Merge pull request #3 from augustin64/macos

Makefile

@@ -1,3 +1,4 @@
+OS           := $(shell uname)
 BUILDDIR     := ./build
 SRCDIR       := ./src
 CACHE_DIR    := ./.cache
@@ -27,8 +28,8 @@ TESTS_SRC_CU += $(wildcard $(TEST_SRCDIR)/*.cu)
 TESTS_OBJ     = $(TESTS_SRC:$(TEST_SRCDIR)/%.c=$(BUILDDIR)/$(TEST_SRCDIR)-%) $(TESTS_SRC_CU:$(TEST_SRCDIR)/%.cu=$(BUILDDIR)/$(TEST_SRCDIR)-%)
 
 # Linker only flags
-LD_CFLAGS    =  -lm -lpthread -ljpeg -fopenmp
+LD_CFLAGS    =  -lm -lpthread -ljpeg
-LD_NVCCFLAGS = -ljpeg -Xcompiler -fopenmp
+LD_NVCCFLAGS = -ljpeg
 
 # Compilation flag
 CFLAGS    = -Wall -Wextra -std=gnu99 -g -O3
@@ -41,6 +42,13 @@ NVCCFLAGS = -g
 # -fsanitize=address -lasan
 #! WARNING: test/cnn-neuron_io fails with this option enabled
 
+# Specify library path of libjpeg on MacOS
+ifeq ($(OS),Darwin)
+	LD_CFLAGS    += -I/opt/homebrew/Cellar/jpeg/9e/include/ -L/opt/homebrew/Cellar/jpeg/9e/lib/
+	LD_NVCCFLAGS += -L/opt/homebrew/Cellar/jpeg/9e/lib/
+
+	CFLAGS       +=  -I/opt/homebrew/Cellar/jpeg/9e/include/
+endif
 
 
 all: dense cnn;

README.md

@@ -178,17 +178,24 @@ Résultats avec VGG16, pour des images de 256x256 pixels (seulement une plus pet
 
 Sur le cloud avec google Colab: bon GPU mais mauvais processeur: [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1LfwSrQRaoC91yC9mx9BKHzuc7odev5r6?usp=sharing)
 
-Les distributions suivantes ont étés essayées, il sera sans doute nécessaire de modifier le code pour l'exécuter sous Windows/ MacOS:
+## Dépendances
+- `cuda` : pour utiliser la carte graphique (NVIDIA seulement)
+- `libjpeg-dev` : n'est pas nécessairement installé par défaut
+- GNU `make` : installé par défaut sur la majorité des distributions Linux et sur MacOS
+- `gcc` : installé par défaut sur la majorité des distributions Linux et sur MacOS
+
+### Linux
+Les distributions suivantes ont étés essayées, il faudra parfois installer `libjpeg`
 - Arch  
 - Fedora  
 - Manjaro  
-- Ubuntu
+- Ubuntu: `apt install libjpeg-dev`
 
-## Dépendances
+### MacOS
-- `cuda` : pour utiliser la carte graphique (NVIDIA seulement)
+Avec [Homebrew](https://brew.sh/):
-- `libjpeg-dev` : n'est pas installé par défaut sur ubuntu notamment
+```bash
-- GNU `make` : installé par défaut sur la majorité des distributions
+brew install libjpeg
-- `gcc` : installé par défaut sur la majorité des distributions
+```
 
 ## Compilation
 

src/cnn/cnn.c

@@ -9,11 +9,6 @@
 #include "../common/include/colors.h"
 #include "../common/include/utils.h"
 #include "include/backpropagation.h"
-#include "include/initialisation.h"
-#include "include/convolution.h"
-#include "include/function.h"
-#include "include/creation.h"
-#include "include/update.h"
 #include "include/make.h"
 
 #include "include/cnn.h"

src/cnn/main.c

@@ -7,12 +7,8 @@
 
 
 #include "../common/include/colors.h"
-#include "include/initialisation.h"
 #include "include/test_network.h"
-#include "include/function.h"
-#include "include/creation.h"
 #include "include/train.h"
-#include "include/cnn.h"
 
