changed StreamKReductionStrategy from Atomic to Reduction as hardcoded

include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp

@@ -1010,10 +1010,16 @@ enum StreamKReductionStrategy
     Reduction,  // let some workgroup responsible for doing the reduction operation
 };
 
+// template <uint32_t MPerBlock_,
+//           uint32_t NPerBlock_,
+//           uint32_t KPerBlock_,
+//           StreamKReductionStrategy ReductionStrategy_ = StreamKReductionStrategy::Atomic,
+//           uint32_t TileSwizzleSubM_                   = 8>
+
 template <uint32_t MPerBlock_,
           uint32_t NPerBlock_,
           uint32_t KPerBlock_,
-          StreamKReductionStrategy ReductionStrategy_ = StreamKReductionStrategy::Atomic,
+          StreamKReductionStrategy ReductionStrategy_ = StreamKReductionStrategy::Reduction,
           uint32_t TileSwizzleSubM_                   = 8>
 struct BlockToCTileMap_GemmStreamK
 {

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp

@@ -539,10 +539,18 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
         const ADataType* p_a_grid;
         const BDataType* p_b_grid;
         CDataType* p_c_grid;
+        // BlockToCTileMap_GemmStreamK_v2<MPerBlock,
+        //                                NPerBlock,
+        //                                KPerBlock,
+        //                                StreamKReductionStrategy::Atomic,
+        //                                8,
+        //                                4>
+        //     block_2_ctile_map_streamk;
+
         BlockToCTileMap_GemmStreamK_v2<MPerBlock,
                                        NPerBlock,
                                        KPerBlock,
-                                       StreamKReductionStrategy::Atomic,
+                                       StreamKReductionStrategy::Reduction,
                                        8,
                                        4>
             block_2_ctile_map_streamk;
@@ -1176,10 +1184,18 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
         }();
         return c_partial_acc_block_m_n;
     }
+    // using Block2CTileMap_streamk = BlockToCTileMap_GemmStreamK_v2<MPerBlock,
+    //                                                               NPerBlock,
+    //                                                               KPerBlock,
+    //                                                               StreamKReductionStrategy::Atomic,
+    //                                                               8,
+    //                                                               4>;
+
+
     using Block2CTileMap_streamk = BlockToCTileMap_GemmStreamK_v2<MPerBlock,
                                                                   NPerBlock,
                                                                   KPerBlock,
-                                                                  StreamKReductionStrategy::Atomic,
+                                                                  StreamKReductionStrategy::Reduction,
                                                                   8,
                                                                   4>;