mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-10-26 06:38:16 +08:00
Apply consistent code formatting across the repo. Add clang-format and pre-commit hooks.
This commit is contained in:
Rob Armstrong
@@ -39,196 +39,201 @@
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
|
||||
#include <iostream>
|
||||
#include <iomanip>
|
||||
#include <stdexcept>
|
||||
#include <cuda_runtime.h>
|
||||
#include <helper_timer.h>
|
||||
#include <helper_cuda.h>
|
||||
|
||||
#include <helper_timer.h>
|
||||
#include <iomanip>
|
||||
#include <iostream>
|
||||
#include <math.h>
|
||||
#include <stdexcept>
|
||||
|
||||
#include "../inc/test.h"
|
||||
|
||||
// Forward declarations
|
||||
void showHelp(const int argc, const char **argv);
|
||||
template <typename Real>
|
||||
void runTest(int argc, const char **argv);
|
||||
void showHelp(const int argc, const char **argv);
|
||||
template <typename Real> void runTest(int argc, const char **argv);
|
||||
|
||||
int main(int argc, char **argv) {
|
||||
using std::invalid_argument;
|
||||
using std::string;
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
using std::invalid_argument;
|
||||
using std::string;
|
||||
|
||||
// Open the log file
|
||||
printf("Monte Carlo Estimate Pi (with inline QRNG)\n");
|
||||
printf("==========================================\n\n");
|
||||
// Open the log file
|
||||
printf("Monte Carlo Estimate Pi (with inline QRNG)\n");
|
||||
printf("==========================================\n\n");
|
||||
|
||||
// If help flag is set, display help and exit immediately
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "help")) {
|
||||
printf("Displaying help on console\n");
|
||||
showHelp(argc, (const char **)argv);
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
|
||||
// Check the precision (checked against the device capability later)
|
||||
try {
|
||||
char *value;
|
||||
|
||||
if (getCmdLineArgumentString(argc, (const char **)argv, "precision",
|
||||
&value)) {
|
||||
// Check requested precision is valid
|
||||
string prec(value);
|
||||
|
||||
if (prec.compare("single") == 0 || prec.compare("\"single\"") == 0) {
|
||||
runTest<float>(argc, (const char **)argv);
|
||||
} else if (prec.compare("double") == 0 ||
|
||||
prec.compare("\"double\"") == 0) {
|
||||
runTest<double>(argc, (const char **)argv);
|
||||
} else {
|
||||
printf(
|
||||
"specified precision (%s) is invalid, must be \"single\" or "
|
||||
"\"double\".\n",
|
||||
value);
|
||||
throw invalid_argument("precision");
|
||||
}
|
||||
} else {
|
||||
runTest<float>(argc, (const char **)argv);
|
||||
// If help flag is set, display help and exit immediately
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "help")) {
|
||||
printf("Displaying help on console\n");
|
||||
showHelp(argc, (const char **)argv);
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
} catch (invalid_argument &e) {
|
||||
printf("invalid command line argument (%s)\n", e.what());
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Finish
|
||||
exit(EXIT_SUCCESS);
|
||||
// Check the precision (checked against the device capability later)
|
||||
try {
|
||||
char *value;
|
||||
|
||||
if (getCmdLineArgumentString(argc, (const char **)argv, "precision", &value)) {
|
||||
// Check requested precision is valid
|
||||
string prec(value);
|
||||
|
||||
if (prec.compare("single") == 0 || prec.compare("\"single\"") == 0) {
|
||||
runTest<float>(argc, (const char **)argv);
|
||||
}
|
||||
else if (prec.compare("double") == 0 || prec.compare("\"double\"") == 0) {
|
||||
runTest<double>(argc, (const char **)argv);
|
||||
}
|
||||
else {
|
||||
printf("specified precision (%s) is invalid, must be \"single\" or "
|
||||
"\"double\".\n",
|
||||
value);
|
||||
throw invalid_argument("precision");
|
||||
}
|
||||
}
|
||||
else {
|
||||
runTest<float>(argc, (const char **)argv);
|
||||
}
|
||||
}
|
||||
catch (invalid_argument &e) {
|
||||
printf("invalid command line argument (%s)\n", e.what());
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
|
||||
// Finish
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
|
||||
template <typename Real>
|
||||
void runTest(int argc, const char **argv) {
|
||||
using std::invalid_argument;
|
||||
using std::runtime_error;
|
||||
template <typename Real> void runTest(int argc, const char **argv)
|
||||
{
|
||||
using std::invalid_argument;
|
||||
using std::runtime_error;
|
||||
|
||||
StopWatchInterface *timer = NULL;
|
||||
StopWatchInterface *timer = NULL;
|
||||
|
||||
try {
|
||||
Test<Real> test;
|
||||
int deviceCount = 0;
|
||||
cudaError_t cudaResult = cudaSuccess;
|
||||
try {
|
||||
Test<Real> test;
|
||||
int deviceCount = 0;
|
||||
cudaError_t cudaResult = cudaSuccess;
|
||||
|
||||
// by default specify GPU Device == 0
|
||||
test.device = 0;
|
||||
// by default specify GPU Device == 0
|
||||
test.device = 0;
|
||||
|
||||
