mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-10-25 22:28:17 +08:00
Apply consistent code formatting across the repo. Add clang-format and pre-commit hooks.
This commit is contained in:
Rob Armstrong
@@ -68,106 +68,118 @@
|
||||
// this
|
||||
// struct by putting an undefined symbol in the function body so it won't
|
||||
// compile.
|
||||
template <typename T>
|
||||
struct SharedMemory {
|
||||
// Ensure that we won't compile any un-specialized types
|
||||
__device__ T *getPointer() {
|
||||
extern __device__ void error(void);
|
||||
error();
|
||||
return NULL;
|
||||
}
|
||||
template <typename T> struct SharedMemory
|
||||
{
|
||||
// Ensure that we won't compile any un-specialized types
|
||||
__device__ T *getPointer()
|
||||
{
|
||||
extern __device__ void error(void);
|
||||
error();
|
||||
return NULL;
|
||||
}
|
||||
};
|
||||
|
||||
// Following are the specializations for the following types.
|
||||
// int, uint, char, uchar, short, ushort, long, ulong, bool, float, and double
|
||||
// One could also specialize it for user-defined types.
|
||||
|
||||
template <>
|
||||
struct SharedMemory<int> {
|
||||
__device__ int *getPointer() {
|
||||
extern __shared__ int s_int[];
|
||||
return s_int;
|
||||
}
|
||||
template <> struct SharedMemory<int>
|
||||
{
|
||||
__device__ int *getPointer()
|
||||
{
|
||||
extern __shared__ int s_int[];
|
||||
return s_int;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<unsigned int> {
|
||||
__device__ unsigned int *getPointer() {
|
||||
extern __shared__ unsigned int s_uint[];
|
||||
return s_uint;
|
||||
}
|
||||
template <> struct SharedMemory<unsigned int>
|
||||
{
|
||||
__device__ unsigned int *getPointer()
|
||||
{
|
||||
extern __shared__ unsigned int s_uint[];
|
||||
return s_uint;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<char> {
|
||||
__device__ char *getPointer() {
|
||||
extern __shared__ char s_char[];
|
||||
return s_char;
|
||||
}
|
||||
template <> struct SharedMemory<char>
|
||||
{
|
||||
__device__ char *getPointer()
|
||||
{
|
||||
extern __shared__ char s_char[];
|
||||
return s_char;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<unsigned char> {
|
||||
__device__ unsigned char *getPointer() {
|
||||
extern __shared__ unsigned char s_uchar[];
|
||||
return s_uchar;
|
||||
}
|
||||
template <> struct SharedMemory<unsigned char>
|
||||
{
|
||||
__device__ unsigned char *getPointer()
|
||||
{
|
||||
extern __shared__ unsigned char s_uchar[];
|
||||
return s_uchar;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<short> {
|
||||
__device__ short *getPointer() {
|
||||
extern __shared__ short s_short[];
|
||||
return s_short;
|
||||
}
|
||||
template <> struct SharedMemory<short>
|
||||
{
|
||||
__device__ short *getPointer()
|
||||
{
|
||||
extern __shared__ short s_short[];
|
||||
return s_short;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<unsigned short> {
|
||||
__device__ unsigned short *getPointer() {
|
||||
extern __shared__ unsigned short s_ushort[];
|
||||
return s_ushort;
|
||||
}
|
||||
template <> struct SharedMemory<unsigned short>
|
||||
{
|
||||
__device__ unsigned short *getPointer()
|
||||
{
|
||||
extern __shared__ unsigned short s_ushort[];
|
||||
return s_ushort;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<long> {
|
||||
__device__ long *getPointer() {
|
||||
extern __shared__ long s_long[];
|
||||
return s_long;
|
||||
}
|
||||
template <> struct SharedMemory<long>
|
||||
{
|
||||
__device__ long *getPointer()
|
||||
{
|
||||
extern __shared__ long s_long[];
|
||||
return s_long;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<unsigned long> {
|
||||
__device__ unsigned long *getPointer() {
|
||||
extern __shared__ unsigned long s_ulong[];
