add rdna3.5, and 3 to mmq configs so they can be tuned independently. (#26199)
This commit is contained in:
@@ -218,6 +218,8 @@ struct ggml_cuda_mmq_config {
|
||||
|
||||
#include "mmq-config-cdna.cuh"
|
||||
#include "mmq-config-rdna2.cuh"
|
||||
#include "mmq-config-rdna3.cuh"
|
||||
#include "mmq-config-rdna3-5.cuh"
|
||||
#include "mmq-config-rdna4.cuh"
|
||||
|
||||
#undef CASE
|
||||
@@ -227,9 +229,15 @@ static __host__ ggml_cuda_mmq_config ggml_cuda_mmq_get_config(const ggml_type ty
|
||||
if (GGML_CUDA_CC_IS_CDNA(cc)) {
|
||||
return ggml_cuda_mmq_get_config_cdna(type, J, fallback);
|
||||
}
|
||||
if (amd_wmma_available(cc)) {
|
||||
if (GGML_CUDA_CC_IS_RDNA4(cc)) {
|
||||
return ggml_cuda_mmq_get_config_rdna4(type, J, fallback);
|
||||
}
|
||||
if (GGML_CUDA_CC_IS_RDNA3_5(cc)) {
|
||||
return ggml_cuda_mmq_get_config_rdna3_5(type, J, fallback);
|
||||
}
|
||||
if (GGML_CUDA_CC_IS_RDNA3(cc)) { // covers RDNA 3.0
|
||||
return ggml_cuda_mmq_get_config_rdna3(type, J, fallback);
|
||||
}
|
||||
return ggml_cuda_mmq_get_config_rdna2(type, J, fallback);
|
||||
}
|
||||
if (blackwell_mma_available(cc)) {
|
||||
@@ -245,8 +253,12 @@ static constexpr __device__ ggml_cuda_mmq_config ggml_cuda_mmq_get_config(ggml_t
|
||||
#ifdef GGML_USE_HIP
|
||||
#ifdef CDNA
|
||||
return ggml_cuda_mmq_get_config_cdna(type, J, fallback);
|
||||
#elif defined(AMD_WMMA_AVAILABLE)
|
||||
#elif defined(RDNA4)
|
||||
return ggml_cuda_mmq_get_config_rdna4(type, J, fallback);
|
||||
#elif defined(RDNA3_5)
|
||||
return ggml_cuda_mmq_get_config_rdna3_5(type, J, fallback);
|
||||
#elif defined(RDNA3)
|
||||
return ggml_cuda_mmq_get_config_rdna3(type, J, fallback);
|
||||
#else
|
||||
return ggml_cuda_mmq_get_config_rdna2(type, J, fallback);
|
||||
#endif // CDNA
|
||||
|
||||
Reference in New Issue
Block a user