CUTLASS 3.1 (#915)

Co-authored-by: Aniket Shivam <ashivam@nvidia.com>
2023-04-14 23:19:34 -04:00
parent 9b8166e3f0
commit d572cc1aab
482 changed files with 37175 additions and 16410 deletions
@@ -159,7 +159,7 @@ class DualGemm {
  using Mma0 = typename cutlass::gemm::threadblock::DefaultMma<
      ElementA, LayoutA, kAlignmentA, ElementB, LayoutB0, kAlignmentB,
      ElementAccumulator, layout::RowMajor, arch::OpClassTensorOp, ArchTag,
-      ThreadblockShape, WarpShape, 
+      ThreadblockShape, WarpShape,
      InstructionShape, Stages, Operator>::ThreadblockMma;
  using Mma1 = typename cutlass::gemm::threadblock::DefaultMma<
      ElementA, LayoutA, kAlignmentA, ElementB, LayoutB1, kAlignmentB,
@@ -348,7 +348,7 @@ public:
      ThreadblockSwizzle threadblock_swizzle;

      cutlass::gemm::GemmCoord tiled_shape = threadblock_swizzle.get_tiled_shape(
-        args.problem_size, 
+        args.problem_size,
        {ThreadblockShape::kM, ThreadblockShape::kN, ThreadblockShape::kK},
        args.split_k_slices);