|
||||
return s_ulong;
|
||||
}
|
||||
template <> struct SharedMemory<unsigned long>
|
||||
{
|
||||
__device__ unsigned long *getPointer()
|
||||
{
|
||||
extern __shared__ unsigned long s_ulong[];
|
||||
return s_ulong;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<bool> {
|
||||
__device__ bool *getPointer() {
|
||||
extern __shared__ bool s_bool[];
|
||||
return s_bool;
|
||||
}
|
||||
template <> struct SharedMemory<bool>
|
||||
{
|
||||
__device__ bool *getPointer()
|
||||
{
|
||||
extern __shared__ bool s_bool[];
|
||||
return s_bool;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<float> {
|
||||
__device__ float *getPointer() {
|
||||
extern __shared__ float s_float[];
|
||||
return s_float;
|
||||
}
|
||||
template <> struct SharedMemory<float>
|
||||
{
|
||||
__device__ float *getPointer()
|
||||
{
|
||||
extern __shared__ float s_float[];
|
||||
return s_float;
|
||||
}
|
||||
};
|
||||
|
||||
template <>
|
||||
struct SharedMemory<double> {
|
||||
__device__ double *getPointer() {
|
||||
extern __shared__ double s_double[];
|
||||
return s_double;
|
||||
}
|
||||
template <> struct SharedMemory<double>
|
||||
{
|
||||
__device__ double *getPointer()
|
||||
{
|
||||
extern __shared__ double s_double[];
|
||||
return s_double;
|
||||
}
|
||||
};
|
||||
|
||||
#endif //_SHAREDMEM_H_
|
||||
#endif //_SHAREDMEM_H_
|
||||
|
||||
@@ -26,23 +26,23 @@
|
||||
*/
|
||||
|
||||
/* This sample is a templatized version of the template project.
|
||||
* It also shows how to correctly templatize dynamically allocated shared
|
||||
* memory arrays.
|
||||
* Host code.
|
||||
*/
|
||||
* It also shows how to correctly templatize dynamically allocated shared
|
||||
* memory arrays.
|
||||
* Host code.
|
||||
*/
|
||||
|
||||
// System includes
|
||||
#include <stdio.h>
|
||||
#include <assert.h>
|
||||
#include <string.h>
|
||||
#include <math.h>
|
||||
#include <stdio.h>
|
||||
#include <string.h>
|
||||
|
||||
// CUDA runtime
|
||||
#include <cuda_runtime.h>
|
||||
|
||||
// helper functions and utilities to work with CUDA
|
||||
#include <helper_functions.h>
|
||||
#include <helper_cuda.h>
|
||||
#include <helper_functions.h>
|
||||
|
||||
#ifndef MAX
|
||||
#define MAX(a, b) (a > b ? a : b)
|
||||
@@ -58,55 +58,55 @@ int g_TotalFailures = 0;
|
||||
//! @param g_idata input data in global memory
|
||||
//! @param g_odata output data in global memory
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
template <class T>
|
||||
__global__ void testKernel(T *g_idata, T *g_odata) {
|
||||
// Shared mem size is determined by the host app at run time
|
||||
SharedMemory<T> smem;
|
||||
T *sdata = smem.getPointer();
|
||||
template <class T> __global__ void testKernel(T *g_idata, T *g_odata)
|
||||
{
|
||||
// Shared mem size is determined by the host app at run time
|
||||
SharedMemory<T> smem;
|
||||
T *sdata = smem.getPointer();
|
||||
|
||||
// access thread id
|
||||
const unsigned int tid = threadIdx.x;
|
||||
// access number of threads in this block
|
||||
const unsigned int num_threads = blockDim.x;
|
||||
// access thread id
|
||||
const unsigned int tid = threadIdx.x;
|
||||
// access number of threads in this block
|
||||
const unsigned int num_threads = blockDim.x;
|
||||
|
||||
// read in input data from global memory
|
||||
sdata[tid] = g_idata[tid];
|
||||
__syncthreads();
|
||||
// read in input data from global memory
|
||||
sdata[tid] = g_idata[tid];
|
||||
__syncthreads();
|
||||
|
||||
// perform some computations
|
||||
sdata[tid] = (T)num_threads * sdata[tid];
|
||||
__syncthreads();
|
||||
// perform some computations
|
||||
sdata[tid] = (T)num_threads * sdata[tid];
|
||||
__syncthreads();
|
||||
|
||||
// write data to global memory
