From a45d17ffe3f6f3435228319171b2efda7484f01d Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 20 Jun 2016 05:37:21 +0000
Subject: [PATCH] yolo demo low latency
---
src/rnn_layer.c | 8 ++++----
1 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/src/rnn_layer.c b/src/rnn_layer.c
index 384169a..b713899 100644
--- a/src/rnn_layer.c
+++ b/src/rnn_layer.c
@@ -10,7 +10,7 @@
#include <stdlib.h>
#include <string.h>
-void increment_layer(layer *l, int steps)
+static void increment_layer(layer *l, int steps)
{
int num = l->outputs*l->batch*steps;
l->output += num;
@@ -242,8 +242,6 @@
increment_layer(&output_layer, l.steps - 1);
l.state_gpu += l.hidden*l.batch*l.steps;
for (i = l.steps-1; i >= 0; --i) {
- copy_ongpu(l.hidden * l.batch, input_layer.output_gpu, 1, l.state_gpu, 1);
- axpy_ongpu(l.hidden * l.batch, 1, self_layer.output_gpu, 1, l.state_gpu, 1);
s.input = l.state_gpu;
s.delta = self_layer.delta_gpu;
@@ -251,12 +249,14 @@
l.state_gpu -= l.hidden*l.batch;
+ copy_ongpu(l.hidden*l.batch, self_layer.delta_gpu, 1, input_layer.delta_gpu, 1);
+
s.input = l.state_gpu;
s.delta = self_layer.delta_gpu - l.hidden*l.batch;
if (i == 0) s.delta = 0;
backward_connected_layer_gpu(self_layer, s);
- copy_ongpu(l.hidden*l.batch, self_layer.delta_gpu, 1, input_layer.delta_gpu, 1);
+ //copy_ongpu(l.hidden*l.batch, self_layer.delta_gpu, 1, input_layer.delta_gpu, 1);
if (i > 0 && l.shortcut) axpy_ongpu(l.hidden*l.batch, 1, self_layer.delta_gpu, 1, self_layer.delta_gpu - l.hidden*l.batch, 1);
s.input = state.input + i*l.inputs*l.batch;
if(state.delta) s.delta = state.delta + i*l.inputs*l.batch;
--
Gitblit v1.10.0