NASP/cuda-samples
85
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2026-10-26 06:38:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

Apply consistent code formatting across the repo. Add clang-format and pre-commit hooks.

Browse Source
This commit is contained in:
Rob Armstrong
2025-03-27 10:30:07 -07:00
parent 2cd58fbc9a
commit ceab6e8bcc
782 changed files with 107230 additions and 106548 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
Samples/4_CUDA_Libraries/cudaNvSci/README.md
View File

@@ -37,4 +37,3 @@ Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-d
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
## References (for more details)

1038
Samples/4_CUDA_Libraries/cudaNvSci/cudaNvSci.cpp
View File

File diff suppressed because it is too large Load Diff

113
Samples/4_CUDA_Libraries/cudaNvSci/cudaNvSci.h
View File

@@ -34,68 +34,79 @@
#include <nvscisync.h>
#include <vector>
#define checkNvSciErrors(call) \
do { \
NvSciError _status = call; \
if (NvSciError_Success != _status) { \
printf( \
"NVSCI call in file '%s' in line %i returned" \
" %d, expected %d\n", \
__FILE__, __LINE__, _status, NvSciError_Success); \
fflush(stdout); \
exit(EXIT_FAILURE); \
} \
} while (0)
#define checkNvSciErrors(call) \
do { \
NvSciError _status = call; \
if (NvSciError_Success != _status) { \
printf("NVSCI call in file '%s' in line %i returned" \
" %d, expected %d\n", \
__FILE__, \
__LINE__, \
_status, \
NvSciError_Success); \
fflush(stdout); \
exit(EXIT_FAILURE); \
} \
} while (0)
extern void rotateKernel(cudaTextureObject_t &texObj, const float angle,
unsigned int *d_outputData, const int imageWidth,
const int imageHeight, cudaStream_t stream);
extern void rotateKernel(cudaTextureObject_t &texObj,
const float angle,
unsigned int *d_outputData,
const int imageWidth,
const int imageHeight,
cudaStream_t stream);
extern void launchGrayScaleKernel(unsigned int *d_rgbaImage,
std::string image_filename, size_t imageWidth,
size_t imageHeight, cudaStream_t stream);
std::string image_filename,
size_t imageWidth,
size_t imageHeight,
cudaStream_t stream);
class cudaNvSci {
private:
int m_isMultiGPU;
int m_cudaNvSciSignalDeviceId;
int m_cudaNvSciWaitDeviceId;
unsigned char *image_data;
size_t m_bufSize;
size_t imageWidth;
size_t imageHeight;
class cudaNvSci
{
private:
int m_isMultiGPU;
int m_cudaNvSciSignalDeviceId;
int m_cudaNvSciWaitDeviceId;
unsigned char *image_data;
size_t m_bufSize;
size_t imageWidth;
size_t imageHeight;
public:
NvSciSyncModule syncModule;
NvSciBufModule buffModule;
NvSciSyncAttrList syncUnreconciledList[2];
NvSciSyncAttrList syncReconciledList;
NvSciSyncAttrList syncConflictList;
public:
NvSciSyncModule syncModule;
NvSciBufModule buffModule;
NvSciSyncAttrList syncUnreconciledList[2];
NvSciSyncAttrList syncReconciledList;
NvSciSyncAttrList syncConflictList;
NvSciBufAttrList rawBufUnreconciledList[2];
NvSciBufAttrList imageBufUnreconciledList[2];
NvSciBufAttrList rawBufReconciledList;
NvSciBufAttrList buffConflictList;
NvSciBufAttrList imageBufReconciledList;
NvSciBufAttrList imageBufConflictList;
NvSciBufAttrList buffAttrListOut;
NvSciBufAttrList rawBufUnreconciledList[2];
NvSciBufAttrList imageBufUnreconciledList[2];
NvSciBufAttrList rawBufReconciledList;
NvSciBufAttrList buffConflictList;
NvSciBufAttrList imageBufReconciledList;
NvSciBufAttrList imageBufConflictList;
NvSciBufAttrList buffAttrListOut;
NvSciSyncObj syncObj;
NvSciBufObj rawBufObj;
NvSciBufObj imageBufObj;
NvSciSyncFence *fence;
NvSciSyncObj syncObj;
NvSciBufObj rawBufObj;
NvSciBufObj imageBufObj;
NvSciSyncFence *fence;
cudaNvSci(int isMultiGPU, std::vector<int> &deviceIds,
unsigned char *image_data, size_t imageWidth, size_t imageHeight);
cudaNvSci(int isMultiGPU,
std::vector<int> &deviceIds,
unsigned char *image_data,
size_t imageWidth,
size_t imageHeight);
void initNvSci();
void initNvSci();
void runCudaNvSci(std::string &image_filename);
void runCudaNvSci(std::string &image_filename);
void createNvSciRawBufObj();
void createNvSciRawBufObj();
void createNvSciSyncObj();
void createNvSciSyncObj();
void createNvSciBufImageObj();
void createNvSciBufImageObj();
};
#endif // CUDANVSCI_H
#endif // CUDANVSCI_H

