summaryrefslogtreecommitdiffstats
path: root/Tests/CudaOnly
diff options
context:
space:
mode:
Diffstat (limited to 'Tests/CudaOnly')
-rw-r--r--Tests/CudaOnly/Architecture/CMakeLists.txt5
-rw-r--r--Tests/CudaOnly/Architecture/main.cu9
-rw-r--r--Tests/CudaOnly/CMakeLists.txt42
-rw-r--r--Tests/CudaOnly/CircularLinkLine/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/CompileFlags/CMakeLists.txt15
-rw-r--r--Tests/CudaOnly/CompileFlags/main.cu13
-rw-r--r--Tests/CudaOnly/DontResolveDeviceSymbols/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/ExportPTX/CMakeLists.txt7
-rw-r--r--Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt15
-rw-r--r--Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/RuntimeControls/CMakeLists.txt3
-rw-r--r--Tests/CudaOnly/SeparateCompilation/CMakeLists.txt4
-rw-r--r--Tests/CudaOnly/Standard98/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/WithDefs/CMakeLists.txt15
-rw-r--r--Tests/CudaOnly/WithDefs/main.notcu6
15 files changed, 91 insertions, 51 deletions
diff --git a/Tests/CudaOnly/Architecture/CMakeLists.txt b/Tests/CudaOnly/Architecture/CMakeLists.txt
new file mode 100644
index 0000000..7270b56
--- /dev/null
+++ b/Tests/CudaOnly/Architecture/CMakeLists.txt
@@ -0,0 +1,5 @@
+cmake_minimum_required(VERSION 3.17)
+project(Architecture CUDA)
+
+set(CMAKE_CUDA_ARCHITECTURES 52)
+add_executable(Architecture main.cu)
diff --git a/Tests/CudaOnly/Architecture/main.cu b/Tests/CudaOnly/Architecture/main.cu
new file mode 100644
index 0000000..8c817d5
--- /dev/null
+++ b/Tests/CudaOnly/Architecture/main.cu
@@ -0,0 +1,9 @@
+#ifdef __CUDA_ARCH__
+# if __CUDA_ARCH__ != 520
+# error "Passed architecture 52, but got something else."
+# endif
+#endif
+
+int main()
+{
+}
diff --git a/Tests/CudaOnly/CMakeLists.txt b/Tests/CudaOnly/CMakeLists.txt
index 3e3e44c..8244b91 100644
--- a/Tests/CudaOnly/CMakeLists.txt
+++ b/Tests/CudaOnly/CMakeLists.txt
@@ -1,16 +1,37 @@
-ADD_TEST_MACRO(CudaOnly.CircularLinkLine CudaOnlyCircularLinkLine)
+ADD_TEST_MACRO(CudaOnly.Architecture Architecture)
+ADD_TEST_MACRO(CudaOnly.CompileFlags CudaOnlyCompileFlags)
ADD_TEST_MACRO(CudaOnly.EnableStandard CudaOnlyEnableStandard)
ADD_TEST_MACRO(CudaOnly.ExportPTX CudaOnlyExportPTX)
-ADD_TEST_MACRO(CudaOnly.GPUDebugFlag CudaOnlyGPUDebugFlag)
-ADD_TEST_MACRO(CudaOnly.ResolveDeviceSymbols CudaOnlyResolveDeviceSymbols)
-ADD_TEST_MACRO(CudaOnly.SeparateCompilation CudaOnlySeparateCompilation)
ADD_TEST_MACRO(CudaOnly.SharedRuntimePlusToolkit CudaOnlySharedRuntimePlusToolkit)
-ADD_TEST_MACRO(CudaOnly.SharedRuntimeViaCUDAFlags CudaOnlySharedRuntimeViaCUDAFlags)
ADD_TEST_MACRO(CudaOnly.Standard98 CudaOnlyStandard98)
ADD_TEST_MACRO(CudaOnly.Toolkit CudaOnlyToolkit)
ADD_TEST_MACRO(CudaOnly.WithDefs CudaOnlyWithDefs)
+if(CMake_TEST_CUDA AND NOT CMake_TEST_CUDA STREQUAL "Clang")
+ ADD_TEST_MACRO(CudaOnly.SharedRuntimeViaCUDAFlags CudaOnlySharedRuntimeViaCUDAFlags)
+
+ # Separable compilation is currently only supported on NVCC. Disable tests
+ # using it for other compilers.
