Merge topic 'cuda_updates_for_10.2'

5341f5e4a1 CUDA: get header deps from compiler invocation when possible
7f15c99851 CUDA: forward unknown flags to host compiler when possible.

Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !4099

3 files changed, 38 insertions, 30 deletions

diff --git a/Modules/CMakeCUDAInformation.cmake b/Modules/CMakeCUDAInformation.cmake
index 806ee3c..974f5fa 100644
--- a/Modules/CMakeCUDAInformation.cmake
+++ b/Modules/CMakeCUDAInformation.cmake
@@ -98,9 +98,7 @@ include(CMakeCommonLanguageInclude)
 # CMAKE_CUDA_LINK_EXECUTABLE
 
 if(CMAKE_CUDA_HOST_COMPILER)
-  set(CMAKE_CUDA_HOST_FLAGS "-ccbin=<CMAKE_CUDA_HOST_COMPILER>")
-else()
-  set(CMAKE_CUDA_HOST_FLAGS "")
+  string(APPEND _CMAKE_CUDA_EXTRA_FLAGS " -ccbin=<CMAKE_CUDA_HOST_COMPILER>")
 endif()
 
 set(__IMPLICT_LINKS )
@@ -140,26 +138,26 @@ endif()
 #Specify how to compile when ptx has been requested
 if(NOT CMAKE_CUDA_COMPILE_PTX_COMPILATION)
   set(CMAKE_CUDA_COMPILE_PTX_COMPILATION
-    "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -ptx <SOURCE> -o <OBJECT>")
+    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -ptx <SOURCE> -o <OBJECT>")
 endif()
 
 #Specify how to compile when separable compilation has been requested
 if(NOT CMAKE_CUDA_COMPILE_SEPARABLE_COMPILATION)
   set(CMAKE_CUDA_COMPILE_SEPARABLE_COMPILATION
-    "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -dc <SOURCE> -o <OBJECT>")
+    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -dc <SOURCE> -o <OBJECT>")
 endif()
 
 #Specify how to compile when whole compilation has been requested
 if(NOT CMAKE_CUDA_COMPILE_WHOLE_COMPILATION)
   set(CMAKE_CUDA_COMPILE_WHOLE_COMPILATION
-    "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -c <SOURCE> -o <OBJECT>")
+    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -c <SOURCE> -o <OBJECT>")
 endif()
 
-if(CMAKE_GENERATOR STREQUAL "Ninja")
+if(CMAKE_GENERATOR STREQUAL "Ninja" AND NOT CMAKE_DEPFILE_FLAGS_CUDA )
   set(CMAKE_CUDA_COMPILE_DEPENDENCY_DETECTION
-    "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -M <SOURCE> -MT <OBJECT> -o $DEP_FILE")
+    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -M <SOURCE> -MT <OBJECT> -o $DEP_FILE")
   #The Ninja generator uses the make file dependency files to determine what
-  #files need to be recompiled. Unfortunately, nvcc doesn't support building
+  #files need to be recompiled. Unfortunately, nvcc < 10.2 doesn't support building
   #a source file and generating the dependencies of said file in a single
   #invocation. Instead we have to state that you need to chain two commands.
   #
@@ -176,13 +174,6 @@ if(NOT CMAKE_CUDA_LINK_EXECUTABLE)
     "<CMAKE_CUDA_HOST_LINK_LAUNCHER> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>${__IMPLICT_LINKS}")
 endif()
 
-if( CMAKE_CUDA_COMPILER_ID STREQUAL "NVIDIA" AND
-    CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "8.0.0")
-  set(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS "-Wno-deprecated-gpu-targets")
-else()
-  set(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS "")
-endif()
-
 # Add implicit host link directories that contain device libraries
 # to the device link line.
 set(__IMPLICT_DLINK_DIRS ${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES})
@@ -197,14 +188,15 @@ foreach(dir ${__IMPLICT_DLINK_DIRS})
 endforeach()
 unset(__IMPLICT_DLINK_DIRS)
 
+
 #These are used when linking relocatable (dc) cuda code
 if(NOT CMAKE_CUDA_DEVICE_LINK_LIBRARY)
   set(CMAKE_CUDA_DEVICE_LINK_LIBRARY
-    "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <LANGUAGE_COMPILE_FLAGS> ${CMAKE_CUDA_COMPILE_OPTIONS_PIC} ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES>${__IMPLICT_DLINK_FLAGS}")
+    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <LANGUAGE_COMPILE_FLAGS> ${CMAKE_CUDA_COMPILE_OPTIONS_PIC} ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES>${__IMPLICT_DLINK_FLAGS}")
 endif()
 if(NOT CMAKE_CUDA_DEVICE_LINK_EXECUTABLE)
   set(CMAKE_CUDA_DEVICE_LINK_EXECUTABLE
-    "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <FLAGS> ${CMAKE_CUDA_COMPILE_OPTIONS_PIC} ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES>${__IMPLICT_DLINK_FLAGS}")
+    "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <FLAGS> ${CMAKE_CUDA_COMPILE_OPTIONS_PIC} ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES>${__IMPLICT_DLINK_FLAGS}")
 endif()
 
 unset(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS)
diff --git a/Modules/Compiler/NVIDIA-CUDA.cmake b/Modules/Compiler/NVIDIA-CUDA.cmake
index b59deda..2b24fa5 100644
--- a/Modules/Compiler/NVIDIA-CUDA.cmake
+++ b/Modules/Compiler/NVIDIA-CUDA.cmake
@@ -2,6 +2,29 @@ set(CMAKE_CUDA_COMPILER_HAS_DEVICE_LINK_PHASE True)
 set(CMAKE_CUDA_VERBOSE_FLAG "-v")
 set(CMAKE_CUDA_VERBOSE_COMPILE_FLAG "-Xcompiler=-v")
 
