// execute kernel
double dslashCUDA() {
printfQuda("Executing %d kernel loops...\n", loops);
fflush(stdout);
if (test_type < 2)
dirac->Tune(*cudaSpinorOut, *cudaSpinor, *tmp);
else
dirac->Tune(cudaSpinorOut->Even(), cudaSpinor->Even(), *tmp);
cudaEvent_t start, end;
cudaEventCreate(&start);
cudaEventRecord(start, 0);
cudaEventSynchronize(start);
for (int i = 0; i < loops; i++) {
switch (test_type) {
case 0:
if (transfer) {
dslashQuda(spinorOut->V(), spinor->V(), &inv_param, parity);
} else {
dirac->Dslash(*cudaSpinorOut, *cudaSpinor, parity);
}
break;
case 1:
case 2:
if (transfer) {
MatQuda(spinorOut->V(), spinor->V(), &inv_param);
} else {
dirac->M(*cudaSpinorOut, *cudaSpinor);
}
break;
}
}
cudaEventCreate(&end);
cudaEventRecord(end, 0);
cudaEventSynchronize(end);
float runTime;
cudaEventElapsedTime(&runTime, start, end);
cudaEventDestroy(start);
cudaEventDestroy(end);
double secs = runTime / 1000; //stopwatchReadSeconds();
// check for errors
cudaError_t stat = cudaGetLastError();
if (stat != cudaSuccess)
printf("with ERROR: %s\n", cudaGetErrorString(stat));
printf("done.\n\n");
return secs;
}
// execute kernel
double dslashCUDA(int niter) {
cudaEvent_t start, end;
cudaEventCreate(&start);
cudaEventCreate(&end);
cudaEventRecord(start, 0);
for (int i = 0; i < niter; i++) {
switch (test_type) {
case 0:
if (transfer) {
dslashQuda(spinorOut->V(), spinor->V(), &inv_param, parity);
} else {
//inv_param.input_location = QUDA_CUDA_FIELD_LOCATION;
//inv_param.output_location = QUDA_CUDA_FIELD_LOCATION;
//dslashQuda(cudaSpinorOut->V(), cudaSpinor->V(), &inv_param, parity);
dirac->Dslash(*cudaSpinorOut, *cudaSpinor, parity);
}
break;
case 1:
case 2:
if (transfer) {
MatQuda(spinorOut->V(), spinor->V(), &inv_param);
} else {
dirac->M(*cudaSpinorOut, *cudaSpinor);
}
break;
case 3:
case 4:
if (transfer) {
MatDagMatQuda(spinorOut->V(), spinor->V(), &inv_param);
} else {
dirac->MdagM(*cudaSpinorOut, *cudaSpinor);
}
break;
}
}
cudaEventRecord(end, 0);
cudaEventSynchronize(end);
float runTime;
cudaEventElapsedTime(&runTime, start, end);
cudaEventDestroy(start);
cudaEventDestroy(end);
double secs = runTime / 1000; //stopwatchReadSeconds();
// check for errors
cudaError_t stat = cudaGetLastError();
if (stat != cudaSuccess)
printfQuda("with ERROR: %s\n", cudaGetErrorString(stat));
return secs;
}