From 0f645836f193e75c4c3b718369e6fab15b5d19c5 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Wed, 11 Feb 2015 03:41:03 +0000
Subject: [PATCH] Detection is back, baby\!
---
src/dropout_layer.c | 78 ++++++++++-----------------------------
1 files changed, 20 insertions(+), 58 deletions(-)
diff --git a/src/dropout_layer.c b/src/dropout_layer.c
index d4616d5..32a3408 100644
--- a/src/dropout_layer.c
+++ b/src/dropout_layer.c
@@ -1,5 +1,6 @@
#include "dropout_layer.h"
#include "utils.h"
+#include "cuda.h"
#include <stdlib.h>
#include <stdio.h>
@@ -10,28 +11,44 @@
layer->probability = probability;
layer->inputs = inputs;
layer->batch = batch;
+ layer->output = calloc(inputs*batch, sizeof(float));
layer->rand = calloc(inputs*batch, sizeof(float));
layer->scale = 1./(1.-probability);
#ifdef GPU
- layer->rand_cl = cl_make_array(layer->rand, inputs*batch);
+ layer->output_gpu = cuda_make_array(layer->output, inputs*batch);
+ layer->rand_gpu = cuda_make_array(layer->rand, inputs*batch);
#endif
return layer;
}
+void resize_dropout_layer(dropout_layer *layer, int inputs)
+{
+ layer->output = realloc(layer->output, layer->inputs*layer->batch*sizeof(float));
+ layer->rand = realloc(layer->rand, layer->inputs*layer->batch*sizeof(float));
+ #ifdef GPU
+ cuda_free(layer->output_gpu);
+ cuda_free(layer->rand_gpu);
+
+ layer->output_gpu = cuda_make_array(layer->output, inputs*layer->batch);
+ layer->rand_gpu = cuda_make_array(layer->rand, inputs*layer->batch);
+ #endif
+}
+
void forward_dropout_layer(dropout_layer layer, float *input)
{
int i;
for(i = 0; i < layer.batch * layer.inputs; ++i){
float r = rand_uniform();
layer.rand[i] = r;
- if(r < layer.probability) input[i] = 0;
- else input[i] *= layer.scale;
+ if(r < layer.probability) layer.output[i] = 0;
+ else layer.output[i] = input[i]*layer.scale;
}
}
void backward_dropout_layer(dropout_layer layer, float *delta)
{
int i;
+ if(!delta) return;
for(i = 0; i < layer.batch * layer.inputs; ++i){
float r = layer.rand[i];
if(r < layer.probability) delta[i] = 0;
@@ -39,58 +56,3 @@
}
}
-#ifdef GPU
-cl_kernel get_dropout_kernel()
-{
- static int init = 0;
- static cl_kernel kernel;
- if(!init){
- kernel = get_kernel("src/dropout_layer.cl", "yoloswag420blazeit360noscope", 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);
- cl.error = clSetKernelArg(kernel, i++, sizeof(layer.scale), (void*) &layer.scale);
- 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);
-}
-
-void backward_dropout_layer_gpu(dropout_layer layer, cl_mem delta)
-{
- int size = 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(delta), (void*) &delta);
- cl.error = clSetKernelArg(kernel, i++, sizeof(layer.rand_cl), (void*) &layer.rand_cl);
- cl.error = clSetKernelArg(kernel, i++, sizeof(layer.probability), (void*) &layer.probability);
- cl.error = clSetKernelArg(kernel, i++, sizeof(layer.scale), (void*) &layer.scale);
- 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