From 1b2c70f82a9ec88c60bb277f9f5f0073c96691e7 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Tue, 03 Apr 2018 13:42:00 +0000
Subject: [PATCH] Detection is accelerated by 7 percent (fused conv and batch_norm layers)

---
 src/network_kernels.cu |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/network_kernels.cu b/src/network_kernels.cu
index 6090bb0..d6bb294 100644
--- a/src/network_kernels.cu
+++ b/src/network_kernels.cu
@@ -117,11 +117,11 @@
 	int i;
 	for (i = 0; i < net.n; ++i) {
 		layer l = net.layers[i];
-		cuda_convert_f32_to_f16(l.weights_gpu, l.c*l.n*l.size*l.size, (half *)l.weights_gpu16);
+		cuda_convert_f32_to_f16(l.weights_gpu, l.c*l.n*l.size*l.size, l.weights_gpu16);
 	}
 #endif
     forward_network_gpu(net, state);
-	cudaStreamSynchronize(get_cuda_stream());
+	//cudaStreamSynchronize(get_cuda_stream());
     backward_network_gpu(net, state);
 }
 

--
Gitblit v1.10.0