////////////////////////////////////////////////////////////////////////////////
// Main program
////////////////////////////////////////////////////////////////////////////////
int main(int argc, char **argv){
const unsigned int OPT_N_MAX = 512;
unsigned int useDoublePrecision;
printf("[binomialOptions]\n");
int devID = cutilDeviceInit(argc, argv);
if (devID < 0) {
printf("exiting...\n");
cutilExit(argc, argv);
exit(0);
}
cutilSafeCall(cudaGetDevice(&devID));
cudaDeviceProp deviceProp;
cutilSafeCall(cudaGetDeviceProperties(&deviceProp, devID));
char *precisionChoice;
cutGetCmdLineArgumentstr(argc, (const char **)argv, "type", &precisionChoice);
if(precisionChoice == NULL) {
useDoublePrecision = 0;
} else {
if(!strcasecmp(precisionChoice, "double"))
useDoublePrecision = 1;
else
useDoublePrecision = 0;
}
printf(useDoublePrecision ? "Using double precision...\n" : "Using single precision...\n");
const int OPT_N = deviceEmulation() ? 1 : OPT_N_MAX;
TOptionData optionData[OPT_N_MAX];
float
callValueBS[OPT_N_MAX],
callValueGPU[OPT_N_MAX],
callValueCPU[OPT_N_MAX];
double
sumDelta, sumRef, gpuTime, errorVal;
unsigned int hTimer;
int i;
cutilCheckError( cutCreateTimer(&hTimer) );
int version = deviceProp.major * 10 + deviceProp.minor;
if(useDoublePrecision && version < 13){
printf("Double precision is not supported.\n");
return 0;
}
printf("Generating input data...\n");
//Generate options set
srand(123);
for(i = 0; i < OPT_N; i++){
optionData[i].S = randData(5.0f, 30.0f);
optionData[i].X = randData(1.0f, 100.0f);
optionData[i].T = randData(0.25f, 10.0f);
optionData[i].R = 0.06f;
optionData[i].V = 0.10f;
BlackScholesCall(callValueBS[i], optionData[i]);
}
printf("Running GPU binomial tree...\n");
cutilSafeCall( cudaThreadSynchronize() );
cutilCheckError( cutResetTimer(hTimer) );
cutilCheckError( cutStartTimer(hTimer) );
if(useDoublePrecision)
binomialOptions_SM13(callValueGPU, optionData, OPT_N);
else
binomialOptions_SM10(callValueGPU, optionData, OPT_N);
cutilSafeCall( cudaThreadSynchronize() );
cutilCheckError( cutStopTimer(hTimer) );
gpuTime = cutGetTimerValue(hTimer);
printf("Options count : %i \n", OPT_N);
printf("Time steps : %i \n", NUM_STEPS);
printf("binomialOptionsGPU() time: %f msec\n", gpuTime);
printf("Options per second : %f \n", OPT_N / (gpuTime * 0.001));
printf("Running CPU binomial tree...\n");
for(i = 0; i < OPT_N; i++)
binomialOptionsCPU(callValueCPU[i], optionData[i]);
printf("Comparing the results...\n");
sumDelta = 0;
sumRef = 0;
printf("GPU binomial vs. Black-Scholes\n");
for(i = 0; i < OPT_N; i++){
sumDelta += fabs(callValueBS[i] - callValueGPU[i]);
sumRef += fabs(callValueBS[i]);
}
if(sumRef >1E-5)
printf("L1 norm: %E\n", sumDelta / sumRef);
else
printf("Avg. diff: %E\n", sumDelta / (double)OPT_N);
printf("CPU binomial vs. Black-Scholes\n");
sumDelta = 0;
sumRef = 0;
for(i = 0; i < OPT_N; i++){
sumDelta += fabs(callValueBS[i]- callValueCPU[i]);
sumRef += fabs(callValueBS[i]);
}
if(sumRef >1E-5)
printf("L1 norm: %E\n", sumDelta / sumRef);
else
printf("Avg. diff: %E\n", sumDelta / (double)OPT_N);
printf("CPU binomial vs. GPU binomial\n");
sumDelta = 0;
sumRef = 0;
for(i = 0; i < OPT_N; i++){
sumDelta += fabs(callValueGPU[i] - callValueCPU[i]);
sumRef += callValueCPU[i];
}
if(sumRef > 1E-5)
printf("L1 norm: %E\n", errorVal = sumDelta / sumRef);
else
printf("Avg. diff: %E\n", errorVal = sumDelta / (double)OPT_N);
printf("Shutting down...\n");
printf("\n[binomialOptions] - Test Summary:\n");
printf((errorVal < 5e-4) ? "PASSED\n" : "FAILED\n");