 #include "include/main.h"
 

src/cnn/test_network.c

@@ -7,7 +7,6 @@
 #include "../common/include/memory_management.h"
 #include "../common/include/mnist.h"
 #include "include/neuron_io.h"
-#include "include/struct.h"
 #include "include/jpeg.h"
 #include "include/free.h"
 #include "include/cnn.h"

src/cnn/train.c

@@ -1,11 +1,18 @@
-#include <sys/sysinfo.h>
 #include <pthread.h>
 #include <stdlib.h>
 #include <stdio.h>
 #include <float.h>
 #include <math.h>
 #include <time.h>
-#include <omp.h>
+#include <time.h>
+
+#ifdef __linux__
+    #include <sys/sysinfo.h>
+#elif defined(__APPLE__)
+    #include <sys/sysctl.h>
+#else
+    #error Unknown platform
+#endif
 
 #include "../common/include/memory_management.h"
 #include "../common/include/colors.h"
@@ -65,7 +72,7 @@ void* train_thread(void* parameters) {
     float loss = 0.;
 
     #ifdef DETAILED_TRAIN_TIMINGS
-        double start_time;
+        clock_t start_time;
     #endif
 
     pthread_t tid;
@@ -82,16 +89,16 @@ void* train_thread(void* parameters) {
             write_image_in_network_32(images[index[i]], height, width, network->input[0][0], param->offset);
 
             #ifdef DETAILED_TRAIN_TIMINGS
-                start_time = omp_get_wtime();
+                start_time = clock();
             #endif
 
             forward_propagation(network);
 
             #ifdef DETAILED_TRAIN_TIMINGS
                 printf("Temps de forward: ");
-                printf_time(omp_get_wtime() - start_time);
+                printf_time(clock() - start_time);
                 printf("\n");
-                start_time = omp_get_wtime();
+                start_time = clock();
             #endif
 
             maxi = indice_max(network->input[network->size-1][0][0], 10);
@@ -109,9 +116,9 @@ void* train_thread(void* parameters) {
 
             #ifdef DETAILED_TRAIN_TIMINGS
                 printf("Temps de backward: ");
-                printf_time(omp_get_wtime() - start_time);
+                printf_time(clock() - start_time);
                 printf("\n");
-                start_time = omp_get_wtime();
+                start_time = clock();
             #endif
 
             if (maxi == labels[index[i]]) {
@@ -132,16 +139,16 @@ void* train_thread(void* parameters) {
             write_256_image_in_network(param->dataset->images[index[i]], width, height, param->dataset->numComponents, network->width[0], network->input[0]);
 
             #ifdef DETAILED_TRAIN_TIMINGS
-                start_time = omp_get_wtime();
+                start_time = clock();
             #endif
 
             forward_propagation(network);
 
             #ifdef DETAILED_TRAIN_TIMINGS
                 printf("Temps de forward: ");
-                printf_time(omp_get_wtime() - start_time);
+                printf_time(clock() - start_time);
                 printf("\n");
-                start_time = omp_get_wtime();
+                start_time = clock();
             #endif
 
             maxi = indice_max(network->input[network->size-1][0][0], param->dataset->numCategories);
@@ -149,9 +156,9 @@ void* train_thread(void* parameters) {
             
             #ifdef DETAILED_TRAIN_TIMINGS
                 printf("Temps de backward: ");
-                printf_time(omp_get_wtime() - start_time);
+                printf_time(clock() - start_time);
                 printf("\n");
-                start_time = omp_get_wtime();
+                start_time = clock();
             #endif
 
 
@@ -180,7 +187,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
         exit(1);
     }
     #endif
-    srand(time(NULL));
+    srand(clock());
     float loss;
     float batch_loss; // May be redundant with loss, but gives more informations
     float test_accuracy = 0.; // Used to decrease Learning rate
@@ -191,12 +198,12 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
 
 
     //* Différents timers pour mesurer les performance en terme de vitesse
-    double start_time, end_time;
+    clock_t start_time, end_time;
-    double elapsed_time;
+    clock_t elapsed_time;
 
-    double algo_start = omp_get_wtime();
+    clock_t algo_start = clock();
 
-    start_time = omp_get_wtime();
+    start_time = clock();
 
 
     //* Chargement du dataset
@@ -266,7 +273,17 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
     #ifdef USE_MULTITHREADING
         int nb_remaining_images; // Nombre d'images restantes à lancer pour une série de threads
         // Récupération du nombre de threads disponibles
-        int nb_threads = get_nprocs();
+        #ifdef __linux__
+            int nb_threads = get_nprocs();
+        #elif defined(__APPLE__)
+            int nb_threads;
+            size_t len = sizeof(nb_threads);
+
+            if (sysctlbyname("hw.logicalcpu", &nb_threads, &len, NULL, 0) == -1) {
+                perror("sysctl");
+                exit(1);
+            }
+        #endif
         pthread_t *tid = (pthread_t*)malloc(nb_threads * sizeof(pthread_t));
 
