Skip to content

LLM inference in C/C++

C++ ggml

SYCL: Refactor GGML_SYCL_ENABLE_MKL_FA to global var (#26863)

Ozymandias_EBONf9f09f02cc44
diff --git a/ggml/src/ggml-sycl/common.hpp b/ggml/src/ggml-sycl/common.hpp
index b5f75ca548..9f2a27b18e 100644
--- a/ggml/src/ggml-sycl/common.hpp
+++ b/ggml/src/ggml-sycl/common.hpp
@@ -68,6 +68,7 @@ extern int g_ggml_sycl_enable_flash_attention;
extern int g_ggml_sycl_dev2dev_memcpy;
extern int g_ggml_sycl_fa_onednn;
extern int g_ggml_sycl_fa_onednn_max_kv;
+extern int g_ggml_sycl_enable_mkl_fa;
#define CHECK_TRY_ERROR(expr) \
diff --git a/ggml/src/ggml-sycl/fattn.cpp b/ggml/src/ggml-sycl/fattn.cpp
index b73e6d46ff..394cda593f 100644
--- a/ggml/src/ggml-sycl/fattn.cpp
+++ b/ggml/src/ggml-sycl/fattn.cpp
@@ -146,14 +146,13 @@ static best_fattn_kernel ggml_sycl_get_best_fattn_kernel(const int device, const
// Set GGML_SYCL_ENABLE_MKL_FA=0 to force TILE/VEC path for A/B testing.
// Example: GGML_SYCL_ENABLE_MKL_FA=0 llama-cli -m model.gguf -fa -ngl 99 ...
// Note: MKL GEMM calls are incompatible with SYCL graph capture replay.
- static int mkl_enable = ggml_sycl_get_env("GGML_SYCL_ENABLE_MKL_FA", 1);
// MKL is validated for the mainstream GQA envelope: grouped-query
// (gqa_ratio >= 2), head_dim a multiple of 64 in [64,512] with matching
// K/V head size, mask, no sinks/ALiBi/softcap. Gemma's global layers use
// head_dim 512, so the cap must include it. Head sizes not a multiple of
// 64 (72/80/96), MHA (gqa_ratio == 1), and MLA (DKQ != DV, e.g. 576/512)
// fall through to TILE/VEC; see follow-up work.
- if (mkl_enable == 1 && mask && !sinks && gqa_ratio >= 2 &&
+ if (g_ggml_sycl_enable_mkl_fa == 1 && mask && !sinks && gqa_ratio >= 2 &&
Q->ne[0] >= 64 && Q->ne[0] <= 512 && Q->ne[0] % 64 == 0 &&
Q->ne[0] == V->ne[0] &&
Q->ne[1] >= 32 && K->ne[1] >= 1024 &&
diff --git a/ggml/src/ggml-sycl/ggml-sycl.cpp b/ggml/src/ggml-sycl/ggml-sycl.cpp
index 60a9c6015d..feced748ce 100644
--- a/ggml/src/ggml-sycl/ggml-sycl.cpp
+++ b/ggml/src/ggml-sycl/ggml-sycl.cpp
@@ -96,6 +96,7 @@ int g_ggml_sycl_enable_graph = 0;
int g_ggml_sycl_enable_dnn = 1;
int g_ggml_sycl_fa_onednn = 1;
int g_ggml_sycl_fa_onednn_max_kv = 0;
+int g_ggml_sycl_enable_mkl_fa = 1;
int g_ggml_sycl_enable_vmm = 1;
int g_ggml_sycl_enable_fusion = 1;
int g_ggml_sycl_enable_esimd = 1;
@@ -333,6 +334,7 @@ static void ggml_check_sycl() try {
g_ggml_sycl_enable_dnn = ggml_sycl_get_env("GGML_SYCL_ENABLE_DNN", 1);
g_ggml_sycl_fa_onednn = ggml_sycl_get_env("GGML_SYCL_FA_ONEDNN", 1);
g_ggml_sycl_fa_onednn_max_kv = ggml_sycl_get_env("GGML_SYCL_FA_ONEDNN_MAX_KV", 0);
+ g_ggml_sycl_enable_mkl_fa = ggml_sycl_get_env("GGML_SYCL_ENABLE_MKL_FA", 1);
g_ggml_sycl_enable_vmm = ggml_sycl_get_env("GGML_SYCL_ENABLE_VMM", 1);
g_ggml_sycl_enable_fusion = ggml_sycl_get_env("GGML_SYCL_ENABLE_FUSION", 1);
g_ggml_sycl_enable_esimd = ggml_sycl_get_env("GGML_SYCL_ENABLE_ESIMD", 1);
@@ -418,6 +420,7 @@ static void ggml_check_sycl() try {
GGML_LOG_INFO(" GGML_SYCL_FA_ONEDNN: %d\n", g_ggml_sycl_fa_onednn);
#endif
GGML_LOG_INFO(" GGML_SYCL_FA_ONEDNN_MAX_KV: %d\n", g_ggml_sycl_fa_onednn_max_kv);
+ GGML_LOG_INFO(" GGML_SYCL_ENABLE_MKL_FA: %d\n", g_ggml_sycl_enable_mkl_fa);
#ifdef SYCL_FLASH_ATTN
GGML_LOG_INFO(" GGML_SYCL_ENABLE_FLASH_ATTN: %d\n", g_ggml_sycl_enable_flash_attention);
#else