// Get number of available devices
|
||||
cudaResult = cudaGetDeviceCount(&deviceCount);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
printf("could not get device count.\n");
|
||||
throw runtime_error("cudaGetDeviceCount");
|
||||
}
|
||||
|
||||
// (default parameters)
|
||||
test.numSims = k_sims_qa;
|
||||
test.threadBlockSize = k_bsize_qa;
|
||||
|
||||
{
|
||||
char *value = 0;
|
||||
|
||||
if (getCmdLineArgumentString(argc, argv, "device", &value)) {
|
||||
test.device = (int)atoi(value);
|
||||
|
||||
if (test.device >= deviceCount) {
|
||||
printf(
|
||||
"invalid target device specified on command line (device %d does "
|
||||
"not exist).\n",
|
||||
test.device);
|
||||
throw invalid_argument("device");
|
||||
}
|
||||
} else {
|
||||
test.device = gpuGetMaxGflopsDeviceId();
|
||||
}
|
||||
|
||||
if (getCmdLineArgumentString(argc, argv, "sims", &value)) {
|
||||
test.numSims = (unsigned int)atoi(value);
|
||||
|
||||
if (test.numSims < k_sims_min || test.numSims > k_sims_max) {
|
||||
printf(
|
||||
"specified number of simulations (%d) is invalid, must be "
|
||||
"between %d and %d.\n",
|
||||
test.numSims, k_sims_min, k_sims_max);
|
||||
throw invalid_argument("sims");
|
||||
}
|
||||
} else {
|
||||
test.numSims = k_sims_def;
|
||||
}
|
||||
|
||||
if (getCmdLineArgumentString(argc, argv, "block-size", &value)) {
|
||||
// Determine max threads per block
|
||||
cudaDeviceProp deviceProperties;
|
||||
cudaResult = cudaGetDeviceProperties(&deviceProperties, test.device);
|
||||
// Get number of available devices
|
||||
cudaResult = cudaGetDeviceCount(&deviceCount);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
printf("cound not get device properties for device %d.\n",
|
||||
test.device);
|
||||
throw runtime_error("cudaGetDeviceProperties");
|
||||
printf("could not get device count.\n");
|
||||
throw runtime_error("cudaGetDeviceCount");
|
||||
}
|
||||
|
||||
// Check requested size is valid
|
||||
test.threadBlockSize = (unsigned int)atoi(value);
|
||||
// (default parameters)
|
||||
test.numSims = k_sims_qa;
|
||||
test.threadBlockSize = k_bsize_qa;
|
||||
|
||||
if (test.threadBlockSize < k_bsize_min ||
|
||||
test.threadBlockSize > static_cast<unsigned int>(
|
||||
deviceProperties.maxThreadsPerBlock)) {
|
||||
printf(
|
||||
"specified block size (%d) is invalid, must be between %d and %d "
|
||||
"for device %d.\n",
|
||||
test.threadBlockSize, k_bsize_min,
|
||||
deviceProperties.maxThreadsPerBlock, test.device);
|
||||
throw invalid_argument("block-size");
|
||||
}
|
||||
{
|
||||
char *value = 0;
|
||||
|
||||
if (test.threadBlockSize & test.threadBlockSize - 1) {
|
||||
printf(
|
||||
"specified block size (%d) is invalid, must be a power of two "
|
||||
"(see reduction function).\n",
|
||||
test.threadBlockSize);
|
||||
throw invalid_argument("block-size");
|
||||
if (getCmdLineArgumentString(argc, argv, "device", &value)) {
|
||||
test.device = (int)atoi(value);
|
||||
|
||||
if (test.device >= deviceCount) {
|
||||
printf("invalid target device specified on command line (device %d does "
|
||||
"not exist).\n",
|
||||
test.device);
|
||||
throw invalid_argument("device");
|
||||
}
|
||||
}
|
||||
else {
|
||||
test.device = gpuGetMaxGflopsDeviceId();
|
||||
}
|
||||
|
||||
if (getCmdLineArgumentString(argc, argv, "sims", &value)) {
|
||||
test.numSims = (unsigned int)atoi(value);
|
||||
|
||||
if (test.numSims < k_sims_min || test.numSims > k_sims_max) {
|
||||
printf("specified number of simulations (%d) is invalid, must be "
|
||||
"between %d and %d.\n",
|
||||
test.numSims,
|
||||
k_sims_min,
|
||||
k_sims_max);
|
||||
throw invalid_argument("sims");
|
||||
}
|
||||
}
|
||||
else {
|
||||
test.numSims = k_sims_def;
|
||||
}
|
||||
|
||||
if (getCmdLineArgumentString(argc, argv, "block-size", &value)) {
|
||||
// Determine max threads per block
|
||||
cudaDeviceProp deviceProperties;
|
||||
cudaResult = cudaGetDeviceProperties(&deviceProperties, test.device);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
printf("cound not get device properties for device %d.\n", test.device);
|
||||
throw runtime_error("cudaGetDeviceProperties");
|
||||
}
|
||||
|
||||
// Check requested size is valid
|
||||
test.threadBlockSize = (unsigned int)atoi(value);
|
||||
|
||||
if (test.threadBlockSize < k_bsize_min
|
||||
|| test.threadBlockSize > static_cast<unsigned int>(deviceProperties.maxThreadsPerBlock)) {
|
||||
printf("specified block size (%d) is invalid, must be between %d and %d "
|
||||
"for device %d.\n",
|
||||
test.threadBlockSize,
|
||||
k_bsize_min,
|
||||
deviceProperties.maxThreadsPerBlock,
|
||||
test.device);
|
||||
throw invalid_argument("block-size");
|
||||
}
|
||||
|
||||
if (test.threadBlockSize & test.threadBlockSize - 1) {
|
||||
printf("specified block size (%d) is invalid, must be a power of two "
|
||||
"(see reduction function).\n",
|
||||
test.threadBlockSize);
|
||||
throw invalid_argument("block-size");
|
||||
}
|
||||
}
|
||||
else {
|
||||
test.threadBlockSize = k_bsize_def;
|
||||
}
|
||||
}
|
||||
} else {
|
||||
test.threadBlockSize = k_bsize_def;
|
||||
}
|
||||
// Execute
|
||||
test();
|
||||
}
|
||||
catch (invalid_argument &e) {
|
||||
printf("invalid command line argument (%s)\n", e.what());
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
catch (runtime_error &e) {
|
||||
printf("runtime error (%s)\n", e.what());
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
// Execute
|
||||
test();
|
||||
} catch (invalid_argument &e) {
|
||||
printf("invalid command line argument (%s)\n", e.what());
|
||||
exit(EXIT_FAILURE);
|
||||
} catch (runtime_error &e) {
|
||||
printf("runtime error (%s)\n", e.what());
|
||||
exit(EXIT_FAILURE);
|
||||
}
|
||||
}
|
||||
|
||||
void showHelp(int argc, const char **argv) {
|
||||
using std::cout;
|
||||
using std::endl;
|
||||
using std::left;
|
||||
using std::setw;
|
||||
void showHelp(int argc, const char **argv)
|
||||
{
|
||||
using std::cout;
|
||||
using std::endl;
|
||||
using std::left;
|
||||
using std::setw;
|
||||
|
||||
if (argc > 0) {
|
||||
cout << endl << argv[0] << endl;
|
||||
}
|
||||
if (argc > 0) {
|
||||
cout << endl << argv[0] << endl;
|
||||
}
|
||||
|
||||
cout << endl << "Syntax:" << endl;
|
||||
cout << left;
|
||||
cout << " " << setw(20) << "--device=<device>"
|
||||
<< "Specify device to use for execution" << endl;
|
||||
cout << " " << setw(20) << "--sims=<N>"
|
||||
<< "Specify number of Monte Carlo simulations" << endl;
|
||||
cout << " " << setw(20) << "--block-size=<N>"
|
||||
<< "Specify number of threads per block" << endl;
|
||||
cout << " " << setw(20) << "--precision=<P>"
|
||||
<< "Specify the precision (\"single\" or \"double\")" << endl;
|
||||
cout << endl;
|
||||
cout << " " << setw(20) << "--noprompt"
|
||||
<< "Skip prompt before exit" << endl;
|
||||
cout << endl;
|
||||
cout << endl << "Syntax:" << endl;
|
||||
cout << left;
|
||||
cout << " " << setw(20) << "--device=<device>"
|
||||
<< "Specify device to use for execution" << endl;
|
||||
cout << " " << setw(20) << "--sims=<N>"
|
||||
<< "Specify number of Monte Carlo simulations" << endl;
|
||||
cout << " " << setw(20) << "--block-size=<N>"
|
||||
<< "Specify number of threads per block" << endl;
|
||||
cout << " " << setw(20) << "--precision=<P>"
|
||||
<< "Specify the precision (\"single\" or \"double\")" << endl;
|
||||
cout << endl;
|
||||
cout << " " << setw(20) << "--noprompt"
|
||||
<< "Skip prompt before exit" << endl;
|
||||
cout << endl;
|
||||
}
|
||||
|
||||
@@ -26,15 +26,15 @@
|
||||
*/
|
||||
|
||||
|
||||
#include "../inc/piestimator.h"
|
||||
|
||||
#include <string>
|
||||
#include <vector>
|
||||
#include <cooperative_groups.h>
|
||||
#include <cuda_runtime.h>
|
||||
#include <numeric>
|
||||
#include <stdexcept>
|
||||
#include <string>
|
||||
#include <typeinfo>
|
||||
#include <cuda_runtime.h>
|
||||
#include <cooperative_groups.h>
|
||||
#include <vector>
|
||||
|
||||
#include "../inc/piestimator.h"
|
||||
|
||||
namespace cg = cooperative_groups;
|
||||
#include <curand.h>
|
||||
@@ -46,346 +46,329 @@ using std::string;
|
||||
using std::vector;
|
||||
|
||||
// Helper templates to support float and double in same code
|
||||
template <typename L, typename R>
|
||||
struct TYPE_IS {
|
||||
static const bool test = false;
|
||||
template <typename L, typename R> struct TYPE_IS
|
||||
{
|
||||
static const bool test = false;
|
||||
};
|
||||
template <typename L>
|
||||
struct TYPE_IS<L, L> {
|
||||
static const bool test = true;
|
||||
template <typename L> struct TYPE_IS<L, L>
|
||||
{
|
||||
static const bool test = true;
|
||||
};
|
||||
template <bool, class L, class R>
|
||||
struct IF {
|
||||
typedef R type;
|
||||
template <bool, class L, class R> struct IF
|
||||
{
|
||||
typedef R type;
|
||||
};
|
||||
template <class L, class R>
|
||||
struct IF<true, L, R> {
|
||||
typedef L type;
|
||||
template <class L, class R> struct IF<true, L, R>
|
||||
{
|
||||
typedef L type;
|
||||
};
|
||||
|
||||
// RNG init kernel
|
||||
template <typename rngState_t, typename rngDirectionVectors_t>
|
||||
__global__ void initRNG(rngState_t *const rngStates,
|
||||
rngDirectionVectors_t *const rngDirections,
|
||||
unsigned int numDrawsPerDirection) {
|
||||
// Determine thread ID
|
||||
unsigned int tid = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
unsigned int step = gridDim.x * blockDim.x;
|
||||
__global__ void
|
||||
initRNG(rngState_t *const rngStates, rngDirectionVectors_t *const rngDirections, unsigned int numDrawsPerDirection)
|
||||
{
|
||||
// Determine thread ID
|
||||
unsigned int tid = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
unsigned int step = gridDim.x * blockDim.x;
|
||||
|
||||
// Determine offset to avoid overlapping sub-sequences
|
||||
unsigned int offset = tid * ((numDrawsPerDirection + step - 1) / step);
|
||||
// Determine offset to avoid overlapping sub-sequences
|
||||
unsigned int offset = tid * ((numDrawsPerDirection + step - 1) / step);
|
||||
|
||||
// Initialise the RNG
|
||||
curand_init(rngDirections[0], offset, &rngStates[tid]);
|
||||
curand_init(rngDirections[1], offset, &rngStates[tid + step]);
|
||||
// Initialise the RNG
|
||||
curand_init(rngDirections[0], offset, &rngStates[tid]);
|
||||
curand_init(rngDirections[1], offset, &rngStates[tid + step]);
|
||||
}
|
||||
|
||||
__device__ unsigned int reduce_sum(unsigned int in, cg::thread_block cta) {
|
||||
extern __shared__ unsigned int sdata[];
|
||||
__device__ unsigned int reduce_sum(unsigned int in, cg::thread_block cta)
|
||||
{
|
||||
extern __shared__ unsigned int sdata[];
|
||||
|
||||
// Perform first level of reduction:
|
||||
// - Write to shared memory
|
||||
unsigned int ltid = threadIdx.x;
|
||||
// Perform first level of reduction:
|
||||
// - Write to shared memory
|
||||
unsigned int ltid = threadIdx.x;
|
||||
|
||||
sdata[ltid] = in;
|
||||
cg::sync(cta);
|
||||
sdata[ltid] = in;
|
||||
cg::sync(cta);
|
||||
|
||||
// Do reduction in shared mem
|
||||
for (unsigned int s = blockDim.x / 2; s > 0; s >>= 1) {
|
||||
if (ltid < s) {
|
||||
sdata[ltid] += sdata[ltid + s];
|
||||
// Do reduction in shared mem
|
||||
for (unsigned int s = blockDim.x / 2; s > 0; s >>= 1) {
|
||||
if (ltid < s) {
|
||||
sdata[ltid] += sdata[ltid + s];
|
||||
}
|
||||
|
||||
cg::sync(cta);
|
||||
}
|
||||
|
||||
cg::sync(cta);
|
||||
}
|
||||
|
||||
return sdata[0];
|
||||
return sdata[0];
|
||||
}
|
||||
|
||||
__device__ inline void getPoint(float &x, float &y, curandStateSobol32 &state1,
|
||||
curandStateSobol32 &state2) {
|
||||
x = curand_uniform(&state1);
|
||||
y = curand_uniform(&state2);
|
||||
__device__ inline void getPoint(float &x, float &y, curandStateSobol32 &state1, curandStateSobol32 &state2)
|
||||
{
|
||||
x = curand_uniform(&state1);
|
||||
y = curand_uniform(&state2);
|
||||
}
|
||||
__device__ inline void getPoint(double &x, double &y,
|
||||
curandStateSobol64 &state1,
|
||||
curandStateSobol64 &state2) {
|
||||
x = curand_uniform_double(&state1);
|
||||
y = curand_uniform_double(&state2);
|
||||
__device__ inline void getPoint(double &x, double &y, curandStateSobol64 &state1, curandStateSobol64 &state2)
|
||||
{
|
||||
x = curand_uniform_double(&state1);
|
||||
y = curand_uniform_double(&state2);
|
||||
}
|
||||
|
||||
// Estimator kernel
|
||||
template <typename Real, typename rngState_t>
|
||||
__global__ void computeValue(unsigned int *const results,
|
||||
rngState_t *const rngStates,
|
||||
const unsigned int numSims) {
|
||||
// Handle to thread block group
|
||||
cg::thread_block cta = cg::this_thread_block();
|
||||
// Determine thread ID
|
||||
unsigned int bid = blockIdx.x;
|
||||
unsigned int tid = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
unsigned int step = gridDim.x * blockDim.x;
|
||||
__global__ void computeValue(unsigned int *const results, rngState_t *const rngStates, const unsigned int numSims)
|
||||
{
|
||||
// Handle to thread block group
|
||||
cg::thread_block cta = cg::this_thread_block();
|
||||
// Determine thread ID
|
||||
unsigned int bid = blockIdx.x;
|
||||
unsigned int tid = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
unsigned int step = gridDim.x * blockDim.x;
|
||||
|
||||
// Initialise the RNG
|
||||
rngState_t localState1 = rngStates[tid];
|
||||
rngState_t localState2 = rngStates[tid + step];
|
||||
// Initialise the RNG
|
||||
rngState_t localState1 = rngStates[tid];
|
||||
rngState_t localState2 = rngStates[tid + step];
|
||||
|
||||
// Count the number of points which lie inside the unit quarter-circle
|
||||
unsigned int pointsInside = 0;
|
||||
// Count the number of points which lie inside the unit quarter-circle
|
||||
unsigned int pointsInside = 0;
|
||||
|
||||
for (unsigned int i = tid; i < numSims; i += step) {
|
||||
Real x;
|
||||
Real y;
|
||||
getPoint(x, y, localState1, localState2);
|
||||
Real l2norm2 = x * x + y * y;
|
||||
for (unsigned int i = tid; i < numSims; i += step) {
|
||||
Real x;
|
||||
Real y;
|
||||
getPoint(x, y, localState1, localState2);
|
||||
Real l2norm2 = x * x + y * y;
|
||||
|
||||
if (l2norm2 < static_cast<Real>(1)) {
|
||||
pointsInside++;
|
||||
if (l2norm2 < static_cast<Real>(1)) {
|
||||
pointsInside++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Reduce within the block
|
||||
pointsInside = reduce_sum(pointsInside, cta);
|
||||
// Reduce within the block
|
||||
pointsInside = reduce_sum(pointsInside, cta);
|
||||
|
||||
// Store the result
|
||||
if (threadIdx.x == 0) {
|
||||
results[bid] = pointsInside;
|
||||
}
|
||||
// Store the result
|
||||
if (threadIdx.x == 0) {
|
||||
results[bid] = pointsInside;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename Real>
|
||||
