void set_batch_network(network *net, int b) { net->batch = b; int i; for(i = 0; i < net->n; ++i){ net->layers[i].batch = b; #ifdef CUDNN if(net->layers[i].type == CONVOLUTIONAL){ cudnn_convolutional_setup(net->layers + i, cudnn_fastest); /* layer *l = net->layers + i; cudnn_convolutional_setup(l, cudnn_fastest); // check for excessive memory consumption size_t free_byte; size_t total_byte; check_error(cudaMemGetInfo(&free_byte, &total_byte)); if (l->workspace_size > free_byte || l->workspace_size >= total_byte / 2) { printf(" used slow CUDNN algo without Workspace! \n"); cudnn_convolutional_setup(l, cudnn_smallest); l->workspace_size = get_workspace_size(*l); } */ } else if (net->layers[i].type == MAXPOOL) { cudnn_maxpool_setup(net->layers + i); } #endif } }
void set_batch_network(network *net, int b) { net->batch = b; int i; for(i = 0; i < net->n; ++i){ net->layers[i].batch = b; #ifdef CUDNN if(net->layers[i].type == CONVOLUTIONAL){ cudnn_convolutional_setup(net->layers + i); } #endif } }
void set_batch_network(network *net, int b) { net->batch = b; int i; for(i = 0; i < net->n; ++i){ net->layers[i].batch = b; #ifdef CUDNN if(net->layers[i].type == CONVOLUTIONAL){ cudnn_convolutional_setup(net->layers + i); } if(net->layers[i].type == DECONVOLUTIONAL){ layer *l = net->layers + i; cudnnSetTensor4dDescriptor(l->dstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, l->out_c, l->out_h, l->out_w); cudnnSetTensor4dDescriptor(l->normTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, l->out_c, 1, 1); } #endif } }
void resize_convolutional_layer(convolutional_layer *l, int w, int h) { l->w = w; l->h = h; int out_w = convolutional_out_width(*l); int out_h = convolutional_out_height(*l); l->out_w = out_w; l->out_h = out_h; l->outputs = l->out_h * l->out_w * l->out_c; l->inputs = l->w * l->h * l->c; l->output = realloc(l->output, l->batch*l->outputs*sizeof(float)); l->delta = realloc(l->delta, l->batch*l->outputs*sizeof(float)); if(l->batch_normalize){ l->x = realloc(l->x, l->batch*l->outputs*sizeof(float)); l->x_norm = realloc(l->x_norm, l->batch*l->outputs*sizeof(float)); } #ifdef GPU cuda_free(l->delta_gpu); cuda_free(l->output_gpu); l->delta_gpu = cuda_make_array(l->delta, l->batch*l->outputs); l->output_gpu = cuda_make_array(l->output, l->batch*l->outputs); if(l->batch_normalize){ cuda_free(l->x_gpu); cuda_free(l->x_norm_gpu); l->x_gpu = cuda_make_array(l->output, l->batch*l->outputs); l->x_norm_gpu = cuda_make_array(l->output, l->batch*l->outputs); } #ifdef CUDNN cudnn_convolutional_setup(l); #endif #endif l->workspace_size = get_workspace_size(*l); }
convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int n, int groups, int size, int stride, int padding, ACTIVATION activation, int batch_normalize, int binary, int xnor, int adam) { int i; convolutional_layer l = {0}; l.type = CONVOLUTIONAL; l.groups = groups; l.h = h; l.w = w; l.c = c; l.n = n; l.binary = binary; l.xnor = xnor; l.batch = batch; l.stride = stride; l.size = size; l.pad = padding; l.batch_normalize = batch_normalize; l.weights = calloc(c/groups*n*size*size, sizeof(float)); l.weight_updates = calloc(c/groups*n*size*size, sizeof(float)); l.biases = calloc(n, sizeof(float)); l.bias_updates = calloc(n, sizeof(float)); l.nweights = c/groups*n*size*size; l.nbiases = n; // float scale = 1./sqrt(size*size*c); float scale = sqrt(2./(size*size*c/l.groups)); //scale = .02; //for(i = 0; i < c*n*size*size; ++i) l.weights[i] = scale*rand_uniform(-1, 1); for(i = 0; i < l.nweights; ++i) l.weights[i] = scale*rand_normal(); int out_w = convolutional_out_width(l); int out_h = convolutional_out_height(l); l.out_h = out_h; l.out_w = out_w; l.out_c = n; l.outputs = l.out_h * l.out_w * l.out_c; l.inputs = l.w * l.h * l.c; l.output = calloc(l.batch*l.outputs, sizeof(float)); l.delta = calloc(l.batch*l.outputs, sizeof(float)); l.forward = forward_convolutional_layer; l.backward = backward_convolutional_layer; l.update = update_convolutional_layer; if(binary){ l.binary_weights = calloc(l.nweights, sizeof(float)); l.cweights = calloc(l.nweights, sizeof(char)); l.scales = calloc(n, sizeof(float)); } if(xnor){ l.binary_weights = calloc(l.nweights, sizeof(float)); l.binary_input = calloc(l.inputs*l.batch, sizeof(float)); } if(batch_normalize){ l.scales = calloc(n, sizeof(float)); l.scale_updates = calloc(n, sizeof(float)); for(i = 0; i < n; ++i){ l.scales[i] = 1; } l.mean = calloc(n, sizeof(float)); l.variance = calloc(n, sizeof(float)); l.mean_delta = calloc(n, sizeof(float)); l.variance_delta = calloc(n, sizeof(float)); l.rolling_mean = calloc(n, sizeof(float)); l.rolling_variance = calloc(n, sizeof(float)); l.x = calloc(l.batch*l.outputs, sizeof(float)); l.x_norm = calloc(l.batch*l.outputs, sizeof(float)); } if(adam){ l.m = calloc(l.nweights, sizeof(float)); l.v = calloc(l.nweights, sizeof(float)); l.bias_m = calloc(n, sizeof(float)); l.scale_m = calloc(n, sizeof(float)); l.bias_v = calloc(n, sizeof(float)); l.scale_v = calloc(n, sizeof(float)); } #ifdef GPU l.forward_gpu = forward_convolutional_layer_gpu; l.backward_gpu = backward_convolutional_layer_gpu; l.update_gpu = update_convolutional_layer_gpu; if(gpu_index >= 0){ if (adam) { l.m_gpu = cuda_make_array(l.m, l.nweights); l.v_gpu = cuda_make_array(l.v, l.nweights); l.bias_m_gpu = cuda_make_array(l.bias_m, n); l.bias_v_gpu = cuda_make_array(l.bias_v, n); l.scale_m_gpu = cuda_make_array(l.scale_m, n); l.scale_v_gpu = cuda_make_array(l.scale_v, n); } l.weights_gpu = cuda_make_array(l.weights, l.nweights); l.weight_updates_gpu = cuda_make_array(l.weight_updates, l.nweights); l.biases_gpu = cuda_make_array(l.biases, n); l.bias_updates_gpu = cuda_make_array(l.bias_updates, n); l.delta_gpu = cuda_make_array(l.delta, l.batch*out_h*out_w*n); l.output_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n); if(binary){ l.binary_weights_gpu = cuda_make_array(l.weights, l.nweights); } if(xnor){ l.binary_weights_gpu = cuda_make_array(l.weights, l.nweights); l.binary_input_gpu = cuda_make_array(0, l.inputs*l.batch); } if(batch_normalize){ l.mean_gpu = cuda_make_array(l.mean, n); l.variance_gpu = cuda_make_array(l.variance, n); l.rolling_mean_gpu = cuda_make_array(l.mean, n); l.rolling_variance_gpu = cuda_make_array(l.variance, n); l.mean_delta_gpu = cuda_make_array(l.mean, n); l.variance_delta_gpu = cuda_make_array(l.variance, n); l.scales_gpu = cuda_make_array(l.scales, n); l.scale_updates_gpu = cuda_make_array(l.scale_updates, n); l.x_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n); l.x_norm_gpu = cuda_make_array(l.output, l.batch*out_h*out_w*n); } #ifdef CUDNN cudnnCreateTensorDescriptor(&l.normTensorDesc); cudnnCreateTensorDescriptor(&l.srcTensorDesc); cudnnCreateTensorDescriptor(&l.dstTensorDesc); cudnnCreateFilterDescriptor(&l.weightDesc); cudnnCreateTensorDescriptor(&l.dsrcTensorDesc); cudnnCreateTensorDescriptor(&l.ddstTensorDesc); cudnnCreateFilterDescriptor(&l.dweightDesc); cudnnCreateConvolutionDescriptor(&l.convDesc); cudnn_convolutional_setup(&l); #endif } #endif l.workspace_size = get_workspace_size(l); l.activation = activation; //fprintf(stderr, "conv %5d %2d x%2d /%2d %4d x%4d x%4d -> %4d x%4d x%4d\n", n, size, size, stride, w, h, c, l.out_w, l.out_h, l.out_c); return l; }