augustin64/tipe
1
1
Fork 0
mirror of https://github.com/augustin64/projet-tipe synced 2025-02-02 19:39:39 +01:00
Code Issues Packages Projects Releases Wiki Activity

Use OpenMP instead of clock time

Browse Source 
It was previously not accurate when using threads
This commit is contained in:
augustin64 2023-01-13 15:58:11 +01:00
parent c914370103
commit 03ad4b01e5
3 changed files with 53 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
src/cnn/train.c
View File

@ -4,6 +4,7 @@
#include <pthread.h>
#include <sys/sysinfo.h>
#include <time.h>
#include <omp.h>
#include "../mnist/include/mnist.h"
#include "include/initialisation.h"
@ -92,6 +93,13 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
jpegDataset* dataset; // Structure de données décrivant un dataset d'images jpeg
int* shuffle_index; // shuffle_index[i] contient le nouvel index de l'élément à l'emplacement i avant mélange
double start_time, end_time;
double elapsed_time;
double algo_start = omp_get_wtime();
start_time = omp_get_wtime();
if (dataset_type == 0) { // Type MNIST
// Chargement des images du set de données MNIST
int* parameters = read_mnist_images_parameters(images_file);
@ -113,7 +121,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
// Initialisation du réseau
if (!recover) {
network = create_network_lenet5(1, 0, TANH, GLOROT, input_dim, input_depth);
network = create_network_lenet5(0.1, 0, TANH, GLOROT, input_dim, input_depth);
} else {
network = read_network(recover);
}
@ -179,8 +187,14 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
train_params->nb_images = BATCHES;
train_params->index = shuffle_index;
#endif
end_time = omp_get_wtime();
elapsed_time = end_time - start_time;
printf("Initialisation: %0.2lf s\n\n", elapsed_time);
for (int i=0; i < epochs; i++) {
start_time = omp_get_wtime();
// La variable accuracy permet d'avoir une ESTIMATION
// du taux de réussite et de l'entraînement du réseau,
// mais n'est en aucun cas une valeur réelle dans le cas
@ -213,12 +227,21 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
train_parameters[k]->start = BATCHES*j + (BATCHES/nb_threads)*k;
train_parameters[k]->network = copy_network(network);
pthread_create( &tid[k], NULL, train_thread, (void*) train_parameters[k]);
if (train_parameters[k]->start+train_parameters[k]->nb_images >= nb_images_total) {
train_parameters[k]->nb_images = nb_images_total - train_parameters[k]->start -1;
}
if (train_parameters[k]->nb_images > 0) {
pthread_create( &tid[k], NULL, train_thread, (void*) train_parameters[k]);
} else {
tid[k] = 0;
}
}
for (int k=0; k < nb_threads; k++) {
// On attend la terminaison de chaque thread un à un
pthread_join( tid[k], NULL );
accuracy += train_parameters[k]->accuracy / (float) nb_images_total;
if (tid[k] != 0) {
pthread_join( tid[k], NULL );
accuracy += train_parameters[k]->accuracy / (float) nb_images_total;
}
}
// On attend que tous les fils aient fini avant d'appliquer des modifications au réseau principal
@ -228,7 +251,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
free_network(train_parameters[k]->network);
}
current_accuracy = accuracy * nb_images_total/((j+1)*BATCHES);
printf("\rThreads [%d]\tÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "YELLOW"%0.1f%%"RESET" ", nb_threads, i, epochs, BATCHES*(j+1), nb_images_total, current_accuracy*100);
printf("\rThreads [%d]\tÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "YELLOW"%0.2f%%"RESET" ", nb_threads, i, epochs, BATCHES*(j+1), nb_images_total, current_accuracy*100);
fflush(stdout);
#else
(void)nb_images_total_remaining; // Juste pour enlever un warning
@ -248,14 +271,16 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
update_weights(network, network, train_params->nb_images);
update_bias(network, network, train_params->nb_images);
printf("\rÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "YELLOW"%0.1f%%"RESET" ", i, epochs, BATCHES*(j+1), nb_images_total, current_accuracy*100);
printf("\rÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "YELLOW"%0.4f%%"RESET" ", i, epochs, BATCHES*(j+1), nb_images_total, current_accuracy*100);
fflush(stdout);
#endif
}
end_time = omp_get_wtime();
elapsed_time = end_time - start_time;
#ifdef USE_MULTITHREADING
printf("\rThreads [%d]\tÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "GREEN"%0.1f%%"RESET" \n", nb_threads, i, epochs, nb_images_total, nb_images_total, accuracy*100);
printf("\rThreads [%d]\tÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "GREEN"%0.4f%%"RESET"\tTemps: %0.2f s\n", nb_threads, i, epochs, nb_images_total, nb_images_total, accuracy*100, elapsed_time);
#else
printf("\rÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "GREEN"%0.1f%%"RESET" \n", i, epochs, nb_images_total, nb_images_total, accuracy*100);
printf("\rÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: "GREEN"%0.4f%%"RESET"\tTemps: %0.2f s\n", i, epochs, nb_images_total, nb_images_total, accuracy*100, elapsed_time);
#endif
write_network(out, network);
}
@ -266,4 +291,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
#else
free(train_params);
#endif
end_time = omp_get_wtime();
elapsed_time = end_time - algo_start;
printf("\nTemps total: %0.1f s\n", elapsed_time);
}

