happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

add comments and code clean up

This commit is contained in:
jiachengjason 2025-11-10 12:57:56 -05:00 committed by jiachengjason
parent c770ca2812
commit 0bf9f09ae4
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -897,7 +897,7 @@ static __device__ __forceinline__ void mma(
GGML_UNUSED(A);
GGML_UNUSED(B);
NO_DEVICE_CODE;
#endif // AMD_MFMA_AVAILABLE
#endif
}
}

6
ggml/src/ggml-cuda/mmq.cuh
View File

@ -1178,7 +1178,7 @@ static __device__ __forceinline__ void vec_dot_q8_0_16_q8_1_mma(
}
}
}
#elif defined(AMD_WMMA_AVAILABLE)
#elif defined(AMD_WMMA_AVAILABLE) //wmma instructions can handle 16x4 tiles, does not require loading 64x2 tiles
typedef tile<16, 4, int> tile_A;
typedef tile<16, 4, int> tile_B;
typedef tile<16, 16, int> tile_C;
@ -1500,7 +1500,7 @@ static __device__ __forceinline__ void vec_dot_q2_K_q8_1_mma(
}
}
}
#elif defined(AMD_WMMA_AVAILABLE)
#elif defined(AMD_WMMA_AVAILABLE) //wmma instructions can handle 16x4 tiles, does not require loading 64x2 tiles
typedef tile<16, 4, int> tile_A;
typedef tile<16, 4, int> tile_B;
@ -2313,7 +2313,7 @@ static __device__ __forceinline__ void vec_dot_q6_K_q8_1_mma(
}
}
}
#elif defined(AMD_WMMA_AVAILABLE)
#elif defined(AMD_WMMA_AVAILABLE) //wmma instructions can handle 16x4 tiles, does not require loading 64x2 tiles
typedef tile<16, 4, int> tile_A;
typedef tile<16, 4, int> tile_B;
typedef tile<16, 16, int> tile_C;