127
Samples/4_CUDA_Libraries/cudaNvSci/imageKernels.cu
View File

@@ -30,91 +30,90 @@
#include <helper_image.h>
// convert floating point rgba color to 32-bit integer
__device__ unsigned int rgbaFloatToInt(float4 rgba) {
rgba.x = __saturatef(rgba.x); // clamp to [0.0, 1.0]
rgba.y = __saturatef(rgba.y);
rgba.z = __saturatef(rgba.z);
rgba.w = __saturatef(rgba.w);
return ((unsigned int)(rgba.w * 255.0f) << 24) |
((unsigned int)(rgba.z * 255.0f) << 16) |
((unsigned int)(rgba.y * 255.0f) << 8) |
((unsigned int)(rgba.x * 255.0f));
__device__ unsigned int rgbaFloatToInt(float4 rgba)
{
rgba.x = __saturatef(rgba.x); // clamp to [0.0, 1.0]
rgba.y = __saturatef(rgba.y);
rgba.z = __saturatef(rgba.z);
rgba.w = __saturatef(rgba.w);
return ((unsigned int)(rgba.w * 255.0f) << 24) | ((unsigned int)(rgba.z * 255.0f) << 16)
| ((unsigned int)(rgba.y * 255.0f) << 8) | ((unsigned int)(rgba.x * 255.0f));
}
////////////////////////////////////////////////////////////////////////////////
//! Rotate an image using texture lookups
//! @param outputData output data in global memory
////////////////////////////////////////////////////////////////////////////////
static __global__ void transformKernel(unsigned int *outputData, int width,
int height, float theta,
cudaTextureObject_t tex) {
// calculate normalized texture coordinates
unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
static __global__ void
transformKernel(unsigned int *outputData, int width, int height, float theta, cudaTextureObject_t tex)
{
// calculate normalized texture coordinates
unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
float u = (float)x - (float)width / 2;
float v = (float)y - (float)height / 2;
float tu = u * cosf(theta) - v * sinf(theta);
float tv = v * cosf(theta) + u * sinf(theta);
float u = (float)x - (float)width / 2;
float v = (float)y - (float)height / 2;
float tu = u * cosf(theta) - v * sinf(theta);
float tv = v * cosf(theta) + u * sinf(theta);
tu /= (float)width;
tv /= (float)height;
tu /= (float)width;
tv /= (float)height;
// read from texture and write to global memory
float4 pix = tex2D<float4>(tex, tu + 0.5f, tv + 0.5f);
unsigned int pixelInt = rgbaFloatToInt(pix);
outputData[y * width + x] = pixelInt;
// read from texture and write to global memory
float4 pix = tex2D<float4>(tex, tu + 0.5f, tv + 0.5f);
unsigned int pixelInt = rgbaFloatToInt(pix);
outputData[y * width + x] = pixelInt;
}
static __global__ void rgbToGrayscaleKernel(unsigned int *rgbaImage,
size_t imageWidth,
size_t imageHeight) {
size_t gidX = blockDim.x * blockIdx.x + threadIdx.x;
static __global__ void rgbToGrayscaleKernel(unsigned int *rgbaImage, size_t imageWidth, size_t imageHeight)
{
size_t gidX = blockDim.x * blockIdx.x + threadIdx.x;
uchar4 *pixArray = (uchar4 *)rgbaImage;
uchar4 *pixArray = (uchar4 *)rgbaImage;
for (int pixId = gidX; pixId < imageWidth * imageHeight;
pixId += gridDim.x * blockDim.x) {
uchar4 dataA = pixArray[pixId];
unsigned char grayscale =
(unsigned char)(dataA.x * 0.3 + dataA.y * 0.59 + dataA.z * 0.11);
