From 1c0fd9bb4726f28b5ccf4491b8d108b00c884ec3 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Thu, 30 Oct 2014 06:26:41 +0000
Subject: [PATCH] im2col slightly faster
---
src/connected_layer.c | 21 +++++++++++++++------
1 files changed, 15 insertions(+), 6 deletions(-)
diff --git a/src/connected_layer.c b/src/connected_layer.c
index ba83dc3..ac4c417 100644
--- a/src/connected_layer.c
+++ b/src/connected_layer.c
@@ -108,6 +108,18 @@
#ifdef GPU
+void pull_connected_layer(connected_layer layer)
+{
+ cl_read_array(layer.weights_cl, layer.weights, layer.inputs*layer.outputs);
+ cl_read_array(layer.biases_cl, layer.biases, layer.outputs);
+}
+
+void push_connected_layer(connected_layer layer)
+{
+ cl_write_array(layer.weights_cl, layer.weights, layer.inputs*layer.outputs);
+ cl_write_array(layer.biases_cl, layer.biases, layer.outputs);
+}
+
void update_connected_layer_gpu(connected_layer layer)
{
axpy_ongpu(layer.outputs, layer.learning_rate, layer.bias_updates_cl, 1, layer.biases_cl, 1);
@@ -116,15 +128,14 @@
scal_ongpu(layer.inputs*layer.outputs, 1.-layer.learning_rate*layer.decay, layer.weights_cl, 1);
axpy_ongpu(layer.inputs*layer.outputs, layer.learning_rate, layer.weight_updates_cl, 1, layer.weights_cl, 1);
scal_ongpu(layer.inputs*layer.outputs, layer.momentum, layer.weight_updates_cl, 1);
+ pull_connected_layer(layer);
}
void forward_connected_layer_gpu(connected_layer layer, cl_mem input)
{
int i;
for(i = 0; i < layer.batch; ++i){
- cl_mem sub = cl_sub_array(layer.output_cl, i*layer.outputs, layer.outputs);
- copy_ongpu(layer.outputs, layer.biases_cl, 1, sub, 1);
- clReleaseMemObject(sub);
+ copy_ongpu_offset(layer.outputs, layer.biases_cl, 0, 1, layer.output_cl, i*layer.outputs, 1);
}
int m = layer.batch;
int k = layer.inputs;
@@ -141,9 +152,7 @@
int i;
gradient_array_ongpu(layer.output_cl, layer.outputs*layer.batch, layer.activation, layer.delta_cl);
for(i = 0; i < layer.batch; ++i){
- cl_mem sub = cl_sub_array(layer.delta_cl, i*layer.outputs, layer.outputs);
- axpy_ongpu(layer.outputs, 1, sub, 1, layer.bias_updates_cl, 1);
- clReleaseMemObject(sub);
+ axpy_ongpu_offset(layer.outputs, 1, layer.delta_cl, i*layer.outputs, 1, layer.bias_updates_cl, 0, 1);
}
int m = layer.inputs;
int k = layer.batch;
--
Gitblit v1.10.0