CUDA: Add test for device linking when host linking uses threads

Convert the `CudaOnly.LinkSystemDeviceLibraries` test to a new
`Cuda.ProperDeviceLibraries` test.  The former covered only the
`cublas_device` library which is removed by CUDA 10.  Extend the new
test to also cover various cases of using threads.

Issue: #18008

7 files changed, 73 insertions, 16 deletions

diff --git a/Tests/Cuda/CMakeLists.txt b/Tests/Cuda/CMakeLists.txt
index 8a43df55ae..1b3daa6c94 100644
--- a/Tests/Cuda/CMakeLists.txt
+++ b/Tests/Cuda/CMakeLists.txt
@@ -4,5 +4,6 @@ ADD_TEST_MACRO(Cuda.ConsumeCompileFeatures CudaConsumeCompileFeatures)
 ADD_TEST_MACRO(Cuda.ObjectLibrary CudaObjectLibrary)
 ADD_TEST_MACRO(Cuda.MixedStandardLevels MixedStandardLevels)
 ADD_TEST_MACRO(Cuda.ToolkitInclude CudaToolkitInclude)
+ADD_TEST_MACRO(Cuda.ProperDeviceLibraries ProperDeviceLibraries)
 ADD_TEST_MACRO(Cuda.ProperLinkFlags ProperLinkFlags)
 ADD_TEST_MACRO(Cuda.WithC CudaWithC)
diff --git a/Tests/Cuda/ProperDeviceLibraries/CMakeLists.txt b/Tests/Cuda/ProperDeviceLibraries/CMakeLists.txt
new file mode 100644
index 0000000000..cb47b09b4e
--- /dev/null
+++ b/Tests/Cuda/ProperDeviceLibraries/CMakeLists.txt
@@ -0,0 +1,45 @@
+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(THREADS_PREFER_PTHREAD_FLAG ON)
+find_package(Threads)
+
+add_executable(ProperDeviceLibraries main.cu)
+set_target_properties(ProperDeviceLibraries
+                      PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
+
+add_library(UseThreadsMixed SHARED use_pthreads.cxx use_pthreads.cu)
+target_link_libraries(UseThreadsMixed Threads::Threads)
+
+add_library(UseThreadsCuda SHARED use_pthreads.cu)
+target_link_libraries(UseThreadsCuda Threads::Threads)
+
+target_link_libraries(ProperDeviceLibraries PRIVATE UseThreadsMixed UseThreadsCuda)
+
+if(THREADS_HAVE_PTHREAD_ARG AND CMAKE_USE_PTHREADS_INIT)
+  add_library(UseExplicitPThreadsFlag SHARED use_pthreads.cu)
+  target_compile_options(UseExplicitPThreadsFlag PUBLIC "-Xcompiler=-pthread")
+  target_link_libraries(UseExplicitPThreadsFlag PUBLIC "-pthread")
+
+  add_library(UseExplicitLThreadsFlag SHARED use_pthreads.cu)
+  target_compile_options(UseExplicitLThreadsFlag PUBLIC "-Xcompiler=-pthread")
+  target_link_libraries(UseExplicitLThreadsFlag PUBLIC "-lpthread")
+
+  add_library(UseExplicitLongThreadsFlag SHARED use_pthreads.cu)
+  target_link_libraries(UseExplicitLongThreadsFlag PUBLIC "--library pthread")
+
+  target_link_libraries(ProperDeviceLibraries PRIVATE UseExplicitPThreadsFlag UseExplicitLThreadsFlag UseExplicitLongThreadsFlag)
+endif()
+
+if(CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 10.0.0)
+  #CUDA 10 removed the cublas_device library
+  target_link_libraries(ProperDeviceLibraries PRIVATE cublas_device)
+endif()
+
+if(APPLE)
+  # Help the static cuda runtime find the driver (libcuda.dyllib) at runtime.
+  set_property(TARGET ProperDeviceLibraries PROPERTY BUILD_RPATH ${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES})
+endif()
diff --git a/Tests/CudaOnly/LinkSystemDeviceLibraries/main.cu b/Tests/Cuda/ProperDeviceLibraries/main.cu
index 2c7c3887bc..8ceb0ccb4a 100644
--- a/Tests/CudaOnly/LinkSystemDeviceLibraries/main.cu
+++ b/Tests/Cuda/ProperDeviceLibraries/main.cu
@@ -3,6 +3,15 @@
 #include <cuda_runtime.h>
 #include <iostream>
 
