Fix TMA: use CU_TENSOR_MAP_DATA_TYPE_BFLOAT16 not UINT16
This commit is contained in:
@@ -92,7 +92,7 @@ int main() {
|
||||
|
||||
CUtensorMap tma_desc_host;
|
||||
CUresult res = cuTensorMapEncodeTiled(
|
||||
&tma_desc_host, CU_TENSOR_MAP_DATA_TYPE_UINT16, 2,
|
||||
&tma_desc_host, CU_TENSOR_MAP_DATA_TYPE_BFLOAT16, 2,
|
||||
d_src, gdim, gstr, tdim, tstr,
|
||||
CU_TENSOR_MAP_INTERLEAVE_NONE,
|
||||
CU_TENSOR_MAP_SWIZZLE_NONE,
|
||||
|
||||
Reference in New Issue
Block a user