CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-11 10:36:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: add dynamic shared mem to softmax, refactor general usage (#14497)

Browse Source
This commit is contained in:
Aman Gupta
2025-07-03 07:45:11 +08:00
committed by GitHub
parent e75ba4c043
commit 55c2646b45
5 changed files with 57 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
ggml/src/ggml-cuda/common.cuh
View File

@@ -175,6 +175,20 @@ static const char * cu_get_error_str(CUresult err) {
#define CU_CHECK(err) CUDA_CHECK_GEN(err, CUDA_SUCCESS, cu_get_error_str)
#endif
#if !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && !defined(GGML_USE_MUSA)
#define CUDA_SET_SHARED_MEMORY_LIMIT(kernel, nbytes) \
do { \
static bool shared_memory_limit_raised[GGML_CUDA_MAX_DEVICES] = {false}; \
const int id = ggml_cuda_get_device(); \
if (!shared_memory_limit_raised[id]) { \
CUDA_CHECK(cudaFuncSetAttribute(kernel, cudaFuncAttributeMaxDynamicSharedMemorySize, nbytes)); \
shared_memory_limit_raised[id] = true; \
} \
} while (0)
#else
#define CUDA_SET_SHARED_MEMORY_LIMIT(kernel, nbytes) do {} while (0)
#endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && !defined(GGML_USE_MUSA)
#if CUDART_VERSION >= 11010 || defined(GGML_USE_MUSA)
#define GGML_CUDA_ASSUME(x) __builtin_assume(x)
#else