add hipEvent based timing to kernels

library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp

@@ -85,7 +85,7 @@ struct ReferenceGemmBiasActivation : public device::BaseOperator
             return 0;
         }
 
-        float Run(const device::BaseArgument* p_arg, int, hipStream_t) override
+        float Run(const device::BaseArgument* p_arg, int, hipStream_t, bool) override
         {
             return Run(*dynamic_cast<const Argument*>(p_arg));
         }

library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp

@@ -91,7 +91,7 @@ struct ReferenceGemmBiasActivationAdd : public device::BaseOperator
             return 0;
         }
 
-        float Run(const device::BaseArgument* p_arg, int, hipStream_t) override
+        float Run(const device::BaseArgument* p_arg, int, hipStream_t, bool) override
         {
             return Run(*dynamic_cast<const Argument*>(p_arg));
         }

library/src/tensor_operation_instance/gpu/CMakeLists.txt

@@ -82,7 +82,6 @@ PUBLIC -DCK_AMD_GPU_GFX908)
 
 target_compile_options(device_operations
 PRIVATE -amdgpu-target=gfx908
-PRIVATE -O3
 )
 # install(TARGETS device_operations LIBRARY DESTINATION lib)
 install(TARGETS device_operations

library/src/tensor_operation_instance/gpu/device_conv2d.cpp

@@ -37,12 +37,12 @@ struct DeviceConvFwdPtr_t::DeviceConvFwdPtrImpl
 				 std::vector<ck::index_t> conv_filter_strides,
 				 std::vector<ck::index_t> conv_filter_dilations,
 				 std::vector<ck::index_t> input_left_pads,
-				 std::vector<ck::index_t> input_right_pads)
+				 std::vector<ck::index_t> input_right_pads) const
     {
         return el->MakeArgumentPointer(in_ptr, wei_ptr, out_ptr, N, K, C, input_spatial_lengths, filter_spatial_lengths, output_spatial_lengths, conv_filter_strides,
         conv_filter_dilations, input_left_pads, input_right_pads, PassThrough{}, PassThrough{}, PassThrough{});
     }
-    std::unique_ptr<DeviceConvFwdPtr_t::BaseInvoker> MakeInvokerPointer()
+    std::unique_ptr<DeviceConvFwdPtr_t::BaseInvoker> MakeInvokerPointer() const
     {
         return el->MakeInvokerPointer();
     }
@@ -73,13 +73,13 @@ std::unique_ptr<DeviceConvFwdPtr_t::BaseArgument> DeviceConvFwdPtr_t::MakeArgume
 				 std::vector<ck::index_t> conv_filter_strides,
 				 std::vector<ck::index_t> conv_filter_dilations,
 				 std::vector<ck::index_t> input_left_pads,
-				 std::vector<ck::index_t> input_right_pads)
+				 std::vector<ck::index_t> input_right_pads) const
 {
     return   pImpl->MakeArgumentPointer(in_ptr, wei_ptr, out_ptr, N, K, C, input_spatial_lengths, filter_spatial_lengths, output_spatial_lengths, conv_filter_strides,
         conv_filter_dilations, input_left_pads, input_right_pads);
 }
 
-std::unique_ptr<DeviceConvFwdPtr_t::BaseInvoker> DeviceConvFwdPtr_t::MakeInvokerPointer()
+std::unique_ptr<DeviceConvFwdPtr_t::BaseInvoker> DeviceConvFwdPtr_t::MakeInvokerPointer() const
 {
     return pImpl->MakeInvokerPointer();
 }

profiler/CMakeLists.txt

@@ -49,3 +49,5 @@ target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_add_instanc
 target_link_libraries(ckProfiler PRIVATE device_conv2d_fwd_bias_relu_atomic_add_instance)
 target_link_libraries(ckProfiler PRIVATE device_conv2d_bwd_data_instance)
 target_link_libraries(ckProfiler PRIVATE device_reduce_instance)
+
+set_target_properties(ckProfiler PROPERTIES EXCLUDE_FROM_ALL 1)
\ No newline at end of file

test/CMakeLists.txt

@@ -28,6 +28,7 @@ function(add_test_executable TEST_NAME)
     add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}> )
     add_dependencies(tests ${TEST_NAME})
     add_dependencies(check ${TEST_NAME})
+    set_target_properties(${TEST_NAME} PROPERTIES EXCLUDE_FROM_ALL 1)
 endfunction(add_test_executable TEST_NAME)
 
 add_subdirectory(magic_number_division)