EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

HIP: remove the use of __HIP_PLATFORM_AMD__, explicitly support only AMD targets (#14945)

Browse Source
This commit is contained in:
uvos
2025-07-29 20:23:04 +02:00
committed by GitHub
parent b77d11179d
commit aa79524c51
10 changed files with 67 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
ggml/src/ggml-cuda/fattn-wmma-f16.cu
View File

@@ -7,7 +7,7 @@
#include "fattn-wmma-f16.cuh"
#ifdef FP16_MMA_AVAILABLE
#if !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__))
#if !defined(GGML_USE_HIP)
#include <mma.h>
#ifdef GGML_USE_MUSA
namespace wmma = mtmusa::wmma;
@@ -18,7 +18,7 @@ namespace wmma = nvcuda::wmma;
#undef HIP_ENABLE_WARP_SYNC_BUILTINS // conflicts with rocWMMA headers
#include <rocwmma/rocwmma.hpp>
namespace wmma = rocwmma;
#endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__))
#endif // !defined(GGML_USE_HIP)
#endif // FP16_MMA_AVAILABLE
// D == head size, VKQ_stride == num VKQ rows calculated in parallel:
@@ -546,7 +546,7 @@ void ggml_cuda_flash_attn_ext_wmma_f16(ggml_backend_cuda_context & ctx, ggml_ten
return;
}
#if !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__))
#if !defined(GGML_USE_HIP)
if (Q->ne[1] <= 8 && Q->ne[0] % warp_size == 0) {
constexpr int cols_per_block = 8;
switch (Q->ne[0]) {
@@ -568,7 +568,7 @@ void ggml_cuda_flash_attn_ext_wmma_f16(ggml_backend_cuda_context & ctx, ggml_ten
}
return;
}
#endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__))
#endif // !defined(GGML_USE_HIP)
if (Q->ne[1] <= 32) {
constexpr int cols_per_block = 16;