From 9bae70b22549b68f5cdeece8b6c3b3de00c22714 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Mon, 16 Apr 2018 23:51:11 +0000
Subject: [PATCH] Accelerated by another 5% using FP16/32 Batch-norm for Tensor Cores.
---
src/layer.h | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/src/layer.h b/src/layer.h
index 75c0358..81e27ad 100644
--- a/src/layer.h
+++ b/src/layer.h
@@ -281,7 +281,7 @@
#ifdef CUDNN
cudnnTensorDescriptor_t srcTensorDesc, dstTensorDesc;
cudnnTensorDescriptor_t dsrcTensorDesc, ddstTensorDesc;
- cudnnTensorDescriptor_t normTensorDesc, normDstTensorDesc;
+ cudnnTensorDescriptor_t normTensorDesc, normDstTensorDesc, normDstTensorDescF16;
cudnnFilterDescriptor_t weightDesc;
cudnnFilterDescriptor_t dweightDesc;
cudnnConvolutionDescriptor_t convDesc;
--
Gitblit v1.10.0