Fix headers

src/mnist/cuda_utils.cu

@@ -1,29 +1,40 @@
 #include <stdio.h>
 #include <stdlib.h>
 
+#include "include/cuda_utils.h"
 
 int*** copy_images_cuda(int*** images, int nb_images, int width, int height) {
     int*** images_cuda;
-    cudaMalloc((int****)&images_cuda, sizeof(int**)*nb_images);
-    cudaMemcpy((int****)&images_cuda, sizeof(int**)*nb_images, images);
+    cudaMalloc(&images_cuda, (size_t)sizeof(int**)*nb_images);
+    cudaMemcpy(images_cuda, &images, (size_t)sizeof(int**)*nb_images, cudaMemcpyHostToDevice);
 
     for (int i=0; i < nb_images; i++) {
-        cudaMalloc((int***)&images_cuda[i], sizeof(int**)*nb_images);
-        cudaMemcpy((int***)&images_cuda[i], sizeof(int**)*nb_images, images[i]);
+        cudaMalloc(&images_cuda[i], sizeof(int**)*nb_images);
+        cudaMemcpy(images_cuda[i], &images[i], sizeof(int**)*nb_images, cudaMemcpyHostToDevice);
             for (int j=0; j < height; j++) {
                 cudaMalloc((int**)&images_cuda[i][j], sizeof(int*)*width);
-                cudaMemcpy((int**)&images_cuda[i][j], sizeof(int*)*width, images[i][j]);
+                cudaMemcpy(images_cuda[i][j], &images[i][j], sizeof(int*)*width, cudaMemcpyHostToDevice);
             }
     }
     return images_cuda;
 }
 
 
-
-
 unsigned int* copy_labels_cuda(unsigned int* labels) {
     unsigned int* labels_cuda;
-    cudaMalloc((unsigned int**)&labels_cuda, sizeof(labels));
-    cudaMemcpy((unsigned int**)&labels_cuda, sizeof(labels), labels);
+    cudaMalloc(&labels_cuda, (size_t)sizeof(labels));
+    cudaMemcpy(labels_cuda, &labels, sizeof(labels), cudaMemcpyHostToDevice);
     return labels_cuda;
+}
+
+
+void check_cuda_compatibility() {
+    int nDevices;
+    cudaError_t err = cudaGetDeviceCount(&nDevices);
+    if (err != cudaSuccess) {
+        printf("%s\n", cudaGetErrorString(err));
+        exit(1);
+    } else {
+        printf("CUDA-capable device is detected\n");
+    }
 }

src/mnist/include/cuda_utils.h

@@ -6,5 +6,6 @@
 
 int*** copy_images_cuda(int*** images, int nb_images, int width, int height);
 unsigned int* copy_labels_cuda(unsigned int* labels);
+void check_cuda_compatibility();
 
 #endif

src/mnist/include/main.h

@@ -0,0 +1,27 @@
+#ifndef DEF_MAIN_H
+#define DEF_MAIN_H
+
+typedef struct TrainParameters {
+    Network* network;
+    int*** images;
+    int* labels;
+    int start;
+    int nb_images;
+    int height;
+    int width;
+    float accuracy;
+} TrainParameters;
+
+
+void print_image(unsigned int width, unsigned int height, int** image, float* previsions);
+int indice_max(float* tab, int n);
+void help(char* call);
+void write_image_in_network(int** image, Network* network, int height, int width);
+void* train_images(void* parameters);
+void train(int epochs, int layers, int neurons, char* recovery, char* image_file, char* label_file, char* out, char* delta, int nb_images_to_process, int start);
+float** recognize(char* modele, char* entree);
+void print_recognize(char* modele, char* entree, char* sortie);
+void test(char* modele, char* fichier_images, char* fichier_labels, bool preview_fails);
+int main(int argc, char* argv[]);
+
+#endif

src/mnist/include/preview.h

@@ -6,10 +6,8 @@
 #ifndef DEF_PREVIEW_H
 #define DEF_PREVIEW_H
 
-uint32_t swap_endian(uint32_t val);
-void print_image(unsigned int width, unsigned int height, FILE* ptr, int start);
-void read_mnist_images(char* filename, unsigned int* labels);
-unsigned int* read_mnist_labels(char* filename);
+void print_image(unsigned int width, unsigned int height, int** image);
+void preview_images(char* images_file, char* labels_file);
 
 #endif
 

src/mnist/main.c

@@ -9,25 +9,17 @@
 #include "neuron_io.c"
 #include "mnist.c"
 
+#include "include/main.h"
+
 #define EPOCHS 10
 #define BATCHES 100
 
 #ifdef __CUDACC__
+#   warning compiling for CUDA compatible device only
 #   include "cuda_utils.cu"
 #   define MAX_CUDA_THREADS 1024 // from NVIDIA documentation
 #endif
 
