groupconv: Sanity check[OK], Performance[Bad]

example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc

@@ -54,7 +54,7 @@ using DeviceConvFwdInstance =
         256,         // BlockSize
         128,         // MPerBlock
         256,         // NPerBlock
-        8,           // K0PerBlock
+        4,           // K0PerBlock
         8,           // K1
         16,          // MPerWMMA
         16,          // NPerWMMA

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

@@ -435,20 +435,12 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
             ds_grid_desc_m_n_ = DeviceOp::MakeDsGridDescriptor_M_N(ds_g_n_k_wos_lengths, ds_g_n_k_wos_strides);
 
             // populate desc for Ds/E
-            if(GridwiseOp::CheckValidity(a_grid_desc_ak0_m_ak1_,
-                                           b_grid_desc_bk0_n_bk1_,
-                                           ds_grid_desc_m_n_,
-                                           e_grid_desc_m_n_,
-                                           block_2_etile_map_))
-            {
-                // e_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    // GridwiseOp::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        // e_grid_desc_m_n_);
-
-                // ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    // GridwiseOp::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        // ds_grid_desc_m_n_);
-            }
+            e_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                GridwiseOp::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                    e_grid_desc_m_n_);
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                GridwiseOp::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                    ds_grid_desc_m_n_);
         }
 
         void Print() const
@@ -520,16 +512,6 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
                 arg.Print();
             }
 
-            if(!GridwiseOp::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                          arg.b_grid_desc_bk0_n_bk1_,
-                                          arg.ds_grid_desc_m_n_,
-                                          arg.e_grid_desc_m_n_,
-                                          arg.block_2_etile_map_))
-            {
-                throw std::runtime_error(
-                    "wrong! GridwiseGemmMultipleD_wmma_cshuffle has invalid setting");
-            }
-
             const index_t grid_size =
                 arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_) * arg.num_group_;