Imported Upstream version 3.18.0upstream/3.18.0

12 files changed, 62 insertions, 37 deletions

diff --git a/Tests/Cuda/CMakeLists.txt b/Tests/Cuda/CMakeLists.txt
index 58b9b03c9..35ceb333c 100644
--- a/Tests/Cuda/CMakeLists.txt
+++ b/Tests/Cuda/CMakeLists.txt
@@ -1,25 +1,35 @@
+macro (add_cuda_test_macro name)
+  add_test_macro("${name}" ${ARGN})
+  set_property(TEST "${name}" APPEND
+    PROPERTY LABELS "CUDA")
+endmacro ()
 
-ADD_TEST_MACRO(Cuda.Complex CudaComplex)
-ADD_TEST_MACRO(Cuda.ConsumeCompileFeatures CudaConsumeCompileFeatures)
-ADD_TEST_MACRO(Cuda.CXXStandardSetTwice CXXStandardSetTwice)
-ADD_TEST_MACRO(Cuda.ObjectLibrary CudaObjectLibrary)
-ADD_TEST_MACRO(Cuda.MixedStandardLevels1 MixedStandardLevels1)
-ADD_TEST_MACRO(Cuda.MixedStandardLevels2 MixedStandardLevels2)
-ADD_TEST_MACRO(Cuda.MixedStandardLevels3 MixedStandardLevels3)
-ADD_TEST_MACRO(Cuda.MixedStandardLevels4 MixedStandardLevels4)
-ADD_TEST_MACRO(Cuda.MixedStandardLevels5 MixedStandardLevels5)
-ADD_TEST_MACRO(Cuda.NotEnabled CudaNotEnabled)
-ADD_TEST_MACRO(Cuda.SeparableCompCXXOnly SeparableCompCXXOnly)
-ADD_TEST_MACRO(Cuda.Toolkit Toolkit)
-ADD_TEST_MACRO(Cuda.IncludePathNoToolkit IncludePathNoToolkit)
-ADD_TEST_MACRO(Cuda.ProperDeviceLibraries ProperDeviceLibraries)
-ADD_TEST_MACRO(Cuda.ProperLinkFlags ProperLinkFlags)
-ADD_TEST_MACRO(Cuda.SharedRuntimePlusToolkit SharedRuntimePlusToolkit)
+add_cuda_test_macro(Cuda.ConsumeCompileFeatures CudaConsumeCompileFeatures)
+add_cuda_test_macro(Cuda.CXXStandardSetTwice CXXStandardSetTwice)
+add_cuda_test_macro(Cuda.ObjectLibrary CudaObjectLibrary)
+add_cuda_test_macro(Cuda.MixedStandardLevels1 MixedStandardLevels1)
+add_cuda_test_macro(Cuda.MixedStandardLevels2 MixedStandardLevels2)
+add_cuda_test_macro(Cuda.MixedStandardLevels3 MixedStandardLevels3)
+add_cuda_test_macro(Cuda.MixedStandardLevels4 MixedStandardLevels4)
+add_cuda_test_macro(Cuda.MixedStandardLevels5 MixedStandardLevels5)
+add_cuda_test_macro(Cuda.NotEnabled CudaNotEnabled)
+add_cuda_test_macro(Cuda.SeparableCompCXXOnly SeparableCompCXXOnly)
+add_cuda_test_macro(Cuda.Toolkit Toolkit)
+add_cuda_test_macro(Cuda.IncludePathNoToolkit IncludePathNoToolkit)
+add_cuda_test_macro(Cuda.SharedRuntimePlusToolkit SharedRuntimePlusToolkit)
+
+# Separable compilation is currently only supported on NVCC. Disable tests
+# using it for other compilers.
+if(CMake_TEST_CUDA AND NOT CMake_TEST_CUDA STREQUAL "Clang")
+  add_cuda_test_macro(Cuda.Complex CudaComplex)
+  add_cuda_test_macro(Cuda.ProperDeviceLibraries ProperDeviceLibraries)
+  add_cuda_test_macro(Cuda.ProperLinkFlags ProperLinkFlags)
+endif()
 
 # The CUDA only ships the shared version of the toolkit libraries
 # on windows
 if(NOT WIN32)
-  ADD_TEST_MACRO(Cuda.StaticRuntimePlusToolkit StaticRuntimePlusToolkit)
+  add_cuda_test_macro(Cuda.StaticRuntimePlusToolkit StaticRuntimePlusToolkit)
 endif()
 
