From f12e9de11124915f804432d3586efba32c13afce Mon Sep 17 00:00:00 2001
From: George Wu <jywu@microsoft.com>
Date: Thu, 13 Aug 2020 22:24:44 +0000
Subject: [PATCH] build fixes for
 https://github.com/microsoft/onnxruntime/pull/4721 (#4784)

* test

* test

* add missing CUDA header include

* debug

* fix

* fix python package for dnnl and tensorrt.

* fix

* fix windows build.

* revert

* target_link_directories for tensorrt shared lib.
---
 cmake/onnxruntime_framework.cmake |  7 ++++++-
 cmake/onnxruntime_providers.cmake |  8 ++++++--
 cmake/onnxruntime_python.cmake    | 10 ++++++++++
 setup.py                          |  2 +-
 4 files changed, 23 insertions(+), 4 deletions(-)

diff --git a/cmake/onnxruntime_framework.cmake b/cmake/onnxruntime_framework.cmake
index 4da6450c00..eebad5791a 100644
--- a/cmake/onnxruntime_framework.cmake
+++ b/cmake/onnxruntime_framework.cmake
@@ -13,7 +13,12 @@ add_library(onnxruntime_framework ${onnxruntime_framework_srcs})
 if(onnxruntime_ENABLE_INSTRUMENT)
   target_compile_definitions(onnxruntime_framework PRIVATE ONNXRUNTIME_ENABLE_INSTRUMENT)
 endif()
-target_include_directories(onnxruntime_framework PRIVATE ${ONNXRUNTIME_ROOT} ${onnxruntime_CUDNN_HOME}/include PUBLIC ${CMAKE_CURRENT_BINARY_DIR})
+if(onnxruntime_USE_TENSORRT)
+# TODO: for now, core framework depends on CUDA. It should be moved to TensorRT EP
+target_include_directories(onnxruntime_framework PRIVATE ${ONNXRUNTIME_ROOT} ${onnxruntime_CUDNN_HOME}/include PUBLIC ${CMAKE_CURRENT_BINARY_DIR} ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
+else()
+target_include_directories(onnxruntime_framework PRIVATE ${ONNXRUNTIME_ROOT} PUBLIC ${CMAKE_CURRENT_BINARY_DIR})
+endif()
 onnxruntime_add_include_to_target(onnxruntime_framework onnxruntime_common onnx onnx_proto protobuf::libprotobuf)
 set_target_properties(onnxruntime_framework PROPERTIES FOLDER "ONNXRuntime")
 # need onnx to build to create headers that this project includes
diff --git a/cmake/onnxruntime_providers.cmake b/cmake/onnxruntime_providers.cmake
index e288a58b57..7ff17f4611 100644
--- a/cmake/onnxruntime_providers.cmake
+++ b/cmake/onnxruntime_providers.cmake
@@ -364,7 +364,6 @@ if (onnxruntime_USE_TENSORRT)
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -include algorithm")
     set(PROTOBUF_LIBRARY libprotobuf)
     set(DISABLED_WARNINGS_FOR_TRT /wd4267 /wd4244 /wd4996 /wd4456)
-    list(APPEND CUDA_LIBRARIES cudart.lib cudadevrt.lib)
   endif()
   if ( CMAKE_COMPILER_IS_GNUCC )
     set(CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} -Wno-unused-parameter -Wno-missing-field-initializers")
@@ -401,9 +400,14 @@ if (onnxruntime_USE_TENSORRT)
 
   source_group(TREE ${ONNXRUNTIME_ROOT}/core FILES ${onnxruntime_providers_tensorrt_cc_srcs})
   add_library(onnxruntime_providers_tensorrt SHARED ${onnxruntime_providers_tensorrt_cc_srcs})
-  target_link_libraries(onnxruntime_providers_tensorrt PRIVATE ${onnxparser_link_libs} ${trt_link_libs} cudart onnxruntime_providers_shared protobuf::libprotobuf)
   onnxruntime_add_include_to_target(onnxruntime_providers_tensorrt onnxruntime_common onnx )
   add_dependencies(onnxruntime_providers_tensorrt onnxruntime_providers_shared ${onnxruntime_EXTERNAL_DEPENDENCIES})
+  if(WIN32)
+    target_link_directories(onnxruntime_providers_tensorrt PRIVATE ${onnxruntime_CUDA_HOME}/x64/lib64)
+  else()
+    target_link_directories(onnxruntime_providers_tensorrt PRIVATE ${onnxruntime_CUDA_HOME}/lib64)
+  endif()
+  target_link_libraries(onnxruntime_providers_tensorrt PRIVATE ${onnxparser_link_libs} ${trt_link_libs} cudart onnxruntime_providers_shared protobuf::libprotobuf)
   target_include_directories(onnxruntime_providers_tensorrt PRIVATE ${ONNXRUNTIME_ROOT} ${onnxruntime_CUDNN_HOME}/include ${eigen_INCLUDE_DIRS} PUBLIC ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
   install(DIRECTORY ${PROJECT_SOURCE_DIR}/../include/onnxruntime/core/providers/tensorrt  DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/onnxruntime/core/providers)
   set_target_properties(onnxruntime_providers_tensorrt PROPERTIES LINKER_LANGUAGE CXX)
diff --git a/cmake/onnxruntime_python.cmake b/cmake/onnxruntime_python.cmake
index 7ea318f9d9..d30c1f0d62 100644
--- a/cmake/onnxruntime_python.cmake
+++ b/cmake/onnxruntime_python.cmake
@@ -270,6 +270,16 @@ if (onnxruntime_USE_DNNL)
   )
 endif()
 
+if (onnxruntime_USE_TENSORRT)
+  add_custom_command(
+    TARGET onnxruntime_pybind11_state POST_BUILD
+    COMMAND ${CMAKE_COMMAND} -E copy
+        ${DNNL_DLL_PATH} $<TARGET_FILE:onnxruntime_providers_tensorrt>
+        $<TARGET_FILE:onnxruntime_providers_shared>
+        $<TARGET_FILE_DIR:${test_data_target}>/onnxruntime/capi/
+  )
+endif()
+
 if (onnxruntime_USE_NGRAPH)
   add_custom_command(
     TARGET onnxruntime_pybind11_state POST_BUILD
diff --git a/setup.py b/setup.py
index 11dc205d8d..5fe14996da 100644
--- a/setup.py
+++ b/setup.py
@@ -183,7 +183,7 @@ elif platform.system() == "Darwin":
 else:
   libs = ['onnxruntime_pybind11_state.pyd', 'dnnl.dll', 'mklml.dll', 'libiomp5md.dll']
   # DNNL & TensorRT EPs are built as shared libs
-  libs.extend(['libonnxruntime_providers_shared.dll'])
+  libs.extend(['onnxruntime_providers_shared.dll'])
   libs.extend(['onnxruntime_providers_dnnl.dll'])
   libs.extend(['onnxruntime_providers_tensorrt.dll'])
   # nGraph Libs