uchar4 dataB = make_uchar4(grayscale, grayscale, grayscale, 0);
pixArray[pixId] = dataB;
}
for (int pixId = gidX; pixId < imageWidth * imageHeight; pixId += gridDim.x * blockDim.x) {
uchar4 dataA = pixArray[pixId];
unsigned char grayscale = (unsigned char)(dataA.x * 0.3 + dataA.y * 0.59 + dataA.z * 0.11);
uchar4 dataB = make_uchar4(grayscale, grayscale, grayscale, 0);
pixArray[pixId] = dataB;
}
}
void launchGrayScaleKernel(unsigned int *d_rgbaImage,
std::string image_filename, size_t imageWidth,
size_t imageHeight, cudaStream_t stream) {
int numThreadsPerBlock = 1024;
int numOfBlocks = (imageWidth * imageHeight) / numThreadsPerBlock;
std::string image_filename,
size_t imageWidth,
size_t imageHeight,
cudaStream_t stream)
{
int numThreadsPerBlock = 1024;
int numOfBlocks = (imageWidth * imageHeight) / numThreadsPerBlock;
rgbToGrayscaleKernel<<<numOfBlocks, numThreadsPerBlock, 0, stream>>>(
d_rgbaImage, imageWidth, imageHeight);
rgbToGrayscaleKernel<<<numOfBlocks, numThreadsPerBlock, 0, stream>>>(d_rgbaImage, imageWidth, imageHeight);
unsigned int *outputData;
checkCudaErrors(cudaMallocHost((void**)&outputData, sizeof(unsigned int) * imageWidth * imageHeight));
checkCudaErrors(cudaMemcpyAsync(
outputData, d_rgbaImage, sizeof(unsigned int) * imageWidth * imageHeight,
cudaMemcpyDeviceToHost, stream));
checkCudaErrors(cudaStreamSynchronize(stream));
unsigned int *outputData;
checkCudaErrors(cudaMallocHost((void **)&outputData, sizeof(unsigned int) * imageWidth * imageHeight));
checkCudaErrors(cudaMemcpyAsync(
outputData, d_rgbaImage, sizeof(unsigned int) * imageWidth * imageHeight, cudaMemcpyDeviceToHost, stream));
checkCudaErrors(cudaStreamSynchronize(stream));
char outputFilename[1024];
strcpy(outputFilename, image_filename.c_str());
strcpy(outputFilename + image_filename.length() - 4, "_out.ppm");
sdkSavePPM4ub(outputFilename, (unsigned char *)outputData, imageWidth,
imageHeight);
printf("Wrote '%s'\n", outputFilename);
char outputFilename[1024];
strcpy(outputFilename, image_filename.c_str());
strcpy(outputFilename + image_filename.length() - 4, "_out.ppm");
sdkSavePPM4ub(outputFilename, (unsigned char *)outputData, imageWidth, imageHeight);
printf("Wrote '%s'\n", outputFilename);
checkCudaErrors(cudaFreeHost(outputData));
checkCudaErrors(cudaFreeHost(outputData));
}
void rotateKernel(cudaTextureObject_t &texObj, const float angle,
unsigned int *d_outputData, const int imageWidth,
const int imageHeight, cudaStream_t stream) {
dim3 dimBlock(8, 8, 1);
dim3 dimGrid(imageWidth / dimBlock.x, imageHeight / dimBlock.y, 1);
void rotateKernel(cudaTextureObject_t &texObj,
const float angle,
unsigned int *d_outputData,
const int imageWidth,
const int imageHeight,
cudaStream_t stream)
{
dim3 dimBlock(8, 8, 1);
dim3 dimGrid(imageWidth / dimBlock.x, imageHeight / dimBlock.y, 1);
transformKernel<<<dimGrid, dimBlock, 0, stream>>>(d_outputData, imageWidth,
imageHeight, angle, texObj);
transformKernel<<<dimGrid, dimBlock, 0, stream>>>(d_outputData, imageWidth, imageHeight, angle, texObj);
}