+ ADD_TEST_MACRO(CudaOnly.CircularLinkLine CudaOnlyCircularLinkLine)
+ ADD_TEST_MACRO(CudaOnly.ResolveDeviceSymbols CudaOnlyResolveDeviceSymbols)
+ ADD_TEST_MACRO(CudaOnly.SeparateCompilation CudaOnlySeparateCompilation)
+
+ add_test(NAME CudaOnly.DontResolveDeviceSymbols COMMAND
+ ${CMAKE_CTEST_COMMAND} -C $<CONFIGURATION>
+ --build-and-test
+ "${CMAKE_CURRENT_SOURCE_DIR}/DontResolveDeviceSymbols/"
+ "${CMAKE_CURRENT_BINARY_DIR}/DontResolveDeviceSymbols/"
+ ${build_generator_args}
+ --build-project DontResolveDeviceSymbols
+ --build-options ${build_options}
+ --test-command ${CMAKE_CTEST_COMMAND} -V -C $<CONFIGURATION>
+ )
+
+ # Only NVCC defines __CUDACC_DEBUG__ when compiling in debug mode.
+ ADD_TEST_MACRO(CudaOnly.GPUDebugFlag CudaOnlyGPUDebugFlag)
+endif()
+
# The CUDA only ships the shared version of the toolkit libraries
# on windows
if(NOT WIN32)
@@ -21,17 +42,6 @@ if(MSVC)
ADD_TEST_MACRO(CudaOnly.PDB CudaOnlyPDB)
endif()
-add_test(NAME CudaOnly.DontResolveDeviceSymbols COMMAND
- ${CMAKE_CTEST_COMMAND} -C $<CONFIGURATION>
- --build-and-test
- "${CMAKE_CURRENT_SOURCE_DIR}/DontResolveDeviceSymbols/"
- "${CMAKE_CURRENT_BINARY_DIR}/DontResolveDeviceSymbols/"
- ${build_generator_args}
- --build-project DontResolveDeviceSymbols
- --build-options ${build_options}
- --test-command ${CMAKE_CTEST_COMMAND} -V -C $<CONFIGURATION>
- )
-
add_test(NAME CudaOnly.RuntimeControls COMMAND
${CMAKE_CTEST_COMMAND} -C $<CONFIGURATION>
--build-and-test
diff --git a/Tests/CudaOnly/CircularLinkLine/CMakeLists.txt b/Tests/CudaOnly/CircularLinkLine/CMakeLists.txt
index 5e6f7ab..e10a348 100644
--- a/Tests/CudaOnly/CircularLinkLine/CMakeLists.txt
+++ b/Tests/CudaOnly/CircularLinkLine/CMakeLists.txt
@@ -5,9 +5,9 @@ project (CircularLinkLine CUDA)
# Verify that we de-duplicate the device link line
# Verify that a de-duplicated link line still works with circular static libraries
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=[compute_30]")
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 30)
add_library(CUDACircularDeviceLinking1 STATIC file1.cu)
add_library(CUDACircularDeviceLinking2 STATIC file2.cu)
diff --git a/Tests/CudaOnly/CompileFlags/CMakeLists.txt b/Tests/CudaOnly/CompileFlags/CMakeLists.txt
new file mode 100644
index 0000000..5e8a8e4
--- /dev/null
+++ b/Tests/CudaOnly/CompileFlags/CMakeLists.txt
@@ -0,0 +1,15 @@
+cmake_minimum_required(VERSION 3.17)
+project(CompileFlags CUDA)
+
+add_executable(CudaOnlyCompileFlags main.cu)
+
+# Try passing CUDA architecture flags explicitly.
+if(CMAKE_CUDA_COMPILER_ID STREQUAL "NVIDIA")
+ target_compile_options(CudaOnlyCompileFlags PRIVATE
+ -gencode arch=compute_50,code=compute_50
+ )
+else()
+ set_property(TARGET CudaOnlyCompileFlags PROPERTY CUDA_ARCHITECTURES 50-real)
+endif()
+
+target_compile_options(CudaOnlyCompileFlags PRIVATE -DALWAYS_DEFINE)
diff --git a/Tests/CudaOnly/CompileFlags/main.cu b/Tests/CudaOnly/CompileFlags/main.cu
new file mode 100644
index 0000000..999c056
--- /dev/null
+++ b/Tests/CudaOnly/CompileFlags/main.cu
@@ -0,0 +1,13 @@
+#ifdef __CUDA_ARCH__
+# if __CUDA_ARCH__ != 500
+# error "Passed architecture 50, but got something else."