-ADD_TEST_MACRO(Cuda.WithC CudaWithC)
+add_cuda_test_macro(Cuda.WithC CudaWithC)
diff --git a/Tests/Cuda/CXXStandardSetTwice/CMakeLists.txt b/Tests/Cuda/CXXStandardSetTwice/CMakeLists.txt
index 1941c4952..f4ad83a30 100644
--- a/Tests/Cuda/CXXStandardSetTwice/CMakeLists.txt
+++ b/Tests/Cuda/CXXStandardSetTwice/CMakeLists.txt
@@ -1,9 +1,8 @@
 cmake_minimum_required(VERSION 3.7)
 project(CXXStandardSetTwice CXX CUDA)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
-
 set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 30)
 
 add_executable(CXXStandardSetTwice main.cu)
 target_compile_features(CXXStandardSetTwice PUBLIC cxx_std_11)
diff --git a/Tests/Cuda/Complex/CMakeLists.txt b/Tests/Cuda/Complex/CMakeLists.txt
index 08d1e16aa..265bd85c1 100644
--- a/Tests/Cuda/Complex/CMakeLists.txt
+++ b/Tests/Cuda/Complex/CMakeLists.txt
@@ -15,7 +15,7 @@ project (Complex CXX CUDA)
 #and also building cpp targets that need cuda implicit libraries
 
 #verify that we can pass explicit cuda arch flags
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
+set(CMAKE_CUDA_ARCHITECTURES 30)
 set(CMAKE_CUDA_STANDARD 11)
 set(CMAKE_CXX_STANDARD 11)
 set(CMAKE_CUDA_STANDARD_REQUIRED TRUE)
diff --git a/Tests/Cuda/MixedStandardLevels1/CMakeLists.txt b/Tests/Cuda/MixedStandardLevels1/CMakeLists.txt
index b03e51e3e..e40ffa6da 100644
--- a/Tests/Cuda/MixedStandardLevels1/CMakeLists.txt
+++ b/Tests/Cuda/MixedStandardLevels1/CMakeLists.txt
@@ -1,10 +1,9 @@
 cmake_minimum_required(VERSION 3.7)
 project(MixedStandardLevels1 CXX CUDA)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
-
 set(CMAKE_CXX_STANDARD 14)
 set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 30)
 
 add_executable(MixedStandardLevels1 main.cu lib.cpp)
 
diff --git a/Tests/Cuda/MixedStandardLevels2/CMakeLists.txt b/Tests/Cuda/MixedStandardLevels2/CMakeLists.txt
index 12dd3285f..7af8081fa 100644
--- a/Tests/Cuda/MixedStandardLevels2/CMakeLists.txt
+++ b/Tests/Cuda/MixedStandardLevels2/CMakeLists.txt
@@ -1,9 +1,8 @@
 cmake_minimum_required(VERSION 3.7)
 project(MixedStandardLevels2 CXX CUDA)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
-
 set(CMAKE_CXX_STANDARD 17) #this can decay
+set(CMAKE_CUDA_ARCHITECTURES 30)
 
 add_executable(MixedStandardLevels2 main.cu lib.cpp)
 target_compile_features(MixedStandardLevels2 PUBLIC cuda_std_11)
diff --git a/Tests/Cuda/MixedStandardLevels3/CMakeLists.txt b/Tests/Cuda/MixedStandardLevels3/CMakeLists.txt
index 2b611be8e..2c420030a 100644
--- a/Tests/Cuda/MixedStandardLevels3/CMakeLists.txt
+++ b/Tests/Cuda/MixedStandardLevels3/CMakeLists.txt
@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.7)
 project(MixedStandardLevels3 CXX CUDA)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
+set(CMAKE_CUDA_ARCHITECTURES 30)
 
 add_executable(MixedStandardLevels3 main.cu lib.cpp)
 target_compile_features(MixedStandardLevels3 PUBLIC cuda_std_03 cxx_std_14)
diff --git a/Tests/Cuda/MixedStandardLevels4/CMakeLists.txt b/Tests/Cuda/MixedStandardLevels4/CMakeLists.txt
index faf68692f..230230d0d 100644
--- a/Tests/Cuda/MixedStandardLevels4/CMakeLists.txt
+++ b/Tests/Cuda/MixedStandardLevels4/CMakeLists.txt
@@ -1,9 +1,8 @@
 cmake_minimum_required(VERSION 3.7)
 project(MixedStandardLevels4 CXX CUDA)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
-
 set(CMAKE_CUDA_STANDARD 03)
+set(CMAKE_CUDA_ARCHITECTURES 30)
 
 add_executable(MixedStandardLevels4 main.cu lib.cpp)
 target_compile_features(MixedStandardLevels4 PUBLIC cxx_std_14)
diff --git a/Tests/Cuda/MixedStandardLevels5/CMakeLists.txt b/Tests/Cuda/MixedStandardLevels5/CMakeLists.txt
index 7209f6085..5f5ee0637 100644
--- a/Tests/Cuda/MixedStandardLevels5/CMakeLists.txt
+++ b/Tests/Cuda/MixedStandardLevels5/CMakeLists.txt
@@ -1,9 +1,8 @@
 cmake_minimum_required(VERSION 3.7)
 project(MixedStandardLevels5 CXX CUDA)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
-
 set(CMAKE_CXX_STANDARD 98)
+set(CMAKE_CUDA_ARCHITECTURES 30)
 
 add_executable(MixedStandardLevels5 main.cu lib.cpp)
 
