int main(int argc, char **argv) {
unsigned int *inputVals;
unsigned int *inputPos;
unsigned int *outputVals;
unsigned int *outputPos;
size_t numElems;
std::string input_file;
std::string template_file;
std::string output_file;
std::string reference_file;
double perPixelError = 0.0;
double globalError = 0.0;
bool useEpsCheck = false;
switch (argc)
{
case 3:
input_file = std::string(argv[1]);
template_file = std::string(argv[2]);
output_file = "HW4_output.png";
break;
case 4:
input_file = std::string(argv[1]);
template_file = std::string(argv[2]);
output_file = std::string(argv[3]);
break;
default:
std::cerr << "Usage: ./HW4 input_file template_file [output_filename]" << std::endl;
exit(1);
}
//load the image and give us our input and output pointers
preProcess(&inputVals, &inputPos, &outputVals, &outputPos, numElems, input_file, template_file);
GpuTimer timer;
timer.Start();
//call the students' code
your_sort(inputVals, inputPos, outputVals, outputPos, numElems);
timer.Stop();
cudaDeviceSynchronize(); checkCudaErrors(cudaGetLastError());
printf("\n");
int err = printf("Your code ran in: %f msecs.\n", timer.Elapsed());
if (err < 0) {
//Couldn't print! Probably the student closed stdout - bad news
std::cerr << "Couldn't print timing information! STDOUT Closed!" << std::endl;
exit(1);
}
//check results and output the red-eye corrected image
postProcess(outputVals, outputPos, numElems, output_file);
// check code moved from HW4.cu
/****************************************************************************
* You can use the code below to help with debugging, but make sure to *
* comment it out again before submitting your assignment for grading, *
* otherwise this code will take too much time and make it seem like your *
* GPU implementation isn't fast enough. *
* *
* This code MUST RUN BEFORE YOUR CODE in case you accidentally change *
* the input values when implementing your radix sort. *
* *
* This code performs the reference radix sort on the host and compares your *
* sorted values to the reference. *
* *
* Thrust containers are used for copying memory from the GPU *
* ************************************************************************* */
thrust::device_ptr<unsigned int> d_inputVals(inputVals);
thrust::device_ptr<unsigned int> d_inputPos(inputPos);
thrust::host_vector<unsigned int> h_inputVals(d_inputVals,
d_inputVals+numElems);
thrust::host_vector<unsigned int> h_inputPos(d_inputPos,
d_inputPos + numElems);
thrust::host_vector<unsigned int> h_outputVals(numElems);
thrust::host_vector<unsigned int> h_outputPos(numElems);
reference_calculation(&h_inputVals[0], &h_inputPos[0],
&h_outputVals[0], &h_outputPos[0],
numElems);
//postProcess(&h_outputVals[0], &h_outputPos[0], numElems, reference_file);
compareImages(reference_file, output_file, useEpsCheck, perPixelError, globalError);
thrust::device_ptr<unsigned int> d_outputVals(outputVals);
thrust::device_ptr<unsigned int> d_outputPos(outputPos);
thrust::host_vector<unsigned int> h_yourOutputVals(d_outputVals,
d_outputVals + numElems);
thrust::host_vector<unsigned int> h_yourOutputPos(d_outputPos,
d_outputPos + numElems);
checkResultsExact(&h_outputVals[0], &h_yourOutputVals[0], numElems);
checkResultsExact(&h_outputPos[0], &h_yourOutputPos[0], numElems);
checkCudaErrors(cudaFree(inputVals));
checkCudaErrors(cudaFree(inputPos));
checkCudaErrors(cudaFree(outputVals));
checkCudaErrors(cudaFree(outputPos));
return 0;
}
int main(int argc, char **argv){
int nTaps = 8;
int nChannels = 1024;
int nStreams = 2;
int NUM_THREADS = 512;
int seg_blocks = 5000;
unsigned int data_size = 10000+nTaps-1;
unsigned int nBlocks = 0;
float error = 1.1f;
bool debug=true;
if (debug) printf("\t\tWelcome\n");
Complex *h_signal, *h_spectra_pinned, *h_spectra_ref, *h_data_pinned;
float *h_coeff;
if (argc >= 2) nChannels = atof(argv[1]);
if (argc >= 3) NUM_THREADS = atof(argv[2]);
if (argc >= 4) nStreams = (atof(argv[3]));
if (argc >= 5) nTaps = (atof(argv[4]));
if (argc >= 6) seg_blocks = (atof(argv[5]));
if (argc >= 7) data_size = (atof(argv[6])+nTaps-1)*nChannels;
nBlocks = (data_size+nTaps-1)/nChannels;nBlocks = data_size/nChannels;
if (debug) printf("\nHost memory allocation...\t");
checkCudaErrors(cudaMallocHost((void**)&h_spectra_pinned, data_size*sizeof(Complex)));
checkCudaErrors(cudaMallocHost((void**)&h_data_pinned, data_size*sizeof(Complex)));
h_signal = (Complex *)malloc(data_size*sizeof(Complex));
h_spectra_ref = (Complex *)malloc(data_size*sizeof(Complex));
h_coeff = (float *)malloc(nTaps*nChannels*sizeof(float));
if (debug) printf("done.");
if (debug) printf("\nHost memory memset...\t\t");
memset(h_spectra_pinned, 0.0, sizeof(Complex)*data_size);
memset(h_spectra_ref, 0.0, sizeof(Complex)*data_size);
if (debug) printf("done.");
if (debug) printf("\nLoad window coefficients...\t");
//Load_window_data(h_coeff);
for (int i = 0; i < nTaps*nChannels; i++)
h_coeff[i] = rand() / (float)RAND_MAX;
if (debug) printf("done.");
if (debug) printf("\nRandom data set...\t\t");
srand(time(NULL));
for (int i=0; i < (int)data_size; i++){
h_signal[i].x = rand() / (float)RAND_MAX;
h_signal[i].y = rand() / (float)RAND_MAX;
}
for (int i = 0; i < (int)data_size; i++){
h_data_pinned[i] = h_signal[i];
}
if (debug) printf("done.");
if (debug) printf("\nReference calculation...\t");
reference_calculation(h_signal, h_spectra_ref, h_coeff, nChannels, nBlocks, nTaps);
if (debug) printf("done.\n");
//printf("CPU jedna %g druha %g", h_spectra_ref[3584], h_spectra_ref[7*512 + 259999].x);
gpu_code(h_data_pinned, h_spectra_pinned, h_coeff, nChannels, nBlocks, data_size, NUM_THREADS, nTaps, nStreams, seg_blocks);
if (debug){
error = reference_code(h_spectra_ref, h_spectra_pinned, nChannels, nTaps, nBlocks);
printf( "error = %lf\n", error);
}
checkCudaErrors(cudaFreeHost(h_spectra_pinned));
checkCudaErrors(cudaFreeHost(h_data_pinned));
delete[] h_signal;
delete[] h_spectra_ref;
delete[] h_coeff;
}