P4: fix OOB fill enum name
This commit is contained in:
@@ -68,7 +68,7 @@ int main() {
|
||||
CU_TENSOR_MAP_DATA_TYPE_BFLOAT16, 2,
|
||||
d_ptr, globalDim, globalStrides, boxDim, elementStrides,
|
||||
CU_TENSOR_MAP_INTERLEAVE_NONE, CU_TENSOR_MAP_SWIZZLE_NONE,
|
||||
CU_TENSOR_MAP_L2_PROMOTION_NONE, CU_TENSOR_MAP_FLOAT_OOB_FILL_ZERO);
|
||||
CU_TENSOR_MAP_L2_PROMOTION_NONE, CU_TENSOR_MAP_FLOAT_OOB_FILL_NAN_REQUEST_ZERO_FMA);
|
||||
if (res == CUDA_SUCCESS) dump("NO swizzle, OOB_FILL_ZERO", tma_desc);
|
||||
else printf("=== NO swizzle, OOB_FILL_ZERO: FAILED (%d) ===\n", res);
|
||||
|
||||
@@ -77,7 +77,7 @@ int main() {
|
||||
CU_TENSOR_MAP_DATA_TYPE_BFLOAT16, 2,
|
||||
d_ptr, globalDim, globalStrides, boxDim, elementStrides,
|
||||
CU_TENSOR_MAP_INTERLEAVE_NONE, CU_TENSOR_MAP_SWIZZLE_128B,
|
||||
CU_TENSOR_MAP_L2_PROMOTION_NONE, CU_TENSOR_MAP_FLOAT_OOB_FILL_ZERO);
|
||||
CU_TENSOR_MAP_L2_PROMOTION_NONE, CU_TENSOR_MAP_FLOAT_OOB_FILL_NAN_REQUEST_ZERO_FMA);
|
||||
if (res == CUDA_SUCCESS) dump("SWIZZLE_128B, OOB_FILL_ZERO", tma_desc);
|
||||
else printf("=== SWIZZLE_128B, OOB_FILL_ZERO: FAILED (%d) ===\n", res);
|
||||
|
||||
|
||||
Reference in New Issue
Block a user