-typedef struct TrainParameters {
-    Network* network;
-    int*** images;
-    int* labels;
-    int start;
-    int nb_images;
-    int height;
-    int width;
-    float accuracy;
-} TrainParameters;
-
 
 void print_image(unsigned int width, unsigned int height, int** image, float* previsions) {
     char tab[] = {' ', '.', ':', '%', '#', '\0'};
@@ -140,7 +132,8 @@ void train(int epochs, int layers, int neurons, char* recovery, char* image_file
     float accuracy;
 
     #ifdef __CUDACC__
-    printf("Utilisation du GPU\n");
+    printf("Testing compatibility...\n");
+    check_cuda_compatibility();
     int nb_threads = MAX_CUDA_THREADS;
     #else
     printf("Pas d'utilisation du GPU\n");

src/mnist/mnist.c

@@ -3,6 +3,7 @@
 #include <stdint.h>
 #include <inttypes.h>
 
+#include "include/mnist.h"
 
 uint32_t swap_endian(uint32_t val) {
     val = ((val << 8) & 0xFF00FF00) | ((val >> 8) & 0xFF00FF);

src/mnist/neural_network.c

@@ -7,6 +7,7 @@
 #include <time.h>
 
 #include "include/neuron.h"
+#include "include/neural_network.h"
 
 // Définit le taux d'apprentissage du réseau neuronal, donc la rapidité d'adaptation du modèle (compris entre 0 et 1)
 // Cette valeur peut évoluer au fur et à mesure des époques (linéaire c'est mieux)
@@ -404,16 +405,16 @@ Network* copy_network_cuda(Network* network) {
     Neuron* neuron1;
     Neuron* neuron;
 
-    cudaMalloc((void**)&network2, sizeof(Network));
+    cudaMalloc(&network2, (size_t)sizeof(Network));
 
     network2->nb_layers = network->nb_layers;
-    cudaMalloc((void***)&network2->layers, sizeof(Layer*)*network->nb_layers);
+    cudaMalloc(&network2->layers, (size_t)sizeof(Layer*)*network->nb_layers);
     for (int i=0; i < network2->nb_layers; i++) {
-        cudaMalloc((void**)&layer, sizeof(Layer));
+        cudaMalloc(&layer, (size_t)sizeof(Layer));
         layer->nb_neurons = network->layers[i]->nb_neurons;
-        cudaMalloc((void***)&layer->neurons, sizeof(Neuron*)*layer->nb_neurons);
+        cudaMalloc(&layer->neurons, (size_t)sizeof(Neuron*)*layer->nb_neurons);
         for (int j=0; j < layer->nb_neurons; j++) {
-            cudaMalloc((void**)neuron, sizeof(Neuron));
+            cudaMalloc(&neuron, (size_t)sizeof(Neuron));
 
             neuron1 = network->layers[i]->neurons[j];
             neuron->bias = neuron1->bias;
@@ -422,9 +423,10 @@ Network* copy_network_cuda(Network* network) {
             neuron->last_back_bias = neuron1->last_back_bias;
             if (i != network2->nb_layers-1) {
                 (void)network2->layers[i+1]->nb_neurons;
-                cudaMalloc((float**)&neuron->weights, sizeof(float)*network->layers[i+1]->nb_neurons);
-                cudaMalloc((float**)&neuron->back_weights, sizeof(float)*network->layers[i+1]->nb_neurons);
-                cudaMalloc((float**)&neuron->last_back_weights, sizeof(float)*network->layers[i+1]->nb_neurons);
+                cudaMalloc(&neuron->weights,            (size_t)sizeof(float)*network->layers[i+1]->nb_neurons);
+                cudaMalloc(&neuron->back_weights,       (size_t)sizeof(float)*network->layers[i+1]->nb_neurons);
+                cudaMalloc(&neuron->last_back_weights,  (size_t)sizeof(float)*network->layers[i+1]->nb_neurons);
+
                 for (int k=0; k < network->layers[i+1]->nb_neurons; k++) {
                     neuron->weights[k] = neuron1->weights[k];
                     neuron->back_weights[k] = neuron1->back_weights[k];

src/mnist/neuron_io.c

@@ -4,6 +4,7 @@
 #include <inttypes.h>
 
 #include "include/neuron.h"
+#include "include/neuron_io.h"
 
 #define MAGIC_NUMBER       2023
 #define DELTA_MAGIC_NUMBER 2024

src/mnist/preview.c

@@ -4,6 +4,7 @@
 #include <inttypes.h>
 
 #include "mnist.c"
+#include "include/preview.h"
 
 
 // Prévisualise un chiffre écrit à la main
@@ -11,8 +12,8 @@
 void print_image(unsigned int width, unsigned int height, int** image) {
     char tab[] = {' ', '.', ':', '%', '#', '\0'};
 
-    for (int i=0; i < height; i++) {
-        for (int j=0; j < width; j++) {
+    for (int i=0; i < (int)height; i++) {
+        for (int j=0; j < (int)width; j++) {
             printf("%c", tab[image[i][j]/52]);
         }
         printf("\n");