diff --git a/Tests/Cuda/ProperDeviceLibraries/CMakeLists.txt b/Tests/Cuda/ProperDeviceLibraries/CMakeLists.txt
index cb47b09b4..fe28c3edc 100644
--- a/Tests/Cuda/ProperDeviceLibraries/CMakeLists.txt
+++ b/Tests/Cuda/ProperDeviceLibraries/CMakeLists.txt
@@ -1,8 +1,8 @@
 cmake_minimum_required(VERSION 3.13)
 project(ProperDeviceLibraries CXX CUDA)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_35,code=compute_35 -gencode arch=compute_35,code=sm_35")
 set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 35)
 
 set(THREADS_PREFER_PTHREAD_FLAG ON)
 find_package(Threads)
diff --git a/Tests/Cuda/ProperLinkFlags/CMakeLists.txt b/Tests/Cuda/ProperLinkFlags/CMakeLists.txt
index b6e0e394b..d38da6d14 100644
--- a/Tests/Cuda/ProperLinkFlags/CMakeLists.txt
+++ b/Tests/Cuda/ProperLinkFlags/CMakeLists.txt
@@ -9,11 +9,17 @@ project (ProperLinkFlags CUDA CXX)
 
 #Specify a set of valid CUDA flags and an invalid set of CXX flags ( for CUDA )
 #to make sure we don't use the CXX flags when linking CUDA executables
-string(APPEND CMAKE_CUDA_FLAGS " -arch=sm_35 --use_fast_math")
+if(CMAKE_CUDA_COMPILER_ID STREQUAL "NVIDIA")
+  string(APPEND CMAKE_CUDA_FLAGS "--use_fast_math")
+elseif(CMAKE_CUDA_COMPILER_ID STREQUAL "Clang")
+  string(APPEND CMAKE_CUDA_FLAGS "-ffast-math")
+endif()
+
 set(CMAKE_CXX_FLAGS "-Wall")
 
 set(CMAKE_CXX_STANDARD 11)
 set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 35)
 add_executable(ProperLinkFlags file1.cu main.cxx)
 
 set_target_properties( ProperLinkFlags
diff --git a/Tests/Cuda/Toolkit/CMakeLists.txt b/Tests/Cuda/Toolkit/CMakeLists.txt
index 86b465219..00125e3b9 100644
--- a/Tests/Cuda/Toolkit/CMakeLists.txt
+++ b/Tests/Cuda/Toolkit/CMakeLists.txt
@@ -14,15 +14,29 @@ message(STATUS "CUDAToolkit_INCLUDE_DIRS: ${CUDAToolkit_INCLUDE_DIRS}")
 message(STATUS "CUDAToolkit_LIBRARY_DIR: ${CUDAToolkit_LIBRARY_DIR}")
 message(STATUS "CUDAToolkit_NVCC_EXECUTABLE ${CUDAToolkit_NVCC_EXECUTABLE}")
 
-# Verify that all the CUDA:: targets exist even when the CUDA language isn't enabled
+set(cuda_libs cudart cuda_driver cublas cufft cufftw curand cusolver cusparse)
+if(CUDAToolkit_VERSION VERSION_GREATER_EQUAL 10.1)
+  list(APPEND cuda_libs cublasLt)
+endif()
+if(CUDAToolkit_VERSION_MAJOR VERSION_LESS 11)
+  list(APPEND cuda_libs nvgraph)
+endif()
 
-foreach (cuda_lib cudart cuda_driver cublas cufft cufftw curand cusolver cusparse nvgraph)
+# Verify that all the CUDA:: targets exist even when the CUDA language isn't enabled
+foreach (cuda_lib IN LISTS cuda_libs)
+  if(NOT CUDA_${cuda_lib}_LIBRARY)
+    message(FATAL_ERROR "expected CUDAToolkit variable CUDA_${cuda_lib}_LIBRARY not found")
+  endif()
   if(NOT TARGET CUDA::${cuda_lib})
-    message(FATAL_ERROR "The CUDA::${cuda_lib} target was expected but couldn't be found")
+    message(FATAL_ERROR "expected CUDAToolkit target CUDA::${cuda_lib} not found")
   endif()
 endforeach()
 
-foreach (cuda_lib nppc nppial nppicc nppidei nppif nppig nppim nppist nppitc npps nppicom nppisu)
+set(npp_libs nppc nppial nppicc nppidei nppif nppig nppim nppist nppitc npps nppisu)
+if(CUDAToolkit_VERSION_MAJOR VERSION_LESS 11)
+  list(APPEND npp_libs nppicom)
+endif()
+foreach (cuda_lib IN LISTS npp_libs)
   if(NOT TARGET CUDA::${cuda_lib})
     message(FATAL_ERROR "The CUDA::${cuda_lib} target was expected but couldn't be found")
   endif()
diff --git a/Tests/Cuda/WithC/CMakeLists.txt b/Tests/Cuda/WithC/CMakeLists.txt
index 69aa3f992..049cbce1f 100644
--- a/Tests/Cuda/WithC/CMakeLists.txt
+++ b/Tests/Cuda/WithC/CMakeLists.txt
@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.7)
 project(WithC CUDA C)
 
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
+set(CMAKE_CUDA_ARCHITECTURES 30)
 
 add_executable(CudaWithC main.c cuda.cu)