrimmingCharactersInSet:[NSCharacterSet whitespaceCharacterSet]];
+ if ([trimmed isEqualToString:@"#pragma once"]) {
+ continue;
+ }
+
+ NSString * include_name = nil;
+ if (ggml_metal_library_parse_quoted_include(line, &include_name)) {
+ NSString * resolved = nil;
+ for (NSString * dir in search_paths) {
+ NSString * candidate = [dir stringByAppendingPathComponent:include_name];
+ if ([fm isReadableFileAtPath:candidate]) {
+ resolved = candidate;
+ break;
}
- } else {
- // The resource couldn't be found in the binary's directory.
- path_lib_default = nil;
}
-
- path_lib = path_lib_default;
+ if (!resolved) {
+ if (error) {
+ NSString * msg = [NSString stringWithFormat:@"could not resolve include \"%@\" from '%@'", include_name, path];
+ *error = [NSError errorWithDomain:@"ggml-metal-source-flatten" code:1
+ userInfo:@{NSLocalizedDescriptionKey: msg}];
+ }
+ return false;
+ }
+ if (!ggml_metal_library_flatten_file(dst, resolved, search_paths, seen, error)) {
+ return false;
+ }
+ continue;
}
- if (path_lib != nil) {
- // pre-compiled library found
- NSURL * libURL = [NSURL fileURLWithPath:path_lib];
- GGML_LOG_INFO("%s: loading '%s'\n", __func__, [path_lib UTF8String]);
+ [dst appendString:line];
+ [dst appendString:@"\n"];
+ }
- library = [device newLibraryWithURL:libURL error:&error];
- if (error) {
- GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
- return nil;
+ return true;
+}
+
+static NSString * ggml_metal_library_flatten_source(NSString * path_source, NSError ** error) {
+ // Search paths cover both runtime layout (build/bin/kernels + build/bin)
+ // and source-tree layout (ggml/src/ggml-metal/kernels + ggml/src/ggml-metal + ggml/src).
+ NSString * path_kernels = [path_source stringByDeletingLastPathComponent];
+ NSString * path_base = [path_kernels stringByDeletingLastPathComponent];
+ NSArray<NSString *> * search_paths = @[
+ path_kernels,
+ path_base,
+ [path_base stringByDeletingLastPathComponent],
+ ];
+
+ NSMutableString * src = [[NSMutableString alloc] init];
+ NSMutableSet<NSString *> * seen = [NSMutableSet set];
+
+ if (!ggml_metal_library_flatten_file(src, path_source, search_paths, seen, error)) {
+ [src release];
+ return nil;
+ }
+ return src;
+}
+
+// Compile all per-kind libraries in parallel. `source_for_kind` returns the MSL
+// source for a kind (the helper takes ownership and releases it), or nil with
+// *err set on failure. On success the objs[] slots are populated and the routing
+// index is built; on any failure every error is logged and false is returned
+// (the caller is responsible for freeing `res`).
+static bool ggml_metal_library_compile_all(
+ ggml_metal_library_t res,
+ id<MTLDevice> device,
+ NSDictionary * prep,
+ NSString * (^source_for_kind)(int kind, NSError ** err),
+ const char * origin) {
+ const int64_t t_start = ggml_time_us();
+
+ int64_t * t_per_lib = calloc(GGML_METAL_LIB_COUNT, sizeof(int64_t));
+ NSError ** err_per_lib = calloc(GGML_METAL_LIB_COUNT, sizeof(NSError *));
+ __block atomic_bool any_failure = false;
+
+ dispatch_group_t group = dispatch_group_create();
+ dispatch_queue_t queue = dispatch_get_global_queue(QOS_CLASS_USER_INITIATED, 0);
+
+ for (int kind = 0; kind < GGML_METAL_LIB_COUNT; ++kind) {
+ dispatch_group_async(group, queue, ^{
+
+ const int64_t t0 = ggml_time_us();
+
+ NSError * error = nil;
+
+ NSString * src = source_for_kind(kind, &error);
+ if (!src) {
+ err_per_lib[kind] = [error retain];
+ atomic_store(&any_failure, true);
+ return;
}
- } else {
- GGML_LOG_INFO("%s: default.metallib not found, loading from source\n", __func__);
- NSString * path_source;
- NSString * path_resource = [[NSProcessInfo processInfo].environment objectForKey:@"GGML_METAL_PATH_RESOURCES"];
+ id<MTLLibrary> lib = nil;
+
+ @autoreleasepool {
+ MTLCompileOptions * options = [MTLCompileOptions new];
+ options.preprocessorMacros = prep;
+
+ lib = [device newLibraryWithSource:src options:options error:&error];
- GGML_LOG_INFO("%s: GGML_METAL_PATH_RESOURCES = %s\n", __func__, path_resource ? [path_resource UTF8String] : "nil");
+ [options release];
- if (path_resource) {
- path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal.metal"];
- } else {
- path_source = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
+ // retain the error before the autorelease pool drains it
+ if (!lib) {
+ err_per_lib[kind] = [error retain];
+ }
}
- if (path_source == nil) {
- GGML_LOG_WARN("%s: error: could not use bundle path to find ggml-metal.metal, falling back to trying cwd\n", __func__);
- path_source = @"ggml-metal.metal";
+ [src release];
+
+ t_per_lib[kind] = ggml_time_us() - t0;
+
+ if (!lib) {
+ atomic_store(&any_failure, true);
+ return;
}
- GGML_LOG_INFO("%s: loading '%s'\n", __func__, [path_source UTF8String]);
+ res->objs[kind] = lib;
+ });
+ }
+ dispatch_group_wait(group, DISPATCH_TIME_FOREVER);
+ dispatch_release(group);
+
+ const bool ok = !atomic_load(&any_failure);
+
+ if (ok) {
+ const int64_t t_total = ggml_time_us() - t_start;
+ int64_t t_max = 0;
+ for (int kind = 0; kind < GGML_METAL_LIB_COUNT; ++kind) {
+ GGML_LOG_DEBUG("%s: compiled '%s' library in %.3f sec\n",
+ __func__, k_lib_names[kind], t_per_lib[kind] / 1e6);
+ if (t_per_lib[kind] > t_max) t_max = t_per_lib[kind];
+ }
+ GGML_LOG_INFO("%s: loaded %d libraries from %s in %.3f sec (max single = %.3f sec)\n",
+ __func__, GGML_METAL_LIB_COUNT, origin, t_total / 1e6, t_max / 1e6);
- src = [NSString stringWithContentsOfFile:path_source encoding:NSUTF8StringEncoding error:&error];
- if (error) {
- GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
- return nil;
+ ggml_metal_library_build_index(res);
+ } else {
+ for (int kind = 0; kind < GGML_METAL_LIB_COUNT; ++kind) {
+ if (err_per_lib[kind]) {
+ GGML_LOG_ERROR("%s: failed to build '%s' library: %s\n", __func__,
+ k_lib_names[kind], [[err_per_lib[kind] description] UTF8String]);
+ [err_per_lib[kind] release];
}
}
-#endif
+ }
- if (!library) {
- @autoreleasepool {
- // dictionary of preprocessor macros
- NSMutableDictionary * prep = [NSMutableDictionary dictionary];
+ free(err_per_lib);
+ free(t_per_lib);
- if (ggml_metal_device_get_props(dev)->has_bfloat) {
- [prep setObject:@"1" forKey:@"GGML_METAL_HAS_BF16"];
- }
+ return ok;
+}
- if (ggml_metal_device_get_props(dev)->has_tensor) {
- [prep setObject:@"1" forKey:@"GGML_METAL_HAS_TENSOR"];
- }
+ggml_metal_library_t ggml_metal_library_init(ggml_metal_device_t dev) {
+ id<MTLDevice> device = ggml_metal_device_get_obj(dev);
+
+ ggml_metal_library_t res = calloc(1, sizeof(struct ggml_metal_library));
+ res->dev = dev;
+ res->pipelines = ggml_metal_pipelines_init();
+ res->lock = [NSLock new];
+ // shared MTLCompileOptions preprocessor macros (matches the build-time defines)
+ NSMutableDictionary * prep = [NSMutableDictionary dictionary];
+ if (ggml_metal_device_get_props(dev)->has_bfloat) {
+ [prep setObject:@"1" forKey:@"GGML_METAL_HAS_BF16"];
+ }
+ if (ggml_metal_device_get_props(dev)->has_tensor) {
+ [prep setObject:@"1" forKey:@"GGML_METAL_HAS_TENSOR"];
+ }
#if GGML_METAL_EMBED_LIBRARY
- [prep setObject:@"1" forKey:@"GGML_METAL_EMBED_LIBRARY"];
+ [prep setObject:@"1" forKey:@"GGML_METAL_EMBED_LIBRARY"];
#endif
- MTLCompileOptions * options = [MTLCompileOptions new];
- options.preprocessorMacros = prep;
+#if GGML_METAL_EMBED_LIBRARY
+ GGML_LOG_INFO("%s: using embedded metal library\n", __func__);
- //[options setFastMathEnabled:false];
+ // start/end symbols emitted by CMake (see CMakeLists.txt), one pair per kind
+#define X(e, s) extern const char ggml_metallib_##s##_start[]; extern const char ggml_metallib_##s##_end[];
+ GGML_METAL_LIBS
+#undef X
- library = [device newLibraryWithSource:src options:options error:&error];
- if (error) {
- GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
- return nil;
- }
+ static const char * const lib_start[GGML_METAL_LIB_COUNT] = {
+#define X(e, s) [GGML_METAL_LIB_##e] = ggml_metallib_##s##_start,
+ GGML_METAL_LIBS
+#undef X
+ };
+ static const char * const lib_end[GGML_METAL_LIB_COUNT] = {
+#define X(e, s) [GGML_METAL_LIB_##e] = ggml_metallib_##s##_end,
+ GGML_METAL_LIBS
+#undef X
+ };
-#if !__has_feature(objc_arc)
- [options release];
+ const bool ok = ggml_metal_library_compile_all(res, device, prep,
+ ^NSString * (int kind, NSError ** err) {
+ (void) err;
+ return [[NSString alloc] initWithBytes:lib_start[kind]
+ length:(lib_end[kind] - lib_start[kind])
+ encoding:NSUTF8StringEncoding];
+ }, "embedded data");
+
+ if (!ok) {
+ ggml_metal_library_free(res);
+ return NULL;
+ }
+
+ return res;
+#else
+#ifdef SWIFT_PACKAGE
+ NSBundle * bundle = SWIFTPM_MODULE_BUNDLE;
+#else
+ NSBundle * bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
#endif
+
+ const int64_t t_start = ggml_time_us();
+
+ NSError * error = nil;
+ NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
+ if (path_lib == nil) {
+ // Try to find the resource in the directory where the current binary located.
+ NSString * bin_cur = [[NSProcessInfo processInfo] arguments][0];
+ NSString * bin_dir = [bin_cur stringByDeletingLastPathComponent];
+
+ NSString * path_lib_default = [NSString pathWithComponents:@[bin_dir, @"default.metallib"]];
+ if ([[NSFileManager defaultManager] isReadableFileAtPath:path_lib_default]) {
+ GGML_LOG_INFO("%s: found '%s'\n", __func__, [path_lib_default UTF8String]);
+
+ NSDictionary * atts = [[NSFileManager defaultManager] attributesOfItemAtPath:path_lib_default error:&error];
+ if (atts && atts[NSFileType] == NSFileTypeSymbolicLink) {
+ // Optionally, if this is a symlink, try to resolve it.
+ path_lib_default = [[NSFileManager defaultManager] destinationOfSymbolicLinkAtPath:path_lib_default error:&error];
+ if (path_lib_default && [path_lib_default length] > 0 && ![[path_lib_default substringToIndex:1] isEqualToString:@"/"]) {
+ // It is a relative path, adding the binary directory as directory prefix.
+ path_lib_default = [NSString pathWithComponents:@[bin_dir, path_lib_default]];
+ }
+ if (!path_lib_default || ![[NSFileManager defaultManager] isReadableFileAtPath:path_lib_default]) {
+ // Link to the resource could not be resolved.
+ path_lib_default = nil;
+ } else {
+ GGML_LOG_INFO("%s: symlink resolved '%s'\n", __func__, [path_lib_default UTF8String]);
+ }
}
+ } else {
+ // The resource couldn't be found in the binary's directory.
+ path_lib_default = nil;
}
-#if GGML_METAL_EMBED_LIBRARY
- [src release];
-#endif // GGML_METAL_EMBED_LIBRARY
+ path_lib = path_lib_default;
+ }
+
+ if (path_lib != nil) {
+ // pre-compiled library found: a single combined default.metallib
+ NSURL * libURL = [NSURL fileURLWithPath:path_lib];
+ GGML_LOG_INFO("%s: loading '%s'\n", __func__, [path_lib UTF8String]);
+
+ res->objs[0] = [device newLibraryWithURL:libURL error:&error];
+ res->single_library = true;
+ if (!res->objs[0]) {
+ GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
+ ggml_metal_library_free(res);
+ return NULL;
+ }
GGML_LOG_INFO("%s: loaded in %.3f sec\n", __func__, (ggml_time_us() - t_start) / 1e6);
+ return res;
}
- ggml_metal_library_t res = calloc(1, sizeof(struct ggml_metal_library));
+ // no pre-compiled metallib: fall back to compiling each kernel source separately
+ GGML_LOG_INFO("%s: default.metallib not found, loading kernel sources\n", __func__);
- res->obj = library;
- res->dev = dev;
- res->pipelines = ggml_metal_pipelines_init();
- res->lock = [NSLock new];
+ NSString * path_resource = [[NSProcessInfo processInfo].environment objectForKey:@"GGML_METAL_PATH_RESOURCES"];
+ if (path_resource) {
+ GGML_LOG_INFO("%s: GGML_METAL_PATH_RESOURCES = %s\n", __func__, [path_resource UTF8String]);
+ }
+
+ // resolve each kind's source path up front (file lookup/logging stays on the calling thread)
+ NSString ** path_per_kind = calloc(GGML_METAL_LIB_COUNT, sizeof(NSString *));
+ for (int kind = 0; kind < GGML_METAL_LIB_COUNT; ++kind) {
+ NSString * rel = [NSString stringWithFormat:@"kernels/%s.metal", k_lib_names[kind]];
+
+ NSString * path_source = nil;
+ if (path_resource) {
+ path_source = [path_resource stringByAppendingPathComponent:rel];
+ } else {
+ NSString * stem = [NSString stringWithFormat:@"kernels/%s", k_lib_names[kind]];
+ path_source = [bundle pathForResource:stem ofType:@"metal"];
+ }
+
+ if (path_source == nil || ![[NSFileManager defaultManager] isReadableFileAtPath:path_source]) {
+ GGML_LOG_WARN("%s: could not locate %s in bundle, falling back to cwd\n", __func__, [rel UTF8String]);
+ path_source = rel;
+ }
+
+ GGML_LOG_DEBUG("%s: loading '%s'\n", __func__, [path_source UTF8String]);
+
+ path_per_kind[kind] = [path_source retain];
+ }
+
+ const bool ok = ggml_metal_library_compile_all(res, device, prep,
+ ^NSString * (int kind, NSError ** err) {
+ return ggml_metal_library_flatten_source(path_per_kind[kind], err);
+ }, "source");
+
+ for (int kind = 0; kind < GGML_METAL_LIB_COUNT; ++kind) {
+ [path_per_kind[kind] release];
+ }
+ free(path_per_kind);
+
+ if (!ok) {
+ ggml_metal_library_free(res);
+ return NULL;
+ }
return res;
+#endif
}
ggml_metal_library_t ggml_metal_library_init_from_source(ggml_metal_device_t dev, const char * source, bool verbose) {
@@ -319,10 +590,11 @@ ggml_metal_library_t ggml_metal_library_init_from_source(ggml_metal_device_t dev
return NULL;
}
- res->obj = library;
- res->dev = dev;
- res->pipelines = ggml_metal_pipelines_init();
- res->lock = [NSLock new];
+ res->objs[0] = library;
+ res->single_library = true;
+ res->dev = dev;
+ res->pipelines = ggml_metal_pipelines_init();
+ res->lock = [NSLock new];
return res;
}
@@ -332,8 +604,14 @@ void ggml_metal_library_free(ggml_metal_library_t lib) {
return;
}
- if (lib->obj) {
- [lib->obj release];
+ for (int kind = 0; kind < GGML_METAL_LIB_COUNT; ++kind) {
+ if (lib->objs[kind]) {
+ [lib->objs[kind] release];
+ }
+ }
+
+ if (lib->fn_to_lib) {
+ [lib->fn_to_lib release];
}
ggml_metal_pipelines_free(lib->pipelines);
@@ -394,11 +672,28 @@ struct ggml_metal_pipeline_with_params ggml_metal_library_compile_pipeline(ggml_
GGML_LOG_DEBUG("%s: compiling pipeline: base = '%s', name = '%s'\n", __func__, base, name);
+ // route to the library that actually defines this kernel; fn_to_lib is
+ // built from -[MTLLibrary functionNames] so it's always in sync
+ int lib_idx = 0;
+ if (!lib->single_library) {
+ NSNumber * idx = lib->fn_to_lib[base_func];
+ if (!idx) {
+ [lib->lock unlock];
+
+ GGML_LOG_ERROR("%s: kernel not found in any metal library: base = '%s', name = '%s'\n", __func__, base, name);
+
+ return res;
+ }
+ lib_idx = [idx intValue];
+ }
+
+ id<MTLLibrary> mtl_lib = lib->objs[lib_idx];
+
id<MTLFunction> mtl_function;
if (!cv) {
- mtl_function = [lib->obj newFunctionWithName:base_func];
+ mtl_function = [mtl_lib newFunctionWithName:base_func];
} else {
- mtl_function = [lib->obj newFunctionWithName:base_func constantValues:cv->obj error:&error];
+ mtl_function = [mtl_lib newFunctionWithName:base_func constantValues:cv->obj error:&error];
}
if (!mtl_function) {
[lib->lock unlock];
diff --git a/ggml/src/ggml-metal/ggml-metal.metal b/ggml/src/ggml-metal/ggml-metal.metal
deleted file mode 100644
index 27f97b5e07..0000000000
--- a/ggml/src/ggml-metal/ggml-metal.metal
+++ /dev/null
@@ -1,11887 +0,0 @@
-#define GGML_COMMON_DECL_METAL
-#define GGML_COMMON_IMPL_METAL
-#if defined(GGML_METAL_EMBED_LIBRARY)
-__embed_ggml-common.h__
-#else
-#include "ggml-common.h"
-#endif
-#include "ggml-metal-impl.h"
-
-#include <metal_stdlib>
-
-#ifdef GGML_METAL_HAS_TENSOR
-#include <metal_tensor>
-
-#include <MetalPerformancePrimitives/MetalPerformancePrimitives.h>
-#endif
-
-using namespace metal;
-
-#define MAX(x, y) ((x) > (y) ? (x) : (y))
-#define MIN(x, y) ((x) < (y) ? (x) : (y))
-#define SWAP(x, y) { auto tmp = (x); (x) = (y); (y) = tmp; }
-
-#define PAD2(x, n) (((x) + (n) - 1) & ~((n) - 1))
-
-#define FOR_UNROLL(x) _Pragma("clang loop unroll(full)") for (x)
-
-#define N_SIMDWIDTH 32 // assuming SIMD group size is 32
-
-// ref: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf
-//
-// cmd:
-// .../usr/bin/metal -dM -E -c ggml/src/ggml-metal/ggml-metal.metal
-// .../usr/bin/metal -dM -E -c -target air64-apple-ios14.0 ggml/src/ggml-metal/ggml-metal.metal
-//
-#if __METAL_VERSION__ < 310 && defined(GGML_METAL_HAS_BF16)
-#undef GGML_METAL_HAS_BF16
-#endif
-
-#if defined(GGML_METAL_HAS_BF16)
-typedef matrix<bfloat, 4, 4> bfloat4x4;
-typedef matrix<bfloat, 2, 4> bfloat2x4;
-#endif
-
-#define QK_NL 16
-
-constexpr constant static float kvalues_iq4nl_f[16] = {
- -127.f, -104.f, -83.f, -65.f, -49.f, -35.f, -22.f, -10.f, 1.f, 13.f, 25.f, 38.f, 53.f, 69.f, 89.f, 113.f
-};
-
-constexpr constant static float kvalues_mxfp4_f[16] = {
- 0, .5f, 1.f, 1.5f, 2.f, 3.f, 4.f, 6.f, -0, -.5f, -1.f, -1.5f, -2.f, -3.f, -4.f, -6.f
-};
-
-static inline int best_index_int8(int n, constant float * val, float x) {
- if (x <= val[0]) return 0;
- if (x >= val[n-1]) return n-1;
- int ml = 0, mu = n-1;
- while (mu-ml > 1) {
- int mav = (ml+mu)/2;
- if (x < val[mav]) mu = mav; else ml = mav;
- }
- return x - val[mu-1] < val[mu] - x ? mu-1 : mu;
-}
-
-static inline float e8m0_to_fp32(uint8_t x) {
- uint32_t bits;
-
- if (x == 0) {
- bits = 0x00400000;
- } else {
- bits = (uint32_t) x << 23;
- }
-
- return as_type<float>(bits);
-}
-
-static inline float dot(float x, float y) {
- return x*y;
-}
-
-static inline float sum(float x) {
- return x;
-}
-
-static inline float sum(float4 x) {
- return x[0] + x[1] + x[2] + x[3];
-}
-
-// NOTE: this is not dequantizing - we are simply fitting the template
-template <typename type4x4>
-void dequantize_f32(device const float4x4 * src, short il, thread type4x4 & reg) {
- reg = (type4x4)(*src);
-}
-
-template <typename type4>
-void dequantize_f32_t4(device const float4 * src, short il, thread type4 & reg) {
- reg = (type4)(*src);
-}
-
-template <typename type4x4>
-void dequantize_f16(device const half4x4 * src, short il, thread type4x4 & reg) {
- reg = (type4x4)(*src);
-}
-
-template <typename type4>
-void dequantize_f16_t4(device const half4 * src, short il, thread type4 & reg) {
- reg = (type4)(*(src));
-}
-
-#if defined(GGML_METAL_HAS_BF16)
-template <typename type4x4>
-void dequantize_bf16(device const bfloat4x4 * src, short il, thread type4x4 & reg) {
- reg = (type4x4)(*src);
-}
-
-template <typename type4>
-void dequantize_bf16_t4(device const bfloat4 * src, short il, thread type4 & reg) {
- reg = (type4)(*(src));
-}
-#endif
-
-template <typename type4x4>
-void dequantize_q1_0(device const block_q1_0 * xb, short il, thread type4x4 & reg) {
- device const uint8_t * qs = xb->qs;
- const float d = xb->d;
- const float neg_d = -d;
-
- const int byte_offset = il * 2; // il*16 bits = il*2 bytes
- const uint8_t b0 = qs[byte_offset];
- const uint8_t b1 = qs[byte_offset + 1];
-
- float4x4 reg_f;
-
- reg_f[0][0] = select(neg_d, d, bool(b0 & 0x01));
- reg_f[0][1] = select(neg_d, d, bool(b0 & 0x02));
- reg_f[0][2] = select(neg_d, d, bool(b0 & 0x04));
- reg_f[0][3] = select(neg_d, d, bool(b0 & 0x08));
- reg_f[1][0] = select(neg_d, d, bool(b0 & 0x10));
- reg_f[1][1] = select(neg_d, d, bool(b0 & 0x20));
- reg_f[1][2] = select(neg_d, d, bool(b0 & 0x40));
- reg_f[1][3] = select(neg_d, d, bool(b0 & 0x80));
-
- reg_f[2][0] = select(neg_d, d, bool(b1 & 0x01));
- reg_f[2][1] = select(neg_d, d, bool(b1 & 0x02));
- reg_f[2][2] = select(neg_d, d, bool(b1 & 0x04));
- reg_f[2][3] = select(neg_d, d, bool(b1 & 0x08));
- reg_f[3][0] = select(neg_d, d, bool(b1 & 0x10));
- reg_f[3][1] = select(neg_d, d, bool(b1 & 0x20));
- reg_f[3][2] = select(neg_d, d, bool(b1 & 0x40));
- reg_f[3][3] = select(neg_d, d, bool(b1 & 0x80));
-
- reg = (type4x4) reg_f;
-}
-
-template <typename type4>
-void dequantize_q1_0_t4(device const block_q1_0 * xb, short il, thread type4 & reg) {
- const float d = xb->d;
- const float neg_d = -d;
- const int base = il * 4;
- const uint8_t byte = xb->qs[base / 8];
- const int s = base % 8;
-
- float4 reg_f;
- reg_f[0] = select(neg_d, d, bool((byte >> (s )) & 1));
- reg_f[1] = select(neg_d, d, bool((byte >> (s + 1)) & 1));
- reg_f[2] = select(neg_d, d, bool((byte >> (s + 2)) & 1));
- reg_f[3] = select(neg_d, d, bool((byte >> (s + 3)) & 1));
-
- reg = (type4) reg_f;
-}
-
-template <typename type4x4>
-void dequantize_q2_0(device const block_q2_0 * xb, short il, thread type4x4 & reg) {
- device const uint8_t * qs = xb->qs;
- const float d = xb->d;
-
- const int byte_offset = il * 4; // il*16 elements = il*4 bytes (4 elements per byte)
- float4x4 reg_f;
-
- for (int i = 0; i < 4; i++) {
- const uint8_t b = qs[byte_offset + i];
- reg_f[i][0] = ((float)((b >> 0) & 3) - 1.0f) * d;
- reg_f[i][1] = ((float)((b >> 2) & 3) - 1.0f) * d;
- reg_f[i][2] = ((float)((b >> 4) & 3) - 1.0f) * d;
- reg_f[i][3] = ((float)((b >> 6) & 3) - 1.0f) * d;
- }
-
- reg = (type4x4) reg_f;
-}
-
-template <typename type4>
-void dequantize_q2_0_t4(device const block_q2_0 * xb, short il, thread type4 & reg) {
- const float d = xb->d;
- const uint8_t b = xb->qs[il];
-
- float4 reg_f;
- reg_f[0] = ((float)((b >> 0) & 3) - 1.0f) * d;
- reg_f[1] = ((float)((b >> 2) & 3) - 1.0f) * d;
- reg_f[2] = ((float)((b >> 4) & 3) - 1.0f) * d;
- reg_f[3] = ((float)((b >> 6) & 3) - 1.0f) * d;
-
- reg = (type4) reg_f;
-}
-
-template <typename type4x4>
-void dequantize_q4_0(device const block_q4_0 * xb, short il, thread type4x4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 1);
- const float d1 = il ? (xb->d / 16.h) : xb->d;
- const float d2 = d1 / 256.f;
- const float md = -8.h * xb->d;
- const ushort mask0 = il ? 0x00F0 : 0x000F;
- const ushort mask1 = mask0 << 8;
-
- float4x4 reg_f;
-
- for (int i = 0; i < 8; i++) {
- reg_f[i/2][2*(i%2) + 0] = d1 * (qs[i] & mask0) + md;
- reg_f[i/2][2*(i%2) + 1] = d2 * (qs[i] & mask1) + md;
- }
-
- reg = (type4x4) reg_f;
-}
-
-template <typename type4>
-void dequantize_q4_0_t4(device const block_q4_0 * xb, short il, thread type4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 1);
- const float d1 = (il/4) ? (xb->d / 16.h) : xb->d;
- const float d2 = d1 / 256.f;
- const float md = -8.h * xb->d;
- const ushort mask0 = (il/4) ? 0x00F0 : 0x000F;
- const ushort mask1 = mask0 << 8;
-
- for (int i = 0; i < 2; i++) {
- reg[2*i + 0] = d1 * (qs[2*(il%4) + i] & mask0) + md;
- reg[2*i + 1] = d2 * (qs[2*(il%4) + i] & mask1) + md;
- }
-}
-
-void quantize_q1_0(device const float * src, device block_q1_0 & dst) {
- float sum_abs = 0.0f;
- for (int j = 0; j < QK1_0; j++) {
- sum_abs += fabs(src[j]);
- }
- dst.d = sum_abs / QK1_0;
-
- for (int j = 0; j < QK1_0 / 8; j++) {
- dst.qs[j] = 0;
- }
- for (int j = 0; j < QK1_0; j++) {
- if (src[j] >= 0.0f) {
- dst.qs[j / 8] |= (1 << (j % 8));
- }
- }
-}
-
-void quantize_q2_0(device const float * src, device block_q2_0 & dst) {
- float amax = 0.0f;
- for (int j = 0; j < QK2_0; j++) {
- float a = fabs(src[j]);
- if (a > amax) amax = a;
- }
- const float d = amax;
- dst.d = d;
-
- const float id = d > 0.0f ? 1.0f / d : 0.0f;
-
- for (int j = 0; j < QK2_0 / 4; j++) {
- dst.qs[j] = 0;
- }
- for (int j = 0; j < QK2_0; j++) {
- int q = (int)round(src[j] * id) + 1;
- q = max(0, min(3, q));
- dst.qs[j / 4] |= (q << (2 * (j % 4)));
- }
-}
-
-void quantize_q4_0(device const float * src, device block_q4_0 & dst) {
-#pragma METAL fp math_mode(safe)
- float amax = 0.0f; // absolute max
- float max = 0.0f;
-
- for (int j = 0; j < QK4_0; j++) {
- const float v = src[j];
- if (amax < fabs(v)) {
- amax = fabs(v);
- max = v;
- }
- }
-
- const float d = max / -8;
- const float id = d ? 1.0f/d : 0.0f;
-
- dst.d = d;
-
- for (int j = 0; j < QK4_0/2; ++j) {
- const float x0 = src[0 + j]*id;
- const float x1 = src[QK4_0/2 + j]*id;
-
- const uint8_t xi0 = MIN(15, (int8_t)(x0 + 8.5f));
- const uint8_t xi1 = MIN(15, (int8_t)(x1 + 8.5f));
-
- dst.qs[j] = xi0;
- dst.qs[j] |= xi1 << 4;
- }
-}
-
-void quantize_q4_1(device const float * src, device block_q4_1 & dst) {
-#pragma METAL fp math_mode(safe)
- float min = FLT_MAX;
- float max = -FLT_MAX;
-
- for (int j = 0; j < QK4_1; j++) {
- const float v = src[j];
- if (min > v) min = v;
- if (max < v) max = v;
- }
-
- const float d = (max - min) / ((1 << 4) - 1);
- const float id = d ? 1.0f/d : 0.0f;
-
- dst.d = d;
- dst.m = min;
-
- for (int j = 0; j < QK4_1/2; ++j) {
- const float x0 = (src[0 + j] - min)*id;
- const float x1 = (src[QK4_1/2 + j] - min)*id;
-
- const uint8_t xi0 = MIN(15, (int8_t)(x0 + 0.5f));
- const uint8_t xi1 = MIN(15, (int8_t)(x1 + 0.5f));
-
- dst.qs[j] = xi0;
- dst.qs[j] |= xi1 << 4;
- }
-}
-
-void quantize_q5_0(device const float * src, device block_q5_0 & dst) {
-#pragma METAL fp math_mode(safe)
- float amax = 0.0f; // absolute max
- float max = 0.0f;
-
- for (int j = 0; j < QK5_0; j++) {
- const float v = src[j];
- if (amax < fabs(v)) {
- amax = fabs(v);
- max = v;
- }
- }
-
- const float d = max / -16;
- const float id = d ? 1.0f/d : 0.0f;
-
- dst.d = d;
-
- uint32_t qh = 0;
- for (int j = 0; j < QK5_0/2; ++j) {
- const float x0 = src[0 + j]*id;
- const float x1 = src[QK5_0/2 + j]*id;
-
- const uint8_t xi0 = MIN(31, (int8_t)(x0 + 16.5f));
- const uint8_t xi1 = MIN(31, (int8_t)(x1 + 16.5f));
-
- dst.qs[j] = (xi0 & 0xf) | ((xi1 & 0xf) << 4);
- qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
- qh |= ((xi1 & 0x10u) >> 4) << (j + QK5_0/2);
- }
-
- thread const uint8_t * qh8 = (thread const uint8_t *)&qh;
-
- for (int j = 0; j < 4; ++j) {
- dst.qh[j] = qh8[j];
- }
-}
-
-void quantize_q5_1(device const float * src, device block_q5_1 & dst) {
-#pragma METAL fp math_mode(safe)
- float max = src[0];
- float min = src[0];
-
- for (int j = 1; j < QK5_1; j++) {
- const float v = src[j];
- min = v < min ? v : min;
- max = v > max ? v : max;
- }
-
- const float d = (max - min) / 31;
- const float id = d ? 1.0f/d : 0.0f;
-
- dst.d = d;
- dst.m = min;
-
- uint32_t qh = 0;
- for (int j = 0; j < QK5_1/2; ++j) {
- const float x0 = (src[0 + j] - min)*id;
- const float x1 = (src[QK5_1/2 + j] - min)*id;
-
- const uint8_t xi0 = (uint8_t)(x0 + 0.5f);
- const uint8_t xi1 = (uint8_t)(x1 + 0.5f);
-
- dst.qs[j] = (xi0 & 0xf) | ((xi1 & 0xf) << 4);
- qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
- qh |= ((xi1 & 0x10u) >> 4) << (j + QK5_1/2);
- }
-
- thread const uint8_t * qh8 = (thread const uint8_t *)&qh;
-
- for (int j = 0; j < 4; ++j) {
- dst.qh[j] = qh8[j];
- }
-}
-
-void quantize_q8_0(device const float * src, device block_q8_0 & dst) {
-#pragma METAL fp math_mode(safe)
- float amax = 0.0f; // absolute max
-
- for (int j = 0; j < QK8_0; j++) {
- const float v = src[j];
- amax = MAX(amax, fabs(v));
- }
-
- const float d = amax / ((1 << 7) - 1);
- const float id = d ? 1.0f/d : 0.0f;
-
- dst.d = d;
-
- for (int j = 0; j < QK8_0; ++j) {
- const float x0 = src[j]*id;
-
- dst.qs[j] = round(x0);
- }
-}
-
-void quantize_iq4_nl(device const float * src, device block_iq4_nl & dst) {
-#pragma METAL fp math_mode(safe)
- float amax = 0.0f; // absolute max
- float max = 0.0f;
-
- for (int j = 0; j < QK4_NL; j++) {
- const float v = src[j];
- if (amax < fabs(v)) {
- amax = fabs(v);
- max = v;
- }
- }
-
- const float d = max / kvalues_iq4nl_f[0];
- const float id = d ? 1.0f/d : 0.0f;
-
- float sumqx = 0, sumq2 = 0;
- for (int j = 0; j < QK4_NL/2; ++j) {
- const float x0 = src[0 + j]*id;
- const float x1 = src[QK4_NL/2 + j]*id;
-
- const uint8_t xi0 = best_index_int8(16, kvalues_iq4nl_f, x0);
- const uint8_t xi1 = best_index_int8(16, kvalues_iq4nl_f, x1);
-
- dst.qs[j] = xi0 | (xi1 << 4);
-
- const float v0 = kvalues_iq4nl_f[xi0];
- const float v1 = kvalues_iq4nl_f[xi1];
- const float w0 = src[0 + j]*src[0 + j];
- const float w1 = src[QK4_NL/2 + j]*src[QK4_NL/2 + j];
- sumqx += w0*v0*src[j] + w1*v1*src[QK4_NL/2 + j];
- sumq2 += w0*v0*v0 + w1*v1*v1;
-
- }
-
- dst.d = sumq2 > 0 ? sumqx/sumq2 : d;
-}
-
-void quantize_tq2_0(device const float * src, device block_tq2_0 & dst) {
-#pragma METAL fp math_mode(safe)
- float amax = 0.0f; // absolute max
-
- for (int j = 0; j < QK_K; j++) {
- const float v = src[j];
- amax = MAX(amax, fabs(v));
- }
-
- const float d = amax;
- const float id = d ? 1.0f/d : 0.0f;
-
- dst.d = (half) d;
-
- for (int j = 0; j < QK_K/4; j += 32) {
- for (int m = 0; m < 32; ++m) {
- uint8_t q = 0;
- for (int n = 0; n < 4; ++n) {
- // -1, 0, 1 -> 0, 1, 2
- int xi = (int)round(src[m + n*32] * id) + 1;
- q += (uint8_t)((xi & 3) << (2*n));
- }
- dst.qs[j + m] = q;
- }
- src += 4*32;
- }
-}
-
-template <typename type4x4>
-void dequantize_q4_1(device const block_q4_1 * xb, short il, thread type4x4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 2);
- const float d1 = il ? (xb->d / 16.h) : xb->d;
- const float d2 = d1 / 256.f;
- const float m = xb->m;
- const ushort mask0 = il ? 0x00F0 : 0x000F;
- const ushort mask1 = mask0 << 8;
-
- float4x4 reg_f;
-
- for (int i = 0; i < 8; i++) {
- reg_f[i/2][2*(i%2) + 0] = ((qs[i] & mask0) * d1) + m;
- reg_f[i/2][2*(i%2) + 1] = ((qs[i] & mask1) * d2) + m;
- }
-
- reg = (type4x4) reg_f;
-}
-
-template <typename type4>
-void dequantize_q4_1_t4(device const block_q4_1 * xb, short il, thread type4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 2);
- const float d1 = (il/4) ? (xb->d / 16.h) : xb->d;
- const float d2 = d1 / 256.f;
- const float m = xb->m;
- const ushort mask0 = (il/4) ? 0x00F0 : 0x000F;
- const ushort mask1 = mask0 << 8;
-
- for (int i = 0; i < 2; i++) {
- reg[2*i + 0] = d1 * (qs[2*(il%4) + i] & mask0) + m;
- reg[2*i + 1] = d2 * (qs[2*(il%4) + i] & mask1) + m;
- }
-}
-
-template <typename type4x4>
-void dequantize_q5_0(device const block_q5_0 * xb, short il, thread type4x4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 3);
- const float d = xb->d;
- const float md = -16.h * xb->d;
- const ushort mask = il ? 0x00F0 : 0x000F;
-
- const uint32_t qh = *((device const uint32_t *)xb->qh);
-
- const int x_mv = il ? 4 : 0;
-
- const int gh_mv = il ? 12 : 0;
- const int gh_bk = il ? 0 : 4;
-
- float4x4 reg_f;
-
- for (int i = 0; i < 8; i++) {
- // extract the 5-th bits for x0 and x1
- const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
- const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
-
- // combine the 4-bits from qs with the 5th bit
- const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
- const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
-
- reg_f[i/2][2*(i%2) + 0] = d * x0 + md;
- reg_f[i/2][2*(i%2) + 1] = d * x1 + md;
- }
-
- reg = (type4x4) reg_f;
-}
-
-template <typename type4>
-void dequantize_q5_0_t4(device const block_q5_0 * xb, short il, thread type4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 3);
- const float d = xb->d;
- const float md = -16.h * xb->d;
- const ushort mask = (il/4) ? 0x00F0 : 0x000F;
-
- const uint32_t qh = *((device const uint32_t *)xb->qh);
-
- const int x_mv = (il/4) ? 4 : 0;
-
- const int gh_mv = (il/4) ? 12 : 0;
- const int gh_bk = (il/4) ? 0 : 4;
-
- for (int ii = 0; ii < 2; ii++) {
- int i = 2*(il%4) + ii;
-
- // extract the 5-th bits for x0 and x1
- const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
- const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
-
- // combine the 4-bits from qs with the 5th bit
- const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
- const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
-
- reg[2*ii + 0] = d * x0 + md;
- reg[2*ii + 1] = d * x1 + md;
- }
-}
-
-template <typename type4x4>
-void dequantize_q5_1(device const block_q5_1 * xb, short il, thread type4x4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 4);
- const float d = xb->d;
- const float m = xb->m;
- const ushort mask = il ? 0x00F0 : 0x000F;
-
- const uint32_t qh = *((device const uint32_t *)xb->qh);
-
- const int x_mv = il ? 4 : 0;
-
- const int gh_mv = il ? 12 : 0;
- const int gh_bk = il ? 0 : 4;
-
- float4x4 reg_f;
-
- for (int i = 0; i < 8; i++) {
- // extract the 5-th bits for x0 and x1
- const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
- const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
-
- // combine the 4-bits from qs with the 5th bit
- const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
- const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
-
- reg_f[i/2][2*(i%2) + 0] = d * x0 + m;
- reg_f[i/2][2*(i%2) + 1] = d * x1 + m;
- }
-
- reg = (type4x4) reg_f;
-}
-
-template <typename type4>
-void dequantize_q5_1_t4(device const block_q5_1 * xb, short il, thread type4 & reg) {
- device const uint16_t * qs = ((device const uint16_t *)xb + 4);
- const float d = xb->d;
- const float m = xb->m;
- const ushort mask = (il/4) ? 0x00F0 : 0x000F;
-
- const uint32_t qh = *((device const uint32_t *)xb->qh);
-
- const int x_mv = (il/4) ? 4 : 0;
-
- const int gh_mv = (il/4) ? 12 : 0;
- const int gh_bk = (il/4) ? 0 : 4;
-
- for (int ii = 0; ii < 2; ii++) {
- int i = 2*(il%4) + ii;
-
- // extract the 5-th bits for x0 and x1
- const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
- const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
-
- // combine the 4-bits from qs with the 5th bit
- const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
- const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
-
- reg[2*ii + 0] = d * x0 + m;
- reg[2*ii + 1] = d * x1 + m;
- }
-}
-
-template <typename type4x4>
-void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg) {
- device const packed_char4 * qs = (device const packed_char4 *) xb->qs;
- const float d = xb->d;
-
- float4x4 reg_f;
-
- for (int i = 0; i < 4; ++i) {
- reg_f[i] = float4(qs[4*il + i]) * d;
- }
-
- reg = (type4x4) reg_f;
-}
-
-template <typename type4>
-void dequantize_q8_0_t4(device const block_q8_0 *xb, short il, thread type4 & reg) {
- device const packed_char4 * qs = (device const packed_char4 *) xb->qs;
- const float d = xb->d;
-
- reg = (type4) (float4(qs[il]) * d);
-}
-
-template <typename type4x4>
-void dequantize_mxfp4(device const block_mxfp4 * xb, short il, thread type4x4 & reg) {
- device const uint8_t * q2 = (device const uint8_t *)xb->qs;
-
- const float d = e8m0_to_fp32(xb->e);
- const uint8_t shr = il >= 1 ? 4 : 0;
-
- for (int i = 0; i < 4; ++i) {
- reg[i][0] = d * kvalues_mxfp4_f[(q2[4*i + 0] >> shr) & 0x0F];
- reg[i][1] = d * kvalues_mxfp4_f[(q2[4*i + 1] >> shr) & 0x0F];
- reg[i][2] = d * kvalues_mxfp4_f[(q2[4*i + 2] >> shr) & 0x0F];
- reg[i][3] = d * kvalues_mxfp4_f[(q2[4*i + 3] >> shr) & 0x0F];
- }
-}
-
-template <typename type4>
-void dequantize_mxfp4_t4(device const block_mxfp4 * xb, short il, thread type4 & reg) {
- device const uint8_t * q2 = (device const uint8_t *)xb->qs;
-
- const float d = e8m0_to_fp32(xb->e);
- const short il4 = il%4;
-
- const uint8_t shr = il >= 4 ? 4 : 0;
-
- reg[0] = d * kvalues_mxfp4_f[(q2[4*il4 + 0] >> shr) & 0x0F];
- reg[1] = d * kvalues_mxfp4_f[(q2[4*il4 + 1] >> shr) & 0x0F];
- reg[2] = d * kvalues_mxfp4_f[(q2[4*il4 + 2] >> shr) & 0x0F];
- reg[3] = d * kvalues_mxfp4_f[(q2[4*il4 + 3] >> shr) & 0x0F];
-}
-
-template <typename type4x4>
-void dequantize_q2_K(device const block_q2_K *xb, short il, thread type4x4 & reg) {
- const float d = xb->d;
- const float min = xb->dmin;
- device const uint8_t * q = (device const uint8_t *)xb->qs;
- float dl, ml;
- uint8_t sc = xb->scales[il];
-
- q = q + 32*(il/8) + 16*(il&1);
- il = (il/2)%4;
-
- half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
- uchar mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
- dl = d * (sc & 0xF) * coef, ml = min * (sc >> 4);
- for (int i = 0; i < 16; ++i) {
- reg[i/4][i%4] = dl * (q[i] & mask) - ml;
- }
-}
-
-template <typename type4x4>
-void dequantize_q3_K(device const block_q3_K *xb, short il, thread type4x4 & reg) {
- const half d_all = xb->d;
- device const uint8_t * q = (device const uint8_t *)xb->qs;
- device const uint8_t * h = (device const uint8_t *)xb->hmask;
- device const int8_t * scales = (device const int8_t *)xb->scales;
-
- q = q + 32 * (il/8) + 16 * (il&1);
- h = h + 16 * (il&1);
- uint8_t m = 1 << (il/2);
- uint16_t kmask1 = (il/4)>1 ? ((il/4)>2 ? 192 : 48) : \
- ((il/4)>0 ? 12 : 3);
- uint16_t kmask2 = il/8 ? 0xF0 : 0x0F;
- uint16_t scale_2 = scales[il%8], scale_1 = scales[8 + il%4];
- int16_t dl_int = (il/4)&1 ? (scale_2&kmask2) | ((scale_1&kmask1) << 2)
- : (scale_2&kmask2) | ((scale_1&kmask1) << 4);
- float dl = il<8 ? d_all * (dl_int - 32.f) : d_all * (dl_int / 16.f - 32.f);
- const float ml = 4.f * dl;
-
- il = (il/2) & 3;
- const half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
- const uint8_t mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
- dl *= coef;
-
- for (int i = 0; i < 16; ++i) {
- reg[i/4][i%4] = dl * (q[i] & mask) - (h[i] & m ? 0 : ml);
- }
-}
-
-static inline uchar2 get_scale_min_k4_just2(int j, int k, device const uchar * q) {
- return j < 4 ? uchar2{uchar(q[j+0+k] & 63), uchar(q[j+4+k] & 63)}
- : uchar2{uchar((q[j+4+k] & 0xF) | ((q[j-4+k] & 0xc0) >> 2)), uchar((q[j+4+k] >> 4) | ((q[j-0+k] & 0xc0) >> 2))};
-}
-
-template <typename type4x4>
-void dequantize_q4_K(device const block_q4_K * xb, short il, thread type4x4 & reg) {
- device const uchar * q = xb->qs;
-
- short is = (il/4) * 2;
- q = q + (il/4) * 32 + 16 * (il&1);
- il = il & 3;
- const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
- const float d = il < 2 ? xb->d : xb->d / 16.h;
- const float min = xb->dmin;
- const float dl = d * sc[0];
- const float ml = min * sc[1];
-
- const ushort mask = il < 2 ? 0x0F : 0xF0;
- for (int i = 0; i < 16; ++i) {
- reg[i/4][i%4] = dl * (q[i] & mask) - ml;
- }
-}
-
-template <typename type4x4>
-void dequantize_q5_K(device const block_q5_K *xb, short il, thread type4x4 & reg) {
- device const uint8_t * q = xb->qs;
- device const uint8_t * qh = xb->qh;
-
- short is = (il/4) * 2;
- q = q + 32 * (il/4) + 16 * (il&1);
- qh = qh + 16 * (il&1);
- uint8_t ul = 1 << (il/2);
- il = il & 3;
- const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
- const float d = il < 2 ? xb->d : xb->d / 16.f;
- const float min = xb->dmin;
- const float dl = d * sc[0];
- const float ml = min * sc[1];
-
- const ushort mask = il<2 ? 0x0F : 0xF0;
- const float qh_val = il<2 ? 16.f : 256.f;
- for (int i = 0; i < 16; ++i) {
- reg[i/4][i%4] = dl * ((q[i] & mask) + (qh[i] & ul ? qh_val : 0)) - ml;
- }
-}
-
-template <typename type4x4>
-void dequantize_q6_K(device const block_q6_K *xb, short il, thread type4x4 & reg) {
- const half d_all = xb->d;
- device const uint16_t * ql = (device const uint16_t *)xb->ql;
- device const uint16_t * qh = (device const uint16_t *)xb->qh;
- device const int8_t * scales = (device const int8_t *)xb->scales;
-
- ql = ql + 32*(il/8) + 16*((il/2)&1) + 8*(il&1);
- qh = qh + 16*(il/8) + 8*(il&1);
- float sc = scales[(il%2) + 2 * ((il/2))];
- il = (il/2) & 3;
-
- const uint32_t kmask1 = il>1 ? (il>2 ? 0xC0C0C0C0 : 0x30303030) : (il>0 ? 0x0C0C0C0C : 0x03030303);
- const uint32_t kmask2 = il>1 ? 0xF0F0F0F0 : 0x0F0F0F0F;
- const float ml = d_all * sc * 32.f;
- const float dl0 = d_all * sc;
- const float dl1 = dl0 / 256.f;
- const float dl2 = dl0 / (256.f * 256.f);
- const float dl3 = dl0 / (256.f * 256.f * 256.f);
- const uint8_t shr_h = il>2 ? 2 : 0;
- const uint8_t shl_h = il>1 ? 0 : (il>0 ? 2 : 4);
- const uint8_t shr_l = il>1 ? 4 : 0;
- for (int i = 0; i < 4; ++i) {
- const uint32_t low = (ql[2*i] | (uint32_t)(ql[2*i+1] << 16)) & kmask2;
- const uint32_t high = (qh[2*i] | (uint32_t)(qh[2*i+1] << 16)) & kmask1;
- const uint32_t q = ((high << shl_h) >> shr_h) | (low >> shr_l);
- reg[i][0] = dl0 * ((half)(q & 0xFF)) - ml;
- reg[i][1] = dl1 * ((float)(q & 0xFF00)) - ml;
- reg[i][2] = dl2 * ((float)(q & 0xFF0000)) - ml;
- reg[i][3] = dl3 * ((float)(q & 0xFF000000)) - ml;
- }
-}
-
-template <typename type4x4>
-void dequantize_iq2_xxs(device const block_iq2_xxs * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const float d = xb->d;
- const int ib32 = il/2;
- il = il%2;
- // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
- // each block of 32 needs 2 uint32_t's for the quants & scale, so 4 uint16_t's.
- device const uint16_t * q2 = xb->qs + 4*ib32;
- const uint32_t aux32_g = q2[0] | (q2[1] << 16);
- const uint32_t aux32_s = q2[2] | (q2[3] << 16);
- thread const uint8_t * aux8 = (thread const uint8_t *)&aux32_g;
- const float dl = d * (0.5f + (aux32_s >> 28)) * 0.25f;
- constant uint8_t * grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+0]);
- uint8_t signs = ksigns_iq2xs[(aux32_s >> 14*il) & 127];
- for (int i = 0; i < 8; ++i) {
- reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
- }
- grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+1]);
- signs = ksigns_iq2xs[(aux32_s >> (14*il+7)) & 127];
- for (int i = 0; i < 8; ++i) {
- reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
- }
-}
-
-template <typename type4x4>
-void dequantize_iq2_xs(device const block_iq2_xs * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const float d = xb->d;
- const int ib32 = il/2;
- il = il%2;
- // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
- device const uint16_t * q2 = xb->qs + 4*ib32;
- const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
- constant uint8_t * grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+0] & 511));
- uint8_t signs = ksigns_iq2xs[q2[2*il+0] >> 9];
- for (int i = 0; i < 8; ++i) {
- reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
- }
- grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+1] & 511));
- signs = ksigns_iq2xs[q2[2*il+1] >> 9];
- for (int i = 0; i < 8; ++i) {
- reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
- }
-}
-
-template <typename type4x4>
-void dequantize_iq3_xxs(device const block_iq3_xxs * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const float d = xb->d;
- const int ib32 = il/2;
- il = il%2;
- // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
- device const uint8_t * q3 = xb->qs + 8*ib32;
- device const uint16_t * gas = (device const uint16_t *)(xb->qs + QK_K/4) + 2*ib32;
- const uint32_t aux32 = gas[0] | (gas[1] << 16);
- const float dl = d * (0.5f + (aux32 >> 28)) * 0.5f;
- constant uint8_t * grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+0]);
- constant uint8_t * grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+1]);
- uint8_t signs = ksigns_iq2xs[(aux32 >> 14*il) & 127];
- for (int i = 0; i < 4; ++i) {
- reg[0][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
- reg[1][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
- }
- grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+2]);
- grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+3]);
- signs = ksigns_iq2xs[(aux32 >> (14*il+7)) & 127];
- for (int i = 0; i < 4; ++i) {
- reg[2][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
- reg[3][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
- }
-}
-
-template <typename type4x4>
-void dequantize_iq3_s(device const block_iq3_s * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const float d = xb->d;
- const int ib32 = il/2;
- il = il%2;
- // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
- device const uint8_t * qs = xb->qs + 8*ib32;
- device const uint8_t * signs = xb->signs + 4*ib32 + 2*il;
- const uint8_t qh = xb->qh[ib32] >> 4*il;
- const float dl = d * (1 + 2*((xb->scales[ib32/2] >> 4*(ib32%2)) & 0xf));
- constant uint8_t * grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+0] | ((qh << 8) & 256)));
- constant uint8_t * grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+1] | ((qh << 7) & 256)));
- for (int i = 0; i < 4; ++i) {
- reg[0][i] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i+0]);
- reg[1][i] = dl * grid2[i] * select(1, -1, signs[0] & kmask_iq2xs[i+4]);
- }
- grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+2] | ((qh << 6) & 256)));
- grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+3] | ((qh << 5) & 256)));
- for (int i = 0; i < 4; ++i) {
- reg[2][i] = dl * grid1[i] * select(1, -1, signs[1] & kmask_iq2xs[i+0]);
- reg[3][i] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i+4]);
- }
-}
-
-template <typename type4x4>
-void dequantize_iq2_s(device const block_iq2_s * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const float d = xb->d;
- const int ib32 = il/2;
- il = il%2;
- // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
- device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
- device const uint8_t * signs = qs + QK_K/8;
- const uint8_t qh = xb->qh[ib32] >> 4*il;
- const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
- constant uint8_t * grid1 = (constant uint8_t *)(iq2s_grid + (qs[0] | ((qh << 8) & 0x300)));
- constant uint8_t * grid2 = (constant uint8_t *)(iq2s_grid + (qs[1] | ((qh << 6) & 0x300)));
- for (int i = 0; i < 8; ++i) {
- reg[i/4+0][i%4] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i]);
- reg[i/4+2][i%4] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i]);
- }
-}
-
-template <typename type4x4>
-void dequantize_iq1_s(device const block_iq1_s * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const int ib32 = il/2;
- il = il%2;
- const float d = xb->d;
- device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
- device const uint16_t * qh = xb->qh;
- const float dl = d * (2*((qh[ib32] >> 12) & 7) + 1);
- const float ml = dl * (qh[ib32] & 0x8000 ? -1 - IQ1S_DELTA : -1 + IQ1S_DELTA);
- const uint16_t h = qh[ib32] >> 6*il;
- constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((h << 8) & 0x700)));
- constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((h << 5) & 0x700)));
- for (int i = 0; i < 4; ++i) {
- reg[0][i] = dl * (grid1[i] & 0xf) + ml;
- reg[1][i] = dl * (grid1[i] >> 4) + ml;
- reg[2][i] = dl * (grid2[i] & 0xf) + ml;
- reg[3][i] = dl * (grid2[i] >> 4) + ml;
- }
-}
-
-template <typename type4x4>
-void dequantize_iq1_m(device const block_iq1_m * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const int ib32 = il/2;
- il = il%2;
- device const uint16_t * sc = (device const uint16_t *)xb->scales;
-
- iq1m_scale_t scale;
- scale.u16 = (sc[0] >> 12) | ((sc[1] >> 8) & 0x00f0) | ((sc[2] >> 4) & 0x0f00) | (sc[3] & 0xf000);
- const float d = scale.f16;
-
- device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
- device const uint8_t * qh = xb->qh + 2*ib32 + il;
-
- const float dl = d * (2*((sc[ib32/2] >> (6*(ib32%2)+3*il)) & 7) + 1);
- const float ml1 = dl * (qh[0] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
- const float ml2 = dl * (qh[0] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
- constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
- constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 4) & 0x700)));
- for (int i = 0; i < 4; ++i) {
- reg[0][i] = dl * (grid1[i] & 0xf) + ml1;
- reg[1][i] = dl * (grid1[i] >> 4) + ml1;
- reg[2][i] = dl * (grid2[i] & 0xf) + ml2;
- reg[3][i] = dl * (grid2[i] >> 4) + ml2;
- }
-}
-
-template <typename type4x4>
-void dequantize_iq4_nl(device const block_iq4_nl * xb, short il, thread type4x4 & reg) {
- device const uint16_t * q4 = (device const uint16_t *)xb->qs;
- const float d = xb->d;
- uint32_t aux32;
- thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
- for (int i = 0; i < 4; ++i) {
- aux32 = ((q4[2*i] | (q4[2*i+1] << 16)) >> 4*il) & 0x0f0f0f0f;
- reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
- reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
- reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
- reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
- }
-}
-
-template <typename type4>
-void dequantize_iq4_nl_t4(device const block_iq4_nl * xb, short il, thread type4 & reg) {
- device const uint16_t * q4 = (device const uint16_t *)xb->qs;
- const float d = xb->d;
- uint32_t aux32;
- thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
- aux32 = ((q4[2*(il%4)] | (q4[2*(il%4)+1] << 16)) >> 4*(il/4)) & 0x0f0f0f0f;
- reg[0] = d * kvalues_iq4nl_f[q8[0]];
- reg[1] = d * kvalues_iq4nl_f[q8[1]];
- reg[2] = d * kvalues_iq4nl_f[q8[2]];
- reg[3] = d * kvalues_iq4nl_f[q8[3]];
-}
-
-template <typename type4x4>
-void dequantize_iq4_xs(device const block_iq4_xs * xb, short il, thread type4x4 & reg) {
- // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
- const int ib32 = il/2;
- il = il%2;
- // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
- device const uint32_t * q4 = (device const uint32_t *)xb->qs + 4*ib32;
- const int ls = ((xb->scales_l[ib32/2] >> 4*(ib32%2)) & 0xf) | (((xb->scales_h >> 2*ib32) & 3) << 4);
- const float d = (float)xb->d * (ls - 32);
- uint32_t aux32;
- thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
- for (int i = 0; i < 4; ++i) {
- aux32 = (q4[i] >> 4*il) & 0x0f0f0f0f;
- reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
- reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
- reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
- reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
- }
-}
-
-template <typename type4x4>
-void dequantize_tq2_0(device const block_tq2_0 * xb, short il, thread type4x4 & reg) {
- device const uint8_t * qs = xb->qs;
- const float d = xb->d;
-
- float4x4 reg_f;
-
- // 2 bits per element, 4 elements per byte, 128 elements per 32-byte group
- const short base = il * 16;
- for (int k = 0; k < 16; k++) {
- const int i = base + k;
- const int byte = ((i >> 7) & 1) * 32 + (i & 31);
- const int l = (i >> 5) & 3;
- reg_f[k/4][k%4] = d * (float)(((qs[byte] >> (2*l)) & 3) - 1);
- }
-
- reg = (type4x4) reg_f;
-}
-
-enum ggml_sort_order {
- GGML_SORT_ORDER_ASC,
- GGML_SORT_ORDER_DESC,
-};
-
-constant float GELU_COEF_A = 0.044715f;
-constant float GELU_QUICK_COEF = -1.702f;
-constant float SQRT_2_OVER_PI = 0.79788456080286535587989211986876f;
-constant float SQRT_2_INV = 0.70710678118654752440084436210484f;
-
-// based on Abramowitz and Stegun formula 7.1.26 or similar Hastings' approximation
-// ref: https://www.johndcook.com/blog/python_erf/
-constant float p_erf = 0.3275911f;
-constant float a1_erf = 0.254829592f;
-constant float a2_erf = -0.284496736f;
-constant float a3_erf = 1.421413741f;
-constant float a4_erf = -1.453152027f;
-constant float a5_erf = 1.061405429f;
-
-template<typename T>
-inline T erf_approx(T x) {
- T sign_x = sign(x);
- x = fabs(x);
- T t = 1.0f / (1.0f + p_erf * x);
- T y = 1.0f - (((((a5_erf * t + a4_erf) * t) + a3_erf) * t + a2_erf) * t + a1_erf) * t * exp(-x * x);
- return sign_x * y;
-}
-
-template<typename T> T elu_approx(T x);
-
-template<> inline float elu_approx<float>(float x) {
- return (x > 0.f) ? x : (exp(x) - 1);
-}
-
-template<> inline float4 elu_approx<float4>(float4 x) {
- float4 res;
-
- res[0] = (x[0] > 0.0f) ? x[0] : (exp(x[0]) - 1.0f);
- res[1] = (x[1] > 0.0f) ? x[1] : (exp(x[1]) - 1.0f);
- res[2] = (x[2] > 0.0f) ? x[2] : (exp(x[2]) - 1.0f);
- res[3] = (x[3] > 0.0f) ? x[3] : (exp(x[3]) - 1.0f);
-
- return res;
-}
-
-constant short FC_unary_op [[function_constant(FC_UNARY + 0)]];
-constant bool FC_unary_cnt[[function_constant(FC_UNARY + 1)]];
-
-template <typename T0, typename T, typename TC>
-kernel void kernel_unary_impl(
- constant ggml_metal_kargs_unary & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
-#define FC_OP FC_unary_op
-#define FC_CNT FC_unary_cnt
-
- device const T0 * src0_ptr;
- device T * dst_ptr;
-
- int i0;
-
- if (FC_CNT) {
- i0 = tgpig.x;
-
- src0_ptr = (device const T0 *) (src0);
- dst_ptr = (device T *) (dst);
- } else {
- const int i03 = tgpig.z;
- const int i02 = tgpig.y;
- const int k0 = tgpig.x/args.ne01;
- const int i01 = tgpig.x - k0*args.ne01;
-
- i0 = k0*ntg.x + tpitg.x;
-
- src0_ptr = (device const T0 *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
- dst_ptr = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1 );
- }
-
- {
- //threadgroup_barrier(mem_flags::mem_none);
-
- if (!FC_CNT) {
- if (i0 >= args.ne0) {
- return;
- }
- }
-
- const TC x = (TC) src0_ptr[i0];
-
- if (FC_OP == OP_UNARY_NUM_SCALE) {
- dst_ptr[i0] = (T) (args.scale * x + args.bias);
- }
-
- if (FC_OP == OP_UNARY_NUM_FILL) {
- dst_ptr[i0] = (T) args.val;
- }
-
- if (FC_OP == OP_UNARY_NUM_CLAMP) {
- dst_ptr[i0] = (T) clamp(x, args.min, args.max);
- }
-
- if (FC_OP == OP_UNARY_NUM_SQR) {
- dst_ptr[i0] = (T) (x * x);
- }
-
- if (FC_OP == OP_UNARY_NUM_SQRT) {
- dst_ptr[i0] = (T) sqrt(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_SIN) {
- dst_ptr[i0] = (T) sin(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_COS) {
- dst_ptr[i0] = (T) cos(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_LOG) {
- dst_ptr[i0] = (T) log(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_LEAKY_RELU) {
- dst_ptr[i0] = (T) (TC(x > 0)*x + TC(x <= 0)*(x * args.slope));
- }
-
- if (FC_OP == OP_UNARY_NUM_TANH) {
- dst_ptr[i0] = (T) precise::tanh(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_RELU) {
- dst_ptr[i0] = (T) fmax(0, x);
- }
-
- if (FC_OP == OP_UNARY_NUM_SIGMOID) {
- dst_ptr[i0] = (T) (1 / (1 + exp(-x)));
- }
-
- if (FC_OP == OP_UNARY_NUM_GELU) {
- dst_ptr[i0] = (T) (0.5*x*(1 + precise::tanh(SQRT_2_OVER_PI*x*(1 + GELU_COEF_A*x*x))));
- }
-
- if (FC_OP == OP_UNARY_NUM_GELU_ERF) {
- dst_ptr[i0] = (T) (0.5*x*(1 + erf_approx(SQRT_2_INV*x)));
- }
-
- if (FC_OP == OP_UNARY_NUM_GELU_QUICK) {
- dst_ptr[i0] = (T) (x * (1/(1 + exp(GELU_QUICK_COEF*x))));
- }
-
- if (FC_OP == OP_UNARY_NUM_SILU) {
- dst_ptr[i0] = (T) (x / (1 + exp(-x)));
- }
-
- if (FC_OP == OP_UNARY_NUM_ELU) {
- dst_ptr[i0] = (T) elu_approx(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_NEG) {
- dst_ptr[i0] = (T) -x;
- }
-
- if (FC_OP == OP_UNARY_NUM_ABS) {
- dst_ptr[i0] = (T) fabs(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_SGN) {
- dst_ptr[i0] = T(x > 0) - T(x < 0);
- }
-
- if (FC_OP == OP_UNARY_NUM_STEP) {
- dst_ptr[i0] = T(x > 0);
- }
-
- if (FC_OP == OP_UNARY_NUM_HARDSWISH) {
- dst_ptr[i0] = (T) (x * fmax(0, fmin(1, x/6 + 0.5)));
- }
-
- if (FC_OP == OP_UNARY_NUM_HARDSIGMOID) {
- dst_ptr[i0] = (T) fmax(0, fmin(1, x/6 + 0.5));
- }
-
- if (FC_OP == OP_UNARY_NUM_EXP) {
- dst_ptr[i0] = (T) exp(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_SOFTPLUS) {
- dst_ptr[i0] = (T) select(log(1 + exp(x)), x, x > 20);
- }
-
- if (FC_OP == OP_UNARY_NUM_EXPM1) {
- // TODO: precise implementation
- dst_ptr[i0] = (T) (exp(x) - 1);
- }
-
- if (FC_OP == OP_UNARY_NUM_FLOOR) {
- dst_ptr[i0] = (T) floor(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_CEIL) {
- dst_ptr[i0] = (T) ceil(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_ROUND) {
- dst_ptr[i0] = (T) round(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_TRUNC) {
- dst_ptr[i0] = (T) trunc(x);
- }
-
- if (FC_OP == OP_UNARY_NUM_XIELU) {
- const TC xi = x;
- const TC gate = TC(xi > TC(0.0f));
- const TC clamped = fmin(xi, TC(args.val));
- const TC y_pos = TC(args.scale) * xi * xi + TC(args.bias) * xi;
- const TC y_neg = (exp(clamped) - TC(1.0f) - xi) * TC(args.slope) + TC(args.bias) * xi;
- dst_ptr[i0] = (T) (gate * y_pos + (TC(1.0f) - gate) * y_neg);
- }
- }
-
-#undef FC_OP
-#undef FC_CNT
-}
-
-typedef decltype(kernel_unary_impl<float, float, float>) kernel_unary_t;
-
-template [[host_name("kernel_unary_f32_f32")]] kernel kernel_unary_t kernel_unary_impl<float, float, float>;
-template [[host_name("kernel_unary_f32_f32_4")]] kernel kernel_unary_t kernel_unary_impl<float4, float4, float4>;
-template [[host_name("kernel_unary_f16_f16")]] kernel kernel_unary_t kernel_unary_impl<half, half, float>;
-template [[host_name("kernel_unary_f16_f16_4")]] kernel kernel_unary_t kernel_unary_impl<half4, half4, float4>;
-
-kernel void kernel_silu_back_f32(
- constant ggml_metal_kargs_silu_back & args,
- device const float * dy,
- device const float * x,
- device float * dx,
- uint gid [[thread_position_in_grid]]) {
- if (gid >= args.ne) {
- return;
- }
-
- const float s = 1.0f / (1.0f + exp(-x[gid]));
- dx[gid] = dy[gid] * s * (1.0f + x[gid] * (1.0f - s));
-}
-
-// OP: 0 - add, 1 - sub, 2 - mul, 3 - div
-constant short FC_bin_op [[function_constant(FC_BIN + 0)]];
-constant short FC_bin_f [[function_constant(FC_BIN + 1)]];
-constant bool FC_bin_rb [[function_constant(FC_BIN + 2)]];
-constant bool FC_bin_cb [[function_constant(FC_BIN + 3)]];
-
-template <typename T0, typename T1, typename T>
-kernel void kernel_bin_fuse_impl(
- constant ggml_metal_kargs_bin & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
-#define FC_OP FC_bin_op
-#define FC_F FC_bin_f
-#define FC_RB FC_bin_rb
-#define FC_CB FC_bin_cb
-
- if (FC_RB) {
- // row broadcast
- const uint i0 = tgpig.y*args.ne00 + tgpig.x;
- const uint i1 = FC_CB ? tgpig.x%args.ne10 : tgpig.x;
-
- device const T0 * src0_row = (device const T0 *) (src0);
- device T * dst_row = (device T *) (dst);
-
- if (FC_F == 1) {
- device const T1 * src1_row = (device const T1 *) (src1 + args.o1[0]);
-
- if (FC_OP == 0) {
- dst_row[i0] = src0_row[i0] + src1_row[i1];
- }
-
- if (FC_OP == 1) {
- dst_row[i0] = src0_row[i0] - src1_row[i1];
- }
-
- if (FC_OP == 2) {
- dst_row[i0] = src0_row[i0] * src1_row[i1];
- }
-
- if (FC_OP == 3) {
- dst_row[i0] = src0_row[i0] / src1_row[i1];
- }
- } else {
- T0 res = src0_row[i0];
-
- if (FC_OP == 0) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res += ((device const T1 *) (src1 + args.o1[j]))[i1];
- }
- }
-
- if (FC_OP == 1) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res -= ((device const T1 *) (src1 + args.o1[j]))[i1];
- }
- }
-
- if (FC_OP == 2) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res *= ((device const T1 *) (src1 + args.o1[j]))[i1];
- }
- }
-
- if (FC_OP == 3) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res /= ((device const T1 *) (src1 + args.o1[j]))[i1];
- }
- }
-
- dst_row[i0] = res;
- }
- } else {
- const int i03 = tgpig.z;
- const int i02 = tgpig.y;
- const int i01 = tgpig.x;
-
- if (i01 >= args.ne01) {
- return;
- }
-
- const int i13 = i03%args.ne13;
- const int i12 = i02%args.ne12;
- const int i11 = i01%args.ne11;
-
- device const T0 * src0_ptr = (device const T0 *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + args.offs);
- device T * dst_ptr = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1 + args.offs);
-
- if (FC_F == 1) {
- device const T1 * src1_ptr = (device const T1 *) (src1 + args.o1[0] + i13*args.nb13 + i12*args.nb12 + i11*args.nb11);
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- const int i10 = FC_CB ? i0%args.ne10 : i0;
-
- if (FC_OP == 0) {
- dst_ptr[i0] = src0_ptr[i0] + src1_ptr[i10];
- }
-
- if (FC_OP == 1) {
- dst_ptr[i0] = src0_ptr[i0] - src1_ptr[i10];
- }
-
- if (FC_OP == 2) {
- dst_ptr[i0] = src0_ptr[i0] * src1_ptr[i10];
- }
-
- if (FC_OP == 3) {
- dst_ptr[i0] = src0_ptr[i0] / src1_ptr[i10];
- }
- }
- } else {
- device const T1 * src1_ptr[8];
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- src1_ptr[j] = (device const T1 *) (src1 + args.o1[j] + i13*args.nb13 + i12*args.nb12 + i11*args.nb11);
- }
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- const int i10 = FC_CB ? i0%args.ne10 : i0;
-
- T res = src0_ptr[i0];
-
- if (FC_OP == 0) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res += src1_ptr[j][i10];
- }
- }
-
- if (FC_OP == 1) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res -= src1_ptr[j][i10];
- }
- }
-
- if (FC_OP == 2) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res *= src1_ptr[j][i10];
- }
- }
-
- if (FC_OP == 3) {
- FOR_UNROLL (short j = 0; j < FC_F; ++j) {
- res /= src1_ptr[j][i10];
- }
- }
-
- dst_ptr[i0] = res;
- }
- }
- }
-
-#undef FC_OP
-#undef FC_F
-#undef FC_RB
-#undef FC_CB
-}
-
-typedef decltype(kernel_bin_fuse_impl<float, float, float>) kernel_bin_fuse_t;
-
-template [[host_name("kernel_bin_fuse_f32_f32_f32")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<float, float, float>;
-template [[host_name("kernel_bin_fuse_f32_f32_f32_4")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<float4, float4, float4>;
-template [[host_name("kernel_bin_fuse_f16_f16_f16")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<half, half, half>;
-template [[host_name("kernel_bin_fuse_f16_f16_f16_4")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<half4, half4, half4>;
-
-kernel void kernel_add_id(
- constant ggml_metal_kargs_add_id & args,
- device const char * src0,
- device const char * src1,
- device const char * src2,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int i1 = tgpig.x;
- const int i2 = tgpig.y;
-
- const int i11 = *((device const int32_t *) (src2 + i1*sizeof(int32_t) + i2*args.nb21));
-
- const size_t nb1 = args.ne0 * sizeof(float);
- const size_t nb2 = args.ne1 * nb1;
-
- device float * dst_row = (device float *)((device char *)dst + i1*nb1 + i2*nb2);
- device const float * src0_row = (device const float *)((device char *)src0 + i1*args.nb01 + i2*args.nb02);
- device const float * src1_row = (device const float *)((device char *)src1 + i11*args.nb11);
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- dst_row[i0] = src0_row[i0] + src1_row[i0];
- }
-}
-
-template<typename T>
-kernel void kernel_repeat(
- constant ggml_metal_kargs_repeat & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int i3 = tgpig.z;
- const int i2 = tgpig.y;
- const int i1 = tgpig.x;
-
- const int i03 = i3%args.ne03;
- const int i02 = i2%args.ne02;
- const int i01 = i1%args.ne01;
-
- device const char * src0_ptr = src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01;
- device char * dst_ptr = dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1;
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- const int i00 = i0%args.ne00;
- *((device T *)(dst_ptr + i0*args.nb0)) = *((device T *)(src0_ptr + i00*args.nb00));
- }
-}
-
-typedef decltype(kernel_repeat<float>) kernel_repeat_t;
-
-template [[host_name("kernel_repeat_f32")]] kernel kernel_repeat_t kernel_repeat<float>;
-template [[host_name("kernel_repeat_f16")]] kernel kernel_repeat_t kernel_repeat<half>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_repeat_bf16")]] kernel kernel_repeat_t kernel_repeat<bfloat>;
-#endif
-template [[host_name("kernel_repeat_i32")]] kernel kernel_repeat_t kernel_repeat<int>;
-template [[host_name("kernel_repeat_i16")]] kernel kernel_repeat_t kernel_repeat<short>;
-
-template<typename T>
-kernel void kernel_reglu(
- constant ggml_metal_kargs_glu & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
- device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
- device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
-
- for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
- const float x0 = src0_row[i0];
- const float x1 = src1_row[i0];
-
- dst_row[i0] = (T)(x0*x1*(x0 > 0.0f));
- }
-}
-
-typedef decltype(kernel_reglu<float>) kernel_reglu_t;
-
-template [[host_name("kernel_reglu_f32")]] kernel kernel_reglu_t kernel_reglu<float>;
-template [[host_name("kernel_reglu_f16")]] kernel kernel_reglu_t kernel_reglu<half>;
-
-template<typename T>
-kernel void kernel_geglu(
- constant ggml_metal_kargs_glu & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
- device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
- device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
-
- for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
- const float x0 = src0_row[i0];
- const float x1 = src1_row[i0];
-
- const float gelu = 0.5f*x0*(1.0f + precise::tanh(SQRT_2_OVER_PI*x0*(1.0f + GELU_COEF_A*x0*x0)));
-
- dst_row[i0] = (T)(gelu*x1);
- }
-}
-
-typedef decltype(kernel_geglu<float>) kernel_geglu_t;
-
-template [[host_name("kernel_geglu_f32")]] kernel kernel_geglu_t kernel_geglu<float>;
-template [[host_name("kernel_geglu_f16")]] kernel kernel_geglu_t kernel_geglu<half>;
-
-template<typename T>
-kernel void kernel_swiglu(
- constant ggml_metal_kargs_glu & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
- device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
- device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
-
- for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
- const float x0 = src0_row[i0];
- const float x1 = src1_row[i0];
-
- const float silu = x0 / (1.0f + exp(-x0));
-
- dst_row[i0] = (T)(silu*x1);
- }
-}
-
-typedef decltype(kernel_swiglu<float>) kernel_swiglu_t;
-
-template [[host_name("kernel_swiglu_f32")]] kernel kernel_swiglu_t kernel_swiglu<float>;
-template [[host_name("kernel_swiglu_f16")]] kernel kernel_swiglu_t kernel_swiglu<half>;
-
-template<typename T>
-kernel void kernel_swiglu_oai(
- constant ggml_metal_kargs_glu & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
- device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
- device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
-
- for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
- float x0 = src0_row[i0];
- float x1 = src1_row[i0];
-
- x0 = min(x0, args.limit);
- x1 = max(min(x1, args.limit), -args.limit);
-
- float out_glu = x0 / (1.0f + exp(-x0 * args.alpha));
- out_glu = out_glu * (1.0f + x1);
-
- dst_row[i0] = (T)out_glu;
- }
-}
-
-typedef decltype(kernel_swiglu_oai<float>) kernel_swiglu_oai_t;
-
-template [[host_name("kernel_swiglu_oai_f32")]] kernel kernel_swiglu_oai_t kernel_swiglu_oai<float>;
-template [[host_name("kernel_swiglu_oai_f16")]] kernel kernel_swiglu_oai_t kernel_swiglu_oai<half>;
-
-template<typename T>
-kernel void kernel_geglu_erf(
- constant ggml_metal_kargs_glu & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
- device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
- device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
-
- for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
- const float x0 = src0_row[i0];
- const float x1 = src1_row[i0];
-
- const float gelu_erf = 0.5f*x0*(1.0f+erf_approx<float>(x0*SQRT_2_INV));
-
- dst_row[i0] = (T)(gelu_erf*x1);
- }
-}
-
-typedef decltype(kernel_geglu_erf<float>) kernel_geglu_erf_t;
-
-template [[host_name("kernel_geglu_erf_f32")]] kernel kernel_geglu_erf_t kernel_geglu_erf<float>;
-template [[host_name("kernel_geglu_erf_f16")]] kernel kernel_geglu_erf_t kernel_geglu_erf<half>;
-
-template<typename T>
-kernel void kernel_geglu_quick(
- constant ggml_metal_kargs_glu & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
- device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
- device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
-
- for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
- const float x0 = src0_row[i0];
- const float x1 = src1_row[i0];
-
- const float gelu_quick = x0*(1.0f/(1.0f+exp(GELU_QUICK_COEF*x0)));
-
- dst_row[i0] = (T)(gelu_quick*x1);
- }
-}
-
-typedef decltype(kernel_geglu_quick<float>) kernel_geglu_quick_t;
-
-template [[host_name("kernel_geglu_quick_f32")]] kernel kernel_geglu_quick_t kernel_geglu_quick<float>;
-template [[host_name("kernel_geglu_quick_f16")]] kernel kernel_geglu_quick_t kernel_geglu_quick<half>;
-
-kernel void kernel_op_sum_f32(
- constant ggml_metal_kargs_sum & args,
- device const float * src0,
- device float * dst,
- threadgroup float * shmem_f32 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
-
- if (args.np == 0) {
- return;
- }
-
- // TODO: become function constant
- const uint nsg = (ntg.x + 31) / 32;
-
- float sumf = 0;
-
- for (uint64_t i0 = tpitg.x; i0 < args.np; i0 += ntg.x) {
- sumf += src0[i0];
- }
-
- sumf = simd_sum(sumf);
-
- if (tiisg == 0) {
- shmem_f32[sgitg] = sumf;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- float total = 0;
-
- if (sgitg == 0) {
- float v = 0;
-
- if (tpitg.x < nsg) {
- v = shmem_f32[tpitg.x];
- }
-
- total = simd_sum(v);
-
- if (tpitg.x == 0) {
- dst[0] = total;
- }
- }
-}
-
-constant short FC_sum_rows_op [[function_constant(FC_SUM_ROWS + 0)]];
-
-template <typename T0, typename T>
-kernel void kernel_sum_rows_impl(
- constant ggml_metal_kargs_sum_rows & args,
- device const char * src0,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
-#define FC_OP FC_sum_rows_op
-
- const int i3 = tgpig.z;
- const int i2 = tgpig.y;
- const int i1 = tgpig.x;
-
- threadgroup T0 * shmem_t = (threadgroup T0 *) shmem;
-
- if (sgitg == 0) {
- shmem_t[tiisg] = 0.0f;
- }
-
- device const T0 * src_row = (device const T0 *) (src0 + i1*args.nb01 + i2*args.nb02 + i3*args.nb03);
- device T * dst_row = (device T *) (dst + i1*args.nb1 + i2*args.nb2 + i3*args.nb3);
-
- T0 sumf = T0(0.0f);
-
- for (int64_t i0 = tpitg.x; i0 < args.ne00; i0 += ntg.x) {
- sumf += src_row[i0];
- }
-
- sumf = simd_sum(sumf);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- shmem_t[sgitg] = sumf;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- sumf = shmem_t[tiisg];
- sumf = simd_sum(sumf);
-
- if (tpitg.x == 0) {
- if (FC_OP == OP_SUM_ROWS_NUM_MEAN) {
- if (is_same<float4, T0>::value) {
- dst_row[0] = sum(sumf) / (4*args.ne00);
- } else {
- dst_row[0] = sum(sumf) / args.ne00;
- }
- } else {
- dst_row[0] = sum(sumf);
- }
- }
-
-#undef FC_OP
-}
-
-typedef decltype(kernel_sum_rows_impl<float, float>) kernel_sum_rows_t;
-
-template [[host_name("kernel_sum_rows_f32_f32")]] kernel kernel_sum_rows_t kernel_sum_rows_impl<float, float>;
-template [[host_name("kernel_sum_rows_f32_f32_4")]] kernel kernel_sum_rows_t kernel_sum_rows_impl<float4, float>;
-
-template<typename T>
-kernel void kernel_cumsum_blk(
- constant ggml_metal_kargs_cumsum_blk & args,
- device const char * src0,
- device char * tmp,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int ib = tgpig[0]/args.ne01;
-
- const int i00 = ib*ntg.x;
- const int i01 = tgpig[0]%args.ne01;
- const int i02 = tgpig[1];
- const int i03 = tgpig[2];
-
- device const float * src0_row = (device const float *) (src0 +
- args.nb01*i01 +
- args.nb02*i02 +
- args.nb03*i03);
-
- threadgroup float * shmem_f32 = (threadgroup float *) shmem;
-
- float v = 0.0f;
-
- if (i00 + tpitg.x < args.ne00) {
- v = src0_row[i00 + tpitg.x];
- }
-
- float s = simd_prefix_inclusive_sum(v);
-
- if (tiisg == N_SIMDWIDTH - 1) {
- shmem_f32[sgitg] = s;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (sgitg == 0) {
- shmem_f32[tiisg] = simd_prefix_exclusive_sum(shmem_f32[tiisg]);
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- s += shmem_f32[sgitg];
-
- device float * dst_row = (device float *) dst +
- args.ne00*i01 +
- args.ne00*args.ne01*i02 +
- args.ne00*args.ne01*args.ne02*i03;
-
- if (i00 + tpitg.x < args.ne00) {
- dst_row[i00 + tpitg.x] = s;
- }
-
- if (args.outb && tpitg.x == ntg.x - 1) {
- device float * tmp_row = (device float *) tmp +
- args.net0*i01 +
- args.net0*args.net1*i02 +
- args.net0*args.net1*args.net2*i03;
-
- tmp_row[ib] = s;
- }
-}
-
-typedef decltype(kernel_cumsum_blk<float>) kernel_cumsum_blk_t;
-
-template [[host_name("kernel_cumsum_blk_f32")]] kernel kernel_cumsum_blk_t kernel_cumsum_blk<float>;
-
-template<typename T>
-kernel void kernel_cumsum_add(
- constant ggml_metal_kargs_cumsum_add & args,
- device const char * tmp,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int ib = tgpig[0]/args.ne01;
-
- if (ib == 0) {
- return;
- }
-
- const int i00 = ib*ntg.x;
- const int i01 = tgpig[0]%args.ne01;
- const int i02 = tgpig[1];
- const int i03 = tgpig[2];
-
- device const float * tmp_row = (device const float *) (tmp +
- args.nbt1*i01 +
- args.nbt2*i02 +
- args.nbt3*i03);
-
- device float * dst_row = (device float *) dst +
- args.ne00*i01 +
- args.ne00*args.ne01*i02 +
- args.ne00*args.ne01*args.ne02*i03;
-
- if (i00 + tpitg.x < args.ne00) {
- dst_row[i00 + tpitg.x] += tmp_row[ib - 1];
- }
-}
-
-typedef decltype(kernel_cumsum_add<float>) kernel_cumsum_add_t;
-
-template [[host_name("kernel_cumsum_add_f32")]] kernel kernel_cumsum_add_t kernel_cumsum_add<float>;
-
-
-template<uint32_t ttype>
-bool _ggml_vec_tri_cmp(const int i, const int r);
-
-template<>
-bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_LOWER */ 3>(const int i, const int r) {
- return i < r;
-}
-
-template<>
-bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_LOWER_DIAG */ 2>(const int i, const int r) {
- return i <= r;
-}
-
-template<>
-bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_UPPER */ 1>(const int i, const int r) {
- return i > r;
-}
-
-template<>
-bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_UPPER_DIAG */ 0>(const int i, const int r) {
- return i >= r;
-}
-
-template<typename T, int ttype>
-kernel void kernel_tri(
- constant ggml_metal_kargs_tri & args,
- device const char * src0,
- device const char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int i3 = tgpig.z;
- const int i2 = tgpig.y;
- const int i1 = tgpig.x;
-
- if (i3 >= args.ne03 || i2 >= args.ne02 || i1 >= args.ne01) {
- return;
- }
-
- device const T * src_row = (device const T *) ((device const char *) src0 + i1*args.nb01 + i2*args.nb02 + i3*args.nb03);
- device T * dst_row = (device T *) ((device char *) dst + i1*args.nb1 + i2*args.nb2 + i3*args.nb3);
-
- // Each thread is a single element of the row if ne00 < max threads per
- // threadgroup, so this will loop once for each index that this thread is
- // responsible for
- for (int64_t i0 = tpitg.x; i0 < args.ne00; i0 += ntg.x) {
- // Use the comparison as a mask for branchless
- dst_row[i0] = static_cast<T>(_ggml_vec_tri_cmp<ttype>(i0, i1)) * src_row[i0];
- }
-}
-
-typedef decltype(kernel_tri<float, 0>) kernel_tri_t;
-
-template [[host_name("kernel_tri_f32_0")]] kernel kernel_tri_t kernel_tri<float, 0>;
-template [[host_name("kernel_tri_f32_1")]] kernel kernel_tri_t kernel_tri<float, 1>;
-template [[host_name("kernel_tri_f32_2")]] kernel kernel_tri_t kernel_tri<float, 2>;
-template [[host_name("kernel_tri_f32_3")]] kernel kernel_tri_t kernel_tri<float, 3>;
-template [[host_name("kernel_tri_f16_0")]] kernel kernel_tri_t kernel_tri<half, 0>;
-template [[host_name("kernel_tri_f16_1")]] kernel kernel_tri_t kernel_tri<half, 1>;
-template [[host_name("kernel_tri_f16_2")]] kernel kernel_tri_t kernel_tri<half, 2>;
-template [[host_name("kernel_tri_f16_3")]] kernel kernel_tri_t kernel_tri<half, 3>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_tri_bf16_0")]] kernel kernel_tri_t kernel_tri<bfloat, 0>;
-template [[host_name("kernel_tri_bf16_1")]] kernel kernel_tri_t kernel_tri<bfloat, 1>;
-template [[host_name("kernel_tri_bf16_2")]] kernel kernel_tri_t kernel_tri<bfloat, 2>;
-template [[host_name("kernel_tri_bf16_3")]] kernel kernel_tri_t kernel_tri<bfloat, 3>;
-#endif
-
-template<typename T>
-kernel void kernel_soft_max(
- constant ggml_metal_kargs_soft_max & args,
- device const char * src0,
- device const char * src1,
- device const char * src2,
- device char * dst,
- threadgroup float * buf [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint sgitg[[simdgroup_index_in_threadgroup]],
- uint tiisg[[thread_index_in_simdgroup]],
- uint3 tptg[[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig.z;
- const int32_t i02 = tgpig.y;
- const int32_t i01 = tgpig.x;
-
- const int32_t i13 = i03%args.ne13;
- const int32_t i12 = i02%args.ne12;
- const int32_t i11 = i01;
-
- device const float * psrc0 = (device const float *) (src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
- device const T * pmask = src1 != src0 ? (device const T * ) (src1 + i11*args.nb11 + i12*args.nb12 + i13*args.nb13) : nullptr;
- device const float * psrc2 = src2 != src0 ? (device const float *) (src2) : nullptr;
- device float * pdst = (device float *) (dst + i01*args.nb1 + i02*args.nb2 + i03*args.nb3);
-
- float slope = 1.0f;
-
- // ALiBi
- if (args.max_bias > 0.0f) {
- const int32_t h = i02;
-
- const float base = h < args.n_head_log2 ? args.m0 : args.m1;
- const int exp = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
-
- slope = pow(base, exp);
- }
-
- // parallel max
- float lmax = psrc2 ? psrc2[i02] : -INFINITY;
-
- for (int i00 = tpitg.x; i00 < args.ne00; i00 += tptg.x) {
- lmax = MAX(lmax, psrc0[i00]*args.scale + (pmask ? slope*pmask[i00] : 0.0f));
- }
-
- // find the max value in the block
- float max_val = simd_max(lmax);
- if (tptg.x > N_SIMDWIDTH) {
- if (sgitg == 0) {
- buf[tiisg] = -INFINITY;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- buf[sgitg] = max_val;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- max_val = buf[tiisg];
- max_val = simd_max(max_val);
- }
-
- // parallel sum
- float lsum = 0.0f;
- for (int i00 = tpitg.x; i00 < args.ne00; i00 += tptg.x) {
- const float exp_psrc0 = exp((psrc0[i00]*args.scale + (pmask ? slope*pmask[i00] : 0.0f)) - max_val);
- lsum += exp_psrc0;
- pdst[i00] = exp_psrc0;
- }
-
- // This barrier fixes a failing test
- // ref: https://github.com/ggml-org/ggml/pull/621#discussion_r1425156335
- threadgroup_barrier(mem_flags::mem_none);
-
- float sum = simd_sum(lsum);
-
- if (tptg.x > N_SIMDWIDTH) {
- if (sgitg == 0) {
- buf[tiisg] = 0.0f;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- buf[sgitg] = sum;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- sum = buf[tiisg];
- sum = simd_sum(sum);
- }
-
- if (psrc2) {
- sum += exp(psrc2[i02] - max_val);
- }
-
- const float inv_sum = 1.0f/sum;
-
- for (int i00 = tpitg.x; i00 < args.ne00; i00 += tptg.x) {
- pdst[i00] *= inv_sum;
- }
-}
-
-template<typename T>
-kernel void kernel_soft_max_4(
- constant ggml_metal_kargs_soft_max & args,
- device const char * src0,
- device const char * src1,
- device const char * src2,
- device char * dst,
- threadgroup float * buf [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint sgitg[[simdgroup_index_in_threadgroup]],
- uint tiisg[[thread_index_in_simdgroup]],
- uint3 tptg[[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig.z;
- const int32_t i02 = tgpig.y;
- const int32_t i01 = tgpig.x;
-
- const int32_t i13 = i03%args.ne13;
- const int32_t i12 = i02%args.ne12;
- const int32_t i11 = i01;
-
- device const float4 * psrc4 = (device const float4 *) (src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
- device const T * pmask = src1 != src0 ? (device const T * ) (src1 + i11*args.nb11 + i12*args.nb12 + i13*args.nb13) : nullptr;
- device const float * psrc2 = src2 != src0 ? (device const float * ) (src2) : nullptr;
- device float4 * pdst4 = (device float4 *) (dst + i01*args.nb1 + i02*args.nb2 + i03*args.nb3);
-
- float slope = 1.0f;
-
- if (args.max_bias > 0.0f) {
- const int32_t h = i02;
-
- const float base = h < args.n_head_log2 ? args.m0 : args.m1;
- const int exp = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
-
- slope = pow(base, exp);
- }
-
- // parallel max
- float4 lmax4 = psrc2 ? psrc2[i02] : -INFINITY;
-
- for (int i00 = tpitg.x; i00 < args.ne00/4; i00 += tptg.x) {
- lmax4 = fmax(lmax4, psrc4[i00]*args.scale + (float4)((pmask ? slope*pmask[i00] : 0.0f)));
- }
-
- const float lmax = MAX(MAX(lmax4[0], lmax4[1]), MAX(lmax4[2], lmax4[3]));
-
- float max_val = simd_max(lmax);
- if (tptg.x > N_SIMDWIDTH) {
- if (sgitg == 0) {
- buf[tiisg] = -INFINITY;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- buf[sgitg] = max_val;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- max_val = buf[tiisg];
- max_val = simd_max(max_val);
- }
-
- // parallel sum
- float4 lsum4 = 0.0f;
- for (int i00 = tpitg.x; i00 < args.ne00/4; i00 += tptg.x) {
- const float4 exp_psrc4 = exp((psrc4[i00]*args.scale + (float4)((pmask ? slope*pmask[i00] : 0.0f))) - max_val);
- lsum4 += exp_psrc4;
- pdst4[i00] = exp_psrc4;
- }
-
- const float lsum = lsum4[0] + lsum4[1] + lsum4[2] + lsum4[3];
-
- // This barrier fixes a failing test
- // ref: https://github.com/ggml-org/ggml/pull/621#discussion_r1425156335
- threadgroup_barrier(mem_flags::mem_none);
-
- float sum = simd_sum(lsum);
-
- if (tptg.x > N_SIMDWIDTH) {
- if (sgitg == 0) {
- buf[tiisg] = 0.0f;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- buf[sgitg] = sum;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- sum = buf[tiisg];
- sum = simd_sum(sum);
- }
-
- if (psrc2) {
- sum += exp(psrc2[i02] - max_val);
- }
-
- const float inv_sum = 1.0f/sum;
-
- for (int i00 = tpitg.x; i00 < args.ne00/4; i00 += tptg.x) {
- pdst4[i00] *= inv_sum;
- }
-}
-
-typedef decltype(kernel_soft_max<float>) kernel_soft_max_t;
-typedef decltype(kernel_soft_max_4<float4>) kernel_soft_max_4_t;
-
-template [[host_name("kernel_soft_max_f16")]] kernel kernel_soft_max_t kernel_soft_max<half>;
-template [[host_name("kernel_soft_max_f32")]] kernel kernel_soft_max_t kernel_soft_max<float>;
-template [[host_name("kernel_soft_max_f16_4")]] kernel kernel_soft_max_4_t kernel_soft_max_4<half4>;
-template [[host_name("kernel_soft_max_f32_4")]] kernel kernel_soft_max_4_t kernel_soft_max_4<float4>;
-
-// ref: ggml.c:ggml_compute_forward_ssm_conv_f32
-kernel void kernel_ssm_conv_f32_f32(
- constant ggml_metal_kargs_ssm_conv & args,
- device const void * src0,
- device const void * src1,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
- const int64_t ir = tgpig.x;
- const int64_t i2 = tgpig.y;
- const int64_t i3 = tgpig.z;
-
- const int64_t nc = args.ne10;
- //const int64_t ncs = args.ne00;
- //const int64_t nr = args.ne01;
- //const int64_t n_t = args.ne1;
- //const int64_t n_s = args.ne2;
-
- device const float * s = (device const float *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
- device const float * c = (device const float *) ((device const char *) src1 + ir*args.nb11);
- device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
-
- float sumf = 0.0f;
-
- for (int64_t i0 = 0; i0 < nc; ++i0) {
- sumf += s[i0] * c[i0];
- }
-
- x[0] = sumf;
-}
-
-kernel void kernel_ssm_conv_f32_f32_4(
- constant ggml_metal_kargs_ssm_conv & args,
- device const void * src0,
- device const void * src1,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
- const int64_t ir = tgpig.x;
- const int64_t i2 = tgpig.y;
- const int64_t i3 = tgpig.z;
-
- const int64_t nc = args.ne10;
- //const int64_t ncs = args.ne00;
- //const int64_t nr = args.ne01;
- //const int64_t n_t = args.ne1;
- //const int64_t n_s = args.ne2;
-
- device const float4 * s = (device const float4 *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
- device const float4 * c = (device const float4 *) ((device const char *) src1 + ir*args.nb11);
- device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
-
- float sumf = 0.0f;
-
- for (int64_t i0 = 0; i0 < nc/4; ++i0) {
- sumf += dot(s[i0], c[i0]);
- }
-
- x[0] = sumf;
-}
-
-constant short FC_ssm_conv_bs [[function_constant(FC_SSM_CONV + 0)]];
-
-// Batched version: each threadgroup processes multiple tokens for better efficiency
-// Thread layout: each thread handles one token, threadgroup covers BATCH_SIZE tokens
-kernel void kernel_ssm_conv_f32_f32_batched(
- constant ggml_metal_kargs_ssm_conv & args,
- device const void * src0,
- device const void * src1,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
- // tgpig.x = row index (ir)
- // tgpig.y = batch of tokens (i2_base / BATCH_SIZE)
- // tgpig.z = sequence index (i3)
- // tpitg.x = thread within batch (0..BATCH_SIZE-1)
- const short BATCH_SIZE = FC_ssm_conv_bs;
-
- const int64_t ir = tgpig.x;
- const int64_t i2_base = tgpig.y * BATCH_SIZE;
- const int64_t i3 = tgpig.z;
- const int64_t i2_off = tpitg.x;
- const int64_t i2 = i2_base + i2_off;
-
- const int64_t nc = args.ne10; // conv kernel size (typically 4)
- const int64_t n_t = args.ne1; // number of tokens
-
- // Bounds check for partial batches at the end
- if (i2 >= n_t) {
- return;
- }
-
- // Load conv weights (shared across all tokens for this row)
- device const float * c = (device const float *) ((device const char *) src1 + ir*args.nb11);
-
- // Load source for this specific token
- device const float * s = (device const float *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
-
- // Output location for this token
- device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
-
- float sumf = 0.0f;
- for (int64_t i0 = 0; i0 < nc; ++i0) {
- sumf += s[i0] * c[i0];
- }
-
- x[0] = sumf;
-}
-
-kernel void kernel_ssm_conv_f32_f32_batched_4(
- constant ggml_metal_kargs_ssm_conv & args,
- device const void * src0,
- device const void * src1,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
- // tgpig.x = row index (ir)
- // tgpig.y = batch of tokens (i2_base / BATCH_SIZE)
- // tgpig.z = sequence index (i3)
- // tpitg.x = thread within batch (0..BATCH_SIZE-1)
- const short BATCH_SIZE = FC_ssm_conv_bs;
-
- const int64_t ir = tgpig.x;
- const int64_t i2_base = tgpig.y * BATCH_SIZE;
- const int64_t i3 = tgpig.z;
- const int64_t i2_off = tpitg.x;
- const int64_t i2 = i2_base + i2_off;
-
- const int64_t nc = args.ne10; // conv kernel size (typically 4)
- const int64_t n_t = args.ne1; // number of tokens
-
- // Bounds check for partial batches at the end
- if (i2 >= n_t) {
- return;
- }
-
- // Load conv weights (shared across all tokens for this row)
- device const float4 * c = (device const float4 *) ((device const char *) src1 + ir*args.nb11);
-
- // Load source for this specific token
- device const float4 * s = (device const float4 *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
-
- // Output location for this token
- device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
-
- float sumf = 0.0f;
- for (int64_t i0 = 0; i0 < nc/4; ++i0) {
- sumf += dot(s[i0], c[i0]);
- }
-
- x[0] = sumf;
-}
-
-// ref: ggml.c:ggml_compute_forward_ssm_scan_f32, Mamba-2 part
-// Optimized version: reduces redundant memory loads by having one thread load shared values
-kernel void kernel_ssm_scan_f32(
- constant ggml_metal_kargs_ssm_scan & args,
- device const void * src0,
- device const void * src1,
- device const void * src2,
- device const void * src3,
- device const void * src4,
- device const void * src5,
- device const void * src6,
- device float * dst,
- threadgroup float * shared [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgptg[[simdgroups_per_threadgroup]],
- uint3 tgpg[[threadgroups_per_grid]]) {
- constexpr short NW = N_SIMDWIDTH;
-
- // Shared memory layout:
- // [0..sgptg*NW-1]: partial sums for reduction (existing)
- // [sgptg*NW..sgptg*NW+sgptg-1]: pre-computed x_dt values for each token in batch
- // [sgptg*NW+sgptg..sgptg*NW+2*sgptg-1]: pre-computed dA values for each token in batch
- threadgroup float * shared_sums = shared;
- threadgroup float * shared_x_dt = shared + sgptg * NW;
- threadgroup float * shared_dA = shared + sgptg * NW + sgptg;
-
- shared_sums[tpitg.x] = 0.0f;
-
- const int32_t i0 = tpitg.x;
- const int32_t i1 = tgpig.x;
- const int32_t ir = tgpig.y; // current head
- const int32_t i3 = tgpig.z; // current seq
-
- const int32_t nc = args.d_state;
- const int32_t nr = args.d_inner;
- const int32_t nh = args.n_head;
- const int32_t ng = args.n_group;
- const int32_t n_t = args.n_seq_tokens;
- const int32_t n_s = args.n_seqs;
- const int32_t K = args.K;
-
- const int32_t s_off = args.s_off;
-
- device const int32_t * ids = (device const int32_t *) src6;
-
- device const float * s0_buff = (device const float *) ((device const char *) src0 + ir*args.nb02 + ids[i3]*args.nb03);
- device float * s_buff = (device float *) ((device char *) dst + ir*args.nb02 + i3*args.nb03 + s_off);
-
- const int32_t i = i0 + i1*nc;
- const int32_t g = ir / (nh / ng); // repeat_interleave
-
- float s0 = s0_buff[i];
- float s = 0.0f;
-
- device const float * A = (device const float *) ((device const char *) src3 + ir*args.nb31); // {ne30, nh}
-
- const float A0 = A[i0%args.ne30];
-
- device const float * x = (device const float *)((device const char *) src1 + i1*args.nb10 + ir*args.nb11 + i3*args.nb13); // {dim, nh, nt, ns}
- device const float * dt = (device const float *)((device const char *) src2 + ir*args.nb20 + i3*args.nb22); // {nh, nt, ns}
- device const float * B = (device const float *)((device const char *) src4 + g*args.nb41 + i3*args.nb43); // {d_state, ng, nt, ns}
- device const float * C = (device const float *)((device const char *) src5 + g*args.nb51 + i3*args.nb53); // {d_state, ng, nt, ns}
-
- device float * y = dst + (i1 + ir*(nr) + i3*(n_t*nh*nr)); // {dim, nh, nt, ns}
-
- for (int i2 = 0; i2 < n_t; i2 += sgptg) {
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // Pre-compute x_dt and dA for this batch of tokens
- // Only first sgptg threads do the loads and expensive math
- if (i0 < sgptg && i2 + i0 < n_t) {
- // ns12 and ns21 are element strides (nb12/nb10, nb21/nb20)
- device const float * x_t = x + i0 * args.ns12;
- device const float * dt_t = dt + i0 * args.ns21;
-
- const float dt0 = dt_t[0];
- const float dtsp = dt0 <= 20.0f ? log(1.0f + exp(dt0)) : dt0;
- shared_x_dt[i0] = x_t[0] * dtsp;
- shared_dA[i0] = dtsp; // Store dtsp, compute exp(dtsp * A0) per-thread since A0 varies
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- for (int t = 0; t < sgptg && i2 + t < n_t; t++) {
- const float x_dt = shared_x_dt[t];
- const float dA = exp(shared_dA[t] * A0);
-
- s = (s0 * dA) + (B[i0] * x_dt);
-
- const float sumf = simd_sum(s * C[i0]);
-
- if (tiisg == 0) {
- shared_sums[t*NW + sgitg] = sumf;
- }
-
- // recurse
- s0 = s;
-
- const int32_t slot = n_t - 1 - (i2 + t);
- if (slot > 0 && slot < K) {
- device float * s_snapshot = (device float *) ((device char *) s_buff + (int64_t) slot*n_s*args.nb03);
- s_snapshot[i] = s;
- }
-
- B += args.ns42;
- C += args.ns52;
- }
-
- // Advance pointers for next batch
- x += sgptg * args.ns12;
- dt += sgptg * args.ns21;
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- const float sumf = simd_sum(shared_sums[sgitg*NW + tiisg]);
-
- if (tiisg == 0 && i2 + sgitg < n_t) {
- y[sgitg*nh*nr] = sumf;
- }
-
- y += sgptg*nh*nr;
- }
-
- s_buff[i] = s;
-}
-
-kernel void kernel_rwkv_wkv6_f32(
- device const float * k,
- device const float * v,
- device const float * r,
- device const float * tf,
- device const float * td,
- device const float * state_in,
- device float * dst,
- constant uint & B,
- constant uint & T,
- constant uint & C,
- constant uint & H,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const uint head_size = 64; // TODO: support head_size = 128
- const uint batch_id = tgpig.x / H;
- const uint head_id = tgpig.x % H;
- const uint tid = tpitg.x;
-
- if (batch_id >= B || head_id >= H) {
- return;
- }
-
- const uint state_size = C * head_size;
- const uint n_seq_tokens = T / B;
-
- threadgroup float _k[head_size];
- threadgroup float _r[head_size];
- threadgroup float _tf[head_size];
- threadgroup float _td[head_size];
-
- float state[head_size];
-
- for (uint i = 0; i < head_size; i++) {
- state[i] = state_in[batch_id * state_size + head_id * head_size * head_size
- + i * head_size + tid];
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- _tf[tid] = tf[head_id * head_size + tid];
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- const uint start_t = batch_id * n_seq_tokens * C + head_id * head_size + tid;
- const uint end_t = (batch_id + 1) * n_seq_tokens * C + head_id * head_size + tid;
-
- for (uint t = start_t; t < end_t; t += C) {
- threadgroup_barrier(mem_flags::mem_threadgroup);
- _k[tid] = k[t];
- _r[tid] = r[t];
- _td[tid] = td[t];
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- const float v_val = v[t];
- float y = 0.0;
-
- for (uint j = 0; j < head_size; j += 4) {
- float4 k_vec = float4(_k[j], _k[j+1], _k[j+2], _k[j+3]);
- float4 r_vec = float4(_r[j], _r[j+1], _r[j+2], _r[j+3]);
- float4 tf_vec = float4(_tf[j], _tf[j+1], _tf[j+2], _tf[j+3]);
- float4 td_vec = float4(_td[j], _td[j+1], _td[j+2], _td[j+3]);
- float4 s_vec = float4(state[j], state[j+1], state[j+2], state[j+3]);
-
- float4 kv = k_vec * v_val;
-
- float4 temp = tf_vec * kv + s_vec;
- y += dot(r_vec, temp);
-
- s_vec = s_vec * td_vec + kv;
- state[j] = s_vec[0];
- state[j+1] = s_vec[1];
- state[j+2] = s_vec[2];
- state[j+3] = s_vec[3];
- }
-
- dst[t] = y;
- }
-
- for (uint i = 0; i < head_size; i++) {
- dst[T * C + batch_id * state_size + head_id * head_size * head_size
- + i * head_size + tid] = state[i];
- }
-}
-
-kernel void kernel_rwkv_wkv7_f32(
- device const float * r,
- device const float * w,
- device const float * k,
- device const float * v,
- device const float * a,
- device const float * b,
- device const float * state_in,
- device float * dst,
- constant uint & B,
- constant uint & T,
- constant uint & C,
- constant uint & H,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const uint head_size = 64; // TODO: support head_size = 128
- const uint batch_id = tgpig.x / H;
- const uint head_id = tgpig.x % H;
- const uint tid = tpitg.x;
-
- if (batch_id >= B || head_id >= H) {
- return;
- }
-
- const uint state_size = C * head_size;
- const uint n_seq_tokens = T / B;
-
- threadgroup float _r[head_size];
- threadgroup float _w[head_size];
- threadgroup float _k[head_size];
- threadgroup float _a[head_size];
- threadgroup float _b[head_size];
-
- float state[head_size];
-
- for (uint i = 0; i < head_size; i++) {
- state[i] = state_in[batch_id * state_size + head_id * head_size * head_size
- + tid * head_size + i];
- }
-
- const uint start_t = batch_id * n_seq_tokens * C + head_id * head_size + tid;
- const uint end_t = (batch_id + 1) * n_seq_tokens * C + head_id * head_size + tid;
-
- for (uint t = start_t; t < end_t; t += C) {
- threadgroup_barrier(mem_flags::mem_threadgroup);
- _r[tid] = r[t];
- _w[tid] = w[t];
- _k[tid] = k[t];
- _a[tid] = a[t];
- _b[tid] = b[t];
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- const float v_val = v[t];
- float y = 0.0, sa = 0.0;
-
- float4 sa_vec(0.0);
-
- for (uint j = 0; j < head_size; j += 4) {
- float4 a_vec = float4(_a[j], _a[j+1], _a[j+2], _a[j+3]);
- float4 s_vec = float4(state[j], state[j+1], state[j+2], state[j+3]);
- sa_vec += a_vec * s_vec;
- }
- sa = sa_vec[0] + sa_vec[1] + sa_vec[2] + sa_vec[3];
-
- for (uint j = 0; j < head_size; j += 4) {
- float4 r_vec = float4(_r[j], _r[j+1], _r[j+2], _r[j+3]);
- float4 w_vec = float4(_w[j], _w[j+1], _w[j+2], _w[j+3]);
- float4 k_vec = float4(_k[j], _k[j+1], _k[j+2], _k[j+3]);
- float4 b_vec = float4(_b[j], _b[j+1], _b[j+2], _b[j+3]);
- float4 s_vec = float4(state[j], state[j+1], state[j+2], state[j+3]);
-
- float4 kv = k_vec * v_val;
-
- s_vec = s_vec * w_vec + kv + sa * b_vec;
- y += dot(s_vec, r_vec);
-
- state[j] = s_vec[0];
- state[j+1] = s_vec[1];
- state[j+2] = s_vec[2];
- state[j+3] = s_vec[3];
- }
-
- dst[t] = y;
- }
-
- for (uint i = 0; i < head_size; i++) {
- dst[T * C + batch_id * state_size + head_id * head_size * head_size
- + tid * head_size + i] = state[i];
- }
-}
-
-constant short FC_gated_delta_net_ne20 [[function_constant(FC_GATED_DELTA_NET + 0)]];
-constant short FC_gated_delta_net_ne30 [[function_constant(FC_GATED_DELTA_NET + 1)]];
-constant short FC_gated_delta_net_K [[function_constant(FC_GATED_DELTA_NET + 2)]];
-
-#if 1
-template<short NSG>
-kernel void kernel_gated_delta_net_impl(
- constant ggml_metal_kargs_gated_delta_net & args,
- device const char * q,
- device const char * k,
- device const char * v,
- device const char * g,
- device const char * b,
- device const char * s,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-#define S_v FC_gated_delta_net_ne20
-#define G FC_gated_delta_net_ne30
-#define K FC_gated_delta_net_K
-
- const uint tx = tpitg.x;
- const uint ty = tpitg.y;
-
- const uint i23 = tgpig.z; // B (n_seqs)
- const uint i21 = tgpig.y; // H (head)
- const uint i20 = tgpig.x*NSG + ty; // row within S_v
-
- const uint i01 = i21 % args.ne01;
- const uint i11 = i21 % args.ne11;
-
- const float scale = 1.0f / sqrt((float)S_v);
-
- // input state layout [S_v, S_v, H, n_seqs] (s0 only): per-seq stride is H*D.
