NASP/cuda-samples
34
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2024-10-06 05:59:19 +08:00
Code Issues Packages Projects Releases Wiki Activity

add and update samples with CUDA 11.3 support

Browse Source
This commit is contained in:
Rutwik Choughule 2021-04-16 11:54:26 +05:30
parent 067cb65523
commit 568b39bd5b
214 changed files with 6590 additions and 3856 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

67
README.md
View File

@ -1,11 +1,17 @@
# CUDA Samples
Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads).
Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads).
## Release Notes
This section describes the release notes for the CUDA Samples on GitHub only.
### CUDA 11.3
* Added `streamOrderedAllocationIPC`. Demonstrates Inter Process Communication using one process per GPU for computation.
* Added `simpleCUBLAS_LU`. Demonstrates batched matrix LU decomposition using cuBLAS API `cublas<t>getrfBatched()`
* Updated `simpleVulkan`. Demonstrates use of timeline semaphore.
* Updated multiple samples to use pinned memory using `cudaMallocHost()`.
### CUDA 11.2
* Added `streamOrderedAllocation`. Demonstrates stream ordered memory allocation on a GPU using cudaMallocAsync and cudaMemPool family of APIs.
* Added `streamOrderedAllocationP2P`. Demonstrates peer-to-peer access of stream ordered memory allocated using cudaMallocAsync and cudaMemPool family of APIs.
@ -103,7 +109,7 @@ This is the first release of CUDA Samples on GitHub:
### Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
For system requirements and installation instructions of cuda toolkit, please refer to the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/), and the [Windows Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html).
### Getting the CUDA Samples
@ -160,38 +166,39 @@ The samples makefiles can take advantage of certain options:
### Samples by OS
#### Linux
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[boxFilterNPP](./Samples/boxFilterNPP)** | **[streamOrderedAllocationP2P](./Samples/streamOrderedAllocationP2P)** | **[binaryPartitionCG](./Samples/binaryPartitionCG)** |
**[bandwidthTest](./Samples/bandwidthTest)** | **[batchedLabelMarkersAndLabelCompressionNPP](./Samples/batchedLabelMarkersAndLabelCompressionNPP)** | **[bf16TensorCoreGemm](./Samples/bf16TensorCoreGemm)** | **[binaryPartitionCG](./Samples/binaryPartitionCG)** |
---|---|---|---|
**[dmmaTensorCoreGemm](./Samples/dmmaTensorCoreGemm)** | **[EGLStream_CUDA_Interop](./Samples/EGLStream_CUDA_Interop)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** |
**[memMapIPCDrv](./Samples/memMapIPCDrv)** | **[vectorAddMMAP](./Samples/vectorAddMMAP)** | **[shfl_scan](./Samples/shfl_scan)** | **[simpleZeroCopy](./Samples/simpleZeroCopy)** |
**[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** | **[globalToShmemAsyncCopy](./Samples/globalToShmemAsyncCopy)** | **[cudaNvSciNvMedia](./Samples/cudaNvSciNvMedia)** | **[nvJPEG](./Samples/nvJPEG)** |
**[batchedLabelMarkersAndLabelCompressionNPP](./Samples/batchedLabelMarkersAndLabelCompressionNPP)** | **[watershedSegmentationNPP](./Samples/watershedSegmentationNPP)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[streamOrderedAllocation](./Samples/streamOrderedAllocation)** |
**[deviceQuery](./Samples/deviceQuery)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[simpleAttributes](./Samples/simpleAttributes)** |
**[cudaNvSci](./Samples/cudaNvSci)** | **[tf32TensorCoreGemm](./Samples/tf32TensorCoreGemm)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[cudaCompressibleMemory](./Samples/cudaCompressibleMemory)** |
**[bf16TensorCoreGemm](./Samples/bf16TensorCoreGemm)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[vulkanImageCUDA](./Samples/vulkanImageCUDA)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** |
**[matrixMulDrv](./Samples/matrixMulDrv)** | **[cuSolverSp_LinearSolver](./Samples/cuSolverSp_LinearSolver)** | **[simpleCUFFT](./Samples/simpleCUFFT)** | **[reduction](./Samples/reduction)** |
**[nvJPEG_encoder](./Samples/nvJPEG_encoder)** | **[simpleDrvRuntime](./Samples/simpleDrvRuntime)** | **[MersenneTwisterGP11213](./Samples/MersenneTwisterGP11213)** | **[simpleAWBarrier](./Samples/simpleAWBarrier)** |
**[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[bandwidthTest](./Samples/bandwidthTest)** | **[concurrentKernels](./Samples/concurrentKernels)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** |
**[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[simpleGL](./Samples/simpleGL)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[jacobiCudaGraphs](./Samples/jacobiCudaGraphs)** |
**[simpleVulkan](./Samples/simpleVulkan)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[cannyEdgeDetectorNPP](./Samples/cannyEdgeDetectorNPP)** | **[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** |
**[simpleVulkanMMAP](./Samples/simpleVulkanMMAP)** | **[cudaOpenMP](./Samples/cudaOpenMP)** | **[matrixMul](./Samples/matrixMul)** | **[systemWideAtomics](./Samples/systemWideAtomics)** |
**[boxFilterNPP](./Samples/boxFilterNPP)** | **[cannyEdgeDetectorNPP](./Samples/cannyEdgeDetectorNPP)** | **[concurrentKernels](./Samples/concurrentKernels)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** |
**[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[cudaCompressibleMemory](./Samples/cudaCompressibleMemory)** | **[cudaNvSci](./Samples/cudaNvSci)** |
**[cudaNvSciNvMedia](./Samples/cudaNvSciNvMedia)** | **[cudaOpenMP](./Samples/cudaOpenMP)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** |
**[cuSolverSp_LinearSolver](./Samples/cuSolverSp_LinearSolver)** | **[deviceQuery](./Samples/deviceQuery)** | **[dmmaTensorCoreGemm](./Samples/dmmaTensorCoreGemm)** | **[EGLStream_CUDA_Interop](./Samples/EGLStream_CUDA_Interop)** |
**[globalToShmemAsyncCopy](./Samples/globalToShmemAsyncCopy)** | **[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[jacobiCudaGraphs](./Samples/jacobiCudaGraphs)** | **[matrixMul](./Samples/matrixMul)** |
**[matrixMulDrv](./Samples/matrixMulDrv)** | **[memMapIPCDrv](./Samples/memMapIPCDrv)** | **[MersenneTwisterGP11213](./Samples/MersenneTwisterGP11213)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** |
**[nvJPEG](./Samples/nvJPEG)** | **[nvJPEG_encoder](./Samples/nvJPEG_encoder)** | **[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[reduction](./Samples/reduction)** |
**[shfl_scan](./Samples/shfl_scan)** | **[simpleAttributes](./Samples/simpleAttributes)** | **[simpleAWBarrier](./Samples/simpleAWBarrier)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** |
**[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[simpleCUBLAS_LU](./Samples/simpleCUBLAS_LU)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[simpleCUFFT](./Samples/simpleCUFFT)** |
**[simpleDrvRuntime](./Samples/simpleDrvRuntime)** | **[simpleGL](./Samples/simpleGL)** | **[simpleIPC](./Samples/simpleIPC)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** |
**[simpleVulkan](./Samples/simpleVulkan)** | **[simpleVulkanMMAP](./Samples/simpleVulkanMMAP)** | **[simpleZeroCopy](./Samples/simpleZeroCopy)** | **[streamOrderedAllocation](./Samples/streamOrderedAllocation)** |
**[streamOrderedAllocationIPC](./Samples/streamOrderedAllocationIPC)** | **[streamOrderedAllocationP2P](./Samples/streamOrderedAllocationP2P)** | **[systemWideAtomics](./Samples/systemWideAtomics)** | **[tf32TensorCoreGemm](./Samples/tf32TensorCoreGemm)** |
**[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[vectorAddMMAP](./Samples/vectorAddMMAP)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[vulkanImageCUDA](./Samples/vulkanImageCUDA)** |
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[watershedSegmentationNPP](./Samples/watershedSegmentationNPP)** |
#### Windows
**[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[boxFilterNPP](./Samples/boxFilterNPP)** | **[streamOrderedAllocationP2P](./Samples/streamOrderedAllocationP2P)** | **[binaryPartitionCG](./Samples/binaryPartitionCG)** |
**[bandwidthTest](./Samples/bandwidthTest)** | **[batchedLabelMarkersAndLabelCompressionNPP](./Samples/batchedLabelMarkersAndLabelCompressionNPP)** | **[bf16TensorCoreGemm](./Samples/bf16TensorCoreGemm)** | **[binaryPartitionCG](./Samples/binaryPartitionCG)** |
---|---|---|---|
**[dmmaTensorCoreGemm](./Samples/dmmaTensorCoreGemm)** | **[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[simpleIPC](./Samples/simpleIPC)** | **[memMapIPCDrv](./Samples/memMapIPCDrv)** |
**[vectorAddMMAP](./Samples/vectorAddMMAP)** | **[shfl_scan](./Samples/shfl_scan)** | **[simpleZeroCopy](./Samples/simpleZeroCopy)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** |
**[globalToShmemAsyncCopy](./Samples/globalToShmemAsyncCopy)** | **[nvJPEG](./Samples/nvJPEG)** | **[batchedLabelMarkersAndLabelCompressionNPP](./Samples/batchedLabelMarkersAndLabelCompressionNPP)** | **[simpleD3D12](./Samples/simpleD3D12)** |
**[watershedSegmentationNPP](./Samples/watershedSegmentationNPP)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** | **[streamOrderedAllocation](./Samples/streamOrderedAllocation)** | **[deviceQuery](./Samples/deviceQuery)** |
**[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[simpleAttributes](./Samples/simpleAttributes)** | **[tf32TensorCoreGemm](./Samples/tf32TensorCoreGemm)** |
**[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[cudaCompressibleMemory](./Samples/cudaCompressibleMemory)** | **[bf16TensorCoreGemm](./Samples/bf16TensorCoreGemm)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** |
**[vulkanImageCUDA](./Samples/vulkanImageCUDA)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[cuSolverSp_LinearSolver](./Samples/cuSolverSp_LinearSolver)** |
**[simpleCUFFT](./Samples/simpleCUFFT)** | **[reduction](./Samples/reduction)** | **[nvJPEG_encoder](./Samples/nvJPEG_encoder)** | **[simpleDrvRuntime](./Samples/simpleDrvRuntime)** |
**[simpleD3D11](./Samples/simpleD3D11)** | **[MersenneTwisterGP11213](./Samples/MersenneTwisterGP11213)** | **[simpleAWBarrier](./Samples/simpleAWBarrier)** | **[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** |
**[bandwidthTest](./Samples/bandwidthTest)** | **[concurrentKernels](./Samples/concurrentKernels)** | **[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[NV12toBGRandResize](./Samples/NV12toBGRandResize)** |
**[simpleGL](./Samples/simpleGL)** | **[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[jacobiCudaGraphs](./Samples/jacobiCudaGraphs)** | **[simpleVulkan](./Samples/simpleVulkan)** |
**[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** | **[cannyEdgeDetectorNPP](./Samples/cannyEdgeDetectorNPP)** | **[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** | **[simpleVulkanMMAP](./Samples/simpleVulkanMMAP)** |
**[cudaOpenMP](./Samples/cudaOpenMP)** | **[matrixMul](./Samples/matrixMul)** |
**[boxFilterNPP](./Samples/boxFilterNPP)** | **[cannyEdgeDetectorNPP](./Samples/cannyEdgeDetectorNPP)** | **[concurrentKernels](./Samples/concurrentKernels)** | **[conjugateGradientCudaGraphs](./Samples/conjugateGradientCudaGraphs)** |
**[conjugateGradientMultiBlockCG](./Samples/conjugateGradientMultiBlockCG)** | **[conjugateGradientMultiDeviceCG](./Samples/conjugateGradientMultiDeviceCG)** | **[cudaCompressibleMemory](./Samples/cudaCompressibleMemory)** | **[cudaOpenMP](./Samples/cudaOpenMP)** |
**[cudaTensorCoreGemm](./Samples/cudaTensorCoreGemm)** | **[cuSolverDn_LinearSolver](./Samples/cuSolverDn_LinearSolver)** | **[cuSolverSp_LinearSolver](./Samples/cuSolverSp_LinearSolver)** | **[deviceQuery](./Samples/deviceQuery)** |
**[dmmaTensorCoreGemm](./Samples/dmmaTensorCoreGemm)** | **[globalToShmemAsyncCopy](./Samples/globalToShmemAsyncCopy)** | **[immaTensorCoreGemm](./Samples/immaTensorCoreGemm)** | **[jacobiCudaGraphs](./Samples/jacobiCudaGraphs)** |
**[matrixMul](./Samples/matrixMul)** | **[matrixMulDrv](./Samples/matrixMulDrv)** | **[memMapIPCDrv](./Samples/memMapIPCDrv)** | **[MersenneTwisterGP11213](./Samples/MersenneTwisterGP11213)** |
**[NV12toBGRandResize](./Samples/NV12toBGRandResize)** | **[nvJPEG](./Samples/nvJPEG)** | **[nvJPEG_encoder](./Samples/nvJPEG_encoder)** | **[p2pBandwidthLatencyTest](./Samples/p2pBandwidthLatencyTest)** |
**[reduction](./Samples/reduction)** | **[shfl_scan](./Samples/shfl_scan)** | **[simpleAttributes](./Samples/simpleAttributes)** | **[simpleAWBarrier](./Samples/simpleAWBarrier)** |
**[simpleCUBLAS](./Samples/simpleCUBLAS)** | **[simpleCUBLASXT](./Samples/simpleCUBLASXT)** | **[simpleCUBLAS_LU](./Samples/simpleCUBLAS_LU)** | **[simpleCudaGraphs](./Samples/simpleCudaGraphs)** |
**[simpleCUFFT](./Samples/simpleCUFFT)** | **[simpleD3D11](./Samples/simpleD3D11)** | **[simpleD3D12](./Samples/simpleD3D12)** | **[simpleDrvRuntime](./Samples/simpleDrvRuntime)** |
**[simpleGL](./Samples/simpleGL)** | **[simpleIPC](./Samples/simpleIPC)** | **[simpleVoteIntrinsics](./Samples/simpleVoteIntrinsics)** | **[simpleVulkan](./Samples/simpleVulkan)** |
**[simpleVulkanMMAP](./Samples/simpleVulkanMMAP)** | **[simpleZeroCopy](./Samples/simpleZeroCopy)** | **[streamOrderedAllocation](./Samples/streamOrderedAllocation)** | **[streamOrderedAllocationP2P](./Samples/streamOrderedAllocationP2P)** |
**[tf32TensorCoreGemm](./Samples/tf32TensorCoreGemm)** | **[UnifiedMemoryPerf](./Samples/UnifiedMemoryPerf)** | **[vectorAddMMAP](./Samples/vectorAddMMAP)** | **[vectorAdd_nvrtc](./Samples/vectorAdd_nvrtc)** |
**[vulkanImageCUDA](./Samples/vulkanImageCUDA)** | **[warpAggregatedAtomicsCG](./Samples/warpAggregatedAtomicsCG)** | **[watershedSegmentationNPP](./Samples/watershedSegmentationNPP)** |
## Dependencies

