void vp9_convolve8_c(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {
convolve_c(src, src_stride, dst, dst_stride,
filter_x, x_step_q4, filter_y, y_step_q4, w, h, 8);
}
int main(int argc, char * argv[])
{
unsigned int nxl, nyl;
PV::Timer timer;
int status = 0;
int argid = 0;
int query = 1;
int device = DEVICE;
if (argc > 1) {
device = atoi(argv[1]);
}
PV::CLDevice * cld = new PV::CLDevice(device);
// query and print information about the devices found
//
if (query) cld->query_device_info();
PV::CLKernel * kernel;
if (device == 1) {
printf("running on gpu, I hope\n");
nxl = NXL;
nyl = NYL;
kernel = cld->createKernel("convolve.cl", "convolve");
}
else {
nxl = 1;
nyl = 1;
kernel = cld->createKernel("convolve_cpu.cl", "convolve_cpu");
}
size_t global; // global domain size for our calculation
size_t local; // local domain size for our calculation
//cl_mem input; // device memory used for the input array
//cl_mem output; // device memory used for the output array
PV::CLBuffer * input; // device memory used for the input array
PV::CLBuffer * output; // device memory used for the output array
//const unsigned int size_ex = SIZE_EX;
//const unsigned int size_img = SIZE_IMG;
size_t size_ex = SIZE_EX * sizeof(float);
size_t size_img = SIZE_IMG * sizeof(float);
const unsigned int nxGlobal = NXGLOBAL;
const unsigned int nyGlobal = NYGLOBAL;
const unsigned int nPad = NPAD;
const unsigned int nPad2 = NPAD2;
const unsigned int sx = 1;
const unsigned int sy = nxGlobal + nPad2;
// float * data = (float *) malloc(size_ex * sizeof(float)); // original data set given to device
// float * results_d = (float *) malloc(size_img * sizeof(float)); // results returned from device
// float * results_l = (float *) malloc(size_img * sizeof(float)); // results returned from local thread
// unsigned char * activity = (unsigned char *) malloc(size_ex * sizeof(unsigned char));
float * data = (float *) malloc(size_ex); // original data set given to device
float * results_d = (float *) malloc(size_img); // results returned from device
float * results_l = (float *) malloc(size_img); // results returned from local thread
//unsigned char * activity = (unsigned char *) malloc(size_ex * sizeof(unsigned char));
assert(data != NULL);
assert(results_d != NULL);
assert(results_l != NULL);
//assert(activity != NULL);
bzero(data, size_ex);
bzero(results_d, size_img);
bzero(results_l, size_img);
// bzero(data, size_ex*sizeof(float));
// bzero(results_d, size_img*sizeof(float));
// bzero(results_l, size_img*sizeof(float));
//bzero(activity, size_ex*sizeof(unsigned char));
size_t local_size_ex = (nxl + nPad2) * (nyl + nPad2) * sizeof(float); // padded image patch
//init_random_data(data, nxGlobal, nyGlobal, nPad);
init_test_data(data, nxGlobal, nyGlobal, nPad);
// time running kernel locally
//
timer.start();
convolve_c(data, results_l, nxGlobal, nyGlobal, nPad);
timer.stop();
printf("Executing on local: "); timer.elapsed_time();
#ifdef USE_ACTIVITY_BYTES
input = cld->addReadBuffer (argid++, activity, size_ex*sizeof(unsigned char));
#else
//input = cld->addReadBuffer (argid++, data, size_ex*sizeof(float));
input = cld->createReadBuffer (size_ex, data);
input->copyToDevice();
status |= kernel->setKernelArg(argid++, input);
#endif
//output = cld->addWriteBuffer(argid++, size_img*sizeof(float));
output = cld->createWriteBuffer(size_img, results_d);
status |= kernel->setKernelArg(argid++, output);
// status = cld->addKernelArg (argid++, nxGlobal);
// status = cld->addKernelArg (argid++, nyGlobal);
// status = cld->addKernelArg (argid++, nPad);
// status = cld->addLocalArg (argid++, local_size_ex);
status |= kernel->setKernelArg (argid++, (int)nxGlobal);
status |= kernel->setKernelArg (argid++, (int)nyGlobal);
status |= kernel->setKernelArg (argid++, (int)nPad);
status |= kernel->setLocalArg (argid++, local_size_ex);
timer.start();
#ifdef USE_ACTIVITY_BYTES
cld->run(nxGlobal/4, nyGlobal, nxl, nyl);
#else
//cld->run(nxGlobal, nyGlobal, nxl, nyl);
printf("starting run...\n");
kernel->run((size_t)nxGlobal, (size_t)nyGlobal, nxl, nyl);
#endif
timer.stop();
printf("Executing on device: "); timer.elapsed_time();
printf("Elapsed time on device: device time == %f \n", ((float)kernel->get_execution_time())/1.0e6);
//cld->copyResultsBuffer(output, results_d, size_img*sizeof(float));
output->copyFromDevice();
// Check results for accuracy
//
check_results(results_d, results_l, nxGlobal, nyGlobal, nPad);
//validate_results(results_d, results_l, nxGlobal, nyGlobal, nPad);
// Shutdown and cleanup
//
//clReleaseMemObject(input);
//clReleaseMemObject(output);
delete input;
delete output;
delete cld;
printf("Finished...\n");
return status;
}