- // state is stored transposed: M[i20][is] = S[is][i20], so row i20 is contiguous
- const uint state_in_base = (i23*args.ne21 + i21)*S_v*S_v + i20*S_v;
- device const float * s_ptr = (device const float *) (s) + state_in_base;
-
- float ls[NSG];
-
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- ls[j] = s_ptr[is];
- }
-
- device float * dst_attn = (device float *) (dst) + (i23*args.ne22*args.ne21 + i21)*S_v + i20;
-
- device const float * q_ptr = (device const float *) (q + i23*args.nb03 + i01*args.nb01);
- device const float * k_ptr = (device const float *) (k + i23*args.nb13 + i11*args.nb11);
- device const float * v_ptr = (device const float *) (v + i23*args.nb23 + i21*args.nb21);
-
- device const float * b_ptr = (device const float *) (b) + (i23*args.ne22*args.ne21 + i21);
- device const float * g_ptr = (device const float *) (g) + (i23*args.ne22*args.ne21 + i21)*G;
-
- // snapshot slot mapping: slot 0 = most recent state, slot s = s tokens back.
- // When n_tokens < K, only slots 0..n_tokens-1 are written; older slots are caller-owned.
-
- // output state base offset: after attention scores
- const uint attn_size = args.ne22 * args.ne21 * S_v * args.ne23;
- // output state per-slot size: S_v * S_v * H * n_seqs
- const uint state_size_per_snap = S_v * S_v * args.ne21 * args.ne23;
- // per-(seq,head) offset within a slot
- const uint state_out_base = (i23*args.ne21 + i21)*S_v*S_v + i20*S_v;
-
- for (short t = 0; t < args.ne22; t++) {
- float s_k = 0.0f;
-
- if (G == 1) {
- const float g_exp = exp(g_ptr[0]);
-
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- ls[j] *= g_exp;
-
- s_k += ls[j]*k_ptr[is];
- }
- } else {
- // KDA
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- ls[j] *= exp(g_ptr[is]);
-
- s_k += ls[j]*k_ptr[is];
- }
- }
-
- s_k = simd_sum(s_k);
-
- const float d = (v_ptr[i20] - s_k)*b_ptr[0];
-
- float y = 0.0f;
-
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- ls[j] += k_ptr[is]*d;
-
- y += ls[j]*q_ptr[is];
- }
-
- y = simd_sum(y);
-
- if (tx == 0) {
- dst_attn[t*args.ne21*S_v] = y*scale;
- }
-
- q_ptr += args.ns02;
- k_ptr += args.ns12;
- v_ptr += args.ns22;
-
- b_ptr += args.ne21;
- g_ptr += args.ne21*G;
-
- if (K > 1) {
- const int target_slot = (int)args.ne22 - 1 - (int)t;
- if (target_slot >= 0 && target_slot < (int)K) {
- device float * dst_state = (device float *) (dst) + attn_size + (uint)target_slot * state_size_per_snap + state_out_base;
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- dst_state[is] = ls[j];
- }
- }
- }
- }
-
- if (K == 1) {
- device float * dst_state = (device float *) (dst) + attn_size + state_out_base;
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- dst_state[is] = ls[j];
- }
- }
-
-#undef S_v
-#undef G
-#undef K
-}
-
-typedef decltype(kernel_gated_delta_net_impl<4>) kernel_gated_delta_net_t;
-
-template [[host_name("kernel_gated_delta_net_f32_1")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<1>;
-template [[host_name("kernel_gated_delta_net_f32_2")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<2>;
-template [[host_name("kernel_gated_delta_net_f32_4")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<4>;
-
-#else
-// a simplified version of the above
-// no performance improvement, so keep the above version for now
-
-template<typename T, short NSG>
-kernel void kernel_gated_delta_net_impl(
- constant ggml_metal_kargs_gated_delta_net & args,
- device const char * q,
- device const char * k,
- device const char * v,
- device const char * g,
- device const char * b,
- device const char * s,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-#define S_v FC_gated_delta_net_ne20
-#define G FC_gated_delta_net_ne30
-
- const uint tx = tpitg.x;
- const uint ty = tpitg.y;
-
- const uint i23 = tgpig.z; // B
- const uint i21 = tgpig.y; // H
- const uint i20 = tgpig.x*NSG + ty;
-
- const uint i01 = i21 % args.ne01;
- const uint i11 = i21 % args.ne11;
-
- const float scale = 1.0f / sqrt((float)S_v);
-
- device const float * s_ptr = (device const float *) (s) + (i23*args.ne21 + i21)*S_v*S_v + i20;
-
- float lsf[NSG];
-
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- lsf[j] = s_ptr[is*S_v];
- }
-
- thread T * ls = (thread T *) (lsf);
-
- device float * dst_attn = (device float *) (dst) + (i23*args.ne22*args.ne21 + i21)*S_v + i20;
-
- device const float * q_ptr = (device const float *) (q + i23*args.nb03 + i01*args.nb01);
- device const float * k_ptr = (device const float *) (k + i23*args.nb13 + i11*args.nb11);
- device const float * v_ptr = (device const float *) (v + i23*args.nb23 + i21*args.nb21);
-
- device const float * b_ptr = (device const float *) (b) + (i23*args.ne22*args.ne21 + i21);
- device const float * g_ptr = (device const float *) (g) + (i23*args.ne22*args.ne21 + i21)*G;
-
- for (short t = 0; t < args.ne22; t++) {
- device const T * qt_ptr = (device const T *) (q_ptr);
- device const T * kt_ptr = (device const T *) (k_ptr);
- device const T * gt_ptr = (device const T *) (g_ptr);
-
- if (G == 1) {
- *ls *= exp(g_ptr[0]);
- } else {
- // KDA
- *ls *= exp(gt_ptr[tx]);
- }
-
- const float s_k = simd_sum(dot(*ls, kt_ptr[tx]));
-
- const float d = (v_ptr[i20] - s_k)*b_ptr[0];
-
- *ls += kt_ptr[tx]*d;
-
- const float y = simd_sum(dot(*ls, qt_ptr[tx]));
-
- if (tx == 0) {
- *dst_attn = y*scale;
- }
-
- q_ptr += args.ns02;
- k_ptr += args.ns12;
- v_ptr += args.ns22;
-
- b_ptr += args.ne21;
- g_ptr += args.ne21*G;
-
- dst_attn += args.ne21*S_v;
- }
-
- device float * dst_state = (device float *) (dst) + args.ne23*args.ne22*args.ne21*S_v + (i23*args.ne21 + i21)*S_v*S_v + i20;
- device T * dstt_state = (device T *) (dst_state);
-
- FOR_UNROLL (short j = 0; j < NSG; j++) {
- const short is = tx*NSG + j;
- dst_state[is*S_v] = lsf[j];
- }
-
-#undef S_v
-#undef G
-}
-
-typedef decltype(kernel_gated_delta_net_impl<float4, 4>) kernel_gated_delta_net_t;
-
-template [[host_name("kernel_gated_delta_net_f32_1")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<float, 1>;
-template [[host_name("kernel_gated_delta_net_f32_2")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<float2, 2>;
-template [[host_name("kernel_gated_delta_net_f32_4")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<float4, 4>;
-#endif
-
-constant short FC_solve_tri_nsg [[function_constant(FC_SOLVE_TRI + 0)]];
-constant short FC_solve_tri_n [[function_constant(FC_SOLVE_TRI + 1)]];
-constant short FC_solve_tri_k [[function_constant(FC_SOLVE_TRI + 2)]];
-
-kernel void kernel_solve_tri_f32(
- constant ggml_metal_kargs_solve_tri & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- ushort3 tgpig[[threadgroup_position_in_grid]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- constexpr short NW = N_SIMDWIDTH;
-
- const short NSG = FC_solve_tri_nsg;
- const short N = FC_solve_tri_n;
- const short K = FC_solve_tri_k;
- const short NP = PAD2(N, NW);
-
- const int32_t i03 = tgpig.z;
- const int32_t i02 = tgpig.y;
- const int32_t i01 = tgpig.x*NSG + sgitg;
-
- threadgroup float * sh0 = (threadgroup float *) shmem;
-
- device const float * src0_ptr = (device const float *)(src0 + i02 * args.nb02 + i03 * args.nb03) + sgitg*N;
- device const float * src1_ptr = (device const float *)(src1 + i02 * args.nb12 + i03 * args.nb13) + i01;
- device float * dst_ptr = (device float *)(dst + i02 * args.nb2 + i03 * args.nb3) + i01;
-
- for (short rr = 0; rr < N; rr += NSG) {
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- {
- threadgroup float * sh0_cur = sh0 + sgitg*NP;
-
- for (short t = 0; t*NW < N; ++t) {
- const short idx = t*NW + tiisg;
- sh0_cur[idx] = src0_ptr[idx];
- }
-
- src0_ptr += NSG*N;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (i01 >= args.ne10) {
- continue;
- }
-
- for (short ir = 0; ir < NSG && rr + ir < N; ++ir) {
- const short r = rr + ir;
-
- threadgroup float * sh0_cur = sh0 + ir*NP;
-
- float sum = 0.0f;
-
- for (short t = 0; t*NW < r; ++t) {
- const short idx = t*NW + tiisg;
- sum += sh0_cur[idx] * dst_ptr[idx*K] * (idx < r);
- }
-
- sum = simd_sum(sum);
-
- if (tiisg == 0) {
- const float diag = sh0_cur[r];
-
- dst_ptr[r*K] = (src1_ptr[r*K] - sum) / diag;
- }
- }
- }
-}
-
-kernel void kernel_argmax_f32(
- constant ggml_metal_kargs_argmax & args,
- device const char * src0,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint sgitg[[simdgroup_index_in_threadgroup]],
- uint tiisg[[thread_index_in_simdgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- device const float * x_row = (device const float *) ((device const char *) src0 + tgpig * args.nb01);
-
- float lmax = -INFINITY;
- int32_t larg = -1;
-
- for (int i00 = tpitg; i00 < args.ne00; i00 += ntg) {
- if (x_row[i00] > lmax) {
- lmax = x_row[i00];
- larg = i00;
- }
- }
-
- // find the argmax value in the block
- float max_val = simd_max(lmax);
- int32_t arg_val = simd_max(select(-1, larg, lmax == max_val));
-
- device int32_t * dst_i32 = (device int32_t *) dst;
-
- threadgroup float * shared_maxval = (threadgroup float *) shmem;
- threadgroup int32_t * shared_argmax = (threadgroup int32_t *) shmem + N_SIMDWIDTH;
-
- if (ntg > N_SIMDWIDTH) {
- if (sgitg == 0) {
- shared_maxval[tiisg] = -INFINITY;
- shared_argmax[tiisg] = -1;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- shared_maxval[sgitg] = max_val;
- shared_argmax[sgitg] = arg_val;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- max_val = shared_maxval[tiisg];
- arg_val = shared_argmax[tiisg];
-
- float max_val_reduced = simd_max(max_val);
- int32_t arg_val_reduced = simd_max(select(-1, arg_val, max_val == max_val_reduced));
-
- dst_i32[tgpig] = arg_val_reduced;
-
- return;
- }
-
- dst_i32[tgpig] = arg_val;
-}
-
-// F == 1 : norm (no fuse)
-// F == 2 : norm + mul
-// F == 3 : norm + mul + add
-template <typename T, short F>
-kernel void kernel_norm_fuse_impl(
- constant ggml_metal_kargs_norm & args,
- device const char * src0,
- device const char * src1_0,
- device const char * src1_1,
- device char * dst,
- threadgroup float * shmem_f32 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- if (sgitg == 0) {
- shmem_f32[tiisg] = 0.0f;
- }
-
- const int i01 = tgpig.x;
- const int i02 = tgpig.y;
- const int i03 = tgpig.z;
-
- device const T * x = (device const T *) (src0 + i03*args.nbf3[0] + i02*args.nbf2[0] + i01*args.nbf1[0]);
-
- device const T * f0 = (device const T *) (src1_0 + (i03%args.nef3[1])*args.nbf3[1] + (i02%args.nef2[1])*args.nbf2[1] + (i01%args.nef1[1])*args.nbf1[1]);
- device const T * f1 = (device const T *) (src1_1 + (i03%args.nef3[2])*args.nbf3[2] + (i02%args.nef2[2])*args.nbf2[2] + (i01%args.nef1[2])*args.nbf1[2]);
-
- T sumft(0.0f);
-
- float sumf = 0.0f;
-
- for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
- sumft += x[i00];
- }
- sumf = dot(sumft, T(1.0f));
- sumf = simd_sum(sumf);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- shmem_f32[sgitg] = sumf;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- sumf = shmem_f32[tiisg];
- sumf = simd_sum(sumf);
-
- const float mean = sumf/args.ne00;
-
- device T * y = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1);
-
- sumf = 0.0f;
- for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
- y[i00] = x[i00] - mean;
- sumf += dot(y[i00], y[i00]);
- }
- sumf = simd_sum(sumf);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- shmem_f32[sgitg] = sumf;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- sumf = shmem_f32[tiisg];
- sumf = simd_sum(sumf);
-
- const float variance = sumf/args.ne00;
-
- const float scale = 1.0f/sqrt(variance + args.eps);
- for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
- if (F == 1) {
- y[i00] = (y[i00]*scale);
- }
- if (F == 2) {
- y[i00] = (y[i00]*scale)*f0[i00];
- }
- if (F == 3) {
- y[i00] = (y[i00]*scale)*f0[i00] + f1[i00];
- }
- }
-}
-
-typedef decltype(kernel_norm_fuse_impl<float4, 1>) kernel_norm_fuse_t;
-
-template [[host_name("kernel_norm_f32")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float, 1>;
-template [[host_name("kernel_norm_mul_f32")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float, 2>;
-template [[host_name("kernel_norm_mul_add_f32")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float, 3>;
-
-template [[host_name("kernel_norm_f32_4")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float4, 1>;
-template [[host_name("kernel_norm_mul_f32_4")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float4, 2>;
-template [[host_name("kernel_norm_mul_add_f32_4")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float4, 3>;
-
-// F == 1 : rms_norm (no fuse)
-// F == 2 : rms_norm + mul
-// F == 3 : rms_norm + mul + add
-template <typename T, short F>
-kernel void kernel_rms_norm_fuse_impl(
- constant ggml_metal_kargs_norm & args,
- device const char * src0,
- device const char * src1_0,
- device const char * src1_1,
- device char * dst,
- threadgroup float * shmem_f32 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- if (sgitg == 0) {
- shmem_f32[tiisg] = 0.0f;
- }
-
- const int i01 = tgpig.x;
- const int i02 = tgpig.y;
- const int i03 = tgpig.z;
-
- device const T * x = (device const T *) (src0 + i03*args.nbf3[0] + i02*args.nbf2[0] + i01*args.nbf1[0]);
-
- device const T * f0 = (device const T *) (src1_0 + (i03%args.nef3[1])*args.nbf3[1] + (i02%args.nef2[1])*args.nbf2[1] + (i01%args.nef1[1])*args.nbf1[1]);
- device const T * f1 = (device const T *) (src1_1 + (i03%args.nef3[2])*args.nbf3[2] + (i02%args.nef2[2])*args.nbf2[2] + (i01%args.nef1[2])*args.nbf1[2]);
-
- float sumf = 0.0f;
-
- // parallel sum
- for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
- sumf += dot(x[i00], x[i00]);
- }
- sumf = simd_sum(sumf);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- shmem_f32[sgitg] = sumf;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- sumf = shmem_f32[tiisg];
- sumf = simd_sum(sumf);
-
- const float mean = sumf/args.ne00;
- const float scale = 1.0f/sqrt(mean + args.eps);
-
- device T * y = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1);
- for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
- if (F == 1) {
- y[i00] = (x[i00]*scale);
- }
- if (F == 2) {
- y[i00] = (x[i00]*scale)*f0[i00];
- }
- if (F == 3) {
- y[i00] = (x[i00]*scale)*f0[i00] + f1[i00];
- }
- }
-}
-
-typedef decltype(kernel_rms_norm_fuse_impl<float4, 1>) kernel_rms_norm_fuse_t;
-
-template [[host_name("kernel_rms_norm_f32")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float, 1>;
-template [[host_name("kernel_rms_norm_mul_f32")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float, 2>;
-template [[host_name("kernel_rms_norm_mul_add_f32")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float, 3>;
-
-template [[host_name("kernel_rms_norm_f32_4")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float4, 1>;
-template [[host_name("kernel_rms_norm_mul_f32_4")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float4, 2>;
-template [[host_name("kernel_rms_norm_mul_add_f32_4")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float4, 3>;
-
-template <typename T0, typename T>
-kernel void kernel_l2_norm_impl(
- constant ggml_metal_kargs_l2_norm & args,
- device const char * src0,
- device char * dst,
- threadgroup float * shmem_f32 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int i03 = tgpig.z;
- const int i02 = tgpig.y;
- const int i01 = tgpig.x;
-
- if (sgitg == 0) {
- shmem_f32[tiisg] = 0.0f;
- }
-
- device const T0 * x = (device const T0 *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
- device T * y = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1);
-
- float sumf = 0.0f;
-
- // parallel sum
- for (int i00 = tpitg.x; i00 < args.ne00; i00 += ntg.x) {
- sumf += dot(x[i00], x[i00]);
- }
- sumf = simd_sum(sumf);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- shmem_f32[sgitg] = sumf;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- sumf = shmem_f32[tiisg];
- sumf = simd_sum(sumf);
-
- const float scale = 1.0f/max(sqrt(sumf), args.eps);
-
- for (int i00 = tpitg.x; i00 < args.ne00; i00 += ntg.x) {
- y[i00] = x[i00] * scale;
- }
-}
-
-typedef decltype(kernel_l2_norm_impl<float, float>) kernel_l2_norm_t;
-
-template [[host_name("kernel_l2_norm_f32_f32")]] kernel kernel_l2_norm_t kernel_l2_norm_impl<float, float>;
-template [[host_name("kernel_l2_norm_f32_f32_4")]] kernel kernel_l2_norm_t kernel_l2_norm_impl<float4, float4>;
-
-kernel void kernel_group_norm_f32(
- constant ggml_metal_kargs_group_norm & args,
- device const float * src0,
- device float * dst,
- threadgroup float * buf [[threadgroup(0)]],
- uint tgpig[[threadgroup_position_in_grid]],
- uint tpitg[[thread_position_in_threadgroup]],
- uint sgitg[[simdgroup_index_in_threadgroup]],
- uint tiisg[[thread_index_in_simdgroup]],
- uint ntg[[threads_per_threadgroup]]) {
- const int64_t ne = args.ne00*args.ne01*args.ne02;
- const int64_t gs = args.ne00*args.ne01*((args.ne02 + args.ngrp - 1) / args.ngrp);
-
- int start = tgpig * gs;
- int end = start + gs;
-
- start += tpitg;
-
- if (end >= ne) {
- end = ne;
- }
-
- float tmp = 0.0f; // partial sum for thread in warp
-
- for (int j = start; j < end; j += ntg) {
- tmp += src0[j];
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- tmp = simd_sum(tmp);
- if (ntg > N_SIMDWIDTH) {
- if (sgitg == 0) {
- buf[tiisg] = 0.0f;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- buf[sgitg] = tmp;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- tmp = buf[tiisg];
- tmp = simd_sum(tmp);
- }
-
- const float mean = tmp / gs;
- tmp = 0.0f;
-
- for (int j = start; j < end; j += ntg) {
- float xi = src0[j] - mean;
- dst[j] = xi;
- tmp += xi * xi;
- }
-
- tmp = simd_sum(tmp);
- if (ntg > N_SIMDWIDTH) {
- if (sgitg == 0) {
- buf[tiisg] = 0.0f;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tiisg == 0) {
- buf[sgitg] = tmp;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- tmp = buf[tiisg];
- tmp = simd_sum(tmp);
- }
-
- const float variance = tmp / gs;
- const float scale = 1.0f/sqrt(variance + args.eps);
- for (int j = start; j < end; j += ntg) {
- dst[j] *= scale;
- }
-}
-
-// Q1_0 dot product: dot = d * (2 * Σ(yl[i] where bit=1) - sumy)
-inline float block_q_n_dot_y(device const block_q1_0 * qb_curr, float sumy, thread float * yl, int il) {
- device const uint8_t * qs = qb_curr->qs + il / 8;
- const uint8_t b0 = qs[0];
- const uint8_t b1 = qs[1];
-
- float acc = 0.0f;
-
- acc += select(0.0f, yl[ 0], bool(b0 & 0x01));
- acc += select(0.0f, yl[ 1], bool(b0 & 0x02));
- acc += select(0.0f, yl[ 2], bool(b0 & 0x04));
- acc += select(0.0f, yl[ 3], bool(b0 & 0x08));
- acc += select(0.0f, yl[ 4], bool(b0 & 0x10));
- acc += select(0.0f, yl[ 5], bool(b0 & 0x20));
- acc += select(0.0f, yl[ 6], bool(b0 & 0x40));
- acc += select(0.0f, yl[ 7], bool(b0 & 0x80));
-
- acc += select(0.0f, yl[ 8], bool(b1 & 0x01));
- acc += select(0.0f, yl[ 9], bool(b1 & 0x02));
- acc += select(0.0f, yl[10], bool(b1 & 0x04));
- acc += select(0.0f, yl[11], bool(b1 & 0x08));
- acc += select(0.0f, yl[12], bool(b1 & 0x10));
- acc += select(0.0f, yl[13], bool(b1 & 0x20));
- acc += select(0.0f, yl[14], bool(b1 & 0x40));
- acc += select(0.0f, yl[15], bool(b1 & 0x80));
-
- return qb_curr->d * (2.0f * acc - sumy);
-}
-
-// Q2_0 dot: d * (sum_lo(y) + 2*sum_hi(y) - sumy) via per-bit conditional adds
-inline float block_q_n_dot_y(device const block_q2_0 * qb_curr, float sumy, thread float * yl, int il) {
- device const uint8_t * qs = qb_curr->qs + (il / 4);
- const uint8_t b0 = qs[0];
- const uint8_t b1 = qs[1];
- const uint8_t b2 = qs[2];
- const uint8_t b3 = qs[3];
-
- // Accumulate where low bit is set (bits 0,2,4,6 of each byte)
- float acc_lo = 0.0f;
- acc_lo += select(0.0f, yl[ 0], bool(b0 & 0x01));
- acc_lo += select(0.0f, yl[ 1], bool(b0 & 0x04));
- acc_lo += select(0.0f, yl[ 2], bool(b0 & 0x10));
- acc_lo += select(0.0f, yl[ 3], bool(b0 & 0x40));
- acc_lo += select(0.0f, yl[ 4], bool(b1 & 0x01));
- acc_lo += select(0.0f, yl[ 5], bool(b1 & 0x04));
- acc_lo += select(0.0f, yl[ 6], bool(b1 & 0x10));
- acc_lo += select(0.0f, yl[ 7], bool(b1 & 0x40));
- acc_lo += select(0.0f, yl[ 8], bool(b2 & 0x01));
- acc_lo += select(0.0f, yl[ 9], bool(b2 & 0x04));
- acc_lo += select(0.0f, yl[10], bool(b2 & 0x10));
- acc_lo += select(0.0f, yl[11], bool(b2 & 0x40));
- acc_lo += select(0.0f, yl[12], bool(b3 & 0x01));
- acc_lo += select(0.0f, yl[13], bool(b3 & 0x04));
- acc_lo += select(0.0f, yl[14], bool(b3 & 0x10));
- acc_lo += select(0.0f, yl[15], bool(b3 & 0x40));
-
- // Accumulate where high bit is set (bits 1,3,5,7 of each byte)
- float acc_hi = 0.0f;
- acc_hi += select(0.0f, yl[ 0], bool(b0 & 0x02));
- acc_hi += select(0.0f, yl[ 1], bool(b0 & 0x08));
- acc_hi += select(0.0f, yl[ 2], bool(b0 & 0x20));
- acc_hi += select(0.0f, yl[ 3], bool(b0 & 0x80));
- acc_hi += select(0.0f, yl[ 4], bool(b1 & 0x02));
- acc_hi += select(0.0f, yl[ 5], bool(b1 & 0x08));
- acc_hi += select(0.0f, yl[ 6], bool(b1 & 0x20));
- acc_hi += select(0.0f, yl[ 7], bool(b1 & 0x80));
- acc_hi += select(0.0f, yl[ 8], bool(b2 & 0x02));
- acc_hi += select(0.0f, yl[ 9], bool(b2 & 0x08));
- acc_hi += select(0.0f, yl[10], bool(b2 & 0x20));
- acc_hi += select(0.0f, yl[11], bool(b2 & 0x80));
- acc_hi += select(0.0f, yl[12], bool(b3 & 0x02));
- acc_hi += select(0.0f, yl[13], bool(b3 & 0x08));
- acc_hi += select(0.0f, yl[14], bool(b3 & 0x20));
- acc_hi += select(0.0f, yl[15], bool(b3 & 0x80));
-
- return qb_curr->d * (acc_lo + 2.0f * acc_hi - sumy);
-}
-
-// function for calculate inner product between half a q4_0 block and 16 floats (yl), sumy is SUM(yl[i])
-// il indicates where the q4 quants begin (0 or QK4_0/4)
-// we assume that the yl's have been multiplied with the appropriate scale factor
-// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
-inline float block_q_n_dot_y(device const block_q4_0 * qb_curr, float sumy, thread float * yl, int il) {
- float d = qb_curr->d;
-
- float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
-
- device const uint16_t * qs = ((device const uint16_t *) qb_curr + 1 + il/2);
-
- for (int i = 0; i < 8; i += 2) {
- acc[0] += yl[i + 0] * (qs[i / 2] & 0x000F);
- acc[1] += yl[i + 1] * (qs[i / 2] & 0x0F00);
- acc[2] += yl[i + 8] * (qs[i / 2] & 0x00F0);
- acc[3] += yl[i + 9] * (qs[i / 2] & 0xF000);
- }
-
- return d * (sumy * -8.f + acc[0] + acc[1] + acc[2] + acc[3]);
-}
-
-// function for calculate inner product between half a q4_1 block and 16 floats (yl), sumy is SUM(yl[i])
-// il indicates where the q4 quants begin (0 or QK4_0/4)
-// we assume that the yl's have been multiplied with the appropriate scale factor
-// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
-inline float block_q_n_dot_y(device const block_q4_1 * qb_curr, float sumy, thread float * yl, int il) {
- float d = qb_curr->d;
- float m = qb_curr->m;
-
- float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
-
- device const uint16_t * qs = ((device const uint16_t *) qb_curr + 2 + il/2);
-
- for (int i = 0; i < 8; i+=2) {
- acc[0] += yl[i + 0] * (qs[i / 2] & 0x000F);
- acc[1] += yl[i + 1] * (qs[i / 2] & 0x0F00);
- acc[2] += yl[i + 8] * (qs[i / 2] & 0x00F0);
- acc[3] += yl[i + 9] * (qs[i / 2] & 0xF000);
- }
-
- return d * (acc[0] + acc[1] + acc[2] + acc[3]) + sumy * m;
-}
-
-// function for calculate inner product between half a q5_0 block and 16 floats (yl), sumy is SUM(yl[i])
-// il indicates where the q5 quants begin (0 or QK5_0/4)
-// we assume that the yl's have been multiplied with the appropriate scale factor
-// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
-inline float block_q_n_dot_y(device const block_q5_0 * qb_curr, float sumy, thread float * yl, int il) {
- float d = qb_curr->d;
-
- float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
-
- device const uint16_t * qs = ((device const uint16_t *)qb_curr + 3 + il/2);
- const uint32_t qh = *((device const uint32_t *)qb_curr->qh);
-
- for (int i = 0; i < 8; i+=2) {
- acc[0] += yl[i + 0] * ((qs[i / 2] & 0x000F) | ((qh >> (i+0+il ) << 4 ) & 0x00010));
- acc[1] += yl[i + 1] * ((qs[i / 2] & 0x0F00) | ((qh >> (i+1+il ) << 12) & 0x01000));
- acc[2] += yl[i + 8] * ((qs[i / 2] & 0x00F0) | ((qh >> (i+0+il+QK5_0/2) << 8 ) & 0x00100));
- acc[3] += yl[i + 9] * ((qs[i / 2] & 0xF000) | ((qh >> (i+1+il+QK5_0/2) << 16) & 0x10000));
- }
-
- return d * (sumy * -16.f + acc[0] + acc[1] + acc[2] + acc[3]);
-}
-
-// function for calculate inner product between half a q5_1 block and 16 floats (yl), sumy is SUM(yl[i])
-// il indicates where the q5 quants begin (0 or QK5_1/4)
-// we assume that the yl's have been multiplied with the appropriate scale factor
-// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
-inline float block_q_n_dot_y(device const block_q5_1 * qb_curr, float sumy, thread float * yl, int il) {
- float d = qb_curr->d;
- float m = qb_curr->m;
-
- float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
-
- device const uint16_t * qs = ((device const uint16_t *)qb_curr + 4 + il/2);
- const uint32_t qh = *((device const uint32_t *)qb_curr->qh);
-
- for (int i = 0; i < 8; i+=2) {
- acc[0] += yl[i + 0] * ((qs[i / 2] & 0x000F) | ((qh >> (i+0+il ) << 4 ) & 0x00010));
- acc[1] += yl[i + 1] * ((qs[i / 2] & 0x0F00) | ((qh >> (i+1+il ) << 12) & 0x01000));
- acc[2] += yl[i + 8] * ((qs[i / 2] & 0x00F0) | ((qh >> (i+0+il+QK5_0/2) << 8 ) & 0x00100));
- acc[3] += yl[i + 9] * ((qs[i / 2] & 0xF000) | ((qh >> (i+1+il+QK5_0/2) << 16) & 0x10000));
- }
-
- return d * (acc[0] + acc[1] + acc[2] + acc[3]) + sumy * m;
-}
-
-template<short NR0>
-static inline void helper_mv_reduce_and_write(
- device float * dst_f32,
- float sumf[NR0],
- const int r0,
- const int ne01,
- ushort tiisg,
- ushort sgitg,
- threadgroup char * shmem) {
- constexpr short NW = N_SIMDWIDTH;
-
- threadgroup float * shmem_f32[NR0];
-
- for (short row = 0; row < NR0; ++row) {
- shmem_f32[row] = (threadgroup float *) shmem + NW*row;
-
- if (sgitg == 0) {
- shmem_f32[row][tiisg] = 0.0f;
- }
-
- sumf[row] = simd_sum(sumf[row]);
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- for (short row = 0; row < NR0; ++row) {
- if (tiisg == 0) {
- shmem_f32[row][sgitg] = sumf[row];
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- for (short row = 0; row < NR0 && r0 + row < ne01; ++row) {
- float tot = simd_sum(shmem_f32[row][tiisg]);
-
- if (tiisg == 0 && sgitg == 0) {
- dst_f32[r0 + row] = tot;
- }
- }
-}
-
-constant short FC_mul_mv_nsg [[function_constant(FC_MUL_MV + 0)]];
-constant short FC_mul_mv_nxpsg [[function_constant(FC_MUL_MV + 1)]];
-constant short FC_mul_mv_ne12 [[function_constant(FC_MUL_MV + 2)]];
-constant short FC_mul_mv_r2 [[function_constant(FC_MUL_MV + 3)]];
-constant short FC_mul_mv_r3 [[function_constant(FC_MUL_MV + 4)]];
-
-template<typename block_q_type, short NR0, typename args_t>
-void mul_vec_q_n_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- constexpr short NW = N_SIMDWIDTH;
- constexpr short NQ = 16;
-
- const int nb = args.ne00/QK4_0;
-
- const int r0 = (tgpig.x*NSG + sgitg)*NR0;
- //const int r0 = tgpig.x*NR0;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- //device const block_q_type * x = (device const block_q_type *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- // pointers to src0 rows
- device const block_q_type * ax[NR0];
- FOR_UNROLL (int row = 0; row < NR0; ++row) {
- const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
-
- ax[row] = (device const block_q_type *) ((device char *) src0 + offset0);
- }
-
- float sumf[NR0] = {0.f};
-
- const short ix = (tiisg/(NW/NQ));
- const short il = (tiisg%(NW/NQ))*8;
-
- //const int ib0 = sgitg*NQ + ix;
- const int ib0 = ix;
-
- float yl[16]; // src1 vector cache
-
- //device const float * yb = y + ix*QK4_0 + il;
- device const float * yb = y + ib0*QK4_0 + il;
-
- // each thread in a SIMD group deals with half a block.
- //for (int ib = ib0; ib < nb; ib += NSG*NQ) {
- for (int ib = ib0; ib < nb; ib += NQ) {
- float sumy[2] = { 0.f, 0.f };
-
- FOR_UNROLL (short i = 0; i < 8; i += 2) {
- sumy[0] += yb[i + 0] + yb[i + 1];
- yl[i + 0] = yb[i + 0];
- yl[i + 1] = yb[i + 1]/256.f;
-
- sumy[1] += yb[i + 16] + yb[i + 17];
- yl[i + 8] = yb[i + 16]/16.f;
- yl[i + 9] = yb[i + 17]/4096.f;
- }
-
- FOR_UNROLL (short row = 0; row < NR0; row++) {
- sumf[row] += block_q_n_dot_y(ax[row] + ib, sumy[0] + sumy[1], yl, il);
- }
-
- yb += QK4_0 * 16;
- //yb += NSG*NQ*QK4_0;
- }
-
- device float * dst_f32 = (device float *) dst + im*args.ne0*args.ne1 + r1*args.ne0;
-
- //helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
-
- for (int row = 0; row < NR0; ++row) {
- const float tot = simd_sum(sumf[row]);
-
- if (tiisg == 0 && r0 + row < args.ne01) {
- dst_f32[r0 + row] = tot;
- }
- }
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_q1_0_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK1_0;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset1 = r1*args.nb11 + (i12)*args.nb12 + (i13)*args.nb13;
-
- device const float * y = (device const float *) (src1 + offset1);
-
- device const block_q1_0 * ax[nr0];
- for (int row = 0; row < nr0; ++row) {
- const uint64_t offset0 = (first_row + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- ax[row] = (device const block_q1_0 *) ((device char *) src0 + offset0);
- }
-
- float yl[16];
- float sumf[nr0] = {0.f};
-
- const short ix = (tiisg/8);
- const short il = (tiisg%8)*16;
-
- device const float * yb = y + ix*QK1_0 + il;
-
- for (int ib = ix; ib < nb; ib += N_SIMDWIDTH/8) {
- float sumy = 0.f;
-
- FOR_UNROLL (short i = 0; i < 16; i++) {
- yl[i] = yb[i];
- sumy += yb[i];
- }
-
- FOR_UNROLL (short row = 0; row < nr0; row++) {
- sumf[row] += block_q_n_dot_y(ax[row] + ib, sumy, yl, il);
- }
-
- yb += QK1_0 * (N_SIMDWIDTH/8);
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0; ++row) {
- const float tot = simd_sum(sumf[row]);
-
- if (tiisg == 0 && first_row + row < args.ne01) {
- dst_f32[first_row + row] = tot;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_q1_0_f32")]]
-kernel void kernel_mul_mv_q1_0_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_q1_0_f32_impl<N_R0_Q1_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_q2_0_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK2_0;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset1 = r1*args.nb11 + (i12)*args.nb12 + (i13)*args.nb13;
-
- device const float * y = (device const float *) (src1 + offset1);
-
- device const block_q2_0 * ax[nr0];
- for (int row = 0; row < nr0; ++row) {
- const uint64_t offset0 = (first_row + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- ax[row] = (device const block_q2_0 *) ((device char *) src0 + offset0);
- }
-
- float yl[16];
- float sumf[nr0] = {0.f};
-
- // group 64: 4 sub-blocks of 16 weights per Q2_0 block
- const short ix = (tiisg/4);
- const short il = (tiisg%4)*16;
-
- device const float * yb = y + ix*QK2_0 + il;
-
- for (int ib = ix; ib < nb; ib += N_SIMDWIDTH/4) {
- float sumy = 0.f;
-
- FOR_UNROLL (short i = 0; i < 16; i++) {
- yl[i] = yb[i];
- sumy += yb[i];
- }
-
- FOR_UNROLL (short row = 0; row < nr0; row++) {
- sumf[row] += block_q_n_dot_y(ax[row] + ib, sumy, yl, il);
- }
-
- yb += QK2_0 * (N_SIMDWIDTH/4);
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0; ++row) {
- const float tot = simd_sum(sumf[row]);
-
- if (tiisg == 0 && first_row + row < args.ne01) {
- dst_f32[first_row + row] = tot;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_q2_0_f32")]]
-kernel void kernel_mul_mv_q2_0_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_q2_0_f32_impl<N_R0_Q2_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-kernel void kernel_mul_mv_q4_0_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- mul_vec_q_n_f32_impl<block_q4_0, N_R0_Q4_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-kernel void kernel_mul_mv_q4_1_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- mul_vec_q_n_f32_impl<block_q4_1, N_R0_Q4_1, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-kernel void kernel_mul_mv_q5_0_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- mul_vec_q_n_f32_impl<block_q5_0, N_R0_Q5_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-kernel void kernel_mul_mv_q5_1_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- mul_vec_q_n_f32_impl<block_q5_1, N_R0_Q5_1, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<short NR0, typename args_t>
-void kernel_mul_mv_q8_0_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- constexpr short NW = N_SIMDWIDTH;
- constexpr short NQ = 8;
-
- const int nb = args.ne00/QK8_0;
-
- const int r0 = tgpig.x*NR0;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- //device const block_q8_0 * x = (device const block_q8_0 *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- // pointers to src0 rows
- device const block_q8_0 * ax[NR0];
- FOR_UNROLL (short row = 0; row < NR0; ++row) {
- const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
-
- ax[row] = (device const block_q8_0 *) ((device char *) src0 + offset0);
- }
-
- float sumf[NR0] = { 0.f };
-
- const short ix = tiisg/(NW/NQ);
- const short il = tiisg%(NW/NQ);
-
- const int ib0 = sgitg*NQ + ix;
-
- float yl[NQ];
-
- device const float * yb = y + ib0*QK8_0 + il*NQ;
-
- // each thread in a SIMD group deals with NQ quants at a time
- for (int ib = ib0; ib < nb; ib += NSG*NQ) {
- for (short i = 0; i < NQ; ++i) {
- yl[i] = yb[i];
- }
-
- for (short row = 0; row < NR0; row++) {
- device const int8_t * qs = ax[row][ib].qs + il*NQ;
-
- float sumq = 0.f;
- FOR_UNROLL (short i = 0; i < NQ; ++i) {
- sumq += qs[i] * yl[i];
- }
-
- sumf[row] += sumq*ax[row][ib].d;
- }
-
- yb += NSG*NQ*QK8_0;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
-}
-
-[[host_name("kernel_mul_mv_q8_0_f32")]]
-kernel void kernel_mul_mv_q8_0_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_q8_0_f32_impl<N_R0_Q8_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-// mat-vec kernel processing in chunks of float4
-// chpb - chunks per quantization block
-template<short r1ptg, typename q_t, short chpb, void (*deq_t4)(device const q_t *, short, thread float4 &) >
-void kernel_mul_mv_ext_q4_f32_impl(
- constant ggml_metal_kargs_mul_mv_ext & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- const short NSG = FC_mul_mv_nsg;
- const short nxpsg = FC_mul_mv_nxpsg;
-
- const short chpt = 4; // chunks per thread
-
- //const short nxpsg = (32);
- const short nypsg = (32/nxpsg);
-
- const short tx = tiisg%nxpsg;
- const short ty = tiisg/nxpsg;
-
- const int i01 = tgpig.x*(nypsg*NSG) + nypsg*sgitg + ty;
- const int i11 = tgpig.y*r1ptg;
- const int i1m = tgpig.z;
-
- const int i12 = i1m%FC_mul_mv_ne12;
- const int i13 = i1m/FC_mul_mv_ne12;
-
- const uint64_t offset0 = i01*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = i11*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const q_t * xq = (i01 < args.ne01) ? (device const q_t *) (src0 + offset0) + tx/chpb : (device const q_t *) src0;
-
- device const float4 * y4[r1ptg];
-
- for (int ir1 = 0; ir1 < r1ptg; ++ir1) {
- y4[ir1] = (i11 + ir1 < args.ne11) ? (device const float4 *) (src1 + offset1 + ir1*args.nb11) + tx : (device const float4 *) src1;
- }
-
- float sumf[r1ptg] = { [ 0 ... r1ptg - 1 ] = 0.0f };
-
- short cch = tx%chpb; // current chunk index
-
- for (int ich = tx; 4*ich < args.ne00; ich += chpt*nxpsg) {
- float4 lx[chpt];
-
-#pragma unroll(chpt)
- for (short ch = 0; ch < chpt; ++ch) {
- deq_t4(xq, cch, lx[ch]);
-
- cch += nxpsg;
- if (cch >= chpb) {
- xq += cch/chpb;
- cch %= chpb;
- }
- }
-
-#pragma unroll(chpt)
- for (short ch = 0; ch < chpt; ++ch) {
-#pragma unroll(r1ptg)
- for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
- sumf[ir1] += dot(lx[ch], y4[ir1][ch*nxpsg]);
- }
- }
-
-#pragma unroll(r1ptg)
- for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
- y4[ir1] += chpt*nxpsg;
- }
- }
-
- // reduce only the threads in each row
- for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
- if (nxpsg >= 32) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 16);
- }
- if (nxpsg >= 16) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 8);
- }
- if (nxpsg >= 8) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 4);
- }
- if (nxpsg >= 4) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 2);
- }
- if (nxpsg >= 2) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 1);
- }
-
- //sumf[ir1] = simd_sum(sumf[ir1]);
- }
-
- if (tx == 0) {
- for (short ir1 = 0; ir1 < r1ptg && i11 + ir1 < args.ne11; ++ir1) {
- device float * dst_f32 = (device float *) dst + (uint64_t)i1m*args.ne0*args.ne1 + (uint64_t)(i11 + ir1)*args.ne0;
-
- if (i01 < args.ne01) {
- dst_f32[i01] = sumf[ir1];
- }
- }
- }
-}
-
-// mat-vec kernel processing in chunks of float4x4
-template<short r1ptg, typename q_t, short chpb, void (*deq_t4x4)(device const q_t *, short, thread float4x4 &) >
-void kernel_mul_mv_ext_q4x4_f32_impl(
- constant ggml_metal_kargs_mul_mv_ext & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- const short NSG = FC_mul_mv_nsg;
- const short nxpsg = FC_mul_mv_nxpsg;
-
- const short chpt = 1;
-
- //const short nxpsg = (32);
- const short nypsg = (32/nxpsg);
-
- const short tx = tiisg%nxpsg;
- const short ty = tiisg/nxpsg;
-
- const int i01 = tgpig.x*(nypsg*NSG) + nypsg*sgitg + ty;
- const int i11 = tgpig.y*r1ptg;
- const int i1m = tgpig.z;
-
- const int i12 = i1m%FC_mul_mv_ne12;
- const int i13 = i1m/FC_mul_mv_ne12;
-
- const uint64_t offset0 = i01*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = i11*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const q_t * xq = (i01 < args.ne01) ? (device const q_t *) (src0 + offset0) + tx/chpb : (device const q_t *) src0;
-
- device const float4x4 * y4x4[r1ptg];
-
- for (int ir1 = 0; ir1 < r1ptg; ++ir1) {
- y4x4[ir1] = (i11 + ir1 < args.ne11) ? (device const float4x4 *) (src1 + offset1 + ir1*args.nb11) + tx : (device const float4x4 *) src1;
- }
-
- float sumf[r1ptg] = { [ 0 ... r1ptg - 1 ] = 0.0f };
-
- short cch = tx%chpb;
-
- for (int ich = tx; 16*ich < args.ne00; ich += chpt*nxpsg) {
- float4x4 lx[chpt];
-
-#pragma unroll(chpt)
- for (short ch = 0; ch < chpt; ++ch) {
- deq_t4x4(xq, cch, lx[ch]);
-
- cch += nxpsg;
- if (cch >= chpb) {
- xq += cch/chpb;
- cch %= chpb;
- }
- }
-
-#pragma unroll(chpt)
- for (short ch = 0; ch < chpt; ++ch) {
-#pragma unroll(r1ptg)
- for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
- sumf[ir1] +=
- dot(lx[ch][0], y4x4[ir1][ch*nxpsg][0]) +
- dot(lx[ch][1], y4x4[ir1][ch*nxpsg][1]) +
- dot(lx[ch][2], y4x4[ir1][ch*nxpsg][2]) +
- dot(lx[ch][3], y4x4[ir1][ch*nxpsg][3]);
-
- }
- }
-
-#pragma unroll(r1ptg)
- for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
- y4x4[ir1] += chpt*nxpsg;
- }
- }
-
- for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
- if (nxpsg >= 32) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 16);
- }
- if (nxpsg >= 16) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 8);
- }
- if (nxpsg >= 8) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 4);
- }
- if (nxpsg >= 4) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 2);
- }
- if (nxpsg >= 2) {
- sumf[ir1] += simd_shuffle_down(sumf[ir1], 1);
- }
-
- //sumf[ir1] = simd_sum(sumf[ir1]);
- }
-
- if (tx == 0) {
- for (short ir1 = 0; ir1 < r1ptg && i11 + ir1 < args.ne11; ++ir1) {
- device float * dst_f32 = (device float *) dst + (uint64_t)i1m*args.ne0*args.ne1 + (uint64_t)(i11 + ir1)*args.ne0;
-
- if (i01 < args.ne01) {
- dst_f32[i01] = sumf[ir1];
- }
- }
- }
-}
-
-// dispatchers needed for compile-time nxpsg
-// epb - elements per quantization block
-template<short r1ptg, typename q_t, short epb, void (*deq_t4)(device const q_t *, short, thread float4 &)>
-kernel void kernel_mul_mv_ext_q4_f32_disp(
- constant ggml_metal_kargs_mul_mv_ext & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_ext_q4_f32_impl<r1ptg, q_t, epb/4, deq_t4>(args, src0, src1, dst, tgpig, tiisg, sgitg);
-}
-
-template<short r1ptg, typename q_t, short epb, void (*deq_t4x4)(device const q_t *, short, thread float4x4 &)>
-kernel void kernel_mul_mv_ext_q4x4_f32_disp(
- constant ggml_metal_kargs_mul_mv_ext & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_ext_q4x4_f32_impl<r1ptg, q_t, epb/16, deq_t4x4>(args, src0, src1, dst, tgpig, tiisg, sgitg);
-}
-
-typedef decltype(kernel_mul_mv_ext_q4_f32_disp <2, block_q8_0, 32, dequantize_q8_0_t4>) mul_mv_ext_q4_f32_t;
-typedef decltype(kernel_mul_mv_ext_q4x4_f32_disp<2, block_q4_K, 256, dequantize_q4_K>) mul_mv_ext_q4x4_f32_t;
-
-template [[host_name("kernel_mul_mv_ext_f32_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, float4, 4, dequantize_f32_t4>;
-template [[host_name("kernel_mul_mv_ext_f32_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, float4, 4, dequantize_f32_t4>;
-template [[host_name("kernel_mul_mv_ext_f32_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, float4, 4, dequantize_f32_t4>;
-template [[host_name("kernel_mul_mv_ext_f32_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, float4, 4, dequantize_f32_t4>;
-
-template [[host_name("kernel_mul_mv_ext_f16_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, half4, 4, dequantize_f16_t4>;
-template [[host_name("kernel_mul_mv_ext_f16_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, half4, 4, dequantize_f16_t4>;
-template [[host_name("kernel_mul_mv_ext_f16_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, half4, 4, dequantize_f16_t4>;
-template [[host_name("kernel_mul_mv_ext_f16_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, half4, 4, dequantize_f16_t4>;
-
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, bfloat4, 4, dequantize_bf16_t4>;
-template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, bfloat4, 4, dequantize_bf16_t4>;
-template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, bfloat4, 4, dequantize_bf16_t4>;
-template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, bfloat4, 4, dequantize_bf16_t4>;
-#endif
-
-template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q1_0, 128, dequantize_q1_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q1_0, 128, dequantize_q1_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q1_0, 128, dequantize_q1_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q1_0, 128, dequantize_q1_0_t4>;
-
-template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q2_0, 64, dequantize_q2_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q2_0, 64, dequantize_q2_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q2_0, 64, dequantize_q2_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q2_0, 64, dequantize_q2_0_t4>;
-
-template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q4_0, 32, dequantize_q4_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q4_0, 32, dequantize_q4_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q4_0, 32, dequantize_q4_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q4_0, 32, dequantize_q4_0_t4>;
-
-template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q4_1, 32, dequantize_q4_1_t4>;
-template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q4_1, 32, dequantize_q4_1_t4>;
-template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q4_1, 32, dequantize_q4_1_t4>;
-template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q4_1, 32, dequantize_q4_1_t4>;
-
-template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q5_0, 32, dequantize_q5_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q5_0, 32, dequantize_q5_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q5_0, 32, dequantize_q5_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q5_0, 32, dequantize_q5_0_t4>;
-
-template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q5_1, 32, dequantize_q5_1_t4>;
-template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q5_1, 32, dequantize_q5_1_t4>;
-template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q5_1, 32, dequantize_q5_1_t4>;
-template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q5_1, 32, dequantize_q5_1_t4>;
-
-template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q8_0, 32, dequantize_q8_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q8_0, 32, dequantize_q8_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q8_0, 32, dequantize_q8_0_t4>;
-template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q8_0, 32, dequantize_q8_0_t4>;
-
-template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_mxfp4, 32, dequantize_mxfp4_t4>;
-template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_mxfp4, 32, dequantize_mxfp4_t4>;
-template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_mxfp4, 32, dequantize_mxfp4_t4>;
-template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_mxfp4, 32, dequantize_mxfp4_t4>;
-
-template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
-template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
-template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
-template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
-
-template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q4_K, 256, dequantize_q4_K>;
-template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q4_K, 256, dequantize_q4_K>;
-template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q4_K, 256, dequantize_q4_K>;
-template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q4_K, 256, dequantize_q4_K>;
-
-template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q5_K, 256, dequantize_q5_K>;
-template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q5_K, 256, dequantize_q5_K>;
-template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q5_K, 256, dequantize_q5_K>;
-template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q5_K, 256, dequantize_q5_K>;
-
-template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q6_K, 256, dequantize_q6_K>;
-template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q6_K, 256, dequantize_q6_K>;
-template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q6_K, 256, dequantize_q6_K>;
-template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q6_K, 256, dequantize_q6_K>;
-
-template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q2_K, 256, dequantize_q2_K>;
-template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q2_K, 256, dequantize_q2_K>;
-template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q2_K, 256, dequantize_q2_K>;
-template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q2_K, 256, dequantize_q2_K>;
-
-template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q3_K, 256, dequantize_q3_K>;
-template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q3_K, 256, dequantize_q3_K>;
-template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q3_K, 256, dequantize_q3_K>;
-template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q3_K, 256, dequantize_q3_K>;
-
-template<typename T0, typename T1, short NR0, typename args_t>
-void kernel_mul_mv_t_t_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- constexpr short NW = N_SIMDWIDTH;
- constexpr short NB = 32;
- constexpr short NF = 8;
-
- const int nb = args.ne00/NB;
-
- const int r0 = tgpig.x*NR0;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- //device const T0 * x = (device const T0 *) (src0 + offset0);
- device const T1 * y = (device const T1 *) (src1 + offset1);
-
- // pointers to src0 rows
- device const T0 * ax [NR0];
- FOR_UNROLL (short row = 0; row < NR0; ++row) {
- const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
-
- ax[row] = (device const T0 *) ((device char *) src0 + offset0);
- }
-
- float sumf[NR0] = { 0.f };
-
- const short ix = tiisg/(NW/NF);
- const short il = tiisg%(NW/NF);
-
- const int ib0 = sgitg*NF + ix;
-
- T1 yl[NF];
-
- device const T1 * yb = y + (ib0*NB + il*NF);
-
- for (int ib = ib0; ib < nb; ib += NSG*NF) {
- for (short i = 0; i < NF; ++i) {
- yl[i] = yb[i];
- }
-
- for (short row = 0; row < NR0; row++) {
- device const T0 * xb = ax[row] + (ib*NB + il*NF);
-
- float sumq = 0.f;
- FOR_UNROLL (short i = 0; i < NF; ++i) {
- sumq += xb[i] * yl[i];
- }
-
- sumf[row] += sumq;
- }
-
- yb += NSG*NF*NW;
- }
-
- for (int i = nb*NB + sgitg*NW + tiisg; i < args.ne00; i += NW*NSG) {
- for (short row = 0; row < NR0; row++) {
- sumf[row] += ax[row][i] * y[i];
- }
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
-}
-
-template<typename T0, typename T1, typename args_t>
-void kernel_mul_mv_t_t_disp(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- switch (args.nr0) {
- //case 1: kernel_mul_mv_t_t_impl<T0, T1, 1, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- case 2: kernel_mul_mv_t_t_impl<T0, T1, 2, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- //case 3: kernel_mul_mv_t_t_impl<T0, T1, 3, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- //case 4: kernel_mul_mv_t_t_impl<T0, T1, 4, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- }
-}
-
-template<typename T0, typename T1>
-kernel void kernel_mul_mv_t_t(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_t_t_disp<T0, T1, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-typedef decltype(kernel_mul_mv_t_t<half, half>) mul_mv_t_t;
-
-template [[host_name("kernel_mul_mv_f32_f32")]] kernel mul_mv_t_t kernel_mul_mv_t_t<float, float>;
-template [[host_name("kernel_mul_mv_f16_f32")]] kernel mul_mv_t_t kernel_mul_mv_t_t<half, float>;
-template [[host_name("kernel_mul_mv_f16_f16")]] kernel mul_mv_t_t kernel_mul_mv_t_t<half, half>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mv_bf16_f32")]] kernel mul_mv_t_t kernel_mul_mv_t_t<bfloat, float>;
-template [[host_name("kernel_mul_mv_bf16_bf16")]] kernel mul_mv_t_t kernel_mul_mv_t_t<bfloat, bfloat>;
-#endif
-
-template<typename T0, typename T04, typename T1, typename T14, short NR0, typename args_t>
-void kernel_mul_mv_t_t_4_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- constexpr short NW = N_SIMDWIDTH;
- constexpr short NB = 32;
- constexpr short NF = 16;
- constexpr short NF4 = NF/4;
-
- const int nb = args.ne00/NB;
-
- const int r0 = tgpig.x*NR0;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const T1 * y = (device const T1 *) (src1 + offset1);
- device const T14 * y4 = (device const T14 *) (src1 + offset1);
-
- // pointers to src0 rows
- device const T0 * ax [NR0];
- device const T04 * ax4[NR0];
- FOR_UNROLL (short row = 0; row < NR0; ++row) {
- const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
-
- ax [row] = (device const T0 *) ((device char *) src0 + offset0);
- ax4[row] = (device const T04 *) ((device char *) src0 + offset0);
- }
-
- float sumf[NR0] = { 0.f };
-
- const short ix = tiisg/(NW/NF);
- const short il = tiisg%(NW/NF);
-
- const int ib0 = sgitg*NF + ix;
-
- T14 yl4[NF4];
-
- device const T14 * yb4 = y4 + (ib0*NB + il*NF)/4;
-
- for (int ib = ib0; ib < nb; ib += NSG*NF) {
- for (short i = 0; i < NF4; ++i) {
- yl4[i] = yb4[i];
- }
-
- for (short row = 0; row < NR0; row++) {
- device const T04 * xb4 = ax4[row] + (ib*NB + il*NF)/4;
-
- float sumq = 0.f;
- FOR_UNROLL (short i = 0; i < NF4; ++i) {
- sumq += dot(float4(xb4[i]), float4(yl4[i]));
- }
-
- sumf[row] += sumq;
- }
-
- yb4 += NSG*NF*NW/4;
- }
-
- for (int i = nb*NB + sgitg*NW + tiisg; i < args.ne00; i += NW*NSG) {
- for (short row = 0; row < NR0; row++) {
- sumf[row] += ax[row][i] * y[i];
- }
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
-}
-
-template<typename T0, typename T04, typename T1, typename T14, typename args_t>
-void kernel_mul_mv_t_t_4_disp(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- switch (args.nr0) {
- //case 1: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 1, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- case 2: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 2, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- //case 3: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 3, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- //case 4: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 4, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
- };
-}
-
-template<typename T0, typename T04, typename T1, typename T14>
-kernel void kernel_mul_mv_t_t_4(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_t_t_4_disp<T0, T04, T1, T14, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-typedef decltype(kernel_mul_mv_t_t_4<half, half4, half, half4>) mul_mv_t_t_4;
-
-template [[host_name("kernel_mul_mv_f32_f32_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<float, float4, float, float4>;
-template [[host_name("kernel_mul_mv_f16_f32_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<half, half4, float, float4>;
-template [[host_name("kernel_mul_mv_f16_f16_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<half, half4, half, half4>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mv_bf16_f32_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<bfloat, bfloat4, float, float4>;
-template [[host_name("kernel_mul_mv_bf16_bf16_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<bfloat, bfloat4, bfloat, bfloat4>;
-#endif
-
-template<typename T0, typename T1, typename args_t>
-void kernel_mul_mv_t_t_short_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig,
- ushort tiisg) {
- const int r0 = tgpig.x*32 + tiisg;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- if (r0 >= args.ne01) {
- return;
- }
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
-
- device const T0 * x = (device const T0 *) (src0 + offset0);
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1;
-
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const T1 * y = (device const T1 *) (src1 + offset1);
-
- float res = 0.0f;
-
- for (int i = 0; i < args.ne00; ++i) {
- res += (float) x[i] * (float) y[i];
- }
-
- dst_f32[(uint64_t)r1*args.ne0 + r0] = res;
-}
-
-template<typename T0, typename T1>
-kernel void kernel_mul_mv_t_t_short(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]]) {
- kernel_mul_mv_t_t_short_impl<T0, T1, constant ggml_metal_kargs_mul_mv &>(
- args,
- src0,
- src1,
- dst,
- tgpig,
- tiisg);
-}
-
-typedef decltype(kernel_mul_mv_t_t_short<half, half>) mul_mv_t_t_short_t;
-
-template [[host_name("kernel_mul_mv_f32_f32_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<float, float>;
-template [[host_name("kernel_mul_mv_f16_f32_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<half, float>;
-template [[host_name("kernel_mul_mv_f16_f16_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<half, half>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mv_bf16_f32_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<bfloat, float>;
-template [[host_name("kernel_mul_mv_bf16_bf16_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<bfloat, bfloat>;
-#endif
-
-constant bool FC_rope_is_imrope [[function_constant(FC_ROPE + 0)]];
-constant bool FC_rope_is_back [[function_constant(FC_ROPE + 1)]];
-
-static float rope_yarn_ramp(const float low, const float high, const int i0) {
- const float y = (i0 / 2 - low) / max(0.001f, high - low);
- return 1.0f - min(1.0f, max(0.0f, y));
-}
-
-// YaRN algorithm based on LlamaYaRNScaledRotaryEmbedding.py from https://github.com/jquesnelle/yarn
-// MIT licensed. Copyright (c) 2023 Jeffrey Quesnelle and Bowen Peng.
-static void rope_yarn(
- float theta_extrap, float freq_scale, float corr_dims[2], int i0, float ext_factor, float mscale,
- thread float * cos_theta, thread float * sin_theta) {
- // Get n-d rotational scaling corrected for extrapolation
- float theta_interp = freq_scale * theta_extrap;
- float theta = theta_interp;
- if (ext_factor != 0.0f) {
- float ramp_mix = rope_yarn_ramp(corr_dims[0], corr_dims[1], i0) * ext_factor;
- theta = theta_interp * (1 - ramp_mix) + theta_extrap * ramp_mix;
-
- // Get n-d magnitude scaling corrected for interpolation
- mscale *= 1.0f + 0.1f * log(1.0f / freq_scale);
- }
- *cos_theta = cos(theta) * mscale;
- *sin_theta = sin(theta) * mscale;
- if (FC_rope_is_back) {
- *sin_theta *= -1.0f;
- }
-}
-
-// Apparently solving `n_rot = 2pi * x * base^((2 * max_pos_emb) / n_dims)` for x, we get
-// `corr_fac(n_rot) = n_dims * log(max_pos_emb / (n_rot * 2pi)) / (2 * log(base))`
-static float rope_yarn_corr_factor(int n_dims, int n_ctx_orig, float n_rot, float base) {
- return n_dims * log(n_ctx_orig / (n_rot * 2 * M_PI_F)) / (2 * log(base));
-}
-
-static void rope_yarn_corr_dims(
- int n_dims, int n_ctx_orig, float freq_base, float beta_fast, float beta_slow, float dims[2]
-) {
- // start and end correction dims
- dims[0] = max(0.0f, floor(rope_yarn_corr_factor(n_dims, n_ctx_orig, beta_fast, freq_base)));
- dims[1] = min(n_dims - 1.0f, ceil(rope_yarn_corr_factor(n_dims, n_ctx_orig, beta_slow, freq_base)));
-}
-
-template<typename T>
-kernel void kernel_rope_norm(
- constant ggml_metal_kargs_rope & args,
- device const char * src0,
- device const char * src1,
- device const char * src2,
- device char * dst,
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort3 tptg [[threads_per_threadgroup]],
- uint3 tgpig[[threadgroup_position_in_grid]]) {
- const int i3 = tgpig[2];
- const int i2 = tgpig[1];
- const int i1 = tgpig[0];
-
- float corr_dims[2];
- rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
-
- device const int32_t * pos = (device const int32_t *) src1;
-
- const float theta_base = (float) pos[i2];
- const float inv_ndims = -1.f/args.n_dims;
-
- float cos_theta;
- float sin_theta;
-
- for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
- if (i0 >= args.n_offs && i0 < args.n_offs + args.n_dims) {
- const int iw = i0 - args.n_offs; // relative idx
- const int ic = iw/2;
-
- const float theta = theta_base * pow(args.freq_base, inv_ndims*iw);
-
- const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
-
- rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, iw, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
-
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- const float x0 = src[0];
- const float x1 = src[1];
-
- dst_data[0] = x0*cos_theta - x1*sin_theta;
- dst_data[1] = x0*sin_theta + x1*cos_theta;
- } else {
- if (args.inplace) {
- continue;
- }
-
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- dst_data[0] = src[0];
- dst_data[1] = src[1];
- }
- }
-}
-
-template<typename T>
-kernel void kernel_rope_neox(
- constant ggml_metal_kargs_rope & args,
- device const char * src0,
- device const char * src1,
- device const char * src2,
- device char * dst,
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort3 tptg [[threads_per_threadgroup]],
- uint3 tgpig[[threadgroup_position_in_grid]]) {
- const int i3 = tgpig[2];
- const int i2 = tgpig[1];
- const int i1 = tgpig[0];
-
- float corr_dims[2];
- rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
-
- device const int32_t * pos = (device const int32_t *) src1;
-
- const float theta_base = (float) pos[i2];
- const float inv_ndims = -1.f/args.n_dims;
-
- float cos_theta;
- float sin_theta;
-
- for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
- if (i0 >= args.n_offs && i0 < args.n_offs + args.n_dims) {
- const int iw = i0 - args.n_offs; // relative idx
- const int ic = iw/2;
-
- const float theta = theta_base * pow(args.freq_base, inv_ndims*iw);
-
- const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
-
- rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, iw, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
-
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + (args.n_offs + ic)*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + (args.n_offs + ic)*args.nb0);
-
- const float x0 = src[0];
- const float x1 = src[args.n_dims/2];
-
- dst_data[0] = x0*cos_theta - x1*sin_theta;
- dst_data[args.n_dims/2] = x0*sin_theta + x1*cos_theta;
- } else {
- if (args.inplace) {
- continue;
- }
-
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- dst_data[0] = src[0];
- dst_data[1] = src[1];
- }
- }
-}
-
-template<typename T>
-kernel void kernel_rope_multi(
- constant ggml_metal_kargs_rope & args,
- device const char * src0,
- device const char * src1,
- device const char * src2,
- device char * dst,
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort3 tptg [[threads_per_threadgroup]],
- uint3 tgpig[[threadgroup_position_in_grid]]) {
- const int i3 = tgpig[2];
- const int i2 = tgpig[1];
- const int i1 = tgpig[0];
-
- float corr_dims[2];
- rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
-
- device const int32_t * pos = (device const int32_t *) src1;
-
- const float inv_ndims = -1.f/args.n_dims;
-
- float cos_theta;
- float sin_theta;
-
- for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
- if (i0 >= args.n_offs && i0 < args.n_offs + args.n_dims) {
- const int iw = i0 - args.n_offs; // relative idx
- const int ic = iw/2;
-
- // mrope theta calculations
- // note: the rest is the same as kernel_rope_neox
- const int sect_dims = args.sect_0 + args.sect_1 + args.sect_2 + args.sect_3;
- const int sec_w01 = args.sect_0 + args.sect_1; // end of section 1
- const int sec_w012 = args.sect_0 + args.sect_1 + args.sect_2; // end of section 2
- const int sector = ic % sect_dims;
-
- float theta_base;
- if (FC_rope_is_imrope) {
- if (sector % 3 == 1 && sector < 3 * args.sect_1) { // h
- theta_base = (float) pos[i2 + args.ne02 * 1];
- } else if (sector % 3 == 2 && sector < 3 * args.sect_2) { // w
- theta_base = (float) pos[i2 + args.ne02 * 2];
- } else if (sector % 3 == 0 && sector < 3 * args.sect_0) { // t
- theta_base = (float) pos[i2 + args.ne02 * 0];
- } else { // e
- theta_base = (float) pos[i2 + args.ne02 * 3];
- }
- } else {
- if (sector < args.sect_0) {
- theta_base = (float) pos[i2];
- } else if (sector < sec_w01) {
- theta_base = (float) pos[i2 + args.ne02 * 1];
- } else if (sector < sec_w012) {
- theta_base = (float) pos[i2 + args.ne02 * 2];
- } else {
- theta_base = (float) pos[i2 + args.ne02 * 3];
- }
- }
- // end of mrope
-
- const float theta = theta_base * pow(args.freq_base, inv_ndims*iw);
-
- const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
-
- rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, iw, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
-
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + (args.n_offs + ic)*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + (args.n_offs + ic)*args.nb0);
-
- const float x0 = src[0];
- const float x1 = src[args.n_dims/2];
-
- dst_data[0] = x0*cos_theta - x1*sin_theta;
- dst_data[args.n_dims/2] = x0*sin_theta + x1*cos_theta;
- } else {
- if (args.inplace) {
- continue;
- }
-
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- dst_data[0] = src[0];
- dst_data[1] = src[1];
- }
- }
-}
-
-template<typename T>
-kernel void kernel_rope_vision(
- constant ggml_metal_kargs_rope & args,
- device const char * src0,
- device const char * src1,
- device const char * src2,
- device char * dst,
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort3 tptg [[threads_per_threadgroup]],
- uint3 tgpig[[threadgroup_position_in_grid]]) {
- const int i3 = tgpig[2];
- const int i2 = tgpig[1];
- const int i1 = tgpig[0];
-
- float corr_dims[2];
- rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
-
- device const int32_t * pos = (device const int32_t *) src1;
-
- const float inv_ndims = -1.f/args.n_dims;
-
- float cos_theta;
- float sin_theta;
-
- for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
- if (i0 < 2*args.n_dims) { // different from kernel_rope_multi
- const int ic = i0/2;
-
- // mrope theta calculations (only support 2 dimensions)
- const int sect_dims = args.sect_0 + args.sect_1;
- const int sector = ic % sect_dims;
-
- float p;
- float theta_base;
- if (sector < args.sect_1) {
- p = (float) sector;
- theta_base = (float) pos[i2];
- } else {
- p = (float) sector - args.sect_0;
- theta_base = (float) pos[i2 + args.ne02];
- }
-
- const float theta = theta_base * pow(args.freq_base, 2.0f * inv_ndims * p);
- // end of mrope
-
- const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
-
- rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, i0, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
-
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + ic*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + ic*args.nb0);
-
- const float x0 = src[0];
- const float x1 = src[args.n_dims]; // different from kernel_rope_multi
-
- dst_data[0] = x0*cos_theta - x1*sin_theta;
- dst_data[args.n_dims] = x0*sin_theta + x1*cos_theta; // different from kernel_rope_multi
- } else {
- device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
- device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- dst_data[0] = src[0];
- dst_data[1] = src[1];
- }
- }
-}
-
-typedef decltype(kernel_rope_norm<float>) kernel_rope_norm_t;
-typedef decltype(kernel_rope_neox<float>) kernel_rope_neox_t;
-typedef decltype(kernel_rope_multi<float>) kernel_rope_multi_t;
-typedef decltype(kernel_rope_vision<float>) kernel_rope_vision_t;
-
-template [[host_name("kernel_rope_norm_f32")]] kernel kernel_rope_norm_t kernel_rope_norm<float>;
-template [[host_name("kernel_rope_norm_f16")]] kernel kernel_rope_norm_t kernel_rope_norm<half>;
-
-template [[host_name("kernel_rope_neox_f32")]] kernel kernel_rope_neox_t kernel_rope_neox<float>;
-template [[host_name("kernel_rope_neox_f16")]] kernel kernel_rope_neox_t kernel_rope_neox<half>;
-
-template [[host_name("kernel_rope_multi_f32")]] kernel kernel_rope_multi_t kernel_rope_multi<float>;
-template [[host_name("kernel_rope_multi_f16")]] kernel kernel_rope_multi_t kernel_rope_multi<half>;
-
-template [[host_name("kernel_rope_vision_f32")]] kernel kernel_rope_vision_t kernel_rope_vision<float>;
-template [[host_name("kernel_rope_vision_f16")]] kernel kernel_rope_vision_t kernel_rope_vision<half>;
-
-typedef void (im2col_t)(
- constant ggml_metal_kargs_im2col & args,
- device const float * x,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-template <typename T>
-kernel void kernel_im2col(
- constant ggml_metal_kargs_im2col & args,
- device const float * x,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-// const int64_t IC = tgpg[0];
- const int64_t OH = tgpg[1];
- const int64_t OW = tgpg[2];
-
- const int64_t KH = ntg[1];
- const int64_t KW = ntg[2];
-
- int64_t in = tpitg[0];
- const int64_t ikh = tpitg[1];
- const int64_t ikw = tpitg[2];
-
- const int64_t iic = tgpig[0];
- const int64_t ioh = tgpig[1];
- const int64_t iow = tgpig[2];
-
- const int64_t iiw = iow*args.s0 + ikw*args.d0 - args.p0;
- const int64_t iih = ioh*args.s1 + ikh*args.d1 - args.p1;
-
- int64_t offset_dst = (in*OH*OW + ioh*OW + iow)*args.CHW + (iic*(KH*KW) + ikh*KW + ikw);
-
- device T * pdst = (device T *) (dst);
-
- if (iih < 0 || iih >= args.IH || iiw < 0 || iiw >= args.IW) {
- while (in < args.N) {
- pdst[offset_dst] = 0.0f;
- offset_dst += ntg[0]*args.CHW*OH*OW;
-
- in += ntg[0];
- }
- } else {
- int64_t offset_src = in*args.ofs0 + iic*args.ofs1 + iih*args.IW + iiw;
-
- while (in < args.N) {
- pdst[offset_dst] = x[offset_src];
-
- offset_dst += ntg[0]*args.CHW*OH*OW;
- offset_src += ntg[0]*args.ofs0;
-
- in += ntg[0];
- }
- }
-}
-
-template [[host_name("kernel_im2col_f32")]] kernel im2col_t kernel_im2col<float>;
-template [[host_name("kernel_im2col_f16")]] kernel im2col_t kernel_im2col<half>;
-
-// TODO: optimize
-typedef void (im2col_ext_t)(
- constant ggml_metal_kargs_im2col & args,
- device const float * x,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-template <typename T>
-kernel void kernel_im2col_ext(
- constant ggml_metal_kargs_im2col & args,
- device const float * x,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]], // tgpg[0] = D x IC x KH x KW, CHW = IC x KH x KW
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) { // [M, 1, 1]
- const int64_t KHW = (int64_t)args.KHW;
-
- const int64_t d = tgpig[0] / args.CHW;
- const int64_t chw = tgpig[0] % args.CHW;
- const int64_t tgpig_0 = chw / KHW; // 0 ~ (IC - 1)
- const int64_t HW = tgpig[0] % KHW;
-
- const int64_t tpitg_0 = (d * ntg[0]) + tpitg[0];
- if (tpitg_0 >= args.N) {
- return;
- }
-
- const int64_t tpitg_1 = HW / args.KW;
- const int64_t tpitg_2 = HW % args.KW;
-
- const int64_t iiw = tgpig[2] * args.s0 + tpitg_2 * args.d0 - args.p0;
- const int64_t iih = tgpig[1] * args.s1 + tpitg_1 * args.d1 - args.p1;
-
- const int64_t offset_dst =
- (tpitg_0 * tgpg[1] * tgpg[2] + tgpig[1] * tgpg[2] + tgpig[2]) * args.CHW +
- (tgpig_0 * KHW + tpitg_1 * args.KW + tpitg_2);
-
- device T * pdst = (device T *) (dst);
-
- if (iih < 0 || iih >= args.IH || iiw < 0 || iiw >= args.IW) {
- pdst[offset_dst] = 0.0f;
- } else {
- const int64_t offset_src = tpitg_0 * args.ofs0 + tgpig_0 * args.ofs1;
- pdst[offset_dst] = x[offset_src + iih * args.IW + iiw];
- }
-}
-
-template [[host_name("kernel_im2col_ext_f32")]] kernel im2col_ext_t kernel_im2col_ext<float>;
-template [[host_name("kernel_im2col_ext_f16")]] kernel im2col_ext_t kernel_im2col_ext<half>;
-
-template <typename TK>
-kernel void kernel_conv_2d(
- constant ggml_metal_kargs_conv_2d & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const uint threads_per_tg = ntg.x * ntg.y * ntg.z;
- const uint tg_index = (tgpig.z * tgpg.y + tgpig.y) * tgpg.x + tgpig.x;
- const uint local_thread = tpitg.z * (ntg.x * ntg.y) + tpitg.y * ntg.x + tpitg.x;
- const uint thread_index = tg_index * threads_per_tg + local_thread;
- const uint64_t total_threads = (uint64_t) threads_per_tg * tgpg.x * tgpg.y * tgpg.z;
- const uint64_t total_outputs = (uint64_t) args.N * args.OC * args.OH * args.OW;
-
- for (uint64_t index = thread_index; index < total_outputs; index += total_threads) {
- uint64_t tmp = index;
-
- const int32_t ow = tmp % args.OW; tmp /= args.OW;
- const int32_t oh = tmp % args.OH; tmp /= args.OH;
- const int32_t oc = tmp % args.OC; tmp /= args.OC;
- const int32_t n = tmp;
-
- float acc = 0.0f;
-
- const int32_t base_x = ow*args.s0 - args.p0;
- const int32_t base_y = oh*args.s1 - args.p1;
-
- int32_t ky_start = 0;
- if (base_y < 0) {
- ky_start = (-base_y + args.d1 - 1)/args.d1;
- }
- int32_t ky_end = args.KH;
- const int32_t y_max = args.IH - 1 - base_y;
- if (y_max < 0) {
- ky_end = ky_start;
- } else if (base_y + (args.KH - 1)*args.d1 >= args.IH) {
- ky_end = min(ky_end, y_max/args.d1 + 1);
- }
-
- int32_t kx_start = 0;
- if (base_x < 0) {
- kx_start = (-base_x + args.d0 - 1)/args.d0;
- }
- int32_t kx_end = args.KW;
- const int32_t x_max = args.IW - 1 - base_x;
- if (x_max < 0) {
- kx_end = kx_start;
- } else if (base_x + (args.KW - 1)*args.d0 >= args.IW) {
- kx_end = min(kx_end, x_max/args.d0 + 1);
- }
-
- if (ky_start < ky_end && kx_start < kx_end) {
- const uint64_t src_base_n = (uint64_t) n * args.nb13;
- const uint64_t w_base_oc = (uint64_t) oc * args.nb03;
-
- for (int32_t ic = 0; ic < args.IC; ++ic) {
- const uint64_t src_base_nc = src_base_n + (uint64_t) ic * args.nb12;
- const uint64_t w_base_ocic = w_base_oc + (uint64_t) ic * args.nb02;
-
- for (int32_t ky = ky_start; ky < ky_end; ++ky) {
- const int32_t iy = base_y + ky*args.d1;
- const uint64_t src_base_row = src_base_nc + (uint64_t) iy * args.nb11;
- const uint64_t w_base_row = w_base_ocic + (uint64_t) ky * args.nb01;
-
- for (int32_t kx = kx_start; kx < kx_end; ++kx) {
- const int32_t ix = base_x + kx*args.d0;
- const uint64_t src_offs = src_base_row + (uint64_t) ix * args.nb10;
- const uint64_t w_offs = w_base_row + (uint64_t) kx * args.nb00;
-
- const float x = *(device const float *)(src + src_offs);
- const float w = (float) (*(device const TK *)(weights + w_offs));
-
- acc += x * w;
- }
- }
- }
- }
-
- const uint64_t dst_offs =
- (uint64_t) n * args.nb3 +
- (uint64_t) oc * args.nb2 +
- (uint64_t) oh * args.nb1 +
- (uint64_t) ow * args.nb0;
-
- *(device float *)(dst + dst_offs) = acc;
- }
-}
-
-template [[host_name("kernel_conv_2d_f32_f32")]]
-kernel void kernel_conv_2d<float>(
- constant ggml_metal_kargs_conv_2d & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-template [[host_name("kernel_conv_2d_f16_f32")]]
-kernel void kernel_conv_2d<half>(
- constant ggml_metal_kargs_conv_2d & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-// grid: x = C tile, y = OH, z = OW * N (for channel-contiguous layouts)
-template <typename TK>
-kernel void kernel_conv_2d_dw_tiled(
- constant ggml_metal_kargs_conv_2d_dw & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int32_t c = (int32_t)(tgpig.x * ntg.x + tpitg.x);
- if (c >= args.C) {
- return;
- }
-
- const int32_t oh = tgpig.y;
- const int32_t own = tgpig.z;
- const int32_t ow = own % args.OW;
- const int32_t n = own / args.OW;
-
- const int32_t base_y = oh*args.s1 - args.p1;
-
- int32_t ky_start = 0;
- if (base_y < 0) {
- ky_start = (-base_y + args.d1 - 1)/args.d1;
- }
- int32_t ky_end = args.KH;
- const int32_t y_max = args.IH - 1 - base_y;
- if (y_max < 0) {
- ky_end = ky_start;
- } else if (base_y + (args.KH - 1)*args.d1 >= args.IH) {
- ky_end = min(ky_end, y_max/args.d1 + 1);
- }
-
- const int32_t base_x = ow*args.s0 - args.p0;
-
- int32_t kx_start = 0;
- if (base_x < 0) {
- kx_start = (-base_x + args.d0 - 1)/args.d0;
- }
- int32_t kx_end = args.KW;
- const int32_t x_max = args.IW - 1 - base_x;
- if (x_max < 0) {
- kx_end = kx_start;
- } else if (base_x + (args.KW - 1)*args.d0 >= args.IW) {
- kx_end = min(kx_end, x_max/args.d0 + 1);
- }
-
- float acc = 0.0f;
-
- if (ky_start < ky_end && kx_start < kx_end) {
- const uint64_t w_base = (uint64_t) c * args.nb02;
- const uint64_t src_base = (uint64_t) n * args.nb13 + (uint64_t) c * args.nb12;
-
- for (int32_t ky = ky_start; ky < ky_end; ++ky) {
- const int32_t iy = base_y + ky*args.d1;
- const uint64_t src_row = src_base + (uint64_t) iy * args.nb11;
- const uint64_t w_row = w_base + (uint64_t) ky * args.nb01;
-
- for (int32_t kx = kx_start; kx < kx_end; ++kx) {
- const int32_t ix = base_x + kx*args.d0;
- const float x = *(device const float *)(src + src_row + (uint64_t) ix * args.nb10);
- const float w = (float)(*(device const TK *)(weights + w_row + (uint64_t) kx * args.nb00));
- acc += x * w;
- }
- }
- }
-
- const uint64_t dst_offs =
- (uint64_t) n * args.nb3 +
- (uint64_t) c * args.nb2 +
- (uint64_t) oh * args.nb1 +
- (uint64_t) ow * args.nb0;
-
- *(device float *)(dst + dst_offs) = acc;
-}
-
-// grid: x = OW tile, y = OH, z = C * N (for spatially-contiguous layouts)
-template <typename TK>
-kernel void kernel_conv_2d_dw(
- constant ggml_metal_kargs_conv_2d_dw & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int32_t oh = tgpig.y;
- const int32_t cn = tgpig.z;
- const int32_t c = cn % args.C;
- const int32_t n = cn / args.C;
-
- const int32_t base_y = oh*args.s1 - args.p1;
-
- int32_t ky_start = 0;
- if (base_y < 0) {
- ky_start = (-base_y + args.d1 - 1)/args.d1;
- }
- int32_t ky_end = args.KH;
- const int32_t y_max = args.IH - 1 - base_y;
- if (y_max < 0) {
- ky_end = ky_start;
- } else if (base_y + (args.KH - 1)*args.d1 >= args.IH) {
- ky_end = min(ky_end, y_max/args.d1 + 1);
- }
-
- const uint64_t w_base = (uint64_t) c * args.nb02;
- const uint64_t src_base = (uint64_t) n * args.nb13 + (uint64_t) c * args.nb12;
-
- const int32_t ow = (int32_t)(tgpig.x * ntg.x + tpitg.x);
- if (ow >= args.OW) {
- return;
- }
-
- float acc = 0.0f;
-
- const int32_t base_x = ow*args.s0 - args.p0;
-
- int32_t kx_start = 0;
- if (base_x < 0) {
- kx_start = (-base_x + args.d0 - 1)/args.d0;
- }
- int32_t kx_end = args.KW;
- const int32_t x_max = args.IW - 1 - base_x;
- if (x_max < 0) {
- kx_end = kx_start;
- } else if (base_x + (args.KW - 1)*args.d0 >= args.IW) {
- kx_end = min(kx_end, x_max/args.d0 + 1);
- }
-
- if (ky_start < ky_end && kx_start < kx_end) {
- for (int32_t ky = ky_start; ky < ky_end; ++ky) {
- const int32_t iy = base_y + ky*args.d1;
- const uint64_t src_row = src_base + (uint64_t) iy * args.nb11;
- const uint64_t w_row = w_base + (uint64_t) ky * args.nb01;
-
- for (int32_t kx = kx_start; kx < kx_end; ++kx) {
- const int32_t ix = base_x + kx*args.d0;
- const float x = *(device const float *)(src + src_row + (uint64_t) ix * args.nb10);
- const float w = (float)(*(device const TK *)(weights + w_row + (uint64_t) kx * args.nb00));
- acc += x * w;
- }
- }
- }
-
- const uint64_t dst_offs =
- (uint64_t) n * args.nb3 +
- (uint64_t) c * args.nb2 +
- (uint64_t) oh * args.nb1 +
- (uint64_t) ow * args.nb0;
-
- *(device float *)(dst + dst_offs) = acc;
-}
-
-template [[host_name("kernel_conv_2d_dw_f32_f32")]]
-kernel void kernel_conv_2d_dw<float>(
- constant ggml_metal_kargs_conv_2d_dw & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-template [[host_name("kernel_conv_2d_dw_f16_f32")]]
-kernel void kernel_conv_2d_dw<half>(
- constant ggml_metal_kargs_conv_2d_dw & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-template [[host_name("kernel_conv_2d_dw_tiled_f32_f32")]]
-kernel void kernel_conv_2d_dw_tiled<float>(
- constant ggml_metal_kargs_conv_2d_dw & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-template [[host_name("kernel_conv_2d_dw_tiled_f16_f32")]]
-kernel void kernel_conv_2d_dw_tiled<half>(
- constant ggml_metal_kargs_conv_2d_dw & args,
- device const char * weights,
- device const char * src,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-typedef void (conv_transpose_1d_t)(
- constant ggml_metal_kargs_conv_transpose_1d & args,
- device const float * src0,
- device const float * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]]);
-
-template <typename T>
-kernel void kernel_conv_transpose_1d(
- constant ggml_metal_kargs_conv_transpose_1d & args,
- device const T * src0,
- device const float * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]]) {
-
- // For output position j on the time axis, only input positions
- // i such that i*s0 <= j < i*s0 + K
- // contribute -- i.e. i in [ceil((j - K + 1)/s0), floor(j/s0)]
- // intersected with [0, IL-1]. That's at most ceil(K/s0) values
- // (typically 2 for stride==K/2 transposed convs).
- const int32_t j = tgpig[0];
- const int32_t s0 = args.s0;
- const int32_t K = args.K;
- const int32_t IL = args.IL;
-
- int32_t i_min;
- {
- int32_t a = j - K + 1;
- i_min = a <= 0 ? 0 : (a + s0 - 1) / s0; // ceil(a/s0) for a>0
- }
- int32_t i_max = j / s0;
- if (i_max > IL - 1) i_max = IL - 1;
-
- float v = 0.0f;
- if (i_min <= i_max) {
- for (int64_t c = 0; c < args.IC; c++) {
- const int32_t kernel_offset = c * tgpg[1] * K + K * tgpig[1];
- const int32_t input_offset = c * IL;
-
- for (int32_t i = i_min; i <= i_max; i++) {
- v += float(src0[kernel_offset + j - i * s0]) * src1[input_offset + i];
- }
- }
- }
-
- device float * dst_ptr = (device float *) (dst + tgpig[0] * args.nb0 + tgpig[1] * args.nb1);
-
- dst_ptr[0] = v;
-}
-
-template [[host_name("kernel_conv_transpose_1d_f32_f32")]]
-kernel void kernel_conv_transpose_1d<float>(
- constant ggml_metal_kargs_conv_transpose_1d & args,
- device const float * src0,
- device const float * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]]);
-
-template [[host_name("kernel_conv_transpose_1d_f16_f32")]]
-kernel void kernel_conv_transpose_1d<half>(
- constant ggml_metal_kargs_conv_transpose_1d & args,
- device const half * src0,
- device const float * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]]);
-
-
-template <typename T>
-kernel void kernel_col2im_1d(
- constant ggml_metal_kargs_col2im_1d & args,
- device const T * col,
- device T * dst,
- uint tgpig [[threadgroup_position_in_grid]],
- uint tpitg [[thread_position_in_threadgroup]],
- uint ntg [[threads_per_threadgroup]]) {
-
- const int idx = tgpig * ntg + tpitg;
- if (idx >= args.T_out * args.OC) {
- return;
- }
-
- const int t_out = idx % args.T_out;
- const int oc = idx / args.T_out;
- const int t_abs = t_out + args.p0; // absolute position in uncropped signal
-
- int t_in_min = (t_abs - args.K + args.s0) / args.s0; // ceil((t_abs - K + 1) / s0)
- if (t_in_min < 0) {
- t_in_min = 0;
- }
- int t_in_max = t_abs / args.s0;
- if (t_in_max >= args.T_in) {
- t_in_max = args.T_in - 1;
- }
-
- float sum = 0.0f;
- for (int t_in = t_in_min; t_in <= t_in_max; t_in++) {
- const int k = t_abs - t_in * args.s0;
- sum += float(col[(oc * args.K + k) + t_in * args.K_OC]);
- }
-
- dst[t_out + oc * args.T_out] = T(sum);
-}
-
-template [[host_name("kernel_col2im_1d_f32")]] kernel void kernel_col2im_1d<float>(constant ggml_metal_kargs_col2im_1d &, device const float *, device float *, uint, uint, uint);
-template [[host_name("kernel_col2im_1d_f16")]] kernel void kernel_col2im_1d<half>(constant ggml_metal_kargs_col2im_1d &, device const half *, device half *, uint, uint, uint);
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_col2im_1d_bf16")]] kernel void kernel_col2im_1d<bfloat>(constant ggml_metal_kargs_col2im_1d &, device const bfloat *, device bfloat *, uint, uint, uint);
-#endif
-
-
-template <typename T>
-kernel void kernel_snake(
- constant ggml_metal_kargs_snake & args,
- device const T * x,
- device const float * a,
- device const float * inv_b,
- device T * dst,
- uint tgpig [[threadgroup_position_in_grid]],
- uint tpitg [[thread_position_in_threadgroup]],
- uint ntg [[threads_per_threadgroup]]) {
-
- const int idx = tgpig * ntg + tpitg;
- if (idx >= args.T * args.C) {
- return;
- }
-
- const int c = idx / args.T; // x is [T, C], a / inv_b collapse to [1, C]
- const float xi = float(x[idx]);
- const float si = sin(a[c] * xi);
- dst[idx] = T(xi + si * si * inv_b[c]);
-}
-
-template [[host_name("kernel_snake_f32")]] kernel void kernel_snake<float>(constant ggml_metal_kargs_snake &, device const float *, device const float *, device const float *, device float *, uint, uint, uint);
-template [[host_name("kernel_snake_f16")]] kernel void kernel_snake<half>(constant ggml_metal_kargs_snake &, device const half *, device const float *, device const float *, device half *, uint, uint, uint);
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_snake_bf16")]] kernel void kernel_snake<bfloat>(constant ggml_metal_kargs_snake &, device const bfloat *, device const float *, device const float *, device bfloat *, uint, uint, uint);
-#endif
-
-
-typedef void (conv_transpose_2d_t)(
- constant ggml_metal_kargs_conv_transpose_2d & args,
- device const float * src0,
- device const float * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]]);
-
-template <typename T>
-kernel void kernel_conv_transpose_2d(
- constant ggml_metal_kargs_conv_transpose_2d & args,
- device const T * src0,
- device const float * src1,
- device char * dst,
- threadgroup float * shared_sum [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int64_t out_x = tgpig[0];
- const int64_t out_y = tgpig[1];
- const int64_t out_c = tgpig[2];
-
- const int64_t kw = tpitg[0];
- const int64_t kh = tpitg[1];
-
- float v = 0.0f;
-
- for (int64_t in_c = 0; in_c < args.IC; in_c++) {
- int64_t in_y = out_y - kh;
-
- if (in_y < 0 || in_y % args.s0) continue;
-
- in_y /= args.s0;
-
- if (in_y >= args.IH) continue;
-
- int64_t in_x = out_x - kw;
-
- if (in_x < 0 || in_x % args.s0) continue;
-
- in_x /= args.s0;
-
- if (in_x >= args.IW) continue;
-
- const int64_t input_idx = (args.IW * args.IH) * in_c + (args.IW) * in_y + in_x;
- const int64_t kernel_idx = (args.KH * args.KW * args.OC) * in_c + (args.KH * args.KW) * out_c + (args.KW) * kh + kw;
-
- v += (float)src0[kernel_idx] * src1[input_idx];
- }
-
- const uint tid = tpitg.y * ntg.x + tpitg.x;
- shared_sum[tid] = v;
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (tid == 0) {
- float total = 0.0f;
- const uint num_threads = ntg.x * ntg.y;
- for (uint i = 0; i < num_threads; i++) {
- total += shared_sum[i];
- }
-
- device float * dst_ptr = (device float *) (dst + out_x*args.nb0 + out_y * args.nb1 + out_c*args.nb2);
- dst_ptr[0] = total;
- }
-}
-
-template [[host_name("kernel_conv_transpose_2d_f32_f32")]]
-kernel void kernel_conv_transpose_2d<float>(
- constant ggml_metal_kargs_conv_transpose_2d & args,
- device const float * src0,
- device const float * src1,
- device char * dst,
- threadgroup float * shared_sum [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-template [[host_name("kernel_conv_transpose_2d_f16_f32")]]
-kernel void kernel_conv_transpose_2d<half>(
- constant ggml_metal_kargs_conv_transpose_2d & args,
- device const half * src0,
- device const float * src1,
- device char * dst,
- threadgroup float * shared_sum [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]);
-
-constant bool FC_upscale_aa [[function_constant(FC_UPSCALE + 0)]];
-
-kernel void kernel_upscale_nearest_f32(
- constant ggml_metal_kargs_upscale & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int64_t i3 = tgpig.z;
- const int64_t i2 = tgpig.y;
- const int64_t i1 = tgpig.x;
-
- const int64_t i03 = i3/args.sf3;
- const int64_t i02 = i2/args.sf2;
- const int64_t i01 = i1/args.sf1;
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- const int64_t i00 = i0/args.sf0;
-
- device const float * src0_ptr = (device const float *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + i00*args.nb00);
- device float * dst_ptr = (device float *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- dst_ptr[0] = src0_ptr[0];
- }
-}
-
-static inline float bilinear_tri(float x) {
- return MAX(0.0f, 1.0f - fabs(x));
-}
-
-kernel void kernel_upscale_bilinear_f32(
- constant ggml_metal_kargs_upscale & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int64_t i3 = tgpig.z;
- const int64_t i2 = tgpig.y;
- const int64_t i1 = tgpig.x;
-
- const int64_t i03 = i3 / args.sf3;
- const int64_t i02 = i2 / args.sf2;
-
- const float f01 = ((float)i1 + args.poffs) / args.sf1 - args.poffs;
- const int64_t i01 = MAX(0, MIN(args.ne01 - 1, (int64_t)floor(f01)));
- const int64_t i01p = MAX(0, MIN(args.ne01 - 1, i01 + 1));
- const float fd1 = MAX(0.0f, MIN(1.0f, f01 - (float)i01));
-
- src0 += i03*args.nb03 + i02*args.nb02;
-
- device float * dst_ptr = (device float *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1);
-
- if (FC_upscale_aa) {
- const float support0 = MAX(1.0f, 1.0f / args.sf0);
- const float invscale0 = 1.0f / support0;
- const float support1 = MAX(1.0f, 1.0f / args.sf1);
- const float invscale1 = 1.0f / support1;
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- const float f00 = ((float)i0 + args.poffs) / args.sf0 - args.poffs;
-
- int64_t x_min = MAX((int64_t)0, (int64_t)floor(f00 - support0 + args.poffs));
- int64_t x_max = MIN(args.ne00, (int64_t)ceil (f00 + support0 + args.poffs));
-
- int64_t y_min = MAX((int64_t)0, (int64_t)floor(f01 - support1 + args.poffs));
- int64_t y_max = MIN(args.ne01, (int64_t)ceil (f01 + support1 + args.poffs));
-
- float sum = 0.0f;
- float wsum = 0.0f;
-
- for (int64_t sy = y_min; sy < y_max; ++sy) {
- const float wy = MAX(0.0f, 1.0f - fabs((float)sy - f01) * invscale1);
- for (int64_t sx = x_min; sx < x_max; ++sx) {
- const float wx = MAX(0.0f, 1.0f - fabs((float)sx - f00) * invscale0);
- const float w = wx * wy;
- device const float * src_ptr = (device const float *)(src0 + sy*args.nb01 + sx*args.nb00);
- sum += (*src_ptr) * w;
- wsum += w;
- }
- }
-
- const float v = (wsum > 0.0f) ? (sum / wsum) : 0.0f;
- dst_ptr[i0] = v;
- }
- } else {
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- const float f00 = ((float)i0 + args.poffs) / args.sf0 - args.poffs;
- const int64_t i00 = MAX(0, MIN(args.ne00 - 1, (int64_t)floor(f00)));
- const int64_t i00p = MAX(0, MIN(args.ne00 - 1, i00 + 1));
- const float fd0 = MAX(0.0f, MIN(1.0f, f00 - (float)i00));
-
- device const float * src00 = (device const float *)(src0 + i01*args.nb01 + i00*args.nb00);
- device const float * src10 = (device const float *)(src0 + i01*args.nb01 + i00p*args.nb00);
- device const float * src01 = (device const float *)(src0 + i01p*args.nb01 + i00*args.nb00);
- device const float * src11 = (device const float *)(src0 + i01p*args.nb01 + i00p*args.nb00);
-
- const float v =
- (*src00) * (1.0f - fd0) * (1.0f - fd1) +
- (*src10) * fd0 * (1.0f - fd1) +
- (*src01) * (1.0f - fd0) * fd1 +
- (*src11) * fd0 * fd1;
-
- dst_ptr[i0] = v;
- }
- }
-}
-
-template <typename T>
-kernel void kernel_conv_3d(
- constant ggml_metal_kargs_conv_3d & args,
- device const char * src0, // Weights [IC * OC, KD, KH, KW]
- device const char * src1, // Inputs [IC * N, ID, IH, IW]
- device char * dst, // Outputs [OC * N, OD, OH, OW]
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]]) {
-
- // 1. Un-flatten the spatial dimension from Grid X
- int64_t spatial_idx = tgpig.x * 32 + tpitg.x;
-
- if (spatial_idx >= args.OW * args.OH * args.OD) {
- return; // Thread falls outside the spatial volume
- }
-
- int64_t od = spatial_idx / (args.OW * args.OH);
- int64_t oh = (spatial_idx / args.OW) % args.OH;
- int64_t ow = spatial_idx % args.OW;
-
- // 2. Map Y to Channels, Z to Batch
- int64_t oc = tgpig.y;
- int64_t batch_idx = tgpig.z;
-
- // 3. Calculate anchor coordinates in the Input volume
- int64_t i_w_base = ow * args.s0 - args.p0;
- int64_t i_h_base = oh * args.s1 - args.p1;
- int64_t i_d_base = od * args.s2 - args.p2;
-
- float sum = 0.0f;
-
- // 4. Gather Loop (Iterate over Input Channels -> Depth -> Height -> Width)
- for (int64_t ic = 0; ic < args.IC; ++ic) {
-
- // ggml packs batch and channel together in the 4th dimension
- int64_t src_cn_idx = batch_idx * args.IC + ic;
- int64_t w_cn_idx = oc * args.IC + ic;
-
- for (int64_t kz = 0; kz < args.KD; ++kz) {
- int64_t id = i_d_base + kz * args.d2;
- if (id < 0 || id >= args.ID) continue; // Boundary check (Padding)
-
- for (int64_t ky = 0; ky < args.KH; ++ky) {
- int64_t ih = i_h_base + ky * args.d1;
- if (ih < 0 || ih >= args.IH) continue;
-
- for (int64_t kx = 0; kx < args.KW; ++kx) {
- int64_t iw = i_w_base + kx * args.d0;
- if (iw < 0 || iw >= args.IW) continue;
-
- // Convert multi-dimensional coordinates to flat byte offsets
- int64_t w_idx = kx*args.nb00 + ky*args.nb01 + kz*args.nb02 + w_cn_idx*args.nb03;
- int64_t i_idx = iw*args.nb10 + ih*args.nb11 + id*args.nb12 + src_cn_idx*args.nb13;
-
- // Dereference memory and cast weights to f32 if they were f16
- float w_val = (float)*(device const T*)((device const char*)src0 + w_idx);
- float i_val = *(device const float*)((device const char*)src1 + i_idx);
-
- sum += w_val * i_val;
- }
- }
- }
- }
-
- // 5. Write the accumulated value out to RAM
- int64_t dst_cn_idx = batch_idx * args.OC + oc;
- int64_t d_idx = ow*args.nb0 + oh*args.nb1 + od*args.nb2 + dst_cn_idx*args.nb3;
-
- *(device float*)(dst + d_idx) = sum;
-}
-
-// Explicit instantiations so the JIT compiler can find them by name
-template [[host_name("kernel_conv_3d_f32_f32")]]
-kernel void kernel_conv_3d<float>(
- constant ggml_metal_kargs_conv_3d & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]]);
-
-// Explicit instantiation for f16 weights
-template [[host_name("kernel_conv_3d_f16_f32")]]
-kernel void kernel_conv_3d<half>(
- constant ggml_metal_kargs_conv_3d & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]]);
-
-
-static inline float bicubic_weight1(float x) {
- const float a = -0.75f;
- return ((a + 2) * x - (a + 3)) * x * x + 1;
-}
-
-static inline float bicubic_weight2(float x) {
- const float a = -0.75f;
- return ((a * x - 5 * a) * x + 8 * a) * x - 4 * a;
-}
-
-kernel void kernel_upscale_bicubic_f32(
- constant ggml_metal_kargs_upscale & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int64_t i3 = tgpig.z;
- const int64_t i2 = tgpig.y;
- const int64_t i1 = tgpig.x;
-
- const int64_t i03 = i3 / args.sf3;
- const int64_t i02 = i2 / args.sf2;
-
- const float f01 = ((float)i1 + args.poffs) / args.sf1 - args.poffs;
- const int64_t i01 = (int64_t)floor(f01);
- const float fd1 = f01 - (float)i01;
-
- const float w_y0 = bicubic_weight2(fd1 + 1.0f);
- const float w_y1 = bicubic_weight1(fd1);
- const float w_y2 = bicubic_weight1(1.0f - fd1);
- const float w_y3 = bicubic_weight2(2.0f - fd1);
-
- const device char * src_slice = src0 + i03 * args.nb03 + i02 * args.nb02;
-
- device float * dst_ptr = (device float *)(dst + i3 * args.nb3 + i2 * args.nb2 + i1 * args.nb1);
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- const float f00 = ((float)i0 + args.poffs) / args.sf0 - args.poffs;
- const int64_t i00 = (int64_t)floor(f00);
- const float fd0 = f00 - (float)i00;
-
- const float w_x0 = bicubic_weight2(fd0 + 1.0f);
- const float w_x1 = bicubic_weight1(fd0);
- const float w_x2 = bicubic_weight1(1.0f - fd0);
- const float w_x3 = bicubic_weight2(2.0f - fd0);
-
- float sum = 0.0f;
-
- for (int dy = -1; dy <= 2; ++dy) {
- const int64_t iy = MAX(0, MIN(args.ne01 - 1, i01 + dy));
- const float wy = (dy == -1) ? w_y0 : (dy == 0) ? w_y1 : (dy == 1) ? w_y2 : w_y3;
-
- for (int dx = -1; dx <= 2; ++dx) {
- const int64_t ix = MAX(0, MIN(args.ne00 - 1, i00 + dx));
- const float wx = (dx == -1) ? w_x0 : (dx == 0) ? w_x1 : (dx == 1) ? w_x2 : w_x3;
-
- device const float * src_ptr = (device const float *)(src_slice + iy * args.nb01 + ix * args.nb00);
- sum += (*src_ptr) * wx * wy;
- }
- }
-
- dst_ptr[i0] = sum;
- }
-}
-
-kernel void kernel_roll_f32(
- constant ggml_metal_kargs_roll & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int64_t i3 = tgpig.z;
- const int64_t i2 = tgpig.y;
- const int64_t i1 = tgpig.x;
-
- device const float * src0_ptr = (device const float *) src0;
- device float * dst_ptr = (device float *) dst;
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- // apply shifts and wrap around
- int64_t i00 = i0 - args.s0;
- int64_t i01 = i1 - args.s1;
- int64_t i02 = i2 - args.s2;
- int64_t i03 = i3 - args.s3;
-
- if (i00 < 0) { i00 += args.ne00; } else if (i00 >= args.ne00) { i00 -= args.ne00; }
- if (i01 < 0) { i01 += args.ne01; } else if (i01 >= args.ne01) { i01 -= args.ne01; }
- if (i02 < 0) { i02 += args.ne02; } else if (i02 >= args.ne02) { i02 -= args.ne02; }
- if (i03 < 0) { i03 += args.ne03; } else if (i03 >= args.ne03) { i03 -= args.ne03; }
-
- int64_t src_idx = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00 + i00;
- int64_t dst_idx = i3 *args.ne2 *args.ne1 *args.ne0 + i2 *args.ne1 *args.ne0 + i1 *args.ne0 + i0;
-
- dst_ptr[dst_idx] = src0_ptr[src_idx];
- }
-}
-
-template <typename T>
-kernel void kernel_pad_impl(
- constant ggml_metal_kargs_pad & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
- const int32_t i3 = tgpig.z;
- const int32_t i2 = tgpig.y;
- const int32_t k0 = tgpig.x/args.ne1;
- const int32_t i1 = tgpig.x - k0*args.ne1;
-
- const int32_t i03 = i3;
- const int32_t i02 = i2;
- const int32_t i01 = i1;
-
- device const T * src0_ptr = (device const T *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
- device T * dst_ptr = (device T *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1);
-
- for (int32_t l0 = 0; l0 < 1024; l0 += ntg.x) {
- const int32_t i0 = k0*1024 + tpitg.x + l0;
- if (i0 >= args.ne0) {
- break;
- }
-
- if (i0 < args.ne00 && i1 < args.ne01 && i2 < args.ne02 && i3 < args.ne03) {
- dst_ptr[i0] = src0_ptr[i0];
- } else {
- dst_ptr[i0] = 0.0f;
- }
- }
-}
-
-typedef decltype(kernel_pad_impl<float>) kernel_pad_t;
-
-template [[host_name("kernel_pad_f32")]] kernel kernel_pad_t kernel_pad_impl<float>;
-template [[host_name("kernel_pad_f32_4")]] kernel kernel_pad_t kernel_pad_impl<float4>;
-
-// TODO: this is slow - optimize
-kernel void kernel_pad_reflect_1d_f32(
- constant ggml_metal_kargs_pad_reflect_1d & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tgpg[[threadgroups_per_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- const int64_t i3 = tgpig.z;
- const int64_t i2 = tgpig.y;
- const int64_t i1 = tgpig.x;
-
- const int64_t i03 = i3;
- const int64_t i02 = i2;
- const int64_t i01 = i1;
-
- device const float * src0_ptr = (device const float *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
- device float * dst_ptr = (device float *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1);
-
- if (i1 < args.ne01 && i2 < args.ne02 && i3 < args.ne03) {
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- if (i0 < args.p0) {
- dst_ptr[i0] = src0_ptr[args.p0 - i0];
- } else if (i0 < args.ne0 - args.p1) {
- dst_ptr[i0] = src0_ptr[i0 - args.p0];
- } else {
- dst_ptr[i0] = src0_ptr[(args.ne0 - args.p1 - args.p0) - (args.p1 + 1 - (args.ne0 - i0)) - 1];
- }
- }
- }
-}
-
-kernel void kernel_arange_f32(
- constant ggml_metal_kargs_arange & args,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- device float * dst_ptr = (device float *) dst;
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- dst_ptr[i0] = args.start + args.step * i0;
- }
-}
-
-kernel void kernel_timestep_embedding_f32(
- constant ggml_metal_kargs_timestep_embedding & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint3 tpitg[[thread_position_in_threadgroup]],
- uint3 ntg[[threads_per_threadgroup]]) {
-
- int i = tgpig.x;
- device float * embed_data = (device float *)(dst + i*args.nb1);
-
- int half_ = args.dim / 2;
- for (int j = tpitg.x; j < half_; j += ntg.x) {
- float timestep = ((device float *)src0)[i];
- float freq = (float)exp(-log((float)args.max_period) * j / half_);
- float arg = timestep * freq;
- embed_data[j ] = cos(arg);
- embed_data[j + half_] = sin(arg);
- }
-
- if (args.dim % 2 != 0 && tpitg.x == 0) {
- embed_data[2 * half_] = 0.f;
- }
-}
-
-// bitonic sort implementation following the CUDA kernels as reference
-typedef void (argsort_t)(
- constant ggml_metal_kargs_argsort & args,
- device const char * src0,
- device int32_t * dst,
- threadgroup int32_t * shmem_i32 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]);
-
-template<ggml_sort_order order>
-kernel void kernel_argsort_f32_i32(
- constant ggml_metal_kargs_argsort & args,
- device const char * src0,
- device int32_t * dst,
- threadgroup int32_t * shmem_i32 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- // bitonic sort
- const int col = tpitg[0];
- const int ib = tgpig[0] / args.ne01;
-
- const int i00 = ib*ntg.x;
- const int i01 = tgpig[0] % args.ne01;
- const int i02 = tgpig[1];
- const int i03 = tgpig[2];
-
- device const float * src0_row = (device const float *) (src0 + args.nb01*i01 + args.nb02*i02 + args.nb03*i03);
-
- // initialize indices
- shmem_i32[col] = i00 + col;
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- for (int k = 2; k <= ntg.x; k *= 2) {
- for (int j = k / 2; j > 0; j /= 2) {
- int ixj = col ^ j;
- if (ixj > col) {
- if ((col & k) == 0) {
- if (shmem_i32[col] >= args.ne00 ||
- (shmem_i32[ixj] < args.ne00 && (order == GGML_SORT_ORDER_ASC ?
