fix bug
This commit is contained in:
parent
94851cec68
commit
4906961bf1
|
|
@ -631,7 +631,7 @@ __global__ void VecQuant4TransposeMatMulHalfKernel(
|
||||||
#if __CUDA_ARCH__ < 700 && __CUDA_ARCH__ > 600
|
#if __CUDA_ARCH__ < 700 && __CUDA_ARCH__ > 600
|
||||||
atomicAddHalf(&mul2[n_cols * height * 8 + n_rows], res);
|
atomicAddHalf(&mul2[n_cols * height * 8 + n_rows], res);
|
||||||
#else
|
#else
|
||||||
atomicAddHalf(&mul2[n_cols * height * 8 + n_rows], res);
|
atomicAdd(&mul2[n_cols * height * 8 + n_rows], res);
|
||||||
#endif
|
#endif
|
||||||
#endif
|
#endif
|
||||||
}
|
}
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue