From e4ab47dfcedb4c87e5eddf484caa4ac0c020fc9b Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Wed, 21 Feb 2018 12:35:09 +0000
Subject: [PATCH] Optimized resizing of region_layer for random=1
---
src/convolutional_layer.c | 115 +++++++++++++++++++++++++--------------------------------
1 files changed, 50 insertions(+), 65 deletions(-)
diff --git a/src/convolutional_layer.c b/src/convolutional_layer.c
index 3864c1b..ca83486 100644
--- a/src/convolutional_layer.c
+++ b/src/convolutional_layer.c
@@ -8,6 +8,10 @@
#include <stdio.h>
#include <time.h>
+#ifdef CUDNN
+#pragma comment(lib, "cudnn.lib")
+#endif
+
#ifdef AI2
#include "xnor_layer.h"
#endif
@@ -142,8 +146,12 @@
cudnnSetTensor4dDescriptor(l->srcTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->c, l->h, l->w);
cudnnSetTensor4dDescriptor(l->dstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, l->batch, l->out_c, l->out_h, l->out_w);
cudnnSetFilter4dDescriptor(l->weightDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c, l->size, l->size);
- cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION);
- cudnnGetConvolutionForwardAlgorithm(cudnn_handle(),
+#if(CUDNN_MAJOR >= 6)
+ cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION, CUDNN_DATA_FLOAT); // cudnn 6.0
+#else
+ cudnnSetConvolution2dDescriptor(l->convDesc, l->pad, l->pad, l->stride, l->stride, 1, 1, CUDNN_CROSS_CORRELATION); // cudnn 5.1
+#endif
+ cudnnGetConvolutionForwardAlgorithm(cudnn_handle(),
l->srcTensorDesc,
l->weightDesc,
l->convDesc,
@@ -206,8 +214,8 @@
l.outputs = l.out_h * l.out_w * l.out_c;
l.inputs = l.w * l.h * l.c;
- l.output = calloc(l.batch*out_h * out_w * n, sizeof(float));
- l.delta = calloc(l.batch*out_h * out_w * n, sizeof(float));
+ 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;
@@ -232,8 +240,13 @@
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.adam = 1;
@@ -346,6 +359,8 @@
void resize_convolutional_layer(convolutional_layer *l, int w, int h)
{
+ int old_w = l->w;
+ int old_h = l->h;
l->w = w;
l->h = h;
int out_w = convolutional_out_width(*l);
@@ -357,25 +372,29 @@
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*out_h * out_w * l->n*sizeof(float));
- l->delta = realloc(l->delta,
- l->batch*out_h * out_w * l->n*sizeof(float));
+ 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);
+ if (old_w < w || old_h < h) {
+ cuda_free(l->delta_gpu);
+ cuda_free(l->output_gpu);
- l->delta_gpu = cuda_make_array(l->delta, l->batch*out_h*out_w*l->n);
- l->output_gpu = cuda_make_array(l->output, l->batch*out_h*out_w*l->n);
+ 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);
+ 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);
- }
+ 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
@@ -423,41 +442,8 @@
int out_w = convolutional_out_width(l);
int i;
-
fill_cpu(l.outputs*l.batch, 0, l.output, 1);
- /*
- if(l.binary){
- binarize_weights(l.weights, l.n, l.c*l.size*l.size, l.binary_weights);
- binarize_weights2(l.weights, l.n, l.c*l.size*l.size, l.cweights, l.scales);
- swap_binary(&l);
- }
- */
-
- /*
- if(l.binary){
- int m = l.n;
- int k = l.size*l.size*l.c;
- int n = out_h*out_w;
-
- char *a = l.cweights;
- float *b = state.workspace;
- float *c = l.output;
-
- for(i = 0; i < l.batch; ++i){
- im2col_cpu(state.input, l.c, l.h, l.w,
- l.size, l.stride, l.pad, b);
- gemm_bin(m,n,k,1,a,k,b,n,c,n);
- c += n*m;
- state.input += l.c*l.h*l.w;
- }
- scale_bias(l.output, l.scales, l.batch, l.n, out_h*out_w);
- add_bias(l.output, l.biases, l.batch, l.n, out_h*out_w);
- activate_array(l.output, m*n*l.batch, l.activation);
- return;
- }
- */
-
if(l.xnor){
binarize_weights(l.weights, l.n, l.c*l.size*l.size, l.binary_weights);
swap_binary(&l);
@@ -469,22 +455,17 @@
int k = l.size*l.size*l.c;
int n = out_h*out_w;
- if (l.xnor && l.c%32 == 0 && AI2) {
- forward_xnor_layer(l, state);
- printf("xnor\n");
- } else {
- float *a = l.weights;
- float *b = state.workspace;
- float *c = l.output;
+ float *a = l.weights;
+ float *b = state.workspace;
+ float *c = l.output;
- for(i = 0; i < l.batch; ++i){
- im2col_cpu(state.input, l.c, l.h, l.w,
- l.size, l.stride, l.pad, b);
- gemm(0,0,m,n,k,1,a,k,b,n,1,c,n);
- c += n*m;
- state.input += l.c*l.h*l.w;
- }
+ for(i = 0; i < l.batch; ++i){
+ im2col_cpu(state.input, l.c, l.h, l.w,
+ l.size, l.stride, l.pad, b);
+ gemm(0,0,m,n,k,1,a,k,b,n,1,c,n);
+ c += n*m;
+ state.input += l.c*l.h*l.w;
}
if(l.batch_normalize){
@@ -507,6 +488,10 @@
gradient_array(l.output, m*k*l.batch, l.activation, l.delta);
backward_bias(l.bias_updates, l.delta, l.batch, l.n, k);
+ if(l.batch_normalize){
+ backward_batchnorm_layer(l, state);
+ }
+
for(i = 0; i < l.batch; ++i){
float *a = l.delta + i*m*k;
float *b = state.workspace;
--
Gitblit v1.10.0