Fix several typos (#1169)

Co-authored-by: isaacw <isaacw@nvidia.com>

include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized_cooperative.hpp

@@ -527,7 +527,7 @@ public:
         auto blk_coord = make_coord(m_coord, n_coord, _, l_coord);
         auto work_k_tile_count = TileScheduler::get_work_k_tile_count(work_tile_info, problem_shape_MNKL, blk_shape);
 
-        // Allocate the the accumulators for the (M,N) blk_shape
+        // Allocate the accumulators for the (M,N) blk_shape
         //
         // MSVC CTAD breaks if we say "Tensor" here, so we use "auto" instead.
         auto accumulators = partition_fragment_C(tiled_mma, take<0,2>(blk_shape));               // (MMA,MMA_M,MMA_N)

include/cutlass/gemm/kernel/sm90_gemm_tma_warpspecialized_pingpong.hpp

@@ -540,7 +540,7 @@ public:
         auto l_coord = idx2crd(work_tile_info.L_idx, shape<4>(gB_nkl));
         auto blk_coord = make_coord(m_coord, n_coord, _, l_coord);
 
-        // Allocate the the accumulators for the (M,N) blk_shape
+        // Allocate the accumulators for the (M,N) blk_shape
         Tensor accumulators = partition_fragment_C(tiled_mma, take<0,2>(blk_shape));               // (MMA,MMA_M,MMA_N)
 
         // Order two Math WG's MMA one after the other, helps hide Epilogue

include/cutlass/gemm/kernel/sm90_tile_scheduler_stream_k.hpp

@@ -347,7 +347,7 @@ public:
     // The number of tiles for which reduction is required is either:
     //   (a) the total number of output tiles (in the case of split-K)
     //   (b) the number of stream-K tiles
-    // To calculate the the total number of output tiles in the split-K case, we
+    // To calcualte the total number of output tiles in the split-K case, we
     // note that, in the split-K case, the units_per_problem_ member of Params will be
     // the total number of output tiles.
     auto reduction_tiles = params.splits_ > 1 ? params.units_per_problem_ : params.sk_tiles_;