CUTLASS 2.10 bug fixes and minor updates. (#626)

2022-09-15 16:20:33 -04:00
parent 2cc2c7ba1f
commit fc9ebc645b
5 changed files with 63 additions and 8 deletions
--- a/include/cutlass/conv/threadblock/conv2d_dgrad_output_gradient_tile_access_iterator_optimized.h
+++ b/include/cutlass/conv/threadblock/conv2d_dgrad_output_gradient_tile_access_iterator_optimized.h
@@ -633,7 +633,7 @@ public:

    CUTLASS_PRAGMA_UNROLL
    for (int v_idx = 0; v_idx < kAccessesPerVector; ++v_idx) {
-      clear_mask(v_idx, filter_k_ >= problem_size.K);
+      clear_mask(v_idx, filter_k_ + v_idx * AccessType::kElements >= problem_size.K);
    }

    set_iteration_index(0);