Increment stride correctly in warp iterator. (NVIDIA#516)

Co-authored-by: peisun1115 <peis@google.com>
lessw2020 · Jun 6, 2022 · dceefe4 · dceefe4
1 parent c3881d0
commit dceefe4
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/include/cutlass/gemm/warp/mma_tensor_op_tile_iterator.h b/include/cutlass/gemm/warp/mma_tensor_op_tile_iterator.h
@@ -403,7 +403,7 @@ class MmaTensorOpMultiplicandTileIterator<
         AccessType const *source_ptr =
             pointer_[c % kPointerCount] +
             Layout::TileShape::kContiguous * (c / kPointerCount) +
-            Policy::LdsmShape::kStrided * s * stride_;
+            Policy::kLdsmOpInner * Policy::LdsmShape::kStrided * s * stride_;
 
         char const *source_byte_ptr = reinterpret_cast<char const *>(source_ptr) + byte_offset + byte_offset_;