cutilCheckError( cutDeleteTimer(hTimer) );
cudaThreadExit();
cutilExit(argc, argv);
}
////////////////////////////////////////////////////////////////////////////////
// Main program
////////////////////////////////////////////////////////////////////////////////
int main(int argc, char **argv)
{
printf("[%s] - Starting...\n", argv[0]);
cudaDeviceProp deviceProp;
int devID = findCudaDevice(argc, (const char **)argv);
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, devID));
if (((deviceProp.major << 4) + deviceProp.minor) < 0x20)
{
fprintf(stderr, "binomialOptions requires Compute Capability of SM 2.0 or higher to run.\n");
cudaDeviceReset();
exit(EXIT_WAIVED);
}
const int OPT_N = MAX_OPTIONS;
TOptionData optionData[MAX_OPTIONS];
real
callValueBS[MAX_OPTIONS],
callValueGPU[MAX_OPTIONS],
callValueCPU[MAX_OPTIONS];
real
sumDelta, sumRef, gpuTime, errorVal;
StopWatchInterface *hTimer = NULL;
int i;
sdkCreateTimer(&hTimer);
printf("Generating input data...\n");
//Generate options set
srand(123);
for (i = 0; i < OPT_N; i++)
{
optionData[i].S = randData(5.0f, 30.0f);
optionData[i].X = randData(1.0f, 100.0f);
optionData[i].T = randData(0.25f, 10.0f);
optionData[i].R = 0.06f;
optionData[i].V = 0.10f;
BlackScholesCall(callValueBS[i], optionData[i]);
}
printf("Running GPU binomial tree...\n");
checkCudaErrors(cudaDeviceSynchronize());
sdkResetTimer(&hTimer);
sdkStartTimer(&hTimer);
binomialOptionsGPU(callValueGPU, optionData, OPT_N);
checkCudaErrors(cudaDeviceSynchronize());
sdkStopTimer(&hTimer);
gpuTime = sdkGetTimerValue(&hTimer);
printf("Options count : %i \n", OPT_N);
printf("Time steps : %i \n", NUM_STEPS);
printf("binomialOptionsGPU() time: %f msec\n", gpuTime);
printf("Options per second : %f \n", OPT_N / (gpuTime * 0.001));
printf("Running CPU binomial tree...\n");
for (i = 0; i < OPT_N; i++)
{
binomialOptionsCPU(callValueCPU[i], optionData[i]);
}
printf("Comparing the results...\n");
sumDelta = 0;
sumRef = 0;
printf("GPU binomial vs. Black-Scholes\n");
for (i = 0; i < OPT_N; i++)
{
sumDelta += fabs(callValueBS[i] - callValueGPU[i]);
sumRef += fabs(callValueBS[i]);
}
if (sumRef >1E-5)
{
printf("L1 norm: %E\n", (double)(sumDelta / sumRef));
}
else
{
printf("Avg. diff: %E\n", (double)(sumDelta / (real)OPT_N));
}
printf("CPU binomial vs. Black-Scholes\n");
sumDelta = 0;
sumRef = 0;
for (i = 0; i < OPT_N; i++)
{
sumDelta += fabs(callValueBS[i]- callValueCPU[i]);
sumRef += fabs(callValueBS[i]);
}
if (sumRef >1E-5)
{
printf("L1 norm: %E\n", sumDelta / sumRef);
}
else
{
printf("Avg. diff: %E\n", (double)(sumDelta / (real)OPT_N));
}
printf("CPU binomial vs. GPU binomial\n");
sumDelta = 0;
sumRef = 0;
for (i = 0; i < OPT_N; i++)
{
sumDelta += fabs(callValueGPU[i] - callValueCPU[i]);
sumRef += callValueCPU[i];
}
if (sumRef > 1E-5)
{
printf("L1 norm: %E\n", errorVal = sumDelta / sumRef);
}
else
{
printf("Avg. diff: %E\n", (double)(sumDelta / (real)OPT_N));
}
printf("Shutting down...\n");
sdkDeleteTimer(&hTimer);
// cudaDeviceReset causes the driver to clean up all state. While
// not mandatory in normal operation, it is good practice. It is also
// needed to ensure correct operation when the application is being
// profiled. Calling cudaDeviceReset causes all profile data to be
// flushed before the application exits
cudaDeviceReset();
printf("\nNOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.\n\n");
if (errorVal > 5e-4)
{
printf("Test failed!\n");
exit(EXIT_FAILURE);
}
printf("Test passed\n");
exit(EXIT_SUCCESS);
}