PiEstimator<Real>::PiEstimator(unsigned int numSims, unsigned int device,
|
||||
unsigned int threadBlockSize)
|
||||
: m_numSims(numSims),
|
||||
m_device(device),
|
||||
m_threadBlockSize(threadBlockSize) {}
|
||||
PiEstimator<Real>::PiEstimator(unsigned int numSims, unsigned int device, unsigned int threadBlockSize)
|
||||
: m_numSims(numSims)
|
||||
, m_device(device)
|
||||
, m_threadBlockSize(threadBlockSize)
|
||||
{
|
||||
}
|
||||
|
||||
template <typename Real>
|
||||
Real PiEstimator<Real>::operator()() {
|
||||
cudaError_t cudaResult = cudaSuccess;
|
||||
struct cudaDeviceProp deviceProperties;
|
||||
struct cudaFuncAttributes funcAttributes;
|
||||
template <typename Real> Real PiEstimator<Real>::operator()()
|
||||
{
|
||||
cudaError_t cudaResult = cudaSuccess;
|
||||
struct cudaDeviceProp deviceProperties;
|
||||
struct cudaFuncAttributes funcAttributes;
|
||||
|
||||
// Determine type of generator to use (32- or 64-bit)
|
||||
typedef typename IF<TYPE_IS<Real, double>::test, curandStateSobol64_t,
|
||||
curandStateSobol32_t>::type curandStateSobol_sz;
|
||||
typedef
|
||||
typename IF<TYPE_IS<Real, double>::test, curandDirectionVectors64_t,
|
||||
curandDirectionVectors32_t>::type curandDirectionVectors_sz;
|
||||
// Determine type of generator to use (32- or 64-bit)
|
||||
typedef
|
||||
typename IF<TYPE_IS<Real, double>::test, curandStateSobol64_t, curandStateSobol32_t>::type curandStateSobol_sz;
|
||||
typedef typename IF<TYPE_IS<Real, double>::test, curandDirectionVectors64_t, curandDirectionVectors32_t>::type
|
||||
curandDirectionVectors_sz;
|
||||
|
||||
// Get device properties
|
||||
cudaResult = cudaGetDeviceProperties(&deviceProperties, m_device);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not get device properties: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Check precision is valid
|
||||
if (typeid(Real) == typeid(double) &&
|
||||
(deviceProperties.major < 1 ||
|
||||
(deviceProperties.major == 1 && deviceProperties.minor < 3))) {
|
||||
throw std::runtime_error("Device does not have double precision support");
|
||||
}
|
||||
|
||||
// Attach to GPU
|
||||
cudaResult = cudaSetDevice(m_device);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not set CUDA device: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Determine how to divide the work between cores
|
||||
dim3 block;
|
||||
dim3 grid;
|
||||
block.x = m_threadBlockSize;
|
||||
grid.x = (m_numSims + m_threadBlockSize - 1) / m_threadBlockSize;
|
||||
|
||||
// Aim to launch around ten or more times as many blocks as there
|
||||
// are multiprocessors on the target device.
|
||||
unsigned int blocksPerSM = 10;
|
||||
unsigned int numSMs = deviceProperties.multiProcessorCount;
|
||||
|
||||
while (grid.x > 2 * blocksPerSM * numSMs) {
|
||||
grid.x >>= 1;
|
||||
}
|
||||
|
||||
// Get initRNG function properties and check the maximum block size
|
||||
cudaResult = cudaFuncGetAttributes(
|
||||
&funcAttributes, initRNG<curandStateSobol_sz, curandDirectionVectors_sz>);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not get function attributes: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
if (block.x > (unsigned int)funcAttributes.maxThreadsPerBlock) {
|
||||
throw std::runtime_error(
|
||||
"Block X dimension is too large for initRNG kernel");
|
||||
}
|
||||
|
||||
// Get computeValue function properties and check the maximum block size
|
||||
cudaResult = cudaFuncGetAttributes(&funcAttributes,
|
||||
computeValue<Real, curandStateSobol_sz>);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not get function attributes: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
if (block.x > (unsigned int)funcAttributes.maxThreadsPerBlock) {
|
||||
throw std::runtime_error(
|
||||
"Block X dimension is too large for computeValue kernel");
|
||||
}
|
||||
|
||||
// Check the dimensions are valid
|
||||
if (block.x > (unsigned int)deviceProperties.maxThreadsDim[0]) {
|
||||
throw std::runtime_error("Block X dimension is too large for device");
|
||||
}
|
||||
|
||||
if (grid.x > (unsigned int)deviceProperties.maxGridSize[0]) {
|
||||
throw std::runtime_error("Grid X dimension is too large for device");
|
||||
}
|
||||
|
||||
// Allocate memory for RNG states and direction vectors
|
||||
curandStateSobol_sz *d_rngStates = 0;
|
||||
curandDirectionVectors_sz *d_rngDirections = 0;
|
||||
cudaResult = cudaMalloc((void **)&d_rngStates,
|
||||
2 * grid.x * block.x * sizeof(curandStateSobol_sz));
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not allocate memory on device for RNG states: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
cudaResult = cudaMalloc((void **)&d_rngDirections,
|
||||
2 * sizeof(curandDirectionVectors_sz));
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg(
|
||||
"Could not allocate memory on device for RNG direction vectors: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Allocate memory for result