6
Samples/EGLStream_CUDA_Interop/Makefile
View File

@ -285,6 +285,12 @@ ifeq ($(TARGET_OS),android)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - EGLStream_CUDA_Interop is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))

2
Samples/EGLStream_CUDA_Interop/README.md
View File

@ -30,7 +30,7 @@ cuDeviceGet, cuDeviceGetAttribute, cuDeviceComputeCapability, cuDeviceGetCount,
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

29
Samples/MersenneTwisterGP11213/Makefile
View File

@ -263,6 +263,14 @@ ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - MersenneTwisterGP11213 is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
@ -297,6 +305,10 @@ ALL_CCFLAGS += --threads 0
LIBRARIES += -lcurand_static -lculibos
ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif
################################################################################
# Target rules
@ -304,16 +316,23 @@ all: build
build: MersenneTwisterGP11213
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@echo "Sample will be waived due to the above missing dependencies"
else
@echo "Sample is ready - all dependencies have been met"
endif
MersenneTwister.o:MersenneTwister.cpp
$(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
MersenneTwisterGP11213: MersenneTwister.o
$(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
$(EXEC) cp $@ ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
run: build
./MersenneTwisterGP11213
$(EXEC) ./MersenneTwisterGP11213
clean:
rm -f MersenneTwisterGP11213 MersenneTwister.o

188
Samples/MersenneTwisterGP11213/MersenneTwister.cpp
View File

@ -47,138 +47,134 @@
float compareResults(int rand_n, float *h_RandGPU, float *h_RandCPU);
const int DEFAULT_RAND_N = 2400000;
const int DEFAULT_RAND_N = 2400000;
const unsigned int DEFAULT_SEED = 777;
///////////////////////////////////////////////////////////////////////////////
// Main program
///////////////////////////////////////////////////////////////////////////////
int main(int argc, char **argv)
{
// Start logs
printf("%s Starting...\n\n", argv[0]);
int main(int argc, char **argv) {
// Start logs
printf("%s Starting...\n\n", argv[0]);
// initialize the GPU, either identified by --device
// or by picking the device with highest flop rate.
int devID = findCudaDevice(argc, (const char **)argv);
// initialize the GPU, either identified by --device
// or by picking the device with highest flop rate.
int devID = findCudaDevice(argc, (const char **)argv);
// parsing the number of random numbers to generate
int rand_n = DEFAULT_RAND_N;
// parsing the number of random numbers to generate
int rand_n = DEFAULT_RAND_N;
if (checkCmdLineFlag(argc, (const char **) argv, "count"))
{
rand_n = getCmdLineArgumentInt(argc, (const char **) argv, "count");
}
if (checkCmdLineFlag(argc, (const char **)argv, "count")) {
rand_n = getCmdLineArgumentInt(argc, (const char **)argv, "count");
}
printf("Allocating data for %i samples...\n", rand_n);
printf("Allocating data for %i samples...\n", rand_n);
// parsing the seed
int seed = DEFAULT_SEED;
// parsing the seed
int seed = DEFAULT_SEED;
if (checkCmdLineFlag(argc, (const char **) argv, "seed"))
{
seed = getCmdLineArgumentInt(argc, (const char **) argv, "seed");
}
if (checkCmdLineFlag(argc, (const char **)argv, "seed")) {
seed = getCmdLineArgumentInt(argc, (const char **)argv, "seed");
}
printf("Seeding with %i ...\n", seed);
printf("Seeding with %i ...\n", seed);
cudaStream_t stream;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
cudaStream_t stream;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
float *d_Rand;
checkCudaErrors(cudaMalloc((void **)&d_Rand, rand_n * sizeof(float)));
float *d_Rand;
checkCudaErrors(cudaMalloc((void **)&d_Rand, rand_n * sizeof(float)));
curandGenerator_t prngGPU;
checkCudaErrors(curandCreateGenerator(&prngGPU, CURAND_RNG_PSEUDO_MTGP32));
checkCudaErrors(curandSetStream(prngGPU, stream));
checkCudaErrors(curandSetPseudoRandomGeneratorSeed(prngGPU, seed));
curandGenerator_t prngGPU;
checkCudaErrors(curandCreateGenerator(&prngGPU, CURAND_RNG_PSEUDO_MTGP32));
checkCudaErrors(curandSetStream(prngGPU, stream));
checkCudaErrors(curandSetPseudoRandomGeneratorSeed(prngGPU, seed));
curandGenerator_t prngCPU;
checkCudaErrors(curandCreateGeneratorHost(&prngCPU, CURAND_RNG_PSEUDO_MTGP32));
checkCudaErrors(curandSetPseudoRandomGeneratorSeed(prngCPU, seed));
curandGenerator_t prngCPU;
checkCudaErrors(
curandCreateGeneratorHost(&prngCPU, CURAND_RNG_PSEUDO_MTGP32));
checkCudaErrors(curandSetPseudoRandomGeneratorSeed(prngCPU, seed));
//
// Example 1: Compare random numbers generated on GPU and CPU
float *h_RandGPU = (float *)malloc(rand_n * sizeof(float));
//
// Example 1: Compare random numbers generated on GPU and CPU
float *h_RandGPU;
checkCudaErrors(cudaMallocHost(&h_RandGPU, rand_n * sizeof(float)));
printf("Generating random numbers on GPU...\n\n");
checkCudaErrors(curandGenerateUniform(prngGPU, (float *) d_Rand, rand_n));
printf("Generating random numbers on GPU...\n\n");
checkCudaErrors(curandGenerateUniform(prngGPU, (float *)d_Rand, rand_n));
printf("\nReading back the results...\n");
checkCudaErrors(cudaMemcpyAsync(h_RandGPU, d_Rand, rand_n * sizeof(float), cudaMemcpyDeviceToHost, stream));
printf("\nReading back the results...\n");
checkCudaErrors(cudaMemcpyAsync(h_RandGPU, d_Rand, rand_n * sizeof(float),
cudaMemcpyDeviceToHost, stream));
float *h_RandCPU = (float *)malloc(rand_n * sizeof(float));
float *h_RandCPU = (float *)malloc(rand_n * sizeof(float));
printf("Generating random numbers on CPU...\n\n");
checkCudaErrors(curandGenerateUniform(prngCPU, (float *)h_RandCPU, rand_n));
printf("Generating random numbers on CPU...\n\n");
checkCudaErrors(curandGenerateUniform(prngCPU, (float *) h_RandCPU, rand_n));
checkCudaErrors(cudaStreamSynchronize(stream));
printf("Comparing CPU/GPU random numbers...\n\n");
float L1norm = compareResults(rand_n, h_RandGPU, h_RandCPU);
checkCudaErrors(cudaStreamSynchronize(stream));
printf("Comparing CPU/GPU random numbers...\n\n");
float L1norm = compareResults(rand_n, h_RandGPU, h_RandCPU);
//
// Example 2: Timing of random number generation on GPU
const int numIterations = 10;
int i;
StopWatchInterface *hTimer;
//
// Example 2: Timing of random number generation on GPU
const int numIterations = 10;
int i;
StopWatchInterface *hTimer;
sdkCreateTimer(&hTimer);
sdkResetTimer(&hTimer);
sdkStartTimer(&hTimer);
sdkCreateTimer(&hTimer);
sdkResetTimer(&hTimer);
sdkStartTimer(&hTimer);
for (i = 0; i < numIterations; i++) {
checkCudaErrors(curandGenerateUniform(prngGPU, (float *)d_Rand, rand_n));
}
for (i = 0; i < numIterations; i++)
{
checkCudaErrors(curandGenerateUniform(prngGPU, (float *) d_Rand, rand_n));
}
checkCudaErrors(cudaStreamSynchronize(stream));
sdkStopTimer(&hTimer);
checkCudaErrors(cudaStreamSynchronize(stream));
sdkStopTimer(&hTimer);
double gpuTime = 1.0e-3 * sdkGetTimerValue(&hTimer) / (double)numIterations;
double gpuTime = 1.0e-3 * sdkGetTimerValue(&hTimer)/(double)numIterations;
printf(
"MersenneTwisterGP11213, Throughput = %.4f GNumbers/s, Time = %.5f s, "
"Size = %u Numbers\n",
1.0e-9 * rand_n / gpuTime, gpuTime, rand_n);
printf("MersenneTwisterGP11213, Throughput = %.4f GNumbers/s, Time = %.5f s, Size = %u Numbers\n",
1.0e-9 * rand_n / gpuTime, gpuTime, rand_n);
printf("Shutting down...\n");
printf("Shutting down...\n");
checkCudaErrors(curandDestroyGenerator(prngGPU));
checkCudaErrors(curandDestroyGenerator(prngCPU));
checkCudaErrors(cudaStreamDestroy(stream));
checkCudaErrors(cudaFree(d_Rand));
sdkDeleteTimer(&hTimer);
checkCudaErrors(cudaFreeHost(h_RandGPU));
free(h_RandCPU);
checkCudaErrors(curandDestroyGenerator(prngGPU));
checkCudaErrors(curandDestroyGenerator(prngCPU));
checkCudaErrors(cudaStreamDestroy(stream));
checkCudaErrors(cudaFree(d_Rand));
sdkDeleteTimer(&hTimer);
free(h_RandGPU);
free(h_RandCPU);
exit(L1norm < 1e-6 ? EXIT_SUCCESS : EXIT_FAILURE);
exit(L1norm < 1e-6 ? EXIT_SUCCESS : EXIT_FAILURE);
}
float compareResults(int rand_n, float *h_RandGPU, float *h_RandCPU) {
int i;
float rCPU, rGPU, delta;
float max_delta = 0.;
float sum_delta = 0.;
float sum_ref = 0.;
float compareResults(int rand_n, float *h_RandGPU, float *h_RandCPU)
{
int i;
float rCPU, rGPU, delta;
float max_delta = 0.;
float sum_delta = 0.;
float sum_ref = 0.;
for (i = 0; i < rand_n; i++) {
rCPU = h_RandCPU[i];
rGPU = h_RandGPU[i];
delta = fabs(rCPU - rGPU);
sum_delta += delta;
sum_ref += fabs(rCPU);
for (i = 0; i < rand_n; i++)
{
rCPU = h_RandCPU[i];
rGPU = h_RandGPU[i];
delta = fabs(rCPU - rGPU);
sum_delta += delta;
sum_ref += fabs(rCPU);
if (delta >= max_delta)
{
max_delta = delta;
}
if (delta >= max_delta) {
max_delta = delta;
}
}
float L1norm = (float)(sum_delta / sum_ref);
printf("Max absolute error: %E\n", max_delta);
printf("L1 norm: %E\n\n", L1norm);
float L1norm = (float)(sum_delta / sum_ref);
printf("Max absolute error: %E\n", max_delta);
printf("L1 norm: %E\n\n", L1norm);
return L1norm;
return L1norm;
}

