From 2b2441313b73c460a60c013c3b7bf9e19c994b6b Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Thu, 30 Oct 2014 18:28:37 +0000
Subject: [PATCH] col2im maybe a little faster
---
src/connected_layer.c | 14 ++++++++------
1 files changed, 8 insertions(+), 6 deletions(-)
diff --git a/src/connected_layer.c b/src/connected_layer.c
index b41ae91..ac4c417 100644
--- a/src/connected_layer.c
+++ b/src/connected_layer.c
@@ -114,6 +114,12 @@
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);
@@ -129,9 +135,7 @@
{
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;
@@ -148,9 +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