|
||||
// Each thread block will produce one result
|
||||
unsigned int *d_results = 0;
|
||||
cudaResult = cudaMalloc((void **)&d_results, grid.x * sizeof(unsigned int));
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not allocate memory on device for partial results: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Generate direction vectors on the host and copy to the device
|
||||
if (typeid(Real) == typeid(float)) {
|
||||
curandDirectionVectors32_t *rngDirections;
|
||||
curandStatus_t curandResult = curandGetDirectionVectors32(
|
||||
&rngDirections, CURAND_DIRECTION_VECTORS_32_JOEKUO6);
|
||||
|
||||
if (curandResult != CURAND_STATUS_SUCCESS) {
|
||||
string msg(
|
||||
"Could not get direction vectors for quasi-random number "
|
||||
"generator: ");
|
||||
msg += curandResult;
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
cudaResult = cudaMemcpy(d_rngDirections, rngDirections,
|
||||
2 * sizeof(curandDirectionVectors32_t),
|
||||
cudaMemcpyHostToDevice);
|
||||
// Get device properties
|
||||
cudaResult = cudaGetDeviceProperties(&deviceProperties, m_device);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not copy direction vectors to device: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
} else if (typeid(Real) == typeid(double)) {
|
||||
curandDirectionVectors64_t *rngDirections;
|
||||
curandStatus_t curandResult = curandGetDirectionVectors64(
|
||||
&rngDirections, CURAND_DIRECTION_VECTORS_64_JOEKUO6);
|
||||
|
||||
if (curandResult != CURAND_STATUS_SUCCESS) {
|
||||
string msg(
|
||||
"Could not get direction vectors for quasi-random number "
|
||||
"generator: ");
|
||||
msg += curandResult;
|
||||
throw std::runtime_error(msg);
|
||||
string msg("Could not get device properties: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
cudaResult = cudaMemcpy(d_rngDirections, rngDirections,
|
||||
2 * sizeof(curandDirectionVectors64_t),
|
||||
cudaMemcpyHostToDevice);
|
||||
// Check precision is valid
|
||||
if (typeid(Real) == typeid(double)
|
||||
&& (deviceProperties.major < 1 || (deviceProperties.major == 1 && deviceProperties.minor < 3))) {
|
||||
throw std::runtime_error("Device does not have double precision support");
|
||||
}
|
||||
|
||||
// Attach to GPU
|
||||
cudaResult = cudaSetDevice(m_device);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not copy direction vectors to device: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
string msg("Could not set CUDA device: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
} else {
|
||||
string msg(
|
||||
"Could not get direction vectors for random number generator of "
|
||||
"specified type");
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Initialise RNG
|
||||
initRNG<<<grid, block>>>(d_rngStates, d_rngDirections, m_numSims);
|
||||
// Determine how to divide the work between cores
|
||||
dim3 block;
|
||||
dim3 grid;
|
||||
block.x = m_threadBlockSize;
|
||||
grid.x = (m_numSims + m_threadBlockSize - 1) / m_threadBlockSize;
|
||||
|
||||
// Count the points inside unit quarter-circle
|
||||
computeValue<Real><<<grid, block, block.x * sizeof(unsigned int)>>>(
|
||||
d_results, d_rngStates, m_numSims);
|
||||
// Aim to launch around ten or more times as many blocks as there
|
||||
// are multiprocessors on the target device.
|
||||
unsigned int blocksPerSM = 10;
|
||||
unsigned int numSMs = deviceProperties.multiProcessorCount;
|
||||
|
||||
// Copy partial results back
|
||||
vector<unsigned int> results(grid.x);
|
||||
cudaResult = cudaMemcpy(&results[0], d_results, grid.x * sizeof(unsigned int),
|
||||
cudaMemcpyDeviceToHost);
|
||||
while (grid.x > 2 * blocksPerSM * numSMs) {
|
||||
grid.x >>= 1;
|
||||
}
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not copy partial results to host: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
// Get initRNG function properties and check the maximum block size
|
||||
cudaResult = cudaFuncGetAttributes(&funcAttributes, initRNG<curandStateSobol_sz, curandDirectionVectors_sz>);
|
||||
|
||||
// Complete sum-reduction on host
|
||||
Real value =
|
||||
static_cast<Real>(std::accumulate(results.begin(), results.end(), 0));
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not get function attributes: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Determine the proportion of points inside the quarter-circle,
|
||||
// i.e. the area of the unit quarter-circle
|
||||
value /= m_numSims;
|
||||
if (block.x > (unsigned int)funcAttributes.maxThreadsPerBlock) {
|
||||
throw std::runtime_error("Block X dimension is too large for initRNG kernel");
|
||||
}
|
||||
|
||||
// Value is currently an estimate of the area of a unit quarter-circle, so we
|
||||
// can scale to a full circle by multiplying by four. Now since the area of a
|
||||
// circle is pi * r^2, and r is one, the value will be an estimate for the
|
||||
// value of pi.