- src0_row[shmem_i32[col]] > src0_row[shmem_i32[ixj]] :
- src0_row[shmem_i32[col]] < src0_row[shmem_i32[ixj]]))
- ) {
- SWAP(shmem_i32[col], shmem_i32[ixj]);
- }
- } else {
- if (shmem_i32[ixj] >= args.ne00 ||
- (shmem_i32[col] < args.ne00 && (order == GGML_SORT_ORDER_ASC ?
- src0_row[shmem_i32[col]] < src0_row[shmem_i32[ixj]] :
- src0_row[shmem_i32[col]] > src0_row[shmem_i32[ixj]]))
- ) {
- SWAP(shmem_i32[col], shmem_i32[ixj]);
- }
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
- }
-
- const int64_t i0 = ib*args.top_k;
-
- // copy the result to dst without the padding
- if (i0 + col < args.ne0 && col < args.top_k) {
- dst += i0 + args.ne0*i01 + args.ne0*args.ne1*i02 + args.ne0*args.ne1*args.ne2*i03;
-
- dst[col] = shmem_i32[col];
- }
-}
-
-template [[host_name("kernel_argsort_f32_i32_asc")]] kernel argsort_t kernel_argsort_f32_i32<GGML_SORT_ORDER_ASC>;
-template [[host_name("kernel_argsort_f32_i32_desc")]] kernel argsort_t kernel_argsort_f32_i32<GGML_SORT_ORDER_DESC>;
-
-typedef void (argsort_merge_t)(
- constant ggml_metal_kargs_argsort_merge & args,
- device const char * src0,
- device const int32_t * tmp,
- device int32_t * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]);
-
-template<ggml_sort_order order>
-kernel void kernel_argsort_merge_f32_i32(
- constant ggml_metal_kargs_argsort_merge & args,
- device const char * src0,
- device const int32_t * tmp,
- device int32_t * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
-
- const int im = tgpig[0] / args.ne01;
- const int i01 = tgpig[0] % args.ne01;
- const int i02 = tgpig[1];
- const int i03 = tgpig[2];
-
- const int start = im * (2 * args.len);
-
- const int len0 = MIN(args.len, MAX(0, args.ne0 - (int)(start)));
- const int len1 = MIN(args.len, MAX(0, args.ne0 - (int)(start + args.len)));
-
- const int total = len0 + len1;
-
- device const int32_t * tmp0 = tmp + start
- + i01*args.ne0
- + i02*args.ne0*args.ne01
- + i03*args.ne0*args.ne01*args.ne02;
-
- device const int32_t * tmp1 = tmp0 + args.len;
-
- dst += start
- + i01*args.top_k
- + i02*args.top_k*args.ne01
- + i03*args.top_k*args.ne01*args.ne02;
-
- device const float * src0_row = (device const float *)(src0
- + args.nb01*i01
- + args.nb02*i02
- + args.nb03*i03);
-
- if (total == 0) {
- return;
- }
-
- const int chunk = (total + ntg.x - 1) / ntg.x;
-
- const int k0 = tpitg.x * chunk;
- const int k1 = MIN(MIN(k0 + chunk, total), args.top_k);
-
- if (k0 >= args.top_k) {
- return;
- }
-
- if (k0 >= total) {
- return;
- }
-
- int low = k0 > len1 ? k0 - len1 : 0;
- int high = MIN(k0, len0);
-
- // binary-search partition (i, j) such that i + j = k
- while (low < high) {
- const int mid = (low + high) >> 1;
-
- const int32_t idx0 = tmp0[mid];
- const int32_t idx1 = tmp1[k0 - mid - 1];
-
- const float val0 = src0_row[idx0];
- const float val1 = src0_row[idx1];
-
- bool take_left;
- if (order == GGML_SORT_ORDER_ASC) {
- take_left = (val0 <= val1);
- } else {
- take_left = (val0 >= val1);
- }
-
- if (take_left) {
- low = mid + 1;
- } else {
- high = mid;
- }
- }
-
- int i = low;
- int j = k0 - i;
-
- // keep the merge fronts into registers
- int32_t idx0 = 0;
- float val0 = 0.0f;
- if (i < len0) {
- idx0 = tmp0[i];
- val0 = src0_row[idx0];
- }
-
- int32_t idx1 = 0;
- float val1 = 0.0f;
- if (j < len1) {
- idx1 = tmp1[j];
- val1 = src0_row[idx1];
- }
-
- for (int k = k0; k < k1; ++k) {
- int32_t out_idx;
-
- if (i >= len0) {
- while (k < k1) {
- dst[k++] = tmp1[j++];
- }
- break;
- } else if (j >= len1) {
- while (k < k1) {
- dst[k++] = tmp0[i++];
- }
- break;
- } else {
- bool take_left;
-
- if (order == GGML_SORT_ORDER_ASC) {
- take_left = (val0 <= val1);
- } else {
- take_left = (val0 >= val1);
- }
-
- if (take_left) {
- out_idx = idx0;
- ++i;
- if (i < len0) {
- idx0 = tmp0[i];
- val0 = src0_row[idx0];
- }
- } else {
- out_idx = idx1;
- ++j;
- if (j < len1) {
- idx1 = tmp1[j];
- val1 = src0_row[idx1];
- }
- }
- }
-
- dst[k] = out_idx;
- }
-}
-
-template [[host_name("kernel_argsort_merge_f32_i32_asc")]] kernel argsort_merge_t kernel_argsort_merge_f32_i32<GGML_SORT_ORDER_ASC>;
-template [[host_name("kernel_argsort_merge_f32_i32_desc")]] kernel argsort_merge_t kernel_argsort_merge_f32_i32<GGML_SORT_ORDER_DESC>;
-
-template<int N>
-kernel void kernel_fwht_f32(
- constant ggml_metal_kargs_fwht & args,
- device const float * src,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
-
- constexpr int NW = N_SIMDWIDTH;
- constexpr int NE = N / NW;
-
- const float scale = 1.0f / sqrt((float) N);
-
- const int sg_per_tg = ntg.x / NW;
- const int64_t r = tgpig.x * sg_per_tg + sgitg;
- if (r >= args.nrows) {
- return;
- }
-
- src += r * N;
- dst += r * N;
-
- const int lane = tiisg;
-
- float reg[NE];
- for (int i = 0; i < NE; i++) {
- reg[i] = src[i*NW + lane]*scale;
- }
- for (int i = 1; i < NW; i *= 2) {
- for (int j = 0; j < NE; j++) {
- const float val = reg[j];
- const float val2 = simd_shuffle_xor(val, i);
- reg[j] = (lane & i) == 0 ? val2 + val : val2 - val;
- }
- }
-
- for (int i = NW; i < N; i *= 2) {
- const int step = i / NW;
- for (int j = 0; j < NE; j += (2 * step)) {
- for (int k = 0; k < step; k++) {
- const float x = reg[j + k ];
- const float y = reg[j + k + step];
- reg[j + k] = x + y;
- reg[j + k + step] = x - y;
- }
- }
- }
-
- for (int i = 0; i < NE; i++) {
- dst[i*NW + lane] = reg[i];
- }
-}
-
-typedef decltype(kernel_fwht_f32<64>) kernel_fwht_t;
-
-template [[host_name("kernel_fwht_f32_64")]] kernel kernel_fwht_t kernel_fwht_f32<64>;
-template [[host_name("kernel_fwht_f32_128")]] kernel kernel_fwht_t kernel_fwht_f32<128>;
-template [[host_name("kernel_fwht_f32_256")]] kernel kernel_fwht_t kernel_fwht_f32<256>;
-template [[host_name("kernel_fwht_f32_512")]] kernel kernel_fwht_t kernel_fwht_f32<512>;
-
-// dequantize a quantized KV cache tensor to contiguous F16 before running the F16 flash attention kernels
-// - one thread per block; dispatched separately for K and V
-// - ref: https://github.com/ggml-org/llama.cpp/pull/27390
-template <
- typename block_t,
- short QK,
- void (*deq_t4x4)(device const block_t *, short, thread float4x4 &)>
-kernel void kernel_flash_attn_ext_kv_f16(
- constant ggml_metal_kargs_flash_attn_ext_kv_f16 & args,
- device const char * x,
- device half * x_dst,
- uint gid [[thread_position_in_grid]]) {
- if (gid >= (uint) args.nblocks) {
- return;
- }
-
- const uint nb = args.ne0/QK;
- const uint i0 = gid%nb;
- uint ib = gid/nb;
- const uint i1 = ib%args.ne1;
- ib /= args.ne1;
- const uint i2 = ib%args.ne2;
- const uint i3 = ib/args.ne2;
-
- const uint64_t offs = i0*args.nb0 + i1*args.nb1 + i2*args.nb2 + i3*args.nb3;
-
- device const block_t * src = (device const block_t *) (x + offs);
- device half4 * dst = (device half4 *) x_dst + (QK/4)*gid;
-
- for (short i = 0; i < QK/16; ++i) {
- float4x4 reg;
- deq_t4x4(src, i, reg);
- dst[4*i + 0] = (half4) reg[0];
- dst[4*i + 1] = (half4) reg[1];
- dst[4*i + 2] = (half4) reg[2];
- dst[4*i + 3] = (half4) reg[3];
- }
-}
-
-typedef decltype(kernel_flash_attn_ext_kv_f16<block_q8_0, 32, dequantize_q8_0>) kernel_flash_attn_ext_kv_f16_t;
-
-template [[host_name("kernel_flash_attn_ext_kv_q4_0_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q4_0, 32, dequantize_q4_0>;
-template [[host_name("kernel_flash_attn_ext_kv_q4_1_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q4_1, 32, dequantize_q4_1>;
-template [[host_name("kernel_flash_attn_ext_kv_q5_0_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q5_0, 32, dequantize_q5_0>;
-template [[host_name("kernel_flash_attn_ext_kv_q5_1_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q5_1, 32, dequantize_q5_1>;
-template [[host_name("kernel_flash_attn_ext_kv_q8_0_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q8_0, 32, dequantize_q8_0>;
-
-constant bool FC_flash_attn_ext_pad_has_mask [[function_constant(FC_FLASH_ATTN_EXT_PAD + 0)]];
-
-constant int32_t FC_flash_attn_ext_pad_ncpsg [[function_constant(FC_FLASH_ATTN_EXT_PAD + 25)]];
-
-// pad the last chunk of C elements of k and v into a an extra pad buffer
-kernel void kernel_flash_attn_ext_pad(
- constant ggml_metal_kargs_flash_attn_ext_pad & args,
- device const char * k,
- device const char * v,
- device const char * mask,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int32_t C = FC_flash_attn_ext_pad_ncpsg;
-
- device char * k_pad = dst;
- device char * v_pad = k_pad + args.nb11*C*args.ne_12_2*args.ne_12_3;
- device char * mask_pad = v_pad + args.nb21*C*args.ne_12_2*args.ne_12_3;
-
- const int32_t icp = args.ne11 % C;
- const int32_t ic0 = args.ne11 - icp;
-
- const int32_t i1 = tgpig[0];
- const int32_t i2 = tgpig[1];
- const int32_t i3 = tgpig[2];
-
- if (i2 < args.ne_12_2 && i3 < args.ne_12_3) {
- device const char * k_src = k + args.nb11*(ic0 + i1) + args.nb12*i2 + args.nb13*i3;
- device const char * v_src = v + args.nb21*(ic0 + i1) + args.nb22*i2 + args.nb23*i3;
-
- device char * k_dst = k_pad + args.nb11*i1 + args.nb11*C*i2 + args.nb11*C*args.ne_12_2*i3;
- device char * v_dst = v_pad + args.nb21*i1 + args.nb21*C*i2 + args.nb21*C*args.ne_12_2*i3;
-
- if (i1 >= icp) {
- // here it is not important the exact value that will be used as we rely on masking out the scores in the attention
- for (uint64_t i = tiitg; i < args.nb11; i += ntg.x) {
- k_dst[i] = 0;
- }
- for (uint64_t i = tiitg; i < args.nb21; i += ntg.x) {
- v_dst[i] = 0;
- }
- } else {
- for (uint64_t i = tiitg; i < args.nb11; i += ntg.x) {
- k_dst[i] = k_src[i];
- }
- for (uint64_t i = tiitg; i < args.nb21; i += ntg.x) {
- v_dst[i] = v_src[i];
- }
- }
- }
-
- if (FC_flash_attn_ext_pad_has_mask) {
- if (i2 < args.ne32 && i3 < args.ne33) {
- for (int ib = i1; ib < args.ne31; ib += C) {
- device const half * mask_src = (device const half *)(mask + args.nb31*ib + args.nb32*i2 + args.nb33*i3) + ic0;
- device half * mask_dst = (device half *)(mask_pad) + C*ib + C*args.ne31*i2 + C*args.ne31*args.ne32*i3;
-
- for (int i = tiitg; i < C; i += ntg.x) {
- if (i >= icp) {
- mask_dst[i] = -MAXHALF;
- } else {
- mask_dst[i] = mask_src[i];
- }
- }
- }
- }
- }
-}
-
-constant int32_t FC_flash_attn_ext_blk_nqptg [[function_constant(FC_FLASH_ATTN_EXT_BLK + 24)]];
-constant int32_t FC_flash_attn_ext_blk_ncpsg [[function_constant(FC_FLASH_ATTN_EXT_BLK + 25)]];
-
-// scan the blocks of the mask that are not masked
-// 0 - masked (i.e. full of -INF, skip)
-// 1 - not masked (i.e. at least one element of the mask is not -INF)
-// 2 - all zero
-kernel void kernel_flash_attn_ext_blk(
- constant ggml_metal_kargs_flash_attn_ext_blk & args,
- device const char * mask,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]]) {
- // block size C x Q
- const int32_t Q = FC_flash_attn_ext_blk_nqptg;
- const int32_t C = FC_flash_attn_ext_blk_ncpsg;
-
- constexpr short NW = N_SIMDWIDTH;
-
- const int32_t i3 = tgpig[2]/args.ne32;
- const int32_t i2 = tgpig[2]%args.ne32;
- const int32_t i1 = tgpig[1];
- const int32_t i0 = tgpig[0];
-
- char res = i0*C + C > args.ne30 ? 1 : 0;
-
- device const half * mask_src = (device const half *) (mask + (i1*Q)*args.nb31 + i2*args.nb32 + i3*args.nb33) + i0*C + tiisg;
-
- // detailed check of the elements of the block
- if ((C > NW || Q > 1) && res == 0) {
- half mmin = MAXHALF;
- half mmax = -MAXHALF;
-
- FOR_UNROLL (short j = 0; j < Q; ++j) {
- FOR_UNROLL (short ii = 0; ii < C/NW; ++ii) {
- mmin = min(mmin, mask_src[ii*NW]);
- mmax = max(mmax, mask_src[ii*NW]);
- }
-
- mask_src += args.nb31/2;
- }
-
- mmin = simd_min(mmin);
- mmax = simd_max(mmax);
-
- if (mmax > -MAXHALF) {
- if (mmin == 0.0 && mmax == 0.0) {
- res = 2;
- } else {
- res = 1;
- }
- }
- }
-
- const int32_t nblk1 = ((args.ne01 + Q - 1)/Q);
- const int32_t nblk0 = ((args.ne30 + C - 1)/C);
-
- if (tiisg == 0) {
- dst[((i3*args.ne32 + i2)*nblk1 + i1)*nblk0 + i0] = res;
- }
-}
-
-constant bool FC_flash_attn_ext_has_mask [[function_constant(FC_FLASH_ATTN_EXT + 0)]];
-constant bool FC_flash_attn_ext_has_sinks [[function_constant(FC_FLASH_ATTN_EXT + 1)]];
-constant bool FC_flash_attn_ext_has_bias [[function_constant(FC_FLASH_ATTN_EXT + 2)]];
-constant bool FC_flash_attn_ext_has_scap [[function_constant(FC_FLASH_ATTN_EXT + 3)]];
-constant bool FC_flash_attn_ext_has_kvpad [[function_constant(FC_FLASH_ATTN_EXT + 4)]];
-
-constant bool FC_flash_attn_ext_bc_mask [[function_constant(FC_FLASH_ATTN_EXT + 10)]];
-
-//constant float FC_flash_attn_ext_scale [[function_constant(FC_FLASH_ATTN_EXT + 10)]];
-//constant float FC_flash_attn_ext_max_bias [[function_constant(FC_FLASH_ATTN_EXT + 11)]];
-//constant float FC_flash_attn_ext_logit_softcap [[function_constant(FC_FLASH_ATTN_EXT + 12)]];
-
-constant int32_t FC_flash_attn_ext_ns10 [[function_constant(FC_FLASH_ATTN_EXT + 20)]];
-constant int32_t FC_flash_attn_ext_ns20 [[function_constant(FC_FLASH_ATTN_EXT + 21)]];
-constant int32_t FC_flash_attn_ext_nsg [[function_constant(FC_FLASH_ATTN_EXT + 22)]];
-
-// ref: https://arxiv.org/pdf/2307.08691.pdf
-template<
- typename q_t, // query types in shared memory
- typename q4_t,
- typename q8x8_t,
- typename k_t, // key types in shared memory
- typename k4x4_t,
- typename k8x8_t,
- typename v_t, // value types in shared memory
- typename v4x4_t,
- typename v8x8_t,
- typename qk_t, // Q*K types
- typename qk8x8_t,
- typename s_t, // soft-max types
- typename s2_t,
- typename s8x8_t,
- typename o_t, // attention accumulation types
- typename o4_t,
- typename o8x8_t,
- typename kd4x4_t, // key type in device memory
- short nl_k,
- void (*deq_k)(device const kd4x4_t *, short, thread k4x4_t &),
- typename vd4x4_t, // value type in device memory
- short nl_v,
- void (*deq_v)(device const vd4x4_t *, short, thread v4x4_t &),
- short DK, // K head size
- short DV, // V head size
- short Q, // queries per threadgroup
- short C, // cache items per threadgroup
- short NSG> // number of simd groups
-void kernel_flash_attn_ext_impl(
- constant ggml_metal_kargs_flash_attn_ext & args,
- device const char * q,
- device const char * k,
- device const char * v,
- device const char * mask,
- device const char * sinks,
- device const char * pad,
- device const char * blk,
- device char * dst,
- threadgroup half * shmem_f16,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const ushort iq3 = tgpig[2];
- const ushort iq2 = tgpig[1];
- const ushort iq1 = tgpig[0]*Q;
-
-#define NS10 (FC_flash_attn_ext_ns10)
-#define NS20 (FC_flash_attn_ext_ns20)
-
- // note: I had some concerns that using this instead of the ugly macros above was affecting performance
- // need to re-check carefully and if no regressions are observerd - remove the macros
- // the concerns is that maybe using const variables requires extra registers? but not sure if the compiler
- // is clever enough to avoid this. unfortunately, using constexpr is not possible with FC
- //const short NS10 = FC_flash_attn_ext_ns10;
- //const short NS20 = FC_flash_attn_ext_ns20;
-
- constexpr short KV = 8;
-
- constexpr short DK4 = DK/4;
- constexpr short DK8 = DK/8;
- constexpr short DK16 = DK/16;
- constexpr short DV4 = DV/4;
- //constexpr short DV8 = DV/8;
- constexpr short DV16 = DV/16;
-
- constexpr short PV = PAD2(DV, 64);
- constexpr short PV4 = PV/4;
- constexpr short PV8 = PV/8;
- //constexpr short PV16 = PV/16;
-
- constexpr short NW = N_SIMDWIDTH;
- constexpr short NQ = Q/NSG;
- constexpr short SH = 2*C; // shared memory per simdgroup (s_t == float)
-
- constexpr short TS = 2*SH;
- constexpr short T = DK + 2*PV; // shared memory size per query in (half)
-
- threadgroup q_t * sq = (threadgroup q_t *) (shmem_f16 + 0*T); // holds the query data
- threadgroup q4_t * sq4 = (threadgroup q4_t *) (shmem_f16 + 0*T); // same as above but in q4_t
- threadgroup o_t * so = (threadgroup o_t *) (shmem_f16 + 0*T + Q*DK); // the result for all queries in 8x8 matrices (the O matrix from the paper)
- threadgroup o4_t * so4 = (threadgroup o4_t *) (shmem_f16 + 0*T + Q*DK);
- threadgroup s_t * ss = (threadgroup s_t *) (shmem_f16 + Q*T); // scratch buffer for attention, mask and diagonal matrix
- threadgroup s2_t * ss2 = (threadgroup s2_t *) (shmem_f16 + Q*T); // same as above but in s2_t
-
- threadgroup k_t * sk = (threadgroup k_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // scratch buffer to load K in shared memory
- threadgroup k4x4_t * sk4x4 = (threadgroup k4x4_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // same as above but in k4x4_t
-
- threadgroup v_t * sv = (threadgroup v_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // scratch buffer to load V in shared memory
- threadgroup v4x4_t * sv4x4 = (threadgroup v4x4_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // same as above but in v4x4_t
-
- // mask storage in shared mem
- threadgroup half2 * sm2 = (threadgroup half2 *) (shmem_f16 + Q*T + 2*C);
-
- // per-query mask pointers
- device const half2 * pm2[NQ];
-
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- pm2[jj] = (device const half2 *) ((device const char *) mask + (iq1 + j)*args.nb31 + (iq2%args.ne32)*args.nb32 + (iq3%args.ne33)*args.nb33);
- }
-
- {
- const int32_t nblk1 = ((args.ne01 + Q - 1)/Q);
- const int32_t nblk0 = ((args.ne11 + C - 1)/C);
-
- blk += (((iq3%args.ne33)*args.ne32 + (iq2%args.ne32))*nblk1 + iq1/Q)*nblk0;
- }
-
- {
- q += iq1*args.nb01 + iq2*args.nb02 + iq3*args.nb03;
-
- const short ikv2 = iq2/(args.ne02/args.ne_12_2);
- const short ikv3 = iq3/(args.ne03/args.ne_12_3);
-
- k += ikv2*args.nb12 + ikv3*args.nb13;
- v += ikv2*args.nb22 + ikv3*args.nb23;
- }
-
- // load heads from Q to shared memory
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- device const float4 * q4 = (device const float4 *) ((device const char *) q + j*args.nb01);
-
- for (short i = tiisg; i < DK4; i += NW) {
- if (iq1 + j < args.ne01) {
- sq4[j*DK4 + i] = (q4_t) q4[i];
- } else {
- sq4[j*DK4 + i] = 0;
- }
- }
- }
-
- // zero out
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- for (short i = tiisg; i < DV4; i += NW) {
- so4[j*PV4 + i] = 0;
- }
-
- for (short i = tiisg; i < SH; i += NW) {
- ss[j*SH + i] = 0.0f;
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- float S[NQ] = { [0 ... NQ-1] = 0.0f };
-
- {
- float M[NQ] = { [0 ... NQ-1] = -FLT_MAX/2 };
-
- float slope = 1.0f;
-
- // ALiBi
- if (FC_flash_attn_ext_has_bias) {
- const short h = iq2;
-
- const float base = h < args.n_head_log2 ? args.m0 : args.m1;
- const short exph = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
-
- slope = pow(base, exph);
- }
-
- // loop over the KV cache
- // each simdgroup handles blocks of Q rows and C columns
- for (int ic0 = 0; ; ++ic0) {
- int ic = ic0*C;
- if (ic >= args.ne11) {
- break;
- }
-
- // the last partial chunk uses the pad buffer as source
- if (FC_flash_attn_ext_has_kvpad && ic + C > args.ne11) {
- k = pad;
- v = k + args.nb11*C*args.ne_12_2*args.ne_12_3;
- mask = v + args.nb21*C*args.ne_12_2*args.ne_12_3;
-
- const short ikv2 = iq2/(args.ne02/args.ne_12_2);
- const short ikv3 = iq3/(args.ne03/args.ne_12_3);
-
- k += (ikv2 + ikv3*args.ne_12_2)*args.nb11*C;
- v += (ikv2 + ikv3*args.ne_12_2)*args.nb21*C;
-
- if (!FC_flash_attn_ext_has_mask) {
- threadgroup half * sm = (threadgroup half *) (sm2);
-
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- for (short i = tiisg; i < C; i += NW) {
- if (ic + i >= args.ne11) {
- sm[2*j*SH + i] = -MAXHALF;
- }
- }
- }
- } else {
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- pm2[jj] = (device const half2 *) ((device const half *) mask +
- (iq1 + j)*C +
- (iq2%args.ne32)*(C*args.ne31) +
- (iq3%args.ne33)*(C*args.ne31*args.ne32));
- }
- }
-
- ic = 0;
- }
-
- char blk_cur = 1;
-
- // read the mask into shared mem
- if (FC_flash_attn_ext_has_mask) {
- blk_cur = blk[ic0];
-
- if (blk_cur == 0) {
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- pm2[jj] += NW;
- }
-
- continue;
- }
-
- if (blk_cur == 1) {
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- if (FC_flash_attn_ext_bc_mask) {
- sm2[j*SH + tiisg] = (iq1 + j) < args.ne31 ? pm2[jj][tiisg] : half2(-MAXHALF, -MAXHALF);
- } else {
- sm2[j*SH + tiisg] = pm2[jj][tiisg];
- }
-
- pm2[jj] += NW;
- }
- } else if (blk_cur == 2) {
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- pm2[jj] += NW;
- }
- }
-
-#if 0
- // note: old -INF block optimization - obsoleted by pre-computing non-masked blocks
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // used to detect blocks full of -INF
- // skip only when the entire threadgroup is masked
- half2 smax2(-MAXHALF/2, -MAXHALF/2);
-
- FOR_UNROLL (short j = 0; j < Q; ++j) {
- smax2 = max(smax2, sm2[j*SH + tiisg]);
- }
-
- smax2 = simd_max(smax2);
-
- if (max(smax2[0], smax2[1]) <= -MAXHALF/2) {
- // this barrier is important
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- continue;
- }
-#endif
- }
-
- // Q*K^T
- // this is compile-time check, so it does not have runtime overhead
- if (is_same<kd4x4_t, k4x4_t>::value) {
- // we can read directly from global memory
- device const k_t * pk = (device const k_t *) (k + ic*args.nb11);
- threadgroup const q_t * pq = sq;
- threadgroup s_t * ps = ss;
-
- pk += sgitg*(8*NS10);
- ps += sgitg*(8*1);
-
- static_assert((C/8) % NSG == 0, "");
-
- constexpr short NC = (C/8)/NSG;
-
- FOR_UNROLL (short cc = 0; cc < NC; ++cc) {
- qk8x8_t mqk = make_filled_simdgroup_matrix<qk_t, 8>((qk_t) 0.0f);
-
- if (DK % 16 != 0) {
- k8x8_t mk;
- q8x8_t mq;
-
- FOR_UNROLL (short i = 0; i < DK8; ++i) {
- simdgroup_barrier(mem_flags::mem_none);
-
- simdgroup_load(mk, pk + 8*i, NS10, 0, true);
- simdgroup_load(mq, pq + 8*i, DK);
-
- simdgroup_barrier(mem_flags::mem_none);
-
- simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
- }
- } else {
- k8x8_t mk[2];
- q8x8_t mq[2];
-
- // note: too much unroll can tank the performance for large heads
- #pragma unroll (MIN(DK8/2, 4*NSG))
- for (short i = 0; i < DK8/2; ++i) {
- simdgroup_barrier(mem_flags::mem_none);
-
- simdgroup_load(mq[0], pq + 0*8 + 16*i, DK);
- simdgroup_load(mq[1], pq + 1*8 + 16*i, DK);
-
- simdgroup_load(mk[0], pk + 0*8 + 16*i, NS10, 0, true);
- simdgroup_load(mk[1], pk + 1*8 + 16*i, NS10, 0, true);
-
- simdgroup_barrier(mem_flags::mem_none);
-
- simdgroup_multiply_accumulate(mqk, mq[0], mk[0], mqk);
- simdgroup_multiply_accumulate(mqk, mq[1], mk[1], mqk);
- }
- }
-
- simdgroup_store(mqk, ps, SH, 0, false);
-
- pk += 8*(NSG*NS10);
- ps += 8*(NSG);
- }
- } else {
- // TODO: this is the quantized K cache branch - not optimized yet
- for (short ccc = 0; ccc < (C/8)/NSG; ++ccc) {
- const short cc = ccc*NSG + sgitg;
-
- const short tx = tiisg%4;
- const short ty = tiisg/4;
-
- qk8x8_t mqk = make_filled_simdgroup_matrix<qk_t, 8>((qk_t) 0.0f);
-
- for (short ii = 0; ii < DK16; ii += 4) {
- device const kd4x4_t * pk4x4 = (device const kd4x4_t *) (k + ((ic + 8*cc + ty)*args.nb11));
-
- if (DK16%4 == 0) {
- // the head is evenly divisible by 4*16 = 64, so no need for bound checks
- {
- k4x4_t tmp;
- deq_k(pk4x4 + (ii + tx)/nl_k, (ii + tx)%nl_k, tmp);
- sk4x4[4*ty + tx] = tmp;
- }
-
- simdgroup_barrier(mem_flags::mem_threadgroup);
-
- FOR_UNROLL (short k = 0; k < 4; ++k) {
- k8x8_t mk;
- q8x8_t mq;
-
- simdgroup_load(mk, sk + 16*k + 0*8, 4*16, 0, true); // transpose
- simdgroup_load(mq, sq + (2*(ii + k) + 0)*8, DK);
- simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
-
- simdgroup_load(mk, sk + 16*k + 1*8, 4*16, 0, true); // transpose
- simdgroup_load(mq, sq + (2*(ii + k) + 1)*8, DK);
- simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
- }
- } else {
- if (ii + tx < DK16) {
- k4x4_t tmp;
- deq_k(pk4x4 + (ii + tx)/nl_k, (ii + tx)%nl_k, tmp);
- sk4x4[4*ty + tx] = tmp;
- }
-
- simdgroup_barrier(mem_flags::mem_threadgroup);
-
- for (short k = 0; k < 4 && ii + k < DK16; ++k) {
- k8x8_t mk;
- q8x8_t mq;
-
- simdgroup_load(mk, sk + 16*k + 0*8, 4*16, 0, true); // transpose
- simdgroup_load(mq, sq + (2*(ii + k) + 0)*8, DK);
- simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
-
- simdgroup_load(mk, sk + 16*k + 1*8, 4*16, 0, true); // transpose
- simdgroup_load(mq, sq + (2*(ii + k) + 1)*8, DK);
- simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
- }
- }
- }
-
- simdgroup_store(mqk, ss + 8*cc, SH, 0, false);
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // online softmax
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- const float m = M[jj];
-
- // scale and apply the logitcap / mask
- float2 s2 = ss2[j*SH/2 + tiisg]*args.scale;
-
- if (FC_flash_attn_ext_has_scap) {
- s2 = args.logit_softcap*precise::tanh(s2);
- }
-
- // mqk = mqk + slope*mask
- if (blk_cur != 2) {
- if (FC_flash_attn_ext_has_bias) {
- s2 += s2_t(sm2[j*SH + tiisg])*slope;
- } else {
- s2 += s2_t(sm2[j*SH + tiisg]);
- }
- }
-
- M[jj] = simd_max(max(M[jj], max(s2[0], s2[1])));
-
- const float ms = exp(m - M[jj]);
- const float2 vs2 = exp(s2 - M[jj]);
-
- S[jj] = S[jj]*ms + simd_sum(vs2[0] + vs2[1]);
-
- // the P matrix from the paper (Q rows, C columns)
- ss2[j*SH/2 + tiisg] = vs2;
-
- if (DV4 % NW == 0) {
- FOR_UNROLL (short ii = 0; ii < DV4/NW; ++ii) {
- const short i = ii*NW + tiisg;
-
- so4[j*PV4 + i] *= ms;
- }
- } else {
- for (short i = tiisg; i < DV4; i += NW) {
- so4[j*PV4 + i] *= ms;
- }
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // O = O + (Q*K^T)*V
- {
- // we can read directly from global memory
- if (is_same<vd4x4_t, v4x4_t>::value) {
- static_assert(PV8 % NSG == 0, "");
-
- constexpr short NO = PV8/NSG;
-
- o8x8_t lo[NO];
-
- {
- auto sot = so + 8*sgitg;
-
- FOR_UNROLL (short ii = 0; ii < NO; ++ii) {
- simdgroup_load(lo[ii], sot, PV, 0, false);
-
- sot += 8*NSG;
- }
- }
-
- {
- device const v_t * pv = (device const v_t *) (v + ic*args.nb21);
-
- pv += 8*sgitg;
-
- if (DV <= 64) {
- FOR_UNROLL (short cc = 0; cc < C/8; ++cc) {
- s8x8_t vs;
- simdgroup_load(vs, ss + 8*cc, SH, 0, false);
-
- FOR_UNROLL (short ii = 0; ii < NO/2; ++ii) {
- v8x8_t mv[2];
-
- simdgroup_load(mv[0], pv + 0*NSG + 16*ii*NSG, NS20, 0, false);
- simdgroup_load(mv[1], pv + 8*NSG + 16*ii*NSG, NS20, 0, false);
-
- simdgroup_multiply_accumulate(lo[2*ii + 0], vs, mv[0], lo[2*ii + 0]);
- simdgroup_multiply_accumulate(lo[2*ii + 1], vs, mv[1], lo[2*ii + 1]);
- }
-
- pv += 8*NS20;
- }
- } else {
- constexpr short NC = (C/8)/2;
-
- FOR_UNROLL (short cc = 0; cc < NC; ++cc) {
- s8x8_t vs[2];
-
- simdgroup_load(vs[0], ss + 16*cc + 0, SH, 0, false);
- simdgroup_load(vs[1], ss + 16*cc + 8, SH, 0, false);
-
- FOR_UNROLL (short ii = 0; ii < NO/2; ++ii) {
- v8x8_t mv[4];
-
- simdgroup_load(mv[0], pv + 0*NSG + 16*ii*NSG + 0*8*NS20, NS20, 0, false);
- simdgroup_load(mv[1], pv + 8*NSG + 16*ii*NSG + 0*8*NS20, NS20, 0, false);
- simdgroup_load(mv[2], pv + 0*NSG + 16*ii*NSG + 1*8*NS20, NS20, 0, false);
- simdgroup_load(mv[3], pv + 8*NSG + 16*ii*NSG + 1*8*NS20, NS20, 0, false);
-
- simdgroup_multiply_accumulate(lo[2*ii + 0], vs[0], mv[0], lo[2*ii + 0]);
- simdgroup_multiply_accumulate(lo[2*ii + 1], vs[0], mv[1], lo[2*ii + 1]);
- simdgroup_multiply_accumulate(lo[2*ii + 0], vs[1], mv[2], lo[2*ii + 0]);
- simdgroup_multiply_accumulate(lo[2*ii + 1], vs[1], mv[3], lo[2*ii + 1]);
- }
-
- pv += 2*8*NS20;
- }
- }
- }
-
- {
- auto sot = so + 8*sgitg;
-
- FOR_UNROLL (short ii = 0; ii < NO; ++ii) {
- simdgroup_store(lo[ii], sot, PV, 0, false);
-
- sot += 8*NSG;
- }
- }
- } else {
- // TODO: this is the quantized V cache branch - not optimized yet
-
- const short tx = tiisg%4;
- const short ty = tiisg/4;
-
- for (short cc = 0; cc < C/8; ++cc) {
- s8x8_t vs;
- simdgroup_load(vs, ss + 8*cc, SH, 0, false);
-
- for (short ii = 4*sgitg; ii < DV16; ii += 4*NSG) {
- device const vd4x4_t * pv4x4 = (device const vd4x4_t *) (v + ((ic + 8*cc + ty)*args.nb21));
-
- if (DV16%4 == 0) {
- // no need for bound checks
- {
- v4x4_t tmp;
- deq_v(pv4x4 + (ii + tx)/nl_v, (ii + tx)%nl_v, tmp);
- sv4x4[4*ty + tx] = tmp;
- }
-
- simdgroup_barrier(mem_flags::mem_threadgroup);
-
- FOR_UNROLL (short k = 0; k < 4; ++k) {
- v8x8_t mv[2];
- o8x8_t lo[2];
-
- simdgroup_load(mv[0], sv + 16*k + 0*8, 4*16, 0, false);
- simdgroup_load(mv[1], sv + 16*k + 1*8, 4*16, 0, false);
- simdgroup_load(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
- simdgroup_load(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
-
- simdgroup_multiply_accumulate(lo[0], vs, mv[0], lo[0]);
- simdgroup_multiply_accumulate(lo[1], vs, mv[1], lo[1]);
-
- simdgroup_store(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
- simdgroup_store(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
- }
- } else {
- if (ii + tx < DV16) {
- v4x4_t tmp;
- deq_v(pv4x4 + (ii + tx)/nl_v, (ii + tx)%nl_v, tmp);
- sv4x4[4*ty + tx] = tmp;
- }
-
- simdgroup_barrier(mem_flags::mem_threadgroup);
-
- for (short k = 0; k < 4 && ii + k < DV16; ++k) {
- v8x8_t mv[2];
- o8x8_t lo[2];
-
- simdgroup_load(mv[0], sv + 16*k + 0*8, 4*16, 0, false);
- simdgroup_load(mv[1], sv + 16*k + 1*8, 4*16, 0, false);
- simdgroup_load(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
- simdgroup_load(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
-
- simdgroup_multiply_accumulate(lo[0], vs, mv[0], lo[0]);
- simdgroup_multiply_accumulate(lo[1], vs, mv[1], lo[1]);
-
- simdgroup_store(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
- simdgroup_store(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
- }
- }
- }
- }
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- if (FC_flash_attn_ext_has_sinks) {
- FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
-
- const float m = M[jj];
- const float s = tiisg == 0 ? ((device const float *) sinks)[iq2] : -FLT_MAX/2;
-
- M[jj] = simd_max(max(M[jj], s));
-
- const float ms = exp(m - M[jj]);
- const float vs = exp(s - M[jj]);
-
- S[jj] = S[jj]*ms + simd_sum(vs);
-
- for (short i = tiisg; i < DV4; i += NW) {
- so4[j*PV4 + i] *= ms;
- }
- }
- }
- }
-
- // store to global memory
- for (short jj = 0; jj < NQ; ++jj) {
- const short j = jj*NSG + sgitg;
- if (iq1 + j >= args.ne01) {
- break;
- }
-
- device float4 * dst4 = (device float4 *) dst + ((uint64_t)iq3*args.ne2*args.ne1 + iq2 + (uint64_t)(iq1 + j)*args.ne1)*DV4;
-
- const float scale = S[jj] == 0.0 ? 0.0f : 1.0f/S[jj];
-
- if (DV4 % NW == 0) {
- FOR_UNROLL (short ii = 0; ii < DV4/NW; ++ii) {
- const short i = ii*NW + tiisg;
-
- dst4[i] = (float4) so4[j*PV4 + i]*scale;
- }
- } else {
- for (short i = tiisg; i < DV4; i += NW) {
- dst4[i] = (float4) so4[j*PV4 + i]*scale;
- }
- }
- }
-
-#undef NS10
-#undef NS20
-}
-
-template<
- typename q_t, // query types in shared memory
- typename q4_t,
- typename q8x8_t,
- typename k_t, // key types in shared memory
- typename k4x4_t,
- typename k8x8_t,
- typename v_t, // value types in shared memory
- typename v4x4_t,
- typename v8x8_t,
- typename qk_t, // Q*K types
- typename qk8x8_t,
- typename s_t, // soft-max types
- typename s2_t,
- typename s8x8_t,
- typename o_t, // attention accumulation types
- typename o4_t,
- typename o8x8_t,
- typename kd4x4_t, // key type in device memory
- short nl_k,
- void (*deq_k)(device const kd4x4_t *, short, thread k4x4_t &),
- typename vd4x4_t, // value type in device memory
- short nl_v,
- void (*deq_v)(device const vd4x4_t *, short, thread v4x4_t &),
- short DK, // K head size
- short DV, // V head size
- short Q = OP_FLASH_ATTN_EXT_NQPSG, // queries per threadgroup
- short C = OP_FLASH_ATTN_EXT_NCPSG> // cache items per threadgroup
-kernel void kernel_flash_attn_ext(
- constant ggml_metal_kargs_flash_attn_ext & args,
- device const char * q,
- device const char * k,
- device const char * v,
- device const char * mask,
- device const char * sinks,
- device const char * pad,
- device const char * blk,
- device char * dst,
- threadgroup half * shmem_f16 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-#define FWD_TMPL q_t, q4_t, q8x8_t, k_t, k4x4_t, k8x8_t, v_t, v4x4_t, v8x8_t, qk_t, qk8x8_t, s_t, s2_t, s8x8_t, o_t, o4_t, o8x8_t, kd4x4_t, nl_k, deq_k, vd4x4_t, nl_v, deq_v, DK, DV, Q, C
-#define FWD_ARGS args, q, k, v, mask, sinks, pad, blk, dst, shmem_f16, tgpig, tiisg, sgitg
- switch (FC_flash_attn_ext_nsg) {
- // note: disabled cases to reduce library load time
- //case 1: kernel_flash_attn_ext_impl<FWD_TMPL, 1>(FWD_ARGS); break;
- //case 2: kernel_flash_attn_ext_impl<FWD_TMPL, 2>(FWD_ARGS); break;
- case 4: kernel_flash_attn_ext_impl<FWD_TMPL, 4>(FWD_ARGS); break;
- case 8: kernel_flash_attn_ext_impl<FWD_TMPL, 8>(FWD_ARGS); break;
- }
-#undef FWD_TMPL
-#undef FWD_ARGS
-}
-
-// TODO: this is quite ugly. in the future these types will be hardcoded in the kernel, but for now keep them as
-// template to be able to explore different combinations
-//
-#define FA_TYPES \
- half, half4, simdgroup_half8x8, \
- half, half4x4, simdgroup_half8x8, \
- half, half4x4, simdgroup_half8x8, \
- float, simdgroup_float8x8, \
- float, float2, simdgroup_float8x8, \
- float, float4, simdgroup_float8x8
- //half, half4, simdgroup_half8x8
-
-#define FA_TYPES_BF \
- bfloat, bfloat4, simdgroup_bfloat8x8, \
- bfloat, bfloat4x4, simdgroup_bfloat8x8, \
- bfloat, bfloat4x4, simdgroup_bfloat8x8, \
- float, simdgroup_float8x8, \
- float, float2, simdgroup_float8x8, \
- half, half4, simdgroup_half8x8
- //float, float4, simdgroup_float8x8
-
-#define FA_TYPES_F32 \
- half, half4, simdgroup_half8x8, \
- float, float4x4, simdgroup_float8x8, \
- float, float4x4, simdgroup_float8x8, \
- float, simdgroup_float8x8, \
- float, float2, simdgroup_float8x8, \
- float, float4, simdgroup_float8x8
- //half, half4, simdgroup_half8x8
-
-typedef decltype(kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 64, 64>) flash_attn_ext_t;
-
-template [[host_name("kernel_flash_attn_ext_f32_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_f32_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_f32_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_f32_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_f32_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_f32_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_f32_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_f32_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_f32_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_f32_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_f32_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_f32_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_f32_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_f32_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_f32_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 576, 512>;
-
-template [[host_name("kernel_flash_attn_ext_f16_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_f16_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_f16_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_f16_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_f16_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_f16_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_f16_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_f16_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_f16_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_f16_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_f16_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_f16_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_f16_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_f16_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_f16_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 576, 512>;
-
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_bf16_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_bf16_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 576, 512>;
-#endif
-
-template [[host_name("kernel_flash_attn_ext_q4_0_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_q4_0_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 576, 512>;
-
-template [[host_name("kernel_flash_attn_ext_q4_1_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_q4_1_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 576, 512>;
-
-template [[host_name("kernel_flash_attn_ext_q5_0_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_q5_0_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 576, 512>;
-
-template [[host_name("kernel_flash_attn_ext_q5_1_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_q5_1_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 576, 512>;
-
-template [[host_name("kernel_flash_attn_ext_q8_0_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 32, 32>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 40, 40>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 48, 48>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 64, 64>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 72, 72>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 80, 80>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 96, 96>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 112, 112>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 128, 128>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 192, 192>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 192, 128>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 256, 256>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 320, 256>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 512, 512>;
-template [[host_name("kernel_flash_attn_ext_q8_0_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 576, 512>;
-
-#undef FA_TYPES
-#undef FA_TYPES_BF
-#undef FA_TYPES_F32
-
-constant bool FC_flash_attn_ext_vec_has_mask [[function_constant(FC_FLASH_ATTN_EXT_VEC + 0)]];
-constant bool FC_flash_attn_ext_vec_has_sinks [[function_constant(FC_FLASH_ATTN_EXT_VEC + 1)]];
-constant bool FC_flash_attn_ext_vec_has_bias [[function_constant(FC_FLASH_ATTN_EXT_VEC + 2)]];
-constant bool FC_flash_attn_ext_vec_has_scap [[function_constant(FC_FLASH_ATTN_EXT_VEC + 3)]];
-constant bool FC_flash_attn_ext_vec_has_kvpad [[function_constant(FC_FLASH_ATTN_EXT_VEC + 4)]];
-
-//constant float FC_flash_attn_ext_vec_scale [[function_constant(FC_FLASH_ATTN_EXT_VEC + 10)]];
-//constant float FC_flash_attn_ext_vec_max_bias [[function_constant(FC_FLASH_ATTN_EXT_VEC + 11)]];
-//constant float FC_flash_attn_ext_vec_logit_softcap [[function_constant(FC_FLASH_ATTN_EXT_VEC + 12)]];
-
-constant int32_t FC_flash_attn_ext_vec_ns10 [[function_constant(FC_FLASH_ATTN_EXT_VEC + 20)]];
-constant int32_t FC_flash_attn_ext_vec_ns20 [[function_constant(FC_FLASH_ATTN_EXT_VEC + 21)]];
-constant int32_t FC_flash_attn_ext_vec_nsg [[function_constant(FC_FLASH_ATTN_EXT_VEC + 22)]];
-constant int32_t FC_flash_attn_ext_vec_nwg [[function_constant(FC_FLASH_ATTN_EXT_VEC + 23)]];
-
-template<
- typename q4_t, // query types in shared memory
- typename k4_t, // key types in shared memory
- typename v4_t, // value types in shared memory
- typename qk_t, // Q*K types
- typename s_t, // soft-max types
- typename s4_t,
- typename o4_t, // attention accumulation types
- typename kd4_t, // key type in device memory
- short nl_k,
- void (*deq_k_t4)(device const kd4_t *, short, thread k4_t &),
- typename vd4_t, // value type in device memory
- short nl_v,
- void (*deq_v_t4)(device const vd4_t *, short, thread v4_t &),
- short DK, // K head size
- short DV, // V head size
- short NE = 4, // head elements per thread
- short Q = OP_FLASH_ATTN_EXT_VEC_NQPSG, // queries per threadgroup
- short C = OP_FLASH_ATTN_EXT_VEC_NCPSG> // cache items per threadgroup
-kernel void kernel_flash_attn_ext_vec(
- constant ggml_metal_kargs_flash_attn_ext_vec & args,
- device const char * q,
- device const char * k,
- device const char * v,
- device const char * mask,
- device const char * sinks,
- device const char * pad,
- device char * dst,
- threadgroup half * shmem_f16 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- static_assert(DK % 32 == 0, "DK must be divisible by 32");
- static_assert(DV % 32 == 0, "DV must be divisible by 32");
-
-#define NWG (FC_flash_attn_ext_vec_nwg)
-#define NSG (FC_flash_attn_ext_vec_nsg)
-
-#define NS10 (FC_flash_attn_ext_vec_ns10)
-#define NS20 (FC_flash_attn_ext_vec_ns20)
-
- const short iwg = tgpig[2]%NWG;
-
- const ushort iq3 = tgpig[2]/NWG;
- const ushort iq2 = tgpig[1];
- const ushort iq1 = tgpig[0];
-
- constexpr short DK4 = DK/4;
- constexpr short DV4 = DV/4;
-
- constexpr short PK = PAD2(DK, 128);
- constexpr short PK4 = PK/4;
-
- constexpr short PV = PAD2(DV, 128);
- constexpr short PV4 = PV/4;
-
- constexpr short NW = N_SIMDWIDTH;
- constexpr short NL = NW/NE; // note: this can be adjusted to support different head sizes and simdgroup work loads
- constexpr short SH = 4*C; // shared memory per simdgroup
-
- static_assert(DK4 % NL == 0, "DK4 must be divisible by NL");
- static_assert(DV4 % NL == 0, "DV4 must be divisible by NL");
-
- //const short T = PK + NSG*SH; // shared memory size per query in (half)
-
- //threadgroup q_t * sq = (threadgroup q_t *) (shmem_f16 + 0*PK); // holds the query data
- threadgroup q4_t * sq4 = (threadgroup q4_t *) (shmem_f16 + 0*PK); // same as above but in q4_t
- threadgroup s_t * ss = (threadgroup s_t *) (shmem_f16 + sgitg*SH + NSG*PK); // scratch buffer for attention
- threadgroup s4_t * ss4 = (threadgroup s4_t *) (shmem_f16 + sgitg*SH + NSG*PK); // same as above but in s4_t
- threadgroup half * sm = (threadgroup half *) (shmem_f16 + sgitg*SH + 2*C + NSG*PK); // scratch buffer for mask
- threadgroup o4_t * so4 = (threadgroup o4_t *) (shmem_f16 + 2*sgitg*PV + NSG*PK + NSG*SH); // scratch buffer for the results
-
- // store the result for all queries in shared memory (the O matrix from the paper)
- so4 += tiisg;
-
- {
- q += iq1*args.nb01 + iq2*args.nb02 + iq3*args.nb03;
-
- const short ikv2 = iq2/(args.ne02/args.ne_12_2);
- const short ikv3 = iq3/(args.ne03/args.ne_12_3);
-
- k += ikv2*args.nb12 + ikv3*args.nb13;
- v += ikv2*args.nb22 + ikv3*args.nb23;
- }
-
- // load heads from Q to shared memory
- device const float4 * q4 = (device const float4 *) ((device const char *) q);
-
- if (iq1 < args.ne01) {
- for (short i = tiisg; i < PK4; i += NW) {
- if (i < DK4) {
- sq4[i] = (q4_t) q4[i];
- } else {
- sq4[i] = (q4_t) 0.0f;
- }
- }
- }
-
- // zero out so
- for (short i = 0; i < DV4/NL; ++i) {
- so4[i*NL] = (o4_t) 0.0f;
- }
-
- // zero out shared memory SH
- for (short i = tiisg; i < SH/4; i += NW) {
- ss4[i] = (s4_t) 0.0f;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- {
- float S = 0.0f;
- float M = -FLT_MAX/2;
-
- // thread indices inside the simdgroup
- const short tx = tiisg%NL;
- const short ty = tiisg/NL;
-
- // pointer to the mask
- device const half * pm = (device const half *) (mask + iq1*args.nb31 + (iq2%args.ne32)*args.nb32 + (iq3%args.ne33)*args.nb33);
-
- float slope = 1.0f;
-
- // ALiBi
- if (FC_flash_attn_ext_vec_has_bias) {
- const short h = iq2;
-
- const float base = h < args.n_head_log2 ? args.m0 : args.m1;
- const short exph = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
-
- slope = pow(base, exph);
- }
-
- // loop over the KV cache
- // each simdgroup handles blocks of Q rows and C columns
- for (int ic0 = iwg*NSG + sgitg; ; ic0 += NWG*NSG) {
- int ic = ic0*C;
- if (ic >= args.ne11) {
- break;
- }
-
- // the last partial chunk uses the pad buffer as source
- if (FC_flash_attn_ext_vec_has_kvpad && ic + C > args.ne11) {
- k = pad;
- v = k + args.nb11*C*args.ne_12_2*args.ne_12_3;
- mask = v + args.nb21*C*args.ne_12_2*args.ne_12_3;
-
- const short ikv2 = iq2/(args.ne02/args.ne_12_2);
- const short ikv3 = iq3/(args.ne03/args.ne_12_3);
-
- k += (ikv2 + ikv3*args.ne_12_2)*args.nb11*C;
- v += (ikv2 + ikv3*args.ne_12_2)*args.nb21*C;
-
- if (!FC_flash_attn_ext_vec_has_mask) {
- if (ic + tiisg >= args.ne11) {
- sm[tiisg] = -MAXHALF;
- }
- } else {
- pm = (device const half *) (mask) +
- iq1*C +
- (iq2%args.ne32)*(C*args.ne31) +
- (iq3%args.ne33)*(C*args.ne31*args.ne32);
- }
-
- ic = 0;
- }
-
- if (FC_flash_attn_ext_vec_has_mask) {
- sm[tiisg] = pm[ic + tiisg];
- }
-
- // skip -INF blocks
- if (simd_max(sm[tiisg]) <= -MAXHALF) {
- continue;
- }
-
- // Q*K^T
- {
- device const k4_t * pk4 = (device const k4_t *) (k + ic*args.nb11);
- threadgroup const q4_t * pq4 = sq4;
-
- pk4 += ty*NS10/4 + tx;
- pq4 += tx;
-
- qk_t mqk[C/NE] = { [ 0 ... C/NE - 1] = 0.0f };
-
- // each simdgroup processes 1 query and NE (NW/NL) cache elements
- FOR_UNROLL (short cc = 0; cc < C/NE; ++cc) {
- if (is_same<kd4_t, k4_t>::value) {
- FOR_UNROLL (short ii = 0; ii < DK4/NL; ++ii) {
- mqk[cc] += dot((float4) pk4[cc*NE*NS10/4 + ii*NL], (float4) pq4[ii*NL]);
- }
- } else {
- device const kd4_t * pk = (device const kd4_t *) (k + ((ic + NE*cc + ty)*args.nb11));
-
- k4_t mk;
-
- FOR_UNROLL (short ii = 0; ii < DK4/NL; ++ii) {
- const short i = ii*NL + tx;
-
- deq_k_t4(pk + i/nl_k, i%nl_k, mk);
-
- mqk[cc] += dot((float4) mk, (float4) sq4[i]);
- }
- }
-
- if (NE == 1) {
- mqk[cc] = simd_sum(mqk[cc]);
- } else {
- // simdgroup reduce (NE = 4)
- // [ 0 .. 7] -> [ 0]
- // [ 8 .. 15] -> [ 8]
- // [16 .. 23] -> [16]
- // [24 .. 31] -> [24]
- if (NE <= 1) {
- mqk[cc] += simd_shuffle_down(mqk[cc], 16);
- }
- if (NE <= 2) {
- mqk[cc] += simd_shuffle_down(mqk[cc], 8);
- }
- if (NE <= 4) {
- mqk[cc] += simd_shuffle_down(mqk[cc], 4);
- }
- if (NE <= 8) {
- mqk[cc] += simd_shuffle_down(mqk[cc], 2);
- }
- if (NE <= 16) {
- mqk[cc] += simd_shuffle_down(mqk[cc], 1);
- }
-
- // broadcast
- mqk[cc] = simd_shuffle(mqk[cc], NL*ty);
- }
- }
-
- if (FC_flash_attn_ext_vec_has_mask &&
- !FC_flash_attn_ext_vec_has_scap &&
- !FC_flash_attn_ext_vec_has_bias) {
- ss[NE*tx + ty] = fma(mqk[tx], args.scale, (qk_t) sm[NE*tx + ty]);
- } else {
- mqk[tx] *= args.scale;
-
- if (FC_flash_attn_ext_vec_has_scap) {
- mqk[tx] = args.logit_softcap*precise::tanh(mqk[tx]);
- }
-
- if (FC_flash_attn_ext_vec_has_bias) {
- mqk[tx] += (qk_t) sm[NE*tx + ty]*slope;
- } else {
- mqk[tx] += (qk_t) sm[NE*tx + ty];
- }
-
- ss[NE*tx + ty] = mqk[tx];
- }
- }
-
- simdgroup_barrier(mem_flags::mem_threadgroup);
-
- // online softmax
- {
- const float m = M;
- const float s = ss[tiisg];
-
- M = simd_max(max(M, s));
-
- const float ms = exp(m - M);
- const float vs = exp(s - M);
-
- S = S*ms + simd_sum(vs);
-
- // the P matrix from the paper (Q rows, C columns)
- ss[tiisg] = vs;
-
- // O = diag(ms)*O
- if ((DV4/NL % NW == 0) || ty == 0) {
- FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
- so4[ii*NL] *= ms;
- }
- }
- }
-
- simdgroup_barrier(mem_flags::mem_threadgroup);
-
- // O = O + (Q*K^T)*V
- {
- o4_t lo[DV4/NL];
- FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
- lo[ii] = 0.0f;
- }
-
- if (is_same<vd4_t, v4_t>::value) {
- device const v4_t * pv4 = (device const v4_t *) (v + ic*args.nb21);
-
- pv4 += ty*NS20/4 + tx;
-
- const auto sst = ss + ty;
-
- FOR_UNROLL (short cc = 0; cc < C/NE; ++cc) {
- FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
- lo[ii] += o4_t(float4(pv4[cc*NE*NS20/4 + ii*NL])*float4(sst[cc*NE]));
- }
- }
- } else {
- FOR_UNROLL (short cc = 0; cc < C/NE; ++cc) {
- device const vd4_t * pv4 = (device const vd4_t *) (v + ((ic + NE*cc + ty)*args.nb21));
-
- FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
- const short i = ii*NL + tx;
-
- v4_t mv;
- deq_v_t4(pv4 + i/nl_v, i%nl_v, mv);
-
- lo[ii] += o4_t(float4(mv)*float4(ss[NE*cc + ty]));
- }
- }
- }
-
- FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
- if (NE > 1) {
- lo[ii][0] += simd_shuffle_down(lo[ii][0], 16);
- lo[ii][1] += simd_shuffle_down(lo[ii][1], 16);
- lo[ii][2] += simd_shuffle_down(lo[ii][2], 16);
- lo[ii][3] += simd_shuffle_down(lo[ii][3], 16);
- }
-
- if (NE > 2) {
- lo[ii][0] += simd_shuffle_down(lo[ii][0], 8);
- lo[ii][1] += simd_shuffle_down(lo[ii][1], 8);
- lo[ii][2] += simd_shuffle_down(lo[ii][2], 8);
- lo[ii][3] += simd_shuffle_down(lo[ii][3], 8);
- }
-
- if (NE > 4) {
- lo[ii][0] += simd_shuffle_down(lo[ii][0], 4);
- lo[ii][1] += simd_shuffle_down(lo[ii][1], 4);
- lo[ii][2] += simd_shuffle_down(lo[ii][2], 4);
- lo[ii][3] += simd_shuffle_down(lo[ii][3], 4);
- }
-
- if (NE > 8) {
- lo[ii][0] += simd_shuffle_down(lo[ii][0], 2);
- lo[ii][1] += simd_shuffle_down(lo[ii][1], 2);
- lo[ii][2] += simd_shuffle_down(lo[ii][2], 2);
- lo[ii][3] += simd_shuffle_down(lo[ii][3], 2);
- }
-
- if (NE > 16) {
- lo[ii][0] += simd_shuffle_down(lo[ii][0], 1);
- lo[ii][1] += simd_shuffle_down(lo[ii][1], 1);
- lo[ii][2] += simd_shuffle_down(lo[ii][2], 1);
- lo[ii][3] += simd_shuffle_down(lo[ii][3], 1);
- }
- }
-
- if ((DV4/NL % NW == 0) || ty == 0) {
- FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
- so4[ii*NL] += lo[ii];
- }
- }
- }
- }
-
- if (FC_flash_attn_ext_vec_has_sinks && sgitg == 0 && iwg == 0) {
- const float m = M;
- const float s = tiisg == 0 ? ((device const float *) sinks)[iq2] : -FLT_MAX/2;
-
- M = simd_max(max(M, s));
-
- const float ms = exp(m - M);
- const float vs = exp(s - M);
-
- S = S*ms + simd_sum(vs);
-
- if ((DV4/NL % NW == 0) || ty == 0) {
- FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
- so4[ii*NL] *= ms;
- }
- }
- }
-
- // these are needed for reducing the results from the simdgroups (reuse the ss buffer)
- if (tiisg == 0) {
- ss[0] = (s_t) S;
- ss[1] = (s_t) M;
- }
- }
-
- so4 -= tiisg;
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // parallel reduce
- for (short r = NSG/2; r > 0; r >>= 1) {
- if (sgitg < r) {
- const float S0 = ss[ 0];
- const float S1 = ss[r*(SH/2) + 0];
-
- const float M0 = ss[ 1];
- const float M1 = ss[r*(SH/2) + 1];
-
- const float M = max(M0, M1);
-
- const float ms0 = exp(M0 - M);
- const float ms1 = exp(M1 - M);
-
- const float S = S0*ms0 + S1*ms1;
-
- if (tiisg == 0) {
- ss[0] = S;
- ss[1] = M;
- }
-
- // O_0 = diag(ms0)*O_0 + diag(ms1)*O_1
- for (short i = tiisg; i < DV4; i += NW) {
- so4[i] = so4[i]*ms0 + so4[i + r*PV4]*ms1;
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- // final rescale with 1/S and store to global memory
- if (sgitg == 0) {
- const int64_t nrows = args.ne3*args.ne2*args.ne1;
- const int64_t rid = iq3*args.ne2*args.ne1 + iq2 + iq1*args.ne1;
-
- device float4 * dst4 = (device float4 *) dst;
- device float * dst1 = (device float *) dst + nrows*DV*NWG; // the S and M are stored after the results
-
- const float S = NWG == 1 ? (ss[0] == 0.0f ? 0.0f : 1.0f/ss[0]) : 1.0f;
-
- // interleave the workgroup data
- for (short i = tiisg; i < DV4; i += NW) {
- dst4[rid*DV4*NWG + NWG*i + iwg] = (float4) so4[i]*S;
- }
-
- // store S and M
- if (NWG > 1) {
- if (tiisg == 0) {
- dst1[rid*(2*NWG) + 2*iwg + 0] = ss[0];
- dst1[rid*(2*NWG) + 2*iwg + 1] = ss[1];
- }
- }
- }
-
-#undef NWG
-#undef NSG
-#undef NS10
-#undef NS20
-}
-
-// note: I think the s_t can be half instead of float, because the Q*K scaling is done before storing to shared mem
-// in the other (non-vec) kernel, we need s_t to also be float because we scale during the soft_max
-//
-#define FA_TYPES \
- half4, \
- half4, \
- half4, \
- float, \
- float, float4, \
- float4
-
-#define FA_TYPES_F32 \
- half4, \
- float4, \
- float4, \
- float, \
- float, float4, \
- float4
-
-typedef decltype(kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 128, 128, 4>) flash_attn_ext_vec_t;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 32, 32, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 32, 32, 4>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 32, 32, 4>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 32, 32, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 32, 32, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 32, 32, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 32, 32, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 32, 32, 4>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 64, 64, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 64, 64, 2>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 64, 64, 2>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 64, 64, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 64, 64, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 64, 64, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 64, 64, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 64, 64, 2>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 96, 96, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 96, 96, 4>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 96, 96, 4>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 96, 96, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 96, 96, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 96, 96, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 96, 96, 4>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 96, 96, 4>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 128, 128, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 128, 128, 1>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 128, 128, 1>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 128, 128, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 128, 128, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 128, 128, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 128, 128, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 128, 128, 1>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 192, 192, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 192, 192, 2>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 192, 192, 2>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 192, 192, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 192, 192, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 192, 192, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 192, 192, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 192, 192, 2>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 192, 128, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 192, 128, 2>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 192, 128, 2>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 192, 128, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 192, 128, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 192, 128, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 192, 128, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 192, 128, 2>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 256, 256, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 256, 256, 1>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 256, 256, 1>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 256, 256, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 256, 256, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 256, 256, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 256, 256, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 256, 256, 1>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 320, 256, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 320, 256, 2>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 320, 256, 2>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 320, 256, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 320, 256, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 320, 256, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 320, 256, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 320, 256, 2>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 512, 512, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 512, 512, 1>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 512, 512, 1>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 512, 512, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 512, 512, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 512, 512, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 512, 512, 1>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 512, 512, 1>;
-
-template [[host_name("kernel_flash_attn_ext_vec_f32_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 576, 512, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_f16_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 576, 512, 2>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_flash_attn_ext_vec_bf16_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 576, 512, 2>;
-#endif
-template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 576, 512, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 576, 512, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 576, 512, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 576, 512, 2>;
-template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 576, 512, 2>;
-
-#undef FA_TYPES
-#undef FA_TYPES_F32
-
-constant int32_t FC_flash_attn_ext_vec_reduce_DV [[function_constant(FC_FLASH_ATTN_EXT_VEC_REDUCE + 0)]];
-constant int32_t FC_flash_attn_ext_vec_reduce_NWG [[function_constant(FC_FLASH_ATTN_EXT_VEC_REDUCE + 1)]];
-
-kernel void kernel_flash_attn_ext_vec_reduce(
- constant ggml_metal_kargs_flash_attn_ext_vec_reduce & args,
- device const char * htmp,
- device char * dst,
- uint tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-#define NWG (FC_flash_attn_ext_vec_reduce_NWG)
-#define DV (FC_flash_attn_ext_vec_reduce_DV)
-
- const uint64_t rid = tgpig;
-
- const short iwg = tiisg;
-
- device const float * ss = (device const float *) htmp + (uint64_t)args.nrows*DV*NWG;
-
- float S = ss[rid*(2*NWG) + 2*iwg + 0];
- float M = ss[rid*(2*NWG) + 2*iwg + 1];
-
- const float m = simd_max(M);
- const float ms = exp(M - m);
-
- S = simd_sum(S*ms);
- S = S == 0.0f ? 0.0f : 1.0f/S;
-
- const short DV4 = DV/4;
-
- device const float4 * htmp4 = (device const float4 *) htmp + rid*DV4*NWG;
- device float4 * dst4 = (device float4 *) dst + rid*DV4;
-
- for (short i = sgitg; i < DV4; i += NWG) {
- const float4 v = simd_sum(htmp4[i*NWG + iwg]*ms);
-
- if (iwg == 0) {
- dst4[i] = v*S;
- }
- }
-
-#undef NWG
-#undef DV
-}
-
-template<typename T0, typename T1>
-kernel void kernel_cpy_t_t(
- constant ggml_metal_kargs_cpy & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig[2];
- const int32_t i02 = tgpig[1];
- const int32_t i01 = ntg[1] == 1 ? tgpig[0]%args.ne01 : tgpig[0]*ntg[1] + tpitg.y;
- const int32_t iw0 = ntg[1] == 1 ? tgpig[0]/args.ne01 : 0;
-
- if (i01 >= args.ne01) {
- return;
- }
-
- const int64_t n = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00;
-
- const int32_t i3 = n/(args.ne2*args.ne1*args.ne0);
- const int32_t i2 = (n - i3*args.ne2*args.ne1*args.ne0)/(args.ne1*args.ne0);
- const int32_t i1 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0)/args.ne0;
- const int32_t i0 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0 - i1*args.ne0);
-
- device T1 * dst_data = (device T1 *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- for (int32_t i00 = iw0*ntg[0] + tpitg.x; i00 < args.ne00;) {
- device const T0 * src = (device T0 *)(src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + i00*args.nb00);
- dst_data[i00] = (T1) src[0];
- break;
- }
-}
-
-typedef decltype(kernel_cpy_t_t<float, float>) kernel_cpy_t;
-
-template [[host_name("kernel_cpy_f32_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<float, float>;
-template [[host_name("kernel_cpy_f32_f16")]] kernel kernel_cpy_t kernel_cpy_t_t<float, half>;
-template [[host_name("kernel_cpy_f32_i32")]] kernel kernel_cpy_t kernel_cpy_t_t<float, int32_t>;
-template [[host_name("kernel_cpy_i32_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<int32_t, float>;
-template [[host_name("kernel_cpy_i32_i32")]] kernel kernel_cpy_t kernel_cpy_t_t<int32_t, int32_t>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_cpy_f32_bf16")]] kernel kernel_cpy_t kernel_cpy_t_t<float, bfloat>;
-#endif
-template [[host_name("kernel_cpy_f16_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<half, float>;
-template [[host_name("kernel_cpy_f16_f16")]] kernel kernel_cpy_t kernel_cpy_t_t<half, half>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_cpy_bf16_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<bfloat, float>;
-template [[host_name("kernel_cpy_bf16_bf16")]] kernel kernel_cpy_t kernel_cpy_t_t<bfloat, bfloat>;
-#endif
-
-template<short QK,
- typename block_q,
- void (*quantize_func)(device const float *, device block_q &)>
-kernel void kernel_cpy_f32_q(
- constant ggml_metal_kargs_cpy & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig[2];
- const int32_t i02 = tgpig[1];
- const int32_t i01 = ntg[1] == 1 ? tgpig[0]%args.ne01 : tgpig[0]*ntg[1] + tpitg.y;
- const int32_t iw0 = ntg[1] == 1 ? tgpig[0]/args.ne01 : 0;
-
- if (i01 >= args.ne01) {
- return;
- }
-
- const int64_t n = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00;
-
- const int32_t i3 = n / (args.ne2*args.ne1*args.ne0);
- const int32_t i2 = (n - i3*args.ne2*args.ne1*args.ne0) / (args.ne1*args.ne0);
- const int32_t i1 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0) / args.ne0;
- const int32_t i0 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0 - i1*args.ne0)/QK;
-
- device block_q * dst_data = (device block_q *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- for (int32_t i00 = iw0*ntg[0] + tpitg.x; i00 < args.nk0;) {
- device const float * src = (device const float *)(src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + (i00*QK)*args.nb00);
-
- quantize_func(src, dst_data[i00]);
-
- break;
- }
-}
-
-typedef decltype(kernel_cpy_f32_q<QK8_0, block_q8_0, quantize_q8_0>) cpy_f_q_t;
-
-template [[host_name("kernel_cpy_f32_q8_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK8_0, block_q8_0, quantize_q8_0>;
-template [[host_name("kernel_cpy_f32_q1_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK1_0, block_q1_0, quantize_q1_0>;
-template [[host_name("kernel_cpy_f32_q2_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK2_0, block_q2_0, quantize_q2_0>;
-template [[host_name("kernel_cpy_f32_q4_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK4_0, block_q4_0, quantize_q4_0>;
-template [[host_name("kernel_cpy_f32_q4_1")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK4_1, block_q4_1, quantize_q4_1>;
-template [[host_name("kernel_cpy_f32_q5_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK5_0, block_q5_0, quantize_q5_0>;
-template [[host_name("kernel_cpy_f32_q5_1")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK5_1, block_q5_1, quantize_q5_1>;
-template [[host_name("kernel_cpy_f32_iq4_nl")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK4_NL, block_iq4_nl, quantize_iq4_nl>;
-template [[host_name("kernel_cpy_f32_tq2_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK_K, block_tq2_0, quantize_tq2_0>;
-
-template<typename T4x4, typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread T4x4 &)>
-kernel void kernel_cpy_q_f32(
- constant ggml_metal_kargs_cpy & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig[2];
- const int32_t i02 = tgpig[1];
- const int32_t i01 = ntg[1] == 1 ? tgpig[0]%args.ne01 : tgpig[0]*ntg[1] + tpitg.y;
- const int32_t iw0 = ntg[1] == 1 ? tgpig[0]/args.ne01 : 0;
-
- if (i01 >= args.ne01) {
- return;
- }
-
- const int64_t n = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00;
-
- const int32_t i3 = n/(args.ne2*args.ne1*args.ne0);
- const int32_t i2 = (n - i3*args.ne2*args.ne1*args.ne0)/(args.ne1*args.ne0);
- const int32_t i1 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0)/args.ne0;
- const int32_t i0 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0 - i1*args.ne0);
-
- device const block_q * src_data = (device const block_q *)(src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
- device T4x4 * dst_data = (device T4x4 *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- for (int32_t i00 = iw0*ntg[0] + tpitg.x; i00 < args.nk0;) {
- T4x4 temp;
- dequantize_func(src_data + i00/nl, i00%nl, temp);
- dst_data[i00] = temp;
-
- break;
- }
-}
-
-typedef decltype(kernel_cpy_q_f32<float4x4, block_q4_0, 2, dequantize_q4_0>) cpy_q_f_t;
-
-template [[host_name("kernel_cpy_q1_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q1_0, 8, dequantize_q1_0>;
-template [[host_name("kernel_cpy_q2_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q2_0, 4, dequantize_q2_0>;
-template [[host_name("kernel_cpy_q4_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q4_0, 2, dequantize_q4_0>;
-template [[host_name("kernel_cpy_q4_1_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q4_1, 2, dequantize_q4_1>;
-template [[host_name("kernel_cpy_q5_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q5_0, 2, dequantize_q5_0>;
-template [[host_name("kernel_cpy_q5_1_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q5_1, 2, dequantize_q5_1>;
-template [[host_name("kernel_cpy_q8_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q8_0, 2, dequantize_q8_0>;
-
-template [[host_name("kernel_cpy_tq2_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_tq2_0, QK_NL, dequantize_tq2_0>;
-
-template [[host_name("kernel_cpy_q1_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q1_0, 8, dequantize_q1_0>;
-template [[host_name("kernel_cpy_q2_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q2_0, 4, dequantize_q2_0>;
-template [[host_name("kernel_cpy_q4_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q4_0, 2, dequantize_q4_0>;
-template [[host_name("kernel_cpy_q4_1_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q4_1, 2, dequantize_q4_1>;
-template [[host_name("kernel_cpy_q5_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q5_0, 2, dequantize_q5_0>;
-template [[host_name("kernel_cpy_q5_1_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q5_1, 2, dequantize_q5_1>;
-template [[host_name("kernel_cpy_q8_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q8_0, 2, dequantize_q8_0>;
-
-template [[host_name("kernel_cpy_tq2_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_tq2_0, QK_NL, dequantize_tq2_0>;
-
-template<typename T>
-kernel void kernel_concat(
- constant ggml_metal_kargs_concat & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
-
- const int i3 = tgpig.z;
- const int i2 = tgpig.y;
- const int i1 = ntg.y == 1 ? tgpig.x : tgpig.x*ntg.y + tpitg.y;
-
- if (i1 >= args.ne1) {
- return;
- }
-
- int o[4] = {0, 0, 0, 0};
- o[args.dim] = args.dim == 0 ? args.ne00 : (args.dim == 1 ? args.ne01 : (args.dim == 2 ? args.ne02 : args.ne03));
-
- for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
- device const T * x;
-
- if (i0 < args.ne00 && i1 < args.ne01 && i2 < args.ne02 && i3 < args.ne03) {
- x = (device const T *)(src0 + (i3 )*args.nb03 + (i2 )*args.nb02 + (i1 )*args.nb01 + (i0 )*args.nb00);
- } else {
- x = (device const T *)(src1 + (i3 - o[3])*args.nb13 + (i2 - o[2])*args.nb12 + (i1 - o[1])*args.nb11 + (i0 - o[0])*args.nb10);
- }
-
- device T * y = (device T *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
-
- *y = *x;
- }
-}
-
-typedef decltype(kernel_concat<float>) kernel_concat_t;
-
-template [[host_name("kernel_concat_f32")]] kernel kernel_concat_t kernel_concat<float>;
-template [[host_name("kernel_concat_f16")]] kernel kernel_concat_t kernel_concat<half>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_concat_bf16")]] kernel kernel_concat_t kernel_concat<bfloat>;
-#endif
-template [[host_name("kernel_concat_i8")]] kernel kernel_concat_t kernel_concat<char>;
-template [[host_name("kernel_concat_i16")]] kernel kernel_concat_t kernel_concat<short>;
-template [[host_name("kernel_concat_i32")]] kernel kernel_concat_t kernel_concat<int>;
-template [[host_name("kernel_concat_i64")]] kernel kernel_concat_t kernel_concat<long>;
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_q2_K_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_q2_K * x = (device const block_q2_K *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const short ix = tiisg/8; // 0...3
- const short it = tiisg%8; // 0...7
- const short iq = it/4; // 0 or 1
- const short ir = it%4; // 0...3
- const short is = (8*ir)/16;// 0 or 1
-
- device const float * y4 = y + ix * QK_K + 128 * iq + 8 * ir;
-
- for (int ib = ix; ib < nb; ib += 4) {
- float4 sumy = {0.f, 0.f, 0.f, 0.f};
- for (short i = 0; i < 8; ++i) {
- yl[i+ 0] = y4[i+ 0]; sumy[0] += yl[i+ 0];
- yl[i+ 8] = y4[i+32]; sumy[1] += yl[i+ 8];
- yl[i+16] = y4[i+64]; sumy[2] += yl[i+16];
- yl[i+24] = y4[i+96]; sumy[3] += yl[i+24];
- }
-
- device const uint8_t * sc = (device const uint8_t *)x[ib].scales + 8*iq + is;
- device const uint16_t * qs = (device const uint16_t *)x[ib].qs + 16 * iq + 4 * ir;
- device const half * dh = &x[ib].d;
-
- for (short row = 0; row < nr0; row++) {
- float4 acc1 = {0.f, 0.f, 0.f, 0.f};
- float4 acc2 = {0.f, 0.f, 0.f, 0.f};
- for (int i = 0; i < 8; i += 2) {
- acc1[0] += yl[i+ 0] * (qs[i/2] & 0x0003);
- acc2[0] += yl[i+ 1] * (qs[i/2] & 0x0300);
- acc1[1] += yl[i+ 8] * (qs[i/2] & 0x000c);
- acc2[1] += yl[i+ 9] * (qs[i/2] & 0x0c00);
- acc1[2] += yl[i+16] * (qs[i/2] & 0x0030);
- acc2[2] += yl[i+17] * (qs[i/2] & 0x3000);
- acc1[3] += yl[i+24] * (qs[i/2] & 0x00c0);
- acc2[3] += yl[i+25] * (qs[i/2] & 0xc000);
- }
- float dall = dh[0];
- float dmin = dh[1] * 1.f/16.f;
- sumf[row] += dall * ((acc1[0] + 1.f/256.f * acc2[0]) * (sc[0] & 0xF) * 1.f/ 1.f +
- (acc1[1] + 1.f/256.f * acc2[1]) * (sc[2] & 0xF) * 1.f/ 4.f +
- (acc1[2] + 1.f/256.f * acc2[2]) * (sc[4] & 0xF) * 1.f/16.f +
- (acc1[3] + 1.f/256.f * acc2[3]) * (sc[6] & 0xF) * 1.f/64.f) -
- dmin * (sumy[0] * (sc[0] & 0xF0) + sumy[1] * (sc[2] & 0xF0) + sumy[2] * (sc[4] & 0xF0) + sumy[3] * (sc[6] & 0xF0));
-
- qs += args.nb01/2;
- sc += args.nb01;
- dh += args.nb01/2;
- }
-
- y4 += 4 * QK_K;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_q2_K_f32")]]
-kernel void kernel_mul_mv_q2_K_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_q2_K_f32_impl<N_R0_Q2_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_q3_K_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_q3_K * x = (device const block_q3_K *) (src0 + offset0);
- device const float * yy = (device const float *) (src1 + offset1);
-
- float yl[32];
-
- //const uint16_t kmask1 = 0x3030;
- //const uint16_t kmask2 = 0x0f0f;
-
- const short tid = tiisg/4;
- const short ix = tiisg%4;
- const short ip = tid/4; // 0 or 1
- const short il = 2*((tid%4)/2); // 0 or 2
- const short ir = tid%2;
- const short l0 = 8*ir;
-
- // One would think that the Metal compiler would figure out that ip and il can only have
- // 4 possible states, and optimize accordingly. Well, no. It needs help, and we do it
- // with these two tales.
