File tree Expand file tree Collapse file tree 1 file changed +4
-1
lines changed
Expand file tree Collapse file tree 1 file changed +4
-1
lines changed Original file line number Diff line number Diff line change @@ -984,6 +984,7 @@ NDArray_Broadcast(NDArray *a, NDArray *b) {
984984 }
985985 }
986986 }
987+ #ifdef HAVE_CUBLAS
987988 if (NDArray_DEVICE (dst ) == NDARRAY_DEVICE_GPU ) {
988989 for (i = 0 ; i < NDArray_SHAPE (dst )[NDArray_NDIM (dst ) - 2 ]; i ++ ) {
989990 for (j = 0 ; j < NDArray_SHAPE (dst )[NDArray_NDIM (dst ) - 1 ]; j ++ ) {
@@ -992,8 +993,8 @@ NDArray_Broadcast(NDArray *a, NDArray *b) {
992993 NDArray_VMEMCPY_D2D (tmp_p , rtn_p , sizeof (float ));
993994 }
994995 }
995- NDArray_Print (rtn ,0 );
996996 }
997+ #endif
997998 }
998999 if (NDArray_SHAPE (src )[NDArray_NDIM (dst ) - 1 ] == NDArray_SHAPE (dst )[NDArray_NDIM (dst ) - 2 ]) {
9991000 if (NDArray_DEVICE (dst ) == NDARRAY_DEVICE_CPU ) {
@@ -1003,6 +1004,7 @@ NDArray_Broadcast(NDArray *a, NDArray *b) {
10031004 rtn_p = rtn_p + (sizeof (float ) * NDArray_SHAPE (src )[NDArray_NDIM (dst ) - 1 ]);
10041005 }
10051006 }
1007+ #ifdef HAVE_CUBLAS
10061008 if (NDArray_DEVICE (dst ) == NDARRAY_DEVICE_GPU ) {
10071009 for (i = 0 ; i < NDArray_SHAPE (dst )[NDArray_NDIM (dst ) - 2 ]; i ++ ) {
10081010 NDArray_VMEMCPY_D2D (NDArray_DATA (src ), rtn_p ,
@@ -1011,6 +1013,7 @@ NDArray_Broadcast(NDArray *a, NDArray *b) {
10111013 (i * NDArray_STRIDES (rtn )[NDArray_NDIM (rtn ) - 2 ] / NDArray_ELSIZE (rtn )) + j );
10121014 }
10131015 }
1016+ #endif
10141017 }
10151018 }
10161019 return rtn ;
You can’t perform that action at this time.
0 commit comments