16
test/cnn_convolution.cu
View File

@ -4,6 +4,7 @@
#include <assert.h>
#include <math.h>
#include <time.h>
#include <omp.h>
#include "../src/cnn/include/convolution.h"
#include "../src/cnn/include/struct.h"
@ -122,22 +123,23 @@ void run_convolution_test(int input_dim, int output_dim, int rows, int columns)
// Lancement des calculs
clock_t start, end;
double start_time, end_time;
double cpu_time_used, gpu_time_used;
start = clock();
start_time = omp_get_wtime();
make_convolution_device(kernel, input, output_gpu, output_dim);
end = clock();
end_time = omp_get_wtime();
gpu_time_used = ((double) (end - start)) / CLOCKS_PER_SEC;
gpu_time_used = end_time - start_time;
printf("(%d, %d, %d, %d) Time used for GPU: %lf seconds\n", rows, columns, input_dim, output_dim, gpu_time_used);
start = clock();
start_time = omp_get_wtime();
make_convolution_cpu(kernel, input, output_cpu, output_dim);
end = clock();
end_time = omp_get_wtime();
cpu_time_used = ((double) (end - start)) / CLOCKS_PER_SEC;
cpu_time_used = end_time - start_time;
printf("(%d, %d, %d, %d) Time used for CPU: %lf seconds\n", rows, columns, input_dim, output_dim, cpu_time_used);
// Vérification de l'égalité des matrices

15
test/cnn_matrix_multiplication.cu
View File

@ -3,6 +3,7 @@
#include <stdbool.h>
#include <math.h>
#include <time.h>
#include <omp.h>
#include "../src/cnn/include/matrix_multiplication.h"
#include "../src/include/colors.h"
@ -70,7 +71,7 @@ bool check_matrices_equality(float** m1, float** m2, int n, int p, int acceptati
}
void run_matrices_test(int n, int p, int q) {
clock_t start, end;
double start_time, end_time;
double cpu_time_used, gpu_time_used;
float** matrix1 = create_matrix(n, p);
@ -79,18 +80,18 @@ void run_matrices_test(int n, int p, int q) {
float** result_cpu = create_empty_matrix(n, q);
printf("(%d,%d)x(%d,%d) Data generation complete.\n", n, p, p, q);
start = clock();
start_time = omp_get_wtime();
matrix_multiplication_device(matrix1, matrix2, result_gpu, n, p, q);
end = clock();
end_time = omp_get_wtime();
cpu_time_used = ((double) (end - start)) / CLOCKS_PER_SEC;
cpu_time_used = end_time - start_time;
printf("(%d,%d)x(%d,%d) Time used for GPU: %lf seconds\n", n, p, p, q, cpu_time_used);
start = clock();
start_time = omp_get_wtime();
matrix_multiplication_host(matrix1, matrix2, result_cpu, n, p, q);
end = clock();
end_time = omp_get_wtime();
gpu_time_used = ((double) (end - start)) / CLOCKS_PER_SEC;
gpu_time_used = end_time - start_time;
printf("(%d,%d)x(%d,%d) Time used for CPU: %lf seconds\n", n, p, p, q, gpu_time_used);
// Vérification de l'égalité des matrices