12 files changed, 120 insertions, 38 deletions
diff --git a/Tests/CudaOnly/CMakeLists.txt b/Tests/CudaOnly/CMakeLists.txt
index 39634ac..db08076 100644
--- a/Tests/CudaOnly/CMakeLists.txt
+++ b/Tests/CudaOnly/CMakeLists.txt
@@ -10,6 +10,7 @@ add_cuda_test_macro(CudaOnly.CompileFlags CudaOnlyCompileFlags)
 add_cuda_test_macro(CudaOnly.EnableStandard CudaOnlyEnableStandard)
 add_cuda_test_macro(CudaOnly.ExportPTX CudaOnlyExportPTX)
 add_cuda_test_macro(CudaOnly.SharedRuntimePlusToolkit CudaOnlySharedRuntimePlusToolkit)
+add_cuda_test_macro(CudaOnly.StaticRuntimePlusToolkit CudaOnlyStaticRuntimePlusToolkit)
 add_cuda_test_macro(CudaOnly.Standard98 CudaOnlyStandard98)
 add_cuda_test_macro(CudaOnly.Toolkit CudaOnlyToolkit)
 add_cuda_test_macro(CudaOnly.ToolkitBeforeLang CudaOnlyToolkitBeforeLang)
@@ -28,12 +29,6 @@ if(CMake_TEST_CUDA AND NOT CMake_TEST_CUDA STREQUAL "Clang")
   add_cuda_test_macro(CudaOnly.GPUDebugFlag CudaOnlyGPUDebugFlag)
 endif()
 
-# The CUDA only ships the shared version of the toolkit libraries
-# on windows
-if(NOT WIN32)
-  add_cuda_test_macro(CudaOnly.StaticRuntimePlusToolkit CudaOnlyStaticRuntimePlusToolkit)
-endif()
-
 add_cuda_test_macro(CudaOnly.DeviceLTO CudaOnlyDeviceLTO)
 
 if(MSVC)
diff --git a/Tests/CudaOnly/SeparateCompilationPTX/main.cu b/Tests/CudaOnly/SeparateCompilationPTX/main.cu
index 164cde5..f94beff 100644
--- a/Tests/CudaOnly/SeparateCompilationPTX/main.cu
+++ b/Tests/CudaOnly/SeparateCompilationPTX/main.cu
@@ -21,10 +21,11 @@ int main()
   cuCtxCreate(&context, 0, device);
 
   CUmodule module;
-  cuModuleLoadData(&module, kernels);
-  if (module == nullptr) {
-    std::cerr << "Failed to load the embedded ptx" << std::endl;
+  CUresult result = cuModuleLoadData(&module, kernels);
+  std::cout << "module pointer: " << module << '\n';
+  if (result != CUDA_SUCCESS || module == nullptr) {
+    std::cerr << "Failed to load the embedded ptx with error: "
+              << static_cast<unsigned int>(result) << '\n';
     return 1;
   }
-  std::cout << module << std::endl;
 }
diff --git a/Tests/CudaOnly/SharedRuntimePlusToolkit/CMakeLists.txt b/Tests/CudaOnly/SharedRuntimePlusToolkit/CMakeLists.txt
index 03fba22..0b01085 100644
--- a/Tests/CudaOnly/SharedRuntimePlusToolkit/CMakeLists.txt
+++ b/Tests/CudaOnly/SharedRuntimePlusToolkit/CMakeLists.txt
@@ -16,16 +16,18 @@ target_link_libraries(SharedToolkit PRIVATE Common PUBLIC CUDA::curand CUDA::npp
 set_target_properties(SharedToolkit PROPERTIES CUDA_RUNTIME_LIBRARY none)
 target_link_libraries(SharedToolkit PUBLIC CUDA::cudart)
 
-# The CUDA only ships the shared version of the toolkit libraries
-# on windows
-if(NOT WIN32)
+# Verify the CUDA Toolkit has static libraries
+if(TARGET CUDA::curand_static AND
+   TARGET CUDA::nppif_static)
   #shared runtime with static toolkit libraries
   add_library(StaticToolkit SHARED static.cu)
+  target_compile_definitions(StaticToolkit INTERFACE "HAS_STATIC_VERSION")
   target_link_libraries(StaticToolkit PRIVATE Common CUDA::curand_static CUDA::nppif_static)
   set_target_properties(StaticToolkit PROPERTIES CUDA_RUNTIME_LIBRARY Shared)
 