|
||||
g_odata[tid] = sdata[tid];
|
||||
// write data to global memory
|
||||
g_odata[tid] = sdata[tid];
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// declaration, forward
|
||||
template <class T>
|
||||
void runTest(int argc, char **argv, int len);
|
||||
template <class T> void runTest(int argc, char **argv, int len);
|
||||
|
||||
template <class T>
|
||||
void computeGold(T *reference, T *idata, const unsigned int len) {
|
||||
const T T_len = static_cast<T>(len);
|
||||
template <class T> void computeGold(T *reference, T *idata, const unsigned int len)
|
||||
{
|
||||
const T T_len = static_cast<T>(len);
|
||||
|
||||
for (unsigned int i = 0; i < len; ++i) {
|
||||
reference[i] = idata[i] * T_len;
|
||||
}
|
||||
for (unsigned int i = 0; i < len; ++i) {
|
||||
reference[i] = idata[i] * T_len;
|
||||
}
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// Program main
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
int main(int argc, char **argv) {
|
||||
printf("> runTest<float,32>\n");
|
||||
runTest<float>(argc, argv, 32);
|
||||
printf("> runTest<int,64>\n");
|
||||
runTest<int>(argc, argv, 64);
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
printf("> runTest<float,32>\n");
|
||||
runTest<float>(argc, argv, 32);
|
||||
printf("> runTest<int,64>\n");
|
||||
runTest<int>(argc, argv, 64);
|
||||
|
||||
printf("\n[simpleTemplates] -> Test Results: %d Failures\n", g_TotalFailures);
|
||||
printf("\n[simpleTemplates] -> Test Results: %d Failures\n", g_TotalFailures);
|
||||
|
||||
exit(g_TotalFailures == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
|
||||
exit(g_TotalFailures == 0 ? EXIT_SUCCESS : EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// To completely templatize runTest (below) with cutil, we need to use
|
||||
@@ -114,151 +114,152 @@ int main(int argc, char **argv) {
|
||||
// functions for different types.
|
||||
|
||||
// Here's the generic wrapper for cutCompare*
|
||||
template <class T>
|
||||
class ArrayComparator {
|
||||
public:
|
||||
bool compare(const T *reference, T *data, unsigned int len) {
|
||||
fprintf(stderr,
|
||||
"Error: no comparison function implemented for this type\n");
|
||||
return false;
|
||||
}
|
||||
template <class T> class ArrayComparator
|
||||
{
|
||||
public:
|
||||
bool compare(const T *reference, T *data, unsigned int len)
|
||||
{
|
||||
fprintf(stderr, "Error: no comparison function implemented for this type\n");
|
||||
return false;
|
||||
}
|
||||
};
|
||||
|
||||
// Here's the specialization for ints:
|
||||
template <>
|
||||
class ArrayComparator<int> {
|
||||
public:
|
||||
bool compare(const int *reference, int *data, unsigned int len) {
|
||||
return compareData(reference, data, len, 0.15f, 0.0f);
|
||||
}
|
||||
template <> class ArrayComparator<int>
|
||||
{
|
||||
public:
|
||||
bool compare(const int *reference, int *data, unsigned int len)
|
||||
{
|
||||
return compareData(reference, data, len, 0.15f, 0.0f);
|
||||
}
|
||||
};
|
||||
|
||||
// Here's the specialization for floats:
|
||||
template <>
|
||||
class ArrayComparator<float> {
|
||||
public:
|
||||
bool compare(const float *reference, float *data, unsigned int len) {
|
||||
return compareData(reference, data, len, 0.15f, 0.15f);
|
||||
}
|
||||
template <> class ArrayComparator<float>
|
||||
{
|
||||
public:
|
||||
bool compare(const float *reference, float *data, unsigned int len)
|
||||
{
|
||||
return compareData(reference, data, len, 0.15f, 0.15f);
|
||||
}
|
||||
};
|
||||
|
||||
// Here's the generic wrapper for cutWriteFile*
|
||||
template <class T>
|
||||
class ArrayFileWriter {
|
||||
public:
|
||||
bool write(const char *filename, T *data, unsigned int len, float epsilon) {
|
||||
fprintf(stderr,