4
Samples/MersenneTwisterGP11213/MersenneTwisterGP11213_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/MersenneTwisterGP11213/MersenneTwisterGP11213_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/MersenneTwisterGP11213/README.md
View File

@ -27,7 +27,7 @@ x86_64, ppc64le, armv7l
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

4
Samples/NV12toBGRandResize/NV12toBGRandResize_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -113,6 +113,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/NV12toBGRandResize/NV12toBGRandResize_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -109,6 +109,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/NV12toBGRandResize/README.md
View File

@ -27,7 +27,7 @@ cudaMemcpy2D, cudaMallocManaged
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

2
Samples/UnifiedMemoryPerf/README.md
View File

@ -28,7 +28,7 @@ cudaMallocManaged, cudaStreamAttachMemAsync, cudaMemcpyAsync, cudaMallocHost, cu
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

4
Samples/UnifiedMemoryPerf/UnifiedMemoryPerf_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -111,6 +111,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/UnifiedMemoryPerf/UnifiedMemoryPerf_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -107,6 +107,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/bandwidthTest/README.md
View File

@ -27,7 +27,7 @@ cudaSetDevice, cudaHostAlloc, cudaFree, cudaMallocHost, cudaFreeHost, cudaMemcpy
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/bandwidthTest/bandwidthTest_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/bandwidthTest/bandwidthTest_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

6
Samples/batchedLabelMarkersAndLabelCompressionNPP/Makefile
View File

@ -271,6 +271,12 @@ ifeq ($(TARGET_OS),darwin)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - batchedLabelMarkersAndLabelCompressionNPP is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))

2
Samples/batchedLabelMarkersAndLabelCompressionNPP/README.md
View File

@ -28,7 +28,7 @@ x86_64, ppc64le, armv7l
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

93
Samples/batchedLabelMarkersAndLabelCompressionNPP/batchedLabelMarkersAndLabelCompressionNPP.cpp
View File

@ -36,11 +36,13 @@
#include <string.h>
#include <fstream>
#include <cuda_runtime.h>
#include <helper_cuda.h>
#include <helper_string.h>
#include <npp.h>
// Note: If you want to view these images we HIGHLY recommend using imagej
// which is free on the internet and works on most platforms
// which is free on the internet and works on most platforms
// because it is one of the few image viewing apps that can display 32
// bit integer image data. While it normalizes the data to floating
// point values for viewing it still provides a good representation of
@ -102,11 +104,12 @@ void tearDown() // Clean up and tear down
if (pUFBatchSrcDstImageListDev != 0) cudaFree(pUFBatchSrcDstImageListDev);
if (pUFBatchSrcImageListDev != 0) cudaFree(pUFBatchSrcImageListDev);
if (pUFBatchPerImageCompressedCountListHost != 0)
free(pUFBatchPerImageCompressedCountListHost);
cudaFreeHost(pUFBatchPerImageCompressedCountListHost);
if (pUFBatchSrcDstScratchBufferListHost != 0)
free(pUFBatchSrcDstScratchBufferListHost);
if (pUFBatchSrcDstImageListHost != 0) free(pUFBatchSrcDstImageListHost);
if (pUFBatchSrcImageListHost != 0) free(pUFBatchSrcImageListHost);
cudaFreeHost(pUFBatchSrcDstScratchBufferListHost);
if (pUFBatchSrcDstImageListHost != 0)
cudaFreeHost(pUFBatchSrcDstImageListHost);
if (pUFBatchSrcImageListHost != 0) cudaFreeHost(pUFBatchSrcImageListHost);
for (int j = 0; j < NUMBER_OF_IMAGES; j++) {
if (pUFCompressedLabelsScratchBufferDev[j] != 0)
@ -115,8 +118,8 @@ void tearDown() // Clean up and tear down
cudaFree(pUFGenerateLabelsScratchBufferDev[j]);
if (pUFLabelDev[j] != 0) cudaFree(pUFLabelDev[j]);
if (pInputImageDev[j] != 0) cudaFree(pInputImageDev[j]);
if (pUFLabelHost[j] != 0) free(pUFLabelHost[j]);
if (pInputImageHost[j] != 0) free(pInputImageHost[j]);
if (pUFLabelHost[j] != 0) cudaFreeHost(pUFLabelHost[j]);
if (pInputImageHost[j] != 0) cudaFreeHost(pInputImageHost[j]);
}
}
@ -177,7 +180,7 @@ int loadRaw8BitImage(Npp8u *pImage, int nWidth, int nHeight, int nImage) {
exit(EXIT_WAIVED);
}
bmpFile = fopen(InputFile, "rb");
FOPEN(bmpFile, InputFile, "rb");
} else if (nImage == 1) {
if (nWidth != 512 || nHeight != 512) return -1;
const char *fileName = "CT_skull_512x512_8u.raw";
@ -187,7 +190,7 @@ int loadRaw8BitImage(Npp8u *pImage, int nWidth, int nHeight, int nImage) {
exit(EXIT_WAIVED);
}
bmpFile = fopen(InputFile, "rb");
FOPEN(bmpFile, InputFile, "rb");
} else if (nImage == 2) {
if (nWidth != 509 || nHeight != 335) return -1;
const char *fileName = "PCB_METAL_509x335_8u.raw";
@ -197,7 +200,7 @@ int loadRaw8BitImage(Npp8u *pImage, int nWidth, int nHeight, int nImage) {
exit(EXIT_WAIVED);
}
bmpFile = fopen(InputFile, "rb");
FOPEN(bmpFile, InputFile, "rb");
} else if (nImage == 3) {
if (nWidth != 1024 || nHeight != 683) return -1;
const char *fileName = "PCB2_1024x683_8u.raw";
@ -207,7 +210,7 @@ int loadRaw8BitImage(Npp8u *pImage, int nWidth, int nHeight, int nImage) {
exit(EXIT_WAIVED);
}
bmpFile = fopen(InputFile, "rb");
FOPEN(bmpFile, InputFile, "rb");
} else if (nImage == 4) {
if (nWidth != 1280 || nHeight != 720) return -1;
const char *fileName = "PCB_1280x720_8u.raw";
@ -217,7 +220,7 @@ int loadRaw8BitImage(Npp8u *pImage, int nWidth, int nHeight, int nImage) {
exit(EXIT_WAIVED);
}
bmpFile = fopen(InputFile, "rb");
FOPEN(bmpFile, InputFile, "rb");
} else {
printf("Input file load failed.\n");
return -1;
@ -347,9 +350,11 @@ int main(int argc, char **argv) {
oSizeROI[nImage].width * sizeof(Npp32u) * oSizeROI[nImage].height);
if (cudaError != cudaSuccess) return NPP_MEMORY_ALLOCATION_ERR;
pInputImageHost[nImage] = reinterpret_cast<Npp8u *>(malloc(
checkCudaErrors(cudaMallocHost(
&(pInputImageHost[nImage]),
oSizeROI[nImage].width * sizeof(Npp8u) * oSizeROI[nImage].height));
pUFLabelHost[nImage] = reinterpret_cast<Npp32u *>(malloc(
checkCudaErrors(cudaMallocHost(
&(pUFLabelHost[nImage]),
oSizeROI[nImage].width * sizeof(Npp32u) * oSizeROI[nImage].height));
// Use UF functions throughout this sample.
@ -409,15 +414,15 @@ int main(int argc, char **argv) {
}
if (nImage == 0)
bmpFile = fopen(LabelMarkersOutputFile0.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersOutputFile0.c_str(), "wb");
else if (nImage == 1)
bmpFile = fopen(LabelMarkersOutputFile1.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersOutputFile1.c_str(), "wb");
else if (nImage == 2)
bmpFile = fopen(LabelMarkersOutputFile2.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersOutputFile2.c_str(), "wb");
else if (nImage == 3)
bmpFile = fopen(LabelMarkersOutputFile3.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersOutputFile3.c_str(), "wb");
else if (nImage == 4)
bmpFile = fopen(LabelMarkersOutputFile4.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersOutputFile4.c_str(), "wb");
if (bmpFile == NULL) return -1;
size_t nSize = 0;
@ -478,15 +483,15 @@ int main(int argc, char **argv) {
}
if (nImage == 0)
bmpFile = fopen(CompressedMarkerLabelsOutputFile0.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsOutputFile0.c_str(), "wb");
else if (nImage == 1)
bmpFile = fopen(CompressedMarkerLabelsOutputFile1.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsOutputFile1.c_str(), "wb");
else if (nImage == 2)
bmpFile = fopen(CompressedMarkerLabelsOutputFile2.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsOutputFile2.c_str(), "wb");
else if (nImage == 3)
bmpFile = fopen(CompressedMarkerLabelsOutputFile3.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsOutputFile3.c_str(), "wb");
else if (nImage == 4)
bmpFile = fopen(CompressedMarkerLabelsOutputFile4.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsOutputFile4.c_str(), "wb");
if (bmpFile == NULL) return -1;
nSize = 0;
@ -554,10 +559,11 @@ int main(int argc, char **argv) {
cudaMalloc((void **)&pUFBatchSrcDstImageListDev, nBatchImageListBytes);
if (cudaError != cudaSuccess) return NPP_MEMORY_ALLOCATION_ERR;
pUFBatchSrcImageListHost =
reinterpret_cast<NppiImageDescriptor *>(malloc(nBatchImageListBytes));
pUFBatchSrcDstImageListHost =
reinterpret_cast<NppiImageDescriptor *>(malloc(nBatchImageListBytes));
checkCudaErrors(
cudaMallocHost((void **)&pUFBatchSrcImageListHost, nBatchImageListBytes));
checkCudaErrors(cudaMallocHost((void **)&pUFBatchSrcDstImageListHost,
nBatchImageListBytes));
NppiSize oMaxROISize = {0, 0};
@ -620,15 +626,15 @@ int main(int argc, char **argv) {
// Save output to files
for (int nImage = 0; nImage < NUMBER_OF_IMAGES; nImage++) {
if (nImage == 0)
bmpFile = fopen(LabelMarkersBatchOutputFile0.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersBatchOutputFile0.c_str(), "wb");
else if (nImage == 1)
bmpFile = fopen(LabelMarkersBatchOutputFile1.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersBatchOutputFile1.c_str(), "wb");
else if (nImage == 2)
bmpFile = fopen(LabelMarkersBatchOutputFile2.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersBatchOutputFile2.c_str(), "wb");
else if (nImage == 3)
bmpFile = fopen(LabelMarkersBatchOutputFile3.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersBatchOutputFile3.c_str(), "wb");
else if (nImage == 4)
bmpFile = fopen(LabelMarkersBatchOutputFile4.c_str(), "wb");
FOPEN(bmpFile, LabelMarkersBatchOutputFile4.c_str(), "wb");
if (bmpFile == NULL) return -1;
size_t nSize = 0;
@ -652,12 +658,13 @@ int main(int argc, char **argv) {
// Allocate host side scratch buffer point and size list and initialize with
// device scratch buffer pointers
pUFBatchSrcDstScratchBufferListHost =
reinterpret_cast<NppiBufferDescriptor *>(
malloc(NUMBER_OF_IMAGES * sizeof(NppiBufferDescriptor)));
checkCudaErrors(
cudaMallocHost((void **)&pUFBatchSrcDstScratchBufferListHost,
NUMBER_OF_IMAGES * sizeof(NppiBufferDescriptor)));
pUFBatchPerImageCompressedCountListHost =
reinterpret_cast<Npp32u *>(malloc(NUMBER_OF_IMAGES * sizeof(Npp32u)));
checkCudaErrors(
cudaMallocHost((void **)&pUFBatchPerImageCompressedCountListHost,
+NUMBER_OF_IMAGES * sizeof(Npp32u)));
// Start buffer pointer at beginning of full per image buffer list sized
// pUFCompressedLabelsScratchBufferDev[0]
@ -728,15 +735,15 @@ int main(int argc, char **argv) {
// Save compressed label images into files
for (int nImage = 0; nImage < NUMBER_OF_IMAGES; nImage++) {
if (nImage == 0)
bmpFile = fopen(CompressedMarkerLabelsBatchOutputFile0.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsBatchOutputFile0.c_str(), "wb");
else if (nImage == 1)
bmpFile = fopen(CompressedMarkerLabelsBatchOutputFile1.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsBatchOutputFile1.c_str(), "wb");
else if (nImage == 2)
bmpFile = fopen(CompressedMarkerLabelsBatchOutputFile2.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsBatchOutputFile2.c_str(), "wb");
else if (nImage == 3)
bmpFile = fopen(CompressedMarkerLabelsBatchOutputFile3.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsBatchOutputFile3.c_str(), "wb");
else if (nImage == 4)
bmpFile = fopen(CompressedMarkerLabelsBatchOutputFile4.c_str(), "wb");
FOPEN(bmpFile, CompressedMarkerLabelsBatchOutputFile4.c_str(), "wb");
if (bmpFile == NULL) return -1;
size_t nSize = 0;