+if (NOT CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 10.2)
+  # The -forward-unknown-to-host-compiler flag was only
+  # added to nvcc in 10.2 so before that we had no good
+  # way to invoke the CUDA compiler and propagate unknown
+  # flags such as -pthread to the host compiler
+  set(_CMAKE_CUDA_EXTRA_FLAGS "-forward-unknown-to-host-compiler")
+else()
+  set(_CMAKE_CUDA_EXTRA_FLAGS "")
+endif()
+
+if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "8.0.0")
+  set(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS "-Wno-deprecated-gpu-targets")
+else()
+  set(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS "")
+endif()
+
+if (NOT CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 10.2)
+  # The -MD flag was only added to nvcc in 10.2 so
+  # before that we had to invoke the compiler twice
+  # to get header dependency information
+  set(CMAKE_DEPFILE_FLAGS_CUDA "-MD -MT <OBJECT> -MF <DEPFILE>")
+endif()
+
 if(NOT "x${CMAKE_CUDA_SIMULATE_ID}" STREQUAL "xMSVC")
   set(CMAKE_CUDA_COMPILE_OPTIONS_PIE -Xcompiler=-fPIE)
   set(CMAKE_CUDA_COMPILE_OPTIONS_PIC -Xcompiler=-fPIC)
diff --git a/Modules/Platform/Windows-NVIDIA-CUDA.cmake b/Modules/Platform/Windows-NVIDIA-CUDA.cmake
index 94d77b9..30b5aa9 100644
--- a/Modules/Platform/Windows-NVIDIA-CUDA.cmake
+++ b/Modules/Platform/Windows-NVIDIA-CUDA.cmake
@@ -1,11 +1,11 @@
 include(Platform/Windows-MSVC)
 
 set(CMAKE_CUDA_COMPILE_PTX_COMPILATION
-  "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -ptx <SOURCE> -o <OBJECT> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS")
+  "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -ptx <SOURCE> -o <OBJECT> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS")
 set(CMAKE_CUDA_COMPILE_SEPARABLE_COMPILATION
-  "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -dc <SOURCE> -o <OBJECT> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS")
+  "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -dc <SOURCE> -o <OBJECT> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS")
 set(CMAKE_CUDA_COMPILE_WHOLE_COMPILATION
-  "<CMAKE_CUDA_COMPILER> ${CMAKE_CUDA_HOST_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -c <SOURCE> -o <OBJECT> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS")
+  "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <DEFINES> <INCLUDES> <FLAGS> -x cu -c <SOURCE> -o <OBJECT> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS")
 
 set(__IMPLICT_LINKS )
 foreach(dir ${CMAKE_CUDA_HOST_IMPLICIT_LINK_DIRECTORIES})
@@ -30,11 +30,6 @@ set(CMAKE_CUDA_LINK_EXECUTABLE
 unset(_CMAKE_VS_LINK_EXE)
 unset(_CMAKE_VS_LINK_EXE)
 
-if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "8.0.0")
-  set(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS "-Wno-deprecated-gpu-targets")
-else()
-  set(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS "")
-endif()
 
 # Add implicit host link directories that contain device libraries
 # to the device link line.
@@ -51,11 +46,9 @@ endforeach()
 unset(__IMPLICT_DLINK_DIRS)
 
 set(CMAKE_CUDA_DEVICE_LINK_LIBRARY
-  "<CMAKE_CUDA_COMPILER> <LANGUAGE_COMPILE_FLAGS> ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS${__IMPLICT_DLINK_FLAGS}")
+  "<CMAKE_CUDA_COMPILER> ${_CMAKE_CUDA_EXTRA_FLAGS} <LANGUAGE_COMPILE_FLAGS> ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS${__IMPLICT_DLINK_FLAGS}")
 set(CMAKE_CUDA_DEVICE_LINK_EXECUTABLE
-  "<CMAKE_CUDA_COMPILER> <FLAGS> ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS${__IMPLICT_DLINK_FLAGS}")
-
-unset(_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS)
+  "<CMAKE_CUDA_COMPILER>  ${_CMAKE_CUDA_EXTRA_FLAGS} <FLAGS> ${_CMAKE_CUDA_EXTRA_DEVICE_LINK_FLAGS} -shared -dlink <OBJECTS> -o <TARGET> <LINK_LIBRARIES> -Xcompiler=-Fd<TARGET_COMPILE_PDB>,-FS${__IMPLICT_DLINK_FLAGS}")
 unset(__IMPLICT_DLINK_FLAGS)
 
 string(REPLACE "/D" "-D" _PLATFORM_DEFINES_CUDA "${_PLATFORM_DEFINES}${_PLATFORM_DEFINES_CXX}")