         // Création des paramètres donnés à chaque thread dans le cas du multi-threading
@@ -325,7 +342,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
         train_params->finetuning = finetuning;
     #endif
 
-    end_time = omp_get_wtime();
+    end_time = clock();
 
     elapsed_time = end_time - start_time;
     printf("Taux d'apprentissage initial: %0.2e\n", network->learning_rate);
@@ -336,7 +353,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
     //* Boucle d'apprentissage
     for (int i=0; i < epochs; i++) {
 
-        start_time = omp_get_wtime();
+        start_time = clock();
         // 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
@@ -428,7 +445,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
             #endif
         }
         //* Fin d'une époque: affichage des résultats et sauvegarde du réseau
-        end_time = omp_get_wtime();
+        end_time = clock();
         elapsed_time = end_time - start_time;
         #ifdef USE_MULTITHREADING
         printf("\rThreads [%d]\tÉpoque [%d/%d]\tImage [%d/%d]\tAccuracy: " GREEN "%0.4f%%" RESET " \tLoss: %lf\tTemps: ", nb_threads, i, epochs, nb_images_total, nb_images_total, accuracy*100, loss);
@@ -489,7 +506,7 @@ void train(int dataset_type, char* images_file, char* labels_file, char* data_di
         free_dataset(dataset);
     }
 
-    end_time = omp_get_wtime();
+    end_time = clock();
     elapsed_time = end_time - algo_start;
     printf("\nTemps total: ");
     printf_time(elapsed_time);

src/common/colors.c

@@ -1,5 +1,6 @@
 #include <stdio.h>
 #include <stdbool.h>
+#include <time.h>
 
 #include "include/colors.h"
 
@@ -15,11 +16,13 @@ void printf_info(char* string) {
     printf(BOLDBLUE "[ INFO  ]" RESET " %s", string);
 }
 
-void printf_time(float time) {
+void printf_time(clock_t time) {
-    int hours = time/3600;
+    double real_time = (double) time / CLOCKS_PER_SEC;
-    int minutes = ((int)time %3600)/60;
+
-    int seconds = ((int)time) %60;
+    int hours = real_time/3600;
-    int milliseconds = (time - (int)time)*1000;
+    int minutes = ((int)real_time %3600)/60;
+    int seconds = ((int)real_time) %60;
+    int milliseconds = (real_time - (int)real_time)*1000;
 
     if (hours != 0) {
         printf("%dh %dmn", hours, minutes);

src/common/include/colors.h

@@ -1,4 +1,5 @@
 #include <stdio.h>
+#include <time.h>
 
 #ifndef DEF_COLORS_H
 #define DEF_COLORS_H
@@ -51,7 +52,7 @@ extern "C"
 /*
 * Affiche un timing en heures minutes secondes millisecondes en limitant la précision aux deux unités les plus significatives
 */
-void printf_time(float time);
+void printf_time(clock_t time);
 
 #ifdef __CUDACC__
 extern "C"

src/dense/main.c

@@ -4,7 +4,14 @@
 #include <float.h>
 #include <stdbool.h>
 #include <pthread.h>
-#include <sys/sysinfo.h>
+
+#ifdef __linux__
+    #include <sys/sysinfo.h>
+#elif defined(__APPLE__)
+    #include <sys/sysctl.h>
+#else
+    #error Unknown platform
+#endif
 
 #include "include/neural_network.h"
 #include "../common/include/colors.h"
@@ -201,7 +208,17 @@ void train(int epochs, char* recovery, char* image_file, char* label_file, char*
     float accuracy;
     float current_accuracy;
 
-    int nb_threads = get_nprocs();
+    #ifdef __linux__
+        int nb_threads = get_nprocs();
+    #elif defined(__APPLE__)
+        int nb_threads;
+        size_t len = sizeof(nb_threads);
+
+        if (sysctlbyname("hw.logicalcpu", &nb_threads, &len, NULL, 0) == -1) {
+            perror("sysctl");
+            exit(1);
+        }
+    #endif
     pthread_t *tid = (pthread_t *)malloc(nb_threads * sizeof(pthread_t));
 
     /*

test/cnn_convolution.cu

@@ -4,7 +4,7 @@
 #include <assert.h>
 #include <math.h>
 #include <time.h>
-#include <omp.h>
+#include <time.h>
 