- //
- // Possible masks for the high bit
- const ushort4 mm[4] = {{0x0001, 0x0100, 0x0002, 0x0200}, // ip = 0, il = 0
- {0x0004, 0x0400, 0x0008, 0x0800}, // ip = 0, il = 2
- {0x0010, 0x1000, 0x0020, 0x2000}, // ip = 1, il = 0
- {0x0040, 0x4000, 0x0080, 0x8000}}; // ip = 1, il = 2
-
- // Possible masks for the low 2 bits
- const int4 qm[2] = {{0x0003, 0x0300, 0x000c, 0x0c00}, {0x0030, 0x3000, 0x00c0, 0xc000}};
-
- const ushort4 hm = mm[2*ip + il/2];
-
- const short shift = 2*il;
-
- const float v1 = il == 0 ? 4.f : 64.f;
- const float v2 = 4.f * v1;
-
- const uint16_t s_shift1 = 4*ip;
- const uint16_t s_shift2 = s_shift1 + il;
-
- const short q_offset = 32*ip + l0;
- const short y_offset = 128*ip + 32*il + l0;
-
- device const float * y1 = yy + ix*QK_K + y_offset;
-
- uint32_t scales32, aux32;
- thread uint16_t * scales16 = (thread uint16_t *)&scales32;
- thread const int8_t * scales = (thread const int8_t *)&scales32;
-
- float sumf1[nr0] = {0.f};
- float sumf2[nr0] = {0.f};
-
- for (int i = ix; i < nb; i += 4) {
- for (short l = 0; l < 8; ++l) {
- yl[l+ 0] = y1[l+ 0];
- yl[l+ 8] = y1[l+16];
- yl[l+16] = y1[l+32];
- yl[l+24] = y1[l+48];
- }
-
- device const uint16_t * q = (device const uint16_t *)(x[i].qs + q_offset);
- device const uint16_t * h = (device const uint16_t *)(x[i].hmask + l0);
- device const uint16_t * a = (device const uint16_t *)(x[i].scales);
- device const half * dh = &x[i].d;
-
- for (short row = 0; row < nr0; ++row) {
- const float d_all = (float)dh[0];
-
- scales16[0] = a[4];
- scales16[1] = a[5];
- aux32 = ((scales32 >> s_shift2) << 4) & 0x30303030;
- scales16[0] = a[il+0];
- scales16[1] = a[il+1];
- scales32 = ((scales32 >> s_shift1) & 0x0f0f0f0f) | aux32;
-
- float s1 = 0, s2 = 0, s3 = 0, s4 = 0, s5 = 0, s6 = 0;
- for (short l = 0; l < 8; l += 2) {
- const int32_t qs = q[l/2];
- s1 += yl[l+0] * (qs & qm[il/2][0]);
- s2 += yl[l+1] * (qs & qm[il/2][1]);
- s3 += ((h[l/2] & hm[0]) ? 0.f : yl[l+0]) + ((h[l/2] & hm[1]) ? 0.f : yl[l+1]);
- s4 += yl[l+16] * (qs & qm[il/2][2]);
- s5 += yl[l+17] * (qs & qm[il/2][3]);
- s6 += ((h[l/2] & hm[2]) ? 0.f : yl[l+16]) + ((h[l/2] & hm[3]) ? 0.f : yl[l+17]);
- }
- float d1 = d_all * (s1 + 1.f/256.f * s2 - s3*v1);
- float d2 = d_all * (s4 + 1.f/256.f * s5 - s6*v2);
- sumf1[row] += d1 * (scales[0] - 32);
- sumf2[row] += d2 * (scales[2] - 32);
-
- s1 = s2 = s3 = s4 = s5 = s6 = 0;
- for (short l = 0; l < 8; l += 2) {
- const int32_t qs = q[l/2+8];
- s1 += yl[l+8] * (qs & qm[il/2][0]);
- s2 += yl[l+9] * (qs & qm[il/2][1]);
- s3 += ((h[l/2+8] & hm[0]) ? 0.f : yl[l+8]) + ((h[l/2+8] & hm[1]) ? 0.f : yl[l+9]);
- s4 += yl[l+24] * (qs & qm[il/2][2]);
- s5 += yl[l+25] * (qs & qm[il/2][3]);
- s6 += ((h[l/2+8] & hm[2]) ? 0.f : yl[l+24]) + ((h[l/2+8] & hm[3]) ? 0.f : yl[l+25]);
- }
- d1 = d_all * (s1 + 1.f/256.f * s2 - s3*v1);
- d2 = d_all * (s4 + 1.f/256.f * s5 - s6*v2);
- sumf1[row] += d1 * (scales[1] - 32);
- sumf2[row] += d2 * (scales[3] - 32);
-
- q += args.nb01/2;
- h += args.nb01/2;
- a += args.nb01/2;
- dh += args.nb01/2;
- }
-
- y1 += 4 * QK_K;
- }
-
- for (int row = 0; row < nr0; ++row) {
- const float sumf = (sumf1[row] + 0.25f * sumf2[row]) / (1 << shift);
- sumf1[row] = simd_sum(sumf);
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- if (tiisg == 0) {
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- dst_f32[first_row + row] = sumf1[row];
- }
- }
-}
-
-[[host_name("kernel_mul_mv_q3_K_f32")]]
-kernel void kernel_mul_mv_q3_K_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_q3_K_f32_impl<N_R0_Q3_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_q4_K_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- constexpr uint16_t kmask1 = 0x3f3f;
- constexpr uint16_t kmask2 = 0x0f0f;
- constexpr uint16_t kmask3 = 0xc0c0;
-
- const short ix = tiisg/8; // 0...3
- const short it = tiisg%8; // 0...7
- const short iq = it/4; // 0 or 1
- const short ir = it%4; // 0...3
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_q4_K * x = (device const block_q4_K *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[16];
- float yh[16];
-
- float sumf[nr0]={0.f};
-
- device const float * y4 = y + ix * QK_K + 64 * iq + 8 * ir;
-
- uint16_t sc16[4];
- thread const uint8_t * sc8 = (thread const uint8_t *)sc16;
-
- for (int ib = ix; ib < nb; ib += 4) {
- float4 sumy = {0.f, 0.f, 0.f, 0.f};
-
- for (short i = 0; i < 8; ++i) {
- yl[i+0] = y4[i+ 0]; sumy[0] += yl[i+0];
- yl[i+8] = y4[i+ 32]; sumy[1] += yl[i+8];
- yh[i+0] = y4[i+128]; sumy[2] += yh[i+0];
- yh[i+8] = y4[i+160]; sumy[3] += yh[i+8];
- }
-
- device const uint16_t * sc = (device const uint16_t *)x[ib].scales + iq;
- device const uint16_t * q1 = (device const uint16_t *)x[ib].qs + 16 * iq + 4 * ir;
- device const half * dh = &x[ib].d;
-
- for (short row = 0; row < nr0; row++) {
- sc16[0] = sc[0] & kmask1;
- sc16[1] = sc[2] & kmask1;
- sc16[2] = ((sc[4] >> 0) & kmask2) | ((sc[0] & kmask3) >> 2);
- sc16[3] = ((sc[4] >> 4) & kmask2) | ((sc[2] & kmask3) >> 2);
-
- device const uint16_t * q2 = q1 + 32;
-
- float4 acc1 = {0.f, 0.f, 0.f, 0.f};
- float4 acc2 = {0.f, 0.f, 0.f, 0.f};
-
- FOR_UNROLL (short i = 0; i < 4; ++i) {
- acc1[0] += yl[2*i + 0] * (q1[i] & 0x000F);
- acc1[1] += yl[2*i + 1] * (q1[i] & 0x0F00);
- acc1[2] += yl[2*i + 8] * (q1[i] & 0x00F0);
- acc1[3] += yl[2*i + 9] * (q1[i] & 0xF000);
- acc2[0] += yh[2*i + 0] * (q2[i] & 0x000F);
- acc2[1] += yh[2*i + 1] * (q2[i] & 0x0F00);
- acc2[2] += yh[2*i + 8] * (q2[i] & 0x00F0);
- acc2[3] += yh[2*i + 9] * (q2[i] & 0xF000);
- }
-
- sumf[row] += dh[0] * ((acc1[0] + 1.f/256.f * acc1[1]) * sc8[0] +
- (acc1[2] + 1.f/256.f * acc1[3]) * sc8[1] * 1.f/16.f +
- (acc2[0] + 1.f/256.f * acc2[1]) * sc8[4] +
- (acc2[2] + 1.f/256.f * acc2[3]) * sc8[5] * 1.f/16.f) -
- dh[1] * (sumy[0] * sc8[2] + sumy[1] * sc8[3] + sumy[2] * sc8[6] + sumy[3] * sc8[7]);
-
- q1 += args.nb01/2;
- sc += args.nb01/2;
- dh += args.nb01/2;
- }
-
- y4 += 4 * QK_K;
- }
-
- device float * dst_f32 = (device float *) dst + (int64_t)im*args.ne0*args.ne1 + (int64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_q4_K_f32")]]
-kernel void kernel_mul_mv_q4_K_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_q4_K_f32_impl<N_R0_Q4_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_q5_K_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_q5_K * x = (device const block_q5_K *) (src0 + offset0);
- device const float * yy = (device const float *) (src1 + offset1);
-
- float sumf[nr0]={0.f};
-
- float yl[16], yh[16];
-
- constexpr uint16_t kmask1 = 0x3f3f;
- constexpr uint16_t kmask2 = 0x0f0f;
- constexpr uint16_t kmask3 = 0xc0c0;
-
- const short tid = tiisg/4;
- const short ix = tiisg%4;
- const short iq = tid/4;
- const short ir = tid%4;
-
- const short l0 = 8*ir;
- const short q_offset = 32*iq + l0;
- const short y_offset = 64*iq + l0;
-
- const uint8_t hm1 = 1u << (2*iq);
- const uint8_t hm2 = hm1 << 1;
- const uint8_t hm3 = hm1 << 4;
- const uint8_t hm4 = hm2 << 4;
-
- uint16_t sc16[4];
- thread const uint8_t * sc8 = (thread const uint8_t *)sc16;
-
- device const float * y1 = yy + ix*QK_K + y_offset;
-
- for (int i = ix; i < nb; i += 4) {
- device const uint8_t * q1 = x[i].qs + q_offset;
- device const uint8_t * qh = x[i].qh + l0;
- device const half * dh = &x[i].d;
- device const uint16_t * a = (device const uint16_t *)x[i].scales + iq;
-
- device const float * y2 = y1 + 128;
- float4 sumy = {0.f, 0.f, 0.f, 0.f};
- for (short l = 0; l < 8; ++l) {
- yl[l+0] = y1[l+ 0]; sumy[0] += yl[l+0];
- yl[l+8] = y1[l+32]; sumy[1] += yl[l+8];
- yh[l+0] = y2[l+ 0]; sumy[2] += yh[l+0];
- yh[l+8] = y2[l+32]; sumy[3] += yh[l+8];
- }
-
- for (short row = 0; row < nr0; ++row) {
- device const uint8_t * q2 = q1 + 64;
-
- sc16[0] = a[0] & kmask1;
- sc16[1] = a[2] & kmask1;
- sc16[2] = ((a[4] >> 0) & kmask2) | ((a[0] & kmask3) >> 2);
- sc16[3] = ((a[4] >> 4) & kmask2) | ((a[2] & kmask3) >> 2);
-
- float4 acc1 = {0.f};
- float4 acc2 = {0.f};
- FOR_UNROLL (short l = 0; l < 8; ++l) {
- uint8_t h = qh[l];
- acc1[0] += yl[l+0] * (q1[l] & 0x0F);
- acc1[1] += yl[l+8] * (q1[l] & 0xF0);
- acc1[2] += yh[l+0] * (q2[l] & 0x0F);
- acc1[3] += yh[l+8] * (q2[l] & 0xF0);
- acc2[0] += h & hm1 ? yl[l+0] : 0.f;
- acc2[1] += h & hm2 ? yl[l+8] : 0.f;
- acc2[2] += h & hm3 ? yh[l+0] : 0.f;
- acc2[3] += h & hm4 ? yh[l+8] : 0.f;
- }
-
- sumf[row] += dh[0] * (sc8[0] * (acc1[0] + 16.f*acc2[0]) +
- sc8[1] * (acc1[1]/16.f + 16.f*acc2[1]) +
- sc8[4] * (acc1[2] + 16.f*acc2[2]) +
- sc8[5] * (acc1[3]/16.f + 16.f*acc2[3])) -
- dh[1] * (sumy[0] * sc8[2] + sumy[1] * sc8[3] + sumy[2] * sc8[6] + sumy[3] * sc8[7]);
-
- q1 += args.nb01;
- qh += args.nb01;
- dh += args.nb01/2;
- a += args.nb01/2;
- }
-
- y1 += 4 * QK_K;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- const float tot = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = tot;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_q5_K_f32")]]
-kernel void kernel_mul_mv_q5_K_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_q5_K_f32_impl<N_R0_Q5_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_q6_K_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- constexpr uint8_t kmask1 = 0x03;
- constexpr uint8_t kmask2 = 0x0C;
- constexpr uint8_t kmask3 = 0x30;
- constexpr uint8_t kmask4 = 0xC0;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_q6_K * x = (device const block_q6_K *) (src0 + offset0);
- device const float * yy = (device const float *) (src1 + offset1);
-
- float sumf[nr0] = { 0.f };
-
- float yl[16];
-
- const short tid = tiisg/2;
- const short ix = tiisg%2;
- const short ip = tid/8; // 0 or 1
- const short il = tid%8;
- const short l0 = 4*il;
- const short is = 8*ip + l0/16;
-
- const short y_offset = 128*ip + l0;
- const short q_offset_l = 64*ip + l0;
- const short q_offset_h = 32*ip + l0;
-
- for (int i = ix; i < nb; i += 2) {
- device const uint8_t * q1 = x[i].ql + q_offset_l;
- device const uint8_t * q2 = q1 + 32;
- device const uint8_t * qh = x[i].qh + q_offset_h;
- device const int8_t * sc = x[i].scales + is;
- device const half * dh = &x[i].d;
-
- device const float * y = yy + i * QK_K + y_offset;
-
- for (short l = 0; l < 4; ++l) {
- yl[4*l + 0] = y[l + 0];
- yl[4*l + 1] = y[l + 32];
- yl[4*l + 2] = y[l + 64];
- yl[4*l + 3] = y[l + 96];
- }
-
- for (short row = 0; row < nr0; ++row) {
- float4 sums = {0.f, 0.f, 0.f, 0.f};
-
- FOR_UNROLL (short l = 0; l < 4; ++l) {
- sums[0] += yl[4*l + 0] * ((int8_t)((q1[l] & 0xF) | ((qh[l] & kmask1) << 4)) - 32);
- sums[1] += yl[4*l + 1] * ((int8_t)((q2[l] & 0xF) | ((qh[l] & kmask2) << 2)) - 32);
- sums[2] += yl[4*l + 2] * ((int8_t)((q1[l] >> 4) | ((qh[l] & kmask3) << 0)) - 32);
- sums[3] += yl[4*l + 3] * ((int8_t)((q2[l] >> 4) | ((qh[l] & kmask4) >> 2)) - 32);
- }
-
- sumf[row] += dh[0] * (sums[0] * sc[0] + sums[1] * sc[2] + sums[2] * sc[4] + sums[3] * sc[6]);
-
- q1 += args.nb01;
- q2 += args.nb01;
- qh += args.nb01;
- sc += args.nb01;
- dh += args.nb01/2;
- }
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_q6_K_f32")]]
-kernel void kernel_mul_mv_q6_K_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_q6_K_f32_impl<N_R0_Q6_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-// ======================= "True" 2-bit
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_iq2_xxs_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq2_xxs * x = (device const block_iq2_xxs *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const int nb32 = nb * (QK_K / 32);
-
- threadgroup uint64_t * svalues = (threadgroup uint64_t *)(shmem);
- threadgroup uint8_t * ssigns = (threadgroup uint8_t *)(svalues + 256);
- {
- int nval = 4;
- int pos = (32*sgitg + tiisg)*nval;
- for (int i = 0; i < nval; ++i) svalues[pos + i] = iq2xxs_grid[pos + i];
- nval = 2;
- pos = (32*sgitg + tiisg)*nval;
- for (int i = 0; i < nval; ++i) ssigns[pos+i] = ksigns_iq2xs[pos+i];
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- const int ix = tiisg;
-
- device const float * y4 = y + 32 * ix;
-
- for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
- for (short i = 0; i < 32; ++i) {
- yl[i] = y4[i];
- }
-
- const int ibl = ib32 / (QK_K / 32);
- const int ib = ib32 % (QK_K / 32);
-
- device const block_iq2_xxs * xr = x + ibl;
- device const uint16_t * q2 = xr->qs + 4 * ib;
- device const half * dh = &xr->d;
-
- for (short row = 0; row < nr0; row++) {
- const float db = dh[0];
- device const uint8_t * aux8 = (device const uint8_t *)q2;
- const uint32_t aux32 = q2[2] | (q2[3] << 16);
- const float d = db * (0.5f + (aux32 >> 28));
-
- float sum = 0;
- for (short l = 0; l < 4; ++l) {
- const threadgroup uint8_t * grid = (const threadgroup uint8_t *)(svalues + aux8[l]);
- const uint8_t signs = ssigns[(aux32 >> 7*l) & 127];
- for (short j = 0; j < 8; ++j) {
- sum += yl[8*l + j] * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
- }
- }
- sumf[row] += d * sum;
-
- dh += args.nb01/2;
- q2 += args.nb01/2;
- }
-
- y4 += 32 * 32;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all * 0.25f;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq2_xxs_f32")]]
-kernel void kernel_mul_mv_iq2_xxs_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- kernel_mul_mv_iq2_xxs_f32_impl<N_R0_IQ2_XXS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_iq2_xs_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq2_xs * x = (device const block_iq2_xs *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const int nb32 = nb * (QK_K / 32);
-
- threadgroup uint64_t * svalues = (threadgroup uint64_t *)(shmem);
- threadgroup uint8_t * ssigns = (threadgroup uint8_t *)(svalues + 512);
- {
- int nval = 8;
- int pos = (32*sgitg + tiisg)*nval;
- for (int i = 0; i < nval; ++i) svalues[pos + i] = iq2xs_grid[pos + i];
- nval = 2;
- pos = (32*sgitg + tiisg)*nval;
- for (int i = 0; i < nval; ++i) ssigns[pos+i] = ksigns_iq2xs[pos+i];
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- const int ix = tiisg;
-
- device const float * y4 = y + 32 * ix;
-
- for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
- for (short i = 0; i < 32; ++i) {
- yl[i] = y4[i];
- }
-
- const int ibl = ib32 / (QK_K / 32);
- const int ib = ib32 % (QK_K / 32);
-
- device const block_iq2_xs * xr = x + ibl;
- device const uint16_t * q2 = xr->qs + 4 * ib;
- device const uint8_t * sc = xr->scales + ib;
- device const half * dh = &xr->d;
-
- for (short row = 0; row < nr0; row++) {
- const float db = dh[0];
- const uint8_t ls1 = sc[0] & 0xf;
- const uint8_t ls2 = sc[0] >> 4;
- const float d1 = db * (0.5f + ls1);
- const float d2 = db * (0.5f + ls2);
-
- float sum1 = 0, sum2 = 0;
- for (short l = 0; l < 2; ++l) {
- const threadgroup uint8_t * grid = (const threadgroup uint8_t *)(svalues + (q2[l] & 511));
- const uint8_t signs = ssigns[(q2[l] >> 9)];
- for (short j = 0; j < 8; ++j) {
- sum1 += yl[8*l + j] * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
- }
- }
- for (short l = 2; l < 4; ++l) {
- const threadgroup uint8_t * grid = (const threadgroup uint8_t *)(svalues + (q2[l] & 511));
- const uint8_t signs = ssigns[(q2[l] >> 9)];
- for (short j = 0; j < 8; ++j) {
- sum2 += yl[8*l + j] * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
- }
- }
- sumf[row] += d1 * sum1 + d2 * sum2;
-
- dh += args.nb01/2;
- q2 += args.nb01/2;
- sc += args.nb01;
- }
-
- y4 += 32 * 32;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all * 0.25f;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq2_xs_f32")]]
-kernel void kernel_mul_mv_iq2_xs_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq2_xs_f32_impl<N_R0_IQ2_XS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_iq3_xxs_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq3_xxs * x = (device const block_iq3_xxs *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const int nb32 = nb * (QK_K / 32);
-
- threadgroup uint32_t * svalues = (threadgroup uint32_t *)(shmem);
- threadgroup uint8_t * ssigns = (threadgroup uint8_t *)(svalues + 256);
- {
- int nval = 4;
- int pos = (32*sgitg + tiisg)*nval;
- for (int i = 0; i < nval; ++i) svalues[pos + i] = iq3xxs_grid[pos + i];
- nval = 2;
- pos = (32*sgitg + tiisg)*nval;
- for (int i = 0; i < nval; ++i) ssigns[pos+i] = ksigns_iq2xs[pos+i];
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- const int ix = tiisg;
-
- device const float * y4 = y + 32 * ix;
-
- for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
- for (short i = 0; i < 32; ++i) {
- yl[i] = y4[i];
- }
-
- const int ibl = ib32 / (QK_K / 32);
- const int ib = ib32 % (QK_K / 32);
-
- device const block_iq3_xxs * xr = x + ibl;
- device const uint8_t * q3 = xr->qs + 8 * ib;
- device const uint16_t * gas = (device const uint16_t *)(xr->qs + QK_K/4) + 2 * ib;
- device const half * dh = &xr->d;
-
- for (short row = 0; row < nr0; row++) {
- const float db = dh[0];
- const uint32_t aux32 = gas[0] | (gas[1] << 16);
- const float d = db * (0.5f + (aux32 >> 28));
-
- float2 sum = {0};
- for (short l = 0; l < 4; ++l) {
- const threadgroup uint8_t * grid1 = (const threadgroup uint8_t *)(svalues + q3[2*l+0]);
- const threadgroup uint8_t * grid2 = (const threadgroup uint8_t *)(svalues + q3[2*l+1]);
- const uint8_t signs = ssigns[(aux32 >> 7*l) & 127];
- for (short j = 0; j < 4; ++j) {
- sum[0] += yl[8*l + j + 0] * grid1[j] * (signs & kmask_iq2xs[j+0] ? -1.f : 1.f);
- sum[1] += yl[8*l + j + 4] * grid2[j] * (signs & kmask_iq2xs[j+4] ? -1.f : 1.f);
- }
- }
- sumf[row] += d * (sum[0] + sum[1]);
-
- dh += args.nb01/2;
- q3 += args.nb01;
- gas += args.nb01/2;
- }
-
- y4 += 32 * 32;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all * 0.5f;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq3_xxs_f32")]]
-kernel void kernel_mul_mv_iq3_xxs_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq3_xxs_f32_impl<N_R0_IQ3_XXS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_iq3_s_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq3_s * x = (device const block_iq3_s *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const int nb32 = nb * (QK_K / 32);
-
- threadgroup uint32_t * svalues = (threadgroup uint32_t *) shmem;
- {
- int nval = 8;
- int pos = (32*sgitg + tiisg)*nval;
- for (int i = 0; i < nval; ++i) svalues[pos + i] = iq3s_grid[pos + i];
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- const int ix = tiisg;
-
- device const float * y4 = y + 32 * ix;
-
- for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
- for (short i = 0; i < 32; ++i) {
- yl[i] = y4[i];
- }
-
- const int ibl = ib32 / (QK_K / 32);
- const int ib = ib32 % (QK_K / 32);
-
- device const block_iq3_s * xr = x + ibl;
- device const uint8_t * qs = xr->qs + 8 * ib;
- device const uint8_t * qh = xr->qh + ib;
- device const uint8_t * sc = xr->scales + (ib/2);
- device const uint8_t * signs = xr->signs + 4 * ib;
- device const half * dh = &xr->d;
-
- for (short row = 0; row < nr0; row++) {
- const float db = dh[0];
- const float d = db * (1 + 2*((sc[0] >> 4*(ib%2)) & 0xf));
-
- float2 sum = {0};
- for (short l = 0; l < 4; ++l) {
- const threadgroup uint32_t * table1 = qh[0] & kmask_iq2xs[2*l+0] ? svalues + 256 : svalues;
- const threadgroup uint32_t * table2 = qh[0] & kmask_iq2xs[2*l+1] ? svalues + 256 : svalues;
- const threadgroup uint8_t * grid1 = (const threadgroup uint8_t *)(table1 + qs[2*l+0]);
- const threadgroup uint8_t * grid2 = (const threadgroup uint8_t *)(table2 + qs[2*l+1]);
- for (short j = 0; j < 4; ++j) {
- sum[0] += yl[8*l + j + 0] * grid1[j] * select(1, -1, signs[l] & kmask_iq2xs[j+0]);
- sum[1] += yl[8*l + j + 4] * grid2[j] * select(1, -1, signs[l] & kmask_iq2xs[j+4]);
- }
- }
- sumf[row] += d * (sum[0] + sum[1]);
-
- dh += args.nb01/2;
- qs += args.nb01;
- qh += args.nb01;
- sc += args.nb01;
- signs += args.nb01;
- }
-
- y4 += 32 * 32;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq3_s_f32")]]
-kernel void kernel_mul_mv_iq3_s_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq3_s_f32_impl<N_R0_IQ3_S, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_iq2_s_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq2_s * x = (device const block_iq2_s *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const int nb32 = nb * (QK_K / 32);
-
- //threadgroup uint64_t * svalues = (threadgroup uint64_t *) shmem;
- //{
- // int nval = 32;
- // int pos = (32*sgitg + tiisg)*nval;
- // for (int i = 0; i < nval; ++i) svalues[pos + i] = iq2s_grid[pos + i];
- // threadgroup_barrier(mem_flags::mem_threadgroup);
- //}
-
- const short ix = tiisg;
-
- device const float * y4 = y + 32 * ix;
-
- for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
- for (short i = 0; i < 32; ++i) {
- yl[i] = y4[i];
- }
-
- const int ibl = ib32 / (QK_K / 32);
- const int ib = ib32 % (QK_K / 32);
-
- device const block_iq2_s * xr = x + ibl;
- device const uint8_t * qs = xr->qs + 4 * ib;
- device const uint8_t * qh = xr->qh + ib;
- device const uint8_t * sc = xr->scales + ib;
- device const uint8_t * signs = qs + QK_K/8;
- device const half * dh = &xr->d;
-
- for (short row = 0; row < nr0; row++) {
- const float db = dh[0];
- const float d1 = db * (0.5f + (sc[0] & 0xf));
- const float d2 = db * (0.5f + (sc[0] >> 4));
-
- float2 sum = {0};
- for (short l = 0; l < 2; ++l) {
- //const threadgroup uint8_t * grid1 = (const threadgroup uint8_t *)(svalues + (qs[l+0] | ((qh[0] << (8-2*l)) & 0x300)));
- //const threadgroup uint8_t * grid2 = (const threadgroup uint8_t *)(svalues + (qs[l+2] | ((qh[0] << (4-2*l)) & 0x300)));
- constant uint8_t * grid1 = (constant uint8_t *)(iq2s_grid + (qs[l+0] | ((qh[0] << (8-2*l)) & 0x300)));
- constant uint8_t * grid2 = (constant uint8_t *)(iq2s_grid + (qs[l+2] | ((qh[0] << (4-2*l)) & 0x300)));
- for (short j = 0; j < 8; ++j) {
- sum[0] += yl[8*l + j + 0] * grid1[j] * select(1, -1, signs[l+0] & kmask_iq2xs[j]);
- sum[1] += yl[8*l + j + 16] * grid2[j] * select(1, -1, signs[l+2] & kmask_iq2xs[j]);
- }
- }
- sumf[row] += d1 * sum[0] + d2 * sum[1];
-
- dh += args.nb01/2;
- qs += args.nb01;
- qh += args.nb01;
- sc += args.nb01;
- signs += args.nb01;
- }
-
- y4 += 32 * 32;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all * 0.25f;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq2_s_f32")]]
-kernel void kernel_mul_mv_iq2_s_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq2_s_f32_impl<N_R0_IQ2_S, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_iq1_s_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq1_s * x = (device const block_iq1_s *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const int nb32 = nb * (QK_K / 32);
-
- const short ix = tiisg;
-
- device const float * y4 = y + 32 * ix;
-
- for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
- float sumy = 0;
- for (short i = 0; i < 32; ++i) {
- yl[i] = y4[i];
- sumy += yl[i];
- }
-
- const int ibl = ib32 / (QK_K / 32);
- const int ib = ib32 % (QK_K / 32);
-
- device const block_iq1_s * xr = x + ibl;
- device const uint8_t * qs = xr->qs + 4 * ib;
- device const uint16_t * qh = xr->qh + ib;
- device const half * dh = &xr->d;
-
- for (short row = 0; row < nr0; row++) {
- constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
- constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 5) & 0x700)));
- constant uint8_t * grid3 = (constant uint8_t *)(iq1s_grid_gpu + (qs[2] | ((qh[0] << 2) & 0x700)));
- constant uint8_t * grid4 = (constant uint8_t *)(iq1s_grid_gpu + (qs[3] | ((qh[0] >> 1) & 0x700)));
-
- float sum = 0;
- for (short j = 0; j < 4; ++j) {
- sum += yl[j+ 0] * (grid1[j] & 0xf) + yl[j+ 4] * (grid1[j] >> 4)
- + yl[j+ 8] * (grid2[j] & 0xf) + yl[j+12] * (grid2[j] >> 4)
- + yl[j+16] * (grid3[j] & 0xf) + yl[j+20] * (grid3[j] >> 4)
- + yl[j+24] * (grid4[j] & 0xf) + yl[j+28] * (grid4[j] >> 4);
- }
- sumf[row] += (float)dh[0] * (sum + sumy * (qh[0] & 0x8000 ? -1 - IQ1S_DELTA : -1 + IQ1S_DELTA)) * (2*((qh[0] >> 12) & 7) + 1);
-
- dh += args.nb01/2;
- qs += args.nb01;
- qh += args.nb01/2;
- }
-
- y4 += 32 * 32;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq1_s_f32")]]
-kernel void kernel_mul_mv_iq1_s_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq1_s_f32_impl<N_R0_IQ1_S, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_iq1_m_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq1_m * x = (device const block_iq1_m *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- float yl[32];
- float sumf[nr0]={0.f};
-
- const int nb32 = nb * (QK_K / 32);
-
- const short ix = tiisg;
-
- device const float * y4 = y + 32 * ix;
-
- iq1m_scale_t scale;
-
- for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
- float4 sumy = {0.f};
- for (short i = 0; i < 8; ++i) {
- yl[i+ 0] = y4[i+ 0]; sumy[0] += yl[i+ 0];
- yl[i+ 8] = y4[i+ 8]; sumy[1] += yl[i+ 8];
- yl[i+16] = y4[i+16]; sumy[2] += yl[i+16];
- yl[i+24] = y4[i+24]; sumy[3] += yl[i+24];
- }
-
- const int ibl = ib32 / (QK_K / 32);
- const int ib = ib32 % (QK_K / 32);
-
- device const block_iq1_m * xr = x + ibl;
- device const uint8_t * qs = xr->qs + 4 * ib;
- device const uint8_t * qh = xr->qh + 2 * ib;
- device const uint16_t * sc = (device const uint16_t *)xr->scales;
-
- for (short row = 0; row < nr0; row++) {
- scale.u16 = (sc[0] >> 12) | ((sc[1] >> 8) & 0x00f0) | ((sc[2] >> 4) & 0x0f00) | (sc[3] & 0xf000);
-
- constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
- constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 4) & 0x700)));
- constant uint8_t * grid3 = (constant uint8_t *)(iq1s_grid_gpu + (qs[2] | ((qh[1] << 8) & 0x700)));
- constant uint8_t * grid4 = (constant uint8_t *)(iq1s_grid_gpu + (qs[3] | ((qh[1] << 4) & 0x700)));
-
- float2 sum = {0.f};
- for (short j = 0; j < 4; ++j) {
- sum[0] += yl[j+ 0] * (grid1[j] & 0xf) + yl[j+ 4] * (grid1[j] >> 4)
- + yl[j+ 8] * (grid2[j] & 0xf) + yl[j+12] * (grid2[j] >> 4);
- sum[1] += yl[j+16] * (grid3[j] & 0xf) + yl[j+20] * (grid3[j] >> 4)
- + yl[j+24] * (grid4[j] & 0xf) + yl[j+28] * (grid4[j] >> 4);
- }
- const float delta1 = sumy[0] * (qh[0] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA) + sumy[1] * (qh[0] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
- const float delta2 = sumy[2] * (qh[1] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA) + sumy[3] * (qh[1] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
-
- sumf[row] += (float)scale.f16 * ((sum[0] + delta1) * (2*((sc[ib/2] >> (6*(ib%2)+0)) & 7) + 1) +
- (sum[1] + delta2) * (2*((sc[ib/2] >> (6*(ib%2)+3)) & 7) + 1));
-
- sc += args.nb01/2;
- qs += args.nb01;
- qh += args.nb01;
- }
-
- y4 += 32 * 32;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq1_m_f32")]]
-kernel void kernel_mul_mv_iq1_m_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq1_m_f32_impl<N_R0_IQ1_M, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<int NR0, typename args_t>
-void kernel_mul_mv_iq4_nl_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- threadgroup float * shmem_f32 = (threadgroup float *) shmem;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * NR0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq4_nl * x = (device const block_iq4_nl *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- const int nb = args.ne00/QK4_NL;
- const int ns01 = args.nb01/args.nb00;
-
- const short ix = tiisg/2; // 0...15
- const short it = tiisg%2; // 0 or 1
-
- shmem_f32[tiisg] = kvalues_iq4nl_f[tiisg%16];
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- float4 yl[4];
- float sumf[NR0]={0.f};
-
- device const float * yb = y + ix*QK4_NL + it*8;
-
- uint32_t aux32[2];
- thread const uint8_t * q8 = (thread const uint8_t *)aux32;
-
- float4 qf1, qf2;
-
- // [TAG_MUL_MV_WEIRD]
- for (int ib = ix; ib < nb && ib < ns01; ib += 16) {
- device const float4 * y4 = (device const float4 *)yb;
- yl[0] = y4[0];
- yl[1] = y4[4];
- yl[2] = y4[1];
- yl[3] = y4[5];
-
- for (short row = 0; row < NR0; row++) {
- device const block_iq4_nl & xb = x[row*ns01 + ib];
- device const uint16_t * q4 = (device const uint16_t *)(xb.qs + 8*it);
-
- float4 acc1 = {0.f}, acc2 = {0.f};
-
- aux32[0] = q4[0] | (q4[1] << 16);
- aux32[1] = (aux32[0] >> 4) & 0x0f0f0f0f;
- aux32[0] &= 0x0f0f0f0f;
- qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
- qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
- acc1 += yl[0] * qf1;
- acc2 += yl[1] * qf2;
-
- aux32[0] = q4[2] | (q4[3] << 16);
- aux32[1] = (aux32[0] >> 4) & 0x0f0f0f0f;
- aux32[0] &= 0x0f0f0f0f;
- qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
- qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
- acc1 += yl[2] * qf1;
- acc2 += yl[3] * qf2;
-
- acc1 += acc2;
-
- sumf[row] += (float)xb.d * (acc1[0] + acc1[1] + acc1[2] + acc1[3]);
- }
-
- yb += 16 * QK4_NL;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < NR0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq4_nl_f32")]]
-kernel void kernel_mul_mv_iq4_nl_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq4_nl_f32_impl<N_R0_IQ4_NL, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int NR0, typename args_t>
-void kernel_mul_mv_iq4_xs_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- threadgroup float * shmem_f32 = (threadgroup float *) shmem;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
- const int first_row = (r0 * NSG + sgitg) * NR0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_iq4_xs * x = (device const block_iq4_xs *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- const int nb = args.ne00/QK_K;
- const int ns01 = args.nb01/args.nb00;
-
- const short ix = tiisg/16; // 0 or 1
- const short it = tiisg%16; // 0...15
- const short ib = it/2;
- const short il = it%2;
-
- shmem_f32[tiisg] = kvalues_iq4nl_f[tiisg%16];
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- float4 yl[4];
- float sumf[NR0]={0.f};
-
- device const float * yb = y + ix * QK_K + ib * 32 + il * 8;
-
- uint32_t aux32[2];
- thread const uint8_t * q8 = (thread const uint8_t *)aux32;
-
- float4 qf1, qf2;
-
- // [TAG_MUL_MV_WEIRD]
- for (int ibl = ix; ibl < nb && ibl < ns01; ibl += 2) {
- device const float4 * y4 = (device const float4 *)yb;
- yl[0] = y4[0];
- yl[1] = y4[4];
- yl[2] = y4[1];
- yl[3] = y4[5];
-
- for (short row = 0; row < NR0; ++row) {
- device const block_iq4_xs & xb = x[row*ns01 + ibl];
- device const uint32_t * q4 = (device const uint32_t *)(xb.qs + 16*ib + 8*il);
-
- float4 acc1 = {0.f}, acc2 = {0.f};
-
- aux32[0] = (q4[0] ) & 0x0f0f0f0f;
- aux32[1] = (q4[0] >> 4) & 0x0f0f0f0f;
- qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
- qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
- acc1 += yl[0] * qf1;
- acc2 += yl[1] * qf2;
-
- aux32[0] = (q4[1] ) & 0x0f0f0f0f;
- aux32[1] = (q4[1] >> 4) & 0x0f0f0f0f;
- qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
- qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
- acc1 += yl[2] * qf1;
- acc2 += yl[3] * qf2;
-
- acc1 += acc2;
-
- const int ls = (((xb.scales_l[ib/2] >> 4*(ib%2)) & 0xf) | (((xb.scales_h >> 2*ib) & 3) << 4)) - 32;
- sumf[row] += (float)xb.d * ls * (acc1[0] + acc1[1] + acc1[2] + acc1[3]);
- }
-
- yb += 2 * QK_K;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < NR0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_iq4_xs_f32")]]
-kernel void kernel_mul_mv_iq4_xs_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_iq4_xs_f32_impl<N_R0_IQ4_XS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int NR0, typename args_t>
-void kernel_mul_mv_mxfp4_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- threadgroup float * shmem_f32 = (threadgroup float *) shmem;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * NR0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const block_mxfp4 * x = (device const block_mxfp4 *) (src0 + offset0);
- device const float * y = (device const float *) (src1 + offset1);
-
- const int nb = args.ne00/QK_MXFP4;
- const int ns01 = args.nb01/args.nb00; // this can be larger than nb for permuted src0 tensors
-
- const short ix = tiisg/2; // 0...15
- const short it = tiisg%2; // 0 or 1
-
- shmem_f32[tiisg] = kvalues_mxfp4_f[tiisg%16];
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- float4 yl[4];
- float sumf[NR0]={0.f};
-
- device const float * yb = y + ix*QK_MXFP4 + it*8;
-
- // note: just the check `ib < nb` is enough, but adding the redundant `&& ib < ns01` check makes the kernel a bit faster
- // no idea why that is - needs some deeper investigation [TAG_MUL_MV_WEIRD]
- for (int ib = ix; ib < nb && ib < ns01; ib += 16) {
- device const float4 * y4 = (device const float4 *) yb;
-
- yl[0] = y4[0];
- yl[1] = y4[4];
- yl[2] = y4[1];
- yl[3] = y4[5];
-
- FOR_UNROLL (short row = 0; row < NR0; row++) {
- device const block_mxfp4 & xb = x[row*ns01 + ib];
- device const uint8_t * q2 = (device const uint8_t *)(xb.qs + 8*it);
-
- float4 acc1 = yl[0]*float4(shmem_f32[q2[0] & 0x0F], shmem_f32[q2[1] & 0x0F], shmem_f32[q2[2] & 0x0F], shmem_f32[q2[3] & 0x0F]);
- float4 acc2 = yl[1]*float4(shmem_f32[q2[0] >> 4 ], shmem_f32[q2[1] >> 4 ], shmem_f32[q2[2] >> 4 ], shmem_f32[q2[3] >> 4 ]);
- float4 acc3 = yl[2]*float4(shmem_f32[q2[4] & 0x0F], shmem_f32[q2[5] & 0x0F], shmem_f32[q2[6] & 0x0F], shmem_f32[q2[7] & 0x0F]);
- float4 acc4 = yl[3]*float4(shmem_f32[q2[4] >> 4 ], shmem_f32[q2[5] >> 4 ], shmem_f32[q2[6] >> 4 ], shmem_f32[q2[7] >> 4 ]);
-
- acc1 = (acc1 + acc3) + (acc2 + acc4);
-
- sumf[row] += e8m0_to_fp32(xb.e) * ((acc1[0] + acc1[1]) + (acc1[2] + acc1[3]));
- }
-
- yb += 16 * QK_MXFP4;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < NR0 && first_row + row < args.ne0; ++row) {
- float sum_all = simd_sum(sumf[row]);
- if (tiisg == 0) {
- dst_f32[first_row + row] = sum_all;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_mxfp4_f32")]]
-kernel void kernel_mul_mv_mxfp4_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_mxfp4_f32_impl<N_R0_MXFP4, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-template<int nr0, typename args_t>
-void kernel_mul_mv_tq2_0_f32_impl(
- args_t args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg) {
- const short NSG = FC_mul_mv_nsg;
-
- const int nb = args.ne00/QK_K;
-
- const int r0 = tgpig.x;
- const int r1 = tgpig.y;
- const int im = tgpig.z;
-
- const int first_row = (r0 * NSG + sgitg) * nr0;
-
- const uint i12 = im%FC_mul_mv_ne12;
- const uint i13 = im/FC_mul_mv_ne12;
-
- const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
-
- device const float * y = (device const float *) (src1 + offset1);
-
- device const block_tq2_0 * ax[nr0];
- for (int row = 0; row < nr0; ++row) {
- const uint64_t offset0 = (first_row + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
- ax[row] = (device const block_tq2_0 *) ((device char *) src0 + offset0);
- }
-
- float sumf[nr0] = {0.f};
-
- // 8 threads per block, NBLOCK blocks per pass, 2 halves per block per pass
- constexpr short NBLOCK = 4;
-
- constexpr short NB = N_SIMDWIDTH/NBLOCK; // threads per block
-
- const short blk = tiisg / NB; // 0..NBLOCK-1, block handled by this thread
- const short htg = tiisg % NB; // 0..NB-1, thread within block (0..7)
-
- // byte and y base offsets within the block (32 elements per thread, 4 per byte)
- device const float4 * yb4 = (device const float4 *)(y + 4*htg + blk*QK_K);
-
- // hoisted per-byte coefficients (from y) and total y-sum, shared across rows
- // ref: https://github.com/ggml-org/llama.cpp/pull/26980
- float4 coef[4];
-
- for (int ib = blk; ib < nb; ib += NBLOCK) {
- FOR_UNROLL (short h0 = 0; h0 < 2; ++h0) {
- const float4 y0 = yb4[ 0 + 32*h0];
- const float4 y1 = yb4[ 8 + 32*h0];
- const float4 y2 = yb4[16 + 32*h0];
- const float4 y3 = yb4[24 + 32*h0];
-
- float sumy = 0.f;
- FOR_UNROLL (short j = 0; j < 4; ++j) {
- coef[j] = float4(
- y0[j],
- y1[j] - 4.0f*y0[j],
- y2[j] - 4.0f*y1[j],
- y3[j] - 4.0f*y2[j]);
-
- sumy += (y0[j] + y1[j]) + (y2[j] + y3[j]);
- }
-
- FOR_UNROLL (short row = 0; row < nr0; ++row) {
- device const block_tq2_0 & xb = ax[row][ib];
- device const uchar * qs = xb.qs + 4*htg + 32*h0;
-
- float sum = -sumy;
- FOR_UNROLL (short j = 0; j < 4; ++j) {
- // express the 2-bit field shifts (v>>2, v>>4, v>>6) as float floor ops
- const float v = (float)qs[j];
-
- const float f0 = v;
- const float f1 = floor(v*0.25f); // v>>2
- const float f2 = floor(v*0.0625); // v>>4
- const float f3 = floor(v*0.015625); // v>>6
-
- sum += coef[j][0]*f0 + coef[j][1]*f1 + coef[j][2]*f2 + coef[j][3]*f3;
- }
-
- sumf[row] += xb.d * sum;
- }
- }
-
- yb4 += QK_K * NBLOCK / 4;
- }
-
- device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
-
- for (int row = 0; row < nr0; ++row) {
- const float tot = simd_sum(sumf[row]);
- if (tiisg == 0 && first_row + row < args.ne01) {
- dst_f32[first_row + row] = tot;
- }
- }
-}
-
-[[host_name("kernel_mul_mv_tq2_0_f32")]]
-kernel void kernel_mul_mv_tq2_0_f32(
- constant ggml_metal_kargs_mul_mv & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- kernel_mul_mv_tq2_0_f32_impl<N_R0_TQ2_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
-}
-
-template<typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread float4x4 &)>
-kernel void kernel_get_rows_q(
- constant ggml_metal_kargs_get_rows & args,
- device const void * src0,
- device const void * src1,
- device void * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort3 ntg [[threads_per_threadgroup]]) {
- const int32_t iw0 = tgpig.x/args.ne10;
- const int32_t i10 = tgpig.x%args.ne10;
- const int32_t i11 = tgpig.y;
- const int32_t i12 = tgpig.z;
-
- const int32_t r = ((const device int32_t *) ((const device char *) src1 + i12*args.nb12 + i11*args.nb11 + i10*args.nb10))[0];
-
- const int32_t i02 = i11;
- const int32_t i03 = i12;
-
- auto psrc = (device const block_q *) ((const device char *) src0 + i03*args.nb03 + i02*args.nb02 + r*args.nb01);
- auto pdst = (device float4x4 *) (( device char *) dst + i12*args.nb3 + i11*args.nb2 + i10*args.nb1);
-
- for (int ind = iw0*ntg.x + tiitg; ind < args.ne00t;) {
- float4x4 temp;
- dequantize_func(psrc + ind/nl, ind%nl, temp);
- pdst[ind] = temp;
-
- break;
- }
-}
-
-template<typename T0, typename T>
-kernel void kernel_get_rows_f(
- constant ggml_metal_kargs_get_rows & args,
- device const void * src0,
- device const void * src1,
- device void * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort3 ntg [[threads_per_threadgroup]]) {
- const int32_t iw0 = tgpig.x/args.ne10;
- const int32_t i10 = tgpig.x%args.ne10;
- const int32_t i11 = tgpig.y;
- const int32_t i12 = tgpig.z;
-
- const int32_t r = ((const device int32_t *) ((const device char *) src1 + i12*args.nb12 + i11*args.nb11 + i10*args.nb10))[0];
-
- const int32_t i02 = i11;
- const int32_t i03 = i12;
-
- auto psrc = (const device T0 *) ((const device char *) src0 + i03*args.nb03 + i02*args.nb02 + r*args.nb01);
- auto pdst = ( device T *) (( device char *) dst + i12*args.nb3 + i11*args.nb2 + i10*args.nb1);
-
- for (int ind = iw0*ntg.x + tiitg; ind < args.ne00t;) {
- pdst[ind] = psrc[ind];
-
- break;
- }
-}
-
-typedef decltype(kernel_get_rows_f<float, float>) get_rows_f_t;
-
-template [[host_name("kernel_get_rows_f32")]] kernel get_rows_f_t kernel_get_rows_f<float, float>;
-template [[host_name("kernel_get_rows_f16")]] kernel get_rows_f_t kernel_get_rows_f<half, float>;
-template [[host_name("kernel_get_rows_i32")]] kernel get_rows_f_t kernel_get_rows_f<int32_t, int32_t>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_get_rows_bf16")]] kernel get_rows_f_t kernel_get_rows_f<bfloat, float>;
-#endif
-
-typedef decltype(kernel_get_rows_q<block_q4_0, 2, dequantize_q4_0>) get_rows_q_t;
-
-template [[host_name("kernel_get_rows_q1_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q1_0, 8, dequantize_q1_0>;
-template [[host_name("kernel_get_rows_q2_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q2_0, 4, dequantize_q2_0>;
-template [[host_name("kernel_get_rows_q4_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q4_0, 2, dequantize_q4_0>;
-template [[host_name("kernel_get_rows_q4_1")]] kernel get_rows_q_t kernel_get_rows_q<block_q4_1, 2, dequantize_q4_1>;
-template [[host_name("kernel_get_rows_q5_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q5_0, 2, dequantize_q5_0>;
-template [[host_name("kernel_get_rows_q5_1")]] kernel get_rows_q_t kernel_get_rows_q<block_q5_1, 2, dequantize_q5_1>;
-template [[host_name("kernel_get_rows_q8_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q8_0, 2, dequantize_q8_0>;
-template [[host_name("kernel_get_rows_mxfp4")]] kernel get_rows_q_t kernel_get_rows_q<block_mxfp4, 2, dequantize_mxfp4>;
-template [[host_name("kernel_get_rows_q2_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q2_K, QK_NL, dequantize_q2_K>;
-template [[host_name("kernel_get_rows_q3_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q3_K, QK_NL, dequantize_q3_K>;
-template [[host_name("kernel_get_rows_q4_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q4_K, QK_NL, dequantize_q4_K>;
-template [[host_name("kernel_get_rows_q5_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q5_K, QK_NL, dequantize_q5_K>;
-template [[host_name("kernel_get_rows_q6_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q6_K, QK_NL, dequantize_q6_K>;
-template [[host_name("kernel_get_rows_iq2_xxs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq2_xxs, QK_NL, dequantize_iq2_xxs>;
-template [[host_name("kernel_get_rows_iq2_xs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq2_xs, QK_NL, dequantize_iq2_xs>;
-template [[host_name("kernel_get_rows_iq3_xxs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq3_xxs, QK_NL, dequantize_iq3_xxs>;
-template [[host_name("kernel_get_rows_iq3_s")]] kernel get_rows_q_t kernel_get_rows_q<block_iq3_s, QK_NL, dequantize_iq3_s>;
-template [[host_name("kernel_get_rows_iq2_s")]] kernel get_rows_q_t kernel_get_rows_q<block_iq2_s, QK_NL, dequantize_iq2_s>;
-template [[host_name("kernel_get_rows_iq1_s")]] kernel get_rows_q_t kernel_get_rows_q<block_iq1_s, QK_NL, dequantize_iq1_s>;
-template [[host_name("kernel_get_rows_iq1_m")]] kernel get_rows_q_t kernel_get_rows_q<block_iq1_m, QK_NL, dequantize_iq1_m>;
-template [[host_name("kernel_get_rows_iq4_nl")]] kernel get_rows_q_t kernel_get_rows_q<block_iq4_nl, 2, dequantize_iq4_nl>;
-template [[host_name("kernel_get_rows_iq4_xs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq4_xs, QK_NL, dequantize_iq4_xs>;
-template [[host_name("kernel_get_rows_tq2_0")]] kernel get_rows_q_t kernel_get_rows_q<block_tq2_0, QK_NL, dequantize_tq2_0>;
-
-template<typename TS, typename TI, short QK, typename block_q, void (*quantize_func)(device const float *, device block_q &)>
-kernel void kernel_set_rows_q(
- constant ggml_metal_kargs_set_rows & args,
- device const void * src0,
- device const void * src1,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint tiitg[[thread_index_in_threadgroup]],
- uint3 tptg [[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig.z;
- const int32_t i02 = tgpig.y;
-
- const int32_t i12 = i03%args.ne12;
- const int32_t i11 = i02%args.ne11;
-
- const int32_t i01 = tgpig.x*tptg.y + tiitg/tptg.x;
- if (i01 >= args.ne01) {
- return;
- }
-
- const int32_t i10 = i01;
- const TI i1 = ((const device TI *) ((const device char *) src1 + i10*args.nb10 + i11*args.nb11 + i12*args.nb12))[0];
-
- device block_q * dst_row = ( device block_q *) (( device char *) dst + i1*args.nb1 + i02*args.nb2 + i03*args.nb3);
- const device TS * src_row = (const device TS *) ((const device char *) src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
-
- for (int ind = tiitg%tptg.x; ind < args.nk0; ind += tptg.x) {
- quantize_func(src_row + QK*ind, dst_row[ind]);
- }
-}
-
-template<typename TS, typename TI, typename block_q, void (*quantize_func)(device const float *, device block_q &)>
-kernel void kernel_set_rows_q32(
- constant ggml_metal_kargs_set_rows & args,
- device const void * src0,
- device const void * src1,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint tiitg[[thread_index_in_threadgroup]],
- uint3 tptg [[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig.z;
- const int32_t i02 = tgpig.y;
-
- const int32_t i12 = i03%args.ne12;
- const int32_t i11 = i02%args.ne11;
-
- const int32_t i01 = tgpig.x*tptg.y + tiitg/tptg.x;
- if (i01 >= args.ne01) {
- return;
- }
-
- const int32_t i10 = i01;
- const TI i1 = ((const device TI *) ((const device char *) src1 + i10*args.nb10 + i11*args.nb11 + i12*args.nb12))[0];
-
- device block_q * dst_row = ( device block_q *) (( device char *) dst + i1*args.nb1 + i02*args.nb2 + i03*args.nb3);
- const device TS * src_row = (const device TS *) ((const device char *) src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
-
- for (int ind = tiitg%tptg.x; ind < args.nk0; ind += tptg.x) {
- quantize_func(src_row + 32*ind, dst_row[ind]);
- }
-}
-
-template<typename TS, typename TI, typename TD>
-kernel void kernel_set_rows_f(
- constant ggml_metal_kargs_set_rows & args,
- device const void * src0,
- device const void * src1,
- device float * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- uint tiitg[[thread_index_in_threadgroup]],
- uint3 tptg [[threads_per_threadgroup]]) {
- const int32_t i03 = tgpig.z;
- const int32_t i02 = tgpig.y;
-
- const int32_t i12 = i03%args.ne12;
- const int32_t i11 = i02%args.ne11;
-
- const int32_t i01 = tgpig.x*tptg.y + tiitg/tptg.x;
- if (i01 >= args.ne01) {
- return;
- }
-
- const int32_t i10 = i01;
- const TI i1 = ((const device TI *) ((const device char *) src1 + i10*args.nb10 + i11*args.nb11 + i12*args.nb12))[0];
-
- device TD * dst_row = ( device TD *) (( device char *) dst + i1*args.nb1 + i02*args.nb2 + i03*args.nb3);
- const device TS * src_row = (const device TS *) ((const device char *) src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
-
- for (int ind = tiitg%tptg.x; ind < args.nk0; ind += tptg.x) {
- dst_row[ind] = (TD) src_row[ind];
- }
-}
-
-typedef decltype(kernel_set_rows_f<float, int64_t, float>) set_rows_f_t;
-
-template [[host_name("kernel_set_rows_f32_i64_f32")]] kernel set_rows_f_t kernel_set_rows_f<float, int64_t, float>;
-template [[host_name("kernel_set_rows_f32_i32_f32")]] kernel set_rows_f_t kernel_set_rows_f<float, int32_t, float>;
-template [[host_name("kernel_set_rows_f32_i64_f16")]] kernel set_rows_f_t kernel_set_rows_f<float, int64_t, half>;
-template [[host_name("kernel_set_rows_f32_i32_f16")]] kernel set_rows_f_t kernel_set_rows_f<float, int32_t, half>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_set_rows_f32_i64_bf16")]] kernel set_rows_f_t kernel_set_rows_f<float, int64_t, bfloat>;
-template [[host_name("kernel_set_rows_f32_i32_bf16")]] kernel set_rows_f_t kernel_set_rows_f<float, int32_t, bfloat>;
-#endif
-
-template [[host_name("kernel_set_rows_f16_i64_f16")]] kernel set_rows_f_t kernel_set_rows_f<half, int64_t, half>;
-template [[host_name("kernel_set_rows_f16_i32_f16")]] kernel set_rows_f_t kernel_set_rows_f<half, int32_t, half>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_set_rows_bf16_i64_bf16")]] kernel set_rows_f_t kernel_set_rows_f<bfloat, int64_t, bfloat>;
-template [[host_name("kernel_set_rows_bf16_i32_bf16")]] kernel set_rows_f_t kernel_set_rows_f<bfloat, int32_t, bfloat>;
-#endif
-
-typedef decltype(kernel_set_rows_q32<float, int64_t, block_q8_0, quantize_q8_0>) set_rows_q32_t;
-
-template [[host_name("kernel_set_rows_f32_i64_q8_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q8_0, quantize_q8_0>;
-template [[host_name("kernel_set_rows_f32_i32_q8_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q8_0, quantize_q8_0>;
-template [[host_name("kernel_set_rows_f32_i64_q4_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q4_0, quantize_q4_0>;
-template [[host_name("kernel_set_rows_f32_i32_q4_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q4_0, quantize_q4_0>;
-template [[host_name("kernel_set_rows_f32_i64_q4_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q4_1, quantize_q4_1>;
-template [[host_name("kernel_set_rows_f32_i32_q4_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q4_1, quantize_q4_1>;
-template [[host_name("kernel_set_rows_f32_i64_q5_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q5_0, quantize_q5_0>;
-template [[host_name("kernel_set_rows_f32_i32_q5_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q5_0, quantize_q5_0>;
-template [[host_name("kernel_set_rows_f32_i64_q5_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q5_1, quantize_q5_1>;
-template [[host_name("kernel_set_rows_f32_i32_q5_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q5_1, quantize_q5_1>;
-template [[host_name("kernel_set_rows_f32_i64_iq4_nl")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_iq4_nl, quantize_iq4_nl>;
-template [[host_name("kernel_set_rows_f32_i32_iq4_nl")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_iq4_nl, quantize_iq4_nl>;
-
-typedef decltype(kernel_set_rows_q<float, int64_t, QK_K, block_tq2_0, quantize_tq2_0>) set_rows_qK_t;
-
-template [[host_name("kernel_set_rows_f32_i64_tq2_0")]] kernel set_rows_qK_t kernel_set_rows_q<float, int64_t, QK_K, block_tq2_0, quantize_tq2_0>;
-template [[host_name("kernel_set_rows_f32_i32_tq2_0")]] kernel set_rows_qK_t kernel_set_rows_q<float, int32_t, QK_K, block_tq2_0, quantize_tq2_0>;
-
-kernel void kernel_diag_f32(
- constant ggml_metal_kargs_diag & args,
- device const char * src0,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]]) {
- constexpr short NW = N_SIMDWIDTH;
-
- const int32_t i3 = tgpig.z;
- const int32_t i2 = tgpig.y;
- const int32_t i1 = tgpig.x;
-
- device const float * src0_ptr = (device const float *)(src0 + i2*args.nb02 + i3*args.nb03);
- device float * dst_ptr = (device float *)(dst + i1*args.nb01 + i2*args.nb2 + i3*args.nb3);
-
- for (int i0 = tiitg; i0 < args.ne0; i0 += NW) {
- dst_ptr[i0] = i0 == i1 ? src0_ptr[i0] : 0.0f;
- }
-}
-
-constant bool FC_mul_mm_bc_inp [[function_constant(FC_MUL_MM + 0)]];
-constant bool FC_mul_mm_bc_out [[function_constant(FC_MUL_MM + 1)]];
-constant short FC_mul_mm_ne12 [[function_constant(FC_MUL_MM + 2)]];
-constant short FC_mul_mm_ne13 [[function_constant(FC_MUL_MM + 3)]];
-constant short FC_mul_mm_r2 [[function_constant(FC_MUL_MM + 4)]];
-constant short FC_mul_mm_r3 [[function_constant(FC_MUL_MM + 5)]];
-
-// each block_q contains 16*nl weights
-#ifdef GGML_METAL_HAS_TENSOR
-template<
- typename SA, typename SA_4x4, typename SA_8x8,
- typename SB, typename SB_2x4, typename SB_8x8,
- typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread SA_4x4 &),
- typename T0, typename T0_4x4, typename T1, typename T1_2x4>
-kernel void kernel_mul_mm(
- constant ggml_metal_kargs_mul_mm & args,
- device const char * srcA,
- device const char * srcB,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig [[threadgroup_position_in_grid]],
- ushort tiitg [[thread_index_in_threadgroup]],
- ushort sgitg [[simdgroup_index_in_threadgroup]]) {
- (void) sgitg;
-
- // Matrix dimensions: A(M,K) x B(K,N) -> C(M,N)
- const int K = args.ne00;
- const int M = args.ne0;
- const int N = args.ne1;
-
- // Batch dimension handling
- const int im = tgpig.z;
- const int i12 = im % FC_mul_mm_ne12;
- const int i13 = im / FC_mul_mm_ne12;
-
- // Batch offsets for srcA and srcB
- const uint64_t offset0 = (i12/FC_mul_mm_r2)*args.nb02 + (i13/FC_mul_mm_r3)*args.nb03;
-
- // Tile dimensions
- constexpr int NRB = SZ_SIMDGROUP * N_MM_BLOCK_X * N_MM_SIMD_GROUP_X;
- constexpr int NRA = SZ_SIMDGROUP * N_MM_BLOCK_Y * N_MM_SIMD_GROUP_Y;
-
- // Tile offsets in output matrix
- const int ra = tgpig.y * NRA;
- const int rb = tgpig.x * NRB;
-
- // Threadgroup memory for dequantized A tile only
- threadgroup SA * sa = (threadgroup SA *)(shmem);
-
- // Work-item count for A loading
- constexpr int A_WORK_ITEMS = NRA * N_MM_NK;
- constexpr int NUM_THREADS = N_SIMDWIDTH * N_MM_SIMD_GROUP_X * N_MM_SIMD_GROUP_Y;
-
- // tA wraps threadgroup memory
- auto tA = tensor(sa, dextents<int32_t, 2>(N_MM_NK_TOTAL, NRA));
-
- // tB wraps device memory directly
- device T1 * ptrB = (device T1 *)(srcB + args.nb12*i12 + args.nb13*i13);
- const int strideB = args.nb11 / sizeof(T1);
- auto tB = tensor(ptrB, dextents<int32_t, 2>(K, N), array<int, 2>({1, strideB}));
-
- // Configure matmul operation
- // note: K is dynamic_extent (clamped to the valid range in PHASE 2), since a static
- // N_MM_NK_TOTAL K tile would read src1 out of bounds when K % N_MM_NK_TOTAL != 0
- // ref: https://github.com/ggml-org/llama.cpp/pull/27064
- mpp::tensor_ops::matmul2d<
- mpp::tensor_ops::matmul2d_descriptor(
- NRB, NRA, static_cast<int>(dynamic_extent), false, true, true,
- mpp::tensor_ops::matmul2d_descriptor::mode::multiply_accumulate),
- execution_simdgroups<N_MM_SIMD_GROUP_X * N_MM_SIMD_GROUP_Y>> mm;
-
- auto cT = mm.get_destination_cooperative_tensor<decltype(tB), decltype(tA), float>();
-
- // Accumulate partial results over K dimension
- for (int loop_k = 0; loop_k < K; loop_k += N_MM_NK_TOTAL) {
- // === PHASE 1: Dequantization of A into threadgroup memory ===
- for (int work = tiitg; work < A_WORK_ITEMS; work += NUM_THREADS) {
- const int row = work / N_MM_NK;
- const int k_chunk = work % N_MM_NK;
- const int k_pos = loop_k + k_chunk * 16;
- const short k_base = k_chunk * 16;
-
- // Bounds check: skip device read if row is out of matrix bounds
- if (ra + row < M) {
- if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
- // Element-wise reads when K is not aligned (nb01 not aligned for half4x4/float4x4).
- // MSL spec Table 2.5: half4x4 requires 8-byte alignment. When K is odd,
- // nb01 = K*2 is not 8-byte aligned, so odd-row pointers are misaligned.
- // Mirrors the legacy kernel's existing guard.
- device const T0 * row_ptr = (device const T0 *)(srcA + args.nb01 * (ra + row) + offset0);
-
- FOR_UNROLL (short i = 0; i < 16; i++) {
- sa[row * N_MM_NK_TOTAL + (k_base + i)] = (k_pos + i < K) ? (SA) row_ptr[k_pos + i] : (SA)0;
- }
- } else {
- const int block_idx = k_pos / (16 * nl);
- const short il = (k_pos / 16) % nl;
-
- device const block_q * row_ptr = (device const block_q *)(srcA + args.nb01 * (ra + row) + offset0);
-
- SA_4x4 temp_a;
- dequantize_func(row_ptr + block_idx, il, temp_a);
-
- FOR_UNROLL (short i = 0; i < 16; i++) {
- // Zero-pad A for K positions beyond valid range (handles partial K iterations)
- sa[row * N_MM_NK_TOTAL + (k_base + i)] = (k_pos + i < K) ? temp_a[i/4][i%4] : (SA)0;
- }
- }
- } else {
- // Zero-pad rows beyond matrix bounds
- FOR_UNROLL (short i = 0; i < 16; i++) {
- sa[row * N_MM_NK_TOTAL + (k_base + i)] = (SA)0;
- }
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // === PHASE 2: Tensor matmul ===
- // Clamp the K extent of both operand tensors to the remaining valid K range so
- // the dynamic-K op never reads past the K extent of src1 (or the staged A tile).
- const int kExt = min(N_MM_NK_TOTAL, K - loop_k);
-
- auto tAv = tensor(sa, dextents<int32_t, 2>(kExt, NRA), array<int, 2>({1, N_MM_NK_TOTAL}));
- auto tBv = tensor(ptrB + loop_k + rb * strideB, dextents<int32_t, 2>(kExt, N - rb), array<int, 2>({1, strideB}));
-
- mm.run(tBv, tAv, cT);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- // Store result tile to output matrix (with batch offset)
- // cT.store handles bounds checking via tD's extents (M, N)
- device float * dstBatch = (device float *)dst + im * N * M;
-
- auto tD = tensor(dstBatch, dextents<int32_t, 2>(M, N), array<int, 2>({1, M}));
- cT.store(tD.slice(ra, rb));
-}
-
-#else
-
-template<
- typename S0, typename S0_4x4, typename S0_8x8,
- typename S1, typename S1_2x4, typename S1_8x8,
- typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread S0_4x4 &),
- typename T0, typename T0_4x4, typename T1, typename T1_2x4>
-kernel void kernel_mul_mm(
- constant ggml_metal_kargs_mul_mm & args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
-
- threadgroup S0 * sa = (threadgroup S0 *)(shmem);
- threadgroup S1 * sb = (threadgroup S1 *)(shmem + 4096);
-
- constexpr int NR0 = 64;
- constexpr int NR1 = 32;
-
- constexpr int NK = 32;
- constexpr int NL0 = NK/16;
- constexpr int NL1 = NK/8;
-
- const int im = tgpig.z;
- const int r0 = tgpig.y*NR0;
- const int r1 = tgpig.x*NR1;
-
- // if this block is of 64x32 shape or smaller
- const short nr0 = (args.ne0 - r0 < NR0) ? (args.ne0 - r0) : NR0;
- const short nr1 = (args.ne1 - r1 < NR1) ? (args.ne1 - r1) : NR1;
-
- // a thread shouldn't load data outside of the matrix
- const short lr0 = ((short)tiitg/NL0) < nr0 ? ((short)tiitg/NL0) : nr0 - 1; // 0 .. 63
- const short lr1 = ((short)tiitg/NL1) < nr1 ? ((short)tiitg/NL1) : nr1 - 1; // 0 .. 31
-
- const short il0 = (tiitg % NL0);
-
- short il = il0;
-
- const int i12 = im % FC_mul_mm_ne12;
- const int i13 = im / FC_mul_mm_ne12;
-
- const uint64_t offset0 = (i12/FC_mul_mm_r2)*args.nb02 + (i13/FC_mul_mm_r3)*args.nb03;
- const short offset1 = il0/nl;
-
- device const block_q * x = (device const block_q *)(src0 + args.nb01*(r0 + lr0) + offset0) + offset1;
-
- const short iy = 8*(tiitg % NL1);
-
- device const T1 * y = (device const T1 *)(src1
- + args.nb13*i13
- + args.nb12*i12
- + args.nb11*(r1 + lr1)
- + args.nb10*iy);
-
- S0_8x8 ma[4];
- S1_8x8 mb[2];
-
- simdgroup_float8x8 mc[8];
-
- for (short i = 0; i < 8; i++){
- mc[i] = make_filled_simdgroup_matrix<float, 8>(0.f);
- }
-
- for (int loop_k = 0; loop_k < args.ne00; loop_k += NK) {
- // load data and store to threadgroup memory
- if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // no need for dequantization
- for (short i = 0; i < 16; i++) {
- const short sx = 2*il0 + i/8;
- const short sy = (tiitg/NL0)/8;
-
- //const short lx = i%8;
- //const short ly = (tiitg/NL0)%8;
- const short lx = (tiitg/NL0)%8;
- const short ly = i%8;
-
- const short ib = 8*sx + sy;
-
- *(sa + 64*ib + 8*ly + lx) = loop_k + 16*il + i < args.ne00 ? *((device T0 *) x + i) : 0;
- }
- } else {
- S0_4x4 temp_a;
- dequantize_func(x, il, temp_a);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- FOR_UNROLL (short i = 0; i < 16; i++) {
- const short sx = 2*il0 + i/8;
- const short sy = (tiitg/NL0)/8;
-
- //const short lx = i%8;
- //const short ly = (tiitg/NL0)%8;
- const short lx = (tiitg/NL0)%8;
- const short ly = i%8;
-
- const short ib = 8*sx + sy;
-
- // NOTE: this is massively slower.. WTF?
- //sa[64*ib + 8*ly + lx] = temp_a[i/4][i%4];
-
- *(sa + 64*ib + 8*ly + lx) = temp_a[i/4][i%4];
- }
- }
-
- if (FC_mul_mm_bc_inp) {
- for (short i = 0; i < 8; ++i) {
- const short sx = (tiitg%NL1);
- const short sy = (tiitg/NL1)/8;
-
- const short lx = i;
- const short ly = (tiitg/NL1)%8;
- //const short lx = (tiitg/NL1)%8;
- //const short ly = i;
-
- const short ib = 4*sx + sy;
-
- *(sb + 64*ib + 8*ly + lx) = loop_k + iy + i < args.ne00 ? (S1) *((device T1 *) y + i) : 0;
- }
- } else {
- const short sx = (tiitg%NL1);
- const short sy = (tiitg/NL1)/8;
-
- //const short dx = sx;
- //const short dy = sy;
-
- const short ly = (tiitg/NL1)%8;
-
- const short ib = 4*sx + sy;
-
- *(threadgroup S1_2x4 *)(sb + 64*ib + 8*ly) = (S1_2x4)(*((device T1_2x4 *) y));
- }
-
- il = (il + 2 < nl) ? il + 2 : il % 2;
- x = (il < 2) ? x + (2 + nl - 1)/nl : x;
-
- y += NK;
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // load matrices from threadgroup memory and conduct outer products
- threadgroup const S0 * lsma = (sa + 4*64*(sgitg%2));
- threadgroup const S1 * lsmb = (sb + 2*64*(sgitg/2));
-
- FOR_UNROLL (short ik = 0; ik < NK/8; ik++) {
- simdgroup_barrier(mem_flags::mem_none);
-
- FOR_UNROLL (short i = 0; i < 4; i++) {
- simdgroup_load(ma[i], lsma + 64*i, 8, 0, false);
- }
-
- simdgroup_barrier(mem_flags::mem_none);
-
- FOR_UNROLL (short i = 0; i < 2; i++) {
- simdgroup_load(mb[i], lsmb + 64*i, 8, 0, false);
- }
-
- simdgroup_barrier(mem_flags::mem_none);
-
- FOR_UNROLL (short i = 0; i < 8; i++){
- simdgroup_multiply_accumulate(mc[i], mb[i/4], ma[i%4], mc[i]);
- }
-
- lsma += 8*64;
- lsmb += 4*64;
- }
- }
-
- if (!FC_mul_mm_bc_out || (r0 + NR0 <= args.ne0 && r1 + NR1 <= args.ne1)) {
- // if no bounds checks on the output are needed, we can directly write to device memory
- device float * C = (device float *) dst +
- (r0 + 32*(sgitg & 1)) + \
- (r1 + 16*(sgitg >> 1)) * args.ne0 + im*args.ne1*args.ne0;
-
- for (short i = 0; i < 8; i++) {
- simdgroup_store(mc[i], C + 8*(i%4) + 8*args.ne0*(i/4), args.ne0, 0, false);
- }
- } else {
- // block is smaller than 64x32, we should avoid writing data outside of the matrix
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- threadgroup float * temp_str = ((threadgroup float *) shmem) + 32*(sgitg&1) + (16*(sgitg >> 1))*NR0;
-
- for (short i = 0; i < 8; i++) {
- simdgroup_store(mc[i], temp_str + 8*(i%4) + 8*NR0*(i/4), NR0, 0, false);
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (sgitg == 0) {
- for (int j = tiitg; j < nr1; j += NR1) {
- device float * D = (device float *) dst + r0 + (r1 + j)*args.ne0 + im*args.ne1*args.ne0;
- device float4 * D4 = (device float4 *) D;
-
- threadgroup float * C = temp_str + (j*NR0);
- threadgroup float4 * C4 = (threadgroup float4 *) C;
-
- int i = 0;
- for (; i < nr0/4; i++) {
- *(D4 + i) = *(C4 + i);
- }
-
- i *= 4;
- for (; i < nr0; i++) {
- *(D + i) = *(C + i);
- }
- }
- }
- }
-}
-
-#endif // GGML_METAL_HAS_TENSOR
-
-template<short ne20> // n_expert_used
-kernel void kernel_mul_mm_id_map0(
- constant ggml_metal_kargs_mul_mm_id_map0 & args,
- device const char * src2,
- device char * htpe,
- device char * hids,
- threadgroup char * shmem [[threadgroup(0)]],
- ushort tpitg[[thread_position_in_threadgroup]],
- ushort ntg[[threads_per_threadgroup]]) {
- const short ide = tpitg; // expert id
-
- uint32_t n_all = 0;
-
- device int32_t * ids_i32 = (device int32_t *) hids + ide*args.ne21;
-
- for (int i21 = 0; i21 < args.ne21; i21 += ntg) { // n_tokens
- if (i21 + tpitg < args.ne21) {
- device const int32_t * src2_i32 = (device const int32_t *) (src2 + (i21 + tpitg)*args.nb21);
-
- threadgroup uint16_t * sids = (threadgroup uint16_t *) shmem + tpitg*ne20;
-
- #pragma unroll(ne20)
- for (short i20 = 0; i20 < ne20; i20++) {
- sids[i20] = src2_i32[i20];
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- for (short t = 0; t < ntg; t++) {
- if (i21 + t >= args.ne21) {
- break;
- }
-
- threadgroup const uint16_t * sids = (threadgroup const uint16_t *) shmem + t*ne20;
-
- short sel = 0;
- #pragma unroll(ne20)
- for (short i20 = 0; i20 < ne20; i20++) {
- sel += (sids[i20] == ide)*(i20 + 1);
- }
-
- ids_i32[n_all] = (i21 + t)*ne20 + sel - 1;
-
- n_all += sel > 0;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- device uint32_t * tpe_u32 = (device uint32_t *) (htpe);
- tpe_u32[ide] = n_all;
-}
-
-typedef decltype(kernel_mul_mm_id_map0<1>) kernel_mul_mm_id_map0_t;
-
-template [[host_name("kernel_mul_mm_id_map0_ne20_1" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<1>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_2" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<2>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_4" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<4>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_5" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<5>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_6" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<6>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_8" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<8>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_10")]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<10>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_16")]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<16>;
-template [[host_name("kernel_mul_mm_id_map0_ne20_22")]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<22>;
-
-template<typename S0, typename S0_4x4, typename S0_8x8, typename S1, typename S1_2x4, typename S1_8x8, typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread S0_4x4 &), typename T0, typename T0_4x4, typename T1, typename T1_2x4>
-kernel void kernel_mul_mm_id(
- constant ggml_metal_kargs_mul_mm_id & args,
- device const char * src0,
- device const char * src1,
- device const char * htpe,
- device const char * hids,
- device char * dst,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- threadgroup S0 * sa = (threadgroup S0 *)(shmem);
- threadgroup S1 * sb = (threadgroup S1 *)(shmem + 4096);
-
-#ifdef GGML_METAL_HAS_TENSOR
- threadgroup float * sc = (threadgroup float *)(shmem);
-#endif
-
- constexpr int NR0 = 64;
- constexpr int NR1 = 32;
-
- constexpr int NK = 32;
- constexpr int NL0 = NK/16;
- constexpr int NL1 = NK/8;
-
- const int im = tgpig.z; // expert
- const int r0 = tgpig.y*NR0;
- const int r1 = tgpig.x*NR1;
-
- device const uint32_t * tpe_u32 = (device const uint32_t *) (htpe);
- device const int32_t * ids_i32 = (device const int32_t *) (hids);
-
- const int32_t neh1 = tpe_u32[im];
-
- if (r1 >= neh1) {
- return;
- }
-
- // if this block is of 64x32 shape or smaller
- const short nr0 = (args.ne0 - r0 < NR0) ? (args.ne0 - r0) : NR0;
- const short nr1 = ( neh1 - r1 < NR1) ? ( neh1 - r1) : NR1;
-
- // a thread shouldn't load data outside of the matrix
- const short lr0 = ((short)tiitg/NL0) < nr0 ? ((short)tiitg/NL0) : nr0 - 1; // 0 .. 63
- const short lr1 = ((short)tiitg/NL1) < nr1 ? ((short)tiitg/NL1) : nr1 - 1; // 0 .. 31
-
- const short il0 = (tiitg % NL0);
-
- short il = il0;
-
- const int id = ids_i32[im*args.ne21 + r1 + lr1];
-
- const short i11 = (id % args.ne20) % args.ne11;
- const short i12 = (id / args.ne20);
- const short i13 = 0;
-
- const uint64_t offset0 = im*args.nb02 + i13*args.nb03;
- const short offset1 = il0/nl;
-
- device const block_q * x = (device const block_q *)(src0 + args.nb01*(r0 + lr0) + offset0) + offset1;
-
- const short iy = 8*(tiitg % NL1);
-
- device const T1 * y = (device const T1 *)(src1
- + args.nb13*i13
- + args.nb12*i12
- + args.nb11*i11
- + args.nb10*iy);
-
-#ifndef GGML_METAL_HAS_TENSOR
- S0_8x8 ma[4];
- S1_8x8 mb[2];
-
- simdgroup_float8x8 mc[8];
-
- for (short i = 0; i < 8; i++){
- mc[i] = make_filled_simdgroup_matrix<float, 8>(0.f);
- }
-#else
- auto tA = tensor<threadgroup S0, dextents<int32_t, 2>, tensor_inline>(sa, dextents<int32_t, 2>(NK, NR0));
- auto tB = tensor<threadgroup S1, dextents<int32_t, 2>, tensor_inline>(sb, dextents<int32_t, 2>(NR1, NK ));
-
- mpp::tensor_ops::matmul2d<
- mpp::tensor_ops::matmul2d_descriptor(NR1, NR0, NK, false, true, false, mpp::tensor_ops::matmul2d_descriptor::mode::multiply_accumulate),
- execution_simdgroups<4>> mm;
-
- auto cT = mm.get_destination_cooperative_tensor<decltype(tA), decltype(tB), float>();
-#endif
-
- for (int loop_k = 0; loop_k < args.ne00; loop_k += NK) {
-#ifndef GGML_METAL_HAS_TENSOR
- // load data and store to threadgroup memory
- if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // no need for dequantization
- for (short i = 0; i < 16; i++) {
- const short sx = 2*il0 + i/8;
- const short sy = (tiitg/NL0)/8;
-
- //const short lx = i%8;
- //const short ly = (tiitg/NL0)%8;
- const short lx = (tiitg/NL0)%8;
- const short ly = i%8;
-
- const short ib = 8*sx + sy;
-
- *(sa + 64*ib + 8*ly + lx) = loop_k + 16*il + i < args.ne00 ? (S0) *((device T0 *) x + i) : (S0) 0;
- }
- } else {
- S0_4x4 temp_a;
- dequantize_func(x, il, temp_a);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- FOR_UNROLL (short i = 0; i < 16; i++) {
- const short sx = 2*il0 + i/8;
- const short sy = (tiitg/NL0)/8;
-
- //const short lx = i%8;
- //const short ly = (tiitg/NL0)%8;
- const short lx = (tiitg/NL0)%8;
- const short ly = i%8;
-
- const short ib = 8*sx + sy;
-
- // NOTE: this is massively slower.. WTF?
- //sa[64*ib + 8*ly + lx] = temp_a[i/4][i%4];
-
- *(sa + 64*ib + 8*ly + lx) = temp_a[i/4][i%4];
- }
- }
-
- if (FC_mul_mm_bc_inp) {
- for (short i = 0; i < 8; ++i) {
- const short sx = (tiitg%NL1);
- const short sy = (tiitg/NL1)/8;
-
- const short lx = i;
- const short ly = (tiitg/NL1)%8;
- //const short lx = (tiitg/NL1)%8;
- //const short ly = i;
-
- const short ib = 4*sx + sy;
-
- *(sb + 64*ib + 8*ly + lx) = loop_k + iy + i < args.ne00 ? (S1) *((device T1 *) y + i) : 0;
- }
- } else {
- const short sx = (tiitg%NL1);
- const short sy = (tiitg/NL1)/8;
-
- //const short dx = sx;
- //const short dy = sy;
-
- const short ly = (tiitg/NL1)%8;
-
- const short ib = 4*sx + sy;
-
- *(threadgroup S1_2x4 *)(sb + 64*ib + 8*ly) = (S1_2x4)(*((device T1_2x4 *) y));
- }
-#else
- // load data and store to threadgroup memory
- if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // no need for dequantization
- for (short i = 0; i < 16; i++) {
- const short sx = 2*il0 + i/8;
- const short sy = (tiitg/NL0)/8;
-
- const short lx = i%8;
- const short ly = (tiitg/NL0)%8;
- //const short lx = (tiitg/NL0)%8;
- //const short ly = i%8;
-
- *(sa + NK*(8*sy + ly) + 8*sx + lx) = loop_k + 16*il + i < args.ne00 ? *((device T0 *) x + i) : 0;
- }
- } else {
- S0_4x4 temp_a;
- dequantize_func(x, il, temp_a);
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- FOR_UNROLL (short i = 0; i < 16; i++) {
- const short sx = 2*il0 + i/8;
- const short sy = (tiitg/NL0)/8;
-
- const short lx = i%8;
- const short ly = (tiitg/NL0)%8;
- //const short lx = (tiitg/NL0)%8;
- //const short ly = i%8;
-
- *(sa + NK*(8*sy + ly) + 8*sx + lx) = temp_a[i/4][i%4];
- }
- }
-
- if (FC_mul_mm_bc_inp) {
- for (short i = 0; i < 8; ++i) {
- const short sx = (tiitg%NL1);
- const short sy = (tiitg/NL1)/8;
-
- const short lx = i;
- const short ly = (tiitg/NL1)%8;
- //const short lx = (tiitg/NL1)%8;
- //const short ly = i;
-
- *(sb + NK*(8*sy + ly) + 8*sx + lx) = loop_k + iy + i < args.ne00 ? (S1) *((device T1 *) y + i) : 0;
- }
- } else {
- const short sx = (tiitg%NL1);
- const short sy = (tiitg/NL1)/8;
-
- //const short lx = i;
- const short ly = (tiitg/NL1)%8;
- //const short lx = (tiitg/NL1)%8;
- //const short ly = i;
-
- *(threadgroup S1_2x4 *)(sb + NK*(8*sy + ly) + 8*sx) = (S1_2x4)(*((device T1_2x4 *) y));
- }
-#endif
-
- il = (il + 2 < nl) ? il + 2 : il % 2;
- x = (il < 2) ? x + (2 + nl - 1)/nl : x;
-
- y += NK;
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
-#ifndef GGML_METAL_HAS_TENSOR
- // load matrices from threadgroup memory and conduct outer products
- threadgroup const S0 * lsma = (sa + 4*64*(sgitg%2));
- threadgroup const S1 * lsmb = (sb + 2*64*(sgitg/2));
-
- FOR_UNROLL (short ik = 0; ik < NK/8; ik++) {
- simdgroup_barrier(mem_flags::mem_none);
-
- FOR_UNROLL (short i = 0; i < 4; i++) {
- simdgroup_load(ma[i], lsma + 64*i, 8, 0, false);
- }
-
- simdgroup_barrier(mem_flags::mem_none);
-
- FOR_UNROLL (short i = 0; i < 2; i++) {
- simdgroup_load(mb[i], lsmb + 64*i, 8, 0, false);
- }
-
- simdgroup_barrier(mem_flags::mem_none);
-
- FOR_UNROLL (short i = 0; i < 8; i++){
- simdgroup_multiply_accumulate(mc[i], mb[i/4], ma[i%4], mc[i]);
- }
-
- lsma += 8*64;
- lsmb += 4*64;
- }
-#else
- auto sA = tA.slice(0, 0);
- auto sB = tB.slice(0, 0);
-
- mm.run(sB, sA, cT);
-#endif
- }
-
- // block is smaller than 64x32, we should avoid writing data outside of the matrix
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
-#ifdef GGML_METAL_HAS_TENSOR
- auto tC = tensor<threadgroup float, dextents<int32_t, 2>, tensor_inline>(sc, dextents<int32_t, 2>(NR0, NR1));
- cT.store(tC);
-#else
- threadgroup float * temp_str = ((threadgroup float *) shmem) + 32*(sgitg&1) + (16*(sgitg >> 1))*NR0;
-
- for (short i = 0; i < 8; i++) {
- simdgroup_store(mc[i], temp_str + 8*(i%4) + 8*NR0*(i/4), NR0, 0, false);
- }
-#endif
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- for (short j = sgitg; j < nr1; j += 4) {
- const int id = ids_i32[im*args.ne21 + r1 + j];
-
- const short ide = id % args.ne20;
- const short idt = id / args.ne20;
-
- device float * D = (device float *) dst + r0 + ide*args.ne0 + idt*args.ne1*args.ne0;
- device float4 * D4 = (device float4 *) D;
-
- threadgroup float * C = (threadgroup float *) shmem + j*NR0;
- threadgroup float4 * C4 = (threadgroup float4 *) C;
-
- int i = tiisg;
- for (; i < nr0/4; i += 32) {
- *(D4 + i) = *(C4 + i);
- }
-
- i = (4*(nr0/4)) + tiisg;
- for (; i < nr0; i += 32) {
- *(D + i) = *(C + i);
- }
- }
-}
-
-//
-// matrix-matrix multiplication
-//
-
-typedef decltype(kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>) mul_mm_t;
-
-template [[host_name("kernel_mul_mm_f32_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_f16_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, float, float2x4>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mm_bf16_f32")]] kernel mul_mm_t kernel_mul_mm<bfloat, bfloat4x4, simdgroup_bfloat8x8, bfloat, bfloat2x4, simdgroup_bfloat8x8, bfloat4x4, 1, dequantize_bf16, bfloat, bfloat4x4, float, float2x4>;
-#endif
-template [[host_name("kernel_mul_mm_q1_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q2_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q4_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q4_1_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q5_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q5_1_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q8_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_mxfp4_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q2_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q3_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q4_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q5_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_q6_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq2_xxs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq2_xs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq3_xxs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq3_s_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq2_s_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq1_s_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq1_m_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq4_nl_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_iq4_xs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_tq2_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, float, float2x4>;
-
-template [[host_name("kernel_mul_mm_f32_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_f16_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q1_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q2_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q4_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q4_1_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q5_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q5_1_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q8_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_mxfp4_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q2_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q3_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q4_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q5_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_q6_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq2_xxs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq2_xs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq3_xxs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq3_s_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq2_s_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq1_s_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq1_m_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq4_nl_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_iq4_xs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_tq2_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, half, half2x4>;
-
-//
-// indirect matrix-matrix multiplication
-//
-
-typedef decltype(kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>) mul_mm_id;
-
-template [[host_name("kernel_mul_mm_id_f32_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_f16_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, float, float2x4>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mm_id_bf16_f32")]] kernel mul_mm_id kernel_mul_mm_id<bfloat, bfloat4x4, simdgroup_bfloat8x8, bfloat, bfloat2x4, simdgroup_bfloat8x8, bfloat4x4, 1, dequantize_bf16, bfloat, bfloat4x4, float, float2x4>;
-#endif
-template [[host_name("kernel_mul_mm_id_q1_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q2_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q4_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q4_1_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q5_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q5_1_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q8_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_mxfp4_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q2_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q3_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q4_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q5_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_q6_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq2_xxs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq2_xs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq3_xxs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq3_s_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq2_s_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq1_s_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq1_m_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq4_nl_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_iq4_xs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, float, float2x4>;
-template [[host_name("kernel_mul_mm_id_tq2_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, float, float2x4>;
-
-template [[host_name("kernel_mul_mm_id_f32_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_f16_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q1_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q2_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q4_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q4_1_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q5_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q5_1_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q8_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_mxfp4_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q2_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q3_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q4_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q5_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_q6_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq2_xxs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq2_xs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq3_xxs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq3_s_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq2_s_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq1_s_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq1_m_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq4_nl_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_iq4_xs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, half, half2x4>;
-template [[host_name("kernel_mul_mm_id_tq2_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, half, half2x4>;
-
-//
-// matrix-vector multiplication
-//
-
-typedef void (kernel_mul_mv_disp_t)(
- ggml_metal_kargs_mul_mv args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- uint3 tgpig,
- ushort tiisg);
-
-typedef void (kernel_mul_mv2_disp_t)(
- ggml_metal_kargs_mul_mv args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiisg,
- ushort sgitg);
-
-template<kernel_mul_mv_disp_t disp_fn>
-void mmv_fn(
- ggml_metal_kargs_mul_mv args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiitg,
- ushort tiisg,
- ushort sgitg) {
- disp_fn(args, src0, src1, dst, tgpig, tiisg);
-}
-
-template<kernel_mul_mv2_disp_t disp_fn>
-void mmv_fn(
- ggml_metal_kargs_mul_mv args,
- device const char * src0,
- device const char * src1,
- device char * dst,
- threadgroup char * shmem,
- uint3 tgpig,
- ushort tiitg,
- ushort tiisg,
- ushort sgitg) {
- disp_fn(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
-}
-
-typedef decltype(mmv_fn<kernel_mul_mv_t_t_disp<half, half, ggml_metal_kargs_mul_mv>>) mul_mv_disp_fn_t;
-
-template<mul_mv_disp_fn_t disp_fn>
-kernel void kernel_mul_mv_id(
- constant ggml_metal_kargs_mul_mv_id & args,
- device const char * src0s,
- device const char * src1,
- device char * dst,
- device const char * ids,
- threadgroup char * shmem [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- const int iid1 = tgpig.z/args.nei0;
- const int idx = tgpig.z%args.nei0;
-
- tgpig.z = 0;
-
- const int32_t i02 = ((device const int32_t *) (ids + iid1*args.nbi1))[idx];
-
- const int64_t i11 = idx % args.ne11;
- const int64_t i12 = iid1;
-
- const int64_t i1 = idx;
- const int64_t i2 = i12;
-
- device const char * src0_cur = src0s + i02*args.nb02;
- device const char * src1_cur = src1 + i11*args.nb11 + i12*args.nb12;
-
- device char * dst_cur = dst + (i1*args.ne0 + i2*args.ne1*args.ne0)*sizeof(float);
-
- ggml_metal_kargs_mul_mv args0 = {
- /*.ne00 =*/ args.ne00,
- /*.ne01 =*/ args.ne01,
- /*.ne02 =*/ 1, // args.ne02,
- /*.nb00 =*/ args.nb00,
- /*.nb01 =*/ args.nb01,
- /*.nb02 =*/ args.nb02,
- /*.nb03 =*/ args.nb02, // args.ne02 == 1
- /*.ne10 =*/ args.ne10,
- /*.ne11 =*/ 1, // args.ne11,
- /*.ne12 =*/ 1, // args.ne12,
- /*.nb10 =*/ args.nb10,
- /*.nb11 =*/ args.nb11,
- /*.nb12 =*/ args.nb12,
- /*.nb13 =*/ args.nb12, // ne12 == 1
- /*.ne0 =*/ args.ne0,
- /*.ne1 =*/ 1, // args.ne1,
- /*.nr0 =*/ args.nr0,
- /*.r2 =*/ 1,
- /*.r3 =*/ 1,
- };
-
- disp_fn(
- args0,
- /* src0 */ src0_cur,
- /* src1 */ src1_cur,
- /* dst */ dst_cur,
- shmem,
- tgpig,
- tiitg,
- tiisg,
- sgitg);
-}
-
-typedef decltype(kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<float, float>>>) kernel_mul_mv_id_t;
-
-typedef decltype(kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<float, float4, float, float4>>>) kernel_mul_mv_id_4_t;
-
-template [[host_name("kernel_mul_mv_id_f32_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<float, float>>>;
-template [[host_name("kernel_mul_mv_id_f16_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<half, float>>>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mv_id_bf16_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<bfloat, float>>>;
-#endif
-template [[host_name("kernel_mul_mv_id_f32_f32_4")]] kernel kernel_mul_mv_id_4_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<float, float4, float, float4>>>;
-template [[host_name("kernel_mul_mv_id_f16_f32_4")]] kernel kernel_mul_mv_id_4_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<half, half4, float, float4>>>;
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_mul_mv_id_bf16_f32_4")]] kernel kernel_mul_mv_id_4_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<bfloat, bfloat4, float, float4>>>;
-#endif
-
-template [[host_name("kernel_mul_mv_id_q8_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q8_0_f32_impl<N_R0_Q8_0>>>;
-
-template [[host_name("kernel_mul_mv_id_q1_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q1_0_f32_impl<N_R0_Q1_0>>>;
-template [[host_name("kernel_mul_mv_id_q2_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q2_0_f32_impl<N_R0_Q2_0>>>;
-template [[host_name("kernel_mul_mv_id_q4_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q4_0, N_R0_Q4_0>>>;
-template [[host_name("kernel_mul_mv_id_q4_1_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q4_1, N_R0_Q4_1>>>;
-template [[host_name("kernel_mul_mv_id_q5_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q5_0, N_R0_Q5_0>>>;
-template [[host_name("kernel_mul_mv_id_q5_1_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q5_1, N_R0_Q5_1>>>;
-
-template [[host_name("kernel_mul_mv_id_mxfp4_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_mxfp4_f32_impl<N_R0_MXFP4>>>;
-
-template [[host_name("kernel_mul_mv_id_q2_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q2_K_f32_impl <N_R0_Q2_K>>>;
-template [[host_name("kernel_mul_mv_id_q3_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q3_K_f32_impl <N_R0_Q3_K>>>;
-template [[host_name("kernel_mul_mv_id_q4_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q4_K_f32_impl <N_R0_Q4_K>>>;
-template [[host_name("kernel_mul_mv_id_q5_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q5_K_f32_impl <N_R0_Q5_K>>>;
-template [[host_name("kernel_mul_mv_id_q6_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q6_K_f32_impl <N_R0_Q6_K>>>;
-template [[host_name("kernel_mul_mv_id_iq1_s_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq1_s_f32_impl <N_R0_IQ1_S>>>;
-template [[host_name("kernel_mul_mv_id_iq1_m_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq1_m_f32_impl <N_R0_IQ1_M>>>;
-template [[host_name("kernel_mul_mv_id_iq2_xxs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq2_xxs_f32_impl<N_R0_IQ2_XXS>>>;
-template [[host_name("kernel_mul_mv_id_iq2_xs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq2_xs_f32_impl <N_R0_IQ2_XS>>>;
-template [[host_name("kernel_mul_mv_id_iq3_xxs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq3_xxs_f32_impl<N_R0_IQ3_XXS>>>;
-template [[host_name("kernel_mul_mv_id_iq3_s_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq3_s_f32_impl <N_R0_IQ3_S>>>;
-template [[host_name("kernel_mul_mv_id_iq2_s_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq2_s_f32_impl <N_R0_IQ2_S>>>;
-template [[host_name("kernel_mul_mv_id_iq4_nl_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq4_nl_f32_impl <N_R0_IQ4_NL>>>;
-template [[host_name("kernel_mul_mv_id_iq4_xs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq4_xs_f32_impl <N_R0_IQ4_XS>>>;
-template [[host_name("kernel_mul_mv_id_tq2_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_tq2_0_f32_impl <N_R0_TQ2_0>>>;
-
-kernel void kernel_pool_2d_max_f32(
- constant ggml_metal_kargs_pool_2d & args,
- device const float * src0,
- device float * dst,
- uint gid[[thread_position_in_grid]]) {
-
- if (gid >= args.np) {
- return;
- }
-
- const int idx = gid;
- const int I_HW = args.IH * args.IW;
- const int O_HW = args.OH * args.OW;
- const int nc = idx / O_HW;
- const int cur_oh = idx % O_HW / args.OW;
- const int cur_ow = idx % O_HW % args.OW;
-
- device const float * i_ptr = src0 + nc * I_HW;
- device float * o_ptr = dst + nc * O_HW;
-
- const int start_h = cur_oh * args.s1 - args.p1;
- const int bh = MAX(0, start_h);
- const int eh = MIN(args.IH, start_h + args.k1);
- const int start_w = cur_ow * args.s0 - args.p0;
- const int bw = MAX(0, start_w);
- const int ew = MIN(args.IW, start_w + args.k0);
-
- float res = -INFINITY;
-
- for (int i = bh; i < eh; i += 1) {
- for (int j = bw; j < ew; j += 1) {
- res = MAX(res, i_ptr[i * args.IW + j]);
- }
- }
-
- o_ptr[cur_oh * args.OW + cur_ow] = res;
-}
-
-kernel void kernel_pool_2d_avg_f32(
- constant ggml_metal_kargs_pool_2d & args,
- device const float * src0,
- device float * dst,
- uint gid[[thread_position_in_grid]]) {
-
- if (gid >= args.np) {
- return;
- }
-
- const int idx = gid;
- const int I_HW = args.IH * args.IW;
- const int O_HW = args.OH * args.OW;
- const int nc = idx / O_HW;
- const int cur_oh = idx % O_HW / args.OW;
- const int cur_ow = idx % O_HW % args.OW;
-
- device const float * i_ptr = src0 + nc * I_HW;
- device float * o_ptr = dst + nc * O_HW;
-
- const int start_h = cur_oh * args.s1 - args.p1;
- const int bh = MAX(0, start_h);
- const int eh = MIN(args.IH, start_h + args.k1);
- const int start_w = cur_ow * args.s0 - args.p0;
- const int bw = MAX(0, start_w);
- const int ew = MIN(args.IW, start_w + args.k0);
- // const float scale = 1. / ((eh - bh) * (ew - bw));
- const float scale = 1. / (args.k0 * args.k1);
-
- float res = 0;
-
- for (int i = bh; i < eh; i += 1) {
- for (int j = bw; j < ew; j += 1) {
- float cur = i_ptr[i * args.IW + j];
- res += cur * scale;
- }
- }
-
- o_ptr[cur_oh * args.OW + cur_ow] = res;
-}
-
-
-kernel void kernel_pool_1d_max_f32(
- constant ggml_metal_kargs_pool_1d & args,
- device const float * src,
- device float * dst,
- uint gid [[thread_position_in_grid]]
-) {
-
- if (gid >= args.np) {
- return;
- }
-
- const int ow = (int)gid % args.OW;
- const int row = (int)gid / args.OW;
-
- const int base = ow * args.s0 - args.p0;
-
- float acc = -INFINITY;
-
- const int src_off = row * args.IW;
- const int dst_off = row * args.OW;
-
- for (int ki = 0; ki < args.k0; ++ki) {
- int j = base + ki;
- if (j < 0 || j >= args.IW){
- continue;
- }
- float v = src[src_off + j];
- acc = max(acc, v);
- }
-
- dst[dst_off + ow] = acc;
-}
-
-kernel void kernel_pool_1d_avg_f32(
- constant ggml_metal_kargs_pool_1d & args,
- device const float * src,
- device float * dst,
- uint gid [[thread_position_in_grid]]
-) {
-
- if (gid >= args.np) {
- return;
- }
-
- const int ow = (int)gid % args.OW;
- const int row = (int)gid / args.OW;
-
- const int base = ow * args.s0 - args.p0;
-
- float acc = 0.0f;
- int cnt = 0;
-
- const int src_off = row * args.IW;
- const int dst_off = row * args.OW;
-
- for (int ki = 0; ki < args.k0; ++ki) {
- const int j = base + ki;
- if (j < 0 || j >= args.IW) {
- continue;
- }
- acc += src[src_off + j];
- cnt += 1;
- }
-
- dst[dst_off + ow] = (cnt > 0) ? (acc / (float)cnt) : 0.0f;
-}
-
-kernel void kernel_opt_step_adamw_f32(
- constant ggml_metal_kargs_opt_step_adamw & args,
- device float * x,
- device const float * g,
- device float * g_m,
- device float * g_v,
- device const float * pars,
- uint gid[[thread_position_in_grid]]) {
-
- if (gid >= args.np) {
- return;
- }
-
- const float alpha = pars[0];
- const float beta1 = pars[1];
- const float beta2 = pars[2];
- const float eps = pars[3];
- const float wd = pars[4];
- const float beta1h = pars[5];
- const float beta2h = pars[6];
-
- const float gi = g[gid];
- const float gmi = g_m[gid] * beta1 + gi * (1.0f - beta1);
- const float gvi = g_v[gid] * beta2 + gi * gi * (1.0f - beta2);
-
- g_m[gid] = gmi;
- g_v[gid] = gvi;
-
- const float mh = gmi * beta1h;
- const float vh = sqrt(gvi * beta2h) + eps;
-
- x[gid] = x[gid] * (1.0f - alpha * wd) - alpha * mh / vh;
-}
-
-kernel void kernel_opt_step_sgd_f32(
- constant ggml_metal_kargs_opt_step_sgd & args,
- device float * x,
- device const float * g,
- device const float * pars,
- uint gid[[thread_position_in_grid]]) {
-
- if (gid >= args.np) {
- return;
- }
-
- x[gid] = x[gid] * (1.0f - pars[0] * pars[1]) - pars[0] * g[gid];
-}
-
-template<typename T>
-kernel void kernel_memset(
- constant ggml_metal_kargs_memset & args,
- device T * dst,
- uint tpig[[thread_position_in_grid]]) {
- dst[tpig] = args.val;
-}
-
-typedef decltype(kernel_memset<int64_t>) kernel_memset_t;
-
-template [[host_name("kernel_memset_i64")]] kernel kernel_memset_t kernel_memset<int64_t>;
-
-constant short FC_count_equal_nsg [[function_constant(FC_COUNT_EQUAL + 0)]];
-
-template<typename T>
-kernel void kernel_count_equal(
- constant ggml_metal_kargs_count_equal & args,
- device const char * src0,
- device const char * src1,
- device atomic_int * dst,
- threadgroup int32_t * shmem_i32 [[threadgroup(0)]],
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort3 tpitg[[thread_position_in_threadgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- const short NSG = FC_count_equal_nsg;
-
- const int i3 = tgpig.z;
- const int i2 = tgpig.y;
- const int i1 = tgpig.x;
-
- if (i3 >= args.ne03 || i2 >= args.ne02 || i1 >= args.ne01) {
- return;
- }
-
- int sum = 0;
-
- device const char * base0 = src0 + i1*args.nb01 + i2*args.nb02 + i3*args.nb03;
- device const char * base1 = src1 + i1*args.nb11 + i2*args.nb12 + i3*args.nb13;
-
- for (int64_t i0 = tpitg.x; i0 < args.ne00; i0 += ntg.x) {
- const T v0 = *(device const T *)(base0 + i0*args.nb00);
- const T v1 = *(device const T *)(base1 + i0*args.nb10);
- sum += (v0 == v1);
- }
-
- sum = simd_sum(sum);
-
- if (tiisg == 0) {
- shmem_i32[sgitg] = sum;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- if (sgitg == 0) {
- float v = 0.0f;
- if (tpitg.x < NSG) {
- v = shmem_i32[tpitg.x];
- }
-
- float total = simd_sum(v);
- if (tpitg.x == 0) {
- atomic_fetch_add_explicit(dst, (int32_t) total, memory_order_relaxed);
- }
- }
-}
-
-typedef decltype(kernel_count_equal<int32_t>) kernel_count_equal_t;
-
-template [[host_name("kernel_count_equal_i32")]] kernel kernel_count_equal_t kernel_count_equal<int32_t>;
-
-template<
- typename kd4x4_t,
- short nl_k,
- void (*deq_k)(device const kd4x4_t *, short, thread half4x4 &)>
-kernel void kernel_lightning_indexer(
- constant ggml_metal_kargs_lightning_indexer & args,
- device const char * q,
- device const char * k,
- device const char * w,
- device const char * m,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiitg[[thread_index_in_threadgroup]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]]) {
- constexpr short DK = OP_LIGHTNING_INDEXER_DK;
- constexpr short NH = OP_LIGHTNING_INDEXER_NH;
- constexpr short NHPTG = OP_LIGHTNING_INDEXER_NHPTG;
- constexpr short NKPSG = OP_LIGHTNING_INDEXER_NKPSG;
- constexpr short NSG = OP_LIGHTNING_INDEXER_NSG;
- constexpr short NBPTG = OP_LIGHTNING_INDEXER_NBPTG;
-
- constexpr short DK4 = DK/4;
- constexpr short DK8 = DK/8;
- constexpr short DK16 = DK/16;
-
- constexpr short NK = NKPSG*NSG; // keys per threadgroup
- constexpr short NTG = 32*NSG; // threads per threadgroup
-
- const int i_stream = tgpig.z;
- const int i_kv_0 = tgpig.x*NK; // first key of this threadgroup
- const int i_kv = i_kv_0 + sgitg*NKPSG; // first key of this simdgroup
-
- threadgroup half sk[NK * DK16 * 16];
- threadgroup half4x4 * sk4x4 = (threadgroup half4x4 *) sk;
-
- for (short i = tiitg; i < NK*DK16; i += NTG) {
- const short ik = i/DK16;
- const short i16 = i%DK16;
-
- half4x4 tmp;
-
- if (i_kv_0 + ik < args.n_kv) {
- device const kd4x4_t * kr = (device const kd4x4_t *) (k + (i_kv_0 + ik)*args.nbk2 + i_stream*args.nbk3);
-
- deq_k(kr + i16/nl_k, i16%nl_k, tmp);
- } else {
- FOR_UNROLL (short j = 0; j < 4; ++j) {
- tmp[j] = half4(0.0h);
- }
- }
-
- sk4x4[i] = tmp;
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- // K tile of this simdgroup, transposed to [DK, NKPSG]
- simdgroup_half8x8 mk[DK8];
-
- FOR_UNROLL (short i = 0; i < DK8; ++i) {
- simdgroup_load(mk[i], sk + sgitg*NKPSG*DK + 8*i, DK, 0, true);
- }
-
- threadgroup half4 sq4[NHPTG*DK4];
- threadgroup half * sq = (threadgroup half *) sq4;
-
- threadgroup float sw [NHPTG];
- threadgroup float sqk[NSG*NHPTG*NKPSG];
-
- const int i_batch_0 = tgpig.y*NBPTG;
- const int n_batch = min((int) NBPTG, args.n_batch - i_batch_0);
-
- for (short ib = 0; ib < n_batch; ++ib) {
- const int i_batch = i_batch_0 + ib;
-
- device const char * pq = q + i_batch*args.nbq2 + i_stream*args.nbq3;
- device const char * pw = w + i_batch*args.nbw1 + i_stream*args.nbw3;
-
- float score = 0.0f;
-
- FOR_UNROLL (short i_head = 0; i_head < NH; i_head += NHPTG) {
- // stage the Q tile [DK, NHPTG] and the (prescaled) head weights
- for (short i = tiitg; i < NHPTG*DK4; i += NTG) {
- const short ih = i/DK4;
- const short i4 = i%DK4;
-
- device const float4 * q4 = (device const float4 *) (pq + (i_head + ih)*args.nbq1);
-
- sq4[ih*DK4 + i4] = half4(q4[i4]);
- }
-
- if (tiitg < NHPTG) {
- sw[tiitg] = ((device const float *) pw)[i_head + tiitg];
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
-
- simdgroup_float8x8 mqk = make_filled_simdgroup_matrix<float, 8>(0.0f);
-
- FOR_UNROLL (short i = 0; i < DK8; ++i) {
- simdgroup_half8x8 mq;
-
- simdgroup_load(mq, sq + 8*i, DK, 0, false);
- simdgroup_multiply_accumulate(mqk, mq, mk[i], mqk);
- }
-
- threadgroup float * pqk = sqk + sgitg*NHPTG*NKPSG;
-
- simdgroup_store(mqk, pqk, NKPSG, 0, false);
- simdgroup_barrier(mem_flags::mem_threadgroup);
-
- // one lane per key: ReLU, apply the head weight and accumulate over the head tile
- if (tiisg < NKPSG) {
- FOR_UNROLL (short ih = 0; ih < NHPTG; ++ih) {
- score += max(pqk[ih*NKPSG + tiisg], 0.0f)*sw[ih];
- }
- }
-
- threadgroup_barrier(mem_flags::mem_threadgroup);
- }
-
- if (tiisg < NKPSG) {
- const int ik = i_kv + tiisg;
- if (ik < args.n_kv) {
- device const half * pm = (device const half *) (m + i_batch*args.nbm1 + (i_stream % args.mask_ne3)*args.nbm3);
- device float * pd = (device float *) (dst + i_batch*args.nb1 + i_stream*args.nb3);
-
- pd[ik] = score + (float) pm[ik];
- }
- }
- }
-}
-
-typedef decltype(kernel_lightning_indexer<half4x4, 1, dequantize_f16>) kernel_lightning_indexer_t;
-
-template [[host_name("kernel_lightning_indexer_f32")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<float4x4, 1, dequantize_f32>;
-template [[host_name("kernel_lightning_indexer_f16")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<half4x4, 1, dequantize_f16>;
-
-#if defined(GGML_METAL_HAS_BF16)
-template [[host_name("kernel_lightning_indexer_bf16")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<bfloat4x4, 1, dequantize_bf16>;
-#endif
-
-template [[host_name("kernel_lightning_indexer_q4_0")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q4_0, 2, dequantize_q4_0>;
-template [[host_name("kernel_lightning_indexer_q4_1")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q4_1, 2, dequantize_q4_1>;
-template [[host_name("kernel_lightning_indexer_q5_0")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q5_0, 2, dequantize_q5_0>;
-template [[host_name("kernel_lightning_indexer_q5_1")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q5_1, 2, dequantize_q5_1>;
-template [[host_name("kernel_lightning_indexer_q8_0")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q8_0, 2, dequantize_q8_0>;
-
-kernel void kernel_dsv4_hc_comb_f32(
- constant ggml_metal_kargs_dsv4_hc_comb & args,
- device const char * mixes,
- device const char * scale,
- device const char * base,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- constexpr ushort hc = 4;
- constexpr ushort comb_offset = 2*hc;
-
- const int it = tgpig.x*ntg.y + sgitg;
- if (it >= args.n_tokens) {
- return;
- }
-
- float scale_lane = 0.0f;
- if (tiisg == 0) {
- scale_lane = *(device const float *) (scale + 2*args.nb_s0);
- }
- const float scale_comb = simd_shuffle(scale_lane, 0);
-
- float v = 0.0f;
- if (tiisg < hc*hc) {
- v = *(device const float *) (mixes + (comb_offset + tiisg)*args.nb_m0 + it*args.nb_m1)*scale_comb
- + *(device const float *) (base + (comb_offset + tiisg)*args.nb_b0);
- }
-
- // Softmax across destinations (the four contiguous lanes for each source).
- float vmax = max(v, simd_shuffle_xor(v, 1));
- vmax = max(vmax, simd_shuffle_xor(vmax, 2));
- v = exp(v - vmax);
-
- float sum = v + simd_shuffle_xor(v, 1);
- sum += simd_shuffle_xor(sum, 2);
- v = v/sum + args.eps;
-
- // Normalize columns: equal destination indices are four lanes apart.
- sum = v + simd_shuffle_xor(v, 4);
- sum += simd_shuffle_xor(sum, 8);
- v /= sum + args.eps;
-
- for (int i = 1; i < args.n_iter; ++i) {
- sum = v + simd_shuffle_xor(v, 1);
- sum += simd_shuffle_xor(sum, 2);
- v /= sum + args.eps;
-
- sum = v + simd_shuffle_xor(v, 4);
- sum += simd_shuffle_xor(sum, 8);
- v /= sum + args.eps;
- }
-
- if (tiisg < hc*hc) {
- const ushort idst = tiisg & 3;
- const ushort isrc = tiisg >> 2;
- *(device float *) (dst + idst*args.nb_d0 + isrc*args.nb_d1 + it*args.nb_d2) = v;
- }
-}
-
-kernel void kernel_dsv4_hc_pre_f32(
- constant ggml_metal_kargs_dsv4_hc_pre & args,
- device const char * x,
- device const char * weights,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- constexpr ushort hc = 4;
-
- const int it = tgpig.y;
- const int i0 = ((int) tgpig.x*ntg.y + sgitg)*32 + tiisg;
-
- float weight_lane = 0.0f;
- if (tiisg < hc) {
- weight_lane = *(device const float *) (weights + tiisg*args.nb_w0 + it*args.nb_w1);
- }
-
- float w[hc];
- FOR_UNROLL (ushort ih = 0; ih < hc; ++ih) {
- w[ih] = simd_shuffle(weight_lane, ih);
- }
-
- if (i0 >= args.n_embd) {
- return;
- }
-
- device const char * xb = x + i0*args.nb_x0 + it*args.nb_x2;
- float result = 0.0f;
- FOR_UNROLL (ushort ih = 0; ih < hc; ++ih) {
- result = fma(*(device const float *) (xb + ih*args.nb_x1), w[ih], result);
- }
-
- *(device float *) (dst + i0*args.nb_d0 + it*args.nb_d1) = result;
-}
-
-kernel void kernel_dsv4_hc_post_f32(
- constant ggml_metal_kargs_dsv4_hc_post & args,
- device const char * x,
- device const char * residual,
- device const char * post,
- device const char * comb,
- device char * dst,
- uint3 tgpig[[threadgroup_position_in_grid]],
- ushort tiisg[[thread_index_in_simdgroup]],
- ushort sgitg[[simdgroup_index_in_threadgroup]],
- ushort3 ntg[[threads_per_threadgroup]]) {
- constexpr ushort hc = 4;
-
- const int it = tgpig.y;
- const int i0 = ((int) tgpig.x*ntg.y + sgitg)*32 + tiisg;
-
- float coeff_lane = 0.0f;
- if (tiisg < hc) {
- coeff_lane = *(device const float *) (post + tiisg*args.nb_p0 + it*args.nb_p1);
- } else if (tiisg < hc + hc*hc) {
- const ushort idx = tiisg - hc;
- const ushort idst = idx & 3;
- const ushort isrc = idx >> 2;
- coeff_lane = *(device const float *) (comb + idst*args.nb_c0 + isrc*args.nb_c1 + it*args.nb_c2);
- }
-
- float post_reg[hc];
- float comb_reg[hc][hc];
- FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
- post_reg[idst] = simd_shuffle(coeff_lane, idst);
- }
- FOR_UNROLL (ushort isrc = 0; isrc < hc; ++isrc) {
- FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
- comb_reg[isrc][idst] = simd_shuffle(coeff_lane, hc + idst + hc*isrc);
- }
- }
-
- if (i0 >= args.n_embd) {
- return;
- }
-
- const float xv = *(device const float *) (x + i0*args.nb_x0 + it*args.nb_x1);
- float result[hc];
- FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
- result[idst] = xv*post_reg[idst];
- }
-
- device const char * rb = residual + i0*args.nb_r0 + it*args.nb_r2;
- FOR_UNROLL (ushort isrc = 0; isrc < hc; ++isrc) {
- const float rv = *(device const float *) (rb + isrc*args.nb_r1);
- FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
- result[idst] = fma(rv, comb_reg[isrc][idst], result[idst]);
- }
- }
-
- FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
- *(device float *) (dst + i0*args.nb_d0 + idst*args.nb_d1 + it*args.nb_d2) = result[idst];
- }
-}
diff --git a/ggml/src/ggml-metal/kernels/argsort.metal b/ggml/src/ggml-metal/kernels/argsort.metal
new file mode 100644
index 0000000000..7d144fbd75
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/argsort.metal
@@ -0,0 +1,232 @@
+#include "common.h"
+
+// bitonic sort implementation following the CUDA kernels as reference
+typedef void (argsort_t)(
+ constant ggml_metal_kargs_argsort & args,
+ device const char * src0,
+ device int32_t * dst,
+ threadgroup int32_t * shmem_i32 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]);
+
+template<ggml_sort_order order>
+kernel void kernel_argsort_f32_i32(
+ constant ggml_metal_kargs_argsort & args,
+ device const char * src0,
+ device int32_t * dst,
+ threadgroup int32_t * shmem_i32 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ // bitonic sort
+ const int col = tpitg[0];
+ const int ib = tgpig[0] / args.ne01;
+
+ const int i00 = ib*ntg.x;
+ const int i01 = tgpig[0] % args.ne01;
+ const int i02 = tgpig[1];
+ const int i03 = tgpig[2];
+
+ device const float * src0_row = (device const float *) (src0 + args.nb01*i01 + args.nb02*i02 + args.nb03*i03);
+
+ // initialize indices
+ shmem_i32[col] = i00 + col;
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (int k = 2; k <= ntg.x; k *= 2) {
+ for (int j = k / 2; j > 0; j /= 2) {
+ int ixj = col ^ j;
+ if (ixj > col) {
+ if ((col & k) == 0) {
+ if (shmem_i32[col] >= args.ne00 ||
+ (shmem_i32[ixj] < args.ne00 && (order == GGML_SORT_ORDER_ASC ?
+ src0_row[shmem_i32[col]] > src0_row[shmem_i32[ixj]] :
+ src0_row[shmem_i32[col]] < src0_row[shmem_i32[ixj]]))
+ ) {
+ SWAP(shmem_i32[col], shmem_i32[ixj]);
+ }
+ } else {
+ if (shmem_i32[ixj] >= args.ne00 ||
+ (shmem_i32[col] < args.ne00 && (order == GGML_SORT_ORDER_ASC ?
