summaryrefslogtreecommitdiffstats
path: root/Tests/CudaOnly
diff options
context:
space:
mode:
Diffstat (limited to 'Tests/CudaOnly')
-rw-r--r--Tests/CudaOnly/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/CircularLinkLine/CMakeLists.txt34
-rw-r--r--Tests/CudaOnly/CircularLinkLine/file1.cu6
-rw-r--r--Tests/CudaOnly/CircularLinkLine/file2.cu6
-rw-r--r--Tests/CudaOnly/CircularLinkLine/file3.cu8
-rw-r--r--Tests/CudaOnly/CircularLinkLine/main.cu5
-rw-r--r--Tests/CudaOnly/EnableStandard/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/ExportPTX/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/LinkSystemDeviceLibraries/CMakeLists.txt15
-rw-r--r--Tests/CudaOnly/LinkSystemDeviceLibraries/main.cu83
-rw-r--r--Tests/CudaOnly/PDB/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/SeparateCompilation/CMakeLists.txt2
-rw-r--r--Tests/CudaOnly/WithDefs/CMakeLists.txt2
15 files changed, 67 insertions, 106 deletions
diff --git a/Tests/CudaOnly/CMakeLists.txt b/Tests/CudaOnly/CMakeLists.txt
index 59f3e84..9c4f86a 100644
--- a/Tests/CudaOnly/CMakeLists.txt
+++ b/Tests/CudaOnly/CMakeLists.txt
@@ -1,8 +1,8 @@
+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/CircularLinkLine/CMakeLists.txt b/Tests/CudaOnly/CircularLinkLine/CMakeLists.txt
new file mode 100644
index 0000000..5e6f7ab
--- /dev/null
+++ b/Tests/CudaOnly/CircularLinkLine/CMakeLists.txt
@@ -0,0 +1,34 @@
+cmake_minimum_required(VERSION 3.7)
+project (CircularLinkLine CUDA)
+
+#Goal for this example:
+# 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)
+
+add_library(CUDACircularDeviceLinking1 STATIC file1.cu)
+add_library(CUDACircularDeviceLinking2 STATIC file2.cu)
+add_library(CUDACircularDeviceLinking3 STATIC file3.cu)
+add_executable(CudaOnlyCircularLinkLine main.cu)
+
+target_link_libraries(CUDACircularDeviceLinking1 PUBLIC CUDACircularDeviceLinking2)
+target_link_libraries(CUDACircularDeviceLinking2 PUBLIC CUDACircularDeviceLinking3)
+target_link_libraries(CUDACircularDeviceLinking3 PUBLIC CUDACircularDeviceLinking1)
+
+target_link_libraries(CudaOnlyCircularLinkLine PRIVATE CUDACircularDeviceLinking3)
+
+
+set_target_properties(CUDACircularDeviceLinking1
+ PROPERTIES
+ CUDA_SEPARABLE_COMPILATION ON)
+
+set_target_properties(CUDACircularDeviceLinking2
+ PROPERTIES
+ CUDA_SEPARABLE_COMPILATION ON)
+
+set_target_properties(CUDACircularDeviceLinking3
+ PROPERTIES
+ CUDA_SEPARABLE_COMPILATION ON)
diff --git a/Tests/CudaOnly/CircularLinkLine/file1.cu b/Tests/CudaOnly/CircularLinkLine/file1.cu
new file mode 100644
index 0000000..88ac4e3
--- /dev/null
+++ b/Tests/CudaOnly/CircularLinkLine/file1.cu
@@ -0,0 +1,6 @@
+
+extern __device__ int file2_func(int);
+int __device__ file1_func(int x)
+{
+ return file2_func(x);
+}
diff --git a/Tests/CudaOnly/CircularLinkLine/file2.cu b/Tests/CudaOnly/CircularLinkLine/file2.cu
new file mode 100644
index 0000000..b32dbff
--- /dev/null
+++ b/Tests/CudaOnly/CircularLinkLine/file2.cu
@@ -0,0 +1,6 @@
+
+extern __device__ int file3_func(int);
+int __device__ file2_func(int x)
+{
+ return x + file3_func(x);
+}