+# endif
+#endif
+
+#ifndef ALWAYS_DEFINE
+# error "ALWAYS_DEFINE not defined!"
+#endif
+
+int main()
+{
+}
diff --git a/Tests/CudaOnly/DontResolveDeviceSymbols/CMakeLists.txt b/Tests/CudaOnly/DontResolveDeviceSymbols/CMakeLists.txt
index 6e3697f..1265660 100644
--- a/Tests/CudaOnly/DontResolveDeviceSymbols/CMakeLists.txt
+++ b/Tests/CudaOnly/DontResolveDeviceSymbols/CMakeLists.txt
@@ -24,9 +24,9 @@ endif()
# Don't resolve the device symbols in the static library
# Don't resolve the device symbols in the executable library
# Verify that we can't use those device symbols from anything
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=[compute_30] -gencode arch=compute_50,code=\\\"compute_50\\\"")
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 30 50)
set(CMAKE_CUDA_RESOLVE_DEVICE_SYMBOLS OFF)
add_library(CUDANoDeviceResolve SHARED file1.cu)
diff --git a/Tests/CudaOnly/ExportPTX/CMakeLists.txt b/Tests/CudaOnly/ExportPTX/CMakeLists.txt
index ff6e77c..ee5f54d 100644
--- a/Tests/CudaOnly/ExportPTX/CMakeLists.txt
+++ b/Tests/CudaOnly/ExportPTX/CMakeLists.txt
@@ -34,16 +34,15 @@ static std::string ptx_paths = "$<TARGET_OBJECTS:CudaPTX>";
# need to also pass the --name option
set(output_file ${CMAKE_CURRENT_BINARY_DIR}/embedded_objs.h)
-get_filename_component(cuda_compiler_bin "${CMAKE_CUDA_COMPILER}" DIRECTORY)
+find_package(CUDAToolkit REQUIRED)
find_program(bin_to_c
NAMES bin2c
- PATHS ${cuda_compiler_bin}
+ PATHS ${CUDAToolkit_BIN_DIR}
)
if(NOT bin_to_c)
message(FATAL_ERROR
"bin2c not found:\n"
- " CMAKE_CUDA_COMPILER='${CMAKE_CUDA_COMPILER}'\n"
- " cuda_compiler_bin='${cuda_compiler_bin}'\n"
+ " CUDAToolkit_BIN_DIR='${CUDAToolkit_BIN_DIR}'\n"
)
endif()
diff --git a/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt b/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt
index fbef15f..6675655 100644
--- a/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt
+++ b/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt
@@ -2,18 +2,19 @@
cmake_minimum_required(VERSION 3.7)
project (GPUDebugFlag CUDA)
-#Goal for this example:
-#verify that -G enables gpu debug flags
-string(APPEND CMAKE_CUDA_FLAGS " -gencode=arch=compute_30,code=compute_30")
-string(APPEND CMAKE_CUDA_FLAGS " -G")
set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 30)
+
+# Goal for this example:
+# Verify that enabling device debug works.
+string(APPEND CMAKE_CUDA_FLAGS " -G")
add_executable(CudaOnlyGPUDebugFlag main.cu)
+#CUDA's __CUDACC_DEBUG__ define was added in NVCC 9.0
+#so if we are below 9.0.0 we will manually add the define so that the test
+#passes
if(CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 9.0.0)
- #CUDA's __CUDACC_DEBUG__ define was added in 9.0
- #so if we are below 9.0.0 we will manually add the define so that the test
- #passes
target_compile_definitions(CudaOnlyGPUDebugFlag PRIVATE "__CUDACC_DEBUG__")
endif()
diff --git a/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt b/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt
index 57aa0b9..bd94ec8 100644
--- a/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt
+++ b/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt
@@ -21,7 +21,7 @@ endif()
# confirming that the first static library is on the device link line
# 3. Verify that we can't use those device symbols from anything that links
# to the static library
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=[sm_30] -gencode arch=compute_50,code=\\\"compute_50\\\"")
+set(CMAKE_CUDA_ARCHITECTURES 30 50)
add_library(CUDAResolveDeviceDepsA STATIC file1.cu)
add_library(CUDAResolveDeviceDepsB STATIC file2.cu)
diff --git a/Tests/CudaOnly/RuntimeControls/CMakeLists.txt b/Tests/CudaOnly/RuntimeControls/CMakeLists.txt
index 8b58fec..0da5739 100644
--- a/Tests/CudaOnly/RuntimeControls/CMakeLists.txt
+++ b/Tests/CudaOnly/RuntimeControls/CMakeLists.txt
@@ -15,9 +15,8 @@ else()
endif()
endif()
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=[compute_30]")
-
set(CMAKE_CUDA_STANDARD 11)
+set(CMAKE_CUDA_ARCHITECTURES 30)
set(CMAKE_CUDA_RUNTIME_LIBRARY static)
if(NOT "x${CMAKE_CUDA_SIMULATE_ID}" STREQUAL "xMSVC")
diff --git a/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt b/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt
index c1bd64a..586be81 100644
--- a/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt
+++ b/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt
@@ -9,9 +9,7 @@ project (SeparateCompilation CUDA)
#and executables.