4
Samples/batchedLabelMarkersAndLabelCompressionNPP/batchedLabelMarkersAndLabelCompressionNPP_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/batchedLabelMarkersAndLabelCompressionNPP/batchedLabelMarkersAndLabelCompressionNPP_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/bf16TensorCoreGemm/README.md
View File

@ -27,7 +27,7 @@ cudaMallocManaged, cudaDeviceSynchronize, cudaFuncSetAttribute, cudaEventCreate,
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/bf16TensorCoreGemm/bf16TensorCoreGemm_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/bf16TensorCoreGemm/bf16TensorCoreGemm_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/binaryPartitionCG/README.md
View File

@ -24,7 +24,7 @@ x86_64, ppc64le, armv7l
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

172
Samples/binaryPartitionCG/binaryPartitionCG.cu
View File

@ -31,14 +31,16 @@
* 1.) Each thread loads a value from random array.
* 2.) then checks if it is odd or even.
* 3.) create binary partition group based on the above predicate
* 4.) we count the number of odd/even in the group based on size of the binary groups
* 4.) we count the number of odd/even in the group based on size of the binary
groups
* 5.) write it global counter of odd.
* 6.) sum the values loaded by individual threads(using reduce) and write it to global
* even & odd elements sum.
* 6.) sum the values loaded by individual threads(using reduce) and write it to
global even & odd elements sum.
*
* **NOTE** : binary_partition results in splitting warp into divergent thread groups
this is not good from performance perspective, but in cases where warp
divergence is inevitable one can use binary_partition group.
* **NOTE** :
* binary_partition results in splitting warp into divergent thread groups
* this is not good from performance perspective, but in cases where warp
* divergence is inevitable one can use binary_partition group.
*/
#include <stdio.h>
@ -48,108 +50,110 @@
namespace cg = cooperative_groups;
void initOddEvenArr(int *inputArr, unsigned int size)
{
for (int i=0; i < size; i++)
{
inputArr[i] = rand() % 50;
}
void initOddEvenArr(int *inputArr, unsigned int size) {
for (int i = 0; i < size; i++) {
inputArr[i] = rand() % 50;
}
}
/**
* CUDA kernel device code
*
*
* Creates cooperative groups and performs odd/even counting & summation.
*/
__global__ void oddEvenCountAndSumCG(int *inputArr, int *numOfOdds, int *sumOfOddAndEvens, unsigned int size)
{
cg::thread_block cta = cg::this_thread_block();
cg::grid_group grid = cg::this_grid();
cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
__global__ void oddEvenCountAndSumCG(int *inputArr, int *numOfOdds,
int *sumOfOddAndEvens, unsigned int size) {
cg::thread_block cta = cg::this_thread_block();
cg::grid_group grid = cg::this_grid();
cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
for (int i = grid.thread_rank(); i < size; i += grid.size())
for (int i = grid.thread_rank(); i < size; i += grid.size()) {
int elem = inputArr[i];
auto subTile = cg::binary_partition(tile32, elem & 1);
if (elem & 1) // Odd numbers group
{
int elem = inputArr[i];
auto subTile = cg::binary_partition(tile32, elem & 1);
if (elem & 1) // Odd numbers group
{
int oddGroupSum = cg::reduce(subTile, elem, cg::plus<int>());
int oddGroupSum = cg::reduce(subTile, elem, cg::plus<int>());
if (subTile.thread_rank() == 0)
{
// Add number of odds present in this group of Odds.
atomicAdd(numOfOdds, subTile.size());
if (subTile.thread_rank() == 0) {
// Add number of odds present in this group of Odds.
atomicAdd(numOfOdds, subTile.size());
// Add local reduction of odds present in this group of Odds.
atomicAdd(&sumOfOddAndEvens[0], oddGroupSum);
// Add local reduction of odds present in this group of Odds.
atomicAdd(&sumOfOddAndEvens[0], oddGroupSum);
}
} else // Even numbers group
{
int evenGroupSum = cg::reduce(subTile, elem, cg::plus<int>());
}
}
else // Even numbers group
{
int evenGroupSum = cg::reduce(subTile, elem, cg::plus<int>());
if (subTile.thread_rank() == 0)
{
// Add local reduction of even present in this group of evens.
atomicAdd(&sumOfOddAndEvens[1], evenGroupSum);
}
}
// reconverge warp so for next loop iteration we ensure convergence of
// above diverged threads to perform coalesced loads of inputArr.
cg::sync(tile32);
if (subTile.thread_rank() == 0) {
// Add local reduction of even present in this group of evens.
atomicAdd(&sumOfOddAndEvens[1], evenGroupSum);
}
}
// reconverge warp so for next loop iteration we ensure convergence of
// above diverged threads to perform coalesced loads of inputArr.
cg::sync(tile32);
}
}
/**
* Host main routine
*/
int main(int argc, const char **argv)
{
int deviceId = findCudaDevice(argc, argv);
int *h_inputArr, *d_inputArr;
int *h_numOfOdds, *d_numOfOdds;
int *h_sumOfOddEvenElems, *d_sumOfOddEvenElems;
unsigned int arrSize = 1024 * 100;
int main(int argc, const char **argv) {
int deviceId = findCudaDevice(argc, argv);
int *h_inputArr, *d_inputArr;
int *h_numOfOdds, *d_numOfOdds;
int *h_sumOfOddEvenElems, *d_sumOfOddEvenElems;
unsigned int arrSize = 1024 * 100;
h_inputArr = new int[arrSize];
h_numOfOdds = new int[1];
h_sumOfOddEvenElems = new int[2];
initOddEvenArr(h_inputArr, arrSize);
cudaStream_t stream;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
checkCudaErrors(cudaMalloc(&d_inputArr, sizeof(int)*arrSize));
checkCudaErrors(cudaMalloc(&d_numOfOdds, sizeof(int)));
checkCudaErrors(cudaMalloc(&d_sumOfOddEvenElems, sizeof(int)*2));
checkCudaErrors(cudaMallocHost(&h_inputArr, sizeof(int) * arrSize));
checkCudaErrors(cudaMallocHost(&h_numOfOdds, sizeof(int)));
checkCudaErrors(cudaMallocHost(&h_sumOfOddEvenElems, sizeof(int) * 2));
initOddEvenArr(h_inputArr, arrSize);
checkCudaErrors(cudaMemcpyAsync(d_inputArr, h_inputArr, sizeof(int)*arrSize, cudaMemcpyHostToDevice, stream));
checkCudaErrors(cudaMemsetAsync(d_numOfOdds, 0, sizeof(int), stream));
checkCudaErrors(cudaMemsetAsync(d_sumOfOddEvenElems, 0, 2*sizeof(int), stream));
cudaStream_t stream;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
checkCudaErrors(cudaMalloc(&d_inputArr, sizeof(int) * arrSize));
checkCudaErrors(cudaMalloc(&d_numOfOdds, sizeof(int)));
checkCudaErrors(cudaMalloc(&d_sumOfOddEvenElems, sizeof(int) * 2));
//Launch the kernel
int threadsPerBlock=1024;
int blocksPerGrid = arrSize / threadsPerBlock;
checkCudaErrors(cudaMemcpyAsync(d_inputArr, h_inputArr, sizeof(int) * arrSize,
cudaMemcpyHostToDevice, stream));
checkCudaErrors(cudaMemsetAsync(d_numOfOdds, 0, sizeof(int), stream));
checkCudaErrors(
cudaMemsetAsync(d_sumOfOddEvenElems, 0, 2 * sizeof(int), stream));
printf("\nLaunching %d blocks with %d threads...\n\n",blocksPerGrid, threadsPerBlock);
// Launch the kernel
int threadsPerBlock = 0;
int blocksPerGrid = 0;
checkCudaErrors(cudaOccupancyMaxPotentialBlockSize(
&blocksPerGrid, &threadsPerBlock, oddEvenCountAndSumCG, 0, 0));
oddEvenCountAndSumCG<<<blocksPerGrid, threadsPerBlock, 0, stream>>>(d_inputArr, d_numOfOdds, d_sumOfOddEvenElems, arrSize);
printf("\nLaunching %d blocks with %d threads...\n\n", blocksPerGrid,
threadsPerBlock);
checkCudaErrors(cudaMemcpyAsync(h_numOfOdds, d_numOfOdds, sizeof(int), cudaMemcpyDeviceToHost, stream));
checkCudaErrors(cudaMemcpyAsync(h_sumOfOddEvenElems, d_sumOfOddEvenElems, 2*sizeof(int), cudaMemcpyDeviceToHost, stream));
printf("Array size = %d Num of Odds = %d Sum of Odds = %d Sum of Evens %d\n", arrSize, h_numOfOdds[0], h_sumOfOddEvenElems[0], h_sumOfOddEvenElems[1]);
printf("\n...Done.\n\n");
oddEvenCountAndSumCG<<<blocksPerGrid, threadsPerBlock, 0, stream>>>(
d_inputArr, d_numOfOdds, d_sumOfOddEvenElems, arrSize);
delete[] h_inputArr;
delete[] h_numOfOdds;
delete[] h_sumOfOddEvenElems;
checkCudaErrors(cudaMemcpyAsync(h_numOfOdds, d_numOfOdds, sizeof(int),
cudaMemcpyDeviceToHost, stream));
checkCudaErrors(cudaMemcpyAsync(h_sumOfOddEvenElems, d_sumOfOddEvenElems,
2 * sizeof(int), cudaMemcpyDeviceToHost,
stream));
checkCudaErrors(cudaStreamSynchronize(stream));
checkCudaErrors(cudaFree(d_inputArr));
checkCudaErrors(cudaFree(d_numOfOdds));
checkCudaErrors(cudaFree(d_sumOfOddEvenElems));
printf("Array size = %d Num of Odds = %d Sum of Odds = %d Sum of Evens %d\n",
arrSize, h_numOfOdds[0], h_sumOfOddEvenElems[0],
h_sumOfOddEvenElems[1]);
printf("\n...Done.\n\n");
return EXIT_SUCCESS;
checkCudaErrors(cudaFreeHost(h_inputArr));
checkCudaErrors(cudaFreeHost(h_numOfOdds));
checkCudaErrors(cudaFreeHost(h_sumOfOddEvenElems));
checkCudaErrors(cudaFree(d_inputArr));
checkCudaErrors(cudaFree(d_numOfOdds));
checkCudaErrors(cudaFree(d_sumOfOddEvenElems));
return EXIT_SUCCESS;
}