111
Samples/4_CUDA_Libraries/cudaNvSci/main.cpp
View File

@@ -26,76 +26,77 @@
*/
#include <cuda.h>
#include <vector>
#include "cudaNvSci.h"
#include <helper_cuda.h>
#include <helper_image.h>
#include <vector>
void loadImageData(const std::string &filename, const char **argv,
unsigned char **image_data, uint32_t &imageWidth,
uint32_t &imageHeight) {
// load image (needed so we can get the width and height before we create
// the window
char *image_path = sdkFindFilePath(filename.c_str(), argv[0]);
#include "cudaNvSci.h"
if (image_path == 0) {
printf("Error finding image file '%s'\n", filename.c_str());
exit(EXIT_FAILURE);
}
void loadImageData(const std::string &filename,
const char **argv,
unsigned char **image_data,
uint32_t &imageWidth,
uint32_t &imageHeight)
{
// load image (needed so we can get the width and height before we create
// the window
char *image_path = sdkFindFilePath(filename.c_str(), argv[0]);
sdkLoadPPM4(image_path, image_data, &imageWidth, &imageHeight);
if (image_path == 0) {
printf("Error finding image file '%s'\n", filename.c_str());
exit(EXIT_FAILURE);
}
if (!image_data) {
printf("Error opening file '%s'\n", image_path);
exit(EXIT_FAILURE);
}
sdkLoadPPM4(image_path, image_data, &imageWidth, &imageHeight);
printf("Loaded '%s', %d x %d pixels\n", image_path, imageWidth, imageHeight);
if (!image_data) {
printf("Error opening file '%s'\n", image_path);
exit(EXIT_FAILURE);
}
printf("Loaded '%s', %d x %d pixels\n", image_path, imageWidth, imageHeight);
}
int main(int argc, const char **argv) {
int numOfGPUs = 0;
std::vector<int> deviceIds;
checkCudaErrors(cudaGetDeviceCount(&numOfGPUs));
int main(int argc, const char **argv)
{
int numOfGPUs = 0;
std::vector<int> deviceIds;
checkCudaErrors(cudaGetDeviceCount(&numOfGPUs));
printf("%d GPUs found\n", numOfGPUs);
if (!numOfGPUs) {
exit(EXIT_WAIVED);
} else {
for (int devID = 0; devID < numOfGPUs; devID++) {
int major = 0, minor = 0;
checkCudaErrors(cudaDeviceGetAttribute(
&major, cudaDevAttrComputeCapabilityMajor, devID));
checkCudaErrors(cudaDeviceGetAttribute(
&minor, cudaDevAttrComputeCapabilityMinor, devID));
if (major >= 6) {
deviceIds.push_back(devID);
}
printf("%d GPUs found\n", numOfGPUs);
if (!numOfGPUs) {
exit(EXIT_WAIVED);
}
if (deviceIds.size() == 0) {
printf(
"cudaNvSci requires one or more GPUs of Pascal(SM 6.0) or higher "
"archs\nWaiving..\n");
exit(EXIT_WAIVED);
else {
for (int devID = 0; devID < numOfGPUs; devID++) {
int major = 0, minor = 0;
checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, devID));
checkCudaErrors(cudaDeviceGetAttribute(&minor, cudaDevAttrComputeCapabilityMinor, devID));
if (major >= 6) {
deviceIds.push_back(devID);
}
}
if (deviceIds.size() == 0) {
printf("cudaNvSci requires one or more GPUs of Pascal(SM 6.0) or higher "
"archs\nWaiving..\n");
exit(EXIT_WAIVED);
}
}
}
std::string image_filename = "teapot1024.ppm";
std::string image_filename = "teapot1024.ppm";
if (checkCmdLineFlag(argc, (const char **)argv, "file")) {
getCmdLineArgumentString(argc, (const char **)argv, "file",
(char **)&image_filename);
}
if (checkCmdLineFlag(argc, (const char **)argv, "file")) {
getCmdLineArgumentString(argc, (const char **)argv, "file", (char **)&image_filename);
}
uint32_t imageWidth = 0;
uint32_t imageHeight = 0;
unsigned char *image_data = NULL;
uint32_t imageWidth = 0;
uint32_t imageHeight = 0;
unsigned char *image_data = NULL;
loadImageData(image_filename, argv, &image_data, imageWidth, imageHeight);
loadImageData(image_filename, argv, &image_data, imageWidth, imageHeight);
cudaNvSci cudaNvSciApp(deviceIds.size() > 1, deviceIds, image_data,
imageWidth, imageHeight);
cudaNvSciApp.runCudaNvSci(image_filename);
cudaNvSci cudaNvSciApp(deviceIds.size() > 1, deviceIds, image_data, imageWidth, imageHeight);
cudaNvSciApp.runCudaNvSci(image_filename);
return EXIT_SUCCESS;
}
return EXIT_SUCCESS;
}