 #include "../src/common/include/memory_management.h"
 #include "../src/cnn/include/convolution.h"
@@ -121,21 +121,21 @@ void run_convolution_test(int input_width, int output_width, int rows, int colum
 
 
     // Lancement des calculs
-    double start_time, end_time;
+    clock_t start_time, end_time;
-    double cpu_time_used, gpu_time_used;
+    clock_t cpu_time_used, gpu_time_used;
 
-    start_time = omp_get_wtime();
+    start_time = clock();
     make_convolution_device(kernel, input, output_gpu, output_width, 1, 0);
-    end_time = omp_get_wtime();
+    end_time = clock();
 
 
     gpu_time_used = end_time - start_time;
     printf("(%d, %d, %d, %d) Time used for GPU: %lf seconds\n", rows, columns, input_width, output_width, gpu_time_used);
 
 
-    start_time = omp_get_wtime();
+    start_time = clock();
     make_convolution_cpu(kernel, input, output_cpu, output_width, 1, 0);
-    end_time = omp_get_wtime();
+    end_time = clock();
 
     cpu_time_used = end_time - start_time;
     printf("(%d, %d, %d, %d) Time used for CPU: %lf seconds\n", rows, columns, input_width, output_width, cpu_time_used);    
@@ -169,7 +169,7 @@ int main() {
     }
     printf(GREEN "OK\n" RESET);
     
-    srand(time(NULL));
+    srand(clock());
 
     run_convolution_test(20, 15, 30, 40);
     run_convolution_test(30, 25, 40, 50);

test/cnn_jpeg.c

@@ -1,6 +1,6 @@
 #include <stdlib.h>
 #include <stdio.h>
-#include <omp.h>
+#include <time.h>
 
 
 #include "../src/common/include/colors.h"
@@ -21,10 +21,10 @@ int main(int argc, char* argv[]) {
     printf("Taille des images:    %dx%d\n", dataset->width, dataset->height);
 
     // Calcul du temps de chargement des images une à une
-    double start_time, end_time;
+    clock_t start_time, end_time;
 
     int N = min(100000, dataset->numImages);
-    start_time = omp_get_wtime();
+    start_time = clock();
     printf("Chargement de %d images\n", N);
     for (int i=0; i < N; i++) {
         imgRawImage* image = loadJpegImageFile(dataset->fileNames[i]);
@@ -32,8 +32,10 @@ int main(int argc, char* argv[]) {
         free(image);
     }
     printf("OK\n");
-    end_time = omp_get_wtime();
+    end_time = clock();
-    printf("Temps par image (calculé sur une moyenne de %d): %lf s\n", N, (end_time - start_time)/N);
+    printf("Temps par image (calculé sur une moyenne de %d): ", N);
+    printf_time((end_time - start_time)/N);
+    printf("\n");
 
     for (int i=0; i < (int)dataset->numImages; i++) {
         if (!dataset->fileNames[i]) {

test/cnn_matrix_multiplication.cu

@@ -3,7 +3,7 @@
 #include <stdbool.h>
 #include <math.h>
 #include <time.h>
-#include <omp.h>
+#include <time.h>
 
 #include "../src/cnn/include/matrix_multiplication.h"
 #include "../src/common/include/memory_management.h"
@@ -72,8 +72,8 @@ bool check_matrices_equality(float** m1, float** m2, int n, int p, int acceptati
 }
 
 void run_matrices_test(int n, int p, int q) {
-    double start_time, end_time;
+    clock_t start_time, end_time;
-    double cpu_time_used, gpu_time_used;
+    clock_t cpu_time_used, gpu_time_used;
 
     float** matrix1 = create_matrix(n, p);
     float** matrix2 = create_matrix(p, q);
@@ -81,16 +81,16 @@ 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_time = omp_get_wtime();
+    start_time = clock();
     matrix_multiplication_device(matrix1, matrix2, result_gpu, n, p, q);
-    end_time = omp_get_wtime();
+    end_time = clock();
 
     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_time = omp_get_wtime();
+    start_time = clock();
     matrix_multiplication_host(matrix1, matrix2, result_cpu, n, p, q);
-    end_time = omp_get_wtime();
+    end_time = clock();
 
     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);
@@ -134,7 +134,7 @@ int main() {
     }
     printf(GREEN "OK\n" RESET);
 
-    srand(time(NULL));
+    srand(clock());
     run_matrices_test(200, 1000, 200);
     run_matrices_test(200, 1000, 20);
     run_matrices_test(20, 1000, 200);