+ src0_row[shmem_i32[col]] < src0_row[shmem_i32[ixj]] :
+ src0_row[shmem_i32[col]] > src0_row[shmem_i32[ixj]]))
+ ) {
+ SWAP(shmem_i32[col], shmem_i32[ixj]);
+ }
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+ }
+
+ const int64_t i0 = ib*args.top_k;
+
+ // copy the result to dst without the padding
+ if (i0 + col < args.ne0 && col < args.top_k) {
+ dst += i0 + args.ne0*i01 + args.ne0*args.ne1*i02 + args.ne0*args.ne1*args.ne2*i03;
+
+ dst[col] = shmem_i32[col];
+ }
+}
+
+template [[host_name("kernel_argsort_f32_i32_asc")]] kernel argsort_t kernel_argsort_f32_i32<GGML_SORT_ORDER_ASC>;
+template [[host_name("kernel_argsort_f32_i32_desc")]] kernel argsort_t kernel_argsort_f32_i32<GGML_SORT_ORDER_DESC>;
+
+typedef void (argsort_merge_t)(
+ constant ggml_metal_kargs_argsort_merge & args,
+ device const char * src0,
+ device const int32_t * tmp,
+ device int32_t * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]);
+
+template<ggml_sort_order order>
+kernel void kernel_argsort_merge_f32_i32(
+ constant ggml_metal_kargs_argsort_merge & args,
+ device const char * src0,
+ device const int32_t * tmp,
+ device int32_t * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+
+ const int im = tgpig[0] / args.ne01;
+ const int i01 = tgpig[0] % args.ne01;
+ const int i02 = tgpig[1];
+ const int i03 = tgpig[2];
+
+ const int start = im * (2 * args.len);
+
+ const int len0 = MIN(args.len, MAX(0, args.ne0 - (int)(start)));
+ const int len1 = MIN(args.len, MAX(0, args.ne0 - (int)(start + args.len)));
+
+ const int total = len0 + len1;
+
+ device const int32_t * tmp0 = tmp + start
+ + i01*args.ne0
+ + i02*args.ne0*args.ne01
+ + i03*args.ne0*args.ne01*args.ne02;
+
+ device const int32_t * tmp1 = tmp0 + args.len;
+
+ dst += start
+ + i01*args.top_k
+ + i02*args.top_k*args.ne01
+ + i03*args.top_k*args.ne01*args.ne02;
+
+ device const float * src0_row = (device const float *)(src0
+ + args.nb01*i01
+ + args.nb02*i02
+ + args.nb03*i03);
+
+ if (total == 0) {
+ return;
+ }
+
+ const int chunk = (total + ntg.x - 1) / ntg.x;
+
+ const int k0 = tpitg.x * chunk;
+ const int k1 = MIN(MIN(k0 + chunk, total), args.top_k);
+
+ if (k0 >= args.top_k) {
+ return;
+ }
+
+ if (k0 >= total) {
+ return;
+ }
+
+ int low = k0 > len1 ? k0 - len1 : 0;
+ int high = MIN(k0, len0);
+
+ // binary-search partition (i, j) such that i + j = k
+ while (low < high) {
+ const int mid = (low + high) >> 1;
+
+ const int32_t idx0 = tmp0[mid];
+ const int32_t idx1 = tmp1[k0 - mid - 1];
+
+ const float val0 = src0_row[idx0];
+ const float val1 = src0_row[idx1];
+
+ bool take_left;
+ if (order == GGML_SORT_ORDER_ASC) {
+ take_left = (val0 <= val1);
+ } else {
+ take_left = (val0 >= val1);
+ }
+
+ if (take_left) {
+ low = mid + 1;
+ } else {
+ high = mid;
+ }
+ }
+
+ int i = low;
+ int j = k0 - i;
+
+ // keep the merge fronts into registers
+ int32_t idx0 = 0;
+ float val0 = 0.0f;
+ if (i < len0) {
+ idx0 = tmp0[i];
+ val0 = src0_row[idx0];
+ }
+
+ int32_t idx1 = 0;
+ float val1 = 0.0f;
+ if (j < len1) {
+ idx1 = tmp1[j];
+ val1 = src0_row[idx1];
+ }
+
+ for (int k = k0; k < k1; ++k) {
+ int32_t out_idx;
+
+ if (i >= len0) {
+ while (k < k1) {
+ dst[k++] = tmp1[j++];
+ }
+ break;
+ } else if (j >= len1) {
+ while (k < k1) {
+ dst[k++] = tmp0[i++];
+ }
+ break;
+ } else {
+ bool take_left;
+
+ if (order == GGML_SORT_ORDER_ASC) {
+ take_left = (val0 <= val1);
+ } else {
+ take_left = (val0 >= val1);
+ }
+
+ if (take_left) {
+ out_idx = idx0;
+ ++i;
+ if (i < len0) {
+ idx0 = tmp0[i];
+ val0 = src0_row[idx0];
+ }
+ } else {
+ out_idx = idx1;
+ ++j;
+ if (j < len1) {
+ idx1 = tmp1[j];
+ val1 = src0_row[idx1];
+ }
+ }
+ }
+
+ dst[k] = out_idx;
+ }
+}
+
+template [[host_name("kernel_argsort_merge_f32_i32_asc")]] kernel argsort_merge_t kernel_argsort_merge_f32_i32<GGML_SORT_ORDER_ASC>;
+template [[host_name("kernel_argsort_merge_f32_i32_desc")]] kernel argsort_merge_t kernel_argsort_merge_f32_i32<GGML_SORT_ORDER_DESC>;
diff --git a/ggml/src/ggml-metal/kernels/binbcast.metal b/ggml/src/ggml-metal/kernels/binbcast.metal
new file mode 100644
index 0000000000..7c7ab9b5eb
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/binbcast.metal
@@ -0,0 +1,228 @@
+#include "common.h"
+
+// OP: 0 - add, 1 - sub, 2 - mul, 3 - div
+constant short FC_bin_op [[function_constant(FC_BIN + 0)]];
+constant short FC_bin_f [[function_constant(FC_BIN + 1)]];
+constant bool FC_bin_rb [[function_constant(FC_BIN + 2)]];
+constant bool FC_bin_cb [[function_constant(FC_BIN + 3)]];
+
+template <typename T0, typename T1, typename T>
+kernel void kernel_bin_fuse_impl(
+ constant ggml_metal_kargs_bin & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+#define FC_OP FC_bin_op
+#define FC_F FC_bin_f
+#define FC_RB FC_bin_rb
+#define FC_CB FC_bin_cb
+
+ if (FC_RB) {
+ // row broadcast
+ const uint i0 = tgpig.y*args.ne00 + tgpig.x;
+ const uint i1 = FC_CB ? tgpig.x%args.ne10 : tgpig.x;
+
+ device const T0 * src0_row = (device const T0 *) (src0);
+ device T * dst_row = (device T *) (dst);
+
+ if (FC_F == 1) {
+ device const T1 * src1_row = (device const T1 *) (src1 + args.o1[0]);
+
+ if (FC_OP == 0) {
+ dst_row[i0] = src0_row[i0] + src1_row[i1];
+ }
+
+ if (FC_OP == 1) {
+ dst_row[i0] = src0_row[i0] - src1_row[i1];
+ }
+
+ if (FC_OP == 2) {
+ dst_row[i0] = src0_row[i0] * src1_row[i1];
+ }
+
+ if (FC_OP == 3) {
+ dst_row[i0] = src0_row[i0] / src1_row[i1];
+ }
+ } else {
+ T0 res = src0_row[i0];
+
+ if (FC_OP == 0) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res += ((device const T1 *) (src1 + args.o1[j]))[i1];
+ }
+ }
+
+ if (FC_OP == 1) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res -= ((device const T1 *) (src1 + args.o1[j]))[i1];
+ }
+ }
+
+ if (FC_OP == 2) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res *= ((device const T1 *) (src1 + args.o1[j]))[i1];
+ }
+ }
+
+ if (FC_OP == 3) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res /= ((device const T1 *) (src1 + args.o1[j]))[i1];
+ }
+ }
+
+ dst_row[i0] = res;
+ }
+ } else {
+ const int i03 = tgpig.z;
+ const int i02 = tgpig.y;
+ const int i01 = tgpig.x;
+
+ if (i01 >= args.ne01) {
+ return;
+ }
+
+ const int i13 = i03%args.ne13;
+ const int i12 = i02%args.ne12;
+ const int i11 = i01%args.ne11;
+
+ device const T0 * src0_ptr = (device const T0 *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + args.offs);
+ device T * dst_ptr = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1 + args.offs);
+
+ if (FC_F == 1) {
+ device const T1 * src1_ptr = (device const T1 *) (src1 + args.o1[0] + i13*args.nb13 + i12*args.nb12 + i11*args.nb11);
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ const int i10 = FC_CB ? i0%args.ne10 : i0;
+
+ if (FC_OP == 0) {
+ dst_ptr[i0] = src0_ptr[i0] + src1_ptr[i10];
+ }
+
+ if (FC_OP == 1) {
+ dst_ptr[i0] = src0_ptr[i0] - src1_ptr[i10];
+ }
+
+ if (FC_OP == 2) {
+ dst_ptr[i0] = src0_ptr[i0] * src1_ptr[i10];
+ }
+
+ if (FC_OP == 3) {
+ dst_ptr[i0] = src0_ptr[i0] / src1_ptr[i10];
+ }
+ }
+ } else {
+ device const T1 * src1_ptr[8];
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ src1_ptr[j] = (device const T1 *) (src1 + args.o1[j] + i13*args.nb13 + i12*args.nb12 + i11*args.nb11);
+ }
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ const int i10 = FC_CB ? i0%args.ne10 : i0;
+
+ T res = src0_ptr[i0];
+
+ if (FC_OP == 0) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res += src1_ptr[j][i10];
+ }
+ }
+
+ if (FC_OP == 1) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res -= src1_ptr[j][i10];
+ }
+ }
+
+ if (FC_OP == 2) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res *= src1_ptr[j][i10];
+ }
+ }
+
+ if (FC_OP == 3) {
+ FOR_UNROLL (short j = 0; j < FC_F; ++j) {
+ res /= src1_ptr[j][i10];
+ }
+ }
+
+ dst_ptr[i0] = res;
+ }
+ }
+ }
+
+#undef FC_OP
+#undef FC_F
+#undef FC_RB
+#undef FC_CB
+}
+
+typedef decltype(kernel_bin_fuse_impl<float, float, float>) kernel_bin_fuse_t;
+
+template [[host_name("kernel_bin_fuse_f32_f32_f32")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<float, float, float>;
+template [[host_name("kernel_bin_fuse_f32_f32_f32_4")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<float4, float4, float4>;
+template [[host_name("kernel_bin_fuse_f16_f16_f16")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<half, half, half>;
+template [[host_name("kernel_bin_fuse_f16_f16_f16_4")]] kernel kernel_bin_fuse_t kernel_bin_fuse_impl<half4, half4, half4>;
+
+kernel void kernel_add_id(
+ constant ggml_metal_kargs_add_id & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * src2,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int i1 = tgpig.x;
+ const int i2 = tgpig.y;
+
+ const int i11 = *((device const int32_t *) (src2 + i1*sizeof(int32_t) + i2*args.nb21));
+
+ const size_t nb1 = args.ne0 * sizeof(float);
+ const size_t nb2 = args.ne1 * nb1;
+
+ device float * dst_row = (device float *)((device char *)dst + i1*nb1 + i2*nb2);
+ device const float * src0_row = (device const float *)((device char *)src0 + i1*args.nb01 + i2*args.nb02);
+ device const float * src1_row = (device const float *)((device char *)src1 + i11*args.nb11);
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ dst_row[i0] = src0_row[i0] + src1_row[i0];
+ }
+}
+
+template<typename T>
+kernel void kernel_repeat(
+ constant ggml_metal_kargs_repeat & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int i3 = tgpig.z;
+ const int i2 = tgpig.y;
+ const int i1 = tgpig.x;
+
+ const int i03 = i3%args.ne03;
+ const int i02 = i2%args.ne02;
+ const int i01 = i1%args.ne01;
+
+ device const char * src0_ptr = src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01;
+ device char * dst_ptr = dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1;
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ const int i00 = i0%args.ne00;
+ *((device T *)(dst_ptr + i0*args.nb0)) = *((device T *)(src0_ptr + i00*args.nb00));
+ }
+}
+
+typedef decltype(kernel_repeat<float>) kernel_repeat_t;
+
+template [[host_name("kernel_repeat_f32")]] kernel kernel_repeat_t kernel_repeat<float>;
+template [[host_name("kernel_repeat_f16")]] kernel kernel_repeat_t kernel_repeat<half>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_repeat_bf16")]] kernel kernel_repeat_t kernel_repeat<bfloat>;
+#endif
+template [[host_name("kernel_repeat_i32")]] kernel kernel_repeat_t kernel_repeat<int>;
+template [[host_name("kernel_repeat_i16")]] kernel kernel_repeat_t kernel_repeat<short>;
diff --git a/ggml/src/ggml-metal/kernels/common.h b/ggml/src/ggml-metal/kernels/common.h
new file mode 100644
index 0000000000..c4d6743944
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/common.h
@@ -0,0 +1,126 @@
+#pragma once
+
+#include "ggml-metal-impl.h"
+
+#include <metal_stdlib>
+
+#ifdef GGML_METAL_HAS_TENSOR
+#include <metal_tensor>
+
+#include <MetalPerformancePrimitives/MetalPerformancePrimitives.h>
+#endif
+
+using namespace metal;
+
+#define MAX(x, y) ((x) > (y) ? (x) : (y))
+#define MIN(x, y) ((x) < (y) ? (x) : (y))
+#define SWAP(x, y) { auto tmp = (x); (x) = (y); (y) = tmp; }
+
+#define PAD2(x, n) (((x) + (n) - 1) & ~((n) - 1))
+
+#define FOR_UNROLL(x) _Pragma("clang loop unroll(full)") for (x)
+
+#define N_SIMDWIDTH 32 // assuming SIMD group size is 32
+
+// ref: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf
+//
+// cmd:
+// .../usr/bin/metal -dM -E -c ggml/src/ggml-metal/kernels/<src>.metal
+// .../usr/bin/metal -dM -E -c -target air64-apple-ios14.0 ggml/src/ggml-metal/kernels/<src>.metal
+//
+#if __METAL_VERSION__ < 310 && defined(GGML_METAL_HAS_BF16)
+#undef GGML_METAL_HAS_BF16
+#endif
+
+#if defined(GGML_METAL_HAS_BF16)
+typedef matrix<bfloat, 4, 4> bfloat4x4;
+typedef matrix<bfloat, 2, 4> bfloat2x4;
+#endif
+
+constexpr constant static float kvalues_iq4nl_f[16] = {
+ -127.f, -104.f, -83.f, -65.f, -49.f, -35.f, -22.f, -10.f, 1.f, 13.f, 25.f, 38.f, 53.f, 69.f, 89.f, 113.f
+};
+
+constexpr constant static float kvalues_mxfp4_f[16] = {
+ 0, .5f, 1.f, 1.5f, 2.f, 3.f, 4.f, 6.f, -0, -.5f, -1.f, -1.5f, -2.f, -3.f, -4.f, -6.f
+};
+
+static inline int best_index_int8(int n, constant float * val, float x) {
+ if (x <= val[0]) return 0;
+ if (x >= val[n-1]) return n-1;
+ int ml = 0, mu = n-1;
+ while (mu-ml > 1) {
+ int mav = (ml+mu)/2;
+ if (x < val[mav]) mu = mav; else ml = mav;
+ }
+ return x - val[mu-1] < val[mu] - x ? mu-1 : mu;
+}
+
+static inline float e8m0_to_fp32(uint8_t x) {
+ uint32_t bits;
+
+ if (x == 0) {
+ bits = 0x00400000;
+ } else {
+ bits = (uint32_t) x << 23;
+ }
+
+ return as_type<float>(bits);
+}
+
+static inline float dot(float x, float y) {
+ return x*y;
+}
+
+static inline float sum(float x) {
+ return x;
+}
+
+static inline float sum(float4 x) {
+ return x[0] + x[1] + x[2] + x[3];
+}
+
+enum ggml_sort_order {
+ GGML_SORT_ORDER_ASC,
+ GGML_SORT_ORDER_DESC,
+};
+
+constant float GELU_COEF_A = 0.044715f;
+constant float GELU_QUICK_COEF = -1.702f;
+constant float SQRT_2_OVER_PI = 0.79788456080286535587989211986876f;
+constant float SQRT_2_INV = 0.70710678118654752440084436210484f;
+
+// based on Abramowitz and Stegun formula 7.1.26 or similar Hastings' approximation
+// ref: https://www.johndcook.com/blog/python_erf/
+constant float p_erf = 0.3275911f;
+constant float a1_erf = 0.254829592f;
+constant float a2_erf = -0.284496736f;
+constant float a3_erf = 1.421413741f;
+constant float a4_erf = -1.453152027f;
+constant float a5_erf = 1.061405429f;
+
+template<typename T>
+inline T erf_approx(T x) {
+ T sign_x = sign(x);
+ x = fabs(x);
+ T t = 1.0f / (1.0f + p_erf * x);
+ T y = 1.0f - (((((a5_erf * t + a4_erf) * t) + a3_erf) * t + a2_erf) * t + a1_erf) * t * exp(-x * x);
+ return sign_x * y;
+}
+
+template<typename T> T elu_approx(T x);
+
+template<> inline float elu_approx<float>(float x) {
+ return (x > 0.f) ? x : (exp(x) - 1);
+}
+
+template<> inline float4 elu_approx<float4>(float4 x) {
+ float4 res;
+
+ res[0] = (x[0] > 0.0f) ? x[0] : (exp(x[0]) - 1.0f);
+ res[1] = (x[1] > 0.0f) ? x[1] : (exp(x[1]) - 1.0f);
+ res[2] = (x[2] > 0.0f) ? x[2] : (exp(x[2]) - 1.0f);
+ res[3] = (x[3] > 0.0f) ? x[3] : (exp(x[3]) - 1.0f);
+
+ return res;
+}
diff --git a/ggml/src/ggml-metal/kernels/conv.metal b/ggml/src/ggml-metal/kernels/conv.metal
new file mode 100644
index 0000000000..5685b5cd49
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/conv.metal
@@ -0,0 +1,723 @@
+#include "common.h"
+
+typedef void (im2col_t)(
+ constant ggml_metal_kargs_im2col & args,
+ device const float * x,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template <typename T>
+kernel void kernel_im2col(
+ constant ggml_metal_kargs_im2col & args,
+ device const float * x,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+// const int64_t IC = tgpg[0];
+ const int64_t OH = tgpg[1];
+ const int64_t OW = tgpg[2];
+
+ const int64_t KH = ntg[1];
+ const int64_t KW = ntg[2];
+
+ int64_t in = tpitg[0];
+ const int64_t ikh = tpitg[1];
+ const int64_t ikw = tpitg[2];
+
+ const int64_t iic = tgpig[0];
+ const int64_t ioh = tgpig[1];
+ const int64_t iow = tgpig[2];
+
+ const int64_t iiw = iow*args.s0 + ikw*args.d0 - args.p0;
+ const int64_t iih = ioh*args.s1 + ikh*args.d1 - args.p1;
+
+ int64_t offset_dst = (in*OH*OW + ioh*OW + iow)*args.CHW + (iic*(KH*KW) + ikh*KW + ikw);
+
+ device T * pdst = (device T *) (dst);
+
+ if (iih < 0 || iih >= args.IH || iiw < 0 || iiw >= args.IW) {
+ while (in < args.N) {
+ pdst[offset_dst] = 0.0f;
+ offset_dst += ntg[0]*args.CHW*OH*OW;
+
+ in += ntg[0];
+ }
+ } else {
+ int64_t offset_src = in*args.ofs0 + iic*args.ofs1 + iih*args.IW + iiw;
+
+ while (in < args.N) {
+ pdst[offset_dst] = x[offset_src];
+
+ offset_dst += ntg[0]*args.CHW*OH*OW;
+ offset_src += ntg[0]*args.ofs0;
+
+ in += ntg[0];
+ }
+ }
+}
+
+template [[host_name("kernel_im2col_f32")]] kernel im2col_t kernel_im2col<float>;
+template [[host_name("kernel_im2col_f16")]] kernel im2col_t kernel_im2col<half>;
+
+// TODO: optimize
+typedef void (im2col_ext_t)(
+ constant ggml_metal_kargs_im2col & args,
+ device const float * x,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template <typename T>
+kernel void kernel_im2col_ext(
+ constant ggml_metal_kargs_im2col & args,
+ device const float * x,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]], // tgpg[0] = D x IC x KH x KW, CHW = IC x KH x KW
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) { // [M, 1, 1]
+ const int64_t KHW = (int64_t)args.KHW;
+
+ const int64_t d = tgpig[0] / args.CHW;
+ const int64_t chw = tgpig[0] % args.CHW;
+ const int64_t tgpig_0 = chw / KHW; // 0 ~ (IC - 1)
+ const int64_t HW = tgpig[0] % KHW;
+
+ const int64_t tpitg_0 = (d * ntg[0]) + tpitg[0];
+ if (tpitg_0 >= args.N) {
+ return;
+ }
+
+ const int64_t tpitg_1 = HW / args.KW;
+ const int64_t tpitg_2 = HW % args.KW;
+
+ const int64_t iiw = tgpig[2] * args.s0 + tpitg_2 * args.d0 - args.p0;
+ const int64_t iih = tgpig[1] * args.s1 + tpitg_1 * args.d1 - args.p1;
+
+ const int64_t offset_dst =
+ (tpitg_0 * tgpg[1] * tgpg[2] + tgpig[1] * tgpg[2] + tgpig[2]) * args.CHW +
+ (tgpig_0 * KHW + tpitg_1 * args.KW + tpitg_2);
+
+ device T * pdst = (device T *) (dst);
+
+ if (iih < 0 || iih >= args.IH || iiw < 0 || iiw >= args.IW) {
+ pdst[offset_dst] = 0.0f;
+ } else {
+ const int64_t offset_src = tpitg_0 * args.ofs0 + tgpig_0 * args.ofs1;
+ pdst[offset_dst] = x[offset_src + iih * args.IW + iiw];
+ }
+}
+
+template [[host_name("kernel_im2col_ext_f32")]] kernel im2col_ext_t kernel_im2col_ext<float>;
+template [[host_name("kernel_im2col_ext_f16")]] kernel im2col_ext_t kernel_im2col_ext<half>;
+
+template <typename T>
+kernel void kernel_col2im_1d(
+ constant ggml_metal_kargs_col2im_1d & args,
+ device const T * col,
+ device T * dst,
+ uint tgpig [[threadgroup_position_in_grid]],
+ uint tpitg [[thread_position_in_threadgroup]],
+ uint ntg [[threads_per_threadgroup]]) {
+
+ const int idx = tgpig * ntg + tpitg;
+ if (idx >= args.T_out * args.OC) {
+ return;
+ }
+
+ const int t_out = idx % args.T_out;
+ const int oc = idx / args.T_out;
+ const int t_abs = t_out + args.p0; // absolute position in uncropped signal
+
+ int t_in_min = (t_abs - args.K + args.s0) / args.s0; // ceil((t_abs - K + 1) / s0)
+ if (t_in_min < 0) {
+ t_in_min = 0;
+ }
+ int t_in_max = t_abs / args.s0;
+ if (t_in_max >= args.T_in) {
+ t_in_max = args.T_in - 1;
+ }
+
+ float sum = 0.0f;
+ for (int t_in = t_in_min; t_in <= t_in_max; t_in++) {
+ const int k = t_abs - t_in * args.s0;
+ sum += float(col[(oc * args.K + k) + t_in * args.K_OC]);
+ }
+
+ dst[t_out + oc * args.T_out] = T(sum);
+}
+
+template [[host_name("kernel_col2im_1d_f32")]] kernel void kernel_col2im_1d<float>(constant ggml_metal_kargs_col2im_1d &, device const float *, device float *, uint, uint, uint);
+template [[host_name("kernel_col2im_1d_f16")]] kernel void kernel_col2im_1d<half>(constant ggml_metal_kargs_col2im_1d &, device const half *, device half *, uint, uint, uint);
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_col2im_1d_bf16")]] kernel void kernel_col2im_1d<bfloat>(constant ggml_metal_kargs_col2im_1d &, device const bfloat *, device bfloat *, uint, uint, uint);
+#endif
+
+template <typename TK>
+kernel void kernel_conv_2d(
+ constant ggml_metal_kargs_conv_2d & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const uint threads_per_tg = ntg.x * ntg.y * ntg.z;
+ const uint tg_index = (tgpig.z * tgpg.y + tgpig.y) * tgpg.x + tgpig.x;
+ const uint local_thread = tpitg.z * (ntg.x * ntg.y) + tpitg.y * ntg.x + tpitg.x;
+ const uint thread_index = tg_index * threads_per_tg + local_thread;
+ const uint64_t total_threads = (uint64_t) threads_per_tg * tgpg.x * tgpg.y * tgpg.z;
+ const uint64_t total_outputs = (uint64_t) args.N * args.OC * args.OH * args.OW;
+
+ for (uint64_t index = thread_index; index < total_outputs; index += total_threads) {
+ uint64_t tmp = index;
+
+ const int32_t ow = tmp % args.OW; tmp /= args.OW;
+ const int32_t oh = tmp % args.OH; tmp /= args.OH;
+ const int32_t oc = tmp % args.OC; tmp /= args.OC;
+ const int32_t n = tmp;
+
+ float acc = 0.0f;
+
+ const int32_t base_x = ow*args.s0 - args.p0;
+ const int32_t base_y = oh*args.s1 - args.p1;
+
+ int32_t ky_start = 0;
+ if (base_y < 0) {
+ ky_start = (-base_y + args.d1 - 1)/args.d1;
+ }
+ int32_t ky_end = args.KH;
+ const int32_t y_max = args.IH - 1 - base_y;
+ if (y_max < 0) {
+ ky_end = ky_start;
+ } else if (base_y + (args.KH - 1)*args.d1 >= args.IH) {
+ ky_end = min(ky_end, y_max/args.d1 + 1);
+ }
+
+ int32_t kx_start = 0;
+ if (base_x < 0) {
+ kx_start = (-base_x + args.d0 - 1)/args.d0;
+ }
+ int32_t kx_end = args.KW;
+ const int32_t x_max = args.IW - 1 - base_x;
+ if (x_max < 0) {
+ kx_end = kx_start;
+ } else if (base_x + (args.KW - 1)*args.d0 >= args.IW) {
+ kx_end = min(kx_end, x_max/args.d0 + 1);
+ }
+
+ if (ky_start < ky_end && kx_start < kx_end) {
+ const uint64_t src_base_n = (uint64_t) n * args.nb13;
+ const uint64_t w_base_oc = (uint64_t) oc * args.nb03;
+
+ for (int32_t ic = 0; ic < args.IC; ++ic) {
+ const uint64_t src_base_nc = src_base_n + (uint64_t) ic * args.nb12;
+ const uint64_t w_base_ocic = w_base_oc + (uint64_t) ic * args.nb02;
+
+ for (int32_t ky = ky_start; ky < ky_end; ++ky) {
+ const int32_t iy = base_y + ky*args.d1;
+ const uint64_t src_base_row = src_base_nc + (uint64_t) iy * args.nb11;
+ const uint64_t w_base_row = w_base_ocic + (uint64_t) ky * args.nb01;
+
+ for (int32_t kx = kx_start; kx < kx_end; ++kx) {
+ const int32_t ix = base_x + kx*args.d0;
+ const uint64_t src_offs = src_base_row + (uint64_t) ix * args.nb10;
+ const uint64_t w_offs = w_base_row + (uint64_t) kx * args.nb00;
+
+ const float x = *(device const float *)(src + src_offs);
+ const float w = (float) (*(device const TK *)(weights + w_offs));
+
+ acc += x * w;
+ }
+ }
+ }
+ }
+
+ const uint64_t dst_offs =
+ (uint64_t) n * args.nb3 +
+ (uint64_t) oc * args.nb2 +
+ (uint64_t) oh * args.nb1 +
+ (uint64_t) ow * args.nb0;
+
+ *(device float *)(dst + dst_offs) = acc;
+ }
+}
+
+template [[host_name("kernel_conv_2d_f32_f32")]]
+kernel void kernel_conv_2d<float>(
+ constant ggml_metal_kargs_conv_2d & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template [[host_name("kernel_conv_2d_f16_f32")]]
+kernel void kernel_conv_2d<half>(
+ constant ggml_metal_kargs_conv_2d & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+typedef void (conv_transpose_1d_t)(
+ constant ggml_metal_kargs_conv_transpose_1d & args,
+ device const float * src0,
+ device const float * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]]);
+
+template <typename T>
+kernel void kernel_conv_transpose_1d(
+ constant ggml_metal_kargs_conv_transpose_1d & args,
+ device const T * src0,
+ device const float * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]]) {
+
+ // For output position j on the time axis, only input positions
+ // i such that i*s0 <= j < i*s0 + K
+ // contribute -- i.e. i in [ceil((j - K + 1)/s0), floor(j/s0)]
+ // intersected with [0, IL-1]. That's at most ceil(K/s0) values
+ // (typically 2 for stride==K/2 transposed convs).
+ const int32_t j = tgpig[0];
+ const int32_t s0 = args.s0;
+ const int32_t K = args.K;
+ const int32_t IL = args.IL;
+
+ int32_t i_min;
+ {
+ int32_t a = j - K + 1;
+ i_min = a <= 0 ? 0 : (a + s0 - 1) / s0; // ceil(a/s0) for a>0
+ }
+ int32_t i_max = j / s0;
+ if (i_max > IL - 1) i_max = IL - 1;
+
+ float v = 0.0f;
+ if (i_min <= i_max) {
+ for (int64_t c = 0; c < args.IC; c++) {
+ const int32_t kernel_offset = c * tgpg[1] * K + K * tgpig[1];
+ const int32_t input_offset = c * IL;
+
+ for (int32_t i = i_min; i <= i_max; i++) {
+ v += float(src0[kernel_offset + j - i * s0]) * src1[input_offset + i];
+ }
+ }
+ }
+
+ device float * dst_ptr = (device float *) (dst + tgpig[0] * args.nb0 + tgpig[1] * args.nb1);
+
+ dst_ptr[0] = v;
+}
+
+template [[host_name("kernel_conv_transpose_1d_f32_f32")]]
+kernel void kernel_conv_transpose_1d<float>(
+ constant ggml_metal_kargs_conv_transpose_1d & args,
+ device const float * src0,
+ device const float * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]]);
+
+template [[host_name("kernel_conv_transpose_1d_f16_f32")]]
+kernel void kernel_conv_transpose_1d<half>(
+ constant ggml_metal_kargs_conv_transpose_1d & args,
+ device const half * src0,
+ device const float * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]]);
+
+
+typedef void (conv_transpose_2d_t)(
+ constant ggml_metal_kargs_conv_transpose_2d & args,
+ device const float * src0,
+ device const float * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]]);
+
+template <typename T>
+kernel void kernel_conv_transpose_2d(
+ constant ggml_metal_kargs_conv_transpose_2d & args,
+ device const T * src0,
+ device const float * src1,
+ device char * dst,
+ threadgroup float * shared_sum [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int64_t out_x = tgpig[0];
+ const int64_t out_y = tgpig[1];
+ const int64_t out_c = tgpig[2];
+
+ const int64_t kw = tpitg[0];
+ const int64_t kh = tpitg[1];
+
+ float v = 0.0f;
+
+ for (int64_t in_c = 0; in_c < args.IC; in_c++) {
+ int64_t in_y = out_y - kh;
+
+ if (in_y < 0 || in_y % args.s0) continue;
+
+ in_y /= args.s0;
+
+ if (in_y >= args.IH) continue;
+
+ int64_t in_x = out_x - kw;
+
+ if (in_x < 0 || in_x % args.s0) continue;
+
+ in_x /= args.s0;
+
+ if (in_x >= args.IW) continue;
+
+ const int64_t input_idx = (args.IW * args.IH) * in_c + (args.IW) * in_y + in_x;
+ const int64_t kernel_idx = (args.KH * args.KW * args.OC) * in_c + (args.KH * args.KW) * out_c + (args.KW) * kh + kw;
+
+ v += (float)src0[kernel_idx] * src1[input_idx];
+ }
+
+ const uint tid = tpitg.y * ntg.x + tpitg.x;
+ shared_sum[tid] = v;
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tid == 0) {
+ float total = 0.0f;
+ const uint num_threads = ntg.x * ntg.y;
+ for (uint i = 0; i < num_threads; i++) {
+ total += shared_sum[i];
+ }
+
+ device float * dst_ptr = (device float *) (dst + out_x*args.nb0 + out_y * args.nb1 + out_c*args.nb2);
+ dst_ptr[0] = total;
+ }
+}
+
+template [[host_name("kernel_conv_transpose_2d_f32_f32")]]
+kernel void kernel_conv_transpose_2d<float>(
+ constant ggml_metal_kargs_conv_transpose_2d & args,
+ device const float * src0,
+ device const float * src1,
+ device char * dst,
+ threadgroup float * shared_sum [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template [[host_name("kernel_conv_transpose_2d_f16_f32")]]
+kernel void kernel_conv_transpose_2d<half>(
+ constant ggml_metal_kargs_conv_transpose_2d & args,
+ device const half * src0,
+ device const float * src1,
+ device char * dst,
+ threadgroup float * shared_sum [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+// grid: x = C tile, y = OH, z = OW * N (for channel-contiguous layouts)
+template <typename TK>
+kernel void kernel_conv_2d_dw_tiled(
+ constant ggml_metal_kargs_conv_2d_dw & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int32_t c = (int32_t)(tgpig.x * ntg.x + tpitg.x);
+ if (c >= args.C) {
+ return;
+ }
+
+ const int32_t oh = tgpig.y;
+ const int32_t own = tgpig.z;
+ const int32_t ow = own % args.OW;
+ const int32_t n = own / args.OW;
+
+ const int32_t base_y = oh*args.s1 - args.p1;
+
+ int32_t ky_start = 0;
+ if (base_y < 0) {
+ ky_start = (-base_y + args.d1 - 1)/args.d1;
+ }
+ int32_t ky_end = args.KH;
+ const int32_t y_max = args.IH - 1 - base_y;
+ if (y_max < 0) {
+ ky_end = ky_start;
+ } else if (base_y + (args.KH - 1)*args.d1 >= args.IH) {
+ ky_end = min(ky_end, y_max/args.d1 + 1);
+ }
+
+ const int32_t base_x = ow*args.s0 - args.p0;
+
+ int32_t kx_start = 0;
+ if (base_x < 0) {
+ kx_start = (-base_x + args.d0 - 1)/args.d0;
+ }
+ int32_t kx_end = args.KW;
+ const int32_t x_max = args.IW - 1 - base_x;
+ if (x_max < 0) {
+ kx_end = kx_start;
+ } else if (base_x + (args.KW - 1)*args.d0 >= args.IW) {
+ kx_end = min(kx_end, x_max/args.d0 + 1);
+ }
+
+ float acc = 0.0f;
+
+ if (ky_start < ky_end && kx_start < kx_end) {
+ const uint64_t w_base = (uint64_t) c * args.nb02;
+ const uint64_t src_base = (uint64_t) n * args.nb13 + (uint64_t) c * args.nb12;
+
+ for (int32_t ky = ky_start; ky < ky_end; ++ky) {
+ const int32_t iy = base_y + ky*args.d1;
+ const uint64_t src_row = src_base + (uint64_t) iy * args.nb11;
+ const uint64_t w_row = w_base + (uint64_t) ky * args.nb01;
+
+ for (int32_t kx = kx_start; kx < kx_end; ++kx) {
+ const int32_t ix = base_x + kx*args.d0;
+ const float x = *(device const float *)(src + src_row + (uint64_t) ix * args.nb10);
+ const float w = (float)(*(device const TK *)(weights + w_row + (uint64_t) kx * args.nb00));
+ acc += x * w;
+ }
+ }
+ }
+
+ const uint64_t dst_offs =
+ (uint64_t) n * args.nb3 +
+ (uint64_t) c * args.nb2 +
+ (uint64_t) oh * args.nb1 +
+ (uint64_t) ow * args.nb0;
+
+ *(device float *)(dst + dst_offs) = acc;
+}
+
+// grid: x = OW tile, y = OH, z = C * N (for spatially-contiguous layouts)
+template <typename TK>
+kernel void kernel_conv_2d_dw(
+ constant ggml_metal_kargs_conv_2d_dw & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int32_t oh = tgpig.y;
+ const int32_t cn = tgpig.z;
+ const int32_t c = cn % args.C;
+ const int32_t n = cn / args.C;
+
+ const int32_t base_y = oh*args.s1 - args.p1;
+
+ int32_t ky_start = 0;
+ if (base_y < 0) {
+ ky_start = (-base_y + args.d1 - 1)/args.d1;
+ }
+ int32_t ky_end = args.KH;
+ const int32_t y_max = args.IH - 1 - base_y;
+ if (y_max < 0) {
+ ky_end = ky_start;
+ } else if (base_y + (args.KH - 1)*args.d1 >= args.IH) {
+ ky_end = min(ky_end, y_max/args.d1 + 1);
+ }
+
+ const uint64_t w_base = (uint64_t) c * args.nb02;
+ const uint64_t src_base = (uint64_t) n * args.nb13 + (uint64_t) c * args.nb12;
+
+ const int32_t ow = (int32_t)(tgpig.x * ntg.x + tpitg.x);
+ if (ow >= args.OW) {
+ return;
+ }
+
+ float acc = 0.0f;
+
+ const int32_t base_x = ow*args.s0 - args.p0;
+
+ int32_t kx_start = 0;
+ if (base_x < 0) {
+ kx_start = (-base_x + args.d0 - 1)/args.d0;
+ }
+ int32_t kx_end = args.KW;
+ const int32_t x_max = args.IW - 1 - base_x;
+ if (x_max < 0) {
+ kx_end = kx_start;
+ } else if (base_x + (args.KW - 1)*args.d0 >= args.IW) {
+ kx_end = min(kx_end, x_max/args.d0 + 1);
+ }
+
+ if (ky_start < ky_end && kx_start < kx_end) {
+ for (int32_t ky = ky_start; ky < ky_end; ++ky) {
+ const int32_t iy = base_y + ky*args.d1;
+ const uint64_t src_row = src_base + (uint64_t) iy * args.nb11;
+ const uint64_t w_row = w_base + (uint64_t) ky * args.nb01;
+
+ for (int32_t kx = kx_start; kx < kx_end; ++kx) {
+ const int32_t ix = base_x + kx*args.d0;
+ const float x = *(device const float *)(src + src_row + (uint64_t) ix * args.nb10);
+ const float w = (float)(*(device const TK *)(weights + w_row + (uint64_t) kx * args.nb00));
+ acc += x * w;
+ }
+ }
+ }
+
+ const uint64_t dst_offs =
+ (uint64_t) n * args.nb3 +
+ (uint64_t) c * args.nb2 +
+ (uint64_t) oh * args.nb1 +
+ (uint64_t) ow * args.nb0;
+
+ *(device float *)(dst + dst_offs) = acc;
+}
+
+template [[host_name("kernel_conv_2d_dw_f32_f32")]]
+kernel void kernel_conv_2d_dw<float>(
+ constant ggml_metal_kargs_conv_2d_dw & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template [[host_name("kernel_conv_2d_dw_f16_f32")]]
+kernel void kernel_conv_2d_dw<half>(
+ constant ggml_metal_kargs_conv_2d_dw & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template [[host_name("kernel_conv_2d_dw_tiled_f32_f32")]]
+kernel void kernel_conv_2d_dw_tiled<float>(
+ constant ggml_metal_kargs_conv_2d_dw & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template [[host_name("kernel_conv_2d_dw_tiled_f16_f32")]]
+kernel void kernel_conv_2d_dw_tiled<half>(
+ constant ggml_metal_kargs_conv_2d_dw & args,
+ device const char * weights,
+ device const char * src,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]);
+
+template <typename T>
+kernel void kernel_conv_3d(
+ constant ggml_metal_kargs_conv_3d & args,
+ device const char * src0, // Weights [IC * OC, KD, KH, KW]
+ device const char * src1, // Inputs [IC * N, ID, IH, IW]
+ device char * dst, // Outputs [OC * N, OD, OH, OW]
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]]) {
+
+ // 1. Un-flatten the spatial dimension from Grid X
+ int64_t spatial_idx = tgpig.x * 32 + tpitg.x;
+
+ if (spatial_idx >= args.OW * args.OH * args.OD) {
+ return; // Thread falls outside the spatial volume
+ }
+
+ int64_t od = spatial_idx / (args.OW * args.OH);
+ int64_t oh = (spatial_idx / args.OW) % args.OH;
+ int64_t ow = spatial_idx % args.OW;
+
+ // 2. Map Y to Channels, Z to Batch
+ int64_t oc = tgpig.y;
+ int64_t batch_idx = tgpig.z;
+
+ // 3. Calculate anchor coordinates in the Input volume
+ int64_t i_w_base = ow * args.s0 - args.p0;
+ int64_t i_h_base = oh * args.s1 - args.p1;
+ int64_t i_d_base = od * args.s2 - args.p2;
+
+ float sum = 0.0f;
+
+ // 4. Gather Loop (Iterate over Input Channels -> Depth -> Height -> Width)
+ for (int64_t ic = 0; ic < args.IC; ++ic) {
+
+ // ggml packs batch and channel together in the 4th dimension
+ int64_t src_cn_idx = batch_idx * args.IC + ic;
+ int64_t w_cn_idx = oc * args.IC + ic;
+
+ for (int64_t kz = 0; kz < args.KD; ++kz) {
+ int64_t id = i_d_base + kz * args.d2;
+ if (id < 0 || id >= args.ID) continue; // Boundary check (Padding)
+
+ for (int64_t ky = 0; ky < args.KH; ++ky) {
+ int64_t ih = i_h_base + ky * args.d1;
+ if (ih < 0 || ih >= args.IH) continue;
+
+ for (int64_t kx = 0; kx < args.KW; ++kx) {
+ int64_t iw = i_w_base + kx * args.d0;
+ if (iw < 0 || iw >= args.IW) continue;
+
+ // Convert multi-dimensional coordinates to flat byte offsets
+ int64_t w_idx = kx*args.nb00 + ky*args.nb01 + kz*args.nb02 + w_cn_idx*args.nb03;
+ int64_t i_idx = iw*args.nb10 + ih*args.nb11 + id*args.nb12 + src_cn_idx*args.nb13;
+
+ // Dereference memory and cast weights to f32 if they were f16
+ float w_val = (float)*(device const T*)((device const char*)src0 + w_idx);
+ float i_val = *(device const float*)((device const char*)src1 + i_idx);
+
+ sum += w_val * i_val;
+ }
+ }
+ }
+ }
+
+ // 5. Write the accumulated value out to RAM
+ int64_t dst_cn_idx = batch_idx * args.OC + oc;
+ int64_t d_idx = ow*args.nb0 + oh*args.nb1 + od*args.nb2 + dst_cn_idx*args.nb3;
+
+ *(device float*)(dst + d_idx) = sum;
+}
+
+// Explicit instantiations so the JIT compiler can find them by name
+template [[host_name("kernel_conv_3d_f32_f32")]]
+kernel void kernel_conv_3d<float>(
+ constant ggml_metal_kargs_conv_3d & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]]);
+
+// Explicit instantiation for f16 weights
+template [[host_name("kernel_conv_3d_f16_f32")]]
+kernel void kernel_conv_3d<half>(
+ constant ggml_metal_kargs_conv_3d & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]]);
diff --git a/ggml/src/ggml-metal/kernels/dequantize.h b/ggml/src/ggml-metal/kernels/dequantize.h
new file mode 100644
index 0000000000..0d1429d9d3
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/dequantize.h
@@ -0,0 +1,735 @@
+#pragma once
+
+#include "common.h"
+
+#define GGML_COMMON_DECL_METAL
+#define GGML_COMMON_IMPL_METAL
+#if defined(GGML_METAL_EMBED_LIBRARY)
+__embed_ggml-common.h__
+#else
+#include "ggml-common.h"
+#endif
+
+#define QK_NL 16 // shared by mul_mm and get_rows_q instantiations
+
+// NOTE: this is not dequantizing - we are simply fitting the template
+template <typename type4x4>
+void dequantize_f32(device const float4x4 * src, short il, thread type4x4 & reg) {
+ reg = (type4x4)(*src);
+}
+
+template <typename type4>
+void dequantize_f32_t4(device const float4 * src, short il, thread type4 & reg) {
+ reg = (type4)(*src);
+}
+
+template <typename type4x4>
+void dequantize_f16(device const half4x4 * src, short il, thread type4x4 & reg) {
+ reg = (type4x4)(*src);
+}
+
+template <typename type4>
+void dequantize_f16_t4(device const half4 * src, short il, thread type4 & reg) {
+ reg = (type4)(*(src));
+}
+
+#if defined(GGML_METAL_HAS_BF16)
+template <typename type4x4>
+void dequantize_bf16(device const bfloat4x4 * src, short il, thread type4x4 & reg) {
+ reg = (type4x4)(*src);
+}
+
+template <typename type4>
+void dequantize_bf16_t4(device const bfloat4 * src, short il, thread type4 & reg) {
+ reg = (type4)(*(src));
+}
+#endif
+
+template <typename type4x4>
+void dequantize_q1_0(device const block_q1_0 * xb, short il, thread type4x4 & reg) {
+ device const uint8_t * qs = xb->qs;
+ const float d = xb->d;
+ const float neg_d = -d;
+
+ const int byte_offset = il * 2; // il*16 bits = il*2 bytes
+ const uint8_t b0 = qs[byte_offset];
+ const uint8_t b1 = qs[byte_offset + 1];
+
+ float4x4 reg_f;
+
+ reg_f[0][0] = select(neg_d, d, bool(b0 & 0x01));
+ reg_f[0][1] = select(neg_d, d, bool(b0 & 0x02));
+ reg_f[0][2] = select(neg_d, d, bool(b0 & 0x04));
+ reg_f[0][3] = select(neg_d, d, bool(b0 & 0x08));
+ reg_f[1][0] = select(neg_d, d, bool(b0 & 0x10));
+ reg_f[1][1] = select(neg_d, d, bool(b0 & 0x20));
+ reg_f[1][2] = select(neg_d, d, bool(b0 & 0x40));
+ reg_f[1][3] = select(neg_d, d, bool(b0 & 0x80));
+
+ reg_f[2][0] = select(neg_d, d, bool(b1 & 0x01));
+ reg_f[2][1] = select(neg_d, d, bool(b1 & 0x02));
+ reg_f[2][2] = select(neg_d, d, bool(b1 & 0x04));
+ reg_f[2][3] = select(neg_d, d, bool(b1 & 0x08));
+ reg_f[3][0] = select(neg_d, d, bool(b1 & 0x10));
+ reg_f[3][1] = select(neg_d, d, bool(b1 & 0x20));
+ reg_f[3][2] = select(neg_d, d, bool(b1 & 0x40));
+ reg_f[3][3] = select(neg_d, d, bool(b1 & 0x80));
+
+ reg = (type4x4) reg_f;
+}
+
+template <typename type4>
+void dequantize_q1_0_t4(device const block_q1_0 * xb, short il, thread type4 & reg) {
+ const float d = xb->d;
+ const float neg_d = -d;
+ const int base = il * 4;
+ const uint8_t byte = xb->qs[base / 8];
+ const int s = base % 8;
+
+ float4 reg_f;
+ reg_f[0] = select(neg_d, d, bool((byte >> (s )) & 1));
+ reg_f[1] = select(neg_d, d, bool((byte >> (s + 1)) & 1));
+ reg_f[2] = select(neg_d, d, bool((byte >> (s + 2)) & 1));
+ reg_f[3] = select(neg_d, d, bool((byte >> (s + 3)) & 1));
+
+ reg = (type4) reg_f;
+}
+
+template <typename type4x4>
+void dequantize_q2_0(device const block_q2_0 * xb, short il, thread type4x4 & reg) {
+ device const uint8_t * qs = xb->qs;
+ const float d = xb->d;
+
+ const int byte_offset = il * 4; // il*16 elements = il*4 bytes (4 elements per byte)
+ float4x4 reg_f;
+
+ for (int i = 0; i < 4; i++) {
+ const uint8_t b = qs[byte_offset + i];
+ reg_f[i][0] = ((float)((b >> 0) & 3) - 1.0f) * d;
+ reg_f[i][1] = ((float)((b >> 2) & 3) - 1.0f) * d;
+ reg_f[i][2] = ((float)((b >> 4) & 3) - 1.0f) * d;
+ reg_f[i][3] = ((float)((b >> 6) & 3) - 1.0f) * d;
+ }
+
+ reg = (type4x4) reg_f;
+}
+
+template <typename type4>
+void dequantize_q2_0_t4(device const block_q2_0 * xb, short il, thread type4 & reg) {
+ const float d = xb->d;
+ const uint8_t b = xb->qs[il];
+
+ float4 reg_f;
+ reg_f[0] = ((float)((b >> 0) & 3) - 1.0f) * d;
+ reg_f[1] = ((float)((b >> 2) & 3) - 1.0f) * d;
+ reg_f[2] = ((float)((b >> 4) & 3) - 1.0f) * d;
+ reg_f[3] = ((float)((b >> 6) & 3) - 1.0f) * d;
+
+ reg = (type4) reg_f;
+}
+
+template <typename type4x4>
+void dequantize_q4_0(device const block_q4_0 * xb, short il, thread type4x4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 1);
+ const float d1 = il ? (xb->d / 16.h) : xb->d;
+ const float d2 = d1 / 256.f;
+ const float md = -8.h * xb->d;
+ const ushort mask0 = il ? 0x00F0 : 0x000F;
+ const ushort mask1 = mask0 << 8;
+
+ float4x4 reg_f;
+
+ for (int i = 0; i < 8; i++) {
+ reg_f[i/2][2*(i%2) + 0] = d1 * (qs[i] & mask0) + md;
+ reg_f[i/2][2*(i%2) + 1] = d2 * (qs[i] & mask1) + md;
+ }
+
+ reg = (type4x4) reg_f;
+}
+
+template <typename type4>
+void dequantize_q4_0_t4(device const block_q4_0 * xb, short il, thread type4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 1);
+ const float d1 = (il/4) ? (xb->d / 16.h) : xb->d;
+ const float d2 = d1 / 256.f;
+ const float md = -8.h * xb->d;
+ const ushort mask0 = (il/4) ? 0x00F0 : 0x000F;
+ const ushort mask1 = mask0 << 8;
+
+ for (int i = 0; i < 2; i++) {
+ reg[2*i + 0] = d1 * (qs[2*(il%4) + i] & mask0) + md;
+ reg[2*i + 1] = d2 * (qs[2*(il%4) + i] & mask1) + md;
+ }
+}
+
+
+
+template <typename type4x4>
+void dequantize_q4_1(device const block_q4_1 * xb, short il, thread type4x4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 2);
+ const float d1 = il ? (xb->d / 16.h) : xb->d;
+ const float d2 = d1 / 256.f;
+ const float m = xb->m;
+ const ushort mask0 = il ? 0x00F0 : 0x000F;
+ const ushort mask1 = mask0 << 8;
+
+ float4x4 reg_f;
+
+ for (int i = 0; i < 8; i++) {
+ reg_f[i/2][2*(i%2) + 0] = ((qs[i] & mask0) * d1) + m;
+ reg_f[i/2][2*(i%2) + 1] = ((qs[i] & mask1) * d2) + m;
+ }
+
+ reg = (type4x4) reg_f;
+}
+
+template <typename type4>
+void dequantize_q4_1_t4(device const block_q4_1 * xb, short il, thread type4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 2);
+ const float d1 = (il/4) ? (xb->d / 16.h) : xb->d;
+ const float d2 = d1 / 256.f;
+ const float m = xb->m;
+ const ushort mask0 = (il/4) ? 0x00F0 : 0x000F;
+ const ushort mask1 = mask0 << 8;
+
+ for (int i = 0; i < 2; i++) {
+ reg[2*i + 0] = d1 * (qs[2*(il%4) + i] & mask0) + m;
+ reg[2*i + 1] = d2 * (qs[2*(il%4) + i] & mask1) + m;
+ }
+}
+
+template <typename type4x4>
+void dequantize_q5_0(device const block_q5_0 * xb, short il, thread type4x4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 3);
+ const float d = xb->d;
+ const float md = -16.h * xb->d;
+ const ushort mask = il ? 0x00F0 : 0x000F;
+
+ const uint32_t qh = *((device const uint32_t *)xb->qh);
+
+ const int x_mv = il ? 4 : 0;
+
+ const int gh_mv = il ? 12 : 0;
+ const int gh_bk = il ? 0 : 4;
+
+ float4x4 reg_f;
+
+ for (int i = 0; i < 8; i++) {
+ // extract the 5-th bits for x0 and x1
+ const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
+ const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
+
+ // combine the 4-bits from qs with the 5th bit
+ const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
+ const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
+
+ reg_f[i/2][2*(i%2) + 0] = d * x0 + md;
+ reg_f[i/2][2*(i%2) + 1] = d * x1 + md;
+ }
+
+ reg = (type4x4) reg_f;
+}
+
+template <typename type4>
+void dequantize_q5_0_t4(device const block_q5_0 * xb, short il, thread type4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 3);
+ const float d = xb->d;
+ const float md = -16.h * xb->d;
+ const ushort mask = (il/4) ? 0x00F0 : 0x000F;
+
+ const uint32_t qh = *((device const uint32_t *)xb->qh);
+
+ const int x_mv = (il/4) ? 4 : 0;
+
+ const int gh_mv = (il/4) ? 12 : 0;
+ const int gh_bk = (il/4) ? 0 : 4;
+
+ for (int ii = 0; ii < 2; ii++) {
+ int i = 2*(il%4) + ii;
+
+ // extract the 5-th bits for x0 and x1
+ const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
+ const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
+
+ // combine the 4-bits from qs with the 5th bit
+ const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
+ const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
+
+ reg[2*ii + 0] = d * x0 + md;
+ reg[2*ii + 1] = d * x1 + md;
+ }
+}
+
+template <typename type4x4>
+void dequantize_q5_1(device const block_q5_1 * xb, short il, thread type4x4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 4);
+ const float d = xb->d;
+ const float m = xb->m;
+ const ushort mask = il ? 0x00F0 : 0x000F;
+
+ const uint32_t qh = *((device const uint32_t *)xb->qh);
+
+ const int x_mv = il ? 4 : 0;
+
+ const int gh_mv = il ? 12 : 0;
+ const int gh_bk = il ? 0 : 4;
+
+ float4x4 reg_f;
+
+ for (int i = 0; i < 8; i++) {
+ // extract the 5-th bits for x0 and x1
+ const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
+ const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
+
+ // combine the 4-bits from qs with the 5th bit
+ const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
+ const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
+
+ reg_f[i/2][2*(i%2) + 0] = d * x0 + m;
+ reg_f[i/2][2*(i%2) + 1] = d * x1 + m;
+ }
+
+ reg = (type4x4) reg_f;
+}
+
+template <typename type4>
+void dequantize_q5_1_t4(device const block_q5_1 * xb, short il, thread type4 & reg) {
+ device const uint16_t * qs = ((device const uint16_t *)xb + 4);
+ const float d = xb->d;
+ const float m = xb->m;
+ const ushort mask = (il/4) ? 0x00F0 : 0x000F;
+
+ const uint32_t qh = *((device const uint32_t *)xb->qh);
+
+ const int x_mv = (il/4) ? 4 : 0;
+
+ const int gh_mv = (il/4) ? 12 : 0;
+ const int gh_bk = (il/4) ? 0 : 4;
+
+ for (int ii = 0; ii < 2; ii++) {
+ int i = 2*(il%4) + ii;
+
+ // extract the 5-th bits for x0 and x1
+ const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
+ const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
+
+ // combine the 4-bits from qs with the 5th bit
+ const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
+ const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
+
+ reg[2*ii + 0] = d * x0 + m;
+ reg[2*ii + 1] = d * x1 + m;
+ }
+}
+
+template <typename type4x4>
+void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg) {
+ device const packed_char4 * qs = (device const packed_char4 *) xb->qs;
+ const float d = xb->d;
+
+ float4x4 reg_f;
+
+ for (int i = 0; i < 4; ++i) {
+ reg_f[i] = float4(qs[4*il + i]) * d;
+ }
+
+ reg = (type4x4) reg_f;
+}
+
+template <typename type4>
+void dequantize_q8_0_t4(device const block_q8_0 *xb, short il, thread type4 & reg) {
+ device const packed_char4 * qs = (device const packed_char4 *) xb->qs;
+ const float d = xb->d;
+
+ reg = (type4) (float4(qs[il]) * d);
+}
+
+template <typename type4x4>
+void dequantize_mxfp4(device const block_mxfp4 * xb, short il, thread type4x4 & reg) {
+ device const uint8_t * q2 = (device const uint8_t *)xb->qs;
+
+ const float d = e8m0_to_fp32(xb->e);
+ const uint8_t shr = il >= 1 ? 4 : 0;
+
+ for (int i = 0; i < 4; ++i) {
+ reg[i][0] = d * kvalues_mxfp4_f[(q2[4*i + 0] >> shr) & 0x0F];
+ reg[i][1] = d * kvalues_mxfp4_f[(q2[4*i + 1] >> shr) & 0x0F];
+ reg[i][2] = d * kvalues_mxfp4_f[(q2[4*i + 2] >> shr) & 0x0F];
+ reg[i][3] = d * kvalues_mxfp4_f[(q2[4*i + 3] >> shr) & 0x0F];
+ }
+}
+
+template <typename type4>
+void dequantize_mxfp4_t4(device const block_mxfp4 * xb, short il, thread type4 & reg) {
+ device const uint8_t * q2 = (device const uint8_t *)xb->qs;
+
+ const float d = e8m0_to_fp32(xb->e);
+ const short il4 = il%4;
+
+ const uint8_t shr = il >= 4 ? 4 : 0;
+
+ reg[0] = d * kvalues_mxfp4_f[(q2[4*il4 + 0] >> shr) & 0x0F];
+ reg[1] = d * kvalues_mxfp4_f[(q2[4*il4 + 1] >> shr) & 0x0F];
+ reg[2] = d * kvalues_mxfp4_f[(q2[4*il4 + 2] >> shr) & 0x0F];
+ reg[3] = d * kvalues_mxfp4_f[(q2[4*il4 + 3] >> shr) & 0x0F];
+}
+
+template <typename type4x4>
+void dequantize_q2_K(device const block_q2_K *xb, short il, thread type4x4 & reg) {
+ const float d = xb->d;
+ const float min = xb->dmin;
+ device const uint8_t * q = (device const uint8_t *)xb->qs;
+ float dl, ml;
+ uint8_t sc = xb->scales[il];
+
+ q = q + 32*(il/8) + 16*(il&1);
+ il = (il/2)%4;
+
+ half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
+ uchar mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
+ dl = d * (sc & 0xF) * coef, ml = min * (sc >> 4);
+ for (int i = 0; i < 16; ++i) {
+ reg[i/4][i%4] = dl * (q[i] & mask) - ml;
+ }
+}
+
+template <typename type4x4>
+void dequantize_q3_K(device const block_q3_K *xb, short il, thread type4x4 & reg) {
+ const half d_all = xb->d;
+ device const uint8_t * q = (device const uint8_t *)xb->qs;
+ device const uint8_t * h = (device const uint8_t *)xb->hmask;
+ device const int8_t * scales = (device const int8_t *)xb->scales;
+
+ q = q + 32 * (il/8) + 16 * (il&1);
+ h = h + 16 * (il&1);
+ uint8_t m = 1 << (il/2);
+ uint16_t kmask1 = (il/4)>1 ? ((il/4)>2 ? 192 : 48) : \
+ ((il/4)>0 ? 12 : 3);
+ uint16_t kmask2 = il/8 ? 0xF0 : 0x0F;
+ uint16_t scale_2 = scales[il%8], scale_1 = scales[8 + il%4];
+ int16_t dl_int = (il/4)&1 ? (scale_2&kmask2) | ((scale_1&kmask1) << 2)
+ : (scale_2&kmask2) | ((scale_1&kmask1) << 4);
+ float dl = il<8 ? d_all * (dl_int - 32.f) : d_all * (dl_int / 16.f - 32.f);
+ const float ml = 4.f * dl;
+
+ il = (il/2) & 3;
+ const half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
+ const uint8_t mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
+ dl *= coef;
+
+ for (int i = 0; i < 16; ++i) {
+ reg[i/4][i%4] = dl * (q[i] & mask) - (h[i] & m ? 0 : ml);
+ }
+}
+
+static inline uchar2 get_scale_min_k4_just2(int j, int k, device const uchar * q) {
+ return j < 4 ? uchar2{uchar(q[j+0+k] & 63), uchar(q[j+4+k] & 63)}
+ : uchar2{uchar((q[j+4+k] & 0xF) | ((q[j-4+k] & 0xc0) >> 2)), uchar((q[j+4+k] >> 4) | ((q[j-0+k] & 0xc0) >> 2))};
+}
+
+template <typename type4x4>
+void dequantize_q4_K(device const block_q4_K * xb, short il, thread type4x4 & reg) {
+ device const uchar * q = xb->qs;
+
+ short is = (il/4) * 2;
+ q = q + (il/4) * 32 + 16 * (il&1);
+ il = il & 3;
+ const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
+ const float d = il < 2 ? xb->d : xb->d / 16.h;
+ const float min = xb->dmin;
+ const float dl = d * sc[0];
+ const float ml = min * sc[1];
+
+ const ushort mask = il < 2 ? 0x0F : 0xF0;
+ for (int i = 0; i < 16; ++i) {
+ reg[i/4][i%4] = dl * (q[i] & mask) - ml;
+ }
+}
+
+template <typename type4x4>
+void dequantize_q5_K(device const block_q5_K *xb, short il, thread type4x4 & reg) {
+ device const uint8_t * q = xb->qs;
+ device const uint8_t * qh = xb->qh;
+
+ short is = (il/4) * 2;
+ q = q + 32 * (il/4) + 16 * (il&1);
+ qh = qh + 16 * (il&1);
+ uint8_t ul = 1 << (il/2);
+ il = il & 3;
+ const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
+ const float d = il < 2 ? xb->d : xb->d / 16.f;
+ const float min = xb->dmin;
+ const float dl = d * sc[0];
+ const float ml = min * sc[1];
+
+ const ushort mask = il<2 ? 0x0F : 0xF0;
+ const float qh_val = il<2 ? 16.f : 256.f;
+ for (int i = 0; i < 16; ++i) {
+ reg[i/4][i%4] = dl * ((q[i] & mask) + (qh[i] & ul ? qh_val : 0)) - ml;
+ }
+}
+
+template <typename type4x4>
+void dequantize_q6_K(device const block_q6_K *xb, short il, thread type4x4 & reg) {
+ const half d_all = xb->d;
+ device const uint16_t * ql = (device const uint16_t *)xb->ql;
+ device const uint16_t * qh = (device const uint16_t *)xb->qh;
+ device const int8_t * scales = (device const int8_t *)xb->scales;
+
+ ql = ql + 32*(il/8) + 16*((il/2)&1) + 8*(il&1);
+ qh = qh + 16*(il/8) + 8*(il&1);
+ float sc = scales[(il%2) + 2 * ((il/2))];
+ il = (il/2) & 3;
+
+ const uint32_t kmask1 = il>1 ? (il>2 ? 0xC0C0C0C0 : 0x30303030) : (il>0 ? 0x0C0C0C0C : 0x03030303);
+ const uint32_t kmask2 = il>1 ? 0xF0F0F0F0 : 0x0F0F0F0F;
+ const float ml = d_all * sc * 32.f;
+ const float dl0 = d_all * sc;
+ const float dl1 = dl0 / 256.f;
+ const float dl2 = dl0 / (256.f * 256.f);
+ const float dl3 = dl0 / (256.f * 256.f * 256.f);
+ const uint8_t shr_h = il>2 ? 2 : 0;
+ const uint8_t shl_h = il>1 ? 0 : (il>0 ? 2 : 4);
+ const uint8_t shr_l = il>1 ? 4 : 0;
+ for (int i = 0; i < 4; ++i) {
+ const uint32_t low = (ql[2*i] | (uint32_t)(ql[2*i+1] << 16)) & kmask2;
+ const uint32_t high = (qh[2*i] | (uint32_t)(qh[2*i+1] << 16)) & kmask1;
+ const uint32_t q = ((high << shl_h) >> shr_h) | (low >> shr_l);
+ reg[i][0] = dl0 * ((half)(q & 0xFF)) - ml;
+ reg[i][1] = dl1 * ((float)(q & 0xFF00)) - ml;
+ reg[i][2] = dl2 * ((float)(q & 0xFF0000)) - ml;
+ reg[i][3] = dl3 * ((float)(q & 0xFF000000)) - ml;
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq2_xxs(device const block_iq2_xxs * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const float d = xb->d;
+ const int ib32 = il/2;
+ il = il%2;
+ // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
+ // each block of 32 needs 2 uint32_t's for the quants & scale, so 4 uint16_t's.
+ device const uint16_t * q2 = xb->qs + 4*ib32;
+ const uint32_t aux32_g = q2[0] | (q2[1] << 16);
+ const uint32_t aux32_s = q2[2] | (q2[3] << 16);
+ thread const uint8_t * aux8 = (thread const uint8_t *)&aux32_g;
+ const float dl = d * (0.5f + (aux32_s >> 28)) * 0.25f;
+ constant uint8_t * grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+0]);
+ uint8_t signs = ksigns_iq2xs[(aux32_s >> 14*il) & 127];
+ for (int i = 0; i < 8; ++i) {
+ reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
+ }
+ grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+1]);
+ signs = ksigns_iq2xs[(aux32_s >> (14*il+7)) & 127];
+ for (int i = 0; i < 8; ++i) {
+ reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq2_xs(device const block_iq2_xs * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const float d = xb->d;
+ const int ib32 = il/2;
+ il = il%2;
+ // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
+ device const uint16_t * q2 = xb->qs + 4*ib32;
+ const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
+ constant uint8_t * grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+0] & 511));
+ uint8_t signs = ksigns_iq2xs[q2[2*il+0] >> 9];
+ for (int i = 0; i < 8; ++i) {
+ reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
+ }
+ grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+1] & 511));
+ signs = ksigns_iq2xs[q2[2*il+1] >> 9];
+ for (int i = 0; i < 8; ++i) {
+ reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq3_xxs(device const block_iq3_xxs * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const float d = xb->d;
+ const int ib32 = il/2;
+ il = il%2;
+ // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
+ device const uint8_t * q3 = xb->qs + 8*ib32;
+ device const uint16_t * gas = (device const uint16_t *)(xb->qs + QK_K/4) + 2*ib32;
+ const uint32_t aux32 = gas[0] | (gas[1] << 16);
+ const float dl = d * (0.5f + (aux32 >> 28)) * 0.5f;
+ constant uint8_t * grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+0]);
+ constant uint8_t * grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+1]);
+ uint8_t signs = ksigns_iq2xs[(aux32 >> 14*il) & 127];
+ for (int i = 0; i < 4; ++i) {
+ reg[0][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
+ reg[1][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
+ }
+ grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+2]);
+ grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+3]);
+ signs = ksigns_iq2xs[(aux32 >> (14*il+7)) & 127];
+ for (int i = 0; i < 4; ++i) {
+ reg[2][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
+ reg[3][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq3_s(device const block_iq3_s * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const float d = xb->d;
+ const int ib32 = il/2;
+ il = il%2;
+ // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
+ device const uint8_t * qs = xb->qs + 8*ib32;
+ device const uint8_t * signs = xb->signs + 4*ib32 + 2*il;
+ const uint8_t qh = xb->qh[ib32] >> 4*il;
+ const float dl = d * (1 + 2*((xb->scales[ib32/2] >> 4*(ib32%2)) & 0xf));
+ constant uint8_t * grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+0] | ((qh << 8) & 256)));
+ constant uint8_t * grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+1] | ((qh << 7) & 256)));
+ for (int i = 0; i < 4; ++i) {
+ reg[0][i] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i+0]);
+ reg[1][i] = dl * grid2[i] * select(1, -1, signs[0] & kmask_iq2xs[i+4]);
+ }
+ grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+2] | ((qh << 6) & 256)));
+ grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+3] | ((qh << 5) & 256)));
+ for (int i = 0; i < 4; ++i) {
+ reg[2][i] = dl * grid1[i] * select(1, -1, signs[1] & kmask_iq2xs[i+0]);
+ reg[3][i] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i+4]);
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq2_s(device const block_iq2_s * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const float d = xb->d;
+ const int ib32 = il/2;
+ il = il%2;
+ // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
+ device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
+ device const uint8_t * signs = qs + QK_K/8;
+ const uint8_t qh = xb->qh[ib32] >> 4*il;
+ const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
+ constant uint8_t * grid1 = (constant uint8_t *)(iq2s_grid + (qs[0] | ((qh << 8) & 0x300)));
+ constant uint8_t * grid2 = (constant uint8_t *)(iq2s_grid + (qs[1] | ((qh << 6) & 0x300)));
+ for (int i = 0; i < 8; ++i) {
+ reg[i/4+0][i%4] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i]);
+ reg[i/4+2][i%4] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i]);
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq1_s(device const block_iq1_s * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const int ib32 = il/2;
+ il = il%2;
+ const float d = xb->d;
+ device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
+ device const uint16_t * qh = xb->qh;
+ const float dl = d * (2*((qh[ib32] >> 12) & 7) + 1);
+ const float ml = dl * (qh[ib32] & 0x8000 ? -1 - IQ1S_DELTA : -1 + IQ1S_DELTA);
+ const uint16_t h = qh[ib32] >> 6*il;
+ constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((h << 8) & 0x700)));
+ constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((h << 5) & 0x700)));
+ for (int i = 0; i < 4; ++i) {
+ reg[0][i] = dl * (grid1[i] & 0xf) + ml;
+ reg[1][i] = dl * (grid1[i] >> 4) + ml;
+ reg[2][i] = dl * (grid2[i] & 0xf) + ml;
+ reg[3][i] = dl * (grid2[i] >> 4) + ml;
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq1_m(device const block_iq1_m * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const int ib32 = il/2;
+ il = il%2;
+ device const uint16_t * sc = (device const uint16_t *)xb->scales;
+
+ iq1m_scale_t scale;
+ scale.u16 = (sc[0] >> 12) | ((sc[1] >> 8) & 0x00f0) | ((sc[2] >> 4) & 0x0f00) | (sc[3] & 0xf000);
+ const float d = scale.f16;
+
+ device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
+ device const uint8_t * qh = xb->qh + 2*ib32 + il;
+
+ const float dl = d * (2*((sc[ib32/2] >> (6*(ib32%2)+3*il)) & 7) + 1);
+ const float ml1 = dl * (qh[0] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
+ const float ml2 = dl * (qh[0] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
+ constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
+ constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 4) & 0x700)));
+ for (int i = 0; i < 4; ++i) {
+ reg[0][i] = dl * (grid1[i] & 0xf) + ml1;
+ reg[1][i] = dl * (grid1[i] >> 4) + ml1;
+ reg[2][i] = dl * (grid2[i] & 0xf) + ml2;
+ reg[3][i] = dl * (grid2[i] >> 4) + ml2;
+ }
+}
+
+template <typename type4x4>
+void dequantize_iq4_nl(device const block_iq4_nl * xb, short il, thread type4x4 & reg) {
+ device const uint16_t * q4 = (device const uint16_t *)xb->qs;
+ const float d = xb->d;
+ uint32_t aux32;
+ thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
+ for (int i = 0; i < 4; ++i) {
+ aux32 = ((q4[2*i] | (q4[2*i+1] << 16)) >> 4*il) & 0x0f0f0f0f;
+ reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
+ reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
+ reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
+ reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
+ }
+}
+
+template <typename type4>
+void dequantize_iq4_nl_t4(device const block_iq4_nl * xb, short il, thread type4 & reg) {
+ device const uint16_t * q4 = (device const uint16_t *)xb->qs;
+ const float d = xb->d;
+ uint32_t aux32;
+ thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
+ aux32 = ((q4[2*(il%4)] | (q4[2*(il%4)+1] << 16)) >> 4*(il/4)) & 0x0f0f0f0f;
+ reg[0] = d * kvalues_iq4nl_f[q8[0]];
+ reg[1] = d * kvalues_iq4nl_f[q8[1]];
+ reg[2] = d * kvalues_iq4nl_f[q8[2]];
+ reg[3] = d * kvalues_iq4nl_f[q8[3]];
+}
+
+template <typename type4x4>
+void dequantize_iq4_xs(device const block_iq4_xs * xb, short il, thread type4x4 & reg) {
+ // il is 0...15 for QK_K = 256 => index of block of 32 is il/2
+ const int ib32 = il/2;
+ il = il%2;
+ // il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
+ device const uint32_t * q4 = (device const uint32_t *)xb->qs + 4*ib32;
+ const int ls = ((xb->scales_l[ib32/2] >> 4*(ib32%2)) & 0xf) | (((xb->scales_h >> 2*ib32) & 3) << 4);
+ const float d = (float)xb->d * (ls - 32);
+ uint32_t aux32;
+ thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
+ for (int i = 0; i < 4; ++i) {
+ aux32 = (q4[i] >> 4*il) & 0x0f0f0f0f;
+ reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
+ reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
+ reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
+ reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
+ }
+}
+
+template <typename type4x4>
+void dequantize_tq2_0(device const block_tq2_0 * xb, short il, thread type4x4 & reg) {
+ device const uint8_t * qs = xb->qs;
+ const float d = xb->d;
+
+ float4x4 reg_f;
+
+ // 2 bits per element, 4 elements per byte, 128 elements per 32-byte group
+ const short base = il * 16;
+ for (int k = 0; k < 16; k++) {
+ const int i = base + k;
+ const int byte = ((i >> 7) & 1) * 32 + (i & 31);
+ const int l = (i >> 5) & 3;
+ reg_f[k/4][k%4] = d * (float)(((qs[byte] >> (2*l)) & 3) - 1);
+ }
+
+ reg = (type4x4) reg_f;
+}
diff --git a/ggml/src/ggml-metal/kernels/fa.metal b/ggml/src/ggml-metal/kernels/fa.metal
new file mode 100644
index 0000000000..60f4bb4a74
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/fa.metal
@@ -0,0 +1,1852 @@
+#include "common.h"
+#include "dequantize.h"
+
+// dequantize a quantized KV cache tensor to contiguous F16 before running the F16 flash attention kernels
+// - one thread per block; dispatched separately for K and V
+// - ref: https://github.com/ggml-org/llama.cpp/pull/27390
+template <
+ typename block_t,
+ short QK,
+ void (*deq_t4x4)(device const block_t *, short, thread float4x4 &)>
+kernel void kernel_flash_attn_ext_kv_f16(
+ constant ggml_metal_kargs_flash_attn_ext_kv_f16 & args,
+ device const char * x,
+ device half * x_dst,
+ uint gid [[thread_position_in_grid]]) {
+ if (gid >= (uint) args.nblocks) {
+ return;
+ }
+
+ const uint nb = args.ne0/QK;
+ const uint i0 = gid%nb;
+ uint ib = gid/nb;
+ const uint i1 = ib%args.ne1;
+ ib /= args.ne1;
+ const uint i2 = ib%args.ne2;
+ const uint i3 = ib/args.ne2;
+
+ const uint64_t offs = i0*args.nb0 + i1*args.nb1 + i2*args.nb2 + i3*args.nb3;
+
+ device const block_t * src = (device const block_t *) (x + offs);
+ device half4 * dst = (device half4 *) x_dst + (QK/4)*gid;
+
+ for (short i = 0; i < QK/16; ++i) {
+ float4x4 reg;
+ deq_t4x4(src, i, reg);
+ dst[4*i + 0] = (half4) reg[0];
+ dst[4*i + 1] = (half4) reg[1];
+ dst[4*i + 2] = (half4) reg[2];
+ dst[4*i + 3] = (half4) reg[3];
+ }
+}
+
+typedef decltype(kernel_flash_attn_ext_kv_f16<block_q8_0, 32, dequantize_q8_0>) kernel_flash_attn_ext_kv_f16_t;
+
+template [[host_name("kernel_flash_attn_ext_kv_q4_0_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q4_0, 32, dequantize_q4_0>;
+template [[host_name("kernel_flash_attn_ext_kv_q4_1_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q4_1, 32, dequantize_q4_1>;
+template [[host_name("kernel_flash_attn_ext_kv_q5_0_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q5_0, 32, dequantize_q5_0>;
+template [[host_name("kernel_flash_attn_ext_kv_q5_1_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q5_1, 32, dequantize_q5_1>;
+template [[host_name("kernel_flash_attn_ext_kv_q8_0_f16")]] kernel kernel_flash_attn_ext_kv_f16_t kernel_flash_attn_ext_kv_f16<block_q8_0, 32, dequantize_q8_0>;
+
+constant bool FC_flash_attn_ext_pad_has_mask [[function_constant(FC_FLASH_ATTN_EXT_PAD + 0)]];
+
+constant int32_t FC_flash_attn_ext_pad_ncpsg [[function_constant(FC_FLASH_ATTN_EXT_PAD + 25)]];
+
+// pad the last chunk of C elements of k and v into a an extra pad buffer
+kernel void kernel_flash_attn_ext_pad(
+ constant ggml_metal_kargs_flash_attn_ext_pad & args,
+ device const char * k,
+ device const char * v,
+ device const char * mask,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int32_t C = FC_flash_attn_ext_pad_ncpsg;
+
+ device char * k_pad = dst;
+ device char * v_pad = k_pad + args.nb11*C*args.ne_12_2*args.ne_12_3;
+ device char * mask_pad = v_pad + args.nb21*C*args.ne_12_2*args.ne_12_3;
+
+ const int32_t icp = args.ne11 % C;
+ const int32_t ic0 = args.ne11 - icp;
+
+ const int32_t i1 = tgpig[0];
+ const int32_t i2 = tgpig[1];
+ const int32_t i3 = tgpig[2];
+
+ if (i2 < args.ne_12_2 && i3 < args.ne_12_3) {
+ device const char * k_src = k + args.nb11*(ic0 + i1) + args.nb12*i2 + args.nb13*i3;
+ device const char * v_src = v + args.nb21*(ic0 + i1) + args.nb22*i2 + args.nb23*i3;
+
+ device char * k_dst = k_pad + args.nb11*i1 + args.nb11*C*i2 + args.nb11*C*args.ne_12_2*i3;
+ device char * v_dst = v_pad + args.nb21*i1 + args.nb21*C*i2 + args.nb21*C*args.ne_12_2*i3;
+
+ if (i1 >= icp) {
+ // here it is not important the exact value that will be used as we rely on masking out the scores in the attention
+ for (uint64_t i = tiitg; i < args.nb11; i += ntg.x) {
+ k_dst[i] = 0;
+ }
+ for (uint64_t i = tiitg; i < args.nb21; i += ntg.x) {
+ v_dst[i] = 0;
+ }
+ } else {
+ for (uint64_t i = tiitg; i < args.nb11; i += ntg.x) {
+ k_dst[i] = k_src[i];
+ }
+ for (uint64_t i = tiitg; i < args.nb21; i += ntg.x) {
+ v_dst[i] = v_src[i];
+ }
+ }
+ }
+
+ if (FC_flash_attn_ext_pad_has_mask) {
+ if (i2 < args.ne32 && i3 < args.ne33) {
+ for (int ib = i1; ib < args.ne31; ib += C) {
+ device const half * mask_src = (device const half *)(mask + args.nb31*ib + args.nb32*i2 + args.nb33*i3) + ic0;
+ device half * mask_dst = (device half *)(mask_pad) + C*ib + C*args.ne31*i2 + C*args.ne31*args.ne32*i3;
+
+ for (int i = tiitg; i < C; i += ntg.x) {
+ if (i >= icp) {
+ mask_dst[i] = -MAXHALF;
+ } else {
+ mask_dst[i] = mask_src[i];
+ }
+ }
+ }
+ }
+ }
+}
+
+constant int32_t FC_flash_attn_ext_blk_nqptg [[function_constant(FC_FLASH_ATTN_EXT_BLK + 24)]];
+constant int32_t FC_flash_attn_ext_blk_ncpsg [[function_constant(FC_FLASH_ATTN_EXT_BLK + 25)]];
+
+// scan the blocks of the mask that are not masked
+// 0 - masked (i.e. full of -INF, skip)
+// 1 - not masked (i.e. at least one element of the mask is not -INF)
+// 2 - all zero
+kernel void kernel_flash_attn_ext_blk(
+ constant ggml_metal_kargs_flash_attn_ext_blk & args,
+ device const char * mask,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]]) {
+ // block size C x Q
+ const int32_t Q = FC_flash_attn_ext_blk_nqptg;
+ const int32_t C = FC_flash_attn_ext_blk_ncpsg;
+
+ constexpr short NW = N_SIMDWIDTH;
+
+ const int32_t i3 = tgpig[2]/args.ne32;
+ const int32_t i2 = tgpig[2]%args.ne32;
+ const int32_t i1 = tgpig[1];
+ const int32_t i0 = tgpig[0];
+
+ char res = i0*C + C > args.ne30 ? 1 : 0;
+
+ device const half * mask_src = (device const half *) (mask + (i1*Q)*args.nb31 + i2*args.nb32 + i3*args.nb33) + i0*C + tiisg;
+
+ // detailed check of the elements of the block
+ if ((C > NW || Q > 1) && res == 0) {
+ half mmin = MAXHALF;
+ half mmax = -MAXHALF;
+
+ FOR_UNROLL (short j = 0; j < Q; ++j) {
+ FOR_UNROLL (short ii = 0; ii < C/NW; ++ii) {
+ mmin = min(mmin, mask_src[ii*NW]);
+ mmax = max(mmax, mask_src[ii*NW]);
+ }
+
+ mask_src += args.nb31/2;
+ }
+
+ mmin = simd_min(mmin);
+ mmax = simd_max(mmax);
+
+ if (mmax > -MAXHALF) {
+ if (mmin == 0.0 && mmax == 0.0) {
+ res = 2;
+ } else {
+ res = 1;
+ }
+ }
+ }
+
+ const int32_t nblk1 = ((args.ne01 + Q - 1)/Q);
+ const int32_t nblk0 = ((args.ne30 + C - 1)/C);
+
+ if (tiisg == 0) {
+ dst[((i3*args.ne32 + i2)*nblk1 + i1)*nblk0 + i0] = res;
+ }
+}
+
+constant bool FC_flash_attn_ext_has_mask [[function_constant(FC_FLASH_ATTN_EXT + 0)]];
+constant bool FC_flash_attn_ext_has_sinks [[function_constant(FC_FLASH_ATTN_EXT + 1)]];
+constant bool FC_flash_attn_ext_has_bias [[function_constant(FC_FLASH_ATTN_EXT + 2)]];
+constant bool FC_flash_attn_ext_has_scap [[function_constant(FC_FLASH_ATTN_EXT + 3)]];
+constant bool FC_flash_attn_ext_has_kvpad [[function_constant(FC_FLASH_ATTN_EXT + 4)]];
+
+constant bool FC_flash_attn_ext_bc_mask [[function_constant(FC_FLASH_ATTN_EXT + 10)]];
+
+//constant float FC_flash_attn_ext_scale [[function_constant(FC_FLASH_ATTN_EXT + 10)]];
+//constant float FC_flash_attn_ext_max_bias [[function_constant(FC_FLASH_ATTN_EXT + 11)]];
+//constant float FC_flash_attn_ext_logit_softcap [[function_constant(FC_FLASH_ATTN_EXT + 12)]];
+
+constant int32_t FC_flash_attn_ext_ns10 [[function_constant(FC_FLASH_ATTN_EXT + 20)]];
+constant int32_t FC_flash_attn_ext_ns20 [[function_constant(FC_FLASH_ATTN_EXT + 21)]];
+constant int32_t FC_flash_attn_ext_nsg [[function_constant(FC_FLASH_ATTN_EXT + 22)]];
+
+// ref: https://arxiv.org/pdf/2307.08691.pdf
+template<
+ typename q_t, // query types in shared memory
+ typename q4_t,
+ typename q8x8_t,
+ typename k_t, // key types in shared memory
+ typename k4x4_t,
+ typename k8x8_t,
+ typename v_t, // value types in shared memory
+ typename v4x4_t,
+ typename v8x8_t,
+ typename qk_t, // Q*K types
+ typename qk8x8_t,
+ typename s_t, // soft-max types
+ typename s2_t,
+ typename s8x8_t,
+ typename o_t, // attention accumulation types
+ typename o4_t,
+ typename o8x8_t,
+ typename kd4x4_t, // key type in device memory
+ short nl_k,
+ void (*deq_k)(device const kd4x4_t *, short, thread k4x4_t &),
+ typename vd4x4_t, // value type in device memory
+ short nl_v,
+ void (*deq_v)(device const vd4x4_t *, short, thread v4x4_t &),
+ short DK, // K head size
+ short DV, // V head size
+ short Q, // queries per threadgroup
+ short C, // cache items per threadgroup
+ short NSG> // number of simd groups
+void kernel_flash_attn_ext_impl(
+ constant ggml_metal_kargs_flash_attn_ext & args,
+ device const char * q,
+ device const char * k,
+ device const char * v,
+ device const char * mask,
+ device const char * sinks,
+ device const char * pad,
+ device const char * blk,
+ device char * dst,
+ threadgroup half * shmem_f16,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const ushort iq3 = tgpig[2];
+ const ushort iq2 = tgpig[1];
+ const ushort iq1 = tgpig[0]*Q;
+
+#define NS10 (FC_flash_attn_ext_ns10)
+#define NS20 (FC_flash_attn_ext_ns20)
+
+ // note: I had some concerns that using this instead of the ugly macros above was affecting performance
+ // need to re-check carefully and if no regressions are observerd - remove the macros
+ // the concerns is that maybe using const variables requires extra registers? but not sure if the compiler
+ // is clever enough to avoid this. unfortunately, using constexpr is not possible with FC
+ //const short NS10 = FC_flash_attn_ext_ns10;
+ //const short NS20 = FC_flash_attn_ext_ns20;
+
+ constexpr short KV = 8;
+
+ constexpr short DK4 = DK/4;
+ constexpr short DK8 = DK/8;
+ constexpr short DK16 = DK/16;
+ constexpr short DV4 = DV/4;
+ //constexpr short DV8 = DV/8;
+ constexpr short DV16 = DV/16;
+
+ constexpr short PV = PAD2(DV, 64);
+ constexpr short PV4 = PV/4;
+ constexpr short PV8 = PV/8;
+ //constexpr short PV16 = PV/16;
+
+ constexpr short NW = N_SIMDWIDTH;
+ constexpr short NQ = Q/NSG;
+ constexpr short SH = 2*C; // shared memory per simdgroup (s_t == float)
+
+ constexpr short TS = 2*SH;
+ constexpr short T = DK + 2*PV; // shared memory size per query in (half)
+
+ threadgroup q_t * sq = (threadgroup q_t *) (shmem_f16 + 0*T); // holds the query data
+ threadgroup q4_t * sq4 = (threadgroup q4_t *) (shmem_f16 + 0*T); // same as above but in q4_t
+ threadgroup o_t * so = (threadgroup o_t *) (shmem_f16 + 0*T + Q*DK); // the result for all queries in 8x8 matrices (the O matrix from the paper)
+ threadgroup o4_t * so4 = (threadgroup o4_t *) (shmem_f16 + 0*T + Q*DK);
+ threadgroup s_t * ss = (threadgroup s_t *) (shmem_f16 + Q*T); // scratch buffer for attention, mask and diagonal matrix
+ threadgroup s2_t * ss2 = (threadgroup s2_t *) (shmem_f16 + Q*T); // same as above but in s2_t
+
+ threadgroup k_t * sk = (threadgroup k_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // scratch buffer to load K in shared memory
+ threadgroup k4x4_t * sk4x4 = (threadgroup k4x4_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // same as above but in k4x4_t
+
+ threadgroup v_t * sv = (threadgroup v_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // scratch buffer to load V in shared memory
+ threadgroup v4x4_t * sv4x4 = (threadgroup v4x4_t *) (shmem_f16 + sgitg*(4*16*KV) + Q*T + Q*TS); // same as above but in v4x4_t
+
+ // mask storage in shared mem
+ threadgroup half2 * sm2 = (threadgroup half2 *) (shmem_f16 + Q*T + 2*C);
+
+ // per-query mask pointers
+ device const half2 * pm2[NQ];
+
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ pm2[jj] = (device const half2 *) ((device const char *) mask + (iq1 + j)*args.nb31 + (iq2%args.ne32)*args.nb32 + (iq3%args.ne33)*args.nb33);
+ }
+
+ {
+ const int32_t nblk1 = ((args.ne01 + Q - 1)/Q);
+ const int32_t nblk0 = ((args.ne11 + C - 1)/C);
+
+ blk += (((iq3%args.ne33)*args.ne32 + (iq2%args.ne32))*nblk1 + iq1/Q)*nblk0;
+ }
+
+ {
+ q += iq1*args.nb01 + iq2*args.nb02 + iq3*args.nb03;
+
+ const short ikv2 = iq2/(args.ne02/args.ne_12_2);
+ const short ikv3 = iq3/(args.ne03/args.ne_12_3);
+
+ k += ikv2*args.nb12 + ikv3*args.nb13;
+ v += ikv2*args.nb22 + ikv3*args.nb23;
+ }
+
+ // load heads from Q to shared memory
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ device const float4 * q4 = (device const float4 *) ((device const char *) q + j*args.nb01);
+
+ for (short i = tiisg; i < DK4; i += NW) {
+ if (iq1 + j < args.ne01) {
+ sq4[j*DK4 + i] = (q4_t) q4[i];
+ } else {
+ sq4[j*DK4 + i] = 0;
+ }
+ }
+ }
+
+ // zero out
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ for (short i = tiisg; i < DV4; i += NW) {
+ so4[j*PV4 + i] = 0;
+ }
+
+ for (short i = tiisg; i < SH; i += NW) {
+ ss[j*SH + i] = 0.0f;
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ float S[NQ] = { [0 ... NQ-1] = 0.0f };
+
+ {
+ float M[NQ] = { [0 ... NQ-1] = -FLT_MAX/2 };
+
+ float slope = 1.0f;
+
+ // ALiBi
+ if (FC_flash_attn_ext_has_bias) {
+ const short h = iq2;
+
+ const float base = h < args.n_head_log2 ? args.m0 : args.m1;
+ const short exph = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
+
+ slope = pow(base, exph);
+ }
+
+ // loop over the KV cache
+ // each simdgroup handles blocks of Q rows and C columns
+ for (int ic0 = 0; ; ++ic0) {
+ int ic = ic0*C;
+ if (ic >= args.ne11) {
+ break;
+ }
+
+ // the last partial chunk uses the pad buffer as source
+ if (FC_flash_attn_ext_has_kvpad && ic + C > args.ne11) {
+ k = pad;
+ v = k + args.nb11*C*args.ne_12_2*args.ne_12_3;
+ mask = v + args.nb21*C*args.ne_12_2*args.ne_12_3;
+
+ const short ikv2 = iq2/(args.ne02/args.ne_12_2);
+ const short ikv3 = iq3/(args.ne03/args.ne_12_3);
+
+ k += (ikv2 + ikv3*args.ne_12_2)*args.nb11*C;
+ v += (ikv2 + ikv3*args.ne_12_2)*args.nb21*C;
+
+ if (!FC_flash_attn_ext_has_mask) {
+ threadgroup half * sm = (threadgroup half *) (sm2);
+
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ for (short i = tiisg; i < C; i += NW) {
+ if (ic + i >= args.ne11) {
+ sm[2*j*SH + i] = -MAXHALF;
+ }
+ }
+ }
+ } else {
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ pm2[jj] = (device const half2 *) ((device const half *) mask +
+ (iq1 + j)*C +
+ (iq2%args.ne32)*(C*args.ne31) +
+ (iq3%args.ne33)*(C*args.ne31*args.ne32));
+ }
+ }
+
+ ic = 0;
+ }
+
+ char blk_cur = 1;
+
+ // read the mask into shared mem
+ if (FC_flash_attn_ext_has_mask) {
+ blk_cur = blk[ic0];
+
+ if (blk_cur == 0) {
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ pm2[jj] += NW;
+ }
+
+ continue;
+ }
+
+ if (blk_cur == 1) {
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ if (FC_flash_attn_ext_bc_mask) {
+ sm2[j*SH + tiisg] = (iq1 + j) < args.ne31 ? pm2[jj][tiisg] : half2(-MAXHALF, -MAXHALF);
+ } else {
+ sm2[j*SH + tiisg] = pm2[jj][tiisg];
+ }
+
+ pm2[jj] += NW;
+ }
+ } else if (blk_cur == 2) {
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ pm2[jj] += NW;
+ }
+ }
+
+#if 0
+ // note: old -INF block optimization - obsoleted by pre-computing non-masked blocks
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // used to detect blocks full of -INF
+ // skip only when the entire threadgroup is masked
+ half2 smax2(-MAXHALF/2, -MAXHALF/2);
+
+ FOR_UNROLL (short j = 0; j < Q; ++j) {
+ smax2 = max(smax2, sm2[j*SH + tiisg]);
+ }
+
+ smax2 = simd_max(smax2);
+
+ if (max(smax2[0], smax2[1]) <= -MAXHALF/2) {
+ // this barrier is important
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ continue;
+ }
+#endif
+ }
+
+ // Q*K^T
+ // this is compile-time check, so it does not have runtime overhead
+ if (is_same<kd4x4_t, k4x4_t>::value) {
+ // we can read directly from global memory
+ device const k_t * pk = (device const k_t *) (k + ic*args.nb11);
+ threadgroup const q_t * pq = sq;
+ threadgroup s_t * ps = ss;
+
+ pk += sgitg*(8*NS10);
+ ps += sgitg*(8*1);
+
+ static_assert((C/8) % NSG == 0, "");
+
+ constexpr short NC = (C/8)/NSG;
+
+ FOR_UNROLL (short cc = 0; cc < NC; ++cc) {
+ qk8x8_t mqk = make_filled_simdgroup_matrix<qk_t, 8>((qk_t) 0.0f);
+
+ if (DK % 16 != 0) {
+ k8x8_t mk;
+ q8x8_t mq;
+
+ FOR_UNROLL (short i = 0; i < DK8; ++i) {
+ simdgroup_barrier(mem_flags::mem_none);
+
+ simdgroup_load(mk, pk + 8*i, NS10, 0, true);
+ simdgroup_load(mq, pq + 8*i, DK);
+
+ simdgroup_barrier(mem_flags::mem_none);
+
+ simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
+ }
+ } else {
+ k8x8_t mk[2];
+ q8x8_t mq[2];
+
+ // note: too much unroll can tank the performance for large heads
+ #pragma unroll (MIN(DK8/2, 4*NSG))
+ for (short i = 0; i < DK8/2; ++i) {
+ simdgroup_barrier(mem_flags::mem_none);
+
+ simdgroup_load(mq[0], pq + 0*8 + 16*i, DK);
+ simdgroup_load(mq[1], pq + 1*8 + 16*i, DK);
+
+ simdgroup_load(mk[0], pk + 0*8 + 16*i, NS10, 0, true);
+ simdgroup_load(mk[1], pk + 1*8 + 16*i, NS10, 0, true);
+
+ simdgroup_barrier(mem_flags::mem_none);
+
+ simdgroup_multiply_accumulate(mqk, mq[0], mk[0], mqk);
+ simdgroup_multiply_accumulate(mqk, mq[1], mk[1], mqk);
+ }
+ }
+
+ simdgroup_store(mqk, ps, SH, 0, false);
+
+ pk += 8*(NSG*NS10);
+ ps += 8*(NSG);
+ }
+ } else {
+ // TODO: this is the quantized K cache branch - not optimized yet
+ for (short ccc = 0; ccc < (C/8)/NSG; ++ccc) {
+ const short cc = ccc*NSG + sgitg;
+
+ const short tx = tiisg%4;
+ const short ty = tiisg/4;
+
+ qk8x8_t mqk = make_filled_simdgroup_matrix<qk_t, 8>((qk_t) 0.0f);
+
+ for (short ii = 0; ii < DK16; ii += 4) {
+ device const kd4x4_t * pk4x4 = (device const kd4x4_t *) (k + ((ic + 8*cc + ty)*args.nb11));
+
+ if (DK16%4 == 0) {
+ // the head is evenly divisible by 4*16 = 64, so no need for bound checks
+ {
+ k4x4_t tmp;
+ deq_k(pk4x4 + (ii + tx)/nl_k, (ii + tx)%nl_k, tmp);
+ sk4x4[4*ty + tx] = tmp;
+ }
+
+ simdgroup_barrier(mem_flags::mem_threadgroup);
+
+ FOR_UNROLL (short k = 0; k < 4; ++k) {
+ k8x8_t mk;
+ q8x8_t mq;
+
+ simdgroup_load(mk, sk + 16*k + 0*8, 4*16, 0, true); // transpose
+ simdgroup_load(mq, sq + (2*(ii + k) + 0)*8, DK);
+ simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
+
+ simdgroup_load(mk, sk + 16*k + 1*8, 4*16, 0, true); // transpose
+ simdgroup_load(mq, sq + (2*(ii + k) + 1)*8, DK);
+ simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
+ }
+ } else {
+ if (ii + tx < DK16) {
+ k4x4_t tmp;
+ deq_k(pk4x4 + (ii + tx)/nl_k, (ii + tx)%nl_k, tmp);
+ sk4x4[4*ty + tx] = tmp;
+ }
+
+ simdgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (short k = 0; k < 4 && ii + k < DK16; ++k) {
+ k8x8_t mk;
+ q8x8_t mq;
+
+ simdgroup_load(mk, sk + 16*k + 0*8, 4*16, 0, true); // transpose
+ simdgroup_load(mq, sq + (2*(ii + k) + 0)*8, DK);
+ simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
+
+ simdgroup_load(mk, sk + 16*k + 1*8, 4*16, 0, true); // transpose
+ simdgroup_load(mq, sq + (2*(ii + k) + 1)*8, DK);
+ simdgroup_multiply_accumulate(mqk, mq, mk, mqk);
+ }
+ }
+ }
+
+ simdgroup_store(mqk, ss + 8*cc, SH, 0, false);
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // online softmax
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ const float m = M[jj];
+
+ // scale and apply the logitcap / mask
+ float2 s2 = ss2[j*SH/2 + tiisg]*args.scale;
+
+ if (FC_flash_attn_ext_has_scap) {
+ s2 = args.logit_softcap*precise::tanh(s2);
+ }
+
+ // mqk = mqk + slope*mask
+ if (blk_cur != 2) {
+ if (FC_flash_attn_ext_has_bias) {
+ s2 += s2_t(sm2[j*SH + tiisg])*slope;
+ } else {
+ s2 += s2_t(sm2[j*SH + tiisg]);
+ }
+ }
+
+ M[jj] = simd_max(max(M[jj], max(s2[0], s2[1])));
+
+ const float ms = exp(m - M[jj]);
+ const float2 vs2 = exp(s2 - M[jj]);
+
+ S[jj] = S[jj]*ms + simd_sum(vs2[0] + vs2[1]);
+
+ // the P matrix from the paper (Q rows, C columns)
+ ss2[j*SH/2 + tiisg] = vs2;
+
+ if (DV4 % NW == 0) {
+ FOR_UNROLL (short ii = 0; ii < DV4/NW; ++ii) {
+ const short i = ii*NW + tiisg;
+
+ so4[j*PV4 + i] *= ms;
+ }
+ } else {
+ for (short i = tiisg; i < DV4; i += NW) {
+ so4[j*PV4 + i] *= ms;
+ }
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // O = O + (Q*K^T)*V
+ {
+ // we can read directly from global memory
+ if (is_same<vd4x4_t, v4x4_t>::value) {
+ static_assert(PV8 % NSG == 0, "");
+
+ constexpr short NO = PV8/NSG;
+
+ o8x8_t lo[NO];
+
+ {
+ auto sot = so + 8*sgitg;
+
+ FOR_UNROLL (short ii = 0; ii < NO; ++ii) {
+ simdgroup_load(lo[ii], sot, PV, 0, false);
+
+ sot += 8*NSG;
+ }
+ }
+
+ {
+ device const v_t * pv = (device const v_t *) (v + ic*args.nb21);
+
+ pv += 8*sgitg;
+
+ if (DV <= 64) {
+ FOR_UNROLL (short cc = 0; cc < C/8; ++cc) {
+ s8x8_t vs;
+ simdgroup_load(vs, ss + 8*cc, SH, 0, false);
+
+ FOR_UNROLL (short ii = 0; ii < NO/2; ++ii) {
+ v8x8_t mv[2];
+
+ simdgroup_load(mv[0], pv + 0*NSG + 16*ii*NSG, NS20, 0, false);
+ simdgroup_load(mv[1], pv + 8*NSG + 16*ii*NSG, NS20, 0, false);
+
+ simdgroup_multiply_accumulate(lo[2*ii + 0], vs, mv[0], lo[2*ii + 0]);
+ simdgroup_multiply_accumulate(lo[2*ii + 1], vs, mv[1], lo[2*ii + 1]);
+ }
+
+ pv += 8*NS20;
+ }
+ } else {
+ constexpr short NC = (C/8)/2;
+
+ FOR_UNROLL (short cc = 0; cc < NC; ++cc) {
+ s8x8_t vs[2];
+
+ simdgroup_load(vs[0], ss + 16*cc + 0, SH, 0, false);
+ simdgroup_load(vs[1], ss + 16*cc + 8, SH, 0, false);
+
+ FOR_UNROLL (short ii = 0; ii < NO/2; ++ii) {
+ v8x8_t mv[4];
+
+ simdgroup_load(mv[0], pv + 0*NSG + 16*ii*NSG + 0*8*NS20, NS20, 0, false);
+ simdgroup_load(mv[1], pv + 8*NSG + 16*ii*NSG + 0*8*NS20, NS20, 0, false);
+ simdgroup_load(mv[2], pv + 0*NSG + 16*ii*NSG + 1*8*NS20, NS20, 0, false);
+ simdgroup_load(mv[3], pv + 8*NSG + 16*ii*NSG + 1*8*NS20, NS20, 0, false);
+
+ simdgroup_multiply_accumulate(lo[2*ii + 0], vs[0], mv[0], lo[2*ii + 0]);
+ simdgroup_multiply_accumulate(lo[2*ii + 1], vs[0], mv[1], lo[2*ii + 1]);
+ simdgroup_multiply_accumulate(lo[2*ii + 0], vs[1], mv[2], lo[2*ii + 0]);
+ simdgroup_multiply_accumulate(lo[2*ii + 1], vs[1], mv[3], lo[2*ii + 1]);
+ }
+
+ pv += 2*8*NS20;
+ }
+ }
+ }
+
+ {
+ auto sot = so + 8*sgitg;
+
+ FOR_UNROLL (short ii = 0; ii < NO; ++ii) {
+ simdgroup_store(lo[ii], sot, PV, 0, false);
+
+ sot += 8*NSG;
+ }
+ }
+ } else {
+ // TODO: this is the quantized V cache branch - not optimized yet
+
+ const short tx = tiisg%4;
+ const short ty = tiisg/4;
+
+ for (short cc = 0; cc < C/8; ++cc) {
+ s8x8_t vs;
+ simdgroup_load(vs, ss + 8*cc, SH, 0, false);
+
+ for (short ii = 4*sgitg; ii < DV16; ii += 4*NSG) {
+ device const vd4x4_t * pv4x4 = (device const vd4x4_t *) (v + ((ic + 8*cc + ty)*args.nb21));
+
+ if (DV16%4 == 0) {
+ // no need for bound checks
+ {
+ v4x4_t tmp;
+ deq_v(pv4x4 + (ii + tx)/nl_v, (ii + tx)%nl_v, tmp);
+ sv4x4[4*ty + tx] = tmp;
+ }
+
+ simdgroup_barrier(mem_flags::mem_threadgroup);
+
+ FOR_UNROLL (short k = 0; k < 4; ++k) {
+ v8x8_t mv[2];
+ o8x8_t lo[2];
+
+ simdgroup_load(mv[0], sv + 16*k + 0*8, 4*16, 0, false);
+ simdgroup_load(mv[1], sv + 16*k + 1*8, 4*16, 0, false);
+ simdgroup_load(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
+ simdgroup_load(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
+
+ simdgroup_multiply_accumulate(lo[0], vs, mv[0], lo[0]);
+ simdgroup_multiply_accumulate(lo[1], vs, mv[1], lo[1]);
+
+ simdgroup_store(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
+ simdgroup_store(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
+ }
+ } else {
+ if (ii + tx < DV16) {
+ v4x4_t tmp;
+ deq_v(pv4x4 + (ii + tx)/nl_v, (ii + tx)%nl_v, tmp);
+ sv4x4[4*ty + tx] = tmp;
+ }
+
+ simdgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (short k = 0; k < 4 && ii + k < DV16; ++k) {
+ v8x8_t mv[2];
+ o8x8_t lo[2];
+
+ simdgroup_load(mv[0], sv + 16*k + 0*8, 4*16, 0, false);
+ simdgroup_load(mv[1], sv + 16*k + 1*8, 4*16, 0, false);
+ simdgroup_load(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
+ simdgroup_load(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
+
+ simdgroup_multiply_accumulate(lo[0], vs, mv[0], lo[0]);
+ simdgroup_multiply_accumulate(lo[1], vs, mv[1], lo[1]);
+
+ simdgroup_store(lo[0], so + 8*(2*(ii + k) + 0), PV, 0, false);
+ simdgroup_store(lo[1], so + 8*(2*(ii + k) + 1), PV, 0, false);
+ }
+ }
+ }
+ }
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ if (FC_flash_attn_ext_has_sinks) {
+ FOR_UNROLL (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+
+ const float m = M[jj];
+ const float s = tiisg == 0 ? ((device const float *) sinks)[iq2] : -FLT_MAX/2;
+
+ M[jj] = simd_max(max(M[jj], s));
+
+ const float ms = exp(m - M[jj]);
+ const float vs = exp(s - M[jj]);
+
+ S[jj] = S[jj]*ms + simd_sum(vs);
+
+ for (short i = tiisg; i < DV4; i += NW) {
+ so4[j*PV4 + i] *= ms;
+ }
+ }
+ }
+ }
+
+ // store to global memory
+ for (short jj = 0; jj < NQ; ++jj) {
+ const short j = jj*NSG + sgitg;
+ if (iq1 + j >= args.ne01) {
+ break;
+ }
+
+ device float4 * dst4 = (device float4 *) dst + ((uint64_t)iq3*args.ne2*args.ne1 + iq2 + (uint64_t)(iq1 + j)*args.ne1)*DV4;
+
+ const float scale = S[jj] == 0.0 ? 0.0f : 1.0f/S[jj];
+
+ if (DV4 % NW == 0) {
+ FOR_UNROLL (short ii = 0; ii < DV4/NW; ++ii) {
+ const short i = ii*NW + tiisg;
+
+ dst4[i] = (float4) so4[j*PV4 + i]*scale;
+ }
+ } else {
+ for (short i = tiisg; i < DV4; i += NW) {
+ dst4[i] = (float4) so4[j*PV4 + i]*scale;
+ }
+ }
+ }
+
+#undef NS10
+#undef NS20
+}
+
+template<
+ typename q_t, // query types in shared memory
+ typename q4_t,
+ typename q8x8_t,
+ typename k_t, // key types in shared memory
+ typename k4x4_t,
+ typename k8x8_t,
+ typename v_t, // value types in shared memory
+ typename v4x4_t,
+ typename v8x8_t,
+ typename qk_t, // Q*K types
+ typename qk8x8_t,
+ typename s_t, // soft-max types
+ typename s2_t,
+ typename s8x8_t,
+ typename o_t, // attention accumulation types
+ typename o4_t,
+ typename o8x8_t,
+ typename kd4x4_t, // key type in device memory
+ short nl_k,
+ void (*deq_k)(device const kd4x4_t *, short, thread k4x4_t &),
+ typename vd4x4_t, // value type in device memory
+ short nl_v,
+ void (*deq_v)(device const vd4x4_t *, short, thread v4x4_t &),
+ short DK, // K head size
+ short DV, // V head size
+ short Q = OP_FLASH_ATTN_EXT_NQPSG, // queries per threadgroup
+ short C = OP_FLASH_ATTN_EXT_NCPSG> // cache items per threadgroup
+kernel void kernel_flash_attn_ext(
+ constant ggml_metal_kargs_flash_attn_ext & args,
+ device const char * q,
+ device const char * k,
+ device const char * v,
+ device const char * mask,
+ device const char * sinks,
+ device const char * pad,
+ device const char * blk,
+ device char * dst,
+ threadgroup half * shmem_f16 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+#define FWD_TMPL q_t, q4_t, q8x8_t, k_t, k4x4_t, k8x8_t, v_t, v4x4_t, v8x8_t, qk_t, qk8x8_t, s_t, s2_t, s8x8_t, o_t, o4_t, o8x8_t, kd4x4_t, nl_k, deq_k, vd4x4_t, nl_v, deq_v, DK, DV, Q, C
+#define FWD_ARGS args, q, k, v, mask, sinks, pad, blk, dst, shmem_f16, tgpig, tiisg, sgitg
+ switch (FC_flash_attn_ext_nsg) {
+ // note: disabled cases to reduce library load time
+ //case 1: kernel_flash_attn_ext_impl<FWD_TMPL, 1>(FWD_ARGS); break;
+ //case 2: kernel_flash_attn_ext_impl<FWD_TMPL, 2>(FWD_ARGS); break;
+ case 4: kernel_flash_attn_ext_impl<FWD_TMPL, 4>(FWD_ARGS); break;
+ case 8: kernel_flash_attn_ext_impl<FWD_TMPL, 8>(FWD_ARGS); break;
+ }
+#undef FWD_TMPL
+#undef FWD_ARGS
+}
+
+// TODO: this is quite ugly. in the future these types will be hardcoded in the kernel, but for now keep them as
+// template to be able to explore different combinations
+//
+#define FA_TYPES \
+ half, half4, simdgroup_half8x8, \
+ half, half4x4, simdgroup_half8x8, \
+ half, half4x4, simdgroup_half8x8, \
+ float, simdgroup_float8x8, \
+ float, float2, simdgroup_float8x8, \
+ float, float4, simdgroup_float8x8
+ //half, half4, simdgroup_half8x8
+
+#define FA_TYPES_BF \
+ bfloat, bfloat4, simdgroup_bfloat8x8, \
+ bfloat, bfloat4x4, simdgroup_bfloat8x8, \
+ bfloat, bfloat4x4, simdgroup_bfloat8x8, \
+ float, simdgroup_float8x8, \
+ float, float2, simdgroup_float8x8, \
+ half, half4, simdgroup_half8x8
+ //float, float4, simdgroup_float8x8
+
+#define FA_TYPES_F32 \
+ half, half4, simdgroup_half8x8, \
+ float, float4x4, simdgroup_float8x8, \
+ float, float4x4, simdgroup_float8x8, \
+ float, simdgroup_float8x8, \
+ float, float2, simdgroup_float8x8, \
+ float, float4, simdgroup_float8x8
+ //half, half4, simdgroup_half8x8
+
+typedef decltype(kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 64, 64>) flash_attn_ext_t;
+
+template [[host_name("kernel_flash_attn_ext_f32_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_f32_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_f32_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_f32_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_f32_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_f32_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_f32_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_f32_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_f32_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_f32_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_f32_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_f32_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_f32_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_f32_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_f32_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_F32, float4x4, 1, dequantize_f32, float4x4, 1, dequantize_f32, 576, 512>;
+
+template [[host_name("kernel_flash_attn_ext_f16_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_f16_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_f16_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_f16_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_f16_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_f16_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_f16_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_f16_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_f16_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_f16_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_f16_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_f16_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_f16_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_f16_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_f16_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, half4x4, 1, dequantize_f16, half4x4, 1, dequantize_f16, 576, 512>;
+
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_bf16_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_bf16_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES_BF, bfloat4x4, 1, dequantize_bf16, bfloat4x4, 1, dequantize_bf16, 576, 512>;
+#endif
+
+template [[host_name("kernel_flash_attn_ext_q4_0_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_q4_0_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_0, 2, dequantize_q4_0, block_q4_0, 2, dequantize_q4_0, 576, 512>;
+
+template [[host_name("kernel_flash_attn_ext_q4_1_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_q4_1_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q4_1, 2, dequantize_q4_1, block_q4_1, 2, dequantize_q4_1, 576, 512>;
+
+template [[host_name("kernel_flash_attn_ext_q5_0_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_q5_0_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_0, 2, dequantize_q5_0, block_q5_0, 2, dequantize_q5_0, 576, 512>;
+
+template [[host_name("kernel_flash_attn_ext_q5_1_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_q5_1_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q5_1, 2, dequantize_q5_1, block_q5_1, 2, dequantize_q5_1, 576, 512>;
+
+template [[host_name("kernel_flash_attn_ext_q8_0_dk32_dv32" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 32, 32>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk40_dv40" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 40, 40>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk48_dv48" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 48, 48>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk64_dv64" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 64, 64>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk72_dv72" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 72, 72>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk80_dv80" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 80, 80>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk96_dv96" )]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 96, 96>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk112_dv112")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 112, 112>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk128_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 128, 128>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk192_dv192")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 192, 192>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk192_dv128")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 192, 128>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk256_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 256, 256>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk320_dv256")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 320, 256>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk512_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 512, 512>;