|
||||
value *= 4;
|
||||
// Get computeValue function properties and check the maximum block size
|
||||
cudaResult = cudaFuncGetAttributes(&funcAttributes, computeValue<Real, curandStateSobol_sz>);
|
||||
|
||||
// Cleanup
|
||||
if (d_rngStates) {
|
||||
cudaFree(d_rngStates);
|
||||
d_rngStates = 0;
|
||||
}
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not get function attributes: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
if (d_rngDirections) {
|
||||
cudaFree(d_rngDirections);
|
||||
d_rngDirections = 0;
|
||||
}
|
||||
if (block.x > (unsigned int)funcAttributes.maxThreadsPerBlock) {
|
||||
throw std::runtime_error("Block X dimension is too large for computeValue kernel");
|
||||
}
|
||||
|
||||
if (d_results) {
|
||||
cudaFree(d_results);
|
||||
d_results = 0;
|
||||
}
|
||||
// Check the dimensions are valid
|
||||
if (block.x > (unsigned int)deviceProperties.maxThreadsDim[0]) {
|
||||
throw std::runtime_error("Block X dimension is too large for device");
|
||||
}
|
||||
|
||||
return value;
|
||||
if (grid.x > (unsigned int)deviceProperties.maxGridSize[0]) {
|
||||
throw std::runtime_error("Grid X dimension is too large for device");
|
||||
}
|
||||
|
||||
// Allocate memory for RNG states and direction vectors
|
||||
curandStateSobol_sz *d_rngStates = 0;
|
||||
curandDirectionVectors_sz *d_rngDirections = 0;
|
||||
cudaResult = cudaMalloc((void **)&d_rngStates, 2 * grid.x * block.x * sizeof(curandStateSobol_sz));
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not allocate memory on device for RNG states: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
cudaResult = cudaMalloc((void **)&d_rngDirections, 2 * sizeof(curandDirectionVectors_sz));
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not allocate memory on device for RNG direction vectors: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Allocate memory for result
|
||||
// Each thread block will produce one result
|
||||
unsigned int *d_results = 0;
|
||||
cudaResult = cudaMalloc((void **)&d_results, grid.x * sizeof(unsigned int));
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not allocate memory on device for partial results: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Generate direction vectors on the host and copy to the device
|
||||
if (typeid(Real) == typeid(float)) {
|
||||
curandDirectionVectors32_t *rngDirections;
|
||||
curandStatus_t curandResult = curandGetDirectionVectors32(&rngDirections, CURAND_DIRECTION_VECTORS_32_JOEKUO6);
|
||||
|
||||
if (curandResult != CURAND_STATUS_SUCCESS) {
|
||||
string msg("Could not get direction vectors for quasi-random number "
|
||||
"generator: ");
|
||||
msg += curandResult;
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
cudaResult =
|
||||
cudaMemcpy(d_rngDirections, rngDirections, 2 * sizeof(curandDirectionVectors32_t), cudaMemcpyHostToDevice);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not copy direction vectors to device: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
}
|
||||
else if (typeid(Real) == typeid(double)) {
|
||||
curandDirectionVectors64_t *rngDirections;
|
||||
curandStatus_t curandResult = curandGetDirectionVectors64(&rngDirections, CURAND_DIRECTION_VECTORS_64_JOEKUO6);
|
||||
|
||||
if (curandResult != CURAND_STATUS_SUCCESS) {
|
||||
string msg("Could not get direction vectors for quasi-random number "
|
||||
"generator: ");
|
||||
msg += curandResult;
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
cudaResult =
|
||||
cudaMemcpy(d_rngDirections, rngDirections, 2 * sizeof(curandDirectionVectors64_t), cudaMemcpyHostToDevice);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not copy direction vectors to device: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
}
|
||||
else {
|
||||
string msg("Could not get direction vectors for random number generator of "
|
||||
"specified type");
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Initialise RNG
|
||||
initRNG<<<grid, block>>>(d_rngStates, d_rngDirections, m_numSims);
|
||||
|
||||
// Count the points inside unit quarter-circle
|
||||
computeValue<Real><<<grid, block, block.x * sizeof(unsigned int)>>>(d_results, d_rngStates, m_numSims);
|
||||
|
||||
// Copy partial results back
|
||||
vector<unsigned int> results(grid.x);
|
||||
cudaResult = cudaMemcpy(&results[0], d_results, grid.x * sizeof(unsigned int), cudaMemcpyDeviceToHost);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
string msg("Could not copy partial results to host: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Complete sum-reduction on host
|
||||
Real value = static_cast<Real>(std::accumulate(results.begin(), results.end(), 0));
|
||||
|
||||
// Determine the proportion of points inside the quarter-circle,
|
||||
// i.e. the area of the unit quarter-circle
|
||||
value /= m_numSims;
|
||||
|
||||
// Value is currently an estimate of the area of a unit quarter-circle, so we
|
||||
// can scale to a full circle by multiplying by four. Now since the area of a
|
||||
// circle is pi * r^2, and r is one, the value will be an estimate for the
|
||||
// value of pi.
