From 6e1d5b45de988bb795c4c505f22f2170a78b7746 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Tue, 20 Jan 2015 06:06:18 +0000
Subject: [PATCH] fast sort of working
---
src/convolutional_layer.c | 39 ++++++++++++++++++++++++++++++++++-----
1 files changed, 34 insertions(+), 5 deletions(-)
diff --git a/src/convolutional_layer.c b/src/convolutional_layer.c
index 18d00e6..4e8c44b 100644
--- a/src/convolutional_layer.c
+++ b/src/convolutional_layer.c
@@ -170,7 +170,9 @@
int n = layer.size*layer.size*layer.c;
int k = convolutional_out_height(layer)*
convolutional_out_width(layer);
+
gradient_array(layer.output, m*k*layer.batch, layer.activation, layer.delta);
+
learn_bias_convolutional_layer(layer);
if(delta) memset(delta, 0, layer.batch*layer.h*layer.w*layer.c*sizeof(float));
@@ -264,13 +266,18 @@
}
#ifdef GPU
+#define BLOCK 32
+
+#define STR_HELPER(x) #x
+#define STR(x) STR_HELPER(x)
+
cl_kernel get_convolutional_learn_bias_kernel()
{
static int init = 0;
static cl_kernel kernel;
if(!init){
- kernel = get_kernel("src/convolutional_layer.cl", "learn_bias", 0);
+ kernel = get_kernel("src/convolutional_layer.cl", "learn_bias", "-D BLOCK=" STR(BLOCK));
init = 1;
}
return kernel;
@@ -291,18 +298,40 @@
cl.error = clSetKernelArg(kernel, i++, sizeof(layer.bias_updates_cl), (void*) &layer.bias_updates_cl);
check_error(cl);
- const size_t global_size[] = {layer.n};
+ const size_t global_size[] = {layer.n*BLOCK};
+ const size_t local_size[] = {BLOCK};
- cl.error = clEnqueueNDRangeKernel(queue, kernel, 1, 0, global_size, 0, 0, 0, 0);
+ cl.error = clEnqueueNDRangeKernel(queue, kernel, 1, 0, global_size, local_size, 0, 0, 0);
check_error(cl);
}
+void test_learn_bias(convolutional_layer l)
+{
+ int i;
+ int size = convolutional_out_height(l) * convolutional_out_width(l);
+ for(i = 0; i < size*l.batch*l.n; ++i){
+ l.delta[i] = rand_uniform();
+ }
+ for(i = 0; i < l.n; ++i){
+ l.bias_updates[i] = rand_uniform();
+ }
+ cl_write_array(l.delta_cl, l.delta, size*l.batch*l.n);
+ cl_write_array(l.bias_updates_cl, l.bias_updates, l.n);
+ float *gpu = calloc(l.n, sizeof(float));
+ cl_read_array(l.bias_updates_cl, gpu, l.n);
+ for(i = 0; i < l.n; ++i) printf("%.9g %.9g\n", l.bias_updates[i], gpu[i]);
+ learn_bias_convolutional_layer_ongpu(l);
+ learn_bias_convolutional_layer(l);
+ cl_read_array(l.bias_updates_cl, gpu, l.n);
+ for(i = 0; i < l.n; ++i) printf("%.9g %.9g\n", l.bias_updates[i], gpu[i]);
+}
+
cl_kernel get_convolutional_bias_kernel()
{
static int init = 0;
static cl_kernel kernel;
if(!init){
- kernel = get_kernel("src/convolutional_layer.cl", "bias", 0);
+ kernel = get_kernel("src/convolutional_layer.cl", "bias", "-D BLOCK=" STR(BLOCK));
init = 1;
}
return kernel;
@@ -410,7 +439,7 @@
axpy_ongpu(size, -layer.decay, layer.filters_cl, 1, layer.filter_updates_cl, 1);
axpy_ongpu(size, layer.learning_rate, layer.filter_updates_cl, 1, layer.filters_cl, 1);
scal_ongpu(size, layer.momentum, layer.filter_updates_cl, 1);
- pull_convolutional_layer(layer);
+ //pull_convolutional_layer(layer);
}
--
Gitblit v1.10.0