Remove redundant data member

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp

@@ -252,8 +252,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
               K{K_},
               a_grid_desc_k0_m_k1_{},
               b_grid_desc_k0_n_k1_{},
-              c_grid_desc_m_n_{},
-              kraw_{K}
+              c_grid_desc_m_n_{}
         {
             a_grid_desc_k0_m_k1_ = DeviceGemmXdl::MakeAGridDescriptor_K0_M_K1(M_, K_, StrideA);
             b_grid_desc_k0_n_k1_ = DeviceGemmXdl::MakeBGridDescriptor_K0_N_K1(K_, N_, StrideB);
@@ -270,7 +269,6 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
         AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
         BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
         CGridDesc_M_N c_grid_desc_m_n_;
-        index_t kraw_;
     };
 
     // Invoker
@@ -379,7 +377,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
             return false;
         }
 
-        if(arg.kraw_ % K1 != 0)
+        if(arg.K % K1 != 0)
         {
             return false;
         }

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp

@@ -99,11 +99,27 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+#if defined(INTEGER_DIVIDE_CEIL)
+#error "macro INTEGER_DIVIDE_CEIL() was already defined somewhere else"
+#endif
+
+#define INTEGER_DIVIDE_CEIL(x, y) (((x) + (y)-1) / (y))
     __host__ static auto CalculateGridSize(index_t M, index_t N)
     {
         return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
     }
 
+    __host__ static auto CalculateMPadded(index_t M)
+    {
+        return INTEGER_DIVIDE_CEIL(M, MPerBlock) * MPerBlock;
+    }
+
+    __host__ static auto CalculateNPadded(index_t N)
+    {
+        return INTEGER_DIVIDE_CEIL(N, NPerBlock) * NPerBlock;
+    }
+#undef INTEGER_DIVIDE_CEIL
+
     using GridwiseGemmPipe = remove_cvref_t<decltype(
         GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;