#We complicate the matter by also testing that multiple static libraries
#all containing cuda separable compilation code links properly
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=\\\"compute_30,sm_30,sm_35\\\"")
-string(APPEND CMAKE_CUDA_FLAGS " --generate-code=arch=compute_50,code=[compute_50,sm_50,sm_52]")
-
+set(CMAKE_CUDA_ARCHITECTURES 30 35 50 52)
set(CMAKE_CUDA_SEPARABLE_COMPILATION ON)
add_library(CUDASeparateLibA STATIC file1.cu file2.cu file3.cu)
target_compile_features(CUDASeparateLibA PRIVATE cuda_std_11)
diff --git a/Tests/CudaOnly/Standard98/CMakeLists.txt b/Tests/CudaOnly/Standard98/CMakeLists.txt
index ef9a685..3ba0360 100644
--- a/Tests/CudaOnly/Standard98/CMakeLists.txt
+++ b/Tests/CudaOnly/Standard98/CMakeLists.txt
@@ -1,7 +1,7 @@
cmake_minimum_required(VERSION 3.7)
project(CudaOnlyStandard98 CUDA)
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
+set(CMAKE_CUDA_ARCHITECTURES 30)
# Support setting CUDA Standard to 98 which internally gets transformed to
# CUDA03
diff --git a/Tests/CudaOnly/WithDefs/CMakeLists.txt b/Tests/CudaOnly/WithDefs/CMakeLists.txt
index ba9bf04..0ed81d8 100644
--- a/Tests/CudaOnly/WithDefs/CMakeLists.txt
+++ b/Tests/CudaOnly/WithDefs/CMakeLists.txt
@@ -3,17 +3,7 @@ cmake_minimum_required(VERSION 3.7)
project (WithDefs CUDA)
#verify that we can pass explicit cuda arch flags
-string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
-if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 9)
- set(debug_compile_flags --generate-code arch=compute_32,code=sm_32)
-else()
- set(debug_compile_flags --generate-code arch=compute_20,code=sm_20)
-endif()
-if(CMAKE_CUDA_SIMULATE_ID STREQUAL "MSVC")
- list(APPEND debug_compile_flags -Xcompiler=-WX)
-else()
- list(APPEND debug_compile_flags -Xcompiler=-Werror)
-endif()
+set(CMAKE_CUDA_ARCHITECTURES 30)
set(release_compile_defs DEFREL)
#Goal for this example:
@@ -28,8 +18,7 @@ target_compile_options(CudaOnlyWithDefs
PRIVATE
-DFLAG_COMPILE_LANG_$<COMPILE_LANGUAGE>
-DFLAG_LANG_IS_CUDA=$<COMPILE_LANGUAGE:CUDA>
- --compiler-options=-DHOST_DEFINE
- $<$<CONFIG:DEBUG>:$<BUILD_INTERFACE:${debug_compile_flags}>>
+ $<$<CUDA_COMPILER_ID:NVIDIA>:--compiler-options=-DHOST_DEFINE> # Host-only defines are possible only on NVCC.
)
target_compile_definitions(CudaOnlyWithDefs
diff --git a/Tests/CudaOnly/WithDefs/main.notcu b/Tests/CudaOnly/WithDefs/main.notcu
index a5f4ed6..9119eba 100644
--- a/Tests/CudaOnly/WithDefs/main.notcu
+++ b/Tests/CudaOnly/WithDefs/main.notcu
@@ -7,8 +7,10 @@
# error "INC_CUDA not defined!"
#endif
-#ifndef HOST_DEFINE
-# error "HOST_DEFINE not defined!"
+#ifdef __NVCC__
+# ifndef HOST_DEFINE
+# error "HOST_DEFINE not defined!"
+# endif
#endif
#ifndef PACKED_DEFINE
generated by cgit v0.12 at 2025-09-26 23:49:05 (GMT)