|
||||
value *= 4;
|
||||
|
||||
// Cleanup
|
||||
if (d_rngStates) {
|
||||
cudaFree(d_rngStates);
|
||||
d_rngStates = 0;
|
||||
}
|
||||
|
||||
if (d_rngDirections) {
|
||||
cudaFree(d_rngDirections);
|
||||
d_rngDirections = 0;
|
||||
}
|
||||
|
||||
if (d_results) {
|
||||
cudaFree(d_results);
|
||||
d_results = 0;
|
||||
}
|
||||
|
||||
return value;
|
||||
}
|
||||
|
||||
// Explicit template instantiation
|
||||
|
||||
@@ -28,84 +28,84 @@
|
||||
|
||||
#include "../inc/test.h"
|
||||
|
||||
#include <sstream>
|
||||
#include <iomanip>
|
||||
#include <stdexcept>
|
||||
#include <memory>
|
||||
#include <iostream>
|
||||
#include <cassert>
|
||||
#include <typeinfo>
|
||||
#include <stdio.h>
|
||||
#include <helper_timer.h>
|
||||
#include <cuda_runtime.h>
|
||||
|
||||
#include <helper_timer.h>
|
||||
#include <iomanip>
|
||||
#include <iostream>
|
||||
#include <math.h>
|
||||
#include <memory>
|
||||
#include <sstream>
|
||||
#include <stdexcept>
|
||||
#include <stdio.h>
|
||||
#include <typeinfo>
|
||||
|
||||
#include "../inc/piestimator.h"
|
||||
|
||||
template <typename Real>
|
||||
bool Test<Real>::operator()() {
|
||||
using std::endl;
|
||||
using std::setw;
|
||||
using std::stringstream;
|
||||
template <typename Real> bool Test<Real>::operator()()
|
||||
{
|
||||
using std::endl;
|
||||
using std::setw;
|
||||
using std::stringstream;
|
||||
|
||||
StopWatchInterface *timer = NULL;
|
||||
sdkCreateTimer(&timer);
|
||||
StopWatchInterface *timer = NULL;
|
||||
sdkCreateTimer(&timer);
|
||||
|
||||
// Get device properties
|
||||
struct cudaDeviceProp deviceProperties;
|
||||
cudaError_t cudaResult = cudaGetDeviceProperties(&deviceProperties, device);
|
||||
// Get device properties
|
||||
struct cudaDeviceProp deviceProperties;
|
||||
cudaError_t cudaResult = cudaGetDeviceProperties(&deviceProperties, device);
|
||||
|
||||
if (cudaResult != cudaSuccess) {
|
||||
std::string msg("Could not get device properties: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
if (cudaResult != cudaSuccess) {
|
||||
std::string msg("Could not get device properties: ");
|
||||
msg += cudaGetErrorString(cudaResult);
|
||||
throw std::runtime_error(msg);
|
||||
}
|
||||
|
||||
// Evaluate on GPU
|
||||
printf("Estimating Pi on GPU (%s)\n\n", deviceProperties.name);
|
||||
PiEstimator<Real> estimator(numSims, device, threadBlockSize);
|
||||
sdkStartTimer(&timer);
|
||||
Real result = estimator();
|
||||
sdkStopTimer(&timer);
|
||||
elapsedTime = sdkGetAverageTimerValue(&timer) / 1000.0f;
|
||||
// Evaluate on GPU
|
||||
printf("Estimating Pi on GPU (%s)\n\n", deviceProperties.name);
|
||||
PiEstimator<Real> estimator(numSims, device, threadBlockSize);
|
||||
sdkStartTimer(&timer);
|
||||
Real result = estimator();
|
||||
sdkStopTimer(&timer);
|
||||
elapsedTime = sdkGetAverageTimerValue(&timer) / 1000.0f;
|
||||
|
||||
// Tolerance to compare result with expected
|
||||
// This is just to check that nothing has gone very wrong with the
|
||||
// test, the actual accuracy of the result depends on the number of
|
||||
// Monte Carlo trials
|
||||
const Real tolerance = static_cast<Real>(0.01);
|
||||
// Tolerance to compare result with expected
|
||||
// This is just to check that nothing has gone very wrong with the
|
||||
// test, the actual accuracy of the result depends on the number of
|
||||
// Monte Carlo trials
|
||||
const Real tolerance = static_cast<Real>(0.01);
|
||||
|
||||
// Display results
|
||||
Real abserror = fabs(result - static_cast<Real>(PI));
|
||||
Real relerror = abserror / static_cast<Real>(PI);
|
||||
printf("Precision: %s\n",
|
||||
(typeid(Real) == typeid(double)) ? "double" : "single");
|
||||
printf("Number of sims: %d\n", numSims);
|
||||
printf("Tolerance: %e\n", tolerance);
|
||||
printf("GPU result: %e\n", result);
|
||||
printf("Expected: %e\n", PI);
|
||||
printf("Absolute error: %e\n", abserror);
|
||||
printf("Relative error: %e\n\n", relerror);
|
||||
// Display results
|
||||
Real abserror = fabs(result - static_cast<Real>(PI));
|
||||
Real relerror = abserror / static_cast<Real>(PI);
|
||||
printf("Precision: %s\n", (typeid(Real) == typeid(double)) ? "double" : "single");
|
||||
printf("Number of sims: %d\n", numSims);
|
||||
printf("Tolerance: %e\n", tolerance);
|
||||
printf("GPU result: %e\n", result);
|
||||
printf("Expected: %e\n", PI);
|
||||
printf("Absolute error: %e\n", abserror);
|
||||
printf("Relative error: %e\n\n", relerror);
|
||||
|
||||
// Check result
|
||||
if (relerror > tolerance) {
|
||||
printf("computed result (%e) does not match expected result (%e).\n",
|
||||
result, PI);
|
||||
pass = false;
|
||||
} else {
|
||||
pass = true;
|
||||
}
|
||||
// Check result
|
||||
if (relerror > tolerance) {
|
||||
printf("computed result (%e) does not match expected result (%e).\n", result, PI);
|
||||
pass = false;
|
||||
}
|
||||
else {
|
||||
pass = true;
|
||||
}
|
||||
|
||||
// Print results
|
||||
printf(
|
||||
"MonteCarloEstimatePiInlineQ, Performance = %.2f sims/s, Time = "
|
||||
"%.2f(ms), NumDevsUsed = %u, Blocksize = %u\n",
|
||||
numSims / elapsedTime, elapsedTime * 1000.0f, 1, threadBlockSize);
|
||||
// Print results
|
||||
printf("MonteCarloEstimatePiInlineQ, Performance = %.2f sims/s, Time = "
|
||||
"%.2f(ms), NumDevsUsed = %u, Blocksize = %u\n",
|
||||
numSims / elapsedTime,
|
||||
elapsedTime * 1000.0f,
|
||||
1,
|
||||
threadBlockSize);
|
||||
|
||||
sdkDeleteTimer(&timer);
|
||||
sdkDeleteTimer(&timer);
|
||||
|
||||
return pass;
|
||||
return pass;
|
||||
}
|
||||
|
||||
// Explicit template instantiation
|
||||
|
||||
Reference in New Issue
Block a user