|
||||
"Error: no file write function implemented for this type\n");
|
||||
return false;
|
||||
}
|
||||
template <class T> class ArrayFileWriter
|
||||
{
|
||||
public:
|
||||
bool write(const char *filename, T *data, unsigned int len, float epsilon)
|
||||
{
|
||||
fprintf(stderr, "Error: no file write function implemented for this type\n");
|
||||
return false;
|
||||
}
|
||||
};
|
||||
|
||||
// Here's the specialization for ints:
|
||||
template <>
|
||||
class ArrayFileWriter<int> {
|
||||
public:
|
||||
bool write(const char *filename, int *data, unsigned int len, float epsilon) {
|
||||
return sdkWriteFile(filename, data, len, epsilon, false);
|
||||
}
|
||||
template <> class ArrayFileWriter<int>
|
||||
{
|
||||
public:
|
||||
bool write(const char *filename, int *data, unsigned int len, float epsilon)
|
||||
{
|
||||
return sdkWriteFile(filename, data, len, epsilon, false);
|
||||
}
|
||||
};
|
||||
|
||||
// Here's the specialization for floats:
|
||||
template <>
|
||||
class ArrayFileWriter<float> {
|
||||
public:
|
||||
bool write(const char *filename, float *data, unsigned int len,
|
||||
float epsilon) {
|
||||
return sdkWriteFile(filename, data, len, epsilon, false);
|
||||
}
|
||||
template <> class ArrayFileWriter<float>
|
||||
{
|
||||
public:
|
||||
bool write(const char *filename, float *data, unsigned int len, float epsilon)
|
||||
{
|
||||
return sdkWriteFile(filename, data, len, epsilon, false);
|
||||
}
|
||||
};
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
//! Run a simple test for CUDA
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
template <class T>
|
||||
void runTest(int argc, char **argv, int len) {
|
||||
int devID;
|
||||
cudaDeviceProp deviceProps;
|
||||
template <class T> void runTest(int argc, char **argv, int len)
|
||||
{
|
||||
int devID;
|
||||
cudaDeviceProp deviceProps;
|
||||
|
||||
devID = findCudaDevice(argc, (const char **)argv);
|
||||
devID = findCudaDevice(argc, (const char **)argv);
|
||||
|
||||
// get number of SMs on this GPU
|
||||
checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
|
||||
printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name,
|
||||
deviceProps.multiProcessorCount);
|
||||
// get number of SMs on this GPU
|
||||
checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
|
||||
printf("CUDA device [%s] has %d Multi-Processors\n", deviceProps.name, deviceProps.multiProcessorCount);
|
||||
|
||||
// create and start timer
|
||||
StopWatchInterface *timer = NULL;
|
||||
sdkCreateTimer(&timer);
|
||||
// create and start timer
|
||||
StopWatchInterface *timer = NULL;
|
||||
sdkCreateTimer(&timer);
|
||||
|
||||
// start the timer
|
||||
sdkStartTimer(&timer);
|
||||
// start the timer
|
||||
sdkStartTimer(&timer);
|
||||
|
||||
unsigned int num_threads = len;
|
||||
unsigned int mem_size = sizeof(float) * num_threads;
|
||||
unsigned int num_threads = len;
|
||||
unsigned int mem_size = sizeof(float) * num_threads;
|
||||
|
||||
// allocate host memory
|
||||
T *h_idata = (T *)malloc(mem_size);
|
||||
// allocate host memory
|
||||
T *h_idata = (T *)malloc(mem_size);
|
||||
|
||||
// initialize the memory
|
||||
for (unsigned int i = 0; i < num_threads; ++i) {
|
||||
h_idata[i] = (T)i;
|
||||
}
|
||||
// initialize the memory
|
||||
for (unsigned int i = 0; i < num_threads; ++i) {
|
||||
h_idata[i] = (T)i;
|
||||
}
|
||||
|
||||
// allocate device memory
|
||||
T *d_idata;
|
||||
checkCudaErrors(cudaMalloc((void **)&d_idata, mem_size));
|
||||
// copy host memory to device
|
||||
checkCudaErrors(
|
||||
cudaMemcpy(d_idata, h_idata, mem_size, cudaMemcpyHostToDevice));
|
||||
// allocate device memory
|
||||
T *d_idata;
|
||||
checkCudaErrors(cudaMalloc((void **)&d_idata, mem_size));
