Fix Copy_Atom type mismatch in sgemm_sm80.cu (#2582)
This commit is contained in:
@ -496,8 +496,8 @@ gemm_nt(int m, int n, int k,
|
|||||||
size(ceil_div(N, bN)));
|
size(ceil_div(N, bN)));
|
||||||
gemm_device<<<dimGrid, dimBlock, smem_size, stream>>>
|
gemm_device<<<dimGrid, dimBlock, smem_size, stream>>>
|
||||||
(prob_shape, cta_tiler,
|
(prob_shape, cta_tiler,
|
||||||
A, dA, sA, copyA, AutoVectorizingCopy{},
|
A, dA, sA, copyA, Copy_Atom<AutoVectorizingCopy, TA>{},
|
||||||
B, dB, sB, copyB, AutoVectorizingCopy{},
|
B, dB, sB, copyB, Copy_Atom<AutoVectorizingCopy, TB>{},
|
||||||
C, dC, sC, mmaC,
|
C, dC, sC, mmaC,
|
||||||
alpha, beta);
|
alpha, beta);
|
||||||
}
|
}
|
||||||
@ -573,8 +573,8 @@ gemm_tn(int m, int n, int k,
|
|||||||
size(ceil_div(N, bN)));
|
size(ceil_div(N, bN)));
|
||||||
gemm_device<<<dimGrid, dimBlock, smem_size, stream>>>
|
gemm_device<<<dimGrid, dimBlock, smem_size, stream>>>
|
||||||
(prob_shape, cta_tiler,
|
(prob_shape, cta_tiler,
|
||||||
A, dA, sA, copyA, AutoVectorizingCopy{},
|
A, dA, sA, copyA, Copy_Atom<AutoVectorizingCopy, TA>{},
|
||||||
B, dB, sB, copyB, AutoVectorizingCopy{},
|
B, dB, sB, copyB, Copy_Atom<AutoVectorizingCopy, TB>{},
|
||||||
C, dC, sC, mmaC,
|
C, dC, sC, mmaC,
|
||||||
alpha, beta);
|
alpha, beta);
|
||||||
}
|
}
|
||||||
|
|||||||
Reference in New Issue
Block a user