diff --git a/Tests/CudaOnly/CircularLinkLine/file3.cu b/Tests/CudaOnly/CircularLinkLine/file3.cu
new file mode 100644
index 0000000..7f67187
--- /dev/null
+++ b/Tests/CudaOnly/CircularLinkLine/file3.cu
@@ -0,0 +1,8 @@
+
+extern __device__ int file1_func(int);
+int __device__ file3_func(int x)
+{
+ if (x > 0)
+ return file1_func(-x);
+ return x;
+}
diff --git a/Tests/CudaOnly/CircularLinkLine/main.cu b/Tests/CudaOnly/CircularLinkLine/main.cu
new file mode 100644
index 0000000..1c19e8d
--- /dev/null
+++ b/Tests/CudaOnly/CircularLinkLine/main.cu
@@ -0,0 +1,5 @@
+
+int main(int argc, char** argv)
+{
+ return 0;
+}
diff --git a/Tests/CudaOnly/EnableStandard/CMakeLists.txt b/Tests/CudaOnly/EnableStandard/CMakeLists.txt
index 35a1deb..54e2c14 100644
--- a/Tests/CudaOnly/EnableStandard/CMakeLists.txt
+++ b/Tests/CudaOnly/EnableStandard/CMakeLists.txt
@@ -1,6 +1,6 @@
cmake_minimum_required(VERSION 3.7)
-project (CudaOnlyEnableStandard CUDA)
+project (EnableStandard CUDA)
#Goal for this example:
#build cuda sources that require C++11 to be enabled.
diff --git a/Tests/CudaOnly/ExportPTX/CMakeLists.txt b/Tests/CudaOnly/ExportPTX/CMakeLists.txt
index 65d5243..ff6e77c 100644
--- a/Tests/CudaOnly/ExportPTX/CMakeLists.txt
+++ b/Tests/CudaOnly/ExportPTX/CMakeLists.txt
@@ -1,5 +1,5 @@
cmake_minimum_required(VERSION 3.8)
-project (CudaOnlyExportPTX CUDA)
+project (ExportPTX CUDA)
#Goal for this example:
# How to generate PTX files instead of OBJECT files
diff --git a/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt b/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt
index 5b96906..fbef15f 100644
--- a/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt
+++ b/Tests/CudaOnly/GPUDebugFlag/CMakeLists.txt
@@ -1,6 +1,6 @@
cmake_minimum_required(VERSION 3.7)
-project (CudaOnlGPUDebugFlag CUDA)
+project (GPUDebugFlag CUDA)
#Goal for this example:
#verify that -G enables gpu debug flags
diff --git a/Tests/CudaOnly/LinkSystemDeviceLibraries/CMakeLists.txt b/Tests/CudaOnly/LinkSystemDeviceLibraries/CMakeLists.txt
deleted file mode 100644
index 62be1e6..0000000
--- a/Tests/CudaOnly/LinkSystemDeviceLibraries/CMakeLists.txt
+++ /dev/null
@@ -1,15 +0,0 @@
-cmake_minimum_required(VERSION 3.8)
-project(CudaOnlyLinkSystemDeviceLibraries 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()
diff --git a/Tests/CudaOnly/LinkSystemDeviceLibraries/main.cu b/Tests/CudaOnly/LinkSystemDeviceLibraries/main.cu
deleted file mode 100644
index 2c7c388..0000000
--- a/Tests/CudaOnly/LinkSystemDeviceLibraries/main.cu
+++ /dev/null
@@ -1,83 +0,0 @@
-
-#include <cublas_v2.h>
-#include <cuda_runtime.h>
-#include <iostream>
-
-// this test only makes sense for versions of CUDA that ships
-// static libraries that have separable compilation device symbols
-#if __CUDACC_VER_MAJOR__ <= 9
-__global__ void deviceCublasSgemm(int n, float alpha, float beta,
- const float* d_A, const float* d_B,
- float* d_C)
-{
- cublasHandle_t cnpHandle;
- cublasStatus_t status = cublasCreate(&cnpHandle);
-
- if (status != CUBLAS_STATUS_SUCCESS) {
- return;
- }
-
- // Call function defined in the cublas_device system static library.
