fixed comments

composable_kernel/include/driver/driver_dynamic_convolution_forward_implicit_gemm_v5r1_nchw_kcyx_nkhw_outpad.hpp

@@ -180,7 +180,6 @@ struct DriverDynamicConvolutionForwardImplicitGemm_v5r1_nchw_kcyx_nkhw_outpad
                                   Sequence<0, 0, 0, 0, 0>{},
                                   Sequence<0, 0, 0, 0, 0>{}));
 
-#if 1
         // GEMM
         using gridwise_gemm = GridwiseDynamicGemm_km_kn_mn_v3<
             BlockSize,
@@ -363,7 +362,6 @@ struct DriverDynamicConvolutionForwardImplicitGemm_v5r1_nchw_kcyx_nkhw_outpad
             std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s"
                       << std::endl;
         }
-#endif
     }
 };
 } // namespace ck

composable_kernel/include/tensor_operation/blockwise_gemm_v3.hpp

@@ -133,12 +133,13 @@ struct BlockwiseGemm_km_kn_m0m1n0n1_v3
         constexpr auto EPerBlock = a_block_mtx.GetLength(I0);
 
         constexpr auto KPerThreadSubC = 4;
-        constexpr auto HPerThreadSubC = 2;
-        constexpr auto WPerThreadSubC = 2;
+
+        constexpr auto HoPerThreadSubC = 2;
+        constexpr auto WoPerThreadSubC = 2;
 
         static_assert(KPerThread % KPerThreadSubC == 0, "");
-        static_assert(HPerThread % HPerThreadSubC == 0, "");
-        static_assert(WPerThread % WPerThreadSubC == 0, "");
+        static_assert(HPerThread % HoPerThreadSubC == 0, "");
+        static_assert(WPerThread % WoPerThreadSubC == 0, "");
 
         // thread A, B for GEMM
         constexpr auto a_thread_mtx = make_dynamic_naive_tensor_descriptor_packed_v2(
@@ -161,8 +162,8 @@ struct BlockwiseGemm_km_kn_m0m1n0n1_v3
         constexpr auto threadwise_gemm = ThreadwiseGemm_km_kn_mn_v3<decltype(a_thread_mtx),
                                                                     decltype(b_thread_mtx),
                                                                     decltype(c_thread_mtx),
-                                                                    HPerThreadSubC,
-                                                                    WPerThreadSubC>{};
+                                                                    HoPerThreadSubC,
+                                                                    WoPerThreadSubC>{};
         // loop over k
 #pragma unroll
         for(index_t e_begin = 0; e_begin < EPerBlock; e_begin += EPerThreadLoop)
@@ -176,10 +177,10 @@ struct BlockwiseGemm_km_kn_m0m1n0n1_v3
                                   p_a_thread);
 
 #pragma unroll
-                for(index_t h_begin = 0; h_begin < HPerThread; h_begin += HPerThreadSubC)
+                for(index_t h_begin = 0; h_begin < HPerThread; h_begin += HoPerThreadSubC)
                 {
 #pragma unroll
-                    for(index_t w_begin = 0; w_begin < WPerThread; w_begin += WPerThreadSubC)
+                    for(index_t w_begin = 0; w_begin < WPerThread; w_begin += WoPerThreadSubC)
                     {
                         threadwise_gemm.Run(p_a_thread,
                                             p_b_thread + b_thread_mtx.CalculateOffset(make_tuple(

composable_kernel/include/utility/config.amd.hpp.in

@@ -36,9 +36,9 @@
 #endif
 
 // buffer resourse
-#if CK_AMD_GPU_GFX906 || CK_AMD_GPU_GFX908
+#if defined(CK_AMD_GPU_GFX906) || defined(CK_AMD_GPU_GFX908)
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x00020000
-#elif CK_AMD_GPU_GFX1030
+#elif defined(CK_AMD_GPU_GFX1030)
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
 #endif
 

driver/include/device_dynamic_convolution_forward_implicit_gemm_v5r1_nchw_kcyx_nkhw.hpp

@@ -164,26 +164,18 @@ void device_dynamic_convolution_forward_implicit_gemm_v5r1_nchw_kcyx_nkhw(
 #endif
 
     constexpr auto conv_driver =
-        // DriverDynamicConvolutionForwardImplicitGemm_v5r1_nchw_kcyx_nkhw_pad<
+#if 0
+        DriverDynamicConvolutionForwardImplicitGemm_v5r1_nchw_kcyx_nkhw_pad<
+#else
         DriverDynamicConvolutionForwardImplicitGemm_v5r1_nchw_kcyx_nkhw_outpad<
+#endif
                        BlockSize,
-            typename vector_type<TInWei, InWeiVectorSize>::type,
-            TAcc,
-            TOut,
-            KPerBlock,
-            HoPerBlock,
-            WoPerBlock,
-            EPerBlock,
-            KPerThread,
-            HoPerThread,
-            WoPerThread,
-            EPerThread,
-            ABlockTransferThreadSliceLengths_E_K,
-            ABlockTransferThreadClusterLengths_E_K,
-            ABlockTransferSrcScalarPerVector_E,
-            ABlockTransferDstScalarPerVector_K,
-            BThreadTransferSrcScalarPerVector_W,
-            CThreadTransferDstScalarPerVector_W>{};
+                   typename vector_type<TInWei, InWeiVectorSize>::type, TAcc, TOut, KPerBlock,
+                   HoPerBlock, WoPerBlock, EPerBlock, KPerThread, HoPerThread, WoPerThread,
+                   EPerThread, ABlockTransferThreadSliceLengths_E_K,
+                   ABlockTransferThreadClusterLengths_E_K, ABlockTransferSrcScalarPerVector_E,
+                   ABlockTransferDstScalarPerVector_K, BThreadTransferSrcScalarPerVector_W,
+                   CThreadTransferDstScalarPerVector_W > {};
 
     conv_driver.Run(wei_k_c0_y_x_desc,
                     in_n_c0_hi_wi_desc,