From aea3bceeb16e553ff75a2f28c7f44f04b81513d7 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 08 Dec 2014 20:11:04 +0000
Subject: [PATCH] timing code
---
src/dropout_layer.c | 46 +++++++++++++++++++++++++++++++++++++++++++---
1 files changed, 43 insertions(+), 3 deletions(-)
diff --git a/src/dropout_layer.c b/src/dropout_layer.c
index fcad7b9..ad13034 100644
--- a/src/dropout_layer.c
+++ b/src/dropout_layer.c
@@ -1,6 +1,7 @@
#include "dropout_layer.h"
-#include "stdlib.h"
-#include "stdio.h"
+#include "utils.h"
+#include <stdlib.h>
+#include <stdio.h>
dropout_layer *make_dropout_layer(int batch, int inputs, float probability)
{
@@ -9,6 +10,10 @@
layer->probability = probability;
layer->inputs = inputs;
layer->batch = batch;
+ #ifdef GPU
+ layer->rand = calloc(inputs*batch, sizeof(float));
+ layer->rand_cl = cl_make_array(layer->rand, inputs*batch);
+ #endif
return layer;
}
@@ -16,7 +21,7 @@
{
int i;
for(i = 0; i < layer.batch * layer.inputs; ++i){
- if((float)rand()/RAND_MAX < layer.probability) input[i] = 0;
+ if(rand_uniform() < layer.probability) input[i] = 0;
else input[i] /= (1-layer.probability);
}
}
@@ -24,3 +29,38 @@
{
// Don't do shit LULZ
}
+
+#ifdef GPU
+cl_kernel get_dropout_kernel()
+{
+ static int init = 0;
+ static cl_kernel kernel;
+ if(!init){
+ kernel = get_kernel("src/dropout_layer.cl", "forward", 0);
+ init = 1;
+ }
+ return kernel;
+}
+
+void forward_dropout_layer_gpu(dropout_layer layer, cl_mem input)
+{
+ int j;
+ int size = layer.inputs*layer.batch;
+ for(j = 0; j < size; ++j) layer.rand[j] = rand_uniform();
+ cl_write_array(layer.rand_cl, layer.rand, layer.inputs*layer.batch);
+
+ cl_kernel kernel = get_dropout_kernel();
+ cl_command_queue queue = cl.queue;
+
+ cl_uint i = 0;
+ cl.error = clSetKernelArg(kernel, i++, sizeof(input), (void*) &input);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(layer.rand_cl), (void*) &layer.rand_cl);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(layer.probability), (void*) &layer.probability);
+ check_error(cl);
+
+ const size_t global_size[] = {size};
+
+ cl.error = clEnqueueNDRangeKernel(queue, kernel, 1, 0, global_size, 0, 0, 0, 0);
+ check_error(cl);
+}
+#endif
--
Gitblit v1.10.0