fix out_ptr bug

composable_kernel/include/kernel_algorithm/gridwise_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw_lds_double_buffer.hpp

@@ -397,9 +397,9 @@ struct GridwiseConvolutionImplicitGemm_v4r1_nchw_kcyx_nkhw_lds_double_buffer
 
             //     origin of dst in device memory
             Float* p_out_thread_on_global =
-                p_out_global +
-                out_k_n1_b_n2_global_merged_desc.GetOffsetFromMultiIndex(
-                    k_thread_data_on_global, 0, b_thread_data_on_global, 0);
+                p_out_global  +
+                 out_k_n1_b_n2_global_merged_desc.GetOffsetFromMultiIndex(
+                     k_thread_data_on_global, 0, b_thread_data_on_global, 0);
 #if 1
             threadwise_generic_tensor_slice_copy_v1(
                 out_n0_n1_n2_k0_k1_k2_h_w_thread_desc,
@@ -411,7 +411,7 @@ struct GridwiseConvolutionImplicitGemm_v4r1_nchw_kcyx_nkhw_lds_double_buffer
                 out_n0_n1_n2_k0_k1_k2_h_w_thread_desc.GetLengths(),
                 arithmetic_sequence_gen<0, 8, 1>::type{},
                 Number<1>{});
-#elif 0
+#elif 1
 	p_out_global[0] = p_out_thread[0];
 #endif
         }

driver/include/device_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw.hpp

@@ -241,8 +241,8 @@ void device_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw(InDesc,
                                    dim3(BlockSize),
                                    0,
                                    static_cast<T*>(in_nchw_device_buf.GetDeviceBuffer()),
-                                   static_cast<T*>(out_nkhw_device_buf.GetDeviceBuffer()));
-                                   static_cast<T*>(wei_kcyx_device_buf.GetDeviceBuffer()),
+                                   static_cast<T*>(out_nkhw_device_buf.GetDeviceBuffer()),
+                                   static_cast<T*>(wei_kcyx_device_buf.GetDeviceBuffer()));
 
         printf("Elapsed time : %f ms, %f TFlop/s\n",
                time,