+#if defined(USE_THREADS_POSIX) && defined(HAVE_PTHREAD_H)
+
+#  include <pthread.h>
+static int verify_linking_to_pthread()
+{
+  return static_cast<int>(pthread_self());
+}
+#endif
+
 // this test only makes sense for versions of CUDA that ships
 // static libraries that have separable compilation device symbols
 #if __CUDACC_VER_MAJOR__ <= 9
diff --git a/Tests/Cuda/ProperDeviceLibraries/use_pthreads.cu b/Tests/Cuda/ProperDeviceLibraries/use_pthreads.cu
new file mode 100644
index 0000000000..c57b8a828e
--- /dev/null
+++ b/Tests/Cuda/ProperDeviceLibraries/use_pthreads.cu
@@ -0,0 +1,9 @@
+
+#if defined(USE_THREADS_POSIX) && defined(HAVE_PTHREAD_H)
+
+#  include <pthread.h>
+static int verify_linking_to_pthread_cuda()
+{
+  return static_cast<int>(pthread_self());
+}
+#endif
diff --git a/Tests/Cuda/ProperDeviceLibraries/use_pthreads.cxx b/Tests/Cuda/ProperDeviceLibraries/use_pthreads.cxx
new file mode 100644
index 0000000000..dc7c208a48
--- /dev/null
+++ b/Tests/Cuda/ProperDeviceLibraries/use_pthreads.cxx
@@ -0,0 +1,9 @@
+
+#if defined(USE_THREADS_POSIX) && defined(HAVE_PTHREAD_H)
+
+#  include <pthread.h>
+static int verify_linking_to_pthread_cxx()
+{
+  return static_cast<int>(pthread_self());
+}
+#endif
diff --git a/Tests/CudaOnly/CMakeLists.txt b/Tests/CudaOnly/CMakeLists.txt
index 5b7c0e673f..9c4f86a31d 100644
--- a/Tests/CudaOnly/CMakeLists.txt
+++ b/Tests/CudaOnly/CMakeLists.txt
@@ -3,7 +3,6 @@ ADD_TEST_MACRO(CudaOnly.CircularLinkLine CudaOnlyCircularLinkLine)
 ADD_TEST_MACRO(CudaOnly.EnableStandard CudaOnlyEnableStandard)
 ADD_TEST_MACRO(CudaOnly.ExportPTX CudaOnlyExportPTX)
 ADD_TEST_MACRO(CudaOnly.GPUDebugFlag CudaOnlyGPUDebugFlag)
-ADD_TEST_MACRO(CudaOnly.LinkSystemDeviceLibraries CudaOnlyLinkSystemDeviceLibraries)
 ADD_TEST_MACRO(CudaOnly.ResolveDeviceSymbols CudaOnlyResolveDeviceSymbols)
 ADD_TEST_MACRO(CudaOnly.SeparateCompilation CudaOnlySeparateCompilation)
 ADD_TEST_MACRO(CudaOnly.WithDefs CudaOnlyWithDefs)
diff --git a/Tests/CudaOnly/LinkSystemDeviceLibraries/CMakeLists.txt b/Tests/CudaOnly/LinkSystemDeviceLibraries/CMakeLists.txt
deleted file mode 100644
index 7f7f606a35..0000000000
--- a/Tests/CudaOnly/LinkSystemDeviceLibraries/CMakeLists.txt
+++ /dev/null
@@ -1,15 +0,0 @@
-cmake_minimum_required(VERSION 3.8)
-project(LinkSystemDeviceLibraries 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)
-
-add_executable(CudaOnlyLinkSystemDeviceLibraries main.cu)
-set_target_properties( CudaOnlyLinkSystemDeviceLibraries
-                       PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
-target_link_libraries( CudaOnlyLinkSystemDeviceLibraries PRIVATE cublas_device)
-
-if(APPLE)
-  # Help the static cuda runtime find the driver (libcuda.dyllib) at runtime.
-  set_property(TARGET CudaOnlyLinkSystemDeviceLibraries PROPERTY BUILD_RPATH ${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES})
-endif()