Skip to content

Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.

C++

metal: the dsv41 host side compiles - ds4_model.cu goes from 20 errors to 0

Niko1221a798d03858b7
diff --git a/metal/include/metal_compat/cublas_v2.h b/metal/include/metal_compat/cublas_v2.h
index dfc87e74a..50a966fd3 100644
--- a/metal/include/metal_compat/cublas_v2.h
+++ b/metal/include/metal_compat/cublas_v2.h
@@ -19,6 +19,7 @@ enum cublasStatus_t { CUBLAS_STATUS_SUCCESS = 0, CUBLAS_STATUS_NOT_INITIALIZED =
CUBLAS_STATUS_INVALID_VALUE = 7, CUBLAS_STATUS_ARCH_MISMATCH = 8, CUBLAS_STATUS_NOT_SUPPORTED = 14 };
enum cublasOperation_t { CUBLAS_OP_N = 0, CUBLAS_OP_T = 1, CUBLAS_OP_C = 2 };
enum cublasComputeType_t { CUBLAS_COMPUTE_16F = 64, CUBLAS_COMPUTE_32F = 68, CUBLAS_COMPUTE_32F_FAST_16F = 74 };
+enum cublasGemmAlgo_t { CUBLAS_GEMM_DEFAULT = -1, CUBLAS_GEMM_DEFAULT_TENSOR_OP = 99 };
enum cudaDataType_t { CUDA_R_32F = 0, CUDA_R_16F = 2, CUDA_R_16BF = 14, CUDA_R_8F_E4M3 = 28 };
static inline cublasStatus_t cublasCreate(cublasHandle_t* h) {
@@ -43,9 +44,9 @@ static inline cublasStatus_t cublasSgemm(cublasHandle_t, cublasOperation_t, cubl
strata_metal_dense_gemm_missing("cublasSgemm");
}
-// the mixed-precision form: the engine passes a compute type and a scale; on Metal the kernels read A/B by their own type
-static inline cublasStatus_t cublasGemmEx(cublasHandle_t, cublasOperation_t, cublasOperation_t, int, int, int,
- const void*, const void*, cudaDataType_t, int, const void*, cudaDataType_t, int,
- const void*, void*, cudaDataType_t, cublasComputeType_t, int) {
+// the mixed-precision form: any argument list, because the engine passes cuBLAS's full 19-argument spelling; the body
+// names kernels/dense_mm.metal, which is where the real implementation goes.
+template <typename... A>
+static inline cublasStatus_t cublasGemmEx(A...) {
strata_metal_dense_gemm_missing("cublasGemmEx");
}
diff --git a/metal/include/metal_compat/cuda_runtime.h b/metal/include/metal_compat/cuda_runtime.h
index dca9b6a78..04dd19494 100644
--- a/metal/include/metal_compat/cuda_runtime.h
+++ b/metal/include/metal_compat/cuda_runtime.h
@@ -1,3 +1,31 @@
+// The CUDA function-space qualifiers and the few launch/kernel spellings the shared sources use. On Metal they say
+// nothing: a "kernel" here is an MSL function compiled at run time and dispatched by the host runtime, so the qualifiers
+// are empty and __forceinline__ is plain inline. Part of the metal_compat layer (M5 Max, 2026-10-23).
+#ifndef __host__
+#define __host__
+#endif
+#ifndef __device__
+#define __device__
+#endif
+#ifndef __global__
+#define __global__
+#endif
+#ifndef __shared__
+#define __shared__
+#endif
+#ifndef __constant__
+#define __constant__
+#endif
+#ifndef __forceinline__
+#define __forceinline__ inline
+#endif
+#ifndef __launch_bounds__
+#define __launch_bounds__(...)
+#endif
+#ifndef __restrict__
+#define __restrict__
+#endif
+
// metal/include/metal_compat/cuda_runtime.h - the CUDA runtime names the host files use, on Metal.
//
// The HIP build's approach (include/strata/hip_compat/cuda_runtime.h maps every cudaX the host code calls to hipX),
@@ -141,3 +169,10 @@ inline cudaError_t cudaGraphLaunch(cudaGraphExec_t x, cudaStream_t s) { return s
inline cudaError_t cudaGraphUpload(cudaGraphExec_t x, cudaStream_t s) { return strata::gpu::graph_upload(x, s); }
inline cudaError_t cudaGraphDestroy(cudaGraph_t g) { return strata::gpu::graph_destroy(g); }
inline cudaError_t cudaGraphExecDestroy(cudaGraphExec_t x) { return strata::gpu::graph_exec_destroy(x); }
+
+// cudaMemcpy2D(dst, dpitch, src, spitch, width, height, kind): a strided copy, used by the vision tower's patch rows and
+// the Engram tables' column slices. The Metal runtime has no pitched copy yet, so this reports failure rather than
+// copying nothing silently.
+inline strata::gpu::Error cudaMemcpy2D(void*, size_t, const void*, size_t, size_t, size_t, strata::gpu::CopyKind) {
+ return strata::gpu::kErrorUnsupported; // cudaMemcpy2D_unimplemented_marker
+}
diff --git a/src/dsv41/ds4_os_apple.h b/src/dsv41/ds4_os_apple.h
index 37eb664ef..c670a539f 100644
--- a/src/dsv41/ds4_os_apple.h
+++ b/src/dsv41/ds4_os_apple.h
@@ -46,6 +46,10 @@ static inline long syscall(long, ...) { return -1; }
// CPU affinity: macOS schedules on its own; the engine only uses this to pin the token thread. No-op, reported as success.
typedef struct { unsigned long __bits[16]; } cpu_set_t;
+#ifndef CPU_ZERO
+#define CPU_ZERO(set) ((void) (set))
+#define CPU_SET(cpu, set) ((void) (set))
+#endif
static inline int sched_setaffinity(pid_t, size_t, const cpu_set_t*) { return 0; }
static inline int pthread_setaffinity_np(pthread_t, size_t, const cpu_set_t*) { return 0; }