From eb9c88ef734d693e65ec35036811363a35e6b5d3 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Sat, 14 Apr 2018 22:51:21 +0000
Subject: [PATCH] Fixed bug in Tensor Cores V100 (1. Desc in Batch norm, 2