Skip to content

Commit 171f08a

Browse files
committed
fix typo
1 parent 477b237 commit 171f08a

File tree

2 files changed

+7
-12
lines changed

2 files changed

+7
-12
lines changed

CMakeLists.txt

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -96,8 +96,8 @@ option(LLAMA_LLAMAFILE "llama: use llamafile SGEMM"
9696
set(LLAMA_BLAS_VENDOR "Generic" CACHE STRING "llama: BLAS library vendor")
9797
option(LLAMA_CUDA "llama: use CUDA" OFF)
9898
option(LLAMA_CUBLAS "llama: use CUDA (deprecated, use LLAMA_CUDA)" OFF)
99-
option(LLAMA_FORCE_DMMV "llama: use dmmv instead of mmvq CUDA kernels" OFF)
100-
option(LLAMA_FORCE_MMQ "llama: use mmq kernels instead of cuBLAS" OFF)
99+
option(LLAMA_FORCE_DMMV "llama: use dmmv instead of mmvq kernels on GPU" OFF)
100+
option(LLAMA_FORCE_MMQ "llama: use mmq kernels instead of Math Lib" OFF)
101101
set(LLAMA_CUDA_DMMV_X "32" CACHE STRING "llama: x stride for dmmv CUDA kernels")
102102
set(LLAMA_CUDA_MMV_Y "1" CACHE STRING "llama: y block size for mmv CUDA kernels")
103103
option(LLAMA_CUDA_F16 "llama: use 16 bit floats for some calculations" OFF)
@@ -628,10 +628,10 @@ if (LLAMA_SYCL)
628628
add_compile_definitions(GGML_SYCL_F16)
629629
endif()
630630

631-
if (LLAMA_SYCL_FORCE_DMMV)
631+
if (LLAMA_FORCE_DMMV)
632632
add_compile_definitions(GGML_SYCL_FORCE_DMMV)
633633
endif()
634-
if (LLAMA_SYCL_FORCE_MMQ)
634+
if (LLAMA_FORCE_MMQ)
635635
add_compile_definitions(GGML_SYCL_FORCE_MMQ)
636636
endif()
637637

ggml-sycl.cpp

Lines changed: 3 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -2978,10 +2978,9 @@ static int g_work_group_size = 0;
29782978

29792979
#define GGML_SYCL_MAX_NODES 8192 //TODO: adapt to hardwares
29802980

2981-
2982-
//define for XMX in Intel GPU
2983-
//TODO: currently, it's not used for XMX really.
2984-
#define SYCL_USE_XMX
2981+
#if !defined(GGML_SYCL_FORCE_MMQ)
2982+
#define SYCL_USE_XMX
2983+
#endif
29852984

29862985
// max batch size to use MMQ kernels when tensor cores are available
29872986
#define MMQ_MAX_BATCH_SIZE 32
@@ -15208,10 +15207,6 @@ static void ggml_sycl_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1
1520815207
}
1520915208
}
1521015209

15211-
#if !defined(GGML_SYCL_FORCE_MMQ)
15212-
#define SYCL_USE_XMX
15213-
#endif
15214-
1521515210
#ifdef SYCL_USE_XMX
1521615211
bool use_xmx = true;
1521715212
#else

0 commit comments

Comments
 (0)