HIP: remove the use of __HIP_PLATFORM_AMD__, explicitly support only AMD targets (#14945)
This commit is contained in:
@@ -1391,24 +1391,24 @@ void ggml_cuda_flash_attn_ext_mma_f16_case(ggml_backend_cuda_context & ctx, ggml
|
||||
constexpr bool use_logit_softcap = false;
|
||||
fattn_kernel = flash_attn_ext_f16<DKQ, DV, ncols1, ncols2, nwarps, ntiles, use_logit_softcap, mla>;
|
||||
|
||||
#if !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && !defined(GGML_USE_MUSA)
|
||||
#if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
|
||||
static bool shared_memory_limit_raised[GGML_CUDA_MAX_DEVICES] = {false};
|
||||
if (!shared_memory_limit_raised[id]) {
|
||||
CUDA_CHECK(cudaFuncSetAttribute(fattn_kernel, cudaFuncAttributeMaxDynamicSharedMemorySize, nbytes_shared_total));
|
||||
shared_memory_limit_raised[id] = true;
|
||||
}
|
||||
#endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && !defined(GGML_USE_MUSA)
|
||||
#endif // !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
|
||||
} else {
|
||||
constexpr bool use_logit_softcap = true;
|
||||
fattn_kernel = flash_attn_ext_f16<DKQ, DV, ncols1, ncols2, nwarps, ntiles, use_logit_softcap, mla>;
|
||||
|
||||
#if !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && !defined(GGML_USE_MUSA)
|
||||
#if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
|
||||
static bool shared_memory_limit_raised[GGML_CUDA_MAX_DEVICES] = {false};
|
||||
if (!shared_memory_limit_raised[id]) {
|
||||
CUDA_CHECK(cudaFuncSetAttribute(fattn_kernel, cudaFuncAttributeMaxDynamicSharedMemorySize, nbytes_shared_total));
|
||||
shared_memory_limit_raised[id] = true;
|
||||
}
|
||||
#endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && !defined(GGML_USE_MUSA)
|
||||
#endif // !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
|
||||
}
|
||||
|
||||
launch_fattn<DV, ncols1, ncols2>
|
||||
|
||||
Reference in New Issue
Block a user