From 75e3936aa1ed86cc44eedbd891a270c0c3c9c328 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Thu, 19 Oct 2017 20:22:43 +0000
Subject: [PATCH] Fixed Makefile for the CUDA 9.0
---
src/batchnorm_layer.c | 50 +++++++++++++++++++++++++++++++++++++++++++-------
1 files changed, 43 insertions(+), 7 deletions(-)
diff --git a/src/batchnorm_layer.c b/src/batchnorm_layer.c
index 6ea4040..b53548b 100644
--- a/src/batchnorm_layer.c
+++ b/src/batchnorm_layer.c
@@ -28,7 +28,13 @@
layer.rolling_mean = calloc(c, sizeof(float));
layer.rolling_variance = calloc(c, sizeof(float));
+
+ layer.forward = forward_batchnorm_layer;
+ layer.backward = backward_batchnorm_layer;
#ifdef GPU
+ layer.forward_gpu = forward_batchnorm_layer_gpu;
+ layer.backward_gpu = backward_batchnorm_layer_gpu;
+
layer.output_gpu = cuda_make_array(layer.output, h * w * c * batch);
layer.delta_gpu = cuda_make_array(layer.delta, h * w * c * batch);
@@ -121,20 +127,50 @@
l.out_h = l.out_w = 1;
}
if(state.train){
- mean_cpu(l.output, l.batch, l.out_c, l.out_h*l.out_w, l.mean);
- variance_cpu(l.output, l.mean, l.batch, l.out_c, l.out_h*l.out_w, l.variance);
+ mean_cpu(l.output, l.batch, l.out_c, l.out_h*l.out_w, l.mean);
+ variance_cpu(l.output, l.mean, l.batch, l.out_c, l.out_h*l.out_w, l.variance);
+
+ scal_cpu(l.out_c, .9, l.rolling_mean, 1);
+ axpy_cpu(l.out_c, .1, l.mean, 1, l.rolling_mean, 1);
+ scal_cpu(l.out_c, .9, l.rolling_variance, 1);
+ axpy_cpu(l.out_c, .1, l.variance, 1, l.rolling_variance, 1);
+
+ copy_cpu(l.outputs*l.batch, l.output, 1, l.x, 1);
normalize_cpu(l.output, l.mean, l.variance, l.batch, l.out_c, l.out_h*l.out_w);
+ copy_cpu(l.outputs*l.batch, l.output, 1, l.x_norm, 1);
} else {
normalize_cpu(l.output, l.rolling_mean, l.rolling_variance, l.batch, l.out_c, l.out_h*l.out_w);
}
scale_bias(l.output, l.scales, l.batch, l.out_c, l.out_h*l.out_w);
}
-void backward_batchnorm_layer(const layer layer, network_state state)
+void backward_batchnorm_layer(const layer l, network_state state)
{
+ backward_scale_cpu(l.x_norm, l.delta, l.batch, l.out_c, l.out_w*l.out_h, l.scale_updates);
+
+ scale_bias(l.delta, l.scales, l.batch, l.out_c, l.out_h*l.out_w);
+
+ mean_delta_cpu(l.delta, l.variance, l.batch, l.out_c, l.out_w*l.out_h, l.mean_delta);
+ variance_delta_cpu(l.x, l.delta, l.mean, l.variance, l.batch, l.out_c, l.out_w*l.out_h, l.variance_delta);
+ normalize_delta_cpu(l.x, l.mean, l.variance, l.mean_delta, l.variance_delta, l.batch, l.out_c, l.out_w*l.out_h, l.delta);
+ if(l.type == BATCHNORM) copy_cpu(l.outputs*l.batch, l.delta, 1, state.delta, 1);
}
#ifdef GPU
+
+void pull_batchnorm_layer(layer l)
+{
+ cuda_pull_array(l.scales_gpu, l.scales, l.c);
+ cuda_pull_array(l.rolling_mean_gpu, l.rolling_mean, l.c);
+ cuda_pull_array(l.rolling_variance_gpu, l.rolling_variance, l.c);
+}
+void push_batchnorm_layer(layer l)
+{
+ cuda_push_array(l.scales_gpu, l.scales, l.c);
+ cuda_push_array(l.rolling_mean_gpu, l.rolling_mean, l.c);
+ cuda_push_array(l.rolling_variance_gpu, l.rolling_variance, l.c);
+}
+
void forward_batchnorm_layer_gpu(layer l, network_state state)
{
if(l.type == BATCHNORM) copy_ongpu(l.outputs*l.batch, state.input, 1, l.output_gpu, 1);
@@ -146,10 +182,10 @@
fast_mean_gpu(l.output_gpu, l.batch, l.out_c, l.out_h*l.out_w, l.mean_gpu);
fast_variance_gpu(l.output_gpu, l.mean_gpu, l.batch, l.out_c, l.out_h*l.out_w, l.variance_gpu);
- scal_ongpu(l.out_c, .95, l.rolling_mean_gpu, 1);
- axpy_ongpu(l.out_c, .05, l.mean_gpu, 1, l.rolling_mean_gpu, 1);
- scal_ongpu(l.out_c, .95, l.rolling_variance_gpu, 1);
- axpy_ongpu(l.out_c, .05, l.variance_gpu, 1, l.rolling_variance_gpu, 1);
+ scal_ongpu(l.out_c, .99, l.rolling_mean_gpu, 1);
+ axpy_ongpu(l.out_c, .01, l.mean_gpu, 1, l.rolling_mean_gpu, 1);
+ scal_ongpu(l.out_c, .99, l.rolling_variance_gpu, 1);
+ axpy_ongpu(l.out_c, .01, l.variance_gpu, 1, l.rolling_variance_gpu, 1);
copy_ongpu(l.outputs*l.batch, l.output_gpu, 1, l.x_gpu, 1);
normalize_gpu(l.output_gpu, l.mean_gpu, l.variance_gpu, l.batch, l.out_c, l.out_h*l.out_w);
--
Gitblit v1.10.0