From 537d135feba179636e9bbfe296e078d51f59914c Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Mon, 19 Mar 2018 23:16:51 +0000
Subject: [PATCH] Improve training performance - batch-norm using cuDNN.
---
src/region_layer.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/src/region_layer.c b/src/region_layer.c
index 9ca71c6..f179906 100644
--- a/src/region_layer.c
+++ b/src/region_layer.c
@@ -434,7 +434,7 @@
cuda_pull_array(state.truth, truth_cpu, num_truth);
}
cuda_pull_array(l.output_gpu, in_cpu, l.batch*l.inputs);
- cudaStreamSynchronize(get_cuda_stream());
+ //cudaStreamSynchronize(get_cuda_stream());
network_state cpu_state = state;
cpu_state.train = state.train;
cpu_state.truth = truth_cpu;
@@ -444,7 +444,7 @@
free(cpu_state.input);
if(!state.train) return;
cuda_push_array(l.delta_gpu, l.delta, l.batch*l.outputs);
- cudaStreamSynchronize(get_cuda_stream());
+ //cudaStreamSynchronize(get_cuda_stream());
if(cpu_state.truth) free(cpu_state.truth);
}
--
Gitblit v1.10.0