root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-24 18:34:36 +00:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

Allow all RDNA2 archs to use sdot4 intrinsic (#8629)

Browse Source 
The check gating the use of `__builtin_amdgc_sdot4` specifically checks for gfx1030. This causes a severe perf regression for anything gfx103? that's not gfx1030 and not using `HSA_OVERRIDE_GFX_VERSION` (if you've built ROCm to support it). We already have a generic RDNA2 define, let's use it.
This commit is contained in:
Jeroen Mostert 2024-07-23 10:50:40 +02:00 committed by GitHub
parent e7e6487ba0
commit 46e47417aa
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda/common.cuh
View File

@ -459,7 +459,7 @@ static __device__ __forceinline__ uint32_t __hgt2_mask(const half2 a, const half
static __device__ __forceinline__ int ggml_cuda_dp4a(const int a, const int b, int c) { static __device__ __forceinline__ int ggml_cuda_dp4a(const int a, const int b, int c) {
#if defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__) #if defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
#if defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__) #if defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(RDNA2)
c = __builtin_amdgcn_sdot4(a, b, c, false); c = __builtin_amdgcn_sdot4(a, b, c, false);
#elif defined(RDNA3) #elif defined(RDNA3)
c = __builtin_amdgcn_sudot4( true, a, true, b, c, false); c = __builtin_amdgcn_sudot4( true, a, true, b, c, false);