clean up

example/6_conv2d_fwd_xdl_c_shuffle_bias_relu_add/conv2d_fwd_xdl_c_shuffle_bias_relu_add.cpp

@@ -28,7 +28,7 @@ using OutLayout = ck::tensor_layout::convolution::NHWK;
 
 using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
 using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd_v2;
+using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
 
 // clang-format off
 using DeviceConvFwdInstance = ck::tensor_operation::device::

profiler/CMakeLists.txt

@@ -48,6 +48,7 @@ install(TARGETS device_conv2d_fwd_instance LIBRARY DESTINATION lib)
 # device_conv2d_fwd_bias_relu_instance
 set(DEVICE_CONV2D_FWD_BIAS_RELU_INSTANCE_SOURCE 
    ${PROJECT_SOURCE_DIR}/device_operation/device_conv2d_fwd_xdl_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp;
+   ${PROJECT_SOURCE_DIR}/device_operation/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp;
 ) 
 
 add_library(device_conv2d_fwd_bias_relu_instance SHARED ${DEVICE_CONV2D_FWD_BIAS_RELU_INSTANCE_SOURCE}) 

profiler/include/profile_conv_fwd_impl.hpp

@@ -149,7 +149,6 @@ void profile_conv_fwd_impl(int do_verification,
                       ck::is_same_v<ck::remove_cv_t<WeiDataType>, ck::half_t> &&
                       ck::is_same_v<ck::remove_cv_t<OutDataType>, ck::half_t>)
     {
-#if 0 // debug
         ck::tensor_operation::device::device_conv2d_fwd_instance::
             add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
 
@@ -158,12 +157,9 @@ void profile_conv_fwd_impl(int do_verification,
 
         ck::tensor_operation::device::device_conv2d_fwd_instance::
             add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_1x1_s1_p0_f16_instances(conv_ptrs);
-#endif
 
-#if 1 // debug
         ck::tensor_operation::device::device_conv2d_fwd_instance::
             add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
-#endif
     }
 
     if(conv_ptrs.size() <= 0)