replica
/
llama.cpp
mirror of https://github.com/ggml-org/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path. (#14930) 

This is useful for testing for regressions on GCN with CDNA hardware.

With GGML_HIP_MMQ_MFMA=Off and GGML_CUDA_FORCE_MMQ=On we can conveniently test the GCN code path on CDNA. As CDNA is just GCN renamed with MFMA added and limited use ACC registers, this provides a good alternative for regression testing when GCN hardware is not available.
This commit is contained in:
uvos 2025-07-29 17:44:30 +02:00 committed by GitHub
parent c7aa1364fd
commit b77d11179d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 11 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/CMakeLists.txt
View File

@ -174,6 +174,7 @@ option(GGML_HIP_GRAPHS "ggml: use HIP graph, experimental,
option(GGML_HIP_NO_VMM "ggml: do not try to use HIP VMM" ON)
option(GGML_HIP_ROCWMMA_FATTN "ggml: enable rocWMMA for FlashAttention" OFF)
option(GGML_HIP_FORCE_ROCWMMA_FATTN_GFX12 "ggml: enable rocWMMA FlashAttention on GFX12" OFF)
option(GGML_HIP_MMQ_MFMA "ggml: enable MFMA MMA for CDNA in MMQ" ON)
option(GGML_MUSA_GRAPHS "ggml: use MUSA graph, experimental, unstable" OFF)
option(GGML_MUSA_MUDNN_COPY "ggml: enable muDNN for accelerated copy" OFF)
option(GGML_VULKAN "ggml: use Vulkan" OFF)

8
ggml/src/ggml-cuda/common.cuh
View File

@ -227,7 +227,7 @@ typedef float2 dfloat2;
#define FP16_MMA_AVAILABLE
#endif // defined(GGML_HIP_ROCWMMA_FATTN) && (defined(CDNA) || defined(RDNA3) || (defined(GGML_HIP_ROCWMMA_FATTN_GFX12) && defined(RDNA4)))
#if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && defined(CDNA3)
#if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && defined(CDNA3) && !defined(GGML_HIP_NO_MMQ_MFMA)
#define AMD_MFMA_AVAILABLE
#endif // defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && defined(CDNA3)
@ -295,7 +295,11 @@ static bool fp32_mma_hardware_available(const int cc) {
// AMD CDNA3 matrix cores.. Will add support for other CDNA generations later.
static bool amd_mfma_available(const int cc) {
return cc >= GGML_CUDA_CC_OFFSET_AMD && GGML_CUDA_CC_IS_CDNA3(cc);
#if !defined(GGML_HIP_NO_MMQ_MFMA)
return GGML_CUDA_CC_IS_CDNA3(cc);
#else
return false;
#endif //!defined(GGML_HIP_NO_MMQ_MFMA)
}
// Volta technically had FP16 tensor cores but they work very differently compared to Turing and later.

4
ggml/src/ggml-hip/CMakeLists.txt
View File

@ -113,6 +113,10 @@ if (GGML_HIP_ROCWMMA_FATTN)
add_compile_definitions(GGML_HIP_ROCWMMA_FATTN)
endif()
if (NOT GGML_HIP_MMQ_MFMA)
add_compile_definitions(GGML_HIP_NO_MMQ_MFMA)
endif()
if (GGML_HIP_FORCE_ROCWMMA_FATTN_GFX12 OR ${hip_VERSION} VERSION_GREATER_EQUAL 7.0)
add_compile_definitions(GGML_HIP_ROCWMMA_FATTN_GFX12)
endif()