-  #static runtime with mixed toolkit libraries
+  #shared runtime with mixed toolkit libraries
   add_library(MixedToolkit SHARED mixed.cu)
+  target_compile_definitions(MixedToolkit INTERFACE "HAS_MIXED_VERSION")
   target_link_libraries(MixedToolkit PRIVATE Common CUDA::curand_static CUDA::nppif)
   set_target_properties(MixedToolkit PROPERTIES CUDA_RUNTIME_LIBRARY Shared)
 endif()
diff --git a/Tests/CudaOnly/SharedRuntimePlusToolkit/main.cu b/Tests/CudaOnly/SharedRuntimePlusToolkit/main.cu
index 2a4da22..d958c3a 100644
--- a/Tests/CudaOnly/SharedRuntimePlusToolkit/main.cu
+++ b/Tests/CudaOnly/SharedRuntimePlusToolkit/main.cu
@@ -1,19 +1,28 @@
 
 #ifdef _WIN32
 #  define IMPORT __declspec(dllimport)
+#else
+#  define IMPORT
+#endif
+
 IMPORT int shared_version();
+
+#ifdef HAS_STATIC_VERSION
+IMPORT int static_version();
+#else
 int static_version()
 {
   return 0;
 }
+#endif
+
+#ifdef HAS_MIXED_VERSION
+IMPORT int mixed_version();
+#else
 int mixed_version()
 {
   return 0;
 }
-#else
-int shared_version();
-int static_version();
-int mixed_version();
 #endif
 
 int main()
diff --git a/Tests/CudaOnly/StaticRuntimePlusToolkit/CMakeLists.txt b/Tests/CudaOnly/StaticRuntimePlusToolkit/CMakeLists.txt
index 534bab2..ae03b66 100644
--- a/Tests/CudaOnly/StaticRuntimePlusToolkit/CMakeLists.txt
+++ b/Tests/CudaOnly/StaticRuntimePlusToolkit/CMakeLists.txt
@@ -16,17 +16,25 @@ target_link_libraries(SharedToolkit PRIVATE Common CUDA::curand CUDA::nppif )
 set_target_properties(SharedToolkit PROPERTIES CUDA_RUNTIME_LIBRARY none)
 target_link_libraries(SharedToolkit PUBLIC CUDA::cudart_static)
 
-#static runtime with static toolkit libraries
-add_library(StaticToolkit SHARED static.cu)
-target_link_libraries(StaticToolkit PRIVATE Common CUDA::curand_static CUDA::nppif_static)
+# Verify the CUDA Toolkit has static libraries
+if(TARGET CUDA::curand_static AND
+   TARGET CUDA::nppif_static)
+  #static runtime with static toolkit libraries
+  add_library(StaticToolkit SHARED static.cu)
+  target_compile_definitions(StaticToolkit INTERFACE "HAS_STATIC_VERSION")
+  target_link_libraries(StaticToolkit PRIVATE Common CUDA::curand_static CUDA::nppif_static)
 
-#static runtime with mixed toolkit libraries
-add_library(MixedToolkit SHARED mixed.cu)
-target_link_libraries(MixedToolkit PRIVATE Common CUDA::curand CUDA::nppif_static)
-set_target_properties(MixedToolkit PROPERTIES CUDA_RUNTIME_LIBRARY Static)
+  #static runtime with mixed toolkit libraries
+  add_library(MixedToolkit SHARED mixed.cu)
+  target_compile_definitions(MixedToolkit INTERFACE "HAS_MIXED_VERSION")
+  target_link_libraries(MixedToolkit PRIVATE Common CUDA::curand CUDA::nppif_static)
+  set_target_properties(MixedToolkit PROPERTIES CUDA_RUNTIME_LIBRARY Static)
+endif()
 
 add_executable(CudaOnlyStaticRuntimePlusToolkit main.cu)
-target_link_libraries(CudaOnlyStaticRuntimePlusToolkit PRIVATE SharedToolkit StaticToolkit MixedToolkit)
+target_link_libraries(CudaOnlyStaticRuntimePlusToolkit PRIVATE SharedToolkit
+                      $<TARGET_NAME_IF_EXISTS:StaticToolkit>
+                      $<TARGET_NAME_IF_EXISTS:MixedToolkit>)
 