4
Samples/binaryPartitionCG/binaryPartitionCG_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/binaryPartitionCG/binaryPartitionCG_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/boxFilterNPP/README.md
View File

@ -27,7 +27,7 @@ x86_64, ppc64le, armv7l
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

4
Samples/boxFilterNPP/boxFilterNPP_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -118,6 +118,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/boxFilterNPP/boxFilterNPP_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -114,6 +114,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/cannyEdgeDetectorNPP/README.md
View File

@ -27,7 +27,7 @@ x86_64, ppc64le, armv7l
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

4
Samples/cannyEdgeDetectorNPP/cannyEdgeDetectorNPP_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -118,6 +118,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/cannyEdgeDetectorNPP/cannyEdgeDetectorNPP_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -114,6 +114,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/concurrentKernels/README.md
View File

@ -24,7 +24,7 @@ x86_64, ppc64le, armv7l
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/concurrentKernels/concurrentKernels_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/concurrentKernels/concurrentKernels_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

6
Samples/conjugateGradientCudaGraphs/Makefile
View File

@ -265,6 +265,12 @@ ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - conjugateGradientCudaGraphs is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))

2
Samples/conjugateGradientCudaGraphs/README.md
View File

@ -30,7 +30,7 @@ cudaStreamBeginCapture, cudaStreamEndCapture, cudaGraphCreate, cudaGraphLaunch,
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

78
Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs.cu
View File

@ -25,7 +25,6 @@
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
* This sample implements a conjugate gradient solver on GPU
* using CUBLAS and CUSPARSE with CUDA Graphs
@ -46,7 +45,6 @@
#include <helper_cuda.h> // helper function CUDA error checking and initialization
#include <helper_functions.h> // helper for shared functions common to CUDA Samples
const char *sSDKname = "conjugateGradientCudaGraphs";
#ifndef WITH_GRAPH
@ -145,12 +143,12 @@ int main(int argc, char **argv) {
/* Generate a random tridiagonal symmetric matrix in CSR format */
N = 1048576;
nz = (N - 2) * 3 + 4;
I = (int *)malloc(sizeof(int) * (N + 1));
J = (int *)malloc(sizeof(int) * nz);
val = (float *)malloc(sizeof(float) * nz);
checkCudaErrors(cudaMallocHost(&I, sizeof(int) * (N + 1)));
checkCudaErrors(cudaMallocHost(&J, sizeof(int) * nz));
checkCudaErrors(cudaMallocHost(&val, sizeof(float) * nz));
genTridiag(I, J, val, N, nz);
x = (float *)malloc(sizeof(float) * N);
checkCudaErrors(cudaMallocHost(&x, sizeof(float) * N));
rhs = (float *)malloc(sizeof(float) * N);
for (int i = 0; i < N; i++) {
@ -192,9 +190,9 @@ int main(int argc, char **argv) {
/* Wrap raw data into cuSPARSE generic API objects */
cusparseSpMatDescr_t matA = NULL;
checkCudaErrors(cusparseCreateCsr(
&matA, N, N, nz, d_row, d_col, d_val, CUSPARSE_INDEX_32I,
CUSPARSE_INDEX_32I, CUSPARSE_INDEX_BASE_ZERO, CUDA_R_32F));
checkCudaErrors(cusparseCreateCsr(&matA, N, N, nz, d_row, d_col, d_val,
CUSPARSE_INDEX_32I, CUSPARSE_INDEX_32I,
CUSPARSE_INDEX_BASE_ZERO, CUDA_R_32F));
cusparseDnVecDescr_t vecx = NULL;
checkCudaErrors(cusparseCreateDnVec(&vecx, N, d_x, CUDA_R_32F));
cusparseDnVecDescr_t vecp = NULL;
@ -206,7 +204,7 @@ int main(int argc, char **argv) {
size_t bufferSize = 0;
checkCudaErrors(cusparseSpMV_bufferSize(
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecx,
&beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &bufferSize));
&beta, vecAx, CUDA_R_32F, CUSPARSE_SPMV_ALG_DEFAULT, &bufferSize));
void *buffer = NULL;
checkCudaErrors(cudaMalloc(&buffer, bufferSize));
@ -234,9 +232,9 @@ int main(int argc, char **argv) {
beta = 0.0;
checkCudaErrors(cusparseSetStream(cusparseHandle, stream1));
checkCudaErrors(cusparseSpMV(
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecx,
&beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
&alpha, matA, vecx, &beta, vecAx, CUDA_R_32F,
CUSPARSE_SPMV_ALG_DEFAULT, buffer));
checkCudaErrors(cublasSetStream(cublasHandle, stream1));
checkCudaErrors(cublasSaxpy(cublasHandle, N, &alpham1, d_Ax, 1, d_r, 1));
@ -248,9 +246,9 @@ int main(int argc, char **argv) {
k = 1;
// First Iteration when k=1 starts
checkCudaErrors(cublasScopy(cublasHandle, N, d_r, 1, d_p, 1));
checkCudaErrors(cusparseSpMV(
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
&beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
&alpha, matA, vecp, &beta, vecAx, CUDA_R_32F,
CUSPARSE_SPMV_ALG_DEFAULT, buffer));
checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
@ -290,9 +288,9 @@ int main(int argc, char **argv) {
checkCudaErrors(
cusparseSetPointerMode(cusparseHandle, CUSPARSE_POINTER_MODE_HOST));
checkCudaErrors(cusparseSpMV(
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
&beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
&alpha, matA, vecp, &beta, vecAx, CUDA_R_32F,
CUSPARSE_SPMV_ALG_DEFAULT, buffer));
checkCudaErrors(cudaMemsetAsync(d_dot, 0, sizeof(float), stream1));
checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
@ -335,8 +333,8 @@ int main(int argc, char **argv) {
checkCudaErrors(cublasSaxpy(cublasHandle, N, &alpha, d_r, 1, d_p, 1));
checkCudaErrors(cusparseSpMV(
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
&beta, vecAx, CUDA_R_32F, CUSPARSE_MV_ALG_DEFAULT, &buffer));
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA, vecp,
&beta, vecAx, CUDA_R_32F, CUSPARSE_SPMV_ALG_DEFAULT, buffer));
cublasSetPointerMode(cublasHandle, CUBLAS_POINTER_MODE_DEVICE);
checkCudaErrors(cublasSdot(cublasHandle, N, d_p, 1, d_Ax, 1, d_dot));
@ -395,23 +393,31 @@ int main(int argc, char **argv) {
cusparseDestroy(cusparseHandle);
cublasDestroy(cublasHandle);
if (matA ) { checkCudaErrors(cusparseDestroySpMat(matA)); }
if (vecx ) { checkCudaErrors(cusparseDestroyDnVec(vecx)); }
if (vecAx ) { checkCudaErrors(cusparseDestroyDnVec(vecAx)); }
if (vecp ) { checkCudaErrors(cusparseDestroyDnVec(vecp)); }
if (matA) {
checkCudaErrors(cusparseDestroySpMat(matA));
}
if (vecx) {
checkCudaErrors(cusparseDestroyDnVec(vecx));
}
if (vecAx) {
checkCudaErrors(cusparseDestroyDnVec(vecAx));
}
if (vecp) {
checkCudaErrors(cusparseDestroyDnVec(vecp));
}
free(I);
free(J);
free(val);
free(x);
checkCudaErrors(cudaFreeHost(I));
checkCudaErrors(cudaFreeHost(J));
checkCudaErrors(cudaFreeHost(val));
checkCudaErrors(cudaFreeHost(x));
free(rhs);
cudaFree(d_col);
cudaFree(d_row);
cudaFree(d_val);
cudaFree(d_x);
cudaFree(d_r);
cudaFree(d_p);
cudaFree(d_Ax);
checkCudaErrors(cudaFree(d_col));
checkCudaErrors(cudaFree(d_row));
checkCudaErrors(cudaFree(d_val));
checkCudaErrors(cudaFree(d_x));
checkCudaErrors(cudaFree(d_r));
checkCudaErrors(cudaFree(d_p));
checkCudaErrors(cudaFree(d_Ax));
printf("Test Summary: Error amount = %f\n", err);
exit((k <= max_iter) ? 0 : 1);