|
||||
// copy host memory to device
|
||||
checkCudaErrors(cudaMemcpy(d_idata, h_idata, mem_size, cudaMemcpyHostToDevice));
|
||||
|
||||
// allocate device memory for result
|
||||
T *d_odata;
|
||||
checkCudaErrors(cudaMalloc((void **)&d_odata, mem_size));
|
||||
// allocate device memory for result
|
||||
T *d_odata;
|
||||
checkCudaErrors(cudaMalloc((void **)&d_odata, mem_size));
|
||||
|
||||
// setup execution parameters
|
||||
dim3 grid(1, 1, 1);
|
||||
dim3 threads(num_threads, 1, 1);
|
||||
// setup execution parameters
|
||||
dim3 grid(1, 1, 1);
|
||||
dim3 threads(num_threads, 1, 1);
|
||||
|
||||
// execute the kernel
|
||||
testKernel<T><<<grid, threads, mem_size>>>(d_idata, d_odata);
|
||||
// execute the kernel
|
||||
testKernel<T><<<grid, threads, mem_size>>>(d_idata, d_odata);
|
||||
|
||||
// check if kernel execution generated and error
|
||||
getLastCudaError("Kernel execution failed");
|
||||
// check if kernel execution generated and error
|
||||
getLastCudaError("Kernel execution failed");
|
||||
|
||||
// allocate mem for the result on host side
|
||||
T *h_odata = (T *)malloc(mem_size);
|
||||
// copy result from device to host
|
||||
checkCudaErrors(cudaMemcpy(h_odata, d_odata, sizeof(T) * num_threads,
|
||||
cudaMemcpyDeviceToHost));
|
||||
// allocate mem for the result on host side
|
||||
T *h_odata = (T *)malloc(mem_size);
|
||||
// copy result from device to host
|
||||
checkCudaErrors(cudaMemcpy(h_odata, d_odata, sizeof(T) * num_threads, cudaMemcpyDeviceToHost));
|
||||
|
||||
sdkStopTimer(&timer);
|
||||
printf("Processing time: %f (ms)\n", sdkGetTimerValue(&timer));
|
||||
sdkDeleteTimer(&timer);
|
||||
sdkStopTimer(&timer);
|
||||
printf("Processing time: %f (ms)\n", sdkGetTimerValue(&timer));
|
||||
sdkDeleteTimer(&timer);
|
||||
|
||||
// compute reference solution
|
||||
T *reference = (T *)malloc(mem_size);
|
||||
computeGold<T>(reference, h_idata, num_threads);
|
||||
// compute reference solution
|
||||
T *reference = (T *)malloc(mem_size);
|
||||
computeGold<T>(reference, h_idata, num_threads);
|
||||
|
||||
ArrayComparator<T> comparator;
|
||||
ArrayFileWriter<T> writer;
|
||||
ArrayComparator<T> comparator;
|
||||
ArrayFileWriter<T> writer;
|
||||
|
||||
// check result
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
|
||||
// write file for regression test
|
||||
writer.write("./data/regression.dat", h_odata, num_threads, 0.0f);
|
||||
} else {
|
||||
// custom output handling when no regression test running
|
||||
// in this case check if the result is equivalent to the expected solution
|
||||
bool res = comparator.compare(reference, h_odata, num_threads);
|
||||
printf("Compare %s\n\n", (1 == res) ? "OK" : "MISMATCH");
|
||||
g_TotalFailures += (1 != res);
|
||||
}
|
||||
// check result
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
|
||||
// write file for regression test
|
||||
writer.write("./data/regression.dat", h_odata, num_threads, 0.0f);
|
||||
}
|
||||
else {
|
||||
// custom output handling when no regression test running
|
||||
// in this case check if the result is equivalent to the expected solution
|
||||
bool res = comparator.compare(reference, h_odata, num_threads);
|
||||
printf("Compare %s\n\n", (1 == res) ? "OK" : "MISMATCH");
|
||||
g_TotalFailures += (1 != res);
|
||||
}
|
||||
|
||||
// cleanup memory
|
||||
free(h_idata);
|
||||
free(h_odata);
|
||||
free(reference);
|
||||
checkCudaErrors(cudaFree(d_idata));
|
||||
checkCudaErrors(cudaFree(d_odata));
|
||||
// cleanup memory
|
||||
free(h_idata);
|
||||
free(h_odata);
|
||||
free(reference);
|
||||
checkCudaErrors(cudaFree(d_idata));
|
||||
checkCudaErrors(cudaFree(d_odata));
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user