chundoong-lab-ta/APWS23/project/main.cpp

88 lines
1.8 KiB
C++
Raw Normal View History

2023-02-15 01:33:28 +09:00
#include <cuda_runtime.h>
#include <unistd.h>
2023-02-01 22:30:00 +09:00
#include <cstdio>
#include <cstdlib>
2023-02-02 19:02:36 +09:00
#include "uNet.h"
2023-02-01 22:30:00 +09:00
#include "util.h"
2023-02-06 15:07:20 +09:00
int WARMUP = 0;
int MEASURE = 1;
2023-02-01 22:30:00 +09:00
// Global variables
int N = 1;
char *parameter_fname;
char *output_fname;
char *input_fname;
2023-02-01 22:30:00 +09:00
/*
Execute Parameters
argv[1] = unet model parameters path
argv[2] = input image path
argv[3] = output save path
argv[4] = the number of images to inference
// optional
argv[5] = warming up count
argv[6] = performance measuring count
*/
2023-02-06 15:07:20 +09:00
int main(int argc, char **argv) {
2023-02-01 22:30:00 +09:00
check_and_parse_args(argc, argv);
2023-02-06 01:38:42 +09:00
print_model();
2023-02-01 22:30:00 +09:00
// Initialize model
2023-02-06 01:38:42 +09:00
uNet_initialize(N, parameter_fname);
2023-02-01 22:30:00 +09:00
Tensor *input = new Tensor({N, 3, 128, 191});
Tensor *output = new Tensor({N, 2, 128, 191});
2023-02-15 01:33:28 +09:00
2023-02-06 01:38:42 +09:00
size_t input_size = 0;
2023-02-15 01:33:28 +09:00
read_binary((void *) input->buf, input_fname, &input_size);
2023-02-03 01:25:19 +09:00
if (argc > 5){
WARMUP = atoi(argv[5]);
MEASURE = atoi(argv[6]);
}
2023-02-06 15:07:20 +09:00
printf(" process %d image(s)...", N);
printf(" Warm up [%d], Performance measure [%d]\n",WARMUP, MEASURE);
2023-02-15 01:33:28 +09:00
fflush(stdout);
2023-02-01 22:30:00 +09:00
// warming up
2023-02-06 15:21:16 +09:00
printf("\nWarmimg up.");
2023-02-14 17:45:28 +09:00
fflush(stdout);
for (int i = 0; i < WARMUP; ++i) {
2023-02-06 15:07:20 +09:00
uNet(input, output, N);
2023-02-14 17:45:28 +09:00
fflush(stdout);
2023-02-06 15:21:16 +09:00
printf(".");
}
cudaDeviceSynchronize();
printf("\n");
2023-02-06 15:07:20 +09:00
// performance measure
2023-02-06 15:21:16 +09:00
printf("\nProcess.");
2023-02-02 23:37:40 +09:00
double uNet_st = get_time();
2023-02-15 01:33:28 +09:00
for (int j = 0; j < MEASURE; ++j) {
2023-02-06 15:21:16 +09:00
uNet(input, output, N);
printf(".");
}
2023-02-06 15:07:20 +09:00
cudaDeviceSynchronize();
2023-02-15 01:33:28 +09:00
double uNet_en = get_time();
2023-02-06 15:21:16 +09:00
printf("\n");
2023-02-15 01:33:28 +09:00
double elapsed_time = uNet_en - uNet_st;
elapsed_time = elapsed_time / MEASURE;
printf("%lfsec (%lf img/sec)\n", elapsed_time, N / elapsed_time);
2023-02-06 15:07:20 +09:00
2023-02-15 01:33:28 +09:00
write_binary((void *) output->buf, output_fname, (size_t)(N * 2 * 128 * 191));
2023-02-06 01:38:42 +09:00
printf("Writing final result to %s ...", output_fname);
2023-02-06 01:38:42 +09:00
fflush(stdout);
2023-02-15 01:33:28 +09:00
2023-02-06 01:38:42 +09:00
printf("Done!\n\n");
2023-02-01 22:30:00 +09:00
// Finalize program
2023-02-02 23:37:40 +09:00
uNet_finalize();
2023-02-01 22:30:00 +09:00
}