+template [[host_name("kernel_flash_attn_ext_q8_0_dk576_dv512")]] kernel flash_attn_ext_t kernel_flash_attn_ext<FA_TYPES, block_q8_0, 2, dequantize_q8_0, block_q8_0, 2, dequantize_q8_0, 576, 512>;
+
+#undef FA_TYPES
+#undef FA_TYPES_BF
+#undef FA_TYPES_F32
+
+constant bool FC_flash_attn_ext_vec_has_mask [[function_constant(FC_FLASH_ATTN_EXT_VEC + 0)]];
+constant bool FC_flash_attn_ext_vec_has_sinks [[function_constant(FC_FLASH_ATTN_EXT_VEC + 1)]];
+constant bool FC_flash_attn_ext_vec_has_bias [[function_constant(FC_FLASH_ATTN_EXT_VEC + 2)]];
+constant bool FC_flash_attn_ext_vec_has_scap [[function_constant(FC_FLASH_ATTN_EXT_VEC + 3)]];
+constant bool FC_flash_attn_ext_vec_has_kvpad [[function_constant(FC_FLASH_ATTN_EXT_VEC + 4)]];
+
+//constant float FC_flash_attn_ext_vec_scale [[function_constant(FC_FLASH_ATTN_EXT_VEC + 10)]];
+//constant float FC_flash_attn_ext_vec_max_bias [[function_constant(FC_FLASH_ATTN_EXT_VEC + 11)]];
+//constant float FC_flash_attn_ext_vec_logit_softcap [[function_constant(FC_FLASH_ATTN_EXT_VEC + 12)]];
+
+constant int32_t FC_flash_attn_ext_vec_ns10 [[function_constant(FC_FLASH_ATTN_EXT_VEC + 20)]];
+constant int32_t FC_flash_attn_ext_vec_ns20 [[function_constant(FC_FLASH_ATTN_EXT_VEC + 21)]];
+constant int32_t FC_flash_attn_ext_vec_nsg [[function_constant(FC_FLASH_ATTN_EXT_VEC + 22)]];
+constant int32_t FC_flash_attn_ext_vec_nwg [[function_constant(FC_FLASH_ATTN_EXT_VEC + 23)]];
+
+template<
+ typename q4_t, // query types in shared memory
+ typename k4_t, // key types in shared memory
+ typename v4_t, // value types in shared memory
+ typename qk_t, // Q*K types
+ typename s_t, // soft-max types
+ typename s4_t,
+ typename o4_t, // attention accumulation types
+ typename kd4_t, // key type in device memory
+ short nl_k,
+ void (*deq_k_t4)(device const kd4_t *, short, thread k4_t &),
+ typename vd4_t, // value type in device memory
+ short nl_v,
+ void (*deq_v_t4)(device const vd4_t *, short, thread v4_t &),
+ short DK, // K head size
+ short DV, // V head size
+ short NE = 4, // head elements per thread
+ short Q = OP_FLASH_ATTN_EXT_VEC_NQPSG, // queries per threadgroup
+ short C = OP_FLASH_ATTN_EXT_VEC_NCPSG> // cache items per threadgroup
+kernel void kernel_flash_attn_ext_vec(
+ constant ggml_metal_kargs_flash_attn_ext_vec & args,
+ device const char * q,
+ device const char * k,
+ device const char * v,
+ device const char * mask,
+ device const char * sinks,
+ device const char * pad,
+ device char * dst,
+ threadgroup half * shmem_f16 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ static_assert(DK % 32 == 0, "DK must be divisible by 32");
+ static_assert(DV % 32 == 0, "DV must be divisible by 32");
+
+#define NWG (FC_flash_attn_ext_vec_nwg)
+#define NSG (FC_flash_attn_ext_vec_nsg)
+
+#define NS10 (FC_flash_attn_ext_vec_ns10)
+#define NS20 (FC_flash_attn_ext_vec_ns20)
+
+ const short iwg = tgpig[2]%NWG;
+
+ const ushort iq3 = tgpig[2]/NWG;
+ const ushort iq2 = tgpig[1];
+ const ushort iq1 = tgpig[0];
+
+ constexpr short DK4 = DK/4;
+ constexpr short DV4 = DV/4;
+
+ constexpr short PK = PAD2(DK, 128);
+ constexpr short PK4 = PK/4;
+
+ constexpr short PV = PAD2(DV, 128);
+ constexpr short PV4 = PV/4;
+
+ constexpr short NW = N_SIMDWIDTH;
+ constexpr short NL = NW/NE; // note: this can be adjusted to support different head sizes and simdgroup work loads
+ constexpr short SH = 4*C; // shared memory per simdgroup
+
+ static_assert(DK4 % NL == 0, "DK4 must be divisible by NL");
+ static_assert(DV4 % NL == 0, "DV4 must be divisible by NL");
+
+ //const short T = PK + NSG*SH; // shared memory size per query in (half)
+
+ //threadgroup q_t * sq = (threadgroup q_t *) (shmem_f16 + 0*PK); // holds the query data
+ threadgroup q4_t * sq4 = (threadgroup q4_t *) (shmem_f16 + 0*PK); // same as above but in q4_t
+ threadgroup s_t * ss = (threadgroup s_t *) (shmem_f16 + sgitg*SH + NSG*PK); // scratch buffer for attention
+ threadgroup s4_t * ss4 = (threadgroup s4_t *) (shmem_f16 + sgitg*SH + NSG*PK); // same as above but in s4_t
+ threadgroup half * sm = (threadgroup half *) (shmem_f16 + sgitg*SH + 2*C + NSG*PK); // scratch buffer for mask
+ threadgroup o4_t * so4 = (threadgroup o4_t *) (shmem_f16 + 2*sgitg*PV + NSG*PK + NSG*SH); // scratch buffer for the results
+
+ // store the result for all queries in shared memory (the O matrix from the paper)
+ so4 += tiisg;
+
+ {
+ q += iq1*args.nb01 + iq2*args.nb02 + iq3*args.nb03;
+
+ const short ikv2 = iq2/(args.ne02/args.ne_12_2);
+ const short ikv3 = iq3/(args.ne03/args.ne_12_3);
+
+ k += ikv2*args.nb12 + ikv3*args.nb13;
+ v += ikv2*args.nb22 + ikv3*args.nb23;
+ }
+
+ // load heads from Q to shared memory
+ device const float4 * q4 = (device const float4 *) ((device const char *) q);
+
+ if (iq1 < args.ne01) {
+ for (short i = tiisg; i < PK4; i += NW) {
+ if (i < DK4) {
+ sq4[i] = (q4_t) q4[i];
+ } else {
+ sq4[i] = (q4_t) 0.0f;
+ }
+ }
+ }
+
+ // zero out so
+ for (short i = 0; i < DV4/NL; ++i) {
+ so4[i*NL] = (o4_t) 0.0f;
+ }
+
+ // zero out shared memory SH
+ for (short i = tiisg; i < SH/4; i += NW) {
+ ss4[i] = (s4_t) 0.0f;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ {
+ float S = 0.0f;
+ float M = -FLT_MAX/2;
+
+ // thread indices inside the simdgroup
+ const short tx = tiisg%NL;
+ const short ty = tiisg/NL;
+
+ // pointer to the mask
+ device const half * pm = (device const half *) (mask + iq1*args.nb31 + (iq2%args.ne32)*args.nb32 + (iq3%args.ne33)*args.nb33);
+
+ float slope = 1.0f;
+
+ // ALiBi
+ if (FC_flash_attn_ext_vec_has_bias) {
+ const short h = iq2;
+
+ const float base = h < args.n_head_log2 ? args.m0 : args.m1;
+ const short exph = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
+
+ slope = pow(base, exph);
+ }
+
+ // loop over the KV cache
+ // each simdgroup handles blocks of Q rows and C columns
+ for (int ic0 = iwg*NSG + sgitg; ; ic0 += NWG*NSG) {
+ int ic = ic0*C;
+ if (ic >= args.ne11) {
+ break;
+ }
+
+ // the last partial chunk uses the pad buffer as source
+ if (FC_flash_attn_ext_vec_has_kvpad && ic + C > args.ne11) {
+ k = pad;
+ v = k + args.nb11*C*args.ne_12_2*args.ne_12_3;
+ mask = v + args.nb21*C*args.ne_12_2*args.ne_12_3;
+
+ const short ikv2 = iq2/(args.ne02/args.ne_12_2);
+ const short ikv3 = iq3/(args.ne03/args.ne_12_3);
+
+ k += (ikv2 + ikv3*args.ne_12_2)*args.nb11*C;
+ v += (ikv2 + ikv3*args.ne_12_2)*args.nb21*C;
+
+ if (!FC_flash_attn_ext_vec_has_mask) {
+ if (ic + tiisg >= args.ne11) {
+ sm[tiisg] = -MAXHALF;
+ }
+ } else {
+ pm = (device const half *) (mask) +
+ iq1*C +
+ (iq2%args.ne32)*(C*args.ne31) +
+ (iq3%args.ne33)*(C*args.ne31*args.ne32);
+ }
+
+ ic = 0;
+ }
+
+ if (FC_flash_attn_ext_vec_has_mask) {
+ sm[tiisg] = pm[ic + tiisg];
+ }
+
+ // skip -INF blocks
+ if (simd_max(sm[tiisg]) <= -MAXHALF) {
+ continue;
+ }
+
+ // Q*K^T
+ {
+ device const k4_t * pk4 = (device const k4_t *) (k + ic*args.nb11);
+ threadgroup const q4_t * pq4 = sq4;
+
+ pk4 += ty*NS10/4 + tx;
+ pq4 += tx;
+
+ qk_t mqk[C/NE] = { [ 0 ... C/NE - 1] = 0.0f };
+
+ // each simdgroup processes 1 query and NE (NW/NL) cache elements
+ FOR_UNROLL (short cc = 0; cc < C/NE; ++cc) {
+ if (is_same<kd4_t, k4_t>::value) {
+ FOR_UNROLL (short ii = 0; ii < DK4/NL; ++ii) {
+ mqk[cc] += dot((float4) pk4[cc*NE*NS10/4 + ii*NL], (float4) pq4[ii*NL]);
+ }
+ } else {
+ device const kd4_t * pk = (device const kd4_t *) (k + ((ic + NE*cc + ty)*args.nb11));
+
+ k4_t mk;
+
+ FOR_UNROLL (short ii = 0; ii < DK4/NL; ++ii) {
+ const short i = ii*NL + tx;
+
+ deq_k_t4(pk + i/nl_k, i%nl_k, mk);
+
+ mqk[cc] += dot((float4) mk, (float4) sq4[i]);
+ }
+ }
+
+ if (NE == 1) {
+ mqk[cc] = simd_sum(mqk[cc]);
+ } else {
+ // simdgroup reduce (NE = 4)
+ // [ 0 .. 7] -> [ 0]
+ // [ 8 .. 15] -> [ 8]
+ // [16 .. 23] -> [16]
+ // [24 .. 31] -> [24]
+ if (NE <= 1) {
+ mqk[cc] += simd_shuffle_down(mqk[cc], 16);
+ }
+ if (NE <= 2) {
+ mqk[cc] += simd_shuffle_down(mqk[cc], 8);
+ }
+ if (NE <= 4) {
+ mqk[cc] += simd_shuffle_down(mqk[cc], 4);
+ }
+ if (NE <= 8) {
+ mqk[cc] += simd_shuffle_down(mqk[cc], 2);
+ }
+ if (NE <= 16) {
+ mqk[cc] += simd_shuffle_down(mqk[cc], 1);
+ }
+
+ // broadcast
+ mqk[cc] = simd_shuffle(mqk[cc], NL*ty);
+ }
+ }
+
+ if (FC_flash_attn_ext_vec_has_mask &&
+ !FC_flash_attn_ext_vec_has_scap &&
+ !FC_flash_attn_ext_vec_has_bias) {
+ ss[NE*tx + ty] = fma(mqk[tx], args.scale, (qk_t) sm[NE*tx + ty]);
+ } else {
+ mqk[tx] *= args.scale;
+
+ if (FC_flash_attn_ext_vec_has_scap) {
+ mqk[tx] = args.logit_softcap*precise::tanh(mqk[tx]);
+ }
+
+ if (FC_flash_attn_ext_vec_has_bias) {
+ mqk[tx] += (qk_t) sm[NE*tx + ty]*slope;
+ } else {
+ mqk[tx] += (qk_t) sm[NE*tx + ty];
+ }
+
+ ss[NE*tx + ty] = mqk[tx];
+ }
+ }
+
+ simdgroup_barrier(mem_flags::mem_threadgroup);
+
+ // online softmax
+ {
+ const float m = M;
+ const float s = ss[tiisg];
+
+ M = simd_max(max(M, s));
+
+ const float ms = exp(m - M);
+ const float vs = exp(s - M);
+
+ S = S*ms + simd_sum(vs);
+
+ // the P matrix from the paper (Q rows, C columns)
+ ss[tiisg] = vs;
+
+ // O = diag(ms)*O
+ if ((DV4/NL % NW == 0) || ty == 0) {
+ FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
+ so4[ii*NL] *= ms;
+ }
+ }
+ }
+
+ simdgroup_barrier(mem_flags::mem_threadgroup);
+
+ // O = O + (Q*K^T)*V
+ {
+ o4_t lo[DV4/NL];
+ FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
+ lo[ii] = 0.0f;
+ }
+
+ if (is_same<vd4_t, v4_t>::value) {
+ device const v4_t * pv4 = (device const v4_t *) (v + ic*args.nb21);
+
+ pv4 += ty*NS20/4 + tx;
+
+ const auto sst = ss + ty;
+
+ FOR_UNROLL (short cc = 0; cc < C/NE; ++cc) {
+ FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
+ lo[ii] += o4_t(float4(pv4[cc*NE*NS20/4 + ii*NL])*float4(sst[cc*NE]));
+ }
+ }
+ } else {
+ FOR_UNROLL (short cc = 0; cc < C/NE; ++cc) {
+ device const vd4_t * pv4 = (device const vd4_t *) (v + ((ic + NE*cc + ty)*args.nb21));
+
+ FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
+ const short i = ii*NL + tx;
+
+ v4_t mv;
+ deq_v_t4(pv4 + i/nl_v, i%nl_v, mv);
+
+ lo[ii] += o4_t(float4(mv)*float4(ss[NE*cc + ty]));
+ }
+ }
+ }
+
+ FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
+ if (NE > 1) {
+ lo[ii][0] += simd_shuffle_down(lo[ii][0], 16);
+ lo[ii][1] += simd_shuffle_down(lo[ii][1], 16);
+ lo[ii][2] += simd_shuffle_down(lo[ii][2], 16);
+ lo[ii][3] += simd_shuffle_down(lo[ii][3], 16);
+ }
+
+ if (NE > 2) {
+ lo[ii][0] += simd_shuffle_down(lo[ii][0], 8);
+ lo[ii][1] += simd_shuffle_down(lo[ii][1], 8);
+ lo[ii][2] += simd_shuffle_down(lo[ii][2], 8);
+ lo[ii][3] += simd_shuffle_down(lo[ii][3], 8);
+ }
+
+ if (NE > 4) {
+ lo[ii][0] += simd_shuffle_down(lo[ii][0], 4);
+ lo[ii][1] += simd_shuffle_down(lo[ii][1], 4);
+ lo[ii][2] += simd_shuffle_down(lo[ii][2], 4);
+ lo[ii][3] += simd_shuffle_down(lo[ii][3], 4);
+ }
+
+ if (NE > 8) {
+ lo[ii][0] += simd_shuffle_down(lo[ii][0], 2);
+ lo[ii][1] += simd_shuffle_down(lo[ii][1], 2);
+ lo[ii][2] += simd_shuffle_down(lo[ii][2], 2);
+ lo[ii][3] += simd_shuffle_down(lo[ii][3], 2);
+ }
+
+ if (NE > 16) {
+ lo[ii][0] += simd_shuffle_down(lo[ii][0], 1);
+ lo[ii][1] += simd_shuffle_down(lo[ii][1], 1);
+ lo[ii][2] += simd_shuffle_down(lo[ii][2], 1);
+ lo[ii][3] += simd_shuffle_down(lo[ii][3], 1);
+ }
+ }
+
+ if ((DV4/NL % NW == 0) || ty == 0) {
+ FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
+ so4[ii*NL] += lo[ii];
+ }
+ }
+ }
+ }
+
+ if (FC_flash_attn_ext_vec_has_sinks && sgitg == 0 && iwg == 0) {
+ const float m = M;
+ const float s = tiisg == 0 ? ((device const float *) sinks)[iq2] : -FLT_MAX/2;
+
+ M = simd_max(max(M, s));
+
+ const float ms = exp(m - M);
+ const float vs = exp(s - M);
+
+ S = S*ms + simd_sum(vs);
+
+ if ((DV4/NL % NW == 0) || ty == 0) {
+ FOR_UNROLL (short ii = 0; ii < DV4/NL; ++ii) {
+ so4[ii*NL] *= ms;
+ }
+ }
+ }
+
+ // these are needed for reducing the results from the simdgroups (reuse the ss buffer)
+ if (tiisg == 0) {
+ ss[0] = (s_t) S;
+ ss[1] = (s_t) M;
+ }
+ }
+
+ so4 -= tiisg;
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // parallel reduce
+ for (short r = NSG/2; r > 0; r >>= 1) {
+ if (sgitg < r) {
+ const float S0 = ss[ 0];
+ const float S1 = ss[r*(SH/2) + 0];
+
+ const float M0 = ss[ 1];
+ const float M1 = ss[r*(SH/2) + 1];
+
+ const float M = max(M0, M1);
+
+ const float ms0 = exp(M0 - M);
+ const float ms1 = exp(M1 - M);
+
+ const float S = S0*ms0 + S1*ms1;
+
+ if (tiisg == 0) {
+ ss[0] = S;
+ ss[1] = M;
+ }
+
+ // O_0 = diag(ms0)*O_0 + diag(ms1)*O_1
+ for (short i = tiisg; i < DV4; i += NW) {
+ so4[i] = so4[i]*ms0 + so4[i + r*PV4]*ms1;
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ // final rescale with 1/S and store to global memory
+ if (sgitg == 0) {
+ const int64_t nrows = args.ne3*args.ne2*args.ne1;
+ const int64_t rid = iq3*args.ne2*args.ne1 + iq2 + iq1*args.ne1;
+
+ device float4 * dst4 = (device float4 *) dst;
+ device float * dst1 = (device float *) dst + nrows*DV*NWG; // the S and M are stored after the results
+
+ const float S = NWG == 1 ? (ss[0] == 0.0f ? 0.0f : 1.0f/ss[0]) : 1.0f;
+
+ // interleave the workgroup data
+ for (short i = tiisg; i < DV4; i += NW) {
+ dst4[rid*DV4*NWG + NWG*i + iwg] = (float4) so4[i]*S;
+ }
+
+ // store S and M
+ if (NWG > 1) {
+ if (tiisg == 0) {
+ dst1[rid*(2*NWG) + 2*iwg + 0] = ss[0];
+ dst1[rid*(2*NWG) + 2*iwg + 1] = ss[1];
+ }
+ }
+ }
+
+#undef NWG
+#undef NSG
+#undef NS10
+#undef NS20
+}
+
+// note: I think the s_t can be half instead of float, because the Q*K scaling is done before storing to shared mem
+// in the other (non-vec) kernel, we need s_t to also be float because we scale during the soft_max
+//
+#define FA_TYPES \
+ half4, \
+ half4, \
+ half4, \
+ float, \
+ float, float4, \
+ float4
+
+#define FA_TYPES_F32 \
+ half4, \
+ float4, \
+ float4, \
+ float, \
+ float, float4, \
+ float4
+
+typedef decltype(kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 128, 128, 4>) flash_attn_ext_vec_t;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 32, 32, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 32, 32, 4>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 32, 32, 4>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 32, 32, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 32, 32, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 32, 32, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 32, 32, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk32_dv32")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 32, 32, 4>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 64, 64, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 64, 64, 2>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 64, 64, 2>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 64, 64, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 64, 64, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 64, 64, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 64, 64, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk64_dv64")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 64, 64, 2>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 96, 96, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 96, 96, 4>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 96, 96, 4>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 96, 96, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 96, 96, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 96, 96, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 96, 96, 4>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk96_dv96")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 96, 96, 4>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 128, 128, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 128, 128, 1>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 128, 128, 1>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 128, 128, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 128, 128, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 128, 128, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 128, 128, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk128_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 128, 128, 1>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 192, 192, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 192, 192, 2>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 192, 192, 2>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 192, 192, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 192, 192, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 192, 192, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 192, 192, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk192_dv192")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 192, 192, 2>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 192, 128, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 192, 128, 2>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 192, 128, 2>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 192, 128, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 192, 128, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 192, 128, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 192, 128, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk192_dv128")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 192, 128, 2>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 256, 256, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 256, 256, 1>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 256, 256, 1>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 256, 256, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 256, 256, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 256, 256, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 256, 256, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk256_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 256, 256, 1>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 320, 256, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 320, 256, 2>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 320, 256, 2>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 320, 256, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 320, 256, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 320, 256, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 320, 256, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk320_dv256")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 320, 256, 2>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 512, 512, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 512, 512, 1>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 512, 512, 1>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 512, 512, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 512, 512, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 512, 512, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 512, 512, 1>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk512_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 512, 512, 1>;
+
+template [[host_name("kernel_flash_attn_ext_vec_f32_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES_F32, float4, 1, dequantize_f32_t4, float4, 1, dequantize_f32_t4, 576, 512, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_f16_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, half4, 1, dequantize_f16_t4, half4, 1, dequantize_f16_t4, 576, 512, 2>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_flash_attn_ext_vec_bf16_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, bfloat4, 1, dequantize_bf16_t4, bfloat4, 1, dequantize_bf16_t4, 576, 512, 2>;
+#endif
+template [[host_name("kernel_flash_attn_ext_vec_q4_0_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_0, 8, dequantize_q4_0_t4, block_q4_0, 8, dequantize_q4_0_t4, 576, 512, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q4_1_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q4_1, 8, dequantize_q4_1_t4, block_q4_1, 8, dequantize_q4_1_t4, 576, 512, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_0_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_0, 8, dequantize_q5_0_t4, block_q5_0, 8, dequantize_q5_0_t4, 576, 512, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q5_1_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q5_1, 8, dequantize_q5_1_t4, block_q5_1, 8, dequantize_q5_1_t4, 576, 512, 2>;
+template [[host_name("kernel_flash_attn_ext_vec_q8_0_dk576_dv512")]] kernel flash_attn_ext_vec_t kernel_flash_attn_ext_vec<FA_TYPES, block_q8_0, 8, dequantize_q8_0_t4, block_q8_0, 8, dequantize_q8_0_t4, 576, 512, 2>;
+
+#undef FA_TYPES
+#undef FA_TYPES_F32
+
+constant int32_t FC_flash_attn_ext_vec_reduce_DV [[function_constant(FC_FLASH_ATTN_EXT_VEC_REDUCE + 0)]];
+constant int32_t FC_flash_attn_ext_vec_reduce_NWG [[function_constant(FC_FLASH_ATTN_EXT_VEC_REDUCE + 1)]];
+
+kernel void kernel_flash_attn_ext_vec_reduce(
+ constant ggml_metal_kargs_flash_attn_ext_vec_reduce & args,
+ device const char * htmp,
+ device char * dst,
+ uint tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+#define NWG (FC_flash_attn_ext_vec_reduce_NWG)
+#define DV (FC_flash_attn_ext_vec_reduce_DV)
+
+ const uint64_t rid = tgpig;
+
+ const short iwg = tiisg;
+
+ device const float * ss = (device const float *) htmp + (uint64_t)args.nrows*DV*NWG;
+
+ float S = ss[rid*(2*NWG) + 2*iwg + 0];
+ float M = ss[rid*(2*NWG) + 2*iwg + 1];
+
+ const float m = simd_max(M);
+ const float ms = exp(M - m);
+
+ S = simd_sum(S*ms);
+ S = S == 0.0f ? 0.0f : 1.0f/S;
+
+ const short DV4 = DV/4;
+
+ device const float4 * htmp4 = (device const float4 *) htmp + rid*DV4*NWG;
+ device float4 * dst4 = (device float4 *) dst + rid*DV4;
+
+ for (short i = sgitg; i < DV4; i += NWG) {
+ const float4 v = simd_sum(htmp4[i*NWG + iwg]*ms);
+
+ if (iwg == 0) {
+ dst4[i] = v*S;
+ }
+ }
+
+#undef NWG
+#undef DV
+}
+
+template<
+ typename kd4x4_t,
+ short nl_k,
+ void (*deq_k)(device const kd4x4_t *, short, thread half4x4 &)>
+kernel void kernel_lightning_indexer(
+ constant ggml_metal_kargs_lightning_indexer & args,
+ device const char * q,
+ device const char * k,
+ device const char * w,
+ device const char * m,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ constexpr short DK = OP_LIGHTNING_INDEXER_DK;
+ constexpr short NH = OP_LIGHTNING_INDEXER_NH;
+ constexpr short NHPTG = OP_LIGHTNING_INDEXER_NHPTG;
+ constexpr short NKPSG = OP_LIGHTNING_INDEXER_NKPSG;
+ constexpr short NSG = OP_LIGHTNING_INDEXER_NSG;
+ constexpr short NBPTG = OP_LIGHTNING_INDEXER_NBPTG;
+
+ constexpr short DK4 = DK/4;
+ constexpr short DK8 = DK/8;
+ constexpr short DK16 = DK/16;
+
+ constexpr short NK = NKPSG*NSG; // keys per threadgroup
+ constexpr short NTG = 32*NSG; // threads per threadgroup
+
+ const int i_stream = tgpig.z;
+ const int i_kv_0 = tgpig.x*NK; // first key of this threadgroup
+ const int i_kv = i_kv_0 + sgitg*NKPSG; // first key of this simdgroup
+
+ threadgroup half sk[NK * DK16 * 16];
+ threadgroup half4x4 * sk4x4 = (threadgroup half4x4 *) sk;
+
+ for (short i = tiitg; i < NK*DK16; i += NTG) {
+ const short ik = i/DK16;
+ const short i16 = i%DK16;
+
+ half4x4 tmp;
+
+ if (i_kv_0 + ik < args.n_kv) {
+ device const kd4x4_t * kr = (device const kd4x4_t *) (k + (i_kv_0 + ik)*args.nbk2 + i_stream*args.nbk3);
+
+ deq_k(kr + i16/nl_k, i16%nl_k, tmp);
+ } else {
+ FOR_UNROLL (short j = 0; j < 4; ++j) {
+ tmp[j] = half4(0.0h);
+ }
+ }
+
+ sk4x4[i] = tmp;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // K tile of this simdgroup, transposed to [DK, NKPSG]
+ simdgroup_half8x8 mk[DK8];
+
+ FOR_UNROLL (short i = 0; i < DK8; ++i) {
+ simdgroup_load(mk[i], sk + sgitg*NKPSG*DK + 8*i, DK, 0, true);
+ }
+
+ threadgroup half4 sq4[NHPTG*DK4];
+ threadgroup half * sq = (threadgroup half *) sq4;
+
+ threadgroup float sw [NHPTG];
+ threadgroup float sqk[NSG*NHPTG*NKPSG];
+
+ const int i_batch_0 = tgpig.y*NBPTG;
+ const int n_batch = min((int) NBPTG, args.n_batch - i_batch_0);
+
+ for (short ib = 0; ib < n_batch; ++ib) {
+ const int i_batch = i_batch_0 + ib;
+
+ device const char * pq = q + i_batch*args.nbq2 + i_stream*args.nbq3;
+ device const char * pw = w + i_batch*args.nbw1 + i_stream*args.nbw3;
+
+ float score = 0.0f;
+
+ FOR_UNROLL (short i_head = 0; i_head < NH; i_head += NHPTG) {
+ // stage the Q tile [DK, NHPTG] and the (prescaled) head weights
+ for (short i = tiitg; i < NHPTG*DK4; i += NTG) {
+ const short ih = i/DK4;
+ const short i4 = i%DK4;
+
+ device const float4 * q4 = (device const float4 *) (pq + (i_head + ih)*args.nbq1);
+
+ sq4[ih*DK4 + i4] = half4(q4[i4]);
+ }
+
+ if (tiitg < NHPTG) {
+ sw[tiitg] = ((device const float *) pw)[i_head + tiitg];
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ simdgroup_float8x8 mqk = make_filled_simdgroup_matrix<float, 8>(0.0f);
+
+ FOR_UNROLL (short i = 0; i < DK8; ++i) {
+ simdgroup_half8x8 mq;
+
+ simdgroup_load(mq, sq + 8*i, DK, 0, false);
+ simdgroup_multiply_accumulate(mqk, mq, mk[i], mqk);
+ }
+
+ threadgroup float * pqk = sqk + sgitg*NHPTG*NKPSG;
+
+ simdgroup_store(mqk, pqk, NKPSG, 0, false);
+ simdgroup_barrier(mem_flags::mem_threadgroup);
+
+ // one lane per key: ReLU, apply the head weight and accumulate over the head tile
+ if (tiisg < NKPSG) {
+ FOR_UNROLL (short ih = 0; ih < NHPTG; ++ih) {
+ score += max(pqk[ih*NKPSG + tiisg], 0.0f)*sw[ih];
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ if (tiisg < NKPSG) {
+ const int ik = i_kv + tiisg;
+ if (ik < args.n_kv) {
+ device const half * pm = (device const half *) (m + i_batch*args.nbm1 + (i_stream % args.mask_ne3)*args.nbm3);
+ device float * pd = (device float *) (dst + i_batch*args.nb1 + i_stream*args.nb3);
+
+ pd[ik] = score + (float) pm[ik];
+ }
+ }
+ }
+}
+
+typedef decltype(kernel_lightning_indexer<half4x4, 1, dequantize_f16>) kernel_lightning_indexer_t;
+
+template [[host_name("kernel_lightning_indexer_f32")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<float4x4, 1, dequantize_f32>;
+template [[host_name("kernel_lightning_indexer_f16")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<half4x4, 1, dequantize_f16>;
+
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_lightning_indexer_bf16")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<bfloat4x4, 1, dequantize_bf16>;
+#endif
+
+template [[host_name("kernel_lightning_indexer_q4_0")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q4_0, 2, dequantize_q4_0>;
+template [[host_name("kernel_lightning_indexer_q4_1")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q4_1, 2, dequantize_q4_1>;
+template [[host_name("kernel_lightning_indexer_q5_0")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q5_0, 2, dequantize_q5_0>;
+template [[host_name("kernel_lightning_indexer_q5_1")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q5_1, 2, dequantize_q5_1>;
+template [[host_name("kernel_lightning_indexer_q8_0")]] kernel kernel_lightning_indexer_t kernel_lightning_indexer<block_q8_0, 2, dequantize_q8_0>;
diff --git a/ggml/src/ggml-metal/kernels/gated_delta_net.metal b/ggml/src/ggml-metal/kernels/gated_delta_net.metal
new file mode 100644
index 0000000000..8422d8e29f
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/gated_delta_net.metal
@@ -0,0 +1,250 @@
+#include "common.h"
+
+constant short FC_gated_delta_net_ne20 [[function_constant(FC_GATED_DELTA_NET + 0)]];
+constant short FC_gated_delta_net_ne30 [[function_constant(FC_GATED_DELTA_NET + 1)]];
+constant short FC_gated_delta_net_K [[function_constant(FC_GATED_DELTA_NET + 2)]];
+
+#if 1
+template<short NSG>
+kernel void kernel_gated_delta_net_impl(
+ constant ggml_metal_kargs_gated_delta_net & args,
+ device const char * q,
+ device const char * k,
+ device const char * v,
+ device const char * g,
+ device const char * b,
+ device const char * s,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+#define S_v FC_gated_delta_net_ne20
+#define G FC_gated_delta_net_ne30
+#define K FC_gated_delta_net_K
+
+ const uint tx = tpitg.x;
+ const uint ty = tpitg.y;
+
+ const uint i23 = tgpig.z; // B (n_seqs)
+ const uint i21 = tgpig.y; // H (head)
+ const uint i20 = tgpig.x*NSG + ty; // row within S_v
+
+ const uint i01 = i21 % args.ne01;
+ const uint i11 = i21 % args.ne11;
+
+ const float scale = 1.0f / sqrt((float)S_v);
+
+ // input state layout [S_v, S_v, H, n_seqs] (s0 only): per-seq stride is H*D.
+ // state is stored transposed: M[i20][is] = S[is][i20], so row i20 is contiguous
+ const uint state_in_base = (i23*args.ne21 + i21)*S_v*S_v + i20*S_v;
+ device const float * s_ptr = (device const float *) (s) + state_in_base;
+
+ float ls[NSG];
+
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ ls[j] = s_ptr[is];
+ }
+
+ device float * dst_attn = (device float *) (dst) + (i23*args.ne22*args.ne21 + i21)*S_v + i20;
+
+ device const float * q_ptr = (device const float *) (q + i23*args.nb03 + i01*args.nb01);
+ device const float * k_ptr = (device const float *) (k + i23*args.nb13 + i11*args.nb11);
+ device const float * v_ptr = (device const float *) (v + i23*args.nb23 + i21*args.nb21);
+
+ device const float * b_ptr = (device const float *) (b) + (i23*args.ne22*args.ne21 + i21);
+ device const float * g_ptr = (device const float *) (g) + (i23*args.ne22*args.ne21 + i21)*G;
+
+ // snapshot slot mapping: slot 0 = most recent state, slot s = s tokens back.
+ // When n_tokens < K, only slots 0..n_tokens-1 are written; older slots are caller-owned.
+
+ // output state base offset: after attention scores
+ const uint attn_size = args.ne22 * args.ne21 * S_v * args.ne23;
+ // output state per-slot size: S_v * S_v * H * n_seqs
+ const uint state_size_per_snap = S_v * S_v * args.ne21 * args.ne23;
+ // per-(seq,head) offset within a slot
+ const uint state_out_base = (i23*args.ne21 + i21)*S_v*S_v + i20*S_v;
+
+ for (short t = 0; t < args.ne22; t++) {
+ float s_k = 0.0f;
+
+ if (G == 1) {
+ const float g_exp = exp(g_ptr[0]);
+
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ ls[j] *= g_exp;
+
+ s_k += ls[j]*k_ptr[is];
+ }
+ } else {
+ // KDA
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ ls[j] *= exp(g_ptr[is]);
+
+ s_k += ls[j]*k_ptr[is];
+ }
+ }
+
+ s_k = simd_sum(s_k);
+
+ const float d = (v_ptr[i20] - s_k)*b_ptr[0];
+
+ float y = 0.0f;
+
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ ls[j] += k_ptr[is]*d;
+
+ y += ls[j]*q_ptr[is];
+ }
+
+ y = simd_sum(y);
+
+ if (tx == 0) {
+ dst_attn[t*args.ne21*S_v] = y*scale;
+ }
+
+ q_ptr += args.ns02;
+ k_ptr += args.ns12;
+ v_ptr += args.ns22;
+
+ b_ptr += args.ne21;
+ g_ptr += args.ne21*G;
+
+ if (K > 1) {
+ const int target_slot = (int)args.ne22 - 1 - (int)t;
+ if (target_slot >= 0 && target_slot < (int)K) {
+ device float * dst_state = (device float *) (dst) + attn_size + (uint)target_slot * state_size_per_snap + state_out_base;
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ dst_state[is] = ls[j];
+ }
+ }
+ }
+ }
+
+ if (K == 1) {
+ device float * dst_state = (device float *) (dst) + attn_size + state_out_base;
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ dst_state[is] = ls[j];
+ }
+ }
+
+#undef S_v
+#undef G
+#undef K
+}
+
+typedef decltype(kernel_gated_delta_net_impl<4>) kernel_gated_delta_net_t;
+
+template [[host_name("kernel_gated_delta_net_f32_1")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<1>;
+template [[host_name("kernel_gated_delta_net_f32_2")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<2>;
+template [[host_name("kernel_gated_delta_net_f32_4")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<4>;
+
+#else
+// a simplified version of the above
+// no performance improvement, so keep the above version for now
+
+template<typename T, short NSG>
+kernel void kernel_gated_delta_net_impl(
+ constant ggml_metal_kargs_gated_delta_net & args,
+ device const char * q,
+ device const char * k,
+ device const char * v,
+ device const char * g,
+ device const char * b,
+ device const char * s,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+#define S_v FC_gated_delta_net_ne20
+#define G FC_gated_delta_net_ne30
+
+ const uint tx = tpitg.x;
+ const uint ty = tpitg.y;
+
+ const uint i23 = tgpig.z; // B
+ const uint i21 = tgpig.y; // H
+ const uint i20 = tgpig.x*NSG + ty;
+
+ const uint i01 = i21 % args.ne01;
+ const uint i11 = i21 % args.ne11;
+
+ const float scale = 1.0f / sqrt((float)S_v);
+
+ device const float * s_ptr = (device const float *) (s) + (i23*args.ne21 + i21)*S_v*S_v + i20;
+
+ float lsf[NSG];
+
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ lsf[j] = s_ptr[is*S_v];
+ }
+
+ thread T * ls = (thread T *) (lsf);
+
+ device float * dst_attn = (device float *) (dst) + (i23*args.ne22*args.ne21 + i21)*S_v + i20;
+
+ device const float * q_ptr = (device const float *) (q + i23*args.nb03 + i01*args.nb01);
+ device const float * k_ptr = (device const float *) (k + i23*args.nb13 + i11*args.nb11);
+ device const float * v_ptr = (device const float *) (v + i23*args.nb23 + i21*args.nb21);
+
+ device const float * b_ptr = (device const float *) (b) + (i23*args.ne22*args.ne21 + i21);
+ device const float * g_ptr = (device const float *) (g) + (i23*args.ne22*args.ne21 + i21)*G;
+
+ for (short t = 0; t < args.ne22; t++) {
+ device const T * qt_ptr = (device const T *) (q_ptr);
+ device const T * kt_ptr = (device const T *) (k_ptr);
+ device const T * gt_ptr = (device const T *) (g_ptr);
+
+ if (G == 1) {
+ *ls *= exp(g_ptr[0]);
+ } else {
+ // KDA
+ *ls *= exp(gt_ptr[tx]);
+ }
+
+ const float s_k = simd_sum(dot(*ls, kt_ptr[tx]));
+
+ const float d = (v_ptr[i20] - s_k)*b_ptr[0];
+
+ *ls += kt_ptr[tx]*d;
+
+ const float y = simd_sum(dot(*ls, qt_ptr[tx]));
+
+ if (tx == 0) {
+ *dst_attn = y*scale;
+ }
+
+ q_ptr += args.ns02;
+ k_ptr += args.ns12;
+ v_ptr += args.ns22;
+
+ b_ptr += args.ne21;
+ g_ptr += args.ne21*G;
+
+ dst_attn += args.ne21*S_v;
+ }
+
+ device float * dst_state = (device float *) (dst) + args.ne23*args.ne22*args.ne21*S_v + (i23*args.ne21 + i21)*S_v*S_v + i20;
+ device T * dstt_state = (device T *) (dst_state);
+
+ FOR_UNROLL (short j = 0; j < NSG; j++) {
+ const short is = tx*NSG + j;
+ dst_state[is*S_v] = lsf[j];
+ }
+
+#undef S_v
+#undef G
+}
+
+typedef decltype(kernel_gated_delta_net_impl<float4, 4>) kernel_gated_delta_net_t;
+
+template [[host_name("kernel_gated_delta_net_f32_1")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<float, 1>;
+template [[host_name("kernel_gated_delta_net_f32_2")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<float2, 2>;
+template [[host_name("kernel_gated_delta_net_f32_4")]] kernel kernel_gated_delta_net_t kernel_gated_delta_net_impl<float4, 4>;
+#endif
diff --git a/ggml/src/ggml-metal/kernels/misc.metal b/ggml/src/ggml-metal/kernels/misc.metal
new file mode 100644
index 0000000000..11104b4d8d
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/misc.metal
@@ -0,0 +1,595 @@
+#include "common.h"
+
+kernel void kernel_argmax_f32(
+ constant ggml_metal_kargs_argmax & args,
+ device const char * src0,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint sgitg[[simdgroup_index_in_threadgroup]],
+ uint tiisg[[thread_index_in_simdgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ device const float * x_row = (device const float *) ((device const char *) src0 + tgpig * args.nb01);
+
+ float lmax = -INFINITY;
+ int32_t larg = -1;
+
+ for (int i00 = tpitg; i00 < args.ne00; i00 += ntg) {
+ if (x_row[i00] > lmax) {
+ lmax = x_row[i00];
+ larg = i00;
+ }
+ }
+
+ // find the argmax value in the block
+ float max_val = simd_max(lmax);
+ int32_t arg_val = simd_max(select(-1, larg, lmax == max_val));
+
+ device int32_t * dst_i32 = (device int32_t *) dst;
+
+ threadgroup float * shared_maxval = (threadgroup float *) shmem;
+ threadgroup int32_t * shared_argmax = (threadgroup int32_t *) shmem + N_SIMDWIDTH;
+
+ if (ntg > N_SIMDWIDTH) {
+ if (sgitg == 0) {
+ shared_maxval[tiisg] = -INFINITY;
+ shared_argmax[tiisg] = -1;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ shared_maxval[sgitg] = max_val;
+ shared_argmax[sgitg] = arg_val;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ max_val = shared_maxval[tiisg];
+ arg_val = shared_argmax[tiisg];
+
+ float max_val_reduced = simd_max(max_val);
+ int32_t arg_val_reduced = simd_max(select(-1, arg_val, max_val == max_val_reduced));
+
+ dst_i32[tgpig] = arg_val_reduced;
+
+ return;
+ }
+
+ dst_i32[tgpig] = arg_val;
+}
+
+kernel void kernel_diag_f32(
+ constant ggml_metal_kargs_diag & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]]) {
+ constexpr short NW = N_SIMDWIDTH;
+
+ const int32_t i3 = tgpig.z;
+ const int32_t i2 = tgpig.y;
+ const int32_t i1 = tgpig.x;
+
+ device const float * src0_ptr = (device const float *)(src0 + i2*args.nb02 + i3*args.nb03);
+ device float * dst_ptr = (device float *)(dst + i1*args.nb01 + i2*args.nb2 + i3*args.nb3);
+
+ for (int i0 = tiitg; i0 < args.ne0; i0 += NW) {
+ dst_ptr[i0] = i0 == i1 ? src0_ptr[i0] : 0.0f;
+ }
+}
+
+kernel void kernel_roll_f32(
+ constant ggml_metal_kargs_roll & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int64_t i3 = tgpig.z;
+ const int64_t i2 = tgpig.y;
+ const int64_t i1 = tgpig.x;
+
+ device const float * src0_ptr = (device const float *) src0;
+ device float * dst_ptr = (device float *) dst;
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ // apply shifts and wrap around
+ int64_t i00 = i0 - args.s0;
+ int64_t i01 = i1 - args.s1;
+ int64_t i02 = i2 - args.s2;
+ int64_t i03 = i3 - args.s3;
+
+ if (i00 < 0) { i00 += args.ne00; } else if (i00 >= args.ne00) { i00 -= args.ne00; }
+ if (i01 < 0) { i01 += args.ne01; } else if (i01 >= args.ne01) { i01 -= args.ne01; }
+ if (i02 < 0) { i02 += args.ne02; } else if (i02 >= args.ne02) { i02 -= args.ne02; }
+ if (i03 < 0) { i03 += args.ne03; } else if (i03 >= args.ne03) { i03 -= args.ne03; }
+
+ int64_t src_idx = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00 + i00;
+ int64_t dst_idx = i3 *args.ne2 *args.ne1 *args.ne0 + i2 *args.ne1 *args.ne0 + i1 *args.ne0 + i0;
+
+ dst_ptr[dst_idx] = src0_ptr[src_idx];
+ }
+}
+
+template <typename T>
+kernel void kernel_pad_impl(
+ constant ggml_metal_kargs_pad & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+ const int32_t i3 = tgpig.z;
+ const int32_t i2 = tgpig.y;
+ const int32_t k0 = tgpig.x/args.ne1;
+ const int32_t i1 = tgpig.x - k0*args.ne1;
+
+ const int32_t i03 = i3;
+ const int32_t i02 = i2;
+ const int32_t i01 = i1;
+
+ device const T * src0_ptr = (device const T *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
+ device T * dst_ptr = (device T *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1);
+
+ for (int32_t l0 = 0; l0 < 1024; l0 += ntg.x) {
+ const int32_t i0 = k0*1024 + tpitg.x + l0;
+ if (i0 >= args.ne0) {
+ break;
+ }
+
+ if (i0 < args.ne00 && i1 < args.ne01 && i2 < args.ne02 && i3 < args.ne03) {
+ dst_ptr[i0] = src0_ptr[i0];
+ } else {
+ dst_ptr[i0] = 0.0f;
+ }
+ }
+}
+
+typedef decltype(kernel_pad_impl<float>) kernel_pad_t;
+
+template [[host_name("kernel_pad_f32")]] kernel kernel_pad_t kernel_pad_impl<float>;
+template [[host_name("kernel_pad_f32_4")]] kernel kernel_pad_t kernel_pad_impl<float4>;
+
+// TODO: this is slow - optimize
+kernel void kernel_pad_reflect_1d_f32(
+ constant ggml_metal_kargs_pad_reflect_1d & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tgpg[[threadgroups_per_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int64_t i3 = tgpig.z;
+ const int64_t i2 = tgpig.y;
+ const int64_t i1 = tgpig.x;
+
+ const int64_t i03 = i3;
+ const int64_t i02 = i2;
+ const int64_t i01 = i1;
+
+ device const float * src0_ptr = (device const float *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
+ device float * dst_ptr = (device float *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1);
+
+ if (i1 < args.ne01 && i2 < args.ne02 && i3 < args.ne03) {
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ if (i0 < args.p0) {
+ dst_ptr[i0] = src0_ptr[args.p0 - i0];
+ } else if (i0 < args.ne0 - args.p1) {
+ dst_ptr[i0] = src0_ptr[i0 - args.p0];
+ } else {
+ dst_ptr[i0] = src0_ptr[(args.ne0 - args.p1 - args.p0) - (args.p1 + 1 - (args.ne0 - i0)) - 1];
+ }
+ }
+ }
+}
+
+kernel void kernel_arange_f32(
+ constant ggml_metal_kargs_arange & args,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ device float * dst_ptr = (device float *) dst;
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ dst_ptr[i0] = args.start + args.step * i0;
+ }
+}
+
+kernel void kernel_timestep_embedding_f32(
+ constant ggml_metal_kargs_timestep_embedding & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ int i = tgpig.x;
+ device float * embed_data = (device float *)(dst + i*args.nb1);
+
+ int half_ = args.dim / 2;
+ for (int j = tpitg.x; j < half_; j += ntg.x) {
+ float timestep = ((device float *)src0)[i];
+ float freq = (float)exp(-log((float)args.max_period) * j / half_);
+ float arg = timestep * freq;
+ embed_data[j ] = cos(arg);
+ embed_data[j + half_] = sin(arg);
+ }
+
+ if (args.dim % 2 != 0 && tpitg.x == 0) {
+ embed_data[2 * half_] = 0.f;
+ }
+}
+
+kernel void kernel_opt_step_adamw_f32(
+ constant ggml_metal_kargs_opt_step_adamw & args,
+ device float * x,
+ device const float * g,
+ device float * g_m,
+ device float * g_v,
+ device const float * pars,
+ uint gid[[thread_position_in_grid]]) {
+
+ if (gid >= args.np) {
+ return;
+ }
+
+ const float alpha = pars[0];
+ const float beta1 = pars[1];
+ const float beta2 = pars[2];
+ const float eps = pars[3];
+ const float wd = pars[4];
+ const float beta1h = pars[5];
+ const float beta2h = pars[6];
+
+ const float gi = g[gid];
+ const float gmi = g_m[gid] * beta1 + gi * (1.0f - beta1);
+ const float gvi = g_v[gid] * beta2 + gi * gi * (1.0f - beta2);
+
+ g_m[gid] = gmi;
+ g_v[gid] = gvi;
+
+ const float mh = gmi * beta1h;
+ const float vh = sqrt(gvi * beta2h) + eps;
+
+ x[gid] = x[gid] * (1.0f - alpha * wd) - alpha * mh / vh;
+}
+
+kernel void kernel_opt_step_sgd_f32(
+ constant ggml_metal_kargs_opt_step_sgd & args,
+ device float * x,
+ device const float * g,
+ device const float * pars,
+ uint gid[[thread_position_in_grid]]) {
+
+ if (gid >= args.np) {
+ return;
+ }
+
+ x[gid] = x[gid] * (1.0f - pars[0] * pars[1]) - pars[0] * g[gid];
+}
+
+template<typename T>
+kernel void kernel_memset(
+ constant ggml_metal_kargs_memset & args,
+ device T * dst,
+ uint tpig[[thread_position_in_grid]]) {
+ dst[tpig] = args.val;
+}
+
+typedef decltype(kernel_memset<int64_t>) kernel_memset_t;
+
+template [[host_name("kernel_memset_i64")]] kernel kernel_memset_t kernel_memset<int64_t>;
+
+constant short FC_count_equal_nsg [[function_constant(FC_COUNT_EQUAL + 0)]];
+
+template<typename T>
+kernel void kernel_count_equal(
+ constant ggml_metal_kargs_count_equal & args,
+ device const char * src0,
+ device const char * src1,
+ device atomic_int * dst,
+ threadgroup int32_t * shmem_i32 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const short NSG = FC_count_equal_nsg;
+
+ const int i3 = tgpig.z;
+ const int i2 = tgpig.y;
+ const int i1 = tgpig.x;
+
+ if (i3 >= args.ne03 || i2 >= args.ne02 || i1 >= args.ne01) {
+ return;
+ }
+
+ int sum = 0;
+
+ device const char * base0 = src0 + i1*args.nb01 + i2*args.nb02 + i3*args.nb03;
+ device const char * base1 = src1 + i1*args.nb11 + i2*args.nb12 + i3*args.nb13;
+
+ for (int64_t i0 = tpitg.x; i0 < args.ne00; i0 += ntg.x) {
+ const T v0 = *(device const T *)(base0 + i0*args.nb00);
+ const T v1 = *(device const T *)(base1 + i0*args.nb10);
+ sum += (v0 == v1);
+ }
+
+ sum = simd_sum(sum);
+
+ if (tiisg == 0) {
+ shmem_i32[sgitg] = sum;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (sgitg == 0) {
+ float v = 0.0f;
+ if (tpitg.x < NSG) {
+ v = shmem_i32[tpitg.x];
+ }
+
+ float total = simd_sum(v);
+ if (tpitg.x == 0) {
+ atomic_fetch_add_explicit(dst, (int32_t) total, memory_order_relaxed);
+ }
+ }
+}
+
+typedef decltype(kernel_count_equal<int32_t>) kernel_count_equal_t;
+
+template [[host_name("kernel_count_equal_i32")]] kernel kernel_count_equal_t kernel_count_equal<int32_t>;
+
+template <typename T>
+kernel void kernel_snake(
+ constant ggml_metal_kargs_snake & args,
+ device const T * x,
+ device const float * a,
+ device const float * inv_b,
+ device T * dst,
+ uint tgpig [[threadgroup_position_in_grid]],
+ uint tpitg [[thread_position_in_threadgroup]],
+ uint ntg [[threads_per_threadgroup]]) {
+
+ const int idx = tgpig * ntg + tpitg;
+ if (idx >= args.T * args.C) {
+ return;
+ }
+
+ const int c = idx / args.T; // x is [T, C], a / inv_b collapse to [1, C]
+ const float xi = float(x[idx]);
+ const float si = sin(a[c] * xi);
+ dst[idx] = T(xi + si * si * inv_b[c]);
+}
+
+template [[host_name("kernel_snake_f32")]] kernel void kernel_snake<float>(constant ggml_metal_kargs_snake &, device const float *, device const float *, device const float *, device float *, uint, uint, uint);
+template [[host_name("kernel_snake_f16")]] kernel void kernel_snake<half>(constant ggml_metal_kargs_snake &, device const half *, device const float *, device const float *, device half *, uint, uint, uint);
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_snake_bf16")]] kernel void kernel_snake<bfloat>(constant ggml_metal_kargs_snake &, device const bfloat *, device const float *, device const float *, device bfloat *, uint, uint, uint);
+#endif
+
+template<int N>
+kernel void kernel_fwht_f32(
+ constant ggml_metal_kargs_fwht & args,
+ device const float * src,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+
+ constexpr int NW = N_SIMDWIDTH;
+ constexpr int NE = N / NW;
+
+ const float scale = 1.0f / sqrt((float) N);
+
+ const int sg_per_tg = ntg.x / NW;
+ const int64_t r = tgpig.x * sg_per_tg + sgitg;
+ if (r >= args.nrows) {
+ return;
+ }
+
+ src += r * N;
+ dst += r * N;
+
+ const int lane = tiisg;
+
+ float reg[NE];
+ for (int i = 0; i < NE; i++) {
+ reg[i] = src[i*NW + lane]*scale;
+ }
+ for (int i = 1; i < NW; i *= 2) {
+ for (int j = 0; j < NE; j++) {
+ const float val = reg[j];
+ const float val2 = simd_shuffle_xor(val, i);
+ reg[j] = (lane & i) == 0 ? val2 + val : val2 - val;
+ }
+ }
+
+ for (int i = NW; i < N; i *= 2) {
+ const int step = i / NW;
+ for (int j = 0; j < NE; j += (2 * step)) {
+ for (int k = 0; k < step; k++) {
+ const float x = reg[j + k ];
+ const float y = reg[j + k + step];
+ reg[j + k] = x + y;
+ reg[j + k + step] = x - y;
+ }
+ }
+ }
+
+ for (int i = 0; i < NE; i++) {
+ dst[i*NW + lane] = reg[i];
+ }
+}
+
+typedef decltype(kernel_fwht_f32<64>) kernel_fwht_t;
+
+template [[host_name("kernel_fwht_f32_64")]] kernel kernel_fwht_t kernel_fwht_f32<64>;
+template [[host_name("kernel_fwht_f32_128")]] kernel kernel_fwht_t kernel_fwht_f32<128>;
+template [[host_name("kernel_fwht_f32_256")]] kernel kernel_fwht_t kernel_fwht_f32<256>;
+template [[host_name("kernel_fwht_f32_512")]] kernel kernel_fwht_t kernel_fwht_f32<512>;
+
+kernel void kernel_dsv4_hc_comb_f32(
+ constant ggml_metal_kargs_dsv4_hc_comb & args,
+ device const char * mixes,
+ device const char * scale,
+ device const char * base,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ constexpr ushort hc = 4;
+ constexpr ushort comb_offset = 2*hc;
+
+ const int it = tgpig.x*ntg.y + sgitg;
+ if (it >= args.n_tokens) {
+ return;
+ }
+
+ float scale_lane = 0.0f;
+ if (tiisg == 0) {
+ scale_lane = *(device const float *) (scale + 2*args.nb_s0);
+ }
+ const float scale_comb = simd_shuffle(scale_lane, 0);
+
+ float v = 0.0f;
+ if (tiisg < hc*hc) {
+ v = *(device const float *) (mixes + (comb_offset + tiisg)*args.nb_m0 + it*args.nb_m1)*scale_comb
+ + *(device const float *) (base + (comb_offset + tiisg)*args.nb_b0);
+ }
+
+ // Softmax across destinations (the four contiguous lanes for each source).
+ float vmax = max(v, simd_shuffle_xor(v, 1));
+ vmax = max(vmax, simd_shuffle_xor(vmax, 2));
+ v = exp(v - vmax);
+
+ float sum = v + simd_shuffle_xor(v, 1);
+ sum += simd_shuffle_xor(sum, 2);
+ v = v/sum + args.eps;
+
+ // Normalize columns: equal destination indices are four lanes apart.
+ sum = v + simd_shuffle_xor(v, 4);
+ sum += simd_shuffle_xor(sum, 8);
+ v /= sum + args.eps;
+
+ for (int i = 1; i < args.n_iter; ++i) {
+ sum = v + simd_shuffle_xor(v, 1);
+ sum += simd_shuffle_xor(sum, 2);
+ v /= sum + args.eps;
+
+ sum = v + simd_shuffle_xor(v, 4);
+ sum += simd_shuffle_xor(sum, 8);
+ v /= sum + args.eps;
+ }
+
+ if (tiisg < hc*hc) {
+ const ushort idst = tiisg & 3;
+ const ushort isrc = tiisg >> 2;
+ *(device float *) (dst + idst*args.nb_d0 + isrc*args.nb_d1 + it*args.nb_d2) = v;
+ }
+}
+
+kernel void kernel_dsv4_hc_pre_f32(
+ constant ggml_metal_kargs_dsv4_hc_pre & args,
+ device const char * x,
+ device const char * weights,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ constexpr ushort hc = 4;
+
+ const int it = tgpig.y;
+ const int i0 = ((int) tgpig.x*ntg.y + sgitg)*32 + tiisg;
+
+ float weight_lane = 0.0f;
+ if (tiisg < hc) {
+ weight_lane = *(device const float *) (weights + tiisg*args.nb_w0 + it*args.nb_w1);
+ }
+
+ float w[hc];
+ FOR_UNROLL (ushort ih = 0; ih < hc; ++ih) {
+ w[ih] = simd_shuffle(weight_lane, ih);
+ }
+
+ if (i0 >= args.n_embd) {
+ return;
+ }
+
+ device const char * xb = x + i0*args.nb_x0 + it*args.nb_x2;
+ float result = 0.0f;
+ FOR_UNROLL (ushort ih = 0; ih < hc; ++ih) {
+ result = fma(*(device const float *) (xb + ih*args.nb_x1), w[ih], result);
+ }
+
+ *(device float *) (dst + i0*args.nb_d0 + it*args.nb_d1) = result;
+}
+
+kernel void kernel_dsv4_hc_post_f32(
+ constant ggml_metal_kargs_dsv4_hc_post & args,
+ device const char * x,
+ device const char * residual,
+ device const char * post,
+ device const char * comb,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ constexpr ushort hc = 4;
+
+ const int it = tgpig.y;
+ const int i0 = ((int) tgpig.x*ntg.y + sgitg)*32 + tiisg;
+
+ float coeff_lane = 0.0f;
+ if (tiisg < hc) {
+ coeff_lane = *(device const float *) (post + tiisg*args.nb_p0 + it*args.nb_p1);
+ } else if (tiisg < hc + hc*hc) {
+ const ushort idx = tiisg - hc;
+ const ushort idst = idx & 3;
+ const ushort isrc = idx >> 2;
+ coeff_lane = *(device const float *) (comb + idst*args.nb_c0 + isrc*args.nb_c1 + it*args.nb_c2);
+ }
+
+ float post_reg[hc];
+ float comb_reg[hc][hc];
+ FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
+ post_reg[idst] = simd_shuffle(coeff_lane, idst);
+ }
+ FOR_UNROLL (ushort isrc = 0; isrc < hc; ++isrc) {
+ FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
+ comb_reg[isrc][idst] = simd_shuffle(coeff_lane, hc + idst + hc*isrc);
+ }
+ }
+
+ if (i0 >= args.n_embd) {
+ return;
+ }
+
+ const float xv = *(device const float *) (x + i0*args.nb_x0 + it*args.nb_x1);
+ float result[hc];
+ FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
+ result[idst] = xv*post_reg[idst];
+ }
+
+ device const char * rb = residual + i0*args.nb_r0 + it*args.nb_r2;
+ FOR_UNROLL (ushort isrc = 0; isrc < hc; ++isrc) {
+ const float rv = *(device const float *) (rb + isrc*args.nb_r1);
+ FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
+ result[idst] = fma(rv, comb_reg[isrc][idst], result[idst]);
+ }
+ }
+
+ FOR_UNROLL (ushort idst = 0; idst < hc; ++idst) {
+ *(device float *) (dst + i0*args.nb_d0 + idst*args.nb_d1 + it*args.nb_d2) = result[idst];
+ }
+}
diff --git a/ggml/src/ggml-metal/kernels/mul_mm.metal b/ggml/src/ggml-metal/kernels/mul_mm.metal
new file mode 100644
index 0000000000..ee848eed6d
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/mul_mm.metal
@@ -0,0 +1,853 @@
+#include "common.h"
+#include "dequantize.h"
+
+constant bool FC_mul_mm_bc_inp [[function_constant(FC_MUL_MM + 0)]];
+constant bool FC_mul_mm_bc_out [[function_constant(FC_MUL_MM + 1)]];
+constant short FC_mul_mm_ne12 [[function_constant(FC_MUL_MM + 2)]];
+constant short FC_mul_mm_ne13 [[function_constant(FC_MUL_MM + 3)]];
+constant short FC_mul_mm_r2 [[function_constant(FC_MUL_MM + 4)]];
+constant short FC_mul_mm_r3 [[function_constant(FC_MUL_MM + 5)]];
+
+// each block_q contains 16*nl weights
+#ifdef GGML_METAL_HAS_TENSOR
+template<
+ typename SA, typename SA_4x4, typename SA_8x8,
+ typename SB, typename SB_2x4, typename SB_8x8,
+ typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread SA_4x4 &),
+ typename T0, typename T0_4x4, typename T1, typename T1_2x4>
+kernel void kernel_mul_mm(
+ constant ggml_metal_kargs_mul_mm & args,
+ device const char * srcA,
+ device const char * srcB,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig [[threadgroup_position_in_grid]],
+ ushort tiitg [[thread_index_in_threadgroup]],
+ ushort sgitg [[simdgroup_index_in_threadgroup]]) {
+ (void) sgitg;
+
+ // Matrix dimensions: A(M,K) x B(K,N) -> C(M,N)
+ const int K = args.ne00;
+ const int M = args.ne0;
+ const int N = args.ne1;
+
+ // Batch dimension handling
+ const int im = tgpig.z;
+ const int i12 = im % FC_mul_mm_ne12;
+ const int i13 = im / FC_mul_mm_ne12;
+
+ // Batch offsets for srcA and srcB
+ const uint64_t offset0 = (i12/FC_mul_mm_r2)*args.nb02 + (i13/FC_mul_mm_r3)*args.nb03;
+
+ // Tile dimensions
+ constexpr int NRB = SZ_SIMDGROUP * N_MM_BLOCK_X * N_MM_SIMD_GROUP_X;
+ constexpr int NRA = SZ_SIMDGROUP * N_MM_BLOCK_Y * N_MM_SIMD_GROUP_Y;
+
+ // Tile offsets in output matrix
+ const int ra = tgpig.y * NRA;
+ const int rb = tgpig.x * NRB;
+
+ // Threadgroup memory for dequantized A tile only
+ threadgroup SA * sa = (threadgroup SA *)(shmem);
+
+ // Work-item count for A loading
+ constexpr int A_WORK_ITEMS = NRA * N_MM_NK;
+ constexpr int NUM_THREADS = N_SIMDWIDTH * N_MM_SIMD_GROUP_X * N_MM_SIMD_GROUP_Y;
+
+ // tA wraps threadgroup memory
+ auto tA = tensor(sa, dextents<int32_t, 2>(N_MM_NK_TOTAL, NRA));
+
+ // tB wraps device memory directly
+ device T1 * ptrB = (device T1 *)(srcB + args.nb12*i12 + args.nb13*i13);
+ const int strideB = args.nb11 / sizeof(T1);
+ auto tB = tensor(ptrB, dextents<int32_t, 2>(K, N), array<int, 2>({1, strideB}));
+
+ // Configure matmul operation
+ // note: K is dynamic_extent (clamped to the valid range in PHASE 2), since a static
+ // N_MM_NK_TOTAL K tile would read src1 out of bounds when K % N_MM_NK_TOTAL != 0
+ // ref: https://github.com/ggml-org/llama.cpp/pull/27064
+ mpp::tensor_ops::matmul2d<
+ mpp::tensor_ops::matmul2d_descriptor(
+ NRB, NRA, static_cast<int>(dynamic_extent), false, true, true,
+ mpp::tensor_ops::matmul2d_descriptor::mode::multiply_accumulate),
+ execution_simdgroups<N_MM_SIMD_GROUP_X * N_MM_SIMD_GROUP_Y>> mm;
+
+ auto cT = mm.get_destination_cooperative_tensor<decltype(tB), decltype(tA), float>();
+
+ // Accumulate partial results over K dimension
+ for (int loop_k = 0; loop_k < K; loop_k += N_MM_NK_TOTAL) {
+ // === PHASE 1: Dequantization of A into threadgroup memory ===
+ for (int work = tiitg; work < A_WORK_ITEMS; work += NUM_THREADS) {
+ const int row = work / N_MM_NK;
+ const int k_chunk = work % N_MM_NK;
+ const int k_pos = loop_k + k_chunk * 16;
+ const short k_base = k_chunk * 16;
+
+ // Bounds check: skip device read if row is out of matrix bounds
+ if (ra + row < M) {
+ if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
+ // Element-wise reads when K is not aligned (nb01 not aligned for half4x4/float4x4).
+ // MSL spec Table 2.5: half4x4 requires 8-byte alignment. When K is odd,
+ // nb01 = K*2 is not 8-byte aligned, so odd-row pointers are misaligned.
+ // Mirrors the legacy kernel's existing guard.
+ device const T0 * row_ptr = (device const T0 *)(srcA + args.nb01 * (ra + row) + offset0);
+
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ sa[row * N_MM_NK_TOTAL + (k_base + i)] = (k_pos + i < K) ? (SA) row_ptr[k_pos + i] : (SA)0;
+ }
+ } else {
+ const int block_idx = k_pos / (16 * nl);
+ const short il = (k_pos / 16) % nl;
+
+ device const block_q * row_ptr = (device const block_q *)(srcA + args.nb01 * (ra + row) + offset0);
+
+ SA_4x4 temp_a;
+ dequantize_func(row_ptr + block_idx, il, temp_a);
+
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ // Zero-pad A for K positions beyond valid range (handles partial K iterations)
+ sa[row * N_MM_NK_TOTAL + (k_base + i)] = (k_pos + i < K) ? temp_a[i/4][i%4] : (SA)0;
+ }
+ }
+ } else {
+ // Zero-pad rows beyond matrix bounds
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ sa[row * N_MM_NK_TOTAL + (k_base + i)] = (SA)0;
+ }
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // === PHASE 2: Tensor matmul ===
+ // Clamp the K extent of both operand tensors to the remaining valid K range so
+ // the dynamic-K op never reads past the K extent of src1 (or the staged A tile).
+ const int kExt = min(N_MM_NK_TOTAL, K - loop_k);
+
+ auto tAv = tensor(sa, dextents<int32_t, 2>(kExt, NRA), array<int, 2>({1, N_MM_NK_TOTAL}));
+ auto tBv = tensor(ptrB + loop_k + rb * strideB, dextents<int32_t, 2>(kExt, N - rb), array<int, 2>({1, strideB}));
+
+ mm.run(tBv, tAv, cT);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ // Store result tile to output matrix (with batch offset)
+ // cT.store handles bounds checking via tD's extents (M, N)
+ device float * dstBatch = (device float *)dst + im * N * M;
+
+ auto tD = tensor(dstBatch, dextents<int32_t, 2>(M, N), array<int, 2>({1, M}));
+ cT.store(tD.slice(ra, rb));
+}
+
+#else
+
+template<
+ typename S0, typename S0_4x4, typename S0_8x8,
+ typename S1, typename S1_2x4, typename S1_8x8,
+ typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread S0_4x4 &),
+ typename T0, typename T0_4x4, typename T1, typename T1_2x4>
+kernel void kernel_mul_mm(
+ constant ggml_metal_kargs_mul_mm & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ threadgroup S0 * sa = (threadgroup S0 *)(shmem);
+ threadgroup S1 * sb = (threadgroup S1 *)(shmem + 4096);
+
+ constexpr int NR0 = 64;
+ constexpr int NR1 = 32;
+
+ constexpr int NK = 32;
+ constexpr int NL0 = NK/16;
+ constexpr int NL1 = NK/8;
+
+ const int im = tgpig.z;
+ const int r0 = tgpig.y*NR0;
+ const int r1 = tgpig.x*NR1;
+
+ // if this block is of 64x32 shape or smaller
+ const short nr0 = (args.ne0 - r0 < NR0) ? (args.ne0 - r0) : NR0;
+ const short nr1 = (args.ne1 - r1 < NR1) ? (args.ne1 - r1) : NR1;
+
+ // a thread shouldn't load data outside of the matrix
+ const short lr0 = ((short)tiitg/NL0) < nr0 ? ((short)tiitg/NL0) : nr0 - 1; // 0 .. 63
+ const short lr1 = ((short)tiitg/NL1) < nr1 ? ((short)tiitg/NL1) : nr1 - 1; // 0 .. 31
+
+ const short il0 = (tiitg % NL0);
+
+ short il = il0;
+
+ const int i12 = im % FC_mul_mm_ne12;
+ const int i13 = im / FC_mul_mm_ne12;
+
+ const uint64_t offset0 = (i12/FC_mul_mm_r2)*args.nb02 + (i13/FC_mul_mm_r3)*args.nb03;
+ const short offset1 = il0/nl;
+
+ device const block_q * x = (device const block_q *)(src0 + args.nb01*(r0 + lr0) + offset0) + offset1;
+
+ const short iy = 8*(tiitg % NL1);
+
+ device const T1 * y = (device const T1 *)(src1
+ + args.nb13*i13
+ + args.nb12*i12
+ + args.nb11*(r1 + lr1)
+ + args.nb10*iy);
+
+ S0_8x8 ma[4];
+ S1_8x8 mb[2];
+
+ simdgroup_float8x8 mc[8];
+
+ for (short i = 0; i < 8; i++){
+ mc[i] = make_filled_simdgroup_matrix<float, 8>(0.f);
+ }
+
+ for (int loop_k = 0; loop_k < args.ne00; loop_k += NK) {
+ // load data and store to threadgroup memory
+ if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // no need for dequantization
+ for (short i = 0; i < 16; i++) {
+ const short sx = 2*il0 + i/8;
+ const short sy = (tiitg/NL0)/8;
+
+ //const short lx = i%8;
+ //const short ly = (tiitg/NL0)%8;
+ const short lx = (tiitg/NL0)%8;
+ const short ly = i%8;
+
+ const short ib = 8*sx + sy;
+
+ *(sa + 64*ib + 8*ly + lx) = loop_k + 16*il + i < args.ne00 ? *((device T0 *) x + i) : 0;
+ }
+ } else {
+ S0_4x4 temp_a;
+ dequantize_func(x, il, temp_a);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ const short sx = 2*il0 + i/8;
+ const short sy = (tiitg/NL0)/8;
+
+ //const short lx = i%8;
+ //const short ly = (tiitg/NL0)%8;
+ const short lx = (tiitg/NL0)%8;
+ const short ly = i%8;
+
+ const short ib = 8*sx + sy;
+
+ // NOTE: this is massively slower.. WTF?
+ //sa[64*ib + 8*ly + lx] = temp_a[i/4][i%4];
+
+ *(sa + 64*ib + 8*ly + lx) = temp_a[i/4][i%4];
+ }
+ }
+
+ if (FC_mul_mm_bc_inp) {
+ for (short i = 0; i < 8; ++i) {
+ const short sx = (tiitg%NL1);
+ const short sy = (tiitg/NL1)/8;
+
+ const short lx = i;
+ const short ly = (tiitg/NL1)%8;
+ //const short lx = (tiitg/NL1)%8;
+ //const short ly = i;
+
+ const short ib = 4*sx + sy;
+
+ *(sb + 64*ib + 8*ly + lx) = loop_k + iy + i < args.ne00 ? (S1) *((device T1 *) y + i) : 0;
+ }
+ } else {
+ const short sx = (tiitg%NL1);
+ const short sy = (tiitg/NL1)/8;
+
+ //const short dx = sx;
+ //const short dy = sy;
+
+ const short ly = (tiitg/NL1)%8;
+
+ const short ib = 4*sx + sy;
+
+ *(threadgroup S1_2x4 *)(sb + 64*ib + 8*ly) = (S1_2x4)(*((device T1_2x4 *) y));
+ }
+
+ il = (il + 2 < nl) ? il + 2 : il % 2;
+ x = (il < 2) ? x + (2 + nl - 1)/nl : x;
+
+ y += NK;
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // load matrices from threadgroup memory and conduct outer products
+ threadgroup const S0 * lsma = (sa + 4*64*(sgitg%2));
+ threadgroup const S1 * lsmb = (sb + 2*64*(sgitg/2));
+
+ FOR_UNROLL (short ik = 0; ik < NK/8; ik++) {
+ simdgroup_barrier(mem_flags::mem_none);
+
+ FOR_UNROLL (short i = 0; i < 4; i++) {
+ simdgroup_load(ma[i], lsma + 64*i, 8, 0, false);
+ }
+
+ simdgroup_barrier(mem_flags::mem_none);
+
+ FOR_UNROLL (short i = 0; i < 2; i++) {
+ simdgroup_load(mb[i], lsmb + 64*i, 8, 0, false);
+ }
+
+ simdgroup_barrier(mem_flags::mem_none);
+
+ FOR_UNROLL (short i = 0; i < 8; i++){
+ simdgroup_multiply_accumulate(mc[i], mb[i/4], ma[i%4], mc[i]);
+ }
+
+ lsma += 8*64;
+ lsmb += 4*64;
+ }
+ }
+
+ if (!FC_mul_mm_bc_out || (r0 + NR0 <= args.ne0 && r1 + NR1 <= args.ne1)) {
+ // if no bounds checks on the output are needed, we can directly write to device memory
+ device float * C = (device float *) dst +
+ (r0 + 32*(sgitg & 1)) + \
+ (r1 + 16*(sgitg >> 1)) * args.ne0 + im*args.ne1*args.ne0;
+
+ for (short i = 0; i < 8; i++) {
+ simdgroup_store(mc[i], C + 8*(i%4) + 8*args.ne0*(i/4), args.ne0, 0, false);
+ }
+ } else {
+ // block is smaller than 64x32, we should avoid writing data outside of the matrix
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ threadgroup float * temp_str = ((threadgroup float *) shmem) + 32*(sgitg&1) + (16*(sgitg >> 1))*NR0;
+
+ for (short i = 0; i < 8; i++) {
+ simdgroup_store(mc[i], temp_str + 8*(i%4) + 8*NR0*(i/4), NR0, 0, false);
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (sgitg == 0) {
+ for (int j = tiitg; j < nr1; j += NR1) {
+ device float * D = (device float *) dst + r0 + (r1 + j)*args.ne0 + im*args.ne1*args.ne0;
+ device float4 * D4 = (device float4 *) D;
+
+ threadgroup float * C = temp_str + (j*NR0);
+ threadgroup float4 * C4 = (threadgroup float4 *) C;
+
+ int i = 0;
+ for (; i < nr0/4; i++) {
+ *(D4 + i) = *(C4 + i);
+ }
+
+ i *= 4;
+ for (; i < nr0; i++) {
+ *(D + i) = *(C + i);
+ }
+ }
+ }
+ }
+}
+
+#endif // GGML_METAL_HAS_TENSOR
+
+template<short ne20> // n_expert_used
+kernel void kernel_mul_mm_id_map0(
+ constant ggml_metal_kargs_mul_mm_id_map0 & args,
+ device const char * src2,
+ device char * htpe,
+ device char * hids,
+ threadgroup char * shmem [[threadgroup(0)]],
+ ushort tpitg[[thread_position_in_threadgroup]],
+ ushort ntg[[threads_per_threadgroup]]) {
+ const short ide = tpitg; // expert id
+
+ uint32_t n_all = 0;
+
+ device int32_t * ids_i32 = (device int32_t *) hids + ide*args.ne21;
+
+ for (int i21 = 0; i21 < args.ne21; i21 += ntg) { // n_tokens
+ if (i21 + tpitg < args.ne21) {
+ device const int32_t * src2_i32 = (device const int32_t *) (src2 + (i21 + tpitg)*args.nb21);
+
+ threadgroup uint16_t * sids = (threadgroup uint16_t *) shmem + tpitg*ne20;
+
+ #pragma unroll(ne20)
+ for (short i20 = 0; i20 < ne20; i20++) {
+ sids[i20] = src2_i32[i20];
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (short t = 0; t < ntg; t++) {
+ if (i21 + t >= args.ne21) {
+ break;
+ }
+
+ threadgroup const uint16_t * sids = (threadgroup const uint16_t *) shmem + t*ne20;
+
+ short sel = 0;
+ #pragma unroll(ne20)
+ for (short i20 = 0; i20 < ne20; i20++) {
+ sel += (sids[i20] == ide)*(i20 + 1);
+ }
+
+ ids_i32[n_all] = (i21 + t)*ne20 + sel - 1;
+
+ n_all += sel > 0;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ device uint32_t * tpe_u32 = (device uint32_t *) (htpe);
+ tpe_u32[ide] = n_all;
+}
+
+typedef decltype(kernel_mul_mm_id_map0<1>) kernel_mul_mm_id_map0_t;
+
+template [[host_name("kernel_mul_mm_id_map0_ne20_1" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<1>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_2" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<2>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_4" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<4>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_5" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<5>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_6" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<6>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_8" )]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<8>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_10")]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<10>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_16")]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<16>;
+template [[host_name("kernel_mul_mm_id_map0_ne20_22")]] kernel kernel_mul_mm_id_map0_t kernel_mul_mm_id_map0<22>;
+
+template<typename S0, typename S0_4x4, typename S0_8x8, typename S1, typename S1_2x4, typename S1_8x8, typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread S0_4x4 &), typename T0, typename T0_4x4, typename T1, typename T1_2x4>
+kernel void kernel_mul_mm_id(
+ constant ggml_metal_kargs_mul_mm_id & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * htpe,
+ device const char * hids,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ threadgroup S0 * sa = (threadgroup S0 *)(shmem);
+ threadgroup S1 * sb = (threadgroup S1 *)(shmem + 4096);
+
+#ifdef GGML_METAL_HAS_TENSOR
+ threadgroup float * sc = (threadgroup float *)(shmem);
+#endif
+
+ constexpr int NR0 = 64;
+ constexpr int NR1 = 32;
+
+ constexpr int NK = 32;
+ constexpr int NL0 = NK/16;
+ constexpr int NL1 = NK/8;
+
+ const int im = tgpig.z; // expert
+ const int r0 = tgpig.y*NR0;
+ const int r1 = tgpig.x*NR1;
+
+ device const uint32_t * tpe_u32 = (device const uint32_t *) (htpe);
+ device const int32_t * ids_i32 = (device const int32_t *) (hids);
+
+ const int32_t neh1 = tpe_u32[im];
+
+ if (r1 >= neh1) {
+ return;
+ }
+
+ // if this block is of 64x32 shape or smaller
+ const short nr0 = (args.ne0 - r0 < NR0) ? (args.ne0 - r0) : NR0;
+ const short nr1 = ( neh1 - r1 < NR1) ? ( neh1 - r1) : NR1;
+
+ // a thread shouldn't load data outside of the matrix
+ const short lr0 = ((short)tiitg/NL0) < nr0 ? ((short)tiitg/NL0) : nr0 - 1; // 0 .. 63
+ const short lr1 = ((short)tiitg/NL1) < nr1 ? ((short)tiitg/NL1) : nr1 - 1; // 0 .. 31
+
+ const short il0 = (tiitg % NL0);
+
+ short il = il0;
+
+ const int id = ids_i32[im*args.ne21 + r1 + lr1];
+
+ const short i11 = (id % args.ne20) % args.ne11;
+ const short i12 = (id / args.ne20);
+ const short i13 = 0;
+
+ const uint64_t offset0 = im*args.nb02 + i13*args.nb03;
+ const short offset1 = il0/nl;
+
+ device const block_q * x = (device const block_q *)(src0 + args.nb01*(r0 + lr0) + offset0) + offset1;
+
+ const short iy = 8*(tiitg % NL1);
+
+ device const T1 * y = (device const T1 *)(src1
+ + args.nb13*i13
+ + args.nb12*i12
+ + args.nb11*i11
+ + args.nb10*iy);
+
+#ifndef GGML_METAL_HAS_TENSOR
+ S0_8x8 ma[4];
+ S1_8x8 mb[2];
+
+ simdgroup_float8x8 mc[8];
+
+ for (short i = 0; i < 8; i++){
+ mc[i] = make_filled_simdgroup_matrix<float, 8>(0.f);
+ }
+#else
+ auto tA = tensor<threadgroup S0, dextents<int32_t, 2>, tensor_inline>(sa, dextents<int32_t, 2>(NK, NR0));
+ auto tB = tensor<threadgroup S1, dextents<int32_t, 2>, tensor_inline>(sb, dextents<int32_t, 2>(NR1, NK ));
+
+ mpp::tensor_ops::matmul2d<
+ mpp::tensor_ops::matmul2d_descriptor(NR1, NR0, NK, false, true, false, mpp::tensor_ops::matmul2d_descriptor::mode::multiply_accumulate),
+ execution_simdgroups<4>> mm;
+
+ auto cT = mm.get_destination_cooperative_tensor<decltype(tA), decltype(tB), float>();
+#endif
+
+ for (int loop_k = 0; loop_k < args.ne00; loop_k += NK) {
+#ifndef GGML_METAL_HAS_TENSOR
+ // load data and store to threadgroup memory
+ if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // no need for dequantization
+ for (short i = 0; i < 16; i++) {
+ const short sx = 2*il0 + i/8;
+ const short sy = (tiitg/NL0)/8;
+
+ //const short lx = i%8;
+ //const short ly = (tiitg/NL0)%8;
+ const short lx = (tiitg/NL0)%8;
+ const short ly = i%8;
+
+ const short ib = 8*sx + sy;
+
+ *(sa + 64*ib + 8*ly + lx) = loop_k + 16*il + i < args.ne00 ? (S0) *((device T0 *) x + i) : (S0) 0;
+ }
+ } else {
+ S0_4x4 temp_a;
+ dequantize_func(x, il, temp_a);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ const short sx = 2*il0 + i/8;
+ const short sy = (tiitg/NL0)/8;
+
+ //const short lx = i%8;
+ //const short ly = (tiitg/NL0)%8;
+ const short lx = (tiitg/NL0)%8;
+ const short ly = i%8;
+
+ const short ib = 8*sx + sy;
+
+ // NOTE: this is massively slower.. WTF?
+ //sa[64*ib + 8*ly + lx] = temp_a[i/4][i%4];
+
+ *(sa + 64*ib + 8*ly + lx) = temp_a[i/4][i%4];
+ }
+ }
+
+ if (FC_mul_mm_bc_inp) {
+ for (short i = 0; i < 8; ++i) {
+ const short sx = (tiitg%NL1);
+ const short sy = (tiitg/NL1)/8;
+
+ const short lx = i;
+ const short ly = (tiitg/NL1)%8;
+ //const short lx = (tiitg/NL1)%8;
+ //const short ly = i;
+
+ const short ib = 4*sx + sy;
+
+ *(sb + 64*ib + 8*ly + lx) = loop_k + iy + i < args.ne00 ? (S1) *((device T1 *) y + i) : 0;
+ }
+ } else {
+ const short sx = (tiitg%NL1);
+ const short sy = (tiitg/NL1)/8;
+
+ //const short dx = sx;
+ //const short dy = sy;
+
+ const short ly = (tiitg/NL1)%8;
+
+ const short ib = 4*sx + sy;
+
+ *(threadgroup S1_2x4 *)(sb + 64*ib + 8*ly) = (S1_2x4)(*((device T1_2x4 *) y));
+ }
+#else
+ // load data and store to threadgroup memory
+ if (is_same<T0_4x4, block_q>::value && FC_mul_mm_bc_inp) {
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // no need for dequantization
+ for (short i = 0; i < 16; i++) {
+ const short sx = 2*il0 + i/8;
+ const short sy = (tiitg/NL0)/8;
+
+ const short lx = i%8;
+ const short ly = (tiitg/NL0)%8;
+ //const short lx = (tiitg/NL0)%8;
+ //const short ly = i%8;
+
+ *(sa + NK*(8*sy + ly) + 8*sx + lx) = loop_k + 16*il + i < args.ne00 ? *((device T0 *) x + i) : 0;
+ }
+ } else {
+ S0_4x4 temp_a;
+ dequantize_func(x, il, temp_a);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ const short sx = 2*il0 + i/8;
+ const short sy = (tiitg/NL0)/8;
+
+ const short lx = i%8;
+ const short ly = (tiitg/NL0)%8;
+ //const short lx = (tiitg/NL0)%8;
+ //const short ly = i%8;
+
+ *(sa + NK*(8*sy + ly) + 8*sx + lx) = temp_a[i/4][i%4];
+ }
+ }
+
+ if (FC_mul_mm_bc_inp) {
+ for (short i = 0; i < 8; ++i) {
+ const short sx = (tiitg%NL1);
+ const short sy = (tiitg/NL1)/8;
+
+ const short lx = i;
+ const short ly = (tiitg/NL1)%8;
+ //const short lx = (tiitg/NL1)%8;
+ //const short ly = i;
+
+ *(sb + NK*(8*sy + ly) + 8*sx + lx) = loop_k + iy + i < args.ne00 ? (S1) *((device T1 *) y + i) : 0;
+ }
+ } else {
+ const short sx = (tiitg%NL1);
+ const short sy = (tiitg/NL1)/8;
+
+ //const short lx = i;
+ const short ly = (tiitg/NL1)%8;
+ //const short lx = (tiitg/NL1)%8;
+ //const short ly = i;
+
+ *(threadgroup S1_2x4 *)(sb + NK*(8*sy + ly) + 8*sx) = (S1_2x4)(*((device T1_2x4 *) y));
+ }
+#endif
+
+ il = (il + 2 < nl) ? il + 2 : il % 2;
+ x = (il < 2) ? x + (2 + nl - 1)/nl : x;
+
+ y += NK;
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+#ifndef GGML_METAL_HAS_TENSOR
+ // load matrices from threadgroup memory and conduct outer products
+ threadgroup const S0 * lsma = (sa + 4*64*(sgitg%2));
+ threadgroup const S1 * lsmb = (sb + 2*64*(sgitg/2));
+
+ FOR_UNROLL (short ik = 0; ik < NK/8; ik++) {
+ simdgroup_barrier(mem_flags::mem_none);
+
+ FOR_UNROLL (short i = 0; i < 4; i++) {
+ simdgroup_load(ma[i], lsma + 64*i, 8, 0, false);
+ }
+
+ simdgroup_barrier(mem_flags::mem_none);
+
+ FOR_UNROLL (short i = 0; i < 2; i++) {
+ simdgroup_load(mb[i], lsmb + 64*i, 8, 0, false);
+ }
+
+ simdgroup_barrier(mem_flags::mem_none);
+
+ FOR_UNROLL (short i = 0; i < 8; i++){
+ simdgroup_multiply_accumulate(mc[i], mb[i/4], ma[i%4], mc[i]);
+ }
+
+ lsma += 8*64;
+ lsmb += 4*64;
+ }
+#else
+ auto sA = tA.slice(0, 0);
+ auto sB = tB.slice(0, 0);
+
+ mm.run(sB, sA, cT);
+#endif
+ }
+
+ // block is smaller than 64x32, we should avoid writing data outside of the matrix
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+#ifdef GGML_METAL_HAS_TENSOR
+ auto tC = tensor<threadgroup float, dextents<int32_t, 2>, tensor_inline>(sc, dextents<int32_t, 2>(NR0, NR1));
+ cT.store(tC);
+#else
+ threadgroup float * temp_str = ((threadgroup float *) shmem) + 32*(sgitg&1) + (16*(sgitg >> 1))*NR0;
+
+ for (short i = 0; i < 8; i++) {
+ simdgroup_store(mc[i], temp_str + 8*(i%4) + 8*NR0*(i/4), NR0, 0, false);
+ }
+#endif
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (short j = sgitg; j < nr1; j += 4) {
+ const int id = ids_i32[im*args.ne21 + r1 + j];
+
+ const short ide = id % args.ne20;
+ const short idt = id / args.ne20;
+
+ device float * D = (device float *) dst + r0 + ide*args.ne0 + idt*args.ne1*args.ne0;
+ device float4 * D4 = (device float4 *) D;
+
+ threadgroup float * C = (threadgroup float *) shmem + j*NR0;
+ threadgroup float4 * C4 = (threadgroup float4 *) C;
+
+ int i = tiisg;
+ for (; i < nr0/4; i += 32) {
+ *(D4 + i) = *(C4 + i);
+ }
+
+ i = (4*(nr0/4)) + tiisg;
+ for (; i < nr0; i += 32) {
+ *(D + i) = *(C + i);
+ }
+ }
+}
+
+//
+// matrix-matrix multiplication
+//
+
+typedef decltype(kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>) mul_mm_t;
+
+template [[host_name("kernel_mul_mm_f32_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_f16_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, float, float2x4>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mm_bf16_f32")]] kernel mul_mm_t kernel_mul_mm<bfloat, bfloat4x4, simdgroup_bfloat8x8, bfloat, bfloat2x4, simdgroup_bfloat8x8, bfloat4x4, 1, dequantize_bf16, bfloat, bfloat4x4, float, float2x4>;
+#endif
+template [[host_name("kernel_mul_mm_q1_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q2_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q4_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q4_1_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q5_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q5_1_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q8_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_mxfp4_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q2_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q3_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q4_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q5_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_q6_K_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq2_xxs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq2_xs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq3_xxs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq3_s_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq2_s_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq1_s_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq1_m_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq4_nl_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_iq4_xs_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_tq2_0_f32")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, float, float2x4>;
+
+template [[host_name("kernel_mul_mm_f32_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_f16_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q1_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q2_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q4_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q4_1_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q5_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q5_1_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q8_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_mxfp4_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q2_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q3_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q4_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q5_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_q6_K_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq2_xxs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq2_xs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq3_xxs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq3_s_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq2_s_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq1_s_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq1_m_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq4_nl_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_iq4_xs_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_tq2_0_f16")]] kernel mul_mm_t kernel_mul_mm<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, half, half2x4>;
+
+//
+// indirect matrix-matrix multiplication
+//
+
+typedef decltype(kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>) mul_mm_id;
+
+template [[host_name("kernel_mul_mm_id_f32_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_f16_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, float, float2x4>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mm_id_bf16_f32")]] kernel mul_mm_id kernel_mul_mm_id<bfloat, bfloat4x4, simdgroup_bfloat8x8, bfloat, bfloat2x4, simdgroup_bfloat8x8, bfloat4x4, 1, dequantize_bf16, bfloat, bfloat4x4, float, float2x4>;
+#endif
+template [[host_name("kernel_mul_mm_id_q1_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q2_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q4_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q4_1_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q5_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q5_1_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q8_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_mxfp4_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q2_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q3_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q4_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q5_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_q6_K_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq2_xxs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq2_xs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq3_xxs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq3_s_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq2_s_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq1_s_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq1_m_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq4_nl_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_iq4_xs_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, float, float2x4>;
+template [[host_name("kernel_mul_mm_id_tq2_0_f32")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, float, float2x4>;
+
+template [[host_name("kernel_mul_mm_id_f32_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, float4x4, 1, dequantize_f32, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_f16_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, half4x4, 1, dequantize_f16, half, half4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q1_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q1_0, 8, dequantize_q1_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q2_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_0, 4, dequantize_q2_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q4_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_0, 2, dequantize_q4_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q4_1_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_1, 2, dequantize_q4_1, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q5_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_0, 2, dequantize_q5_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q5_1_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_1, 2, dequantize_q5_1, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q8_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q8_0, 2, dequantize_q8_0, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_mxfp4_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_mxfp4, 2, dequantize_mxfp4, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q2_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q2_K, QK_NL, dequantize_q2_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q3_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q3_K, QK_NL, dequantize_q3_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q4_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q4_K, QK_NL, dequantize_q4_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q5_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q5_K, QK_NL, dequantize_q5_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_q6_K_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_q6_K, QK_NL, dequantize_q6_K, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq2_xxs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xxs, QK_NL, dequantize_iq2_xxs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq2_xs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_xs, QK_NL, dequantize_iq2_xs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq3_xxs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_xxs, QK_NL, dequantize_iq3_xxs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq3_s_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq3_s, QK_NL, dequantize_iq3_s, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq2_s_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq2_s, QK_NL, dequantize_iq2_s, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq1_s_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_s, QK_NL, dequantize_iq1_s, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq1_m_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq1_m, QK_NL, dequantize_iq1_m, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq4_nl_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_nl, 2, dequantize_iq4_nl, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_iq4_xs_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_iq4_xs, QK_NL, dequantize_iq4_xs, float, float4x4, half, half2x4>;
+template [[host_name("kernel_mul_mm_id_tq2_0_f16")]] kernel mul_mm_id kernel_mul_mm_id<half, half4x4, simdgroup_half8x8, half, half2x4, simdgroup_half8x8, block_tq2_0, QK_NL, dequantize_tq2_0, float, float4x4, half, half2x4>;
diff --git a/ggml/src/ggml-metal/kernels/mul_mv.metal b/ggml/src/ggml-metal/kernels/mul_mv.metal
new file mode 100644
index 0000000000..d1800313ed
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/mul_mv.metal
@@ -0,0 +1,3225 @@
+#include "common.h"
+#include "dequantize.h"
+// Q1_0 dot product: dot = d * (2 * Σ(yl[i] where bit=1) - sumy)
+inline float block_q_n_dot_y(device const block_q1_0 * qb_curr, float sumy, thread float * yl, int il) {
+ device const uint8_t * qs = qb_curr->qs + il / 8;
+ const uint8_t b0 = qs[0];
+ const uint8_t b1 = qs[1];
+
+ float acc = 0.0f;
+
+ acc += select(0.0f, yl[ 0], bool(b0 & 0x01));
+ acc += select(0.0f, yl[ 1], bool(b0 & 0x02));
+ acc += select(0.0f, yl[ 2], bool(b0 & 0x04));
+ acc += select(0.0f, yl[ 3], bool(b0 & 0x08));
+ acc += select(0.0f, yl[ 4], bool(b0 & 0x10));
+ acc += select(0.0f, yl[ 5], bool(b0 & 0x20));
+ acc += select(0.0f, yl[ 6], bool(b0 & 0x40));
+ acc += select(0.0f, yl[ 7], bool(b0 & 0x80));
+
+ acc += select(0.0f, yl[ 8], bool(b1 & 0x01));
+ acc += select(0.0f, yl[ 9], bool(b1 & 0x02));
+ acc += select(0.0f, yl[10], bool(b1 & 0x04));
+ acc += select(0.0f, yl[11], bool(b1 & 0x08));
+ acc += select(0.0f, yl[12], bool(b1 & 0x10));
+ acc += select(0.0f, yl[13], bool(b1 & 0x20));
+ acc += select(0.0f, yl[14], bool(b1 & 0x40));
+ acc += select(0.0f, yl[15], bool(b1 & 0x80));
+
+ return qb_curr->d * (2.0f * acc - sumy);
+}
+
+// Q2_0 dot: d * (sum_lo(y) + 2*sum_hi(y) - sumy) via per-bit conditional adds
+inline float block_q_n_dot_y(device const block_q2_0 * qb_curr, float sumy, thread float * yl, int il) {
+ device const uint8_t * qs = qb_curr->qs + (il / 4);
+ const uint8_t b0 = qs[0];
+ const uint8_t b1 = qs[1];
+ const uint8_t b2 = qs[2];
+ const uint8_t b3 = qs[3];
+
+ // Accumulate where low bit is set (bits 0,2,4,6 of each byte)
+ float acc_lo = 0.0f;
+ acc_lo += select(0.0f, yl[ 0], bool(b0 & 0x01));
+ acc_lo += select(0.0f, yl[ 1], bool(b0 & 0x04));
+ acc_lo += select(0.0f, yl[ 2], bool(b0 & 0x10));
+ acc_lo += select(0.0f, yl[ 3], bool(b0 & 0x40));
+ acc_lo += select(0.0f, yl[ 4], bool(b1 & 0x01));
+ acc_lo += select(0.0f, yl[ 5], bool(b1 & 0x04));
+ acc_lo += select(0.0f, yl[ 6], bool(b1 & 0x10));
+ acc_lo += select(0.0f, yl[ 7], bool(b1 & 0x40));
+ acc_lo += select(0.0f, yl[ 8], bool(b2 & 0x01));
+ acc_lo += select(0.0f, yl[ 9], bool(b2 & 0x04));
+ acc_lo += select(0.0f, yl[10], bool(b2 & 0x10));
+ acc_lo += select(0.0f, yl[11], bool(b2 & 0x40));
+ acc_lo += select(0.0f, yl[12], bool(b3 & 0x01));
+ acc_lo += select(0.0f, yl[13], bool(b3 & 0x04));
+ acc_lo += select(0.0f, yl[14], bool(b3 & 0x10));
+ acc_lo += select(0.0f, yl[15], bool(b3 & 0x40));
+
+ // Accumulate where high bit is set (bits 1,3,5,7 of each byte)
+ float acc_hi = 0.0f;
+ acc_hi += select(0.0f, yl[ 0], bool(b0 & 0x02));
+ acc_hi += select(0.0f, yl[ 1], bool(b0 & 0x08));
+ acc_hi += select(0.0f, yl[ 2], bool(b0 & 0x20));
+ acc_hi += select(0.0f, yl[ 3], bool(b0 & 0x80));
+ acc_hi += select(0.0f, yl[ 4], bool(b1 & 0x02));
+ acc_hi += select(0.0f, yl[ 5], bool(b1 & 0x08));
+ acc_hi += select(0.0f, yl[ 6], bool(b1 & 0x20));
+ acc_hi += select(0.0f, yl[ 7], bool(b1 & 0x80));
+ acc_hi += select(0.0f, yl[ 8], bool(b2 & 0x02));
+ acc_hi += select(0.0f, yl[ 9], bool(b2 & 0x08));
+ acc_hi += select(0.0f, yl[10], bool(b2 & 0x20));
+ acc_hi += select(0.0f, yl[11], bool(b2 & 0x80));
+ acc_hi += select(0.0f, yl[12], bool(b3 & 0x02));
+ acc_hi += select(0.0f, yl[13], bool(b3 & 0x08));
+ acc_hi += select(0.0f, yl[14], bool(b3 & 0x20));
+ acc_hi += select(0.0f, yl[15], bool(b3 & 0x80));
+
+ return qb_curr->d * (acc_lo + 2.0f * acc_hi - sumy);
+}
+
+// function for calculate inner product between half a q4_0 block and 16 floats (yl), sumy is SUM(yl[i])
+// il indicates where the q4 quants begin (0 or QK4_0/4)
+// we assume that the yl's have been multiplied with the appropriate scale factor
+// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
+inline float block_q_n_dot_y(device const block_q4_0 * qb_curr, float sumy, thread float * yl, int il) {
+ float d = qb_curr->d;
+
+ float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
+
+ device const uint16_t * qs = ((device const uint16_t *) qb_curr + 1 + il/2);
+
+ for (int i = 0; i < 8; i += 2) {
+ acc[0] += yl[i + 0] * (qs[i / 2] & 0x000F);
+ acc[1] += yl[i + 1] * (qs[i / 2] & 0x0F00);
+ acc[2] += yl[i + 8] * (qs[i / 2] & 0x00F0);
+ acc[3] += yl[i + 9] * (qs[i / 2] & 0xF000);
+ }
+
+ return d * (sumy * -8.f + acc[0] + acc[1] + acc[2] + acc[3]);
+}
+
+// function for calculate inner product between half a q4_1 block and 16 floats (yl), sumy is SUM(yl[i])
+// il indicates where the q4 quants begin (0 or QK4_0/4)
+// we assume that the yl's have been multiplied with the appropriate scale factor
+// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
+inline float block_q_n_dot_y(device const block_q4_1 * qb_curr, float sumy, thread float * yl, int il) {
+ float d = qb_curr->d;
+ float m = qb_curr->m;
+
+ float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
+
+ device const uint16_t * qs = ((device const uint16_t *) qb_curr + 2 + il/2);
+
+ for (int i = 0; i < 8; i+=2) {
+ acc[0] += yl[i + 0] * (qs[i / 2] & 0x000F);
+ acc[1] += yl[i + 1] * (qs[i / 2] & 0x0F00);
+ acc[2] += yl[i + 8] * (qs[i / 2] & 0x00F0);
+ acc[3] += yl[i + 9] * (qs[i / 2] & 0xF000);
+ }
+
+ return d * (acc[0] + acc[1] + acc[2] + acc[3]) + sumy * m;
+}
+
+// function for calculate inner product between half a q5_0 block and 16 floats (yl), sumy is SUM(yl[i])
+// il indicates where the q5 quants begin (0 or QK5_0/4)
+// we assume that the yl's have been multiplied with the appropriate scale factor
+// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
+inline float block_q_n_dot_y(device const block_q5_0 * qb_curr, float sumy, thread float * yl, int il) {
+ float d = qb_curr->d;
+
+ float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
+
+ device const uint16_t * qs = ((device const uint16_t *)qb_curr + 3 + il/2);
+ const uint32_t qh = *((device const uint32_t *)qb_curr->qh);
+
+ for (int i = 0; i < 8; i+=2) {
+ acc[0] += yl[i + 0] * ((qs[i / 2] & 0x000F) | ((qh >> (i+0+il ) << 4 ) & 0x00010));
+ acc[1] += yl[i + 1] * ((qs[i / 2] & 0x0F00) | ((qh >> (i+1+il ) << 12) & 0x01000));
+ acc[2] += yl[i + 8] * ((qs[i / 2] & 0x00F0) | ((qh >> (i+0+il+QK5_0/2) << 8 ) & 0x00100));
+ acc[3] += yl[i + 9] * ((qs[i / 2] & 0xF000) | ((qh >> (i+1+il+QK5_0/2) << 16) & 0x10000));
+ }
+
+ return d * (sumy * -16.f + acc[0] + acc[1] + acc[2] + acc[3]);
+}
+
+// function for calculate inner product between half a q5_1 block and 16 floats (yl), sumy is SUM(yl[i])
+// il indicates where the q5 quants begin (0 or QK5_1/4)
+// we assume that the yl's have been multiplied with the appropriate scale factor
+// that corresponds to the missing bit shifts (1, 1/16, 1/256, 1/4096)
+inline float block_q_n_dot_y(device const block_q5_1 * qb_curr, float sumy, thread float * yl, int il) {
+ float d = qb_curr->d;
+ float m = qb_curr->m;
+
+ float acc[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
+
+ device const uint16_t * qs = ((device const uint16_t *)qb_curr + 4 + il/2);
+ const uint32_t qh = *((device const uint32_t *)qb_curr->qh);
+
+ for (int i = 0; i < 8; i+=2) {
+ acc[0] += yl[i + 0] * ((qs[i / 2] & 0x000F) | ((qh >> (i+0+il ) << 4 ) & 0x00010));
+ acc[1] += yl[i + 1] * ((qs[i / 2] & 0x0F00) | ((qh >> (i+1+il ) << 12) & 0x01000));
+ acc[2] += yl[i + 8] * ((qs[i / 2] & 0x00F0) | ((qh >> (i+0+il+QK5_0/2) << 8 ) & 0x00100));
+ acc[3] += yl[i + 9] * ((qs[i / 2] & 0xF000) | ((qh >> (i+1+il+QK5_0/2) << 16) & 0x10000));
+ }
+
+ return d * (acc[0] + acc[1] + acc[2] + acc[3]) + sumy * m;
+}
+
+template<short NR0>
+static inline void helper_mv_reduce_and_write(
+ device float * dst_f32,
+ float sumf[NR0],
+ const int r0,
+ const int ne01,
+ ushort tiisg,
+ ushort sgitg,
+ threadgroup char * shmem) {
+ constexpr short NW = N_SIMDWIDTH;
+
+ threadgroup float * shmem_f32[NR0];
+
+ for (short row = 0; row < NR0; ++row) {
+ shmem_f32[row] = (threadgroup float *) shmem + NW*row;
+
+ if (sgitg == 0) {
+ shmem_f32[row][tiisg] = 0.0f;
+ }
+
+ sumf[row] = simd_sum(sumf[row]);
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (short row = 0; row < NR0; ++row) {
+ if (tiisg == 0) {
+ shmem_f32[row][sgitg] = sumf[row];
+ }
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (short row = 0; row < NR0 && r0 + row < ne01; ++row) {
+ float tot = simd_sum(shmem_f32[row][tiisg]);
+
+ if (tiisg == 0 && sgitg == 0) {
+ dst_f32[r0 + row] = tot;
+ }
+ }
+}
+
+constant short FC_mul_mv_nsg [[function_constant(FC_MUL_MV + 0)]];
+constant short FC_mul_mv_nxpsg [[function_constant(FC_MUL_MV + 1)]];
+constant short FC_mul_mv_ne12 [[function_constant(FC_MUL_MV + 2)]];
+constant short FC_mul_mv_r2 [[function_constant(FC_MUL_MV + 3)]];
+constant short FC_mul_mv_r3 [[function_constant(FC_MUL_MV + 4)]];
+
+template<typename block_q_type, short NR0, typename args_t>
+void mul_vec_q_n_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ constexpr short NW = N_SIMDWIDTH;
+ constexpr short NQ = 16;
+
+ const int nb = args.ne00/QK4_0;
+
+ const int r0 = (tgpig.x*NSG + sgitg)*NR0;
+ //const int r0 = tgpig.x*NR0;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ //device const block_q_type * x = (device const block_q_type *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ // pointers to src0 rows
+ device const block_q_type * ax[NR0];
+ FOR_UNROLL (int row = 0; row < NR0; ++row) {
+ const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+
+ ax[row] = (device const block_q_type *) ((device char *) src0 + offset0);
+ }
+
+ float sumf[NR0] = {0.f};
+
+ const short ix = (tiisg/(NW/NQ));
+ const short il = (tiisg%(NW/NQ))*8;
+
+ //const int ib0 = sgitg*NQ + ix;
+ const int ib0 = ix;
+
+ float yl[16]; // src1 vector cache
+
+ //device const float * yb = y + ix*QK4_0 + il;
+ device const float * yb = y + ib0*QK4_0 + il;
+
+ // each thread in a SIMD group deals with half a block.