4
Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/conjugateGradientCudaGraphs/conjugateGradientCudaGraphs_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/conjugateGradientMultiBlockCG/README.md
View File

@ -27,7 +27,7 @@ x86_64, ppc64le, aarch64
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

4
Samples/conjugateGradientMultiBlockCG/conjugateGradientMultiBlockCG_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/conjugateGradientMultiBlockCG/conjugateGradientMultiBlockCG_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/conjugateGradientMultiDeviceCG/README.md
View File

@ -30,7 +30,7 @@ cudaMemAdvise, cudaMemPrefetchAsync, cudaLaunchCooperativeKernelMultiDevice, cud
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

138
Samples/conjugateGradientMultiDeviceCG/conjugateGradientMultiDeviceCG.cu
View File

@ -223,8 +223,10 @@ __device__ void gpuDotProduct(float *vecA, float *vecB, int size,
cg::sync(cta);
if (tile32.meta_group_rank() == 0) {
temp_sum = tile32.thread_rank() < tile32.meta_group_size() ? tmp[tile32.thread_rank()] : 0.0;
temp_sum = cg::reduce(tile32, temp_sum, cg::plus<double>());
temp_sum = tile32.thread_rank() < tile32.meta_group_size()
? tmp[tile32.thread_rank()]
: 0.0;
temp_sum = cg::reduce(tile32, temp_sum, cg::plus<double>());
if (tile32.thread_rank() == 0) {
atomicAdd(&grid_dot_result, temp_sum);
@ -239,8 +241,9 @@ __device__ void gpuCopyVector(float *srcA, float *destB, int size,
}
}
__device__ void gpuScaleVectorAndSaxpy(float *x, float *y, float a, float scale, int size,
const cg::multi_grid_group &multi_grid) {
__device__ void gpuScaleVectorAndSaxpy(float *x, float *y, float a, float scale,
int size,
const cg::multi_grid_group &multi_grid) {
for (int i = multi_grid.thread_rank(); i < size; i += multi_grid.size()) {
y[i] = a * x[i] + scale * y[i];
}
@ -360,10 +363,11 @@ std::multimap<std::pair<int, int>, int> getIdenticalGPUs() {
// Filter unsupported devices
if (deviceProp.cooperativeMultiDeviceLaunch &&
deviceProp.concurrentManagedAccess) {
identicalGpus.emplace(std::make_pair(deviceProp.major, deviceProp.minor), i);
identicalGpus.emplace(std::make_pair(deviceProp.major, deviceProp.minor),
i);
}
printf("GPU Device %d: \"%s\" with compute capability %d.%d\n", i,
deviceProp.name, deviceProp.major, deviceProp.minor);
deviceProp.name, deviceProp.major, deviceProp.minor);
}
return identicalGpus;
@ -387,15 +391,17 @@ int main(int argc, char **argv) {
auto bestFit = std::make_pair(it, it);
// use std::distance to find the largest number of GPUs amongst architectures
auto distance = [](decltype(bestFit) p){return std::distance(p.first, p.second);};
auto distance = [](decltype(bestFit) p) {
return std::distance(p.first, p.second);
};
// Read each unique key/pair element in order
for (; it != end; it = gpusByArch.upper_bound(it->first)) {
// first and second are iterators bounded within the architecture group
auto testFit = gpusByArch.equal_range(it->first);
// Always use devices with highest architecture version or whichever has the most devices available
if (distance(bestFit) <= distance(testFit))
bestFit = testFit;
// Always use devices with highest architecture version or whichever has the
// most devices available
if (distance(bestFit) <= distance(testFit)) bestFit = testFit;
}
if (distance(bestFit) < kNumGpusRequired) {
@ -408,33 +414,35 @@ int main(int argc, char **argv) {
std::set<int> bestFitDeviceIds;
// check & select peer-to-peer access capable GPU devices as enabling p2p access between participating
// check & select peer-to-peer access capable GPU devices as enabling p2p
// access between participating
// GPUs gives better performance for multi_grid sync.
for (auto itr = bestFit.first; itr != bestFit.second; itr++) {
int deviceId = itr->second;
checkCudaErrors(cudaSetDevice(deviceId));
std::for_each(itr, bestFit.second, [&deviceId, &bestFitDeviceIds](decltype(*itr) mapPair) {
if (deviceId != mapPair.second)
{
std::for_each(itr, bestFit.second, [&deviceId, &bestFitDeviceIds,
&kNumGpusRequired](
decltype(*itr) mapPair) {
if (deviceId != mapPair.second) {
int access = 0;
checkCudaErrors(cudaDeviceCanAccessPeer(&access, deviceId, mapPair.second));
printf("Device=%d %s Access Peer Device=%d\n", deviceId, access ? "CAN" : "CANNOT", mapPair.second);
checkCudaErrors(
cudaDeviceCanAccessPeer(&access, deviceId, mapPair.second));
printf("Device=%d %s Access Peer Device=%d\n", deviceId,
access ? "CAN" : "CANNOT", mapPair.second);
if (access && bestFitDeviceIds.size() < kNumGpusRequired) {
bestFitDeviceIds.emplace(deviceId);
bestFitDeviceIds.emplace(mapPair.second);
}
else {
} else {
printf("Ignoring device %i (max devices exceeded)\n", mapPair.second);
}
}
});
if (bestFitDeviceIds.size() >= kNumGpusRequired)
{
if (bestFitDeviceIds.size() >= kNumGpusRequired) {
printf("Selected p2p capable devices - ");
for (auto devicesItr = bestFitDeviceIds.begin(); devicesItr != bestFitDeviceIds.end(); devicesItr++)
{
for (auto devicesItr = bestFitDeviceIds.begin();
devicesItr != bestFitDeviceIds.end(); devicesItr++) {
printf("deviceId = %d ", *devicesItr);
}
printf("\n");
@ -442,33 +450,34 @@ int main(int argc, char **argv) {
}
}
// if bestFitDeviceIds.size() == 0 it means the GPUs in system are not p2p capable,
// if bestFitDeviceIds.size() == 0 it means the GPUs in system are not p2p
// capable,
// hence we add it without p2p capability check.
if (!bestFitDeviceIds.size())
{
printf("Devices involved are not p2p capable.. selecting %zu of them\n", kNumGpusRequired);
std::for_each(bestFit.first, bestFit.second, [&bestFitDeviceIds](decltype(*bestFit.first) mapPair) {
if (bestFitDeviceIds.size() < kNumGpusRequired) {
bestFitDeviceIds.emplace(mapPair.second);
}
else {
printf("Ignoring device %i (max devices exceeded)\n", mapPair.second);
}
// Insert the sequence into the deviceIds set
});
}
else
{
// perform cudaDeviceEnablePeerAccess in both directions for all participating devices
// of a cudaLaunchCooperativeKernelMultiDevice call this gives better performance for multi_grid sync.
for (auto p1_itr = bestFitDeviceIds.begin(); p1_itr != bestFitDeviceIds.end(); p1_itr++)
{
if (!bestFitDeviceIds.size()) {
printf("Devices involved are not p2p capable.. selecting %zu of them\n",
kNumGpusRequired);
std::for_each(bestFit.first, bestFit.second,
[&bestFitDeviceIds,
&kNumGpusRequired](decltype(*bestFit.first) mapPair) {
if (bestFitDeviceIds.size() < kNumGpusRequired) {
bestFitDeviceIds.emplace(mapPair.second);
} else {
printf("Ignoring device %i (max devices exceeded)\n",
mapPair.second);
}
// Insert the sequence into the deviceIds set
});
} else {
// perform cudaDeviceEnablePeerAccess in both directions for all
// participating devices of a cudaLaunchCooperativeKernelMultiDevice call
// this gives better performance for multi_grid sync.
for (auto p1_itr = bestFitDeviceIds.begin();
p1_itr != bestFitDeviceIds.end(); p1_itr++) {
checkCudaErrors(cudaSetDevice(*p1_itr));
for (auto p2_itr = bestFitDeviceIds.begin(); p2_itr != bestFitDeviceIds.end(); p2_itr++)
{
if (*p1_itr != *p2_itr)
{
checkCudaErrors(cudaDeviceEnablePeerAccess(*p2_itr, 0 ));
for (auto p2_itr = bestFitDeviceIds.begin();
p2_itr != bestFitDeviceIds.end(); p2_itr++) {
if (*p1_itr != *p2_itr) {
checkCudaErrors(cudaDeviceEnablePeerAccess(*p2_itr, 0));
checkCudaErrors(cudaSetDevice(*p1_itr));
}
}
@ -518,7 +527,7 @@ int main(int argc, char **argv) {
std::cout << "\nRunning on GPUs = " << kNumGpusRequired << std::endl;
cudaStream_t nStreams[kNumGpusRequired];
int sMemSize = sizeof(double) * ((THREADS_PER_BLOCK/32) + 1);
int sMemSize = sizeof(double) * ((THREADS_PER_BLOCK / 32) + 1);
int numBlocksPerSm = INT_MAX;
int numThreads = THREADS_PER_BLOCK;
int numSms = INT_MAX;
@ -530,17 +539,16 @@ int main(int argc, char **argv) {
checkCudaErrors(cudaSetDevice(*deviceId));
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, *deviceId));
int numBlocksPerSm_current=0;
int numBlocksPerSm_current = 0;
checkCudaErrors(cudaOccupancyMaxActiveBlocksPerMultiprocessor(
&numBlocksPerSm_current, multiGpuConjugateGradient, numThreads, sMemSize));
&numBlocksPerSm_current, multiGpuConjugateGradient, numThreads,
sMemSize));
if (numBlocksPerSm > numBlocksPerSm_current)
{
numBlocksPerSm = numBlocksPerSm_current;
if (numBlocksPerSm > numBlocksPerSm_current) {
numBlocksPerSm = numBlocksPerSm_current;
}
if (numSms > deviceProp.multiProcessorCount)
{
numSms = deviceProp.multiProcessorCount;
if (numSms > deviceProp.multiProcessorCount) {
numSms = deviceProp.multiProcessorCount;
}
deviceId++;
}
@ -554,7 +562,7 @@ int main(int argc, char **argv) {
int device_count = 0;
int totalThreadsPerGPU = numSms * numBlocksPerSm * THREADS_PER_BLOCK;
deviceId = bestFitDeviceIds.begin();;
deviceId = bestFitDeviceIds.begin();
while (deviceId != bestFitDeviceIds.end()) {
checkCudaErrors(cudaSetDevice(*deviceId));
checkCudaErrors(cudaStreamCreate(&nStreams[device_count]));
@ -621,14 +629,15 @@ int main(int argc, char **argv) {
printf("Total threads per GPU = %d numBlocksPerSm = %d\n",
numSms * numBlocksPerSm * THREADS_PER_BLOCK, numBlocksPerSm);
dim3 dimGrid(numSms * numBlocksPerSm, 1, 1), dimBlock(THREADS_PER_BLOCK, 1, 1);
dim3 dimGrid(numSms * numBlocksPerSm, 1, 1),
dimBlock(THREADS_PER_BLOCK, 1, 1);
void *kernelArgs[] = {
(void *)&I, (void *)&J, (void *)&val, (void *)&x,
(void *)&Ax, (void *)&p, (void *)&r, (void *)&dot_result,
(void *)&nz, (void *)&N, (void *)&tol,
};
cudaLaunchParams *launchParamsList = (cudaLaunchParams *)malloc(
sizeof(cudaLaunchParams) * kNumGpusRequired);
cudaLaunchParams *launchParamsList =
(cudaLaunchParams *)malloc(sizeof(cudaLaunchParams) * kNumGpusRequired);
for (int i = 0; i < kNumGpusRequired; i++) {
launchParamsList[i].func = (void *)multiGpuConjugateGradient;
launchParamsList[i].gridDim = dimGrid;
@ -645,12 +654,11 @@ int main(int argc, char **argv) {
cudaCooperativeLaunchMultiDeviceNoPreSync |
cudaCooperativeLaunchMultiDeviceNoPostSync));
checkCudaErrors(cudaMemPrefetchAsync(x, sizeof(float) * N, cudaCpuDeviceId));
checkCudaErrors(
cudaMemPrefetchAsync(x, sizeof(float) * N, cudaCpuDeviceId));
checkCudaErrors(
cudaMemPrefetchAsync(dot_result, sizeof(double), cudaCpuDeviceId));
cudaMemPrefetchAsync(dot_result, sizeof(double), cudaCpuDeviceId));
deviceId = bestFitDeviceIds.begin();;
deviceId = bestFitDeviceIds.begin();
device_count = 0;
while (deviceId != bestFitDeviceIds.end()) {
checkCudaErrors(cudaSetDevice(*deviceId));
@ -658,7 +666,7 @@ int main(int argc, char **argv) {
deviceId++;
}
r1 = *dot_result;
r1 = (float)*dot_result;
printf("GPU Final, residual = %e \n ", sqrt(r1));