 if(UNIX)
   # Help the shared cuda runtime find libcurand and libnppif when they are not located
diff --git a/Tests/CudaOnly/StaticRuntimePlusToolkit/curand.cu b/Tests/CudaOnly/StaticRuntimePlusToolkit/curand.cu
index 95872f0..fdd7b53 100644
--- a/Tests/CudaOnly/StaticRuntimePlusToolkit/curand.cu
+++ b/Tests/CudaOnly/StaticRuntimePlusToolkit/curand.cu
@@ -1,6 +1,12 @@
 // Comes from:
 // https://docs.nvidia.com/cuda/curand/host-api-overview.html#host-api-example
 
+#ifdef _WIN32
+#  define EXPORT __declspec(dllexport)
+#else
+#  define EXPORT
+#endif
+
 /*
  * This program uses the host CURAND API to generate 100
  * pseudorandom floats.
@@ -25,7 +31,7 @@
     }                                                                         \
   } while (0)
 
-int curand_main()
+EXPORT int curand_main()
 {
   size_t n = 100;
   size_t i;
diff --git a/Tests/CudaOnly/StaticRuntimePlusToolkit/main.cu b/Tests/CudaOnly/StaticRuntimePlusToolkit/main.cu
index 5a09f8e..d958c3a 100644
--- a/Tests/CudaOnly/StaticRuntimePlusToolkit/main.cu
+++ b/Tests/CudaOnly/StaticRuntimePlusToolkit/main.cu
@@ -1,8 +1,29 @@
 
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+#else
+#  define IMPORT
+#endif
 
-int shared_version();
-int static_version();
-int mixed_version();
+IMPORT int shared_version();
+
+#ifdef HAS_STATIC_VERSION
+IMPORT int static_version();
+#else
+int static_version()
+{
+  return 0;
+}
+#endif
+
+#ifdef HAS_MIXED_VERSION
+IMPORT int mixed_version();
+#else
+int mixed_version()
+{
+  return 0;
+}
+#endif
 
 int main()
 {
diff --git a/Tests/CudaOnly/StaticRuntimePlusToolkit/mixed.cu b/Tests/CudaOnly/StaticRuntimePlusToolkit/mixed.cu
index a05140d..6de6886 100644
--- a/Tests/CudaOnly/StaticRuntimePlusToolkit/mixed.cu
+++ b/Tests/CudaOnly/StaticRuntimePlusToolkit/mixed.cu
@@ -1,8 +1,16 @@
 
-int curand_main();
-int nppif_main();
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+#  define EXPORT __declspec(dllexport)
+#else
+#  define IMPORT
+#  define EXPORT
+#endif
 
-int mixed_version()
+IMPORT int curand_main();
+IMPORT int nppif_main();
+
+EXPORT int mixed_version()
 {
   return curand_main() == 0 && nppif_main() == 0;
 }
diff --git a/Tests/CudaOnly/StaticRuntimePlusToolkit/nppif.cu b/Tests/CudaOnly/StaticRuntimePlusToolkit/nppif.cu
index 2871090..ac5341c 100644
--- a/Tests/CudaOnly/StaticRuntimePlusToolkit/nppif.cu
+++ b/Tests/CudaOnly/StaticRuntimePlusToolkit/nppif.cu
@@ -1,6 +1,12 @@
 // Comes from
 // https://devtalk.nvidia.com/default/topic/1037482/gpu-accelerated-libraries/help-me-help-you-with-modern-cmake-and-cuda-mwe-for-npp/post/5271066/#5271066
 
+#ifdef _WIN32
+#  define EXPORT __declspec(dllexport)
+#else
+#  define EXPORT
+#endif
+
 #include <cstdio>
 #include <iostream>
 
@@ -8,7 +14,7 @@
 #include <cuda_runtime_api.h>
 #include <nppi_filtering_functions.h>
 