+ //for (int ib = ib0; ib < nb; ib += NSG*NQ) {
+ for (int ib = ib0; ib < nb; ib += NQ) {
+ float sumy[2] = { 0.f, 0.f };
+
+ FOR_UNROLL (short i = 0; i < 8; i += 2) {
+ sumy[0] += yb[i + 0] + yb[i + 1];
+ yl[i + 0] = yb[i + 0];
+ yl[i + 1] = yb[i + 1]/256.f;
+
+ sumy[1] += yb[i + 16] + yb[i + 17];
+ yl[i + 8] = yb[i + 16]/16.f;
+ yl[i + 9] = yb[i + 17]/4096.f;
+ }
+
+ FOR_UNROLL (short row = 0; row < NR0; row++) {
+ sumf[row] += block_q_n_dot_y(ax[row] + ib, sumy[0] + sumy[1], yl, il);
+ }
+
+ yb += QK4_0 * 16;
+ //yb += NSG*NQ*QK4_0;
+ }
+
+ device float * dst_f32 = (device float *) dst + im*args.ne0*args.ne1 + r1*args.ne0;
+
+ //helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
+
+ for (int row = 0; row < NR0; ++row) {
+ const float tot = simd_sum(sumf[row]);
+
+ if (tiisg == 0 && r0 + row < args.ne01) {
+ dst_f32[r0 + row] = tot;
+ }
+ }
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_q1_0_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK1_0;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset1 = r1*args.nb11 + (i12)*args.nb12 + (i13)*args.nb13;
+
+ device const float * y = (device const float *) (src1 + offset1);
+
+ device const block_q1_0 * ax[nr0];
+ for (int row = 0; row < nr0; ++row) {
+ const uint64_t offset0 = (first_row + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ ax[row] = (device const block_q1_0 *) ((device char *) src0 + offset0);
+ }
+
+ float yl[16];
+ float sumf[nr0] = {0.f};
+
+ const short ix = (tiisg/8);
+ const short il = (tiisg%8)*16;
+
+ device const float * yb = y + ix*QK1_0 + il;
+
+ for (int ib = ix; ib < nb; ib += N_SIMDWIDTH/8) {
+ float sumy = 0.f;
+
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ yl[i] = yb[i];
+ sumy += yb[i];
+ }
+
+ FOR_UNROLL (short row = 0; row < nr0; row++) {
+ sumf[row] += block_q_n_dot_y(ax[row] + ib, sumy, yl, il);
+ }
+
+ yb += QK1_0 * (N_SIMDWIDTH/8);
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0; ++row) {
+ const float tot = simd_sum(sumf[row]);
+
+ if (tiisg == 0 && first_row + row < args.ne01) {
+ dst_f32[first_row + row] = tot;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_q1_0_f32")]]
+kernel void kernel_mul_mv_q1_0_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_q1_0_f32_impl<N_R0_Q1_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_q2_0_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK2_0;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset1 = r1*args.nb11 + (i12)*args.nb12 + (i13)*args.nb13;
+
+ device const float * y = (device const float *) (src1 + offset1);
+
+ device const block_q2_0 * ax[nr0];
+ for (int row = 0; row < nr0; ++row) {
+ const uint64_t offset0 = (first_row + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ ax[row] = (device const block_q2_0 *) ((device char *) src0 + offset0);
+ }
+
+ float yl[16];
+ float sumf[nr0] = {0.f};
+
+ // group 64: 4 sub-blocks of 16 weights per Q2_0 block
+ const short ix = (tiisg/4);
+ const short il = (tiisg%4)*16;
+
+ device const float * yb = y + ix*QK2_0 + il;
+
+ for (int ib = ix; ib < nb; ib += N_SIMDWIDTH/4) {
+ float sumy = 0.f;
+
+ FOR_UNROLL (short i = 0; i < 16; i++) {
+ yl[i] = yb[i];
+ sumy += yb[i];
+ }
+
+ FOR_UNROLL (short row = 0; row < nr0; row++) {
+ sumf[row] += block_q_n_dot_y(ax[row] + ib, sumy, yl, il);
+ }
+
+ yb += QK2_0 * (N_SIMDWIDTH/4);
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0; ++row) {
+ const float tot = simd_sum(sumf[row]);
+
+ if (tiisg == 0 && first_row + row < args.ne01) {
+ dst_f32[first_row + row] = tot;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_q2_0_f32")]]
+kernel void kernel_mul_mv_q2_0_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_q2_0_f32_impl<N_R0_Q2_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+kernel void kernel_mul_mv_q4_0_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ mul_vec_q_n_f32_impl<block_q4_0, N_R0_Q4_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+kernel void kernel_mul_mv_q4_1_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ mul_vec_q_n_f32_impl<block_q4_1, N_R0_Q4_1, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+kernel void kernel_mul_mv_q5_0_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ mul_vec_q_n_f32_impl<block_q5_0, N_R0_Q5_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+kernel void kernel_mul_mv_q5_1_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ mul_vec_q_n_f32_impl<block_q5_1, N_R0_Q5_1, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<short NR0, typename args_t>
+void kernel_mul_mv_q8_0_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ constexpr short NW = N_SIMDWIDTH;
+ constexpr short NQ = 8;
+
+ const int nb = args.ne00/QK8_0;
+
+ const int r0 = tgpig.x*NR0;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ //device const block_q8_0 * x = (device const block_q8_0 *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ // pointers to src0 rows
+ device const block_q8_0 * ax[NR0];
+ FOR_UNROLL (short row = 0; row < NR0; ++row) {
+ const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+
+ ax[row] = (device const block_q8_0 *) ((device char *) src0 + offset0);
+ }
+
+ float sumf[NR0] = { 0.f };
+
+ const short ix = tiisg/(NW/NQ);
+ const short il = tiisg%(NW/NQ);
+
+ const int ib0 = sgitg*NQ + ix;
+
+ float yl[NQ];
+
+ device const float * yb = y + ib0*QK8_0 + il*NQ;
+
+ // each thread in a SIMD group deals with NQ quants at a time
+ for (int ib = ib0; ib < nb; ib += NSG*NQ) {
+ for (short i = 0; i < NQ; ++i) {
+ yl[i] = yb[i];
+ }
+
+ for (short row = 0; row < NR0; row++) {
+ device const int8_t * qs = ax[row][ib].qs + il*NQ;
+
+ float sumq = 0.f;
+ FOR_UNROLL (short i = 0; i < NQ; ++i) {
+ sumq += qs[i] * yl[i];
+ }
+
+ sumf[row] += sumq*ax[row][ib].d;
+ }
+
+ yb += NSG*NQ*QK8_0;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
+}
+
+[[host_name("kernel_mul_mv_q8_0_f32")]]
+kernel void kernel_mul_mv_q8_0_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_q8_0_f32_impl<N_R0_Q8_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+// mat-vec kernel processing in chunks of float4
+// chpb - chunks per quantization block
+template<short r1ptg, typename q_t, short chpb, void (*deq_t4)(device const q_t *, short, thread float4 &) >
+void kernel_mul_mv_ext_q4_f32_impl(
+ constant ggml_metal_kargs_mul_mv_ext & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ const short NSG = FC_mul_mv_nsg;
+ const short nxpsg = FC_mul_mv_nxpsg;
+
+ const short chpt = 4; // chunks per thread
+
+ //const short nxpsg = (32);
+ const short nypsg = (32/nxpsg);
+
+ const short tx = tiisg%nxpsg;
+ const short ty = tiisg/nxpsg;
+
+ const int i01 = tgpig.x*(nypsg*NSG) + nypsg*sgitg + ty;
+ const int i11 = tgpig.y*r1ptg;
+ const int i1m = tgpig.z;
+
+ const int i12 = i1m%FC_mul_mv_ne12;
+ const int i13 = i1m/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = i01*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = i11*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const q_t * xq = (i01 < args.ne01) ? (device const q_t *) (src0 + offset0) + tx/chpb : (device const q_t *) src0;
+
+ device const float4 * y4[r1ptg];
+
+ for (int ir1 = 0; ir1 < r1ptg; ++ir1) {
+ y4[ir1] = (i11 + ir1 < args.ne11) ? (device const float4 *) (src1 + offset1 + ir1*args.nb11) + tx : (device const float4 *) src1;
+ }
+
+ float sumf[r1ptg] = { [ 0 ... r1ptg - 1 ] = 0.0f };
+
+ short cch = tx%chpb; // current chunk index
+
+ for (int ich = tx; 4*ich < args.ne00; ich += chpt*nxpsg) {
+ float4 lx[chpt];
+
+#pragma unroll(chpt)
+ for (short ch = 0; ch < chpt; ++ch) {
+ deq_t4(xq, cch, lx[ch]);
+
+ cch += nxpsg;
+ if (cch >= chpb) {
+ xq += cch/chpb;
+ cch %= chpb;
+ }
+ }
+
+#pragma unroll(chpt)
+ for (short ch = 0; ch < chpt; ++ch) {
+#pragma unroll(r1ptg)
+ for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
+ sumf[ir1] += dot(lx[ch], y4[ir1][ch*nxpsg]);
+ }
+ }
+
+#pragma unroll(r1ptg)
+ for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
+ y4[ir1] += chpt*nxpsg;
+ }
+ }
+
+ // reduce only the threads in each row
+ for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
+ if (nxpsg >= 32) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 16);
+ }
+ if (nxpsg >= 16) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 8);
+ }
+ if (nxpsg >= 8) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 4);
+ }
+ if (nxpsg >= 4) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 2);
+ }
+ if (nxpsg >= 2) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 1);
+ }
+
+ //sumf[ir1] = simd_sum(sumf[ir1]);
+ }
+
+ if (tx == 0) {
+ for (short ir1 = 0; ir1 < r1ptg && i11 + ir1 < args.ne11; ++ir1) {
+ device float * dst_f32 = (device float *) dst + (uint64_t)i1m*args.ne0*args.ne1 + (uint64_t)(i11 + ir1)*args.ne0;
+
+ if (i01 < args.ne01) {
+ dst_f32[i01] = sumf[ir1];
+ }
+ }
+ }
+}
+
+// mat-vec kernel processing in chunks of float4x4
+template<short r1ptg, typename q_t, short chpb, void (*deq_t4x4)(device const q_t *, short, thread float4x4 &) >
+void kernel_mul_mv_ext_q4x4_f32_impl(
+ constant ggml_metal_kargs_mul_mv_ext & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ const short NSG = FC_mul_mv_nsg;
+ const short nxpsg = FC_mul_mv_nxpsg;
+
+ const short chpt = 1;
+
+ //const short nxpsg = (32);
+ const short nypsg = (32/nxpsg);
+
+ const short tx = tiisg%nxpsg;
+ const short ty = tiisg/nxpsg;
+
+ const int i01 = tgpig.x*(nypsg*NSG) + nypsg*sgitg + ty;
+ const int i11 = tgpig.y*r1ptg;
+ const int i1m = tgpig.z;
+
+ const int i12 = i1m%FC_mul_mv_ne12;
+ const int i13 = i1m/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = i01*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = i11*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const q_t * xq = (i01 < args.ne01) ? (device const q_t *) (src0 + offset0) + tx/chpb : (device const q_t *) src0;
+
+ device const float4x4 * y4x4[r1ptg];
+
+ for (int ir1 = 0; ir1 < r1ptg; ++ir1) {
+ y4x4[ir1] = (i11 + ir1 < args.ne11) ? (device const float4x4 *) (src1 + offset1 + ir1*args.nb11) + tx : (device const float4x4 *) src1;
+ }
+
+ float sumf[r1ptg] = { [ 0 ... r1ptg - 1 ] = 0.0f };
+
+ short cch = tx%chpb;
+
+ for (int ich = tx; 16*ich < args.ne00; ich += chpt*nxpsg) {
+ float4x4 lx[chpt];
+
+#pragma unroll(chpt)
+ for (short ch = 0; ch < chpt; ++ch) {
+ deq_t4x4(xq, cch, lx[ch]);
+
+ cch += nxpsg;
+ if (cch >= chpb) {
+ xq += cch/chpb;
+ cch %= chpb;
+ }
+ }
+
+#pragma unroll(chpt)
+ for (short ch = 0; ch < chpt; ++ch) {
+#pragma unroll(r1ptg)
+ for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
+ sumf[ir1] +=
+ dot(lx[ch][0], y4x4[ir1][ch*nxpsg][0]) +
+ dot(lx[ch][1], y4x4[ir1][ch*nxpsg][1]) +
+ dot(lx[ch][2], y4x4[ir1][ch*nxpsg][2]) +
+ dot(lx[ch][3], y4x4[ir1][ch*nxpsg][3]);
+
+ }
+ }
+
+#pragma unroll(r1ptg)
+ for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
+ y4x4[ir1] += chpt*nxpsg;
+ }
+ }
+
+ for (short ir1 = 0; ir1 < r1ptg; ++ir1) {
+ if (nxpsg >= 32) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 16);
+ }
+ if (nxpsg >= 16) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 8);
+ }
+ if (nxpsg >= 8) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 4);
+ }
+ if (nxpsg >= 4) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 2);
+ }
+ if (nxpsg >= 2) {
+ sumf[ir1] += simd_shuffle_down(sumf[ir1], 1);
+ }
+
+ //sumf[ir1] = simd_sum(sumf[ir1]);
+ }
+
+ if (tx == 0) {
+ for (short ir1 = 0; ir1 < r1ptg && i11 + ir1 < args.ne11; ++ir1) {
+ device float * dst_f32 = (device float *) dst + (uint64_t)i1m*args.ne0*args.ne1 + (uint64_t)(i11 + ir1)*args.ne0;
+
+ if (i01 < args.ne01) {
+ dst_f32[i01] = sumf[ir1];
+ }
+ }
+ }
+}
+
+// dispatchers needed for compile-time nxpsg
+// epb - elements per quantization block
+template<short r1ptg, typename q_t, short epb, void (*deq_t4)(device const q_t *, short, thread float4 &)>
+kernel void kernel_mul_mv_ext_q4_f32_disp(
+ constant ggml_metal_kargs_mul_mv_ext & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_ext_q4_f32_impl<r1ptg, q_t, epb/4, deq_t4>(args, src0, src1, dst, tgpig, tiisg, sgitg);
+}
+
+template<short r1ptg, typename q_t, short epb, void (*deq_t4x4)(device const q_t *, short, thread float4x4 &)>
+kernel void kernel_mul_mv_ext_q4x4_f32_disp(
+ constant ggml_metal_kargs_mul_mv_ext & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_ext_q4x4_f32_impl<r1ptg, q_t, epb/16, deq_t4x4>(args, src0, src1, dst, tgpig, tiisg, sgitg);
+}
+
+typedef decltype(kernel_mul_mv_ext_q4_f32_disp <2, block_q8_0, 32, dequantize_q8_0_t4>) mul_mv_ext_q4_f32_t;
+typedef decltype(kernel_mul_mv_ext_q4x4_f32_disp<2, block_q4_K, 256, dequantize_q4_K>) mul_mv_ext_q4x4_f32_t;
+
+template [[host_name("kernel_mul_mv_ext_f32_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, float4, 4, dequantize_f32_t4>;
+template [[host_name("kernel_mul_mv_ext_f32_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, float4, 4, dequantize_f32_t4>;
+template [[host_name("kernel_mul_mv_ext_f32_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, float4, 4, dequantize_f32_t4>;
+template [[host_name("kernel_mul_mv_ext_f32_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, float4, 4, dequantize_f32_t4>;
+
+template [[host_name("kernel_mul_mv_ext_f16_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, half4, 4, dequantize_f16_t4>;
+template [[host_name("kernel_mul_mv_ext_f16_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, half4, 4, dequantize_f16_t4>;
+template [[host_name("kernel_mul_mv_ext_f16_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, half4, 4, dequantize_f16_t4>;
+template [[host_name("kernel_mul_mv_ext_f16_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, half4, 4, dequantize_f16_t4>;
+
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, bfloat4, 4, dequantize_bf16_t4>;
+template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, bfloat4, 4, dequantize_bf16_t4>;
+template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, bfloat4, 4, dequantize_bf16_t4>;
+template [[host_name("kernel_mul_mv_ext_bf16_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, bfloat4, 4, dequantize_bf16_t4>;
+#endif
+
+template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q1_0, 128, dequantize_q1_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q1_0, 128, dequantize_q1_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q1_0, 128, dequantize_q1_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q1_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q1_0, 128, dequantize_q1_0_t4>;
+
+template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q2_0, 64, dequantize_q2_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q2_0, 64, dequantize_q2_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q2_0, 64, dequantize_q2_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q2_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q2_0, 64, dequantize_q2_0_t4>;
+
+template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q4_0, 32, dequantize_q4_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q4_0, 32, dequantize_q4_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q4_0, 32, dequantize_q4_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q4_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q4_0, 32, dequantize_q4_0_t4>;
+
+template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q4_1, 32, dequantize_q4_1_t4>;
+template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q4_1, 32, dequantize_q4_1_t4>;
+template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q4_1, 32, dequantize_q4_1_t4>;
+template [[host_name("kernel_mul_mv_ext_q4_1_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q4_1, 32, dequantize_q4_1_t4>;
+
+template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q5_0, 32, dequantize_q5_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q5_0, 32, dequantize_q5_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q5_0, 32, dequantize_q5_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q5_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q5_0, 32, dequantize_q5_0_t4>;
+
+template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q5_1, 32, dequantize_q5_1_t4>;
+template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q5_1, 32, dequantize_q5_1_t4>;
+template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q5_1, 32, dequantize_q5_1_t4>;
+template [[host_name("kernel_mul_mv_ext_q5_1_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q5_1, 32, dequantize_q5_1_t4>;
+
+template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_q8_0, 32, dequantize_q8_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_q8_0, 32, dequantize_q8_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_q8_0, 32, dequantize_q8_0_t4>;
+template [[host_name("kernel_mul_mv_ext_q8_0_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_q8_0, 32, dequantize_q8_0_t4>;
+
+template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_mxfp4, 32, dequantize_mxfp4_t4>;
+template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_mxfp4, 32, dequantize_mxfp4_t4>;
+template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_mxfp4, 32, dequantize_mxfp4_t4>;
+template [[host_name("kernel_mul_mv_ext_mxfp4_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_mxfp4, 32, dequantize_mxfp4_t4>;
+
+template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_2")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<2, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
+template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_3")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<3, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
+template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_4")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<4, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
+template [[host_name("kernel_mul_mv_ext_iq4_nl_f32_r1_5")]] kernel mul_mv_ext_q4_f32_t kernel_mul_mv_ext_q4_f32_disp<5, block_iq4_nl, 32, dequantize_iq4_nl_t4>;
+
+template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q4_K, 256, dequantize_q4_K>;
+template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q4_K, 256, dequantize_q4_K>;
+template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q4_K, 256, dequantize_q4_K>;
+template [[host_name("kernel_mul_mv_ext_q4_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q4_K, 256, dequantize_q4_K>;
+
+template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q5_K, 256, dequantize_q5_K>;
+template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q5_K, 256, dequantize_q5_K>;
+template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q5_K, 256, dequantize_q5_K>;
+template [[host_name("kernel_mul_mv_ext_q5_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q5_K, 256, dequantize_q5_K>;
+
+template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q6_K, 256, dequantize_q6_K>;
+template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q6_K, 256, dequantize_q6_K>;
+template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q6_K, 256, dequantize_q6_K>;
+template [[host_name("kernel_mul_mv_ext_q6_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q6_K, 256, dequantize_q6_K>;
+
+template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q2_K, 256, dequantize_q2_K>;
+template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q2_K, 256, dequantize_q2_K>;
+template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q2_K, 256, dequantize_q2_K>;
+template [[host_name("kernel_mul_mv_ext_q2_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q2_K, 256, dequantize_q2_K>;
+
+template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_2")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<2, block_q3_K, 256, dequantize_q3_K>;
+template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_3")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<3, block_q3_K, 256, dequantize_q3_K>;
+template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_4")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<4, block_q3_K, 256, dequantize_q3_K>;
+template [[host_name("kernel_mul_mv_ext_q3_K_f32_r1_5")]] kernel mul_mv_ext_q4x4_f32_t kernel_mul_mv_ext_q4x4_f32_disp<5, block_q3_K, 256, dequantize_q3_K>;
+
+template<typename T0, typename T1, short NR0, typename args_t>
+void kernel_mul_mv_t_t_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ constexpr short NW = N_SIMDWIDTH;
+ constexpr short NB = 32;
+ constexpr short NF = 8;
+
+ const int nb = args.ne00/NB;
+
+ const int r0 = tgpig.x*NR0;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ //device const T0 * x = (device const T0 *) (src0 + offset0);
+ device const T1 * y = (device const T1 *) (src1 + offset1);
+
+ // pointers to src0 rows
+ device const T0 * ax [NR0];
+ FOR_UNROLL (short row = 0; row < NR0; ++row) {
+ const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+
+ ax[row] = (device const T0 *) ((device char *) src0 + offset0);
+ }
+
+ float sumf[NR0] = { 0.f };
+
+ const short ix = tiisg/(NW/NF);
+ const short il = tiisg%(NW/NF);
+
+ const int ib0 = sgitg*NF + ix;
+
+ T1 yl[NF];
+
+ device const T1 * yb = y + (ib0*NB + il*NF);
+
+ for (int ib = ib0; ib < nb; ib += NSG*NF) {
+ for (short i = 0; i < NF; ++i) {
+ yl[i] = yb[i];
+ }
+
+ for (short row = 0; row < NR0; row++) {
+ device const T0 * xb = ax[row] + (ib*NB + il*NF);
+
+ float sumq = 0.f;
+ FOR_UNROLL (short i = 0; i < NF; ++i) {
+ sumq += xb[i] * yl[i];
+ }
+
+ sumf[row] += sumq;
+ }
+
+ yb += NSG*NF*NW;
+ }
+
+ for (int i = nb*NB + sgitg*NW + tiisg; i < args.ne00; i += NW*NSG) {
+ for (short row = 0; row < NR0; row++) {
+ sumf[row] += ax[row][i] * y[i];
+ }
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
+}
+
+template<typename T0, typename T1, typename args_t>
+void kernel_mul_mv_t_t_disp(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ switch (args.nr0) {
+ //case 1: kernel_mul_mv_t_t_impl<T0, T1, 1, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ case 2: kernel_mul_mv_t_t_impl<T0, T1, 2, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ //case 3: kernel_mul_mv_t_t_impl<T0, T1, 3, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ //case 4: kernel_mul_mv_t_t_impl<T0, T1, 4, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ }
+}
+
+template<typename T0, typename T1>
+kernel void kernel_mul_mv_t_t(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_t_t_disp<T0, T1, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+typedef decltype(kernel_mul_mv_t_t<half, half>) mul_mv_t_t;
+
+template [[host_name("kernel_mul_mv_f32_f32")]] kernel mul_mv_t_t kernel_mul_mv_t_t<float, float>;
+template [[host_name("kernel_mul_mv_f16_f32")]] kernel mul_mv_t_t kernel_mul_mv_t_t<half, float>;
+template [[host_name("kernel_mul_mv_f16_f16")]] kernel mul_mv_t_t kernel_mul_mv_t_t<half, half>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mv_bf16_f32")]] kernel mul_mv_t_t kernel_mul_mv_t_t<bfloat, float>;
+template [[host_name("kernel_mul_mv_bf16_bf16")]] kernel mul_mv_t_t kernel_mul_mv_t_t<bfloat, bfloat>;
+#endif
+
+template<typename T0, typename T04, typename T1, typename T14, short NR0, typename args_t>
+void kernel_mul_mv_t_t_4_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ constexpr short NW = N_SIMDWIDTH;
+ constexpr short NB = 32;
+ constexpr short NF = 16;
+ constexpr short NF4 = NF/4;
+
+ const int nb = args.ne00/NB;
+
+ const int r0 = tgpig.x*NR0;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ //const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const T1 * y = (device const T1 *) (src1 + offset1);
+ device const T14 * y4 = (device const T14 *) (src1 + offset1);
+
+ // pointers to src0 rows
+ device const T0 * ax [NR0];
+ device const T04 * ax4[NR0];
+ FOR_UNROLL (short row = 0; row < NR0; ++row) {
+ const uint64_t offset0 = (r0 + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+
+ ax [row] = (device const T0 *) ((device char *) src0 + offset0);
+ ax4[row] = (device const T04 *) ((device char *) src0 + offset0);
+ }
+
+ float sumf[NR0] = { 0.f };
+
+ const short ix = tiisg/(NW/NF);
+ const short il = tiisg%(NW/NF);
+
+ const int ib0 = sgitg*NF + ix;
+
+ T14 yl4[NF4];
+
+ device const T14 * yb4 = y4 + (ib0*NB + il*NF)/4;
+
+ for (int ib = ib0; ib < nb; ib += NSG*NF) {
+ for (short i = 0; i < NF4; ++i) {
+ yl4[i] = yb4[i];
+ }
+
+ for (short row = 0; row < NR0; row++) {
+ device const T04 * xb4 = ax4[row] + (ib*NB + il*NF)/4;
+
+ float sumq = 0.f;
+ FOR_UNROLL (short i = 0; i < NF4; ++i) {
+ sumq += dot(float4(xb4[i]), float4(yl4[i]));
+ }
+
+ sumf[row] += sumq;
+ }
+
+ yb4 += NSG*NF*NW/4;
+ }
+
+ for (int i = nb*NB + sgitg*NW + tiisg; i < args.ne00; i += NW*NSG) {
+ for (short row = 0; row < NR0; row++) {
+ sumf[row] += ax[row][i] * y[i];
+ }
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ helper_mv_reduce_and_write<NR0>(dst_f32, sumf, r0, args.ne01, tiisg, sgitg, shmem);
+}
+
+template<typename T0, typename T04, typename T1, typename T14, typename args_t>
+void kernel_mul_mv_t_t_4_disp(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ switch (args.nr0) {
+ //case 1: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 1, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ case 2: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 2, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ //case 3: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 3, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ //case 4: kernel_mul_mv_t_t_4_impl<T0, T04, T1, T14, 4, args_t>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg); break;
+ };
+}
+
+template<typename T0, typename T04, typename T1, typename T14>
+kernel void kernel_mul_mv_t_t_4(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_t_t_4_disp<T0, T04, T1, T14, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+typedef decltype(kernel_mul_mv_t_t_4<half, half4, half, half4>) mul_mv_t_t_4;
+
+template [[host_name("kernel_mul_mv_f32_f32_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<float, float4, float, float4>;
+template [[host_name("kernel_mul_mv_f16_f32_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<half, half4, float, float4>;
+template [[host_name("kernel_mul_mv_f16_f16_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<half, half4, half, half4>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mv_bf16_f32_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<bfloat, bfloat4, float, float4>;
+template [[host_name("kernel_mul_mv_bf16_bf16_4")]] kernel mul_mv_t_t_4 kernel_mul_mv_t_t_4<bfloat, bfloat4, bfloat, bfloat4>;
+#endif
+
+template<typename T0, typename T1, typename args_t>
+void kernel_mul_mv_t_t_short_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig,
+ ushort tiisg) {
+ const int r0 = tgpig.x*32 + tiisg;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ if (r0 >= args.ne01) {
+ return;
+ }
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = r0*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+
+ device const T0 * x = (device const T0 *) (src0 + offset0);
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1;
+
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const T1 * y = (device const T1 *) (src1 + offset1);
+
+ float res = 0.0f;
+
+ for (int i = 0; i < args.ne00; ++i) {
+ res += (float) x[i] * (float) y[i];
+ }
+
+ dst_f32[(uint64_t)r1*args.ne0 + r0] = res;
+}
+
+template<typename T0, typename T1>
+kernel void kernel_mul_mv_t_t_short(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]]) {
+ kernel_mul_mv_t_t_short_impl<T0, T1, constant ggml_metal_kargs_mul_mv &>(
+ args,
+ src0,
+ src1,
+ dst,
+ tgpig,
+ tiisg);
+}
+
+typedef decltype(kernel_mul_mv_t_t_short<half, half>) mul_mv_t_t_short_t;
+
+template [[host_name("kernel_mul_mv_f32_f32_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<float, float>;
+template [[host_name("kernel_mul_mv_f16_f32_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<half, float>;
+template [[host_name("kernel_mul_mv_f16_f16_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<half, half>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mv_bf16_f32_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<bfloat, float>;
+template [[host_name("kernel_mul_mv_bf16_bf16_short")]] kernel mul_mv_t_t_short_t kernel_mul_mv_t_t_short<bfloat, bfloat>;
+#endif
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_q2_K_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_q2_K * x = (device const block_q2_K *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const short ix = tiisg/8; // 0...3
+ const short it = tiisg%8; // 0...7
+ const short iq = it/4; // 0 or 1
+ const short ir = it%4; // 0...3
+ const short is = (8*ir)/16;// 0 or 1
+
+ device const float * y4 = y + ix * QK_K + 128 * iq + 8 * ir;
+
+ for (int ib = ix; ib < nb; ib += 4) {
+ float4 sumy = {0.f, 0.f, 0.f, 0.f};
+ for (short i = 0; i < 8; ++i) {
+ yl[i+ 0] = y4[i+ 0]; sumy[0] += yl[i+ 0];
+ yl[i+ 8] = y4[i+32]; sumy[1] += yl[i+ 8];
+ yl[i+16] = y4[i+64]; sumy[2] += yl[i+16];
+ yl[i+24] = y4[i+96]; sumy[3] += yl[i+24];
+ }
+
+ device const uint8_t * sc = (device const uint8_t *)x[ib].scales + 8*iq + is;
+ device const uint16_t * qs = (device const uint16_t *)x[ib].qs + 16 * iq + 4 * ir;
+ device const half * dh = &x[ib].d;
+
+ for (short row = 0; row < nr0; row++) {
+ float4 acc1 = {0.f, 0.f, 0.f, 0.f};
+ float4 acc2 = {0.f, 0.f, 0.f, 0.f};
+ for (int i = 0; i < 8; i += 2) {
+ acc1[0] += yl[i+ 0] * (qs[i/2] & 0x0003);
+ acc2[0] += yl[i+ 1] * (qs[i/2] & 0x0300);
+ acc1[1] += yl[i+ 8] * (qs[i/2] & 0x000c);
+ acc2[1] += yl[i+ 9] * (qs[i/2] & 0x0c00);
+ acc1[2] += yl[i+16] * (qs[i/2] & 0x0030);
+ acc2[2] += yl[i+17] * (qs[i/2] & 0x3000);
+ acc1[3] += yl[i+24] * (qs[i/2] & 0x00c0);
+ acc2[3] += yl[i+25] * (qs[i/2] & 0xc000);
+ }
+ float dall = dh[0];
+ float dmin = dh[1] * 1.f/16.f;
+ sumf[row] += dall * ((acc1[0] + 1.f/256.f * acc2[0]) * (sc[0] & 0xF) * 1.f/ 1.f +
+ (acc1[1] + 1.f/256.f * acc2[1]) * (sc[2] & 0xF) * 1.f/ 4.f +
+ (acc1[2] + 1.f/256.f * acc2[2]) * (sc[4] & 0xF) * 1.f/16.f +
+ (acc1[3] + 1.f/256.f * acc2[3]) * (sc[6] & 0xF) * 1.f/64.f) -
+ dmin * (sumy[0] * (sc[0] & 0xF0) + sumy[1] * (sc[2] & 0xF0) + sumy[2] * (sc[4] & 0xF0) + sumy[3] * (sc[6] & 0xF0));
+
+ qs += args.nb01/2;
+ sc += args.nb01;
+ dh += args.nb01/2;
+ }
+
+ y4 += 4 * QK_K;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_q2_K_f32")]]
+kernel void kernel_mul_mv_q2_K_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_q2_K_f32_impl<N_R0_Q2_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_q3_K_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_q3_K * x = (device const block_q3_K *) (src0 + offset0);
+ device const float * yy = (device const float *) (src1 + offset1);
+
+ float yl[32];
+
+ //const uint16_t kmask1 = 0x3030;
+ //const uint16_t kmask2 = 0x0f0f;
+
+ const short tid = tiisg/4;
+ const short ix = tiisg%4;
+ const short ip = tid/4; // 0 or 1
+ const short il = 2*((tid%4)/2); // 0 or 2
+ const short ir = tid%2;
+ const short l0 = 8*ir;
+
+ // One would think that the Metal compiler would figure out that ip and il can only have
+ // 4 possible states, and optimize accordingly. Well, no. It needs help, and we do it
+ // with these two tales.
+ //
+ // Possible masks for the high bit
+ const ushort4 mm[4] = {{0x0001, 0x0100, 0x0002, 0x0200}, // ip = 0, il = 0
+ {0x0004, 0x0400, 0x0008, 0x0800}, // ip = 0, il = 2
+ {0x0010, 0x1000, 0x0020, 0x2000}, // ip = 1, il = 0
+ {0x0040, 0x4000, 0x0080, 0x8000}}; // ip = 1, il = 2
+
+ // Possible masks for the low 2 bits
+ const int4 qm[2] = {{0x0003, 0x0300, 0x000c, 0x0c00}, {0x0030, 0x3000, 0x00c0, 0xc000}};
+
+ const ushort4 hm = mm[2*ip + il/2];
+
+ const short shift = 2*il;
+
+ const float v1 = il == 0 ? 4.f : 64.f;
+ const float v2 = 4.f * v1;
+
+ const uint16_t s_shift1 = 4*ip;
+ const uint16_t s_shift2 = s_shift1 + il;
+
+ const short q_offset = 32*ip + l0;
+ const short y_offset = 128*ip + 32*il + l0;
+
+ device const float * y1 = yy + ix*QK_K + y_offset;
+
+ uint32_t scales32, aux32;
+ thread uint16_t * scales16 = (thread uint16_t *)&scales32;
+ thread const int8_t * scales = (thread const int8_t *)&scales32;
+
+ float sumf1[nr0] = {0.f};
+ float sumf2[nr0] = {0.f};
+
+ for (int i = ix; i < nb; i += 4) {
+ for (short l = 0; l < 8; ++l) {
+ yl[l+ 0] = y1[l+ 0];
+ yl[l+ 8] = y1[l+16];
+ yl[l+16] = y1[l+32];
+ yl[l+24] = y1[l+48];
+ }
+
+ device const uint16_t * q = (device const uint16_t *)(x[i].qs + q_offset);
+ device const uint16_t * h = (device const uint16_t *)(x[i].hmask + l0);
+ device const uint16_t * a = (device const uint16_t *)(x[i].scales);
+ device const half * dh = &x[i].d;
+
+ for (short row = 0; row < nr0; ++row) {
+ const float d_all = (float)dh[0];
+
+ scales16[0] = a[4];
+ scales16[1] = a[5];
+ aux32 = ((scales32 >> s_shift2) << 4) & 0x30303030;
+ scales16[0] = a[il+0];
+ scales16[1] = a[il+1];
+ scales32 = ((scales32 >> s_shift1) & 0x0f0f0f0f) | aux32;
+
+ float s1 = 0, s2 = 0, s3 = 0, s4 = 0, s5 = 0, s6 = 0;
+ for (short l = 0; l < 8; l += 2) {
+ const int32_t qs = q[l/2];
+ s1 += yl[l+0] * (qs & qm[il/2][0]);
+ s2 += yl[l+1] * (qs & qm[il/2][1]);
+ s3 += ((h[l/2] & hm[0]) ? 0.f : yl[l+0]) + ((h[l/2] & hm[1]) ? 0.f : yl[l+1]);
+ s4 += yl[l+16] * (qs & qm[il/2][2]);
+ s5 += yl[l+17] * (qs & qm[il/2][3]);
+ s6 += ((h[l/2] & hm[2]) ? 0.f : yl[l+16]) + ((h[l/2] & hm[3]) ? 0.f : yl[l+17]);
+ }
+ float d1 = d_all * (s1 + 1.f/256.f * s2 - s3*v1);
+ float d2 = d_all * (s4 + 1.f/256.f * s5 - s6*v2);
+ sumf1[row] += d1 * (scales[0] - 32);
+ sumf2[row] += d2 * (scales[2] - 32);
+
+ s1 = s2 = s3 = s4 = s5 = s6 = 0;
+ for (short l = 0; l < 8; l += 2) {
+ const int32_t qs = q[l/2+8];
+ s1 += yl[l+8] * (qs & qm[il/2][0]);
+ s2 += yl[l+9] * (qs & qm[il/2][1]);
+ s3 += ((h[l/2+8] & hm[0]) ? 0.f : yl[l+8]) + ((h[l/2+8] & hm[1]) ? 0.f : yl[l+9]);
+ s4 += yl[l+24] * (qs & qm[il/2][2]);
+ s5 += yl[l+25] * (qs & qm[il/2][3]);
+ s6 += ((h[l/2+8] & hm[2]) ? 0.f : yl[l+24]) + ((h[l/2+8] & hm[3]) ? 0.f : yl[l+25]);
+ }
+ d1 = d_all * (s1 + 1.f/256.f * s2 - s3*v1);
+ d2 = d_all * (s4 + 1.f/256.f * s5 - s6*v2);
+ sumf1[row] += d1 * (scales[1] - 32);
+ sumf2[row] += d2 * (scales[3] - 32);
+
+ q += args.nb01/2;
+ h += args.nb01/2;
+ a += args.nb01/2;
+ dh += args.nb01/2;
+ }
+
+ y1 += 4 * QK_K;
+ }
+
+ for (int row = 0; row < nr0; ++row) {
+ const float sumf = (sumf1[row] + 0.25f * sumf2[row]) / (1 << shift);
+ sumf1[row] = simd_sum(sumf);
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ if (tiisg == 0) {
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ dst_f32[first_row + row] = sumf1[row];
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_q3_K_f32")]]
+kernel void kernel_mul_mv_q3_K_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_q3_K_f32_impl<N_R0_Q3_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_q4_K_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ constexpr uint16_t kmask1 = 0x3f3f;
+ constexpr uint16_t kmask2 = 0x0f0f;
+ constexpr uint16_t kmask3 = 0xc0c0;
+
+ const short ix = tiisg/8; // 0...3
+ const short it = tiisg%8; // 0...7
+ const short iq = it/4; // 0 or 1
+ const short ir = it%4; // 0...3
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_q4_K * x = (device const block_q4_K *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[16];
+ float yh[16];
+
+ float sumf[nr0]={0.f};
+
+ device const float * y4 = y + ix * QK_K + 64 * iq + 8 * ir;
+
+ uint16_t sc16[4];
+ thread const uint8_t * sc8 = (thread const uint8_t *)sc16;
+
+ for (int ib = ix; ib < nb; ib += 4) {
+ float4 sumy = {0.f, 0.f, 0.f, 0.f};
+
+ for (short i = 0; i < 8; ++i) {
+ yl[i+0] = y4[i+ 0]; sumy[0] += yl[i+0];
+ yl[i+8] = y4[i+ 32]; sumy[1] += yl[i+8];
+ yh[i+0] = y4[i+128]; sumy[2] += yh[i+0];
+ yh[i+8] = y4[i+160]; sumy[3] += yh[i+8];
+ }
+
+ device const uint16_t * sc = (device const uint16_t *)x[ib].scales + iq;
+ device const uint16_t * q1 = (device const uint16_t *)x[ib].qs + 16 * iq + 4 * ir;
+ device const half * dh = &x[ib].d;
+
+ for (short row = 0; row < nr0; row++) {
+ sc16[0] = sc[0] & kmask1;
+ sc16[1] = sc[2] & kmask1;
+ sc16[2] = ((sc[4] >> 0) & kmask2) | ((sc[0] & kmask3) >> 2);
+ sc16[3] = ((sc[4] >> 4) & kmask2) | ((sc[2] & kmask3) >> 2);
+
+ device const uint16_t * q2 = q1 + 32;
+
+ float4 acc1 = {0.f, 0.f, 0.f, 0.f};
+ float4 acc2 = {0.f, 0.f, 0.f, 0.f};
+
+ FOR_UNROLL (short i = 0; i < 4; ++i) {
+ acc1[0] += yl[2*i + 0] * (q1[i] & 0x000F);
+ acc1[1] += yl[2*i + 1] * (q1[i] & 0x0F00);
+ acc1[2] += yl[2*i + 8] * (q1[i] & 0x00F0);
+ acc1[3] += yl[2*i + 9] * (q1[i] & 0xF000);
+ acc2[0] += yh[2*i + 0] * (q2[i] & 0x000F);
+ acc2[1] += yh[2*i + 1] * (q2[i] & 0x0F00);
+ acc2[2] += yh[2*i + 8] * (q2[i] & 0x00F0);
+ acc2[3] += yh[2*i + 9] * (q2[i] & 0xF000);
+ }
+
+ sumf[row] += dh[0] * ((acc1[0] + 1.f/256.f * acc1[1]) * sc8[0] +
+ (acc1[2] + 1.f/256.f * acc1[3]) * sc8[1] * 1.f/16.f +
+ (acc2[0] + 1.f/256.f * acc2[1]) * sc8[4] +
+ (acc2[2] + 1.f/256.f * acc2[3]) * sc8[5] * 1.f/16.f) -
+ dh[1] * (sumy[0] * sc8[2] + sumy[1] * sc8[3] + sumy[2] * sc8[6] + sumy[3] * sc8[7]);
+
+ q1 += args.nb01/2;
+ sc += args.nb01/2;
+ dh += args.nb01/2;
+ }
+
+ y4 += 4 * QK_K;
+ }
+
+ device float * dst_f32 = (device float *) dst + (int64_t)im*args.ne0*args.ne1 + (int64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_q4_K_f32")]]
+kernel void kernel_mul_mv_q4_K_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_q4_K_f32_impl<N_R0_Q4_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_q5_K_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_q5_K * x = (device const block_q5_K *) (src0 + offset0);
+ device const float * yy = (device const float *) (src1 + offset1);
+
+ float sumf[nr0]={0.f};
+
+ float yl[16], yh[16];
+
+ constexpr uint16_t kmask1 = 0x3f3f;
+ constexpr uint16_t kmask2 = 0x0f0f;
+ constexpr uint16_t kmask3 = 0xc0c0;
+
+ const short tid = tiisg/4;
+ const short ix = tiisg%4;
+ const short iq = tid/4;
+ const short ir = tid%4;
+
+ const short l0 = 8*ir;
+ const short q_offset = 32*iq + l0;
+ const short y_offset = 64*iq + l0;
+
+ const uint8_t hm1 = 1u << (2*iq);
+ const uint8_t hm2 = hm1 << 1;
+ const uint8_t hm3 = hm1 << 4;
+ const uint8_t hm4 = hm2 << 4;
+
+ uint16_t sc16[4];
+ thread const uint8_t * sc8 = (thread const uint8_t *)sc16;
+
+ device const float * y1 = yy + ix*QK_K + y_offset;
+
+ for (int i = ix; i < nb; i += 4) {
+ device const uint8_t * q1 = x[i].qs + q_offset;
+ device const uint8_t * qh = x[i].qh + l0;
+ device const half * dh = &x[i].d;
+ device const uint16_t * a = (device const uint16_t *)x[i].scales + iq;
+
+ device const float * y2 = y1 + 128;
+ float4 sumy = {0.f, 0.f, 0.f, 0.f};
+ for (short l = 0; l < 8; ++l) {
+ yl[l+0] = y1[l+ 0]; sumy[0] += yl[l+0];
+ yl[l+8] = y1[l+32]; sumy[1] += yl[l+8];
+ yh[l+0] = y2[l+ 0]; sumy[2] += yh[l+0];
+ yh[l+8] = y2[l+32]; sumy[3] += yh[l+8];
+ }
+
+ for (short row = 0; row < nr0; ++row) {
+ device const uint8_t * q2 = q1 + 64;
+
+ sc16[0] = a[0] & kmask1;
+ sc16[1] = a[2] & kmask1;
+ sc16[2] = ((a[4] >> 0) & kmask2) | ((a[0] & kmask3) >> 2);
+ sc16[3] = ((a[4] >> 4) & kmask2) | ((a[2] & kmask3) >> 2);
+
+ float4 acc1 = {0.f};
+ float4 acc2 = {0.f};
+ FOR_UNROLL (short l = 0; l < 8; ++l) {
+ uint8_t h = qh[l];
+ acc1[0] += yl[l+0] * (q1[l] & 0x0F);
+ acc1[1] += yl[l+8] * (q1[l] & 0xF0);
+ acc1[2] += yh[l+0] * (q2[l] & 0x0F);
+ acc1[3] += yh[l+8] * (q2[l] & 0xF0);
+ acc2[0] += h & hm1 ? yl[l+0] : 0.f;
+ acc2[1] += h & hm2 ? yl[l+8] : 0.f;
+ acc2[2] += h & hm3 ? yh[l+0] : 0.f;
+ acc2[3] += h & hm4 ? yh[l+8] : 0.f;
+ }
+
+ sumf[row] += dh[0] * (sc8[0] * (acc1[0] + 16.f*acc2[0]) +
+ sc8[1] * (acc1[1]/16.f + 16.f*acc2[1]) +
+ sc8[4] * (acc1[2] + 16.f*acc2[2]) +
+ sc8[5] * (acc1[3]/16.f + 16.f*acc2[3])) -
+ dh[1] * (sumy[0] * sc8[2] + sumy[1] * sc8[3] + sumy[2] * sc8[6] + sumy[3] * sc8[7]);
+
+ q1 += args.nb01;
+ qh += args.nb01;
+ dh += args.nb01/2;
+ a += args.nb01/2;
+ }
+
+ y1 += 4 * QK_K;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ const float tot = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = tot;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_q5_K_f32")]]
+kernel void kernel_mul_mv_q5_K_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_q5_K_f32_impl<N_R0_Q5_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_q6_K_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ constexpr uint8_t kmask1 = 0x03;
+ constexpr uint8_t kmask2 = 0x0C;
+ constexpr uint8_t kmask3 = 0x30;
+ constexpr uint8_t kmask4 = 0xC0;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_q6_K * x = (device const block_q6_K *) (src0 + offset0);
+ device const float * yy = (device const float *) (src1 + offset1);
+
+ float sumf[nr0] = { 0.f };
+
+ float yl[16];
+
+ const short tid = tiisg/2;
+ const short ix = tiisg%2;
+ const short ip = tid/8; // 0 or 1
+ const short il = tid%8;
+ const short l0 = 4*il;
+ const short is = 8*ip + l0/16;
+
+ const short y_offset = 128*ip + l0;
+ const short q_offset_l = 64*ip + l0;
+ const short q_offset_h = 32*ip + l0;
+
+ for (int i = ix; i < nb; i += 2) {
+ device const uint8_t * q1 = x[i].ql + q_offset_l;
+ device const uint8_t * q2 = q1 + 32;
+ device const uint8_t * qh = x[i].qh + q_offset_h;
+ device const int8_t * sc = x[i].scales + is;
+ device const half * dh = &x[i].d;
+
+ device const float * y = yy + i * QK_K + y_offset;
+
+ for (short l = 0; l < 4; ++l) {
+ yl[4*l + 0] = y[l + 0];
+ yl[4*l + 1] = y[l + 32];
+ yl[4*l + 2] = y[l + 64];
+ yl[4*l + 3] = y[l + 96];
+ }
+
+ for (short row = 0; row < nr0; ++row) {
+ float4 sums = {0.f, 0.f, 0.f, 0.f};
+
+ FOR_UNROLL (short l = 0; l < 4; ++l) {
+ sums[0] += yl[4*l + 0] * ((int8_t)((q1[l] & 0xF) | ((qh[l] & kmask1) << 4)) - 32);
+ sums[1] += yl[4*l + 1] * ((int8_t)((q2[l] & 0xF) | ((qh[l] & kmask2) << 2)) - 32);
+ sums[2] += yl[4*l + 2] * ((int8_t)((q1[l] >> 4) | ((qh[l] & kmask3) << 0)) - 32);
+ sums[3] += yl[4*l + 3] * ((int8_t)((q2[l] >> 4) | ((qh[l] & kmask4) >> 2)) - 32);
+ }
+
+ sumf[row] += dh[0] * (sums[0] * sc[0] + sums[1] * sc[2] + sums[2] * sc[4] + sums[3] * sc[6]);
+
+ q1 += args.nb01;
+ q2 += args.nb01;
+ qh += args.nb01;
+ sc += args.nb01;
+ dh += args.nb01/2;
+ }
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_q6_K_f32")]]
+kernel void kernel_mul_mv_q6_K_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_q6_K_f32_impl<N_R0_Q6_K, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+// ======================= "True" 2-bit
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_iq2_xxs_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq2_xxs * x = (device const block_iq2_xxs *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const int nb32 = nb * (QK_K / 32);
+
+ threadgroup uint64_t * svalues = (threadgroup uint64_t *)(shmem);
+ threadgroup uint8_t * ssigns = (threadgroup uint8_t *)(svalues + 256);
+ {
+ int nval = 4;
+ int pos = (32*sgitg + tiisg)*nval;
+ for (int i = 0; i < nval; ++i) svalues[pos + i] = iq2xxs_grid[pos + i];
+ nval = 2;
+ pos = (32*sgitg + tiisg)*nval;
+ for (int i = 0; i < nval; ++i) ssigns[pos+i] = ksigns_iq2xs[pos+i];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ const int ix = tiisg;
+
+ device const float * y4 = y + 32 * ix;
+
+ for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
+ for (short i = 0; i < 32; ++i) {
+ yl[i] = y4[i];
+ }
+
+ const int ibl = ib32 / (QK_K / 32);
+ const int ib = ib32 % (QK_K / 32);
+
+ device const block_iq2_xxs * xr = x + ibl;
+ device const uint16_t * q2 = xr->qs + 4 * ib;
+ device const half * dh = &xr->d;
+
+ for (short row = 0; row < nr0; row++) {
+ const float db = dh[0];
+ device const uint8_t * aux8 = (device const uint8_t *)q2;
+ const uint32_t aux32 = q2[2] | (q2[3] << 16);
+ const float d = db * (0.5f + (aux32 >> 28));
+
+ float sum = 0;
+ for (short l = 0; l < 4; ++l) {
+ const threadgroup uint8_t * grid = (const threadgroup uint8_t *)(svalues + aux8[l]);
+ const uint8_t signs = ssigns[(aux32 >> 7*l) & 127];
+ for (short j = 0; j < 8; ++j) {
+ sum += yl[8*l + j] * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
+ }
+ }
+ sumf[row] += d * sum;
+
+ dh += args.nb01/2;
+ q2 += args.nb01/2;
+ }
+
+ y4 += 32 * 32;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all * 0.25f;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq2_xxs_f32")]]
+kernel void kernel_mul_mv_iq2_xxs_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ kernel_mul_mv_iq2_xxs_f32_impl<N_R0_IQ2_XXS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_iq2_xs_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq2_xs * x = (device const block_iq2_xs *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const int nb32 = nb * (QK_K / 32);
+
+ threadgroup uint64_t * svalues = (threadgroup uint64_t *)(shmem);
+ threadgroup uint8_t * ssigns = (threadgroup uint8_t *)(svalues + 512);
+ {
+ int nval = 8;
+ int pos = (32*sgitg + tiisg)*nval;
+ for (int i = 0; i < nval; ++i) svalues[pos + i] = iq2xs_grid[pos + i];
+ nval = 2;
+ pos = (32*sgitg + tiisg)*nval;
+ for (int i = 0; i < nval; ++i) ssigns[pos+i] = ksigns_iq2xs[pos+i];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ const int ix = tiisg;
+
+ device const float * y4 = y + 32 * ix;
+
+ for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
+ for (short i = 0; i < 32; ++i) {
+ yl[i] = y4[i];
+ }
+
+ const int ibl = ib32 / (QK_K / 32);
+ const int ib = ib32 % (QK_K / 32);
+
+ device const block_iq2_xs * xr = x + ibl;
+ device const uint16_t * q2 = xr->qs + 4 * ib;
+ device const uint8_t * sc = xr->scales + ib;
+ device const half * dh = &xr->d;
+
+ for (short row = 0; row < nr0; row++) {
+ const float db = dh[0];
+ const uint8_t ls1 = sc[0] & 0xf;
+ const uint8_t ls2 = sc[0] >> 4;
+ const float d1 = db * (0.5f + ls1);
+ const float d2 = db * (0.5f + ls2);
+
+ float sum1 = 0, sum2 = 0;
+ for (short l = 0; l < 2; ++l) {
+ const threadgroup uint8_t * grid = (const threadgroup uint8_t *)(svalues + (q2[l] & 511));
+ const uint8_t signs = ssigns[(q2[l] >> 9)];
+ for (short j = 0; j < 8; ++j) {
+ sum1 += yl[8*l + j] * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
+ }
+ }
+ for (short l = 2; l < 4; ++l) {
+ const threadgroup uint8_t * grid = (const threadgroup uint8_t *)(svalues + (q2[l] & 511));
+ const uint8_t signs = ssigns[(q2[l] >> 9)];
+ for (short j = 0; j < 8; ++j) {
+ sum2 += yl[8*l + j] * grid[j] * (signs & kmask_iq2xs[j] ? -1.f : 1.f);
+ }
+ }
+ sumf[row] += d1 * sum1 + d2 * sum2;
+
+ dh += args.nb01/2;
+ q2 += args.nb01/2;
+ sc += args.nb01;
+ }
+
+ y4 += 32 * 32;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all * 0.25f;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq2_xs_f32")]]
+kernel void kernel_mul_mv_iq2_xs_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq2_xs_f32_impl<N_R0_IQ2_XS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_iq3_xxs_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq3_xxs * x = (device const block_iq3_xxs *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const int nb32 = nb * (QK_K / 32);
+
+ threadgroup uint32_t * svalues = (threadgroup uint32_t *)(shmem);
+ threadgroup uint8_t * ssigns = (threadgroup uint8_t *)(svalues + 256);
+ {
+ int nval = 4;
+ int pos = (32*sgitg + tiisg)*nval;
+ for (int i = 0; i < nval; ++i) svalues[pos + i] = iq3xxs_grid[pos + i];
+ nval = 2;
+ pos = (32*sgitg + tiisg)*nval;
+ for (int i = 0; i < nval; ++i) ssigns[pos+i] = ksigns_iq2xs[pos+i];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ const int ix = tiisg;
+
+ device const float * y4 = y + 32 * ix;
+
+ for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
+ for (short i = 0; i < 32; ++i) {
+ yl[i] = y4[i];
+ }
+
+ const int ibl = ib32 / (QK_K / 32);
+ const int ib = ib32 % (QK_K / 32);
+
+ device const block_iq3_xxs * xr = x + ibl;
+ device const uint8_t * q3 = xr->qs + 8 * ib;
+ device const uint16_t * gas = (device const uint16_t *)(xr->qs + QK_K/4) + 2 * ib;
+ device const half * dh = &xr->d;
+
+ for (short row = 0; row < nr0; row++) {
+ const float db = dh[0];
+ const uint32_t aux32 = gas[0] | (gas[1] << 16);
+ const float d = db * (0.5f + (aux32 >> 28));
+
+ float2 sum = {0};
+ for (short l = 0; l < 4; ++l) {
+ const threadgroup uint8_t * grid1 = (const threadgroup uint8_t *)(svalues + q3[2*l+0]);
+ const threadgroup uint8_t * grid2 = (const threadgroup uint8_t *)(svalues + q3[2*l+1]);
+ const uint8_t signs = ssigns[(aux32 >> 7*l) & 127];
+ for (short j = 0; j < 4; ++j) {
+ sum[0] += yl[8*l + j + 0] * grid1[j] * (signs & kmask_iq2xs[j+0] ? -1.f : 1.f);
+ sum[1] += yl[8*l + j + 4] * grid2[j] * (signs & kmask_iq2xs[j+4] ? -1.f : 1.f);
+ }
+ }
+ sumf[row] += d * (sum[0] + sum[1]);
+
+ dh += args.nb01/2;
+ q3 += args.nb01;
+ gas += args.nb01/2;
+ }
+
+ y4 += 32 * 32;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all * 0.5f;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq3_xxs_f32")]]
+kernel void kernel_mul_mv_iq3_xxs_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq3_xxs_f32_impl<N_R0_IQ3_XXS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_iq3_s_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq3_s * x = (device const block_iq3_s *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const int nb32 = nb * (QK_K / 32);
+
+ threadgroup uint32_t * svalues = (threadgroup uint32_t *) shmem;
+ {
+ int nval = 8;
+ int pos = (32*sgitg + tiisg)*nval;
+ for (int i = 0; i < nval; ++i) svalues[pos + i] = iq3s_grid[pos + i];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ }
+
+ const int ix = tiisg;
+
+ device const float * y4 = y + 32 * ix;
+
+ for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
+ for (short i = 0; i < 32; ++i) {
+ yl[i] = y4[i];
+ }
+
+ const int ibl = ib32 / (QK_K / 32);
+ const int ib = ib32 % (QK_K / 32);
+
+ device const block_iq3_s * xr = x + ibl;
+ device const uint8_t * qs = xr->qs + 8 * ib;
+ device const uint8_t * qh = xr->qh + ib;
+ device const uint8_t * sc = xr->scales + (ib/2);
+ device const uint8_t * signs = xr->signs + 4 * ib;
+ device const half * dh = &xr->d;
+
+ for (short row = 0; row < nr0; row++) {
+ const float db = dh[0];
+ const float d = db * (1 + 2*((sc[0] >> 4*(ib%2)) & 0xf));
+
+ float2 sum = {0};
+ for (short l = 0; l < 4; ++l) {
+ const threadgroup uint32_t * table1 = qh[0] & kmask_iq2xs[2*l+0] ? svalues + 256 : svalues;
+ const threadgroup uint32_t * table2 = qh[0] & kmask_iq2xs[2*l+1] ? svalues + 256 : svalues;
+ const threadgroup uint8_t * grid1 = (const threadgroup uint8_t *)(table1 + qs[2*l+0]);
+ const threadgroup uint8_t * grid2 = (const threadgroup uint8_t *)(table2 + qs[2*l+1]);
+ for (short j = 0; j < 4; ++j) {
+ sum[0] += yl[8*l + j + 0] * grid1[j] * select(1, -1, signs[l] & kmask_iq2xs[j+0]);
+ sum[1] += yl[8*l + j + 4] * grid2[j] * select(1, -1, signs[l] & kmask_iq2xs[j+4]);
+ }
+ }
+ sumf[row] += d * (sum[0] + sum[1]);
+
+ dh += args.nb01/2;
+ qs += args.nb01;
+ qh += args.nb01;
+ sc += args.nb01;
+ signs += args.nb01;
+ }
+
+ y4 += 32 * 32;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq3_s_f32")]]
+kernel void kernel_mul_mv_iq3_s_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq3_s_f32_impl<N_R0_IQ3_S, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_iq2_s_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq2_s * x = (device const block_iq2_s *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const int nb32 = nb * (QK_K / 32);
+
+ //threadgroup uint64_t * svalues = (threadgroup uint64_t *) shmem;
+ //{
+ // int nval = 32;
+ // int pos = (32*sgitg + tiisg)*nval;
+ // for (int i = 0; i < nval; ++i) svalues[pos + i] = iq2s_grid[pos + i];
+ // threadgroup_barrier(mem_flags::mem_threadgroup);
+ //}
+
+ const short ix = tiisg;
+
+ device const float * y4 = y + 32 * ix;
+
+ for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
+ for (short i = 0; i < 32; ++i) {
+ yl[i] = y4[i];
+ }
+
+ const int ibl = ib32 / (QK_K / 32);
+ const int ib = ib32 % (QK_K / 32);
+
+ device const block_iq2_s * xr = x + ibl;
+ device const uint8_t * qs = xr->qs + 4 * ib;
+ device const uint8_t * qh = xr->qh + ib;
+ device const uint8_t * sc = xr->scales + ib;
+ device const uint8_t * signs = qs + QK_K/8;
+ device const half * dh = &xr->d;
+
+ for (short row = 0; row < nr0; row++) {
+ const float db = dh[0];
+ const float d1 = db * (0.5f + (sc[0] & 0xf));
+ const float d2 = db * (0.5f + (sc[0] >> 4));
+
+ float2 sum = {0};
+ for (short l = 0; l < 2; ++l) {
+ //const threadgroup uint8_t * grid1 = (const threadgroup uint8_t *)(svalues + (qs[l+0] | ((qh[0] << (8-2*l)) & 0x300)));
+ //const threadgroup uint8_t * grid2 = (const threadgroup uint8_t *)(svalues + (qs[l+2] | ((qh[0] << (4-2*l)) & 0x300)));
+ constant uint8_t * grid1 = (constant uint8_t *)(iq2s_grid + (qs[l+0] | ((qh[0] << (8-2*l)) & 0x300)));
+ constant uint8_t * grid2 = (constant uint8_t *)(iq2s_grid + (qs[l+2] | ((qh[0] << (4-2*l)) & 0x300)));
+ for (short j = 0; j < 8; ++j) {
+ sum[0] += yl[8*l + j + 0] * grid1[j] * select(1, -1, signs[l+0] & kmask_iq2xs[j]);
+ sum[1] += yl[8*l + j + 16] * grid2[j] * select(1, -1, signs[l+2] & kmask_iq2xs[j]);
+ }
+ }
+ sumf[row] += d1 * sum[0] + d2 * sum[1];
+
+ dh += args.nb01/2;
+ qs += args.nb01;
+ qh += args.nb01;
+ sc += args.nb01;
+ signs += args.nb01;
+ }
+
+ y4 += 32 * 32;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all * 0.25f;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq2_s_f32")]]
+kernel void kernel_mul_mv_iq2_s_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq2_s_f32_impl<N_R0_IQ2_S, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_iq1_s_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq1_s * x = (device const block_iq1_s *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const int nb32 = nb * (QK_K / 32);
+
+ const short ix = tiisg;
+
+ device const float * y4 = y + 32 * ix;
+
+ for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
+ float sumy = 0;
+ for (short i = 0; i < 32; ++i) {
+ yl[i] = y4[i];
+ sumy += yl[i];
+ }
+
+ const int ibl = ib32 / (QK_K / 32);
+ const int ib = ib32 % (QK_K / 32);
+
+ device const block_iq1_s * xr = x + ibl;
+ device const uint8_t * qs = xr->qs + 4 * ib;
+ device const uint16_t * qh = xr->qh + ib;
+ device const half * dh = &xr->d;
+
+ for (short row = 0; row < nr0; row++) {
+ constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
+ constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 5) & 0x700)));
+ constant uint8_t * grid3 = (constant uint8_t *)(iq1s_grid_gpu + (qs[2] | ((qh[0] << 2) & 0x700)));
+ constant uint8_t * grid4 = (constant uint8_t *)(iq1s_grid_gpu + (qs[3] | ((qh[0] >> 1) & 0x700)));
+
+ float sum = 0;
+ for (short j = 0; j < 4; ++j) {
+ sum += yl[j+ 0] * (grid1[j] & 0xf) + yl[j+ 4] * (grid1[j] >> 4)
+ + yl[j+ 8] * (grid2[j] & 0xf) + yl[j+12] * (grid2[j] >> 4)
+ + yl[j+16] * (grid3[j] & 0xf) + yl[j+20] * (grid3[j] >> 4)
+ + yl[j+24] * (grid4[j] & 0xf) + yl[j+28] * (grid4[j] >> 4);
+ }
+ sumf[row] += (float)dh[0] * (sum + sumy * (qh[0] & 0x8000 ? -1 - IQ1S_DELTA : -1 + IQ1S_DELTA)) * (2*((qh[0] >> 12) & 7) + 1);
+
+ dh += args.nb01/2;
+ qs += args.nb01;
+ qh += args.nb01/2;
+ }
+
+ y4 += 32 * 32;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq1_s_f32")]]
+kernel void kernel_mul_mv_iq1_s_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq1_s_f32_impl<N_R0_IQ1_S, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_iq1_m_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq1_m * x = (device const block_iq1_m *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ float yl[32];
+ float sumf[nr0]={0.f};
+
+ const int nb32 = nb * (QK_K / 32);
+
+ const short ix = tiisg;
+
+ device const float * y4 = y + 32 * ix;
+
+ iq1m_scale_t scale;
+
+ for (int ib32 = ix; ib32 < nb32; ib32 += 32) {
+ float4 sumy = {0.f};
+ for (short i = 0; i < 8; ++i) {
+ yl[i+ 0] = y4[i+ 0]; sumy[0] += yl[i+ 0];
+ yl[i+ 8] = y4[i+ 8]; sumy[1] += yl[i+ 8];
+ yl[i+16] = y4[i+16]; sumy[2] += yl[i+16];
+ yl[i+24] = y4[i+24]; sumy[3] += yl[i+24];
+ }
+
+ const int ibl = ib32 / (QK_K / 32);
+ const int ib = ib32 % (QK_K / 32);
+
+ device const block_iq1_m * xr = x + ibl;
+ device const uint8_t * qs = xr->qs + 4 * ib;
+ device const uint8_t * qh = xr->qh + 2 * ib;
+ device const uint16_t * sc = (device const uint16_t *)xr->scales;
+
+ for (short row = 0; row < nr0; row++) {
+ scale.u16 = (sc[0] >> 12) | ((sc[1] >> 8) & 0x00f0) | ((sc[2] >> 4) & 0x0f00) | (sc[3] & 0xf000);
+
+ constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
+ constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 4) & 0x700)));
+ constant uint8_t * grid3 = (constant uint8_t *)(iq1s_grid_gpu + (qs[2] | ((qh[1] << 8) & 0x700)));
+ constant uint8_t * grid4 = (constant uint8_t *)(iq1s_grid_gpu + (qs[3] | ((qh[1] << 4) & 0x700)));
+
+ float2 sum = {0.f};
+ for (short j = 0; j < 4; ++j) {
+ sum[0] += yl[j+ 0] * (grid1[j] & 0xf) + yl[j+ 4] * (grid1[j] >> 4)
+ + yl[j+ 8] * (grid2[j] & 0xf) + yl[j+12] * (grid2[j] >> 4);
+ sum[1] += yl[j+16] * (grid3[j] & 0xf) + yl[j+20] * (grid3[j] >> 4)
+ + yl[j+24] * (grid4[j] & 0xf) + yl[j+28] * (grid4[j] >> 4);
+ }
+ const float delta1 = sumy[0] * (qh[0] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA) + sumy[1] * (qh[0] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
+ const float delta2 = sumy[2] * (qh[1] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA) + sumy[3] * (qh[1] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
+
+ sumf[row] += (float)scale.f16 * ((sum[0] + delta1) * (2*((sc[ib/2] >> (6*(ib%2)+0)) & 7) + 1) +
+ (sum[1] + delta2) * (2*((sc[ib/2] >> (6*(ib%2)+3)) & 7) + 1));
+
+ sc += args.nb01/2;
+ qs += args.nb01;
+ qh += args.nb01;
+ }
+
+ y4 += 32 * 32;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq1_m_f32")]]
+kernel void kernel_mul_mv_iq1_m_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq1_m_f32_impl<N_R0_IQ1_M, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+template<int NR0, typename args_t>
+void kernel_mul_mv_iq4_nl_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ threadgroup float * shmem_f32 = (threadgroup float *) shmem;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * NR0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq4_nl * x = (device const block_iq4_nl *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ const int nb = args.ne00/QK4_NL;
+ const int ns01 = args.nb01/args.nb00;
+
+ const short ix = tiisg/2; // 0...15
+ const short it = tiisg%2; // 0 or 1
+
+ shmem_f32[tiisg] = kvalues_iq4nl_f[tiisg%16];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ float4 yl[4];
+ float sumf[NR0]={0.f};
+
+ device const float * yb = y + ix*QK4_NL + it*8;
+
+ uint32_t aux32[2];
+ thread const uint8_t * q8 = (thread const uint8_t *)aux32;
+
+ float4 qf1, qf2;
+
+ // [TAG_MUL_MV_WEIRD]
+ for (int ib = ix; ib < nb && ib < ns01; ib += 16) {
+ device const float4 * y4 = (device const float4 *)yb;
+ yl[0] = y4[0];
+ yl[1] = y4[4];
+ yl[2] = y4[1];
+ yl[3] = y4[5];
+
+ for (short row = 0; row < NR0; row++) {
+ device const block_iq4_nl & xb = x[row*ns01 + ib];
+ device const uint16_t * q4 = (device const uint16_t *)(xb.qs + 8*it);
+
+ float4 acc1 = {0.f}, acc2 = {0.f};
+
+ aux32[0] = q4[0] | (q4[1] << 16);
+ aux32[1] = (aux32[0] >> 4) & 0x0f0f0f0f;
+ aux32[0] &= 0x0f0f0f0f;
+ qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
+ qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
+ acc1 += yl[0] * qf1;
+ acc2 += yl[1] * qf2;
+
+ aux32[0] = q4[2] | (q4[3] << 16);
+ aux32[1] = (aux32[0] >> 4) & 0x0f0f0f0f;
+ aux32[0] &= 0x0f0f0f0f;
+ qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
+ qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
+ acc1 += yl[2] * qf1;
+ acc2 += yl[3] * qf2;
+
+ acc1 += acc2;
+
+ sumf[row] += (float)xb.d * (acc1[0] + acc1[1] + acc1[2] + acc1[3]);
+ }
+
+ yb += 16 * QK4_NL;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < NR0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq4_nl_f32")]]
+kernel void kernel_mul_mv_iq4_nl_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq4_nl_f32_impl<N_R0_IQ4_NL, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int NR0, typename args_t>
+void kernel_mul_mv_iq4_xs_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ threadgroup float * shmem_f32 = (threadgroup float *) shmem;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+ const int first_row = (r0 * NSG + sgitg) * NR0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_iq4_xs * x = (device const block_iq4_xs *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ const int nb = args.ne00/QK_K;
+ const int ns01 = args.nb01/args.nb00;
+
+ const short ix = tiisg/16; // 0 or 1
+ const short it = tiisg%16; // 0...15
+ const short ib = it/2;
+ const short il = it%2;
+
+ shmem_f32[tiisg] = kvalues_iq4nl_f[tiisg%16];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ float4 yl[4];
+ float sumf[NR0]={0.f};
+
+ device const float * yb = y + ix * QK_K + ib * 32 + il * 8;
+
+ uint32_t aux32[2];
+ thread const uint8_t * q8 = (thread const uint8_t *)aux32;
+
+ float4 qf1, qf2;
+
+ // [TAG_MUL_MV_WEIRD]
+ for (int ibl = ix; ibl < nb && ibl < ns01; ibl += 2) {
+ device const float4 * y4 = (device const float4 *)yb;
+ yl[0] = y4[0];
+ yl[1] = y4[4];
+ yl[2] = y4[1];
+ yl[3] = y4[5];
+
+ for (short row = 0; row < NR0; ++row) {
+ device const block_iq4_xs & xb = x[row*ns01 + ibl];
+ device const uint32_t * q4 = (device const uint32_t *)(xb.qs + 16*ib + 8*il);
+
+ float4 acc1 = {0.f}, acc2 = {0.f};
+
+ aux32[0] = (q4[0] ) & 0x0f0f0f0f;
+ aux32[1] = (q4[0] >> 4) & 0x0f0f0f0f;
+ qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
+ qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
+ acc1 += yl[0] * qf1;
+ acc2 += yl[1] * qf2;
+
+ aux32[0] = (q4[1] ) & 0x0f0f0f0f;
+ aux32[1] = (q4[1] >> 4) & 0x0f0f0f0f;
+ qf1 = {shmem_f32[q8[0]], shmem_f32[q8[1]], shmem_f32[q8[2]], shmem_f32[q8[3]]};
+ qf2 = {shmem_f32[q8[4]], shmem_f32[q8[5]], shmem_f32[q8[6]], shmem_f32[q8[7]]};
+ acc1 += yl[2] * qf1;
+ acc2 += yl[3] * qf2;
+
+ acc1 += acc2;
+
+ const int ls = (((xb.scales_l[ib/2] >> 4*(ib%2)) & 0xf) | (((xb.scales_h >> 2*ib) & 3) << 4)) - 32;
+ sumf[row] += (float)xb.d * ls * (acc1[0] + acc1[1] + acc1[2] + acc1[3]);
+ }
+
+ yb += 2 * QK_K;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < NR0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_iq4_xs_f32")]]
+kernel void kernel_mul_mv_iq4_xs_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_iq4_xs_f32_impl<N_R0_IQ4_XS, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int NR0, typename args_t>
+void kernel_mul_mv_mxfp4_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ threadgroup float * shmem_f32 = (threadgroup float *) shmem;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * NR0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset0 = first_row*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const block_mxfp4 * x = (device const block_mxfp4 *) (src0 + offset0);
+ device const float * y = (device const float *) (src1 + offset1);
+
+ const int nb = args.ne00/QK_MXFP4;
+ const int ns01 = args.nb01/args.nb00; // this can be larger than nb for permuted src0 tensors
+
+ const short ix = tiisg/2; // 0...15
+ const short it = tiisg%2; // 0 or 1
+
+ shmem_f32[tiisg] = kvalues_mxfp4_f[tiisg%16];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ float4 yl[4];
+ float sumf[NR0]={0.f};
+
+ device const float * yb = y + ix*QK_MXFP4 + it*8;
+
+ // note: just the check `ib < nb` is enough, but adding the redundant `&& ib < ns01` check makes the kernel a bit faster
+ // no idea why that is - needs some deeper investigation [TAG_MUL_MV_WEIRD]
+ for (int ib = ix; ib < nb && ib < ns01; ib += 16) {
+ device const float4 * y4 = (device const float4 *) yb;
+
+ yl[0] = y4[0];
+ yl[1] = y4[4];
+ yl[2] = y4[1];
+ yl[3] = y4[5];
+
+ FOR_UNROLL (short row = 0; row < NR0; row++) {
+ device const block_mxfp4 & xb = x[row*ns01 + ib];
+ device const uint8_t * q2 = (device const uint8_t *)(xb.qs + 8*it);
+
+ float4 acc1 = yl[0]*float4(shmem_f32[q2[0] & 0x0F], shmem_f32[q2[1] & 0x0F], shmem_f32[q2[2] & 0x0F], shmem_f32[q2[3] & 0x0F]);
+ float4 acc2 = yl[1]*float4(shmem_f32[q2[0] >> 4 ], shmem_f32[q2[1] >> 4 ], shmem_f32[q2[2] >> 4 ], shmem_f32[q2[3] >> 4 ]);
+ float4 acc3 = yl[2]*float4(shmem_f32[q2[4] & 0x0F], shmem_f32[q2[5] & 0x0F], shmem_f32[q2[6] & 0x0F], shmem_f32[q2[7] & 0x0F]);
+ float4 acc4 = yl[3]*float4(shmem_f32[q2[4] >> 4 ], shmem_f32[q2[5] >> 4 ], shmem_f32[q2[6] >> 4 ], shmem_f32[q2[7] >> 4 ]);
+
+ acc1 = (acc1 + acc3) + (acc2 + acc4);
+
+ sumf[row] += e8m0_to_fp32(xb.e) * ((acc1[0] + acc1[1]) + (acc1[2] + acc1[3]));
+ }
+
+ yb += 16 * QK_MXFP4;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < NR0 && first_row + row < args.ne0; ++row) {
+ float sum_all = simd_sum(sumf[row]);
+ if (tiisg == 0) {
+ dst_f32[first_row + row] = sum_all;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_mxfp4_f32")]]
+kernel void kernel_mul_mv_mxfp4_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_mxfp4_f32_impl<N_R0_MXFP4, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+template<int nr0, typename args_t>
+void kernel_mul_mv_tq2_0_f32_impl(
+ args_t args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg) {
+ const short NSG = FC_mul_mv_nsg;
+
+ const int nb = args.ne00/QK_K;
+
+ const int r0 = tgpig.x;
+ const int r1 = tgpig.y;
+ const int im = tgpig.z;
+
+ const int first_row = (r0 * NSG + sgitg) * nr0;
+
+ const uint i12 = im%FC_mul_mv_ne12;
+ const uint i13 = im/FC_mul_mv_ne12;
+
+ const uint64_t offset1 = r1*args.nb11 + (i12 )*args.nb12 + (i13 )*args.nb13;
+
+ device const float * y = (device const float *) (src1 + offset1);
+
+ device const block_tq2_0 * ax[nr0];
+ for (int row = 0; row < nr0; ++row) {
+ const uint64_t offset0 = (first_row + row)*args.nb01 + (i12/FC_mul_mv_r2)*args.nb02 + (i13/FC_mul_mv_r3)*args.nb03;
+ ax[row] = (device const block_tq2_0 *) ((device char *) src0 + offset0);
+ }
+
+ float sumf[nr0] = {0.f};
+
+ // 8 threads per block, NBLOCK blocks per pass, 2 halves per block per pass
+ constexpr short NBLOCK = 4;
+
+ constexpr short NB = N_SIMDWIDTH/NBLOCK; // threads per block
+
+ const short blk = tiisg / NB; // 0..NBLOCK-1, block handled by this thread
+ const short htg = tiisg % NB; // 0..NB-1, thread within block (0..7)
+
+ // byte and y base offsets within the block (32 elements per thread, 4 per byte)
+ device const float4 * yb4 = (device const float4 *)(y + 4*htg + blk*QK_K);
+
+ // hoisted per-byte coefficients (from y) and total y-sum, shared across rows
+ // ref: https://github.com/ggml-org/llama.cpp/pull/26980
+ float4 coef[4];
+
+ for (int ib = blk; ib < nb; ib += NBLOCK) {
+ FOR_UNROLL (short h0 = 0; h0 < 2; ++h0) {
+ const float4 y0 = yb4[ 0 + 32*h0];
+ const float4 y1 = yb4[ 8 + 32*h0];
+ const float4 y2 = yb4[16 + 32*h0];
+ const float4 y3 = yb4[24 + 32*h0];
+
+ float sumy = 0.f;
+ FOR_UNROLL (short j = 0; j < 4; ++j) {
+ coef[j] = float4(
+ y0[j],
+ y1[j] - 4.0f*y0[j],
+ y2[j] - 4.0f*y1[j],
+ y3[j] - 4.0f*y2[j]);
+
+ sumy += (y0[j] + y1[j]) + (y2[j] + y3[j]);
+ }
+
+ FOR_UNROLL (short row = 0; row < nr0; ++row) {
+ device const block_tq2_0 & xb = ax[row][ib];
+ device const uchar * qs = xb.qs + 4*htg + 32*h0;
+
+ float sum = -sumy;
+ FOR_UNROLL (short j = 0; j < 4; ++j) {
+ // express the 2-bit field shifts (v>>2, v>>4, v>>6) as float floor ops
+ const float v = (float)qs[j];
+
+ const float f0 = v;
+ const float f1 = floor(v*0.25f); // v>>2
+ const float f2 = floor(v*0.0625); // v>>4
+ const float f3 = floor(v*0.015625); // v>>6
+
+ sum += coef[j][0]*f0 + coef[j][1]*f1 + coef[j][2]*f2 + coef[j][3]*f3;
+ }
+
+ sumf[row] += xb.d * sum;
+ }
+ }
+
+ yb4 += QK_K * NBLOCK / 4;
+ }
+
+ device float * dst_f32 = (device float *) dst + (uint64_t)im*args.ne0*args.ne1 + (uint64_t)r1*args.ne0;
+
+ for (int row = 0; row < nr0; ++row) {
+ const float tot = simd_sum(sumf[row]);
+ if (tiisg == 0 && first_row + row < args.ne01) {
+ dst_f32[first_row + row] = tot;
+ }
+ }
+}
+
+[[host_name("kernel_mul_mv_tq2_0_f32")]]
+kernel void kernel_mul_mv_tq2_0_f32(
+ constant ggml_metal_kargs_mul_mv & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+
+ kernel_mul_mv_tq2_0_f32_impl<N_R0_TQ2_0, constant ggml_metal_kargs_mul_mv &>(args, src0, src1, dst, nullptr, tgpig, tiisg, sgitg);
+}
+
+//
+// matrix-vector multiplication
+//
+
+typedef void (kernel_mul_mv_disp_t)(
+ ggml_metal_kargs_mul_mv args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig,
+ ushort tiisg);
+
+typedef void (kernel_mul_mv2_disp_t)(
+ ggml_metal_kargs_mul_mv args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiisg,
+ ushort sgitg);
+
+template<kernel_mul_mv_disp_t disp_fn>
+void mmv_fn(
+ ggml_metal_kargs_mul_mv args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiitg,
+ ushort tiisg,
+ ushort sgitg) {
+ disp_fn(args, src0, src1, dst, tgpig, tiisg);
+}
+
+template<kernel_mul_mv2_disp_t disp_fn>
+void mmv_fn(
+ ggml_metal_kargs_mul_mv args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem,
+ uint3 tgpig,
+ ushort tiitg,
+ ushort tiisg,
+ ushort sgitg) {
+ disp_fn(args, src0, src1, dst, shmem, tgpig, tiisg, sgitg);
+}
+
+typedef decltype(mmv_fn<kernel_mul_mv_t_t_disp<half, half, ggml_metal_kargs_mul_mv>>) mul_mv_disp_fn_t;
+
+template<mul_mv_disp_fn_t disp_fn>
+kernel void kernel_mul_mv_id(
+ constant ggml_metal_kargs_mul_mv_id & args,
+ device const char * src0s,
+ device const char * src1,
+ device char * dst,
+ device const char * ids,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]]) {
+ const int iid1 = tgpig.z/args.nei0;
+ const int idx = tgpig.z%args.nei0;
+
+ tgpig.z = 0;
+
+ const int32_t i02 = ((device const int32_t *) (ids + iid1*args.nbi1))[idx];
+
+ const int64_t i11 = idx % args.ne11;
+ const int64_t i12 = iid1;
+
+ const int64_t i1 = idx;
+ const int64_t i2 = i12;
+
+ device const char * src0_cur = src0s + i02*args.nb02;
+ device const char * src1_cur = src1 + i11*args.nb11 + i12*args.nb12;
+
+ device char * dst_cur = dst + (i1*args.ne0 + i2*args.ne1*args.ne0)*sizeof(float);
+
+ ggml_metal_kargs_mul_mv args0 = {
+ /*.ne00 =*/ args.ne00,
+ /*.ne01 =*/ args.ne01,
+ /*.ne02 =*/ 1, // args.ne02,
+ /*.nb00 =*/ args.nb00,
+ /*.nb01 =*/ args.nb01,
+ /*.nb02 =*/ args.nb02,
+ /*.nb03 =*/ args.nb02, // args.ne02 == 1
+ /*.ne10 =*/ args.ne10,
+ /*.ne11 =*/ 1, // args.ne11,
+ /*.ne12 =*/ 1, // args.ne12,
+ /*.nb10 =*/ args.nb10,
+ /*.nb11 =*/ args.nb11,
+ /*.nb12 =*/ args.nb12,
+ /*.nb13 =*/ args.nb12, // ne12 == 1
+ /*.ne0 =*/ args.ne0,
+ /*.ne1 =*/ 1, // args.ne1,
+ /*.nr0 =*/ args.nr0,
+ /*.r2 =*/ 1,
+ /*.r3 =*/ 1,
+ };
+
+ disp_fn(
+ args0,
+ /* src0 */ src0_cur,
+ /* src1 */ src1_cur,
+ /* dst */ dst_cur,
+ shmem,
+ tgpig,
+ tiitg,
+ tiisg,
+ sgitg);
+}
+
+typedef decltype(kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<float, float>>>) kernel_mul_mv_id_t;
+
+typedef decltype(kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<float, float4, float, float4>>>) kernel_mul_mv_id_4_t;
+
+template [[host_name("kernel_mul_mv_id_f32_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<float, float>>>;
+template [[host_name("kernel_mul_mv_id_f16_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<half, float>>>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mv_id_bf16_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_disp<bfloat, float>>>;
+#endif
+template [[host_name("kernel_mul_mv_id_f32_f32_4")]] kernel kernel_mul_mv_id_4_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<float, float4, float, float4>>>;
+template [[host_name("kernel_mul_mv_id_f16_f32_4")]] kernel kernel_mul_mv_id_4_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<half, half4, float, float4>>>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_mul_mv_id_bf16_f32_4")]] kernel kernel_mul_mv_id_4_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_t_t_4_disp<bfloat, bfloat4, float, float4>>>;
+#endif
+
+template [[host_name("kernel_mul_mv_id_q8_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q8_0_f32_impl<N_R0_Q8_0>>>;
+
+template [[host_name("kernel_mul_mv_id_q1_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q1_0_f32_impl<N_R0_Q1_0>>>;
+template [[host_name("kernel_mul_mv_id_q2_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q2_0_f32_impl<N_R0_Q2_0>>>;
+template [[host_name("kernel_mul_mv_id_q4_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q4_0, N_R0_Q4_0>>>;
+template [[host_name("kernel_mul_mv_id_q4_1_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q4_1, N_R0_Q4_1>>>;
+template [[host_name("kernel_mul_mv_id_q5_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q5_0, N_R0_Q5_0>>>;
+template [[host_name("kernel_mul_mv_id_q5_1_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<mul_vec_q_n_f32_impl<block_q5_1, N_R0_Q5_1>>>;
+
+template [[host_name("kernel_mul_mv_id_mxfp4_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_mxfp4_f32_impl<N_R0_MXFP4>>>;
+
+template [[host_name("kernel_mul_mv_id_q2_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q2_K_f32_impl <N_R0_Q2_K>>>;
+template [[host_name("kernel_mul_mv_id_q3_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q3_K_f32_impl <N_R0_Q3_K>>>;
+template [[host_name("kernel_mul_mv_id_q4_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q4_K_f32_impl <N_R0_Q4_K>>>;
+template [[host_name("kernel_mul_mv_id_q5_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q5_K_f32_impl <N_R0_Q5_K>>>;
+template [[host_name("kernel_mul_mv_id_q6_K_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_q6_K_f32_impl <N_R0_Q6_K>>>;
+template [[host_name("kernel_mul_mv_id_iq1_s_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq1_s_f32_impl <N_R0_IQ1_S>>>;
+template [[host_name("kernel_mul_mv_id_iq1_m_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq1_m_f32_impl <N_R0_IQ1_M>>>;
+template [[host_name("kernel_mul_mv_id_iq2_xxs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq2_xxs_f32_impl<N_R0_IQ2_XXS>>>;
+template [[host_name("kernel_mul_mv_id_iq2_xs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq2_xs_f32_impl <N_R0_IQ2_XS>>>;
+template [[host_name("kernel_mul_mv_id_iq3_xxs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq3_xxs_f32_impl<N_R0_IQ3_XXS>>>;
+template [[host_name("kernel_mul_mv_id_iq3_s_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq3_s_f32_impl <N_R0_IQ3_S>>>;
+template [[host_name("kernel_mul_mv_id_iq2_s_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq2_s_f32_impl <N_R0_IQ2_S>>>;
+template [[host_name("kernel_mul_mv_id_iq4_nl_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq4_nl_f32_impl <N_R0_IQ4_NL>>>;
+template [[host_name("kernel_mul_mv_id_iq4_xs_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_iq4_xs_f32_impl <N_R0_IQ4_XS>>>;
+template [[host_name("kernel_mul_mv_id_tq2_0_f32")]] kernel kernel_mul_mv_id_t kernel_mul_mv_id<mmv_fn<kernel_mul_mv_tq2_0_f32_impl <N_R0_TQ2_0>>>;
diff --git a/ggml/src/ggml-metal/kernels/norm.metal b/ggml/src/ggml-metal/kernels/norm.metal
new file mode 100644
index 0000000000..7e42389fe5
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/norm.metal
@@ -0,0 +1,308 @@
+#include "common.h"
+
+// F == 1 : norm (no fuse)
+// F == 2 : norm + mul
+// F == 3 : norm + mul + add
+template <typename T, short F>
+kernel void kernel_norm_fuse_impl(
+ constant ggml_metal_kargs_norm & args,
+ device const char * src0,
+ device const char * src1_0,
+ device const char * src1_1,
+ device char * dst,
+ threadgroup float * shmem_f32 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ if (sgitg == 0) {
+ shmem_f32[tiisg] = 0.0f;
+ }
+
+ const int i01 = tgpig.x;
+ const int i02 = tgpig.y;
+ const int i03 = tgpig.z;
+
+ device const T * x = (device const T *) (src0 + i03*args.nbf3[0] + i02*args.nbf2[0] + i01*args.nbf1[0]);
+
+ device const T * f0 = (device const T *) (src1_0 + (i03%args.nef3[1])*args.nbf3[1] + (i02%args.nef2[1])*args.nbf2[1] + (i01%args.nef1[1])*args.nbf1[1]);
+ device const T * f1 = (device const T *) (src1_1 + (i03%args.nef3[2])*args.nbf3[2] + (i02%args.nef2[2])*args.nbf2[2] + (i01%args.nef1[2])*args.nbf1[2]);
+
+ T sumft(0.0f);
+
+ float sumf = 0.0f;
+
+ for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
+ sumft += x[i00];
+ }
+ sumf = dot(sumft, T(1.0f));
+ sumf = simd_sum(sumf);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ shmem_f32[sgitg] = sumf;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ sumf = shmem_f32[tiisg];
+ sumf = simd_sum(sumf);
+
+ const float mean = sumf/args.ne00;
+
+ device T * y = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1);
+
+ sumf = 0.0f;
+ for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
+ y[i00] = x[i00] - mean;
+ sumf += dot(y[i00], y[i00]);
+ }
+ sumf = simd_sum(sumf);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ shmem_f32[sgitg] = sumf;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ sumf = shmem_f32[tiisg];
+ sumf = simd_sum(sumf);
+
+ const float variance = sumf/args.ne00;
+
+ const float scale = 1.0f/sqrt(variance + args.eps);
+ for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
+ if (F == 1) {
+ y[i00] = (y[i00]*scale);
+ }
+ if (F == 2) {
+ y[i00] = (y[i00]*scale)*f0[i00];
+ }
+ if (F == 3) {
+ y[i00] = (y[i00]*scale)*f0[i00] + f1[i00];
+ }
+ }
+}
+
+typedef decltype(kernel_norm_fuse_impl<float4, 1>) kernel_norm_fuse_t;
+
+template [[host_name("kernel_norm_f32")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float, 1>;
+template [[host_name("kernel_norm_mul_f32")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float, 2>;
+template [[host_name("kernel_norm_mul_add_f32")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float, 3>;
+
+template [[host_name("kernel_norm_f32_4")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float4, 1>;
+template [[host_name("kernel_norm_mul_f32_4")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float4, 2>;
+template [[host_name("kernel_norm_mul_add_f32_4")]] kernel kernel_norm_fuse_t kernel_norm_fuse_impl<float4, 3>;
+
+// F == 1 : rms_norm (no fuse)
+// F == 2 : rms_norm + mul
+// F == 3 : rms_norm + mul + add
+template <typename T, short F>
+kernel void kernel_rms_norm_fuse_impl(
+ constant ggml_metal_kargs_norm & args,
+ device const char * src0,
+ device const char * src1_0,
+ device const char * src1_1,
+ device char * dst,
+ threadgroup float * shmem_f32 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ if (sgitg == 0) {
+ shmem_f32[tiisg] = 0.0f;
+ }
+
+ const int i01 = tgpig.x;
+ const int i02 = tgpig.y;
+ const int i03 = tgpig.z;
+
+ device const T * x = (device const T *) (src0 + i03*args.nbf3[0] + i02*args.nbf2[0] + i01*args.nbf1[0]);
+
+ device const T * f0 = (device const T *) (src1_0 + (i03%args.nef3[1])*args.nbf3[1] + (i02%args.nef2[1])*args.nbf2[1] + (i01%args.nef1[1])*args.nbf1[1]);
+ device const T * f1 = (device const T *) (src1_1 + (i03%args.nef3[2])*args.nbf3[2] + (i02%args.nef2[2])*args.nbf2[2] + (i01%args.nef1[2])*args.nbf1[2]);
+
+ float sumf = 0.0f;
+
+ // parallel sum
+ for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
+ sumf += dot(x[i00], x[i00]);
+ }
+ sumf = simd_sum(sumf);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ shmem_f32[sgitg] = sumf;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ sumf = shmem_f32[tiisg];
+ sumf = simd_sum(sumf);
+
+ const float mean = sumf/args.ne00;
+ const float scale = 1.0f/sqrt(mean + args.eps);
+
+ device T * y = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1);
+ for (int i00 = tpitg.x; i00 < args.ne00_t; i00 += ntg.x) {
+ if (F == 1) {
+ y[i00] = (x[i00]*scale);
+ }
+ if (F == 2) {
+ y[i00] = (x[i00]*scale)*f0[i00];
+ }
+ if (F == 3) {
+ y[i00] = (x[i00]*scale)*f0[i00] + f1[i00];
+ }
+ }
+}
+
+typedef decltype(kernel_rms_norm_fuse_impl<float4, 1>) kernel_rms_norm_fuse_t;
+
+template [[host_name("kernel_rms_norm_f32")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float, 1>;
+template [[host_name("kernel_rms_norm_mul_f32")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float, 2>;
+template [[host_name("kernel_rms_norm_mul_add_f32")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float, 3>;
+
+template [[host_name("kernel_rms_norm_f32_4")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float4, 1>;
+template [[host_name("kernel_rms_norm_mul_f32_4")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float4, 2>;
+template [[host_name("kernel_rms_norm_mul_add_f32_4")]] kernel kernel_rms_norm_fuse_t kernel_rms_norm_fuse_impl<float4, 3>;
+
+template <typename T0, typename T>
+kernel void kernel_l2_norm_impl(
+ constant ggml_metal_kargs_l2_norm & args,
+ device const char * src0,
+ device char * dst,
+ threadgroup float * shmem_f32 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int i03 = tgpig.z;
+ const int i02 = tgpig.y;
+ const int i01 = tgpig.x;
+
+ if (sgitg == 0) {
+ shmem_f32[tiisg] = 0.0f;
+ }
+
+ device const T0 * x = (device const T0 *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
+ device T * y = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1);
+
+ float sumf = 0.0f;
+
+ // parallel sum
+ for (int i00 = tpitg.x; i00 < args.ne00; i00 += ntg.x) {
+ sumf += dot(x[i00], x[i00]);
+ }
+ sumf = simd_sum(sumf);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ shmem_f32[sgitg] = sumf;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ sumf = shmem_f32[tiisg];
+ sumf = simd_sum(sumf);
+
+ const float scale = 1.0f/max(sqrt(sumf), args.eps);
+
+ for (int i00 = tpitg.x; i00 < args.ne00; i00 += ntg.x) {
+ y[i00] = x[i00] * scale;
+ }
+}
+
+typedef decltype(kernel_l2_norm_impl<float, float>) kernel_l2_norm_t;
+
+template [[host_name("kernel_l2_norm_f32_f32")]] kernel kernel_l2_norm_t kernel_l2_norm_impl<float, float>;
+template [[host_name("kernel_l2_norm_f32_f32_4")]] kernel kernel_l2_norm_t kernel_l2_norm_impl<float4, float4>;
+
+kernel void kernel_group_norm_f32(
+ constant ggml_metal_kargs_group_norm & args,
+ device const float * src0,
+ device float * dst,
+ threadgroup float * buf [[threadgroup(0)]],
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint sgitg[[simdgroup_index_in_threadgroup]],
+ uint tiisg[[thread_index_in_simdgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ const int64_t ne = args.ne00*args.ne01*args.ne02;
+ const int64_t gs = args.ne00*args.ne01*((args.ne02 + args.ngrp - 1) / args.ngrp);
+
+ int start = tgpig * gs;
+ int end = start + gs;
+
+ start += tpitg;
+
+ if (end >= ne) {
+ end = ne;
+ }
+
+ float tmp = 0.0f; // partial sum for thread in warp
+
+ for (int j = start; j < end; j += ntg) {
+ tmp += src0[j];
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ tmp = simd_sum(tmp);
+ if (ntg > N_SIMDWIDTH) {
+ if (sgitg == 0) {
+ buf[tiisg] = 0.0f;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ buf[sgitg] = tmp;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ tmp = buf[tiisg];
+ tmp = simd_sum(tmp);
+ }
+
+ const float mean = tmp / gs;
+ tmp = 0.0f;
+
+ for (int j = start; j < end; j += ntg) {
+ float xi = src0[j] - mean;
+ dst[j] = xi;
+ tmp += xi * xi;
+ }
+
+ tmp = simd_sum(tmp);
+ if (ntg > N_SIMDWIDTH) {
+ if (sgitg == 0) {
+ buf[tiisg] = 0.0f;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ buf[sgitg] = tmp;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ tmp = buf[tiisg];
+ tmp = simd_sum(tmp);
+ }
+
+ const float variance = tmp / gs;
+ const float scale = 1.0f/sqrt(variance + args.eps);
+ for (int j = start; j < end; j += ntg) {
+ dst[j] *= scale;
+ }
+}
diff --git a/ggml/src/ggml-metal/kernels/pool.metal b/ggml/src/ggml-metal/kernels/pool.metal
new file mode 100644
index 0000000000..13d355b9de
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/pool.metal
@@ -0,0 +1,148 @@
+#include "common.h"
+
+kernel void kernel_pool_2d_max_f32(
+ constant ggml_metal_kargs_pool_2d & args,
+ device const float * src0,
+ device float * dst,
+ uint gid[[thread_position_in_grid]]) {
+
+ if (gid >= args.np) {
+ return;
+ }
+
+ const int idx = gid;
+ const int I_HW = args.IH * args.IW;
+ const int O_HW = args.OH * args.OW;
+ const int nc = idx / O_HW;
+ const int cur_oh = idx % O_HW / args.OW;
+ const int cur_ow = idx % O_HW % args.OW;
+
+ device const float * i_ptr = src0 + nc * I_HW;
+ device float * o_ptr = dst + nc * O_HW;
+
+ const int start_h = cur_oh * args.s1 - args.p1;
+ const int bh = MAX(0, start_h);
+ const int eh = MIN(args.IH, start_h + args.k1);
+ const int start_w = cur_ow * args.s0 - args.p0;
+ const int bw = MAX(0, start_w);
+ const int ew = MIN(args.IW, start_w + args.k0);
+
+ float res = -INFINITY;
+
+ for (int i = bh; i < eh; i += 1) {
+ for (int j = bw; j < ew; j += 1) {
+ res = MAX(res, i_ptr[i * args.IW + j]);
+ }
+ }
+
+ o_ptr[cur_oh * args.OW + cur_ow] = res;
+}
+
+kernel void kernel_pool_2d_avg_f32(
+ constant ggml_metal_kargs_pool_2d & args,
+ device const float * src0,
+ device float * dst,
+ uint gid[[thread_position_in_grid]]) {
+
+ if (gid >= args.np) {
+ return;
+ }
+
+ const int idx = gid;
+ const int I_HW = args.IH * args.IW;
+ const int O_HW = args.OH * args.OW;
+ const int nc = idx / O_HW;
+ const int cur_oh = idx % O_HW / args.OW;
+ const int cur_ow = idx % O_HW % args.OW;
+
+ device const float * i_ptr = src0 + nc * I_HW;
+ device float * o_ptr = dst + nc * O_HW;
+
+ const int start_h = cur_oh * args.s1 - args.p1;
+ const int bh = MAX(0, start_h);
+ const int eh = MIN(args.IH, start_h + args.k1);
+ const int start_w = cur_ow * args.s0 - args.p0;
+ const int bw = MAX(0, start_w);
+ const int ew = MIN(args.IW, start_w + args.k0);
+ // const float scale = 1. / ((eh - bh) * (ew - bw));
+ const float scale = 1. / (args.k0 * args.k1);
+
+ float res = 0;
+
+ for (int i = bh; i < eh; i += 1) {
+ for (int j = bw; j < ew; j += 1) {
+ float cur = i_ptr[i * args.IW + j];
+ res += cur * scale;
+ }
+ }
+
+ o_ptr[cur_oh * args.OW + cur_ow] = res;
+}
+
+
+kernel void kernel_pool_1d_max_f32(
+ constant ggml_metal_kargs_pool_1d & args,
+ device const float * src,
+ device float * dst,
+ uint gid [[thread_position_in_grid]]
+) {
+
+ if (gid >= args.np) {
+ return;
+ }
+
+ const int ow = (int)gid % args.OW;
+ const int row = (int)gid / args.OW;
+
+ const int base = ow * args.s0 - args.p0;
+
+ float acc = -INFINITY;
+
+ const int src_off = row * args.IW;
+ const int dst_off = row * args.OW;
+
+ for (int ki = 0; ki < args.k0; ++ki) {
+ int j = base + ki;
+ if (j < 0 || j >= args.IW){
+ continue;
+ }
+ float v = src[src_off + j];
+ acc = max(acc, v);
+ }
+
+ dst[dst_off + ow] = acc;
+}
+
+kernel void kernel_pool_1d_avg_f32(
+ constant ggml_metal_kargs_pool_1d & args,
+ device const float * src,
+ device float * dst,
+ uint gid [[thread_position_in_grid]]
+) {
+
+ if (gid >= args.np) {
+ return;
+ }
+
+ const int ow = (int)gid % args.OW;
+ const int row = (int)gid / args.OW;
+
+ const int base = ow * args.s0 - args.p0;
+
+ float acc = 0.0f;
+ int cnt = 0;
+
+ const int src_off = row * args.IW;
+ const int dst_off = row * args.OW;
+
+ for (int ki = 0; ki < args.k0; ++ki) {
+ const int j = base + ki;
+ if (j < 0 || j >= args.IW) {
+ continue;
+ }
+ acc += src[src_off + j];
+ cnt += 1;
+ }
+
+ dst[dst_off + ow] = (cnt > 0) ? (acc / (float)cnt) : 0.0f;
+}
diff --git a/ggml/src/ggml-metal/kernels/quantize.h b/ggml/src/ggml-metal/kernels/quantize.h
new file mode 100644
index 0000000000..0741b22253
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/quantize.h
