CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-03 09:22:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: use arch list for compatibility check (#11775)

Browse Source 
* CUDA: use arch list for feature availability check

---------

Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit is contained in:
Johannes Gäßler
2025-02-11 00:17:22 +01:00
committed by GitHub
parent 7b891bdc86
commit b9ab0a4d0b
6 changed files with 80 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

65
ggml/src/ggml-cuda/common.cuh
View File

@@ -71,6 +71,47 @@
#define GGML_CUDA_CC_QY1 210
#define GGML_CUDA_CC_QY2 220
#ifdef __CUDA_ARCH_LIST__
constexpr bool ggml_cuda_has_arch_impl(int) {
return false;
}
template<class ... Archs>
constexpr bool ggml_cuda_has_arch_impl(const int arch, const int first, Archs... rest) {
return arch == first || ggml_cuda_has_arch_impl(arch, rest...);
}
constexpr bool ggml_cuda_has_arch(const int arch) {
return ggml_cuda_has_arch_impl(arch, __CUDA_ARCH_LIST__);
}
constexpr int ggml_cuda_highest_compiled_arch_impl(const int arch, const int cur) {
if (cur == 0) {
GGML_ABORT("ggml was not compiled with any CUDA arch <= %d", arch);
}
return cur;
}
template<class ... Archs>
constexpr int ggml_cuda_highest_compiled_arch_impl(const int arch, const int cur, const int first, Archs... rest) {
if (first <= arch && first > cur) {
return ggml_cuda_highest_compiled_arch_impl(arch, first, rest...);
} else {
return ggml_cuda_highest_compiled_arch_impl(arch, cur, rest...);
}
}
constexpr int ggml_cuda_highest_compiled_arch(const int arch) {
return ggml_cuda_highest_compiled_arch_impl(arch, 0, __CUDA_ARCH_LIST__);
}
#else
static int ggml_cuda_highest_compiled_arch(const int arch) {
return arch;
}
#endif // __CUDA_ARCH_LIST__
// ---------------------------------------------------------------------------------------------------------
#define MATRIX_ROW_PADDING 512 // last row of quant. matrices is a multiple of this to avoid out-of-bounds memory accesses
#if defined(_MSC_VER)
@@ -162,18 +203,32 @@ typedef float2 dfloat2;
#define FLASH_ATTN_AVAILABLE
#endif // !(defined(GGML_USE_MUSA) && __MUSA_ARCH__ <= GGML_CUDA_CC_QY1)
static constexpr bool fast_fp16_available(const int cc) {
static bool fp16_available(const int cc) {
return ggml_cuda_highest_compiled_arch(cc) >= GGML_CUDA_CC_PASCAL;
}
static bool fast_fp16_available(const int cc) {
return fp16_available(cc) && cc != 610;
}
// To be used for feature selection of external libraries, e.g. cuBLAS.
static bool fast_fp16_hardware_available(const int cc) {
return cc >= GGML_CUDA_CC_PASCAL && cc != 610;
}
// Any FP16 tensor cores are available.
static constexpr bool fp16_mma_available(const int cc) {
// Any FP16 tensor core instructions are available for ggml code.
static bool fp16_mma_available(const int cc) {
return cc < GGML_CUDA_CC_OFFSET_AMD && ggml_cuda_highest_compiled_arch(cc) >= GGML_CUDA_CC_VOLTA;
}
// To be used for feature selection of external libraries, e.g. cuBLAS.
static bool fp16_mma_hardware_available(const int cc) {
return cc < GGML_CUDA_CC_OFFSET_AMD && cc >= GGML_CUDA_CC_VOLTA;
}
// Volta technically had FP16 tensor cores but they work very differently compared to Turing and later.
static constexpr bool new_mma_available(const int cc) {
return cc < GGML_CUDA_CC_OFFSET_AMD && cc >= GGML_CUDA_CC_TURING;
static bool new_mma_available(const int cc) {
return cc < GGML_CUDA_CC_OFFSET_AMD && ggml_cuda_highest_compiled_arch(cc) >= GGML_CUDA_CC_TURING;
}
static constexpr __device__ int ggml_cuda_get_physical_warp_size() {