4
Samples/conjugateGradientMultiDeviceCG/conjugateGradientMultiDeviceCG_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -109,6 +109,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/conjugateGradientMultiDeviceCG/conjugateGradientMultiDeviceCG_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -105,6 +105,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

6
Samples/cuSolverDn_LinearSolver/Makefile
View File

@ -271,6 +271,12 @@ ifeq ($(TARGET_ARCH),armv7l)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - cuSolverDn_LinearSolver is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(TARGET_OS),linux)
ALL_CCFLAGS += -Xcompiler \"-Wl,--no-as-needed\"
endif

2
Samples/cuSolverDn_LinearSolver/README.md
View File

@ -27,7 +27,7 @@ x86_64, ppc64le, aarch64
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

4
Samples/cuSolverDn_LinearSolver/cuSolverDn_LinearSolver_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -110,6 +110,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/cuSolverDn_LinearSolver/cuSolverDn_LinearSolver_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -106,6 +106,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

6
Samples/cuSolverSp_LinearSolver/Makefile
View File

@ -265,6 +265,12 @@ ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
SAMPLE_ENABLED := 1
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - cuSolverSp_LinearSolver is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ifeq ($(TARGET_OS),linux)
ALL_CCFLAGS += -Xcompiler \"-Wl,--no-as-needed\"
endif

2
Samples/cuSolverSp_LinearSolver/README.md
View File

@ -27,7 +27,7 @@ x86_64, ppc64le, armv7l
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

6
Samples/cuSolverSp_LinearSolver/cuSolverSp_LinearSolver.cpp
View File

@ -495,13 +495,13 @@ int main(int argc, char *argv[]) {
size_t bufferSize = 0;
checkCudaErrors(cusparseSpMV_bufferSize(
cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE, &minus_one, matA, vecx,
&one, vecAx, CUDA_R_64F, CUSPARSE_MV_ALG_DEFAULT, &bufferSize));
&one, vecAx, CUDA_R_64F, CUSPARSE_SPMV_ALG_DEFAULT, &bufferSize));
void *buffer = NULL;
checkCudaErrors(cudaMalloc(&buffer, bufferSize));
checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
&minus_one, matA, vecx, &one, vecAx, CUDA_R_64F,
CUSPARSE_MV_ALG_DEFAULT, &buffer));
CUSPARSE_SPMV_ALG_DEFAULT, buffer));
checkCudaErrors(cudaMemcpyAsync(h_r, d_r, sizeof(double) * rowsA,
cudaMemcpyDeviceToHost, stream));
@ -559,7 +559,7 @@ int main(int argc, char *argv[]) {
checkCudaErrors(cusparseSpMV(cusparseHandle, CUSPARSE_OPERATION_NON_TRANSPOSE,
&minus_one, matA, vecx, &one, vecAx, CUDA_R_64F,
CUSPARSE_MV_ALG_DEFAULT, &buffer));
CUSPARSE_SPMV_ALG_DEFAULT, buffer));
checkCudaErrors(cudaMemcpyAsync(h_x, d_x, sizeof(double) * colsA,
cudaMemcpyDeviceToHost, stream));

4
Samples/cuSolverSp_LinearSolver/cuSolverSp_LinearSolver_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -110,6 +110,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/cuSolverSp_LinearSolver/cuSolverSp_LinearSolver_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -106,6 +106,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/cudaCompressibleMemory/README.md
View File

@ -30,7 +30,7 @@ cudaMalloc, cudaFree
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/cudaCompressibleMemory/cudaCompressibleMemory_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -109,6 +109,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/cudaCompressibleMemory/cudaCompressibleMemory_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -105,6 +105,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

6
Samples/cudaNvSci/Makefile
View File

@ -279,6 +279,12 @@ ifeq ($(TARGET_ARCH),armv7l)
SAMPLE_ENABLED := 0
endif
# This sample is not supported on QNX
ifeq ($(TARGET_OS),qnx)
$(info >>> WARNING - cudaNvSci is not supported on QNX - waiving sample <<<)
SAMPLE_ENABLED := 0
endif
ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))

2
Samples/cudaNvSci/README.md
View File

@ -30,7 +30,7 @@ cudaImportExternalMemory, cudaExternalMemoryGetMappedBuffer, cudaExternalMemoryG
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

2
Samples/cudaNvSciNvMedia/README.md
View File

@ -30,7 +30,7 @@ cudaImportExternalMemory, cudaExternalMemoryGetMappedBuffer, cudaExternalMemoryG
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

2
Samples/cudaOpenMP/README.md
View File

@ -30,7 +30,7 @@ cudaMalloc, cudaFree, cudaMemcpy
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

4
Samples/cudaOpenMP/cudaOpenMP_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -109,6 +109,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/cudaOpenMP/cudaOpenMP_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -105,6 +105,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/cudaTensorCoreGemm/README.md
View File

@ -31,7 +31,7 @@ cudaMallocManaged, cudaDeviceSynchronize, cudaFuncSetAttribute, cudaEventCreate,
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/cudaTensorCoreGemm/cudaTensorCoreGemm_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/cudaTensorCoreGemm/cudaTensorCoreGemm_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/deviceQuery/README.md
View File

@ -27,7 +27,7 @@ cudaSetDevice, cudaGetDeviceCount, cudaGetDeviceProperties, cudaDriverGetVersion
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

21
Samples/deviceQuery/deviceQuery.cpp
View File

@ -112,10 +112,10 @@ int main(int argc, char **argv) {
char msg[256];
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
sprintf_s(msg, sizeof(msg),
" Total amount of global memory: %.0f MBytes "
"(%llu bytes)\n",
static_cast<float>(deviceProp.totalGlobalMem / 1048576.0f),
(unsigned long long)deviceProp.totalGlobalMem);
" Total amount of global memory: %.0f MBytes "
"(%llu bytes)\n",
static_cast<float>(deviceProp.totalGlobalMem / 1048576.0f),
(unsigned long long)deviceProp.totalGlobalMem);
#else
snprintf(msg, sizeof(msg),
" Total amount of global memory: %.0f MBytes "
@ -125,7 +125,7 @@ int main(int argc, char **argv) {
#endif
printf("%s", msg);
printf(" (%2d) Multiprocessors, (%3d) CUDA Cores/MP: %d CUDA Cores\n",
printf(" (%03d) Multiprocessors, (%03d) CUDA Cores/MP: %d CUDA Cores\n",
deviceProp.multiProcessorCount,
_ConvertSMVer2Cores(deviceProp.major, deviceProp.minor),
_ConvertSMVer2Cores(deviceProp.major, deviceProp.minor) *
@ -250,8 +250,7 @@ int main(int argc, char **argv) {
"device)",
"Exclusive Process (many threads in one process is able to use "
"::cudaSetDevice() with this device)",
"Unknown",
NULL};
"Unknown", NULL};
printf(" Compute Mode:\n");
printf(" < %s >\n", sComputeMode[deviceProp.computeMode]);
}
@ -272,7 +271,7 @@ int main(int argc, char **argv) {
// must be enabled to support this
&& prop[i].tccDriver
#endif
) {
) {
// This is an array of P2P capable GPUs
gpuid[gpu_p2p_count++] = i;
}
@ -307,7 +306,8 @@ int main(int argc, char **argv) {
// driver version
sProfileString += ", CUDA Driver Version = ";
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
sprintf_s(cTemp, 10, "%d.%d", driverVersion/1000, (driverVersion%100)/10);
sprintf_s(cTemp, 10, "%d.%d", driverVersion / 1000,
(driverVersion % 100) / 10);
#else
snprintf(cTemp, sizeof(cTemp), "%d.%d", driverVersion / 1000,
(driverVersion % 100) / 10);
@ -317,7 +317,8 @@ int main(int argc, char **argv) {
// Runtime version
sProfileString += ", CUDA Runtime Version = ";
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
sprintf_s(cTemp, 10, "%d.%d", runtimeVersion/1000, (runtimeVersion%100)/10);
sprintf_s(cTemp, 10, "%d.%d", runtimeVersion / 1000,
(runtimeVersion % 100) / 10);
#else
snprintf(cTemp, sizeof(cTemp), "%d.%d", runtimeVersion / 1000,
(runtimeVersion % 100) / 10);

4
Samples/deviceQuery/deviceQuery_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/deviceQuery/deviceQuery_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/dmmaTensorCoreGemm/README.md
View File

@ -27,7 +27,7 @@ cudaMallocManaged, cudaDeviceSynchronize, cudaFuncSetAttribute, cudaEventCreate,
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/dmmaTensorCoreGemm/dmmaTensorCoreGemm_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/dmmaTensorCoreGemm/dmmaTensorCoreGemm_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/globalToShmemAsyncCopy/README.md
View File

@ -30,7 +30,7 @@ cudaEventCreate, cudaEventRecord, cudaEventQuery, cudaEventDestroy, cudaEventEla
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