@@ -0,0 +1,262 @@
+#pragma once
+
+#include "common.h"
+
+void quantize_q1_0(device const float * src, device block_q1_0 & dst) {
+ float sum_abs = 0.0f;
+ for (int j = 0; j < QK1_0; j++) {
+ sum_abs += fabs(src[j]);
+ }
+ dst.d = sum_abs / QK1_0;
+
+ for (int j = 0; j < QK1_0 / 8; j++) {
+ dst.qs[j] = 0;
+ }
+ for (int j = 0; j < QK1_0; j++) {
+ if (src[j] >= 0.0f) {
+ dst.qs[j / 8] |= (1 << (j % 8));
+ }
+ }
+}
+
+void quantize_q2_0(device const float * src, device block_q2_0 & dst) {
+ float amax = 0.0f;
+ for (int j = 0; j < QK2_0; j++) {
+ float a = fabs(src[j]);
+ if (a > amax) amax = a;
+ }
+ const float d = amax;
+ dst.d = d;
+
+ const float id = d > 0.0f ? 1.0f / d : 0.0f;
+
+ for (int j = 0; j < QK2_0 / 4; j++) {
+ dst.qs[j] = 0;
+ }
+ for (int j = 0; j < QK2_0; j++) {
+ int q = (int)round(src[j] * id) + 1;
+ q = max(0, min(3, q));
+ dst.qs[j / 4] |= (q << (2 * (j % 4)));
+ }
+}
+
+void quantize_q4_0(device const float * src, device block_q4_0 & dst) {
+#pragma METAL fp math_mode(safe)
+ float amax = 0.0f; // absolute max
+ float max = 0.0f;
+
+ for (int j = 0; j < QK4_0; j++) {
+ const float v = src[j];
+ if (amax < fabs(v)) {
+ amax = fabs(v);
+ max = v;
+ }
+ }
+
+ const float d = max / -8;
+ const float id = d ? 1.0f/d : 0.0f;
+
+ dst.d = d;
+
+ for (int j = 0; j < QK4_0/2; ++j) {
+ const float x0 = src[0 + j]*id;
+ const float x1 = src[QK4_0/2 + j]*id;
+
+ const uint8_t xi0 = MIN(15, (int8_t)(x0 + 8.5f));
+ const uint8_t xi1 = MIN(15, (int8_t)(x1 + 8.5f));
+
+ dst.qs[j] = xi0;
+ dst.qs[j] |= xi1 << 4;
+ }
+}
+
+void quantize_q4_1(device const float * src, device block_q4_1 & dst) {
+#pragma METAL fp math_mode(safe)
+ float min = FLT_MAX;
+ float max = -FLT_MAX;
+
+ for (int j = 0; j < QK4_1; j++) {
+ const float v = src[j];
+ if (min > v) min = v;
+ if (max < v) max = v;
+ }
+
+ const float d = (max - min) / ((1 << 4) - 1);
+ const float id = d ? 1.0f/d : 0.0f;
+
+ dst.d = d;
+ dst.m = min;
+
+ for (int j = 0; j < QK4_1/2; ++j) {
+ const float x0 = (src[0 + j] - min)*id;
+ const float x1 = (src[QK4_1/2 + j] - min)*id;
+
+ const uint8_t xi0 = MIN(15, (int8_t)(x0 + 0.5f));
+ const uint8_t xi1 = MIN(15, (int8_t)(x1 + 0.5f));
+
+ dst.qs[j] = xi0;
+ dst.qs[j] |= xi1 << 4;
+ }
+}
+
+void quantize_q5_0(device const float * src, device block_q5_0 & dst) {
+#pragma METAL fp math_mode(safe)
+ float amax = 0.0f; // absolute max
+ float max = 0.0f;
+
+ for (int j = 0; j < QK5_0; j++) {
+ const float v = src[j];
+ if (amax < fabs(v)) {
+ amax = fabs(v);
+ max = v;
+ }
+ }
+
+ const float d = max / -16;
+ const float id = d ? 1.0f/d : 0.0f;
+
+ dst.d = d;
+
+ uint32_t qh = 0;
+ for (int j = 0; j < QK5_0/2; ++j) {
+ const float x0 = src[0 + j]*id;
+ const float x1 = src[QK5_0/2 + j]*id;
+
+ const uint8_t xi0 = MIN(31, (int8_t)(x0 + 16.5f));
+ const uint8_t xi1 = MIN(31, (int8_t)(x1 + 16.5f));
+
+ dst.qs[j] = (xi0 & 0xf) | ((xi1 & 0xf) << 4);
+ qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
+ qh |= ((xi1 & 0x10u) >> 4) << (j + QK5_0/2);
+ }
+
+ thread const uint8_t * qh8 = (thread const uint8_t *)&qh;
+
+ for (int j = 0; j < 4; ++j) {
+ dst.qh[j] = qh8[j];
+ }
+}
+
+void quantize_q5_1(device const float * src, device block_q5_1 & dst) {
+#pragma METAL fp math_mode(safe)
+ float max = src[0];
+ float min = src[0];
+
+ for (int j = 1; j < QK5_1; j++) {
+ const float v = src[j];
+ min = v < min ? v : min;
+ max = v > max ? v : max;
+ }
+
+ const float d = (max - min) / 31;
+ const float id = d ? 1.0f/d : 0.0f;
+
+ dst.d = d;
+ dst.m = min;
+
+ uint32_t qh = 0;
+ for (int j = 0; j < QK5_1/2; ++j) {
+ const float x0 = (src[0 + j] - min)*id;
+ const float x1 = (src[QK5_1/2 + j] - min)*id;
+
+ const uint8_t xi0 = (uint8_t)(x0 + 0.5f);
+ const uint8_t xi1 = (uint8_t)(x1 + 0.5f);
+
+ dst.qs[j] = (xi0 & 0xf) | ((xi1 & 0xf) << 4);
+ qh |= ((xi0 & 0x10u) >> 4) << (j + 0);
+ qh |= ((xi1 & 0x10u) >> 4) << (j + QK5_1/2);
+ }
+
+ thread const uint8_t * qh8 = (thread const uint8_t *)&qh;
+
+ for (int j = 0; j < 4; ++j) {
+ dst.qh[j] = qh8[j];
+ }
+}
+
+void quantize_q8_0(device const float * src, device block_q8_0 & dst) {
+#pragma METAL fp math_mode(safe)
+ float amax = 0.0f; // absolute max
+
+ for (int j = 0; j < QK8_0; j++) {
+ const float v = src[j];
+ amax = MAX(amax, fabs(v));
+ }
+
+ const float d = amax / ((1 << 7) - 1);
+ const float id = d ? 1.0f/d : 0.0f;
+
+ dst.d = d;
+
+ for (int j = 0; j < QK8_0; ++j) {
+ const float x0 = src[j]*id;
+
+ dst.qs[j] = round(x0);
+ }
+}
+
+void quantize_iq4_nl(device const float * src, device block_iq4_nl & dst) {
+#pragma METAL fp math_mode(safe)
+ float amax = 0.0f; // absolute max
+ float max = 0.0f;
+
+ for (int j = 0; j < QK4_NL; j++) {
+ const float v = src[j];
+ if (amax < fabs(v)) {
+ amax = fabs(v);
+ max = v;
+ }
+ }
+
+ const float d = max / kvalues_iq4nl_f[0];
+ const float id = d ? 1.0f/d : 0.0f;
+
+ float sumqx = 0, sumq2 = 0;
+ for (int j = 0; j < QK4_NL/2; ++j) {
+ const float x0 = src[0 + j]*id;
+ const float x1 = src[QK4_NL/2 + j]*id;
+
+ const uint8_t xi0 = best_index_int8(16, kvalues_iq4nl_f, x0);
+ const uint8_t xi1 = best_index_int8(16, kvalues_iq4nl_f, x1);
+
+ dst.qs[j] = xi0 | (xi1 << 4);
+
+ const float v0 = kvalues_iq4nl_f[xi0];
+ const float v1 = kvalues_iq4nl_f[xi1];
+ const float w0 = src[0 + j]*src[0 + j];
+ const float w1 = src[QK4_NL/2 + j]*src[QK4_NL/2 + j];
+ sumqx += w0*v0*src[j] + w1*v1*src[QK4_NL/2 + j];
+ sumq2 += w0*v0*v0 + w1*v1*v1;
+
+ }
+
+ dst.d = sumq2 > 0 ? sumqx/sumq2 : d;
+}
+
+void quantize_tq2_0(device const float * src, device block_tq2_0 & dst) {
+#pragma METAL fp math_mode(safe)
+ float amax = 0.0f; // absolute max
+
+ for (int j = 0; j < QK_K; j++) {
+ const float v = src[j];
+ amax = MAX(amax, fabs(v));
+ }
+
+ const float d = amax;
+ const float id = d ? 1.0f/d : 0.0f;
+
+ dst.d = (half) d;
+
+ for (int j = 0; j < QK_K/4; j += 32) {
+ for (int m = 0; m < 32; ++m) {
+ uint8_t q = 0;
+ for (int n = 0; n < 4; ++n) {
+ // -1, 0, 1 -> 0, 1, 2
+ int xi = (int)round(src[m + n*32] * id) + 1;
+ q += (uint8_t)((xi & 3) << (2*n));
+ }
+ dst.qs[j + m] = q;
+ }
+ src += 4*32;
+ }
+}
diff --git a/ggml/src/ggml-metal/kernels/quantize.metal b/ggml/src/ggml-metal/kernels/quantize.metal
new file mode 100644
index 0000000000..59d0afe969
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/quantize.metal
@@ -0,0 +1,435 @@
+#include "common.h"
+#include "dequantize.h"
+#include "quantize.h"
+
+template<typename T0, typename T1>
+kernel void kernel_cpy_t_t(
+ constant ggml_metal_kargs_cpy & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig[2];
+ const int32_t i02 = tgpig[1];
+ const int32_t i01 = ntg[1] == 1 ? tgpig[0]%args.ne01 : tgpig[0]*ntg[1] + tpitg.y;
+ const int32_t iw0 = ntg[1] == 1 ? tgpig[0]/args.ne01 : 0;
+
+ if (i01 >= args.ne01) {
+ return;
+ }
+
+ const int64_t n = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00;
+
+ const int32_t i3 = n/(args.ne2*args.ne1*args.ne0);
+ const int32_t i2 = (n - i3*args.ne2*args.ne1*args.ne0)/(args.ne1*args.ne0);
+ const int32_t i1 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0)/args.ne0;
+ const int32_t i0 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0 - i1*args.ne0);
+
+ device T1 * dst_data = (device T1 *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ for (int32_t i00 = iw0*ntg[0] + tpitg.x; i00 < args.ne00;) {
+ device const T0 * src = (device T0 *)(src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + i00*args.nb00);
+ dst_data[i00] = (T1) src[0];
+ break;
+ }
+}
+
+typedef decltype(kernel_cpy_t_t<float, float>) kernel_cpy_t;
+
+template [[host_name("kernel_cpy_f32_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<float, float>;
+template [[host_name("kernel_cpy_f32_f16")]] kernel kernel_cpy_t kernel_cpy_t_t<float, half>;
+template [[host_name("kernel_cpy_f32_i32")]] kernel kernel_cpy_t kernel_cpy_t_t<float, int32_t>;
+template [[host_name("kernel_cpy_i32_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<int32_t, float>;
+template [[host_name("kernel_cpy_i32_i32")]] kernel kernel_cpy_t kernel_cpy_t_t<int32_t, int32_t>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_cpy_f32_bf16")]] kernel kernel_cpy_t kernel_cpy_t_t<float, bfloat>;
+#endif
+template [[host_name("kernel_cpy_f16_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<half, float>;
+template [[host_name("kernel_cpy_f16_f16")]] kernel kernel_cpy_t kernel_cpy_t_t<half, half>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_cpy_bf16_f32")]] kernel kernel_cpy_t kernel_cpy_t_t<bfloat, float>;
+template [[host_name("kernel_cpy_bf16_bf16")]] kernel kernel_cpy_t kernel_cpy_t_t<bfloat, bfloat>;
+#endif
+
+template<short QK,
+ typename block_q,
+ void (*quantize_func)(device const float *, device block_q &)>
+kernel void kernel_cpy_f32_q(
+ constant ggml_metal_kargs_cpy & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig[2];
+ const int32_t i02 = tgpig[1];
+ const int32_t i01 = ntg[1] == 1 ? tgpig[0]%args.ne01 : tgpig[0]*ntg[1] + tpitg.y;
+ const int32_t iw0 = ntg[1] == 1 ? tgpig[0]/args.ne01 : 0;
+
+ if (i01 >= args.ne01) {
+ return;
+ }
+
+ const int64_t n = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00;
+
+ const int32_t i3 = n / (args.ne2*args.ne1*args.ne0);
+ const int32_t i2 = (n - i3*args.ne2*args.ne1*args.ne0) / (args.ne1*args.ne0);
+ const int32_t i1 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0) / args.ne0;
+ const int32_t i0 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0 - i1*args.ne0)/QK;
+
+ device block_q * dst_data = (device block_q *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ for (int32_t i00 = iw0*ntg[0] + tpitg.x; i00 < args.nk0;) {
+ device const float * src = (device const float *)(src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + (i00*QK)*args.nb00);
+
+ quantize_func(src, dst_data[i00]);
+
+ break;
+ }
+}
+
+typedef decltype(kernel_cpy_f32_q<QK8_0, block_q8_0, quantize_q8_0>) cpy_f_q_t;
+
+template [[host_name("kernel_cpy_f32_q8_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK8_0, block_q8_0, quantize_q8_0>;
+template [[host_name("kernel_cpy_f32_q1_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK1_0, block_q1_0, quantize_q1_0>;
+template [[host_name("kernel_cpy_f32_q2_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK2_0, block_q2_0, quantize_q2_0>;
+template [[host_name("kernel_cpy_f32_q4_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK4_0, block_q4_0, quantize_q4_0>;
+template [[host_name("kernel_cpy_f32_q4_1")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK4_1, block_q4_1, quantize_q4_1>;
+template [[host_name("kernel_cpy_f32_q5_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK5_0, block_q5_0, quantize_q5_0>;
+template [[host_name("kernel_cpy_f32_q5_1")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK5_1, block_q5_1, quantize_q5_1>;
+template [[host_name("kernel_cpy_f32_iq4_nl")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK4_NL, block_iq4_nl, quantize_iq4_nl>;
+template [[host_name("kernel_cpy_f32_tq2_0")]] kernel cpy_f_q_t kernel_cpy_f32_q<QK_K, block_tq2_0, quantize_tq2_0>;
+
+template<typename T4x4, typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread T4x4 &)>
+kernel void kernel_cpy_q_f32(
+ constant ggml_metal_kargs_cpy & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig[2];
+ const int32_t i02 = tgpig[1];
+ const int32_t i01 = ntg[1] == 1 ? tgpig[0]%args.ne01 : tgpig[0]*ntg[1] + tpitg.y;
+ const int32_t iw0 = ntg[1] == 1 ? tgpig[0]/args.ne01 : 0;
+
+ if (i01 >= args.ne01) {
+ return;
+ }
+
+ const int64_t n = i03*args.ne02*args.ne01*args.ne00 + i02*args.ne01*args.ne00 + i01*args.ne00;
+
+ const int32_t i3 = n/(args.ne2*args.ne1*args.ne0);
+ const int32_t i2 = (n - i3*args.ne2*args.ne1*args.ne0)/(args.ne1*args.ne0);
+ const int32_t i1 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0)/args.ne0;
+ const int32_t i0 = (n - i3*args.ne2*args.ne1*args.ne0 - i2*args.ne1*args.ne0 - i1*args.ne0);
+
+ device const block_q * src_data = (device const block_q *)(src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
+ device T4x4 * dst_data = (device T4x4 *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ for (int32_t i00 = iw0*ntg[0] + tpitg.x; i00 < args.nk0;) {
+ T4x4 temp;
+ dequantize_func(src_data + i00/nl, i00%nl, temp);
+ dst_data[i00] = temp;
+
+ break;
+ }
+}
+
+typedef decltype(kernel_cpy_q_f32<float4x4, block_q4_0, 2, dequantize_q4_0>) cpy_q_f_t;
+
+template [[host_name("kernel_cpy_q1_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q1_0, 8, dequantize_q1_0>;
+template [[host_name("kernel_cpy_q2_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q2_0, 4, dequantize_q2_0>;
+template [[host_name("kernel_cpy_q4_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q4_0, 2, dequantize_q4_0>;
+template [[host_name("kernel_cpy_q4_1_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q4_1, 2, dequantize_q4_1>;
+template [[host_name("kernel_cpy_q5_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q5_0, 2, dequantize_q5_0>;
+template [[host_name("kernel_cpy_q5_1_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q5_1, 2, dequantize_q5_1>;
+template [[host_name("kernel_cpy_q8_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_q8_0, 2, dequantize_q8_0>;
+
+template [[host_name("kernel_cpy_tq2_0_f32")]] kernel cpy_q_f_t kernel_cpy_q_f32<float4x4, block_tq2_0, QK_NL, dequantize_tq2_0>;
+
+template [[host_name("kernel_cpy_q1_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q1_0, 8, dequantize_q1_0>;
+template [[host_name("kernel_cpy_q2_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q2_0, 4, dequantize_q2_0>;
+template [[host_name("kernel_cpy_q4_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q4_0, 2, dequantize_q4_0>;
+template [[host_name("kernel_cpy_q4_1_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q4_1, 2, dequantize_q4_1>;
+template [[host_name("kernel_cpy_q5_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q5_0, 2, dequantize_q5_0>;
+template [[host_name("kernel_cpy_q5_1_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q5_1, 2, dequantize_q5_1>;
+template [[host_name("kernel_cpy_q8_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_q8_0, 2, dequantize_q8_0>;
+
+template [[host_name("kernel_cpy_tq2_0_f16")]] kernel cpy_q_f_t kernel_cpy_q_f32<half4x4, block_tq2_0, QK_NL, dequantize_tq2_0>;
+
+template<typename T>
+kernel void kernel_concat(
+ constant ggml_metal_kargs_concat & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+
+ const int i3 = tgpig.z;
+ const int i2 = tgpig.y;
+ const int i1 = ntg.y == 1 ? tgpig.x : tgpig.x*ntg.y + tpitg.y;
+
+ if (i1 >= args.ne1) {
+ return;
+ }
+
+ int o[4] = {0, 0, 0, 0};
+ o[args.dim] = args.dim == 0 ? args.ne00 : (args.dim == 1 ? args.ne01 : (args.dim == 2 ? args.ne02 : args.ne03));
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ device const T * x;
+
+ if (i0 < args.ne00 && i1 < args.ne01 && i2 < args.ne02 && i3 < args.ne03) {
+ x = (device const T *)(src0 + (i3 )*args.nb03 + (i2 )*args.nb02 + (i1 )*args.nb01 + (i0 )*args.nb00);
+ } else {
+ x = (device const T *)(src1 + (i3 - o[3])*args.nb13 + (i2 - o[2])*args.nb12 + (i1 - o[1])*args.nb11 + (i0 - o[0])*args.nb10);
+ }
+
+ device T * y = (device T *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ *y = *x;
+ }
+}
+
+typedef decltype(kernel_concat<float>) kernel_concat_t;
+
+template [[host_name("kernel_concat_f32")]] kernel kernel_concat_t kernel_concat<float>;
+template [[host_name("kernel_concat_f16")]] kernel kernel_concat_t kernel_concat<half>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_concat_bf16")]] kernel kernel_concat_t kernel_concat<bfloat>;
+#endif
+template [[host_name("kernel_concat_i8")]] kernel kernel_concat_t kernel_concat<char>;
+template [[host_name("kernel_concat_i16")]] kernel kernel_concat_t kernel_concat<short>;
+template [[host_name("kernel_concat_i32")]] kernel kernel_concat_t kernel_concat<int>;
+template [[host_name("kernel_concat_i64")]] kernel kernel_concat_t kernel_concat<long>;
+
+template<typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread float4x4 &)>
+kernel void kernel_get_rows_q(
+ constant ggml_metal_kargs_get_rows & args,
+ device const void * src0,
+ device const void * src1,
+ device void * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort3 ntg [[threads_per_threadgroup]]) {
+ const int32_t iw0 = tgpig.x/args.ne10;
+ const int32_t i10 = tgpig.x%args.ne10;
+ const int32_t i11 = tgpig.y;
+ const int32_t i12 = tgpig.z;
+
+ const int32_t r = ((const device int32_t *) ((const device char *) src1 + i12*args.nb12 + i11*args.nb11 + i10*args.nb10))[0];
+
+ const int32_t i02 = i11;
+ const int32_t i03 = i12;
+
+ auto psrc = (device const block_q *) ((const device char *) src0 + i03*args.nb03 + i02*args.nb02 + r*args.nb01);
+ auto pdst = (device float4x4 *) (( device char *) dst + i12*args.nb3 + i11*args.nb2 + i10*args.nb1);
+
+ for (int ind = iw0*ntg.x + tiitg; ind < args.ne00t;) {
+ float4x4 temp;
+ dequantize_func(psrc + ind/nl, ind%nl, temp);
+ pdst[ind] = temp;
+
+ break;
+ }
+}
+
+template<typename T0, typename T>
+kernel void kernel_get_rows_f(
+ constant ggml_metal_kargs_get_rows & args,
+ device const void * src0,
+ device const void * src1,
+ device void * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort3 ntg [[threads_per_threadgroup]]) {
+ const int32_t iw0 = tgpig.x/args.ne10;
+ const int32_t i10 = tgpig.x%args.ne10;
+ const int32_t i11 = tgpig.y;
+ const int32_t i12 = tgpig.z;
+
+ const int32_t r = ((const device int32_t *) ((const device char *) src1 + i12*args.nb12 + i11*args.nb11 + i10*args.nb10))[0];
+
+ const int32_t i02 = i11;
+ const int32_t i03 = i12;
+
+ auto psrc = (const device T0 *) ((const device char *) src0 + i03*args.nb03 + i02*args.nb02 + r*args.nb01);
+ auto pdst = ( device T *) (( device char *) dst + i12*args.nb3 + i11*args.nb2 + i10*args.nb1);
+
+ for (int ind = iw0*ntg.x + tiitg; ind < args.ne00t;) {
+ pdst[ind] = psrc[ind];
+
+ break;
+ }
+}
+
+typedef decltype(kernel_get_rows_f<float, float>) get_rows_f_t;
+
+template [[host_name("kernel_get_rows_f32")]] kernel get_rows_f_t kernel_get_rows_f<float, float>;
+template [[host_name("kernel_get_rows_f16")]] kernel get_rows_f_t kernel_get_rows_f<half, float>;
+template [[host_name("kernel_get_rows_i32")]] kernel get_rows_f_t kernel_get_rows_f<int32_t, int32_t>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_get_rows_bf16")]] kernel get_rows_f_t kernel_get_rows_f<bfloat, float>;
+#endif
+
+typedef decltype(kernel_get_rows_q<block_q4_0, 2, dequantize_q4_0>) get_rows_q_t;
+
+template [[host_name("kernel_get_rows_q1_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q1_0, 8, dequantize_q1_0>;
+template [[host_name("kernel_get_rows_q2_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q2_0, 4, dequantize_q2_0>;
+template [[host_name("kernel_get_rows_q4_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q4_0, 2, dequantize_q4_0>;
+template [[host_name("kernel_get_rows_q4_1")]] kernel get_rows_q_t kernel_get_rows_q<block_q4_1, 2, dequantize_q4_1>;
+template [[host_name("kernel_get_rows_q5_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q5_0, 2, dequantize_q5_0>;
+template [[host_name("kernel_get_rows_q5_1")]] kernel get_rows_q_t kernel_get_rows_q<block_q5_1, 2, dequantize_q5_1>;
+template [[host_name("kernel_get_rows_q8_0")]] kernel get_rows_q_t kernel_get_rows_q<block_q8_0, 2, dequantize_q8_0>;
+template [[host_name("kernel_get_rows_mxfp4")]] kernel get_rows_q_t kernel_get_rows_q<block_mxfp4, 2, dequantize_mxfp4>;
+template [[host_name("kernel_get_rows_q2_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q2_K, QK_NL, dequantize_q2_K>;
+template [[host_name("kernel_get_rows_q3_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q3_K, QK_NL, dequantize_q3_K>;
+template [[host_name("kernel_get_rows_q4_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q4_K, QK_NL, dequantize_q4_K>;
+template [[host_name("kernel_get_rows_q5_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q5_K, QK_NL, dequantize_q5_K>;
+template [[host_name("kernel_get_rows_q6_K")]] kernel get_rows_q_t kernel_get_rows_q<block_q6_K, QK_NL, dequantize_q6_K>;
+template [[host_name("kernel_get_rows_iq2_xxs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq2_xxs, QK_NL, dequantize_iq2_xxs>;
+template [[host_name("kernel_get_rows_iq2_xs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq2_xs, QK_NL, dequantize_iq2_xs>;
+template [[host_name("kernel_get_rows_iq3_xxs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq3_xxs, QK_NL, dequantize_iq3_xxs>;
+template [[host_name("kernel_get_rows_iq3_s")]] kernel get_rows_q_t kernel_get_rows_q<block_iq3_s, QK_NL, dequantize_iq3_s>;
+template [[host_name("kernel_get_rows_iq2_s")]] kernel get_rows_q_t kernel_get_rows_q<block_iq2_s, QK_NL, dequantize_iq2_s>;
+template [[host_name("kernel_get_rows_iq1_s")]] kernel get_rows_q_t kernel_get_rows_q<block_iq1_s, QK_NL, dequantize_iq1_s>;
+template [[host_name("kernel_get_rows_iq1_m")]] kernel get_rows_q_t kernel_get_rows_q<block_iq1_m, QK_NL, dequantize_iq1_m>;
+template [[host_name("kernel_get_rows_iq4_nl")]] kernel get_rows_q_t kernel_get_rows_q<block_iq4_nl, 2, dequantize_iq4_nl>;
+template [[host_name("kernel_get_rows_iq4_xs")]] kernel get_rows_q_t kernel_get_rows_q<block_iq4_xs, QK_NL, dequantize_iq4_xs>;
+template [[host_name("kernel_get_rows_tq2_0")]] kernel get_rows_q_t kernel_get_rows_q<block_tq2_0, QK_NL, dequantize_tq2_0>;
+
+template<typename TS, typename TI, short QK, typename block_q, void (*quantize_func)(device const float *, device block_q &)>
+kernel void kernel_set_rows_q(
+ constant ggml_metal_kargs_set_rows & args,
+ device const void * src0,
+ device const void * src1,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint tiitg[[thread_index_in_threadgroup]],
+ uint3 tptg [[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig.z;
+ const int32_t i02 = tgpig.y;
+
+ const int32_t i12 = i03%args.ne12;
+ const int32_t i11 = i02%args.ne11;
+
+ const int32_t i01 = tgpig.x*tptg.y + tiitg/tptg.x;
+ if (i01 >= args.ne01) {
+ return;
+ }
+
+ const int32_t i10 = i01;
+ const TI i1 = ((const device TI *) ((const device char *) src1 + i10*args.nb10 + i11*args.nb11 + i12*args.nb12))[0];
+
+ device block_q * dst_row = ( device block_q *) (( device char *) dst + i1*args.nb1 + i02*args.nb2 + i03*args.nb3);
+ const device TS * src_row = (const device TS *) ((const device char *) src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
+
+ for (int ind = tiitg%tptg.x; ind < args.nk0; ind += tptg.x) {
+ quantize_func(src_row + QK*ind, dst_row[ind]);
+ }
+}
+
+template<typename TS, typename TI, typename block_q, void (*quantize_func)(device const float *, device block_q &)>
+kernel void kernel_set_rows_q32(
+ constant ggml_metal_kargs_set_rows & args,
+ device const void * src0,
+ device const void * src1,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint tiitg[[thread_index_in_threadgroup]],
+ uint3 tptg [[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig.z;
+ const int32_t i02 = tgpig.y;
+
+ const int32_t i12 = i03%args.ne12;
+ const int32_t i11 = i02%args.ne11;
+
+ const int32_t i01 = tgpig.x*tptg.y + tiitg/tptg.x;
+ if (i01 >= args.ne01) {
+ return;
+ }
+
+ const int32_t i10 = i01;
+ const TI i1 = ((const device TI *) ((const device char *) src1 + i10*args.nb10 + i11*args.nb11 + i12*args.nb12))[0];
+
+ device block_q * dst_row = ( device block_q *) (( device char *) dst + i1*args.nb1 + i02*args.nb2 + i03*args.nb3);
+ const device TS * src_row = (const device TS *) ((const device char *) src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
+
+ for (int ind = tiitg%tptg.x; ind < args.nk0; ind += tptg.x) {
+ quantize_func(src_row + 32*ind, dst_row[ind]);
+ }
+}
+
+template<typename TS, typename TI, typename TD>
+kernel void kernel_set_rows_f(
+ constant ggml_metal_kargs_set_rows & args,
+ device const void * src0,
+ device const void * src1,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint tiitg[[thread_index_in_threadgroup]],
+ uint3 tptg [[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig.z;
+ const int32_t i02 = tgpig.y;
+
+ const int32_t i12 = i03%args.ne12;
+ const int32_t i11 = i02%args.ne11;
+
+ const int32_t i01 = tgpig.x*tptg.y + tiitg/tptg.x;
+ if (i01 >= args.ne01) {
+ return;
+ }
+
+ const int32_t i10 = i01;
+ const TI i1 = ((const device TI *) ((const device char *) src1 + i10*args.nb10 + i11*args.nb11 + i12*args.nb12))[0];
+
+ device TD * dst_row = ( device TD *) (( device char *) dst + i1*args.nb1 + i02*args.nb2 + i03*args.nb3);
+ const device TS * src_row = (const device TS *) ((const device char *) src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
+
+ for (int ind = tiitg%tptg.x; ind < args.nk0; ind += tptg.x) {
+ dst_row[ind] = (TD) src_row[ind];
+ }
+}
+
+typedef decltype(kernel_set_rows_f<float, int64_t, float>) set_rows_f_t;
+
+template [[host_name("kernel_set_rows_f32_i64_f32")]] kernel set_rows_f_t kernel_set_rows_f<float, int64_t, float>;
+template [[host_name("kernel_set_rows_f32_i32_f32")]] kernel set_rows_f_t kernel_set_rows_f<float, int32_t, float>;
+template [[host_name("kernel_set_rows_f32_i64_f16")]] kernel set_rows_f_t kernel_set_rows_f<float, int64_t, half>;
+template [[host_name("kernel_set_rows_f32_i32_f16")]] kernel set_rows_f_t kernel_set_rows_f<float, int32_t, half>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_set_rows_f32_i64_bf16")]] kernel set_rows_f_t kernel_set_rows_f<float, int64_t, bfloat>;
+template [[host_name("kernel_set_rows_f32_i32_bf16")]] kernel set_rows_f_t kernel_set_rows_f<float, int32_t, bfloat>;
+#endif
+
+template [[host_name("kernel_set_rows_f16_i64_f16")]] kernel set_rows_f_t kernel_set_rows_f<half, int64_t, half>;
+template [[host_name("kernel_set_rows_f16_i32_f16")]] kernel set_rows_f_t kernel_set_rows_f<half, int32_t, half>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_set_rows_bf16_i64_bf16")]] kernel set_rows_f_t kernel_set_rows_f<bfloat, int64_t, bfloat>;
+template [[host_name("kernel_set_rows_bf16_i32_bf16")]] kernel set_rows_f_t kernel_set_rows_f<bfloat, int32_t, bfloat>;
+#endif
+
+typedef decltype(kernel_set_rows_q32<float, int64_t, block_q8_0, quantize_q8_0>) set_rows_q32_t;
+
+template [[host_name("kernel_set_rows_f32_i64_q8_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q8_0, quantize_q8_0>;
+template [[host_name("kernel_set_rows_f32_i32_q8_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q8_0, quantize_q8_0>;
+template [[host_name("kernel_set_rows_f32_i64_q4_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q4_0, quantize_q4_0>;
+template [[host_name("kernel_set_rows_f32_i32_q4_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q4_0, quantize_q4_0>;
+template [[host_name("kernel_set_rows_f32_i64_q4_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q4_1, quantize_q4_1>;
+template [[host_name("kernel_set_rows_f32_i32_q4_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q4_1, quantize_q4_1>;
+template [[host_name("kernel_set_rows_f32_i64_q5_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q5_0, quantize_q5_0>;
+template [[host_name("kernel_set_rows_f32_i32_q5_0")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q5_0, quantize_q5_0>;
+template [[host_name("kernel_set_rows_f32_i64_q5_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_q5_1, quantize_q5_1>;
+template [[host_name("kernel_set_rows_f32_i32_q5_1")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_q5_1, quantize_q5_1>;
+template [[host_name("kernel_set_rows_f32_i64_iq4_nl")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int64_t, block_iq4_nl, quantize_iq4_nl>;
+template [[host_name("kernel_set_rows_f32_i32_iq4_nl")]] kernel set_rows_q32_t kernel_set_rows_q32<float, int32_t, block_iq4_nl, quantize_iq4_nl>;
+
+typedef decltype(kernel_set_rows_q<float, int64_t, QK_K, block_tq2_0, quantize_tq2_0>) set_rows_qK_t;
+
+template [[host_name("kernel_set_rows_f32_i64_tq2_0")]] kernel set_rows_qK_t kernel_set_rows_q<float, int64_t, QK_K, block_tq2_0, quantize_tq2_0>;
+template [[host_name("kernel_set_rows_f32_i32_tq2_0")]] kernel set_rows_qK_t kernel_set_rows_q<float, int32_t, QK_K, block_tq2_0, quantize_tq2_0>;
+
diff --git a/ggml/src/ggml-metal/kernels/reduce.metal b/ggml/src/ggml-metal/kernels/reduce.metal
new file mode 100644
index 0000000000..0af9e4f6c2
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/reduce.metal
@@ -0,0 +1,228 @@
+#include "common.h"
+
+kernel void kernel_op_sum_f32(
+ constant ggml_metal_kargs_sum & args,
+ device const float * src0,
+ device float * dst,
+ threadgroup float * shmem_f32 [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+
+ if (args.np == 0) {
+ return;
+ }
+
+ // TODO: become function constant
+ const uint nsg = (ntg.x + 31) / 32;
+
+ float sumf = 0;
+
+ for (uint64_t i0 = tpitg.x; i0 < args.np; i0 += ntg.x) {
+ sumf += src0[i0];
+ }
+
+ sumf = simd_sum(sumf);
+
+ if (tiisg == 0) {
+ shmem_f32[sgitg] = sumf;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ float total = 0;
+
+ if (sgitg == 0) {
+ float v = 0;
+
+ if (tpitg.x < nsg) {
+ v = shmem_f32[tpitg.x];
+ }
+
+ total = simd_sum(v);
+
+ if (tpitg.x == 0) {
+ dst[0] = total;
+ }
+ }
+}
+
+constant short FC_sum_rows_op [[function_constant(FC_SUM_ROWS + 0)]];
+
+template <typename T0, typename T>
+kernel void kernel_sum_rows_impl(
+ constant ggml_metal_kargs_sum_rows & args,
+ device const char * src0,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+#define FC_OP FC_sum_rows_op
+
+ const int i3 = tgpig.z;
+ const int i2 = tgpig.y;
+ const int i1 = tgpig.x;
+
+ threadgroup T0 * shmem_t = (threadgroup T0 *) shmem;
+
+ if (sgitg == 0) {
+ shmem_t[tiisg] = 0.0f;
+ }
+
+ device const T0 * src_row = (device const T0 *) (src0 + i1*args.nb01 + i2*args.nb02 + i3*args.nb03);
+ device T * dst_row = (device T *) (dst + i1*args.nb1 + i2*args.nb2 + i3*args.nb3);
+
+ T0 sumf = T0(0.0f);
+
+ for (int64_t i0 = tpitg.x; i0 < args.ne00; i0 += ntg.x) {
+ sumf += src_row[i0];
+ }
+
+ sumf = simd_sum(sumf);
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ shmem_t[sgitg] = sumf;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ sumf = shmem_t[tiisg];
+ sumf = simd_sum(sumf);
+
+ if (tpitg.x == 0) {
+ if (FC_OP == OP_SUM_ROWS_NUM_MEAN) {
+ if (is_same<float4, T0>::value) {
+ dst_row[0] = sum(sumf) / (4*args.ne00);
+ } else {
+ dst_row[0] = sum(sumf) / args.ne00;
+ }
+ } else {
+ dst_row[0] = sum(sumf);
+ }
+ }
+
+#undef FC_OP
+}
+
+typedef decltype(kernel_sum_rows_impl<float, float>) kernel_sum_rows_t;
+
+template [[host_name("kernel_sum_rows_f32_f32")]] kernel kernel_sum_rows_t kernel_sum_rows_impl<float, float>;
+template [[host_name("kernel_sum_rows_f32_f32_4")]] kernel kernel_sum_rows_t kernel_sum_rows_impl<float4, float>;
+
+template<typename T>
+kernel void kernel_cumsum_blk(
+ constant ggml_metal_kargs_cumsum_blk & args,
+ device const char * src0,
+ device char * tmp,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int ib = tgpig[0]/args.ne01;
+
+ const int i00 = ib*ntg.x;
+ const int i01 = tgpig[0]%args.ne01;
+ const int i02 = tgpig[1];
+ const int i03 = tgpig[2];
+
+ device const float * src0_row = (device const float *) (src0 +
+ args.nb01*i01 +
+ args.nb02*i02 +
+ args.nb03*i03);
+
+ threadgroup float * shmem_f32 = (threadgroup float *) shmem;
+
+ float v = 0.0f;
+
+ if (i00 + tpitg.x < args.ne00) {
+ v = src0_row[i00 + tpitg.x];
+ }
+
+ float s = simd_prefix_inclusive_sum(v);
+
+ if (tiisg == N_SIMDWIDTH - 1) {
+ shmem_f32[sgitg] = s;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (sgitg == 0) {
+ shmem_f32[tiisg] = simd_prefix_exclusive_sum(shmem_f32[tiisg]);
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ s += shmem_f32[sgitg];
+
+ device float * dst_row = (device float *) dst +
+ args.ne00*i01 +
+ args.ne00*args.ne01*i02 +
+ args.ne00*args.ne01*args.ne02*i03;
+
+ if (i00 + tpitg.x < args.ne00) {
+ dst_row[i00 + tpitg.x] = s;
+ }
+
+ if (args.outb && tpitg.x == ntg.x - 1) {
+ device float * tmp_row = (device float *) tmp +
+ args.net0*i01 +
+ args.net0*args.net1*i02 +
+ args.net0*args.net1*args.net2*i03;
+
+ tmp_row[ib] = s;
+ }
+}
+
+typedef decltype(kernel_cumsum_blk<float>) kernel_cumsum_blk_t;
+
+template [[host_name("kernel_cumsum_blk_f32")]] kernel kernel_cumsum_blk_t kernel_cumsum_blk<float>;
+
+template<typename T>
+kernel void kernel_cumsum_add(
+ constant ggml_metal_kargs_cumsum_add & args,
+ device const char * tmp,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int ib = tgpig[0]/args.ne01;
+
+ if (ib == 0) {
+ return;
+ }
+
+ const int i00 = ib*ntg.x;
+ const int i01 = tgpig[0]%args.ne01;
+ const int i02 = tgpig[1];
+ const int i03 = tgpig[2];
+
+ device const float * tmp_row = (device const float *) (tmp +
+ args.nbt1*i01 +
+ args.nbt2*i02 +
+ args.nbt3*i03);
+
+ device float * dst_row = (device float *) dst +
+ args.ne00*i01 +
+ args.ne00*args.ne01*i02 +
+ args.ne00*args.ne01*args.ne02*i03;
+
+ if (i00 + tpitg.x < args.ne00) {
+ dst_row[i00 + tpitg.x] += tmp_row[ib - 1];
+ }
+}
+
+typedef decltype(kernel_cumsum_add<float>) kernel_cumsum_add_t;
+
+template [[host_name("kernel_cumsum_add_f32")]] kernel kernel_cumsum_add_t kernel_cumsum_add<float>;
diff --git a/ggml/src/ggml-metal/kernels/rope.metal b/ggml/src/ggml-metal/kernels/rope.metal
new file mode 100644
index 0000000000..401ceacb01
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/rope.metal
@@ -0,0 +1,333 @@
+#include "common.h"
+
+constant bool FC_rope_is_imrope [[function_constant(FC_ROPE + 0)]];
+constant bool FC_rope_is_back [[function_constant(FC_ROPE + 1)]];
+
+static float rope_yarn_ramp(const float low, const float high, const int i0) {
+ const float y = (i0 / 2 - low) / max(0.001f, high - low);
+ return 1.0f - min(1.0f, max(0.0f, y));
+}
+
+// YaRN algorithm based on LlamaYaRNScaledRotaryEmbedding.py from https://github.com/jquesnelle/yarn
+// MIT licensed. Copyright (c) 2023 Jeffrey Quesnelle and Bowen Peng.
+static void rope_yarn(
+ float theta_extrap, float freq_scale, float corr_dims[2], int i0, float ext_factor, float mscale,
+ thread float * cos_theta, thread float * sin_theta) {
+ // Get n-d rotational scaling corrected for extrapolation
+ float theta_interp = freq_scale * theta_extrap;
+ float theta = theta_interp;
+ if (ext_factor != 0.0f) {
+ float ramp_mix = rope_yarn_ramp(corr_dims[0], corr_dims[1], i0) * ext_factor;
+ theta = theta_interp * (1 - ramp_mix) + theta_extrap * ramp_mix;
+
+ // Get n-d magnitude scaling corrected for interpolation
+ mscale *= 1.0f + 0.1f * log(1.0f / freq_scale);
+ }
+ *cos_theta = cos(theta) * mscale;
+ *sin_theta = sin(theta) * mscale;
+ if (FC_rope_is_back) {
+ *sin_theta *= -1.0f;
+ }
+}
+
+// Apparently solving `n_rot = 2pi * x * base^((2 * max_pos_emb) / n_dims)` for x, we get
+// `corr_fac(n_rot) = n_dims * log(max_pos_emb / (n_rot * 2pi)) / (2 * log(base))`
+static float rope_yarn_corr_factor(int n_dims, int n_ctx_orig, float n_rot, float base) {
+ return n_dims * log(n_ctx_orig / (n_rot * 2 * M_PI_F)) / (2 * log(base));
+}
+
+static void rope_yarn_corr_dims(
+ int n_dims, int n_ctx_orig, float freq_base, float beta_fast, float beta_slow, float dims[2]
+) {
+ // start and end correction dims
+ dims[0] = max(0.0f, floor(rope_yarn_corr_factor(n_dims, n_ctx_orig, beta_fast, freq_base)));
+ dims[1] = min(n_dims - 1.0f, ceil(rope_yarn_corr_factor(n_dims, n_ctx_orig, beta_slow, freq_base)));
+}
+
+template<typename T>
+kernel void kernel_rope_norm(
+ constant ggml_metal_kargs_rope & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * src2,
+ device char * dst,
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort3 tptg [[threads_per_threadgroup]],
+ uint3 tgpig[[threadgroup_position_in_grid]]) {
+ const int i3 = tgpig[2];
+ const int i2 = tgpig[1];
+ const int i1 = tgpig[0];
+
+ float corr_dims[2];
+ rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
+
+ device const int32_t * pos = (device const int32_t *) src1;
+
+ const float theta_base = (float) pos[i2];
+ const float inv_ndims = -1.f/args.n_dims;
+
+ float cos_theta;
+ float sin_theta;
+
+ for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
+ if (i0 >= args.n_offs && i0 < args.n_offs + args.n_dims) {
+ const int iw = i0 - args.n_offs; // relative idx
+ const int ic = iw/2;
+
+ const float theta = theta_base * pow(args.freq_base, inv_ndims*iw);
+
+ const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
+
+ rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, iw, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
+
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ const float x0 = src[0];
+ const float x1 = src[1];
+
+ dst_data[0] = x0*cos_theta - x1*sin_theta;
+ dst_data[1] = x0*sin_theta + x1*cos_theta;
+ } else {
+ if (args.inplace) {
+ continue;
+ }
+
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ dst_data[0] = src[0];
+ dst_data[1] = src[1];
+ }
+ }
+}
+
+template<typename T>
+kernel void kernel_rope_neox(
+ constant ggml_metal_kargs_rope & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * src2,
+ device char * dst,
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort3 tptg [[threads_per_threadgroup]],
+ uint3 tgpig[[threadgroup_position_in_grid]]) {
+ const int i3 = tgpig[2];
+ const int i2 = tgpig[1];
+ const int i1 = tgpig[0];
+
+ float corr_dims[2];
+ rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
+
+ device const int32_t * pos = (device const int32_t *) src1;
+
+ const float theta_base = (float) pos[i2];
+ const float inv_ndims = -1.f/args.n_dims;
+
+ float cos_theta;
+ float sin_theta;
+
+ for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
+ if (i0 >= args.n_offs && i0 < args.n_offs + args.n_dims) {
+ const int iw = i0 - args.n_offs; // relative idx
+ const int ic = iw/2;
+
+ const float theta = theta_base * pow(args.freq_base, inv_ndims*iw);
+
+ const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
+
+ rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, iw, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
+
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + (args.n_offs + ic)*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + (args.n_offs + ic)*args.nb0);
+
+ const float x0 = src[0];
+ const float x1 = src[args.n_dims/2];
+
+ dst_data[0] = x0*cos_theta - x1*sin_theta;
+ dst_data[args.n_dims/2] = x0*sin_theta + x1*cos_theta;
+ } else {
+ if (args.inplace) {
+ continue;
+ }
+
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ dst_data[0] = src[0];
+ dst_data[1] = src[1];
+ }
+ }
+}
+
+template<typename T>
+kernel void kernel_rope_multi(
+ constant ggml_metal_kargs_rope & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * src2,
+ device char * dst,
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort3 tptg [[threads_per_threadgroup]],
+ uint3 tgpig[[threadgroup_position_in_grid]]) {
+ const int i3 = tgpig[2];
+ const int i2 = tgpig[1];
+ const int i1 = tgpig[0];
+
+ float corr_dims[2];
+ rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
+
+ device const int32_t * pos = (device const int32_t *) src1;
+
+ const float inv_ndims = -1.f/args.n_dims;
+
+ float cos_theta;
+ float sin_theta;
+
+ for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
+ if (i0 >= args.n_offs && i0 < args.n_offs + args.n_dims) {
+ const int iw = i0 - args.n_offs; // relative idx
+ const int ic = iw/2;
+
+ // mrope theta calculations
+ // note: the rest is the same as kernel_rope_neox
+ const int sect_dims = args.sect_0 + args.sect_1 + args.sect_2 + args.sect_3;
+ const int sec_w01 = args.sect_0 + args.sect_1; // end of section 1
+ const int sec_w012 = args.sect_0 + args.sect_1 + args.sect_2; // end of section 2
+ const int sector = ic % sect_dims;
+
+ float theta_base;
+ if (FC_rope_is_imrope) {
+ if (sector % 3 == 1 && sector < 3 * args.sect_1) { // h
+ theta_base = (float) pos[i2 + args.ne02 * 1];
+ } else if (sector % 3 == 2 && sector < 3 * args.sect_2) { // w
+ theta_base = (float) pos[i2 + args.ne02 * 2];
+ } else if (sector % 3 == 0 && sector < 3 * args.sect_0) { // t
+ theta_base = (float) pos[i2 + args.ne02 * 0];
+ } else { // e
+ theta_base = (float) pos[i2 + args.ne02 * 3];
+ }
+ } else {
+ if (sector < args.sect_0) {
+ theta_base = (float) pos[i2];
+ } else if (sector < sec_w01) {
+ theta_base = (float) pos[i2 + args.ne02 * 1];
+ } else if (sector < sec_w012) {
+ theta_base = (float) pos[i2 + args.ne02 * 2];
+ } else {
+ theta_base = (float) pos[i2 + args.ne02 * 3];
+ }
+ }
+ // end of mrope
+
+ const float theta = theta_base * pow(args.freq_base, inv_ndims*iw);
+
+ const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
+
+ rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, iw, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
+
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + (args.n_offs + ic)*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + (args.n_offs + ic)*args.nb0);
+
+ const float x0 = src[0];
+ const float x1 = src[args.n_dims/2];
+
+ dst_data[0] = x0*cos_theta - x1*sin_theta;
+ dst_data[args.n_dims/2] = x0*sin_theta + x1*cos_theta;
+ } else {
+ if (args.inplace) {
+ continue;
+ }
+
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ dst_data[0] = src[0];
+ dst_data[1] = src[1];
+ }
+ }
+}
+
+template<typename T>
+kernel void kernel_rope_vision(
+ constant ggml_metal_kargs_rope & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * src2,
+ device char * dst,
+ ushort tiitg[[thread_index_in_threadgroup]],
+ ushort3 tptg [[threads_per_threadgroup]],
+ uint3 tgpig[[threadgroup_position_in_grid]]) {
+ const int i3 = tgpig[2];
+ const int i2 = tgpig[1];
+ const int i1 = tgpig[0];
+
+ float corr_dims[2];
+ rope_yarn_corr_dims(args.n_dims, args.n_ctx_orig, args.freq_base, args.beta_fast, args.beta_slow, corr_dims);
+
+ device const int32_t * pos = (device const int32_t *) src1;
+
+ const float inv_ndims = -1.f/args.n_dims;
+
+ float cos_theta;
+ float sin_theta;
+
+ for (int i0 = 2*tiitg; i0 < args.ne0; i0 += 2*tptg.x) {
+ if (i0 < 2*args.n_dims) { // different from kernel_rope_multi
+ const int ic = i0/2;
+
+ // mrope theta calculations (only support 2 dimensions)
+ const int sect_dims = args.sect_0 + args.sect_1;
+ const int sector = ic % sect_dims;
+
+ float p;
+ float theta_base;
+ if (sector < args.sect_1) {
+ p = (float) sector;
+ theta_base = (float) pos[i2];
+ } else {
+ p = (float) sector - args.sect_0;
+ theta_base = (float) pos[i2 + args.ne02];
+ }
+
+ const float theta = theta_base * pow(args.freq_base, 2.0f * inv_ndims * p);
+ // end of mrope
+
+ const float freq_factor = args.src2 ? ((device const float *) src2)[ic] : 1.0f;
+
+ rope_yarn(theta/freq_factor, args.freq_scale, corr_dims, i0, args.ext_factor, args.attn_factor, &cos_theta, &sin_theta);
+
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + ic*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + ic*args.nb0);
+
+ const float x0 = src[0];
+ const float x1 = src[args.n_dims]; // different from kernel_rope_multi
+
+ dst_data[0] = x0*cos_theta - x1*sin_theta;
+ dst_data[args.n_dims] = x0*sin_theta + x1*cos_theta; // different from kernel_rope_multi
+ } else {
+ device const T * const src = (device T *)(src0 + i3*args.nb03 + i2*args.nb02 + i1*args.nb01 + i0*args.nb00);
+ device T * dst_data = (device T *)( dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ dst_data[0] = src[0];
+ dst_data[1] = src[1];
+ }
+ }
+}
+
+typedef decltype(kernel_rope_norm<float>) kernel_rope_norm_t;
+typedef decltype(kernel_rope_neox<float>) kernel_rope_neox_t;
+typedef decltype(kernel_rope_multi<float>) kernel_rope_multi_t;
+typedef decltype(kernel_rope_vision<float>) kernel_rope_vision_t;
+
+template [[host_name("kernel_rope_norm_f32")]] kernel kernel_rope_norm_t kernel_rope_norm<float>;
+template [[host_name("kernel_rope_norm_f16")]] kernel kernel_rope_norm_t kernel_rope_norm<half>;
+
+template [[host_name("kernel_rope_neox_f32")]] kernel kernel_rope_neox_t kernel_rope_neox<float>;
+template [[host_name("kernel_rope_neox_f16")]] kernel kernel_rope_neox_t kernel_rope_neox<half>;
+
+template [[host_name("kernel_rope_multi_f32")]] kernel kernel_rope_multi_t kernel_rope_multi<float>;
+template [[host_name("kernel_rope_multi_f16")]] kernel kernel_rope_multi_t kernel_rope_multi<half>;
+
+template [[host_name("kernel_rope_vision_f32")]] kernel kernel_rope_vision_t kernel_rope_vision<float>;
+template [[host_name("kernel_rope_vision_f16")]] kernel kernel_rope_vision_t kernel_rope_vision<half>;
diff --git a/ggml/src/ggml-metal/kernels/softmax.metal b/ggml/src/ggml-metal/kernels/softmax.metal
new file mode 100644
index 0000000000..f32fe29379
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/softmax.metal
@@ -0,0 +1,223 @@
+#include "common.h"
+
+template<typename T>
+kernel void kernel_soft_max(
+ constant ggml_metal_kargs_soft_max & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * src2,
+ device char * dst,
+ threadgroup float * buf [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint sgitg[[simdgroup_index_in_threadgroup]],
+ uint tiisg[[thread_index_in_simdgroup]],
+ uint3 tptg[[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig.z;
+ const int32_t i02 = tgpig.y;
+ const int32_t i01 = tgpig.x;
+
+ const int32_t i13 = i03%args.ne13;
+ const int32_t i12 = i02%args.ne12;
+ const int32_t i11 = i01;
+
+ device const float * psrc0 = (device const float *) (src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
+ device const T * pmask = src1 != src0 ? (device const T * ) (src1 + i11*args.nb11 + i12*args.nb12 + i13*args.nb13) : nullptr;
+ device const float * psrc2 = src2 != src0 ? (device const float *) (src2) : nullptr;
+ device float * pdst = (device float *) (dst + i01*args.nb1 + i02*args.nb2 + i03*args.nb3);
+
+ float slope = 1.0f;
+
+ // ALiBi
+ if (args.max_bias > 0.0f) {
+ const int32_t h = i02;
+
+ const float base = h < args.n_head_log2 ? args.m0 : args.m1;
+ const int exp = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
+
+ slope = pow(base, exp);
+ }
+
+ // parallel max
+ float lmax = psrc2 ? psrc2[i02] : -INFINITY;
+
+ for (int i00 = tpitg.x; i00 < args.ne00; i00 += tptg.x) {
+ lmax = MAX(lmax, psrc0[i00]*args.scale + (pmask ? slope*pmask[i00] : 0.0f));
+ }
+
+ // find the max value in the block
+ float max_val = simd_max(lmax);
+ if (tptg.x > N_SIMDWIDTH) {
+ if (sgitg == 0) {
+ buf[tiisg] = -INFINITY;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ buf[sgitg] = max_val;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ max_val = buf[tiisg];
+ max_val = simd_max(max_val);
+ }
+
+ // parallel sum
+ float lsum = 0.0f;
+ for (int i00 = tpitg.x; i00 < args.ne00; i00 += tptg.x) {
+ const float exp_psrc0 = exp((psrc0[i00]*args.scale + (pmask ? slope*pmask[i00] : 0.0f)) - max_val);
+ lsum += exp_psrc0;
+ pdst[i00] = exp_psrc0;
+ }
+
+ // This barrier fixes a failing test
+ // ref: https://github.com/ggml-org/ggml/pull/621#discussion_r1425156335
+ threadgroup_barrier(mem_flags::mem_none);
+
+ float sum = simd_sum(lsum);
+
+ if (tptg.x > N_SIMDWIDTH) {
+ if (sgitg == 0) {
+ buf[tiisg] = 0.0f;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ buf[sgitg] = sum;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ sum = buf[tiisg];
+ sum = simd_sum(sum);
+ }
+
+ if (psrc2) {
+ sum += exp(psrc2[i02] - max_val);
+ }
+
+ const float inv_sum = 1.0f/sum;
+
+ for (int i00 = tpitg.x; i00 < args.ne00; i00 += tptg.x) {
+ pdst[i00] *= inv_sum;
+ }
+}
+
+template<typename T>
+kernel void kernel_soft_max_4(
+ constant ggml_metal_kargs_soft_max & args,
+ device const char * src0,
+ device const char * src1,
+ device const char * src2,
+ device char * dst,
+ threadgroup float * buf [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint sgitg[[simdgroup_index_in_threadgroup]],
+ uint tiisg[[thread_index_in_simdgroup]],
+ uint3 tptg[[threads_per_threadgroup]]) {
+ const int32_t i03 = tgpig.z;
+ const int32_t i02 = tgpig.y;
+ const int32_t i01 = tgpig.x;
+
+ const int32_t i13 = i03%args.ne13;
+ const int32_t i12 = i02%args.ne12;
+ const int32_t i11 = i01;
+
+ device const float4 * psrc4 = (device const float4 *) (src0 + i01*args.nb01 + i02*args.nb02 + i03*args.nb03);
+ device const T * pmask = src1 != src0 ? (device const T * ) (src1 + i11*args.nb11 + i12*args.nb12 + i13*args.nb13) : nullptr;
+ device const float * psrc2 = src2 != src0 ? (device const float * ) (src2) : nullptr;
+ device float4 * pdst4 = (device float4 *) (dst + i01*args.nb1 + i02*args.nb2 + i03*args.nb3);
+
+ float slope = 1.0f;
+
+ if (args.max_bias > 0.0f) {
+ const int32_t h = i02;
+
+ const float base = h < args.n_head_log2 ? args.m0 : args.m1;
+ const int exp = h < args.n_head_log2 ? h + 1 : 2*(h - args.n_head_log2) + 1;
+
+ slope = pow(base, exp);
+ }
+
+ // parallel max
+ float4 lmax4 = psrc2 ? psrc2[i02] : -INFINITY;
+
+ for (int i00 = tpitg.x; i00 < args.ne00/4; i00 += tptg.x) {
+ lmax4 = fmax(lmax4, psrc4[i00]*args.scale + (float4)((pmask ? slope*pmask[i00] : 0.0f)));
+ }
+
+ const float lmax = MAX(MAX(lmax4[0], lmax4[1]), MAX(lmax4[2], lmax4[3]));
+
+ float max_val = simd_max(lmax);
+ if (tptg.x > N_SIMDWIDTH) {
+ if (sgitg == 0) {
+ buf[tiisg] = -INFINITY;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ buf[sgitg] = max_val;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ max_val = buf[tiisg];
+ max_val = simd_max(max_val);
+ }
+
+ // parallel sum
+ float4 lsum4 = 0.0f;
+ for (int i00 = tpitg.x; i00 < args.ne00/4; i00 += tptg.x) {
+ const float4 exp_psrc4 = exp((psrc4[i00]*args.scale + (float4)((pmask ? slope*pmask[i00] : 0.0f))) - max_val);
+ lsum4 += exp_psrc4;
+ pdst4[i00] = exp_psrc4;
+ }
+
+ const float lsum = lsum4[0] + lsum4[1] + lsum4[2] + lsum4[3];
+
+ // This barrier fixes a failing test
+ // ref: https://github.com/ggml-org/ggml/pull/621#discussion_r1425156335
+ threadgroup_barrier(mem_flags::mem_none);
+
+ float sum = simd_sum(lsum);
+
+ if (tptg.x > N_SIMDWIDTH) {
+ if (sgitg == 0) {
+ buf[tiisg] = 0.0f;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (tiisg == 0) {
+ buf[sgitg] = sum;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ sum = buf[tiisg];
+ sum = simd_sum(sum);
+ }
+
+ if (psrc2) {
+ sum += exp(psrc2[i02] - max_val);
+ }
+
+ const float inv_sum = 1.0f/sum;
+
+ for (int i00 = tpitg.x; i00 < args.ne00/4; i00 += tptg.x) {
+ pdst4[i00] *= inv_sum;
+ }
+}
+
+typedef decltype(kernel_soft_max<float>) kernel_soft_max_t;
+typedef decltype(kernel_soft_max_4<float4>) kernel_soft_max_4_t;
+
+template [[host_name("kernel_soft_max_f16")]] kernel kernel_soft_max_t kernel_soft_max<half>;
+template [[host_name("kernel_soft_max_f32")]] kernel kernel_soft_max_t kernel_soft_max<float>;
+template [[host_name("kernel_soft_max_f16_4")]] kernel kernel_soft_max_4_t kernel_soft_max_4<half4>;
+template [[host_name("kernel_soft_max_f32_4")]] kernel kernel_soft_max_4_t kernel_soft_max_4<float4>;
diff --git a/ggml/src/ggml-metal/kernels/solve_tri.metal b/ggml/src/ggml-metal/kernels/solve_tri.metal
new file mode 100644
index 0000000000..50f16facbf
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/solve_tri.metal
@@ -0,0 +1,75 @@
+#include "common.h"
+
+constant short FC_solve_tri_nsg [[function_constant(FC_SOLVE_TRI + 0)]];
+constant short FC_solve_tri_n [[function_constant(FC_SOLVE_TRI + 1)]];
+constant short FC_solve_tri_k [[function_constant(FC_SOLVE_TRI + 2)]];
+
+kernel void kernel_solve_tri_f32(
+ constant ggml_metal_kargs_solve_tri & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ threadgroup char * shmem [[threadgroup(0)]],
+ ushort3 tgpig[[threadgroup_position_in_grid]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ constexpr short NW = N_SIMDWIDTH;
+
+ const short NSG = FC_solve_tri_nsg;
+ const short N = FC_solve_tri_n;
+ const short K = FC_solve_tri_k;
+ const short NP = PAD2(N, NW);
+
+ const int32_t i03 = tgpig.z;
+ const int32_t i02 = tgpig.y;
+ const int32_t i01 = tgpig.x*NSG + sgitg;
+
+ threadgroup float * sh0 = (threadgroup float *) shmem;
+
+ device const float * src0_ptr = (device const float *)(src0 + i02 * args.nb02 + i03 * args.nb03) + sgitg*N;
+ device const float * src1_ptr = (device const float *)(src1 + i02 * args.nb12 + i03 * args.nb13) + i01;
+ device float * dst_ptr = (device float *)(dst + i02 * args.nb2 + i03 * args.nb3) + i01;
+
+ for (short rr = 0; rr < N; rr += NSG) {
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ {
+ threadgroup float * sh0_cur = sh0 + sgitg*NP;
+
+ for (short t = 0; t*NW < N; ++t) {
+ const short idx = t*NW + tiisg;
+ sh0_cur[idx] = src0_ptr[idx];
+ }
+
+ src0_ptr += NSG*N;
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ if (i01 >= args.ne10) {
+ continue;
+ }
+
+ for (short ir = 0; ir < NSG && rr + ir < N; ++ir) {
+ const short r = rr + ir;
+
+ threadgroup float * sh0_cur = sh0 + ir*NP;
+
+ float sum = 0.0f;
+
+ for (short t = 0; t*NW < r; ++t) {
+ const short idx = t*NW + tiisg;
+ sum += sh0_cur[idx] * dst_ptr[idx*K] * (idx < r);
+ }
+
+ sum = simd_sum(sum);
+
+ if (tiisg == 0) {
+ const float diag = sh0_cur[r];
+
+ dst_ptr[r*K] = (src1_ptr[r*K] - sum) / diag;
+ }
+ }
+ }
+}
diff --git a/ggml/src/ggml-metal/kernels/ssm.metal b/ggml/src/ggml-metal/kernels/ssm.metal
new file mode 100644
index 0000000000..be065c4fff
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/ssm.metal
@@ -0,0 +1,287 @@
+#include "common.h"
+
+// ref: ggml.c:ggml_compute_forward_ssm_conv_f32
+kernel void kernel_ssm_conv_f32_f32(
+ constant ggml_metal_kargs_ssm_conv & args,
+ device const void * src0,
+ device const void * src1,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+ const int64_t ir = tgpig.x;
+ const int64_t i2 = tgpig.y;
+ const int64_t i3 = tgpig.z;
+
+ const int64_t nc = args.ne10;
+ //const int64_t ncs = args.ne00;
+ //const int64_t nr = args.ne01;
+ //const int64_t n_t = args.ne1;
+ //const int64_t n_s = args.ne2;
+
+ device const float * s = (device const float *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
+ device const float * c = (device const float *) ((device const char *) src1 + ir*args.nb11);
+ device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
+
+ float sumf = 0.0f;
+
+ for (int64_t i0 = 0; i0 < nc; ++i0) {
+ sumf += s[i0] * c[i0];
+ }
+
+ x[0] = sumf;
+}
+
+kernel void kernel_ssm_conv_f32_f32_4(
+ constant ggml_metal_kargs_ssm_conv & args,
+ device const void * src0,
+ device const void * src1,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+ const int64_t ir = tgpig.x;
+ const int64_t i2 = tgpig.y;
+ const int64_t i3 = tgpig.z;
+
+ const int64_t nc = args.ne10;
+ //const int64_t ncs = args.ne00;
+ //const int64_t nr = args.ne01;
+ //const int64_t n_t = args.ne1;
+ //const int64_t n_s = args.ne2;
+
+ device const float4 * s = (device const float4 *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
+ device const float4 * c = (device const float4 *) ((device const char *) src1 + ir*args.nb11);
+ device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
+
+ float sumf = 0.0f;
+
+ for (int64_t i0 = 0; i0 < nc/4; ++i0) {
+ sumf += dot(s[i0], c[i0]);
+ }
+
+ x[0] = sumf;
+}
+
+constant short FC_ssm_conv_bs [[function_constant(FC_SSM_CONV + 0)]];
+
+// Batched version: each threadgroup processes multiple tokens for better efficiency
+// Thread layout: each thread handles one token, threadgroup covers BATCH_SIZE tokens
+kernel void kernel_ssm_conv_f32_f32_batched(
+ constant ggml_metal_kargs_ssm_conv & args,
+ device const void * src0,
+ device const void * src1,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+ // tgpig.x = row index (ir)
+ // tgpig.y = batch of tokens (i2_base / BATCH_SIZE)
+ // tgpig.z = sequence index (i3)
+ // tpitg.x = thread within batch (0..BATCH_SIZE-1)
+ const short BATCH_SIZE = FC_ssm_conv_bs;
+
+ const int64_t ir = tgpig.x;
+ const int64_t i2_base = tgpig.y * BATCH_SIZE;
+ const int64_t i3 = tgpig.z;
+ const int64_t i2_off = tpitg.x;
+ const int64_t i2 = i2_base + i2_off;
+
+ const int64_t nc = args.ne10; // conv kernel size (typically 4)
+ const int64_t n_t = args.ne1; // number of tokens
+
+ // Bounds check for partial batches at the end
+ if (i2 >= n_t) {
+ return;
+ }
+
+ // Load conv weights (shared across all tokens for this row)
+ device const float * c = (device const float *) ((device const char *) src1 + ir*args.nb11);
+
+ // Load source for this specific token
+ device const float * s = (device const float *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
+
+ // Output location for this token
+ device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
+
+ float sumf = 0.0f;
+ for (int64_t i0 = 0; i0 < nc; ++i0) {
+ sumf += s[i0] * c[i0];
+ }
+
+ x[0] = sumf;
+}
+
+kernel void kernel_ssm_conv_f32_f32_batched_4(
+ constant ggml_metal_kargs_ssm_conv & args,
+ device const void * src0,
+ device const void * src1,
+ device float * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+ // tgpig.x = row index (ir)
+ // tgpig.y = batch of tokens (i2_base / BATCH_SIZE)
+ // tgpig.z = sequence index (i3)
+ // tpitg.x = thread within batch (0..BATCH_SIZE-1)
+ const short BATCH_SIZE = FC_ssm_conv_bs;
+
+ const int64_t ir = tgpig.x;
+ const int64_t i2_base = tgpig.y * BATCH_SIZE;
+ const int64_t i3 = tgpig.z;
+ const int64_t i2_off = tpitg.x;
+ const int64_t i2 = i2_base + i2_off;
+
+ const int64_t nc = args.ne10; // conv kernel size (typically 4)
+ const int64_t n_t = args.ne1; // number of tokens
+
+ // Bounds check for partial batches at the end
+ if (i2 >= n_t) {
+ return;
+ }
+
+ // Load conv weights (shared across all tokens for this row)
+ device const float4 * c = (device const float4 *) ((device const char *) src1 + ir*args.nb11);
+
+ // Load source for this specific token
+ device const float4 * s = (device const float4 *) ((device const char *) src0 + ir*args.nb01 + i2*args.nb00 + i3*args.nb02);
+
+ // Output location for this token
+ device float * x = (device float *) ((device char *) dst + ir*args.nb0 + i2*args.nb1 + i3*args.nb2);
+
+ float sumf = 0.0f;
+ for (int64_t i0 = 0; i0 < nc/4; ++i0) {
+ sumf += dot(s[i0], c[i0]);
+ }
+
+ x[0] = sumf;
+}
+
+// ref: ggml.c:ggml_compute_forward_ssm_scan_f32, Mamba-2 part
+// Optimized version: reduces redundant memory loads by having one thread load shared values
+kernel void kernel_ssm_scan_f32(
+ constant ggml_metal_kargs_ssm_scan & args,
+ device const void * src0,
+ device const void * src1,
+ device const void * src2,
+ device const void * src3,
+ device const void * src4,
+ device const void * src5,
+ device const void * src6,
+ device float * dst,
+ threadgroup float * shared [[threadgroup(0)]],
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort sgitg[[simdgroup_index_in_threadgroup]],
+ ushort tiisg[[thread_index_in_simdgroup]],
+ ushort sgptg[[simdgroups_per_threadgroup]],
+ uint3 tgpg[[threadgroups_per_grid]]) {
+ constexpr short NW = N_SIMDWIDTH;
+
+ // Shared memory layout:
+ // [0..sgptg*NW-1]: partial sums for reduction (existing)
+ // [sgptg*NW..sgptg*NW+sgptg-1]: pre-computed x_dt values for each token in batch
+ // [sgptg*NW+sgptg..sgptg*NW+2*sgptg-1]: pre-computed dA values for each token in batch
+ threadgroup float * shared_sums = shared;
+ threadgroup float * shared_x_dt = shared + sgptg * NW;
+ threadgroup float * shared_dA = shared + sgptg * NW + sgptg;
+
+ shared_sums[tpitg.x] = 0.0f;
+
+ const int32_t i0 = tpitg.x;
+ const int32_t i1 = tgpig.x;
+ const int32_t ir = tgpig.y; // current head
+ const int32_t i3 = tgpig.z; // current seq
+
+ const int32_t nc = args.d_state;
+ const int32_t nr = args.d_inner;
+ const int32_t nh = args.n_head;
+ const int32_t ng = args.n_group;
+ const int32_t n_t = args.n_seq_tokens;
+ const int32_t n_s = args.n_seqs;
+ const int32_t K = args.K;
+
+ const int32_t s_off = args.s_off;
+
+ device const int32_t * ids = (device const int32_t *) src6;
+
+ device const float * s0_buff = (device const float *) ((device const char *) src0 + ir*args.nb02 + ids[i3]*args.nb03);
+ device float * s_buff = (device float *) ((device char *) dst + ir*args.nb02 + i3*args.nb03 + s_off);
+
+ const int32_t i = i0 + i1*nc;
+ const int32_t g = ir / (nh / ng); // repeat_interleave
+
+ float s0 = s0_buff[i];
+ float s = 0.0f;
+
+ device const float * A = (device const float *) ((device const char *) src3 + ir*args.nb31); // {ne30, nh}
+
+ const float A0 = A[i0%args.ne30];
+
+ device const float * x = (device const float *)((device const char *) src1 + i1*args.nb10 + ir*args.nb11 + i3*args.nb13); // {dim, nh, nt, ns}
+ device const float * dt = (device const float *)((device const char *) src2 + ir*args.nb20 + i3*args.nb22); // {nh, nt, ns}
+ device const float * B = (device const float *)((device const char *) src4 + g*args.nb41 + i3*args.nb43); // {d_state, ng, nt, ns}
+ device const float * C = (device const float *)((device const char *) src5 + g*args.nb51 + i3*args.nb53); // {d_state, ng, nt, ns}
+
+ device float * y = dst + (i1 + ir*(nr) + i3*(n_t*nh*nr)); // {dim, nh, nt, ns}
+
+ for (int i2 = 0; i2 < n_t; i2 += sgptg) {
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ // Pre-compute x_dt and dA for this batch of tokens
+ // Only first sgptg threads do the loads and expensive math
+ if (i0 < sgptg && i2 + i0 < n_t) {
+ // ns12 and ns21 are element strides (nb12/nb10, nb21/nb20)
+ device const float * x_t = x + i0 * args.ns12;
+ device const float * dt_t = dt + i0 * args.ns21;
+
+ const float dt0 = dt_t[0];
+ const float dtsp = dt0 <= 20.0f ? log(1.0f + exp(dt0)) : dt0;
+ shared_x_dt[i0] = x_t[0] * dtsp;
+ shared_dA[i0] = dtsp; // Store dtsp, compute exp(dtsp * A0) per-thread since A0 varies
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ for (int t = 0; t < sgptg && i2 + t < n_t; t++) {
+ const float x_dt = shared_x_dt[t];
+ const float dA = exp(shared_dA[t] * A0);
+
+ s = (s0 * dA) + (B[i0] * x_dt);
+
+ const float sumf = simd_sum(s * C[i0]);
+
+ if (tiisg == 0) {
+ shared_sums[t*NW + sgitg] = sumf;
+ }
+
+ // recurse
+ s0 = s;
+
+ const int32_t slot = n_t - 1 - (i2 + t);
+ if (slot > 0 && slot < K) {
+ device float * s_snapshot = (device float *) ((device char *) s_buff + (int64_t) slot*n_s*args.nb03);
+ s_snapshot[i] = s;
+ }
+
+ B += args.ns42;
+ C += args.ns52;
+ }
+
+ // Advance pointers for next batch
+ x += sgptg * args.ns12;
+ dt += sgptg * args.ns21;
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ const float sumf = simd_sum(shared_sums[sgitg*NW + tiisg]);
+
+ if (tiisg == 0 && i2 + sgitg < n_t) {
+ y[sgitg*nh*nr] = sumf;
+ }
+
+ y += sgptg*nh*nr;
+ }
+
+ s_buff[i] = s;
+}
diff --git a/ggml/src/ggml-metal/kernels/tri.metal b/ggml/src/ggml-metal/kernels/tri.metal
new file mode 100644
index 0000000000..862f78678b
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/tri.metal
@@ -0,0 +1,69 @@
+#include "common.h"
+
+template<uint32_t ttype>
+bool _ggml_vec_tri_cmp(const int i, const int r);
+
+template<>
+bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_LOWER */ 3>(const int i, const int r) {
+ return i < r;
+}
+
+template<>
+bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_LOWER_DIAG */ 2>(const int i, const int r) {
+ return i <= r;
+}
+
+template<>
+bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_UPPER */ 1>(const int i, const int r) {
+ return i > r;
+}
+
+template<>
+bool _ggml_vec_tri_cmp</* GGML_TRI_TYPE_UPPER_DIAG */ 0>(const int i, const int r) {
+ return i >= r;
+}
+
+template<typename T, int ttype>
+kernel void kernel_tri(
+ constant ggml_metal_kargs_tri & args,
+ device const char * src0,
+ device const char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+ const int i3 = tgpig.z;
+ const int i2 = tgpig.y;
+ const int i1 = tgpig.x;
+
+ if (i3 >= args.ne03 || i2 >= args.ne02 || i1 >= args.ne01) {
+ return;
+ }
+
+ device const T * src_row = (device const T *) ((device const char *) src0 + i1*args.nb01 + i2*args.nb02 + i3*args.nb03);
+ device T * dst_row = (device T *) ((device char *) dst + i1*args.nb1 + i2*args.nb2 + i3*args.nb3);
+
+ // Each thread is a single element of the row if ne00 < max threads per
+ // threadgroup, so this will loop once for each index that this thread is
+ // responsible for
+ for (int64_t i0 = tpitg.x; i0 < args.ne00; i0 += ntg.x) {
+ // Use the comparison as a mask for branchless
+ dst_row[i0] = static_cast<T>(_ggml_vec_tri_cmp<ttype>(i0, i1)) * src_row[i0];
+ }
+}
+
+typedef decltype(kernel_tri<float, 0>) kernel_tri_t;
+
+template [[host_name("kernel_tri_f32_0")]] kernel kernel_tri_t kernel_tri<float, 0>;
+template [[host_name("kernel_tri_f32_1")]] kernel kernel_tri_t kernel_tri<float, 1>;
+template [[host_name("kernel_tri_f32_2")]] kernel kernel_tri_t kernel_tri<float, 2>;
+template [[host_name("kernel_tri_f32_3")]] kernel kernel_tri_t kernel_tri<float, 3>;
+template [[host_name("kernel_tri_f16_0")]] kernel kernel_tri_t kernel_tri<half, 0>;
+template [[host_name("kernel_tri_f16_1")]] kernel kernel_tri_t kernel_tri<half, 1>;
+template [[host_name("kernel_tri_f16_2")]] kernel kernel_tri_t kernel_tri<half, 2>;
+template [[host_name("kernel_tri_f16_3")]] kernel kernel_tri_t kernel_tri<half, 3>;
+#if defined(GGML_METAL_HAS_BF16)
+template [[host_name("kernel_tri_bf16_0")]] kernel kernel_tri_t kernel_tri<bfloat, 0>;
+template [[host_name("kernel_tri_bf16_1")]] kernel kernel_tri_t kernel_tri<bfloat, 1>;
+template [[host_name("kernel_tri_bf16_2")]] kernel kernel_tri_t kernel_tri<bfloat, 2>;
+template [[host_name("kernel_tri_bf16_3")]] kernel kernel_tri_t kernel_tri<bfloat, 3>;
+#endif
diff --git a/ggml/src/ggml-metal/kernels/unary.metal b/ggml/src/ggml-metal/kernels/unary.metal
new file mode 100644
index 0000000000..39cad0cbee
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/unary.metal
@@ -0,0 +1,374 @@
+#include "common.h"
+
+constant short FC_unary_op [[function_constant(FC_UNARY + 0)]];
+constant bool FC_unary_cnt[[function_constant(FC_UNARY + 1)]];
+
+template <typename T0, typename T, typename TC>
+kernel void kernel_unary_impl(
+ constant ggml_metal_kargs_unary & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ ushort3 tpitg[[thread_position_in_threadgroup]],
+ ushort3 ntg[[threads_per_threadgroup]]) {
+#define FC_OP FC_unary_op
+#define FC_CNT FC_unary_cnt
+
+ device const T0 * src0_ptr;
+ device T * dst_ptr;
+
+ int i0;
+
+ if (FC_CNT) {
+ i0 = tgpig.x;
+
+ src0_ptr = (device const T0 *) (src0);
+ dst_ptr = (device T *) (dst);
+ } else {
+ const int i03 = tgpig.z;
+ const int i02 = tgpig.y;
+ const int k0 = tgpig.x/args.ne01;
+ const int i01 = tgpig.x - k0*args.ne01;
+
+ i0 = k0*ntg.x + tpitg.x;
+
+ src0_ptr = (device const T0 *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01);
+ dst_ptr = (device T *) (dst + i03*args.nb3 + i02*args.nb2 + i01*args.nb1 );
+ }
+
+ {
+ //threadgroup_barrier(mem_flags::mem_none);
+
+ if (!FC_CNT) {
+ if (i0 >= args.ne0) {
+ return;
+ }
+ }
+
+ const TC x = (TC) src0_ptr[i0];
+
+ if (FC_OP == OP_UNARY_NUM_SCALE) {
+ dst_ptr[i0] = (T) (args.scale * x + args.bias);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_FILL) {
+ dst_ptr[i0] = (T) args.val;
+ }
+
+ if (FC_OP == OP_UNARY_NUM_CLAMP) {
+ dst_ptr[i0] = (T) clamp(x, args.min, args.max);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_SQR) {
+ dst_ptr[i0] = (T) (x * x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_SQRT) {
+ dst_ptr[i0] = (T) sqrt(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_SIN) {
+ dst_ptr[i0] = (T) sin(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_COS) {
+ dst_ptr[i0] = (T) cos(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_LOG) {
+ dst_ptr[i0] = (T) log(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_LEAKY_RELU) {
+ dst_ptr[i0] = (T) (TC(x > 0)*x + TC(x <= 0)*(x * args.slope));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_TANH) {
+ dst_ptr[i0] = (T) precise::tanh(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_RELU) {
+ dst_ptr[i0] = (T) fmax(0, x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_SIGMOID) {
+ dst_ptr[i0] = (T) (1 / (1 + exp(-x)));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_GELU) {
+ dst_ptr[i0] = (T) (0.5*x*(1 + precise::tanh(SQRT_2_OVER_PI*x*(1 + GELU_COEF_A*x*x))));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_GELU_ERF) {
+ dst_ptr[i0] = (T) (0.5*x*(1 + erf_approx(SQRT_2_INV*x)));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_GELU_QUICK) {
+ dst_ptr[i0] = (T) (x * (1/(1 + exp(GELU_QUICK_COEF*x))));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_SILU) {
+ dst_ptr[i0] = (T) (x / (1 + exp(-x)));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_ELU) {
+ dst_ptr[i0] = (T) elu_approx(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_NEG) {
+ dst_ptr[i0] = (T) -x;
+ }
+
+ if (FC_OP == OP_UNARY_NUM_ABS) {
+ dst_ptr[i0] = (T) fabs(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_SGN) {
+ dst_ptr[i0] = T(x > 0) - T(x < 0);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_STEP) {
+ dst_ptr[i0] = T(x > 0);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_HARDSWISH) {
+ dst_ptr[i0] = (T) (x * fmax(0, fmin(1, x/6 + 0.5)));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_HARDSIGMOID) {
+ dst_ptr[i0] = (T) fmax(0, fmin(1, x/6 + 0.5));
+ }
+
+ if (FC_OP == OP_UNARY_NUM_EXP) {
+ dst_ptr[i0] = (T) exp(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_SOFTPLUS) {
+ dst_ptr[i0] = (T) select(log(1 + exp(x)), x, x > 20);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_EXPM1) {
+ // TODO: precise implementation
+ dst_ptr[i0] = (T) (exp(x) - 1);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_FLOOR) {
+ dst_ptr[i0] = (T) floor(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_CEIL) {
+ dst_ptr[i0] = (T) ceil(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_ROUND) {
+ dst_ptr[i0] = (T) round(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_TRUNC) {
+ dst_ptr[i0] = (T) trunc(x);
+ }
+
+ if (FC_OP == OP_UNARY_NUM_XIELU) {
+ const TC xi = x;
+ const TC gate = TC(xi > TC(0.0f));
+ const TC clamped = fmin(xi, TC(args.val));
+ const TC y_pos = TC(args.scale) * xi * xi + TC(args.bias) * xi;
+ const TC y_neg = (exp(clamped) - TC(1.0f) - xi) * TC(args.slope) + TC(args.bias) * xi;
+ dst_ptr[i0] = (T) (gate * y_pos + (TC(1.0f) - gate) * y_neg);
+ }
+ }
+
+#undef FC_OP
+#undef FC_CNT
+}
+
+typedef decltype(kernel_unary_impl<float, float, float>) kernel_unary_t;
+
+template [[host_name("kernel_unary_f32_f32")]] kernel kernel_unary_t kernel_unary_impl<float, float, float>;
+template [[host_name("kernel_unary_f32_f32_4")]] kernel kernel_unary_t kernel_unary_impl<float4, float4, float4>;
+template [[host_name("kernel_unary_f16_f16")]] kernel kernel_unary_t kernel_unary_impl<half, half, float>;
+template [[host_name("kernel_unary_f16_f16_4")]] kernel kernel_unary_t kernel_unary_impl<half4, half4, float4>;
+
+kernel void kernel_silu_back_f32(
+ constant ggml_metal_kargs_silu_back & args,
+ device const float * dy,
+ device const float * x,
+ device float * dx,
+ uint gid [[thread_position_in_grid]]) {
+ if (gid >= args.ne) {
+ return;
+ }
+
+ const float s = 1.0f / (1.0f + exp(-x[gid]));
+ dx[gid] = dy[gid] * s * (1.0f + x[gid] * (1.0f - s));
+}
+
+template<typename T>
+kernel void kernel_reglu(
+ constant ggml_metal_kargs_glu & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
+ device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
+ device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
+
+ for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
+ const float x0 = src0_row[i0];
+ const float x1 = src1_row[i0];
+
+ dst_row[i0] = (T)(x0*x1*(x0 > 0.0f));
+ }
+}
+
+typedef decltype(kernel_reglu<float>) kernel_reglu_t;
+
+template [[host_name("kernel_reglu_f32")]] kernel kernel_reglu_t kernel_reglu<float>;
+template [[host_name("kernel_reglu_f16")]] kernel kernel_reglu_t kernel_reglu<half>;
+
+template<typename T>
+kernel void kernel_geglu(
+ constant ggml_metal_kargs_glu & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
+ device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
+ device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
+
+ for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
+ const float x0 = src0_row[i0];
+ const float x1 = src1_row[i0];
+
+ const float gelu = 0.5f*x0*(1.0f + precise::tanh(SQRT_2_OVER_PI*x0*(1.0f + GELU_COEF_A*x0*x0)));
+
+ dst_row[i0] = (T)(gelu*x1);
+ }
+}
+
+typedef decltype(kernel_geglu<float>) kernel_geglu_t;
+
+template [[host_name("kernel_geglu_f32")]] kernel kernel_geglu_t kernel_geglu<float>;
+template [[host_name("kernel_geglu_f16")]] kernel kernel_geglu_t kernel_geglu<half>;
+
+template<typename T>
+kernel void kernel_swiglu(
+ constant ggml_metal_kargs_glu & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
+ device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
+ device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
+
+ for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
+ const float x0 = src0_row[i0];
+ const float x1 = src1_row[i0];
+
+ const float silu = x0 / (1.0f + exp(-x0));
+
+ dst_row[i0] = (T)(silu*x1);
+ }
+}
+
+typedef decltype(kernel_swiglu<float>) kernel_swiglu_t;
+
+template [[host_name("kernel_swiglu_f32")]] kernel kernel_swiglu_t kernel_swiglu<float>;
+template [[host_name("kernel_swiglu_f16")]] kernel kernel_swiglu_t kernel_swiglu<half>;
+
+template<typename T>
+kernel void kernel_swiglu_oai(
+ constant ggml_metal_kargs_glu & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
+ device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
+ device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
+
+ for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
+ float x0 = src0_row[i0];
+ float x1 = src1_row[i0];
+
+ x0 = min(x0, args.limit);
+ x1 = max(min(x1, args.limit), -args.limit);
+
+ float out_glu = x0 / (1.0f + exp(-x0 * args.alpha));
+ out_glu = out_glu * (1.0f + x1);
+
+ dst_row[i0] = (T)out_glu;
+ }
+}
+
+typedef decltype(kernel_swiglu_oai<float>) kernel_swiglu_oai_t;
+
+template [[host_name("kernel_swiglu_oai_f32")]] kernel kernel_swiglu_oai_t kernel_swiglu_oai<float>;
+template [[host_name("kernel_swiglu_oai_f16")]] kernel kernel_swiglu_oai_t kernel_swiglu_oai<half>;
+
+template<typename T>
+kernel void kernel_geglu_erf(
+ constant ggml_metal_kargs_glu & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
+ device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
+ device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
+
+ for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
+ const float x0 = src0_row[i0];
+ const float x1 = src1_row[i0];
+
+ const float gelu_erf = 0.5f*x0*(1.0f+erf_approx<float>(x0*SQRT_2_INV));
+
+ dst_row[i0] = (T)(gelu_erf*x1);
+ }
+}
+
+typedef decltype(kernel_geglu_erf<float>) kernel_geglu_erf_t;
+
+template [[host_name("kernel_geglu_erf_f32")]] kernel kernel_geglu_erf_t kernel_geglu_erf<float>;
+template [[host_name("kernel_geglu_erf_f16")]] kernel kernel_geglu_erf_t kernel_geglu_erf<half>;
+
+template<typename T>
+kernel void kernel_geglu_quick(
+ constant ggml_metal_kargs_glu & args,
+ device const char * src0,
+ device const char * src1,
+ device char * dst,
+ uint tgpig[[threadgroup_position_in_grid]],
+ uint tpitg[[thread_position_in_threadgroup]],
+ uint ntg[[threads_per_threadgroup]]) {
+ device const T * src0_row = (device const T *) ((device const char *) src0 + tgpig*args.nb01) + args.i00;
+ device const T * src1_row = (device const T *) ((device const char *) src1 + tgpig*args.nb11) + args.i10;
+ device T * dst_row = (device T *) ((device char *) dst + tgpig*args.nb1);
+
+ for (int i0 = tpitg; i0 < args.ne0; i0 += ntg) {
+ const float x0 = src0_row[i0];
+ const float x1 = src1_row[i0];
+
+ const float gelu_quick = x0*(1.0f/(1.0f+exp(GELU_QUICK_COEF*x0)));
+
+ dst_row[i0] = (T)(gelu_quick*x1);
+ }
+}
+
+typedef decltype(kernel_geglu_quick<float>) kernel_geglu_quick_t;
+
+template [[host_name("kernel_geglu_quick_f32")]] kernel kernel_geglu_quick_t kernel_geglu_quick<float>;
+template [[host_name("kernel_geglu_quick_f16")]] kernel kernel_geglu_quick_t kernel_geglu_quick<half>;
diff --git a/ggml/src/ggml-metal/kernels/upscale.metal b/ggml/src/ggml-metal/kernels/upscale.metal
new file mode 100644
index 0000000000..8bac13082a
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/upscale.metal
@@ -0,0 +1,179 @@
+#include "common.h"
+
+constant bool FC_upscale_aa [[function_constant(FC_UPSCALE + 0)]];
+
+kernel void kernel_upscale_nearest_f32(
+ constant ggml_metal_kargs_upscale & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int64_t i3 = tgpig.z;
+ const int64_t i2 = tgpig.y;
+ const int64_t i1 = tgpig.x;
+
+ const int64_t i03 = i3/args.sf3;
+ const int64_t i02 = i2/args.sf2;
+ const int64_t i01 = i1/args.sf1;
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ const int64_t i00 = i0/args.sf0;
+
+ device const float * src0_ptr = (device const float *) (src0 + i03*args.nb03 + i02*args.nb02 + i01*args.nb01 + i00*args.nb00);
+ device float * dst_ptr = (device float *) (dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1 + i0*args.nb0);
+
+ dst_ptr[0] = src0_ptr[0];
+ }
+}
+
+static inline float bilinear_tri(float x) {
+ return MAX(0.0f, 1.0f - fabs(x));
+}
+
+kernel void kernel_upscale_bilinear_f32(
+ constant ggml_metal_kargs_upscale & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int64_t i3 = tgpig.z;
+ const int64_t i2 = tgpig.y;
+ const int64_t i1 = tgpig.x;
+
+ const int64_t i03 = i3 / args.sf3;
+ const int64_t i02 = i2 / args.sf2;
+
+ const float f01 = ((float)i1 + args.poffs) / args.sf1 - args.poffs;
+ const int64_t i01 = MAX(0, MIN(args.ne01 - 1, (int64_t)floor(f01)));
+ const int64_t i01p = MAX(0, MIN(args.ne01 - 1, i01 + 1));
+ const float fd1 = MAX(0.0f, MIN(1.0f, f01 - (float)i01));
+
+ src0 += i03*args.nb03 + i02*args.nb02;
+
+ device float * dst_ptr = (device float *)(dst + i3*args.nb3 + i2*args.nb2 + i1*args.nb1);
+
+ if (FC_upscale_aa) {
+ const float support0 = MAX(1.0f, 1.0f / args.sf0);
+ const float invscale0 = 1.0f / support0;
+ const float support1 = MAX(1.0f, 1.0f / args.sf1);
+ const float invscale1 = 1.0f / support1;
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ const float f00 = ((float)i0 + args.poffs) / args.sf0 - args.poffs;
+
+ int64_t x_min = MAX((int64_t)0, (int64_t)floor(f00 - support0 + args.poffs));
+ int64_t x_max = MIN(args.ne00, (int64_t)ceil (f00 + support0 + args.poffs));
+
+ int64_t y_min = MAX((int64_t)0, (int64_t)floor(f01 - support1 + args.poffs));
+ int64_t y_max = MIN(args.ne01, (int64_t)ceil (f01 + support1 + args.poffs));
+
+ float sum = 0.0f;
+ float wsum = 0.0f;
+
+ for (int64_t sy = y_min; sy < y_max; ++sy) {
+ const float wy = MAX(0.0f, 1.0f - fabs((float)sy - f01) * invscale1);
+ for (int64_t sx = x_min; sx < x_max; ++sx) {
+ const float wx = MAX(0.0f, 1.0f - fabs((float)sx - f00) * invscale0);
+ const float w = wx * wy;
+ device const float * src_ptr = (device const float *)(src0 + sy*args.nb01 + sx*args.nb00);
+ sum += (*src_ptr) * w;
+ wsum += w;
+ }
+ }
+
+ const float v = (wsum > 0.0f) ? (sum / wsum) : 0.0f;
+ dst_ptr[i0] = v;
+ }
+ } else {
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ const float f00 = ((float)i0 + args.poffs) / args.sf0 - args.poffs;
+ const int64_t i00 = MAX(0, MIN(args.ne00 - 1, (int64_t)floor(f00)));
+ const int64_t i00p = MAX(0, MIN(args.ne00 - 1, i00 + 1));
+ const float fd0 = MAX(0.0f, MIN(1.0f, f00 - (float)i00));
+
+ device const float * src00 = (device const float *)(src0 + i01*args.nb01 + i00*args.nb00);
+ device const float * src10 = (device const float *)(src0 + i01*args.nb01 + i00p*args.nb00);
+ device const float * src01 = (device const float *)(src0 + i01p*args.nb01 + i00*args.nb00);
+ device const float * src11 = (device const float *)(src0 + i01p*args.nb01 + i00p*args.nb00);
+
+ const float v =
+ (*src00) * (1.0f - fd0) * (1.0f - fd1) +
+ (*src10) * fd0 * (1.0f - fd1) +
+ (*src01) * (1.0f - fd0) * fd1 +
+ (*src11) * fd0 * fd1;
+
+ dst_ptr[i0] = v;
+ }
+ }
+}
+
+static inline float bicubic_weight1(float x) {
+ const float a = -0.75f;
+ return ((a + 2) * x - (a + 3)) * x * x + 1;
+}
+
+static inline float bicubic_weight2(float x) {
+ const float a = -0.75f;
+ return ((a * x - 5 * a) * x + 8 * a) * x - 4 * a;
+}
+
+kernel void kernel_upscale_bicubic_f32(
+ constant ggml_metal_kargs_upscale & args,
+ device const char * src0,
+ device char * dst,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const int64_t i3 = tgpig.z;
+ const int64_t i2 = tgpig.y;
+ const int64_t i1 = tgpig.x;
+
+ const int64_t i03 = i3 / args.sf3;
+ const int64_t i02 = i2 / args.sf2;
+
+ const float f01 = ((float)i1 + args.poffs) / args.sf1 - args.poffs;
+ const int64_t i01 = (int64_t)floor(f01);
+ const float fd1 = f01 - (float)i01;
+
+ const float w_y0 = bicubic_weight2(fd1 + 1.0f);
+ const float w_y1 = bicubic_weight1(fd1);
+ const float w_y2 = bicubic_weight1(1.0f - fd1);
+ const float w_y3 = bicubic_weight2(2.0f - fd1);
+
+ const device char * src_slice = src0 + i03 * args.nb03 + i02 * args.nb02;
+
+ device float * dst_ptr = (device float *)(dst + i3 * args.nb3 + i2 * args.nb2 + i1 * args.nb1);
+
+ for (int i0 = tpitg.x; i0 < args.ne0; i0 += ntg.x) {
+ const float f00 = ((float)i0 + args.poffs) / args.sf0 - args.poffs;
+ const int64_t i00 = (int64_t)floor(f00);
+ const float fd0 = f00 - (float)i00;
+
+ const float w_x0 = bicubic_weight2(fd0 + 1.0f);
+ const float w_x1 = bicubic_weight1(fd0);
+ const float w_x2 = bicubic_weight1(1.0f - fd0);
+ const float w_x3 = bicubic_weight2(2.0f - fd0);
+
+ float sum = 0.0f;
+
+ for (int dy = -1; dy <= 2; ++dy) {
+ const int64_t iy = MAX(0, MIN(args.ne01 - 1, i01 + dy));
+ const float wy = (dy == -1) ? w_y0 : (dy == 0) ? w_y1 : (dy == 1) ? w_y2 : w_y3;
+
+ for (int dx = -1; dx <= 2; ++dx) {
+ const int64_t ix = MAX(0, MIN(args.ne00 - 1, i00 + dx));
+ const float wx = (dx == -1) ? w_x0 : (dx == 0) ? w_x1 : (dx == 1) ? w_x2 : w_x3;
+
+ device const float * src_ptr = (device const float *)(src_slice + iy * args.nb01 + ix * args.nb00);
+ sum += (*src_ptr) * wx * wy;
+ }
+ }
+
+ dst_ptr[i0] = sum;
+ }
+}
diff --git a/ggml/src/ggml-metal/kernels/wkv.metal b/ggml/src/ggml-metal/kernels/wkv.metal
new file mode 100644
index 0000000000..8767581c69
--- /dev/null
+++ b/ggml/src/ggml-metal/kernels/wkv.metal
@@ -0,0 +1,179 @@
+#include "common.h"
+
+kernel void kernel_rwkv_wkv6_f32(
+ device const float * k,
+ device const float * v,
+ device const float * r,
+ device const float * tf,
+ device const float * td,
+ device const float * state_in,
+ device float * dst,
+ constant uint & B,
+ constant uint & T,
+ constant uint & C,
+ constant uint & H,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const uint head_size = 64; // TODO: support head_size = 128
+ const uint batch_id = tgpig.x / H;
+ const uint head_id = tgpig.x % H;
+ const uint tid = tpitg.x;
+
+ if (batch_id >= B || head_id >= H) {
+ return;
+ }
+
+ const uint state_size = C * head_size;
+ const uint n_seq_tokens = T / B;
+
+ threadgroup float _k[head_size];
+ threadgroup float _r[head_size];
+ threadgroup float _tf[head_size];
+ threadgroup float _td[head_size];
+
+ float state[head_size];
+
+ for (uint i = 0; i < head_size; i++) {
+ state[i] = state_in[batch_id * state_size + head_id * head_size * head_size
+ + i * head_size + tid];
+ }
+
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ _tf[tid] = tf[head_id * head_size + tid];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ const uint start_t = batch_id * n_seq_tokens * C + head_id * head_size + tid;
+ const uint end_t = (batch_id + 1) * n_seq_tokens * C + head_id * head_size + tid;
+
+ for (uint t = start_t; t < end_t; t += C) {
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ _k[tid] = k[t];
+ _r[tid] = r[t];
+ _td[tid] = td[t];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ const float v_val = v[t];
+ float y = 0.0;
+
+ for (uint j = 0; j < head_size; j += 4) {
+ float4 k_vec = float4(_k[j], _k[j+1], _k[j+2], _k[j+3]);
+ float4 r_vec = float4(_r[j], _r[j+1], _r[j+2], _r[j+3]);
+ float4 tf_vec = float4(_tf[j], _tf[j+1], _tf[j+2], _tf[j+3]);
+ float4 td_vec = float4(_td[j], _td[j+1], _td[j+2], _td[j+3]);
+ float4 s_vec = float4(state[j], state[j+1], state[j+2], state[j+3]);
+
+ float4 kv = k_vec * v_val;
+
+ float4 temp = tf_vec * kv + s_vec;
+ y += dot(r_vec, temp);
+
+ s_vec = s_vec * td_vec + kv;
+ state[j] = s_vec[0];
+ state[j+1] = s_vec[1];
+ state[j+2] = s_vec[2];
+ state[j+3] = s_vec[3];
+ }
+
+ dst[t] = y;
+ }
+
+ for (uint i = 0; i < head_size; i++) {
+ dst[T * C + batch_id * state_size + head_id * head_size * head_size
+ + i * head_size + tid] = state[i];
+ }
+}
+
+kernel void kernel_rwkv_wkv7_f32(
+ device const float * r,
+ device const float * w,
+ device const float * k,
+ device const float * v,
+ device const float * a,
+ device const float * b,
+ device const float * state_in,
+ device float * dst,
+ constant uint & B,
+ constant uint & T,
+ constant uint & C,
+ constant uint & H,
+ uint3 tgpig[[threadgroup_position_in_grid]],
+ uint3 tpitg[[thread_position_in_threadgroup]],
+ uint3 ntg[[threads_per_threadgroup]]) {
+
+ const uint head_size = 64; // TODO: support head_size = 128
+ const uint batch_id = tgpig.x / H;
+ const uint head_id = tgpig.x % H;
+ const uint tid = tpitg.x;
+
+ if (batch_id >= B || head_id >= H) {
+ return;
+ }
+
+ const uint state_size = C * head_size;
+ const uint n_seq_tokens = T / B;
+
+ threadgroup float _r[head_size];
+ threadgroup float _w[head_size];
+ threadgroup float _k[head_size];
+ threadgroup float _a[head_size];
+ threadgroup float _b[head_size];
+
+ float state[head_size];
+
+ for (uint i = 0; i < head_size; i++) {
+ state[i] = state_in[batch_id * state_size + head_id * head_size * head_size
+ + tid * head_size + i];
+ }
+
+ const uint start_t = batch_id * n_seq_tokens * C + head_id * head_size + tid;
+ const uint end_t = (batch_id + 1) * n_seq_tokens * C + head_id * head_size + tid;
+
+ for (uint t = start_t; t < end_t; t += C) {
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+ _r[tid] = r[t];
+ _w[tid] = w[t];
+ _k[tid] = k[t];
+ _a[tid] = a[t];
+ _b[tid] = b[t];
+ threadgroup_barrier(mem_flags::mem_threadgroup);
+
+ const float v_val = v[t];
+ float y = 0.0, sa = 0.0;
+
+ float4 sa_vec(0.0);
+
+ for (uint j = 0; j < head_size; j += 4) {
+ float4 a_vec = float4(_a[j], _a[j+1], _a[j+2], _a[j+3]);
+ float4 s_vec = float4(state[j], state[j+1], state[j+2], state[j+3]);
+ sa_vec += a_vec * s_vec;
+ }
+ sa = sa_vec[0] + sa_vec[1] + sa_vec[2] + sa_vec[3];
+
+ for (uint j = 0; j < head_size; j += 4) {
+ float4 r_vec = float4(_r[j], _r[j+1], _r[j+2], _r[j+3]);
+ float4 w_vec = float4(_w[j], _w[j+1], _w[j+2], _w[j+3]);
+ float4 k_vec = float4(_k[j], _k[j+1], _k[j+2], _k[j+3]);
+ float4 b_vec = float4(_b[j], _b[j+1], _b[j+2], _b[j+3]);
+ float4 s_vec = float4(state[j], state[j+1], state[j+2], state[j+3]);
+
+ float4 kv = k_vec * v_val;
+
+ s_vec = s_vec * w_vec + kv + sa * b_vec;
+ y += dot(s_vec, r_vec);
+
+ state[j] = s_vec[0];
+ state[j+1] = s_vec[1];
+ state[j+2] = s_vec[2];
+ state[j+3] = s_vec[3];
+ }
+
+ dst[t] = y;
+ }
+
+ for (uint i = 0; i < head_size; i++) {
+ dst[T * C + batch_id * state_size + head_id * head_size * head_size
+ + tid * head_size + i] = state[i];
+ }
+}