- // This way we can verify that we properly pass system libraries to the
- // device link line
- status = cublasSgemm(cnpHandle, CUBLAS_OP_N, CUBLAS_OP_N, n, n, n, &alpha,
- d_A, n, d_B, n, &beta, d_C, n);
-
- cublasDestroy(cnpHandle);
-}
-#endif
-
-int choose_cuda_device()
-{
- int nDevices = 0;
- cudaError_t err = cudaGetDeviceCount(&nDevices);
- if (err != cudaSuccess) {
- std::cerr << "Failed to retrieve the number of CUDA enabled devices"
- << std::endl;
- return 1;
- }
- for (int i = 0; i < nDevices; ++i) {
- cudaDeviceProp prop;
- cudaError_t err = cudaGetDeviceProperties(&prop, i);
- if (err != cudaSuccess) {
- std::cerr << "Could not retrieve properties from CUDA device " << i
- << std::endl;
- return 1;
- }
-
- if (prop.major > 3 || (prop.major == 3 && prop.minor >= 5)) {
- err = cudaSetDevice(i);
- if (err != cudaSuccess) {
- std::cout << "Could not select CUDA device " << i << std::endl;
- } else {
- return 0;
- }
- }
- }
-
- std::cout << "Could not find a CUDA enabled card supporting compute >=3.5"
- << std::endl;
- return 1;
-}
-
-int main(int argc, char** argv)
-{
- int ret = choose_cuda_device();
- if (ret) {
- return 0;
- }
-
-#if __CUDACC_VER_MAJOR__ <= 9
- // initial values that will make sure that the cublasSgemm won't actually
- // do any work
- int n = 0;
- float alpha = 1;
- float beta = 1;
- float* d_A = nullptr;
- float* d_B = nullptr;
- float* d_C = nullptr;
- deviceCublasSgemm<<<1, 1>>>(n, alpha, beta, d_A, d_B, d_C);
-#endif
-
- return 0;
-}
diff --git a/Tests/CudaOnly/PDB/CMakeLists.txt b/Tests/CudaOnly/PDB/CMakeLists.txt
index 34e1e5c..6ecf989 100644
--- a/Tests/CudaOnly/PDB/CMakeLists.txt
+++ b/Tests/CudaOnly/PDB/CMakeLists.txt
@@ -1,5 +1,5 @@
cmake_minimum_required(VERSION 3.11)
-project (CudaOnlyPDB CUDA)
+project (PDB CUDA)
add_executable(CudaOnlyPDB main.cu)
set_target_properties(CudaOnlyPDB PROPERTIES
diff --git a/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt b/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt
index 0c453a9..796e133 100644
--- a/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt
+++ b/Tests/CudaOnly/ResolveDeviceSymbols/CMakeLists.txt
@@ -1,5 +1,5 @@
cmake_minimum_required(VERSION 3.7)
-project (CudaOnlyResolveDeviceSymbols CUDA)
+project (ResolveDeviceSymbols CUDA)
# Find nm and dumpbin
if(CMAKE_NM)
diff --git a/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt b/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt
index c934c51..1e574d6 100644
--- a/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt
+++ b/Tests/CudaOnly/SeparateCompilation/CMakeLists.txt
@@ -1,6 +1,6 @@
cmake_minimum_required(VERSION 3.7)
-project (CudaOnlySeparateCompilation CUDA)
+project (SeparateCompilation CUDA)
#Goal for this example:
#Build a static library that defines multiple methods and kernels that
diff --git a/Tests/CudaOnly/WithDefs/CMakeLists.txt b/Tests/CudaOnly/WithDefs/CMakeLists.txt
index 926d9ed..e58204d 100644
--- a/Tests/CudaOnly/WithDefs/CMakeLists.txt
+++ b/Tests/CudaOnly/WithDefs/CMakeLists.txt
@@ -1,6 +1,6 @@
cmake_minimum_required(VERSION 3.7)
-project (CudaOnlyWithDefs CUDA)
+project (WithDefs CUDA)
#verify that we can pass explicit cuda arch flags
string(APPEND CMAKE_CUDA_FLAGS " -gencode arch=compute_30,code=compute_30")
generated by cgit v0.12 at 2025-01-08 00:26:41 (GMT)