1681
Samples/globalToShmemAsyncCopy/globalToShmemAsyncCopy.cu
View File

File diff suppressed because it is too large Load Diff

4
Samples/globalToShmemAsyncCopy/globalToShmemAsyncCopy_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/globalToShmemAsyncCopy/globalToShmemAsyncCopy_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/immaTensorCoreGemm/README.md
View File

@ -27,7 +27,7 @@ cudaMallocManaged, cudaDeviceSynchronize, cudaFuncSetAttribute, cudaEventCreate,
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/immaTensorCoreGemm/immaTensorCoreGemm_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/immaTensorCoreGemm/immaTensorCoreGemm_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/jacobiCudaGraphs/README.md
View File

@ -25,7 +25,7 @@ cudaStreamBeginCapture, cudaStreamEndCapture, cudaGraphCreate, cudaGraphLaunch,
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/jacobiCudaGraphs/jacobiCudaGraphs_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -109,6 +109,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/jacobiCudaGraphs/jacobiCudaGraphs_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -105,6 +105,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

9
Samples/jacobiCudaGraphs/main.cpp
View File

@ -100,8 +100,10 @@ int main(int argc, char **argv) {
double *b = NULL;
float *A = NULL;
b = (double *)calloc(N_ROWS, sizeof(double));
A = (float *)calloc(N_ROWS * N_ROWS, sizeof(float));
checkCudaErrors(cudaMallocHost(&b, N_ROWS * sizeof(double)));
memset(b, 0, N_ROWS * sizeof(double));
checkCudaErrors(cudaMallocHost(&A, N_ROWS * N_ROWS * sizeof(float)));
memset(A, 0, N_ROWS * N_ROWS * sizeof(float));
createLinearSystem(A, b);
double *x = NULL;
@ -170,6 +172,9 @@ int main(int argc, char **argv) {
checkCudaErrors(cudaFree(d_x));
checkCudaErrors(cudaFree(d_x_new));
checkCudaErrors(cudaFreeHost(A));
checkCudaErrors(cudaFreeHost(b));
printf("&&&& jacobiCudaGraphs %s\n",
(fabs(sum - sumGPU) < conv_threshold) ? "PASSED" : "FAILED");

2
Samples/matrixMul/README.md
View File

@ -27,7 +27,7 @@ cudaEventCreate, cudaEventRecord, cudaEventQuery, cudaEventDestroy, cudaEventEla
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/matrixMul/matrixMul_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/matrixMul/matrixMul_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -104,6 +104,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

2
Samples/matrixMulDrv/README.md
View File

@ -27,7 +27,7 @@ cuModuleLoad, cuModuleLoadDataEx, cuModuleGetFunction, cuMemAlloc, cuMemFree, cu
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
## Build and Run

4
Samples/matrixMulDrv/matrixMulDrv_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -112,6 +112,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

4
Samples/matrixMulDrv/matrixMulDrv_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -108,6 +108,6 @@
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

19
Samples/memMapIPCDrv/Makefile
View File

@ -302,14 +302,10 @@ LIBRARIES :=
################################################################################
FATBIN_FILE := memMapIpc_kernel${TARGET_SIZE}.fatbin
PTX_FILE := memMapIpc_kernel${TARGET_SIZE}.ptx
# Gencode arguments
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
SMS ?= 35 37 50 52 60 61 70 72 75 80 86
else
SMS ?= 35 37 50 52 60 61 70 75 80 86
endif
SMS ?=
ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
@ -395,7 +391,7 @@ endif
# Target rules
all: build
build: memMapIPCDrv $(FATBIN_FILE)
build: memMapIPCDrv $(PTX_FILE)
check.deps:
ifeq ($(SAMPLE_ENABLED),0)
@ -404,8 +400,8 @@ else
@echo "Sample is ready - all dependencies have been met"
endif
$(FATBIN_FILE): memMapIpc_kernel.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -fatbin $<
$(PTX_FILE): memMapIpc_kernel.cu
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -ptx $<
$(EXEC) mkdir -p data
$(EXEC) cp -f $@ ./data
$(EXEC) mkdir -p ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
@ -426,9 +422,8 @@ run: build
$(EXEC) ./memMapIPCDrv
clean:
rm -f memMapIPCDrv helper_multiprocess.o memMapIpc.o data/$(FATBIN_FILE) $(FATBIN_FILE)
rm -f memMapIPCDrv helper_multiprocess.o memMapIpc.o data/$(PTX_FILE) $(PTX_FILE)
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/memMapIPCDrv
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/$(FATBIN_FILE)
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/$(PTX_FILE)
clobber: clean

2
Samples/memMapIPCDrv/README.md
View File

@ -30,7 +30,7 @@ cuModuleLoad, cuModuleLoadDataEx, cuModuleGetFunction, cuLaunchKernel, cuMemcpyD
## Prerequisites
Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Download and install the [CUDA Toolkit 11.3](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## Build and Run

10
Samples/memMapIPCDrv/memMapIPCDrv_vs2017.vcxproj
View File

@ -38,7 +38,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -67,7 +67,7 @@
<OutputFile>$(OutDir)/memMapIPCDrv.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,compute_35;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;</CodeGeneration>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
@ -105,14 +105,14 @@
<ItemGroup>
<ClCompile Include="memMapIpc.cpp" />
<CudaCompile Include="memMapIpc_kernel.cu">
<CompileOut Condition="'$(Platform)'=='x64'">data/%(Filename)64.fatbin</CompileOut>
<NvccCompilation>fatbin</NvccCompilation>
<CompileOut Condition="'$(Platform)'=='x64'">data/%(Filename)64.ptx</CompileOut>
<NvccCompilation>ptx</NvccCompilation>
</CudaCompile>
<ClCompile Include="../../Common/helper_multiprocess.cpp" />
<ClInclude Include="../../Common/helper_multiprocess.h" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

10
Samples/memMapIPCDrv/memMapIPCDrv_vs2019.vcxproj
View File

@ -34,7 +34,7 @@
</PropertyGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
<ImportGroup Label="ExtensionSettings">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.props" />
</ImportGroup>
<ImportGroup Label="PropertySheets">
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@ -63,7 +63,7 @@
<OutputFile>$(OutDir)/memMapIPCDrv.exe</OutputFile>
</Link>
<CudaCompile>
<CodeGeneration>compute_35,compute_35;compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;</CodeGeneration>
<CodeGeneration>compute_35,compute_35;</CodeGeneration>
<AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
<Include>./;../../Common</Include>
<Defines>WIN32</Defines>
@ -101,14 +101,14 @@
<ItemGroup>
<ClCompile Include="memMapIpc.cpp" />
<CudaCompile Include="memMapIpc_kernel.cu">
<CompileOut Condition="'$(Platform)'=='x64'">data/%(Filename)64.fatbin</CompileOut>
<NvccCompilation>fatbin</NvccCompilation>
<CompileOut Condition="'$(Platform)'=='x64'">data/%(Filename)64.ptx</CompileOut>
<NvccCompilation>ptx</NvccCompilation>
</CudaCompile>
<ClCompile Include="../../Common/helper_multiprocess.cpp" />
<ClInclude Include="../../Common/helper_multiprocess.h" />
</ItemGroup>
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
<Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
<Import Project="$(CUDAPropsPath)\CUDA 11.3.targets" />
</ImportGroup>
</Project>

87
Samples/memMapIPCDrv/memMapIpc.cpp
View File

@ -64,9 +64,13 @@ typedef struct shmStruct_st {
int sense;
} shmStruct;
// define input fatbin file
#ifndef FATBIN_FILE
#define FATBIN_FILE "memMapIpc_kernel64.fatbin"
bool findModulePath(const char *, string &, char **, string &);
// define input ptx file for different platforms
#if defined(_WIN64) || defined(__LP64__)
#define PTX_FILE "memMapIpc_kernel64.ptx"
#else
#define PTX_FILE "memMapIpc_kernel32.ptx"
#endif
// `ipcHandleTypeFlag` specifies the platform specific handle type this sample
@ -255,23 +259,44 @@ static void memMapUnmapAndFreeMemory(CUdeviceptr dptr, size_t size) {
static void memMapGetDeviceFunction(char **argv) {
// first search for the module path before we load the results
string module_path;
std::ostringstream fatbin;
if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin)) {
exit(EXIT_FAILURE);
string module_path, ptx_source;
if (!findModulePath(PTX_FILE, module_path, argv, ptx_source)) {
if (!findModulePath("memMapIpc_kernel.cubin", module_path, argv,
ptx_source)) {
printf(
"> findModulePath could not find <simpleMemMapIpc> ptx or cubin\n");
exit(EXIT_FAILURE);
}
} else {
printf("> initCUDA loading module: <%s>\n", module_path.c_str());
}
if (!fatbin.str().size()) {
printf("fatbin file empty. exiting..\n");
exit(EXIT_FAILURE);
// Create module from binary file (PTX or CUBIN)
if (module_path.rfind("ptx") != string::npos) {
// in this branch we use compilation with parameters
const unsigned int jitNumOptions = 3;
CUjit_option *jitOptions = new CUjit_option[jitNumOptions];
void **jitOptVals = new void *[jitNumOptions];
// set up size of compilation log buffer
jitOptions[0] = CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES;
int jitLogBufferSize = 1024;
jitOptVals[0] = (void *)(size_t)jitLogBufferSize;
// set up pointer to the compilation log buffer
jitOptions[1] = CU_JIT_INFO_LOG_BUFFER;
char *jitLogBuffer = new char[jitLogBufferSize];
jitOptVals[1] = jitLogBuffer;
// set up pointer to set the Maximum # of registers for a particular kernel
jitOptions[2] = CU_JIT_MAX_REGISTERS;
int jitRegCount = 32;
jitOptVals[2] = (void *)(size_t)jitRegCount;
checkCudaErrors(cuModuleLoadDataEx(&cuModule, ptx_source.c_str(),
jitNumOptions, jitOptions,
(void **)jitOptVals));
printf("> PTX JIT log:\n%s\n", jitLogBuffer);
} else {
checkCudaErrors(cuModuleLoad(&cuModule, module_path.c_str()));
}
// Create module from binary file (FATBIN)
checkCudaErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));
// Get function handle from module
checkCudaErrors(
cuModuleGetFunction(&_memMapIpc_kernel, cuModule, "memMapIpc_kernel"));
@ -585,3 +610,37 @@ int main(int argc, char **argv) {
return EXIT_SUCCESS;
#endif
}
bool inline findModulePath(const char *module_file, string &module_path,
char **argv, string &ptx_source) {
char *actual_path = sdkFindFilePath(module_file, argv[0]);
if (actual_path) {
module_path = actual_path;
} else {
printf("> findModulePath file not found: <%s> \n", module_file);
return false;
}
if (module_path.empty()) {
printf("> findModulePath could not find file: <%s> \n", module_file);
return false;
} else {
printf("> findModulePath found file at <%s>\n", module_path.c_str());
if (module_path.rfind(".ptx") != string::npos) {
FILE *fp = fopen(module_path.c_str(), "rb");
fseek(fp, 0, SEEK_END);
int file_size = ftell(fp);
char *buf = new char[file_size + 1];
fseek(fp, 0, SEEK_SET);
fread(buf, sizeof(char), file_size, fp);
fclose(fp);
buf[file_size] = '\0';
ptx_source = buf;
delete[] buf;
}
return true;
}
}

Some files were not shown because too many files have changed in this diff Show More