-int nppif_main()
+EXPORT int nppif_main()
 {
   /**
    * 8-bit unsigned single-channel 1D row convolution.
diff --git a/Tests/CudaOnly/StaticRuntimePlusToolkit/shared.cu b/Tests/CudaOnly/StaticRuntimePlusToolkit/shared.cu
index 9967b66..f3c3dbc 100644
--- a/Tests/CudaOnly/StaticRuntimePlusToolkit/shared.cu
+++ b/Tests/CudaOnly/StaticRuntimePlusToolkit/shared.cu
@@ -1,8 +1,16 @@
 
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+#  define EXPORT __declspec(dllexport)
+#else
+#  define IMPORT
+#  define EXPORT
+#endif
+
 int curand_main();
 int nppif_main();
 
-int shared_version()
+EXPORT int shared_version()
 {
   return curand_main() == 0 && nppif_main() == 0;
 }
diff --git a/Tests/CudaOnly/StaticRuntimePlusToolkit/static.cu b/Tests/CudaOnly/StaticRuntimePlusToolkit/static.cu
index ca7eb4c..6932fa3 100644
--- a/Tests/CudaOnly/StaticRuntimePlusToolkit/static.cu
+++ b/Tests/CudaOnly/StaticRuntimePlusToolkit/static.cu
@@ -1,8 +1,16 @@
 
-int curand_main();
-int nppif_main();
+#ifdef _WIN32
+#  define IMPORT __declspec(dllimport)
+#  define EXPORT __declspec(dllexport)
+#else
+#  define IMPORT
+#  define EXPORT
+#endif
 
-int static_version()
+IMPORT int curand_main();
+IMPORT int nppif_main();
+
+EXPORT int static_version()
 {
   return curand_main() == 0 && nppif_main() == 0;
 }
diff --git a/Tests/CudaOnly/Toolkit/CMakeLists.txt b/Tests/CudaOnly/Toolkit/CMakeLists.txt
index 1486c1a..e0801a3 100644
--- a/Tests/CudaOnly/Toolkit/CMakeLists.txt
+++ b/Tests/CudaOnly/Toolkit/CMakeLists.txt
@@ -20,14 +20,20 @@ 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_GREATER 11)
+  list(APPEND cuda_libs nvJitLink)
+endif()
 if(CUDAToolkit_VERSION_MAJOR VERSION_LESS 11)
   list(APPEND cuda_libs nvgraph)
 endif()
 
+
 # Verify that all the CUDA:: targets and variables exist
 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")
+  elseif(CUDA_${cuda_lib}_LIBRARY MATCHES [[\.\./]])
+    message(FATAL_ERROR "expected CUDAToolkit variable CUDA_${cuda_lib}_LIBRARY[\"${CUDA_${cuda_lib}_LIBRARY}\"] to not contain /..")
   endif()
   if(NOT TARGET CUDA::${cuda_lib})
     message(FATAL_ERROR "expected CUDAToolkit target CUDA::${cuda_lib} not found")
@@ -41,6 +47,8 @@ endif()
 foreach (cuda_lib )
   if(NOT CUDA_${cuda_lib}_LIBRARY)
     message(FATAL_ERROR "expected CUDAToolkit variable CUDA_${cuda_lib}_LIBRARY not found")
+  elseif(CUDA_${cuda_lib}_LIBRARY MATCHES [[\.\./]])
+    message(FATAL_ERROR "expected CUDAToolkit variable CUDA_${cuda_lib}_LIBRARY[\"${CUDA_${cuda_lib}_LIBRARY}\"] to not contain /..")
   endif()
   if(NOT TARGET CUDA::${cuda_lib})
     message(FATAL_ERROR "expected CUDAToolkit target CUDA::${cuda_lib} not found")
@@ -50,6 +58,8 @@ endforeach()
 foreach (cuda_lib nvrtc nvToolsExt OpenCL)
   if(NOT CUDA_${cuda_lib}_LIBRARY)
     message(FATAL_ERROR "expected CUDAToolkit variable CUDA_${cuda_lib}_LIBRARY not found")
+  elseif(CUDA_${cuda_lib}_LIBRARY MATCHES [[\.\./]])
+    message(FATAL_ERROR "expected CUDAToolkit variable CUDA_${cuda_lib}_LIBRARY[\"${CUDA_${cuda_lib}_LIBRARY}\"] to not contain /..")
   endif()
 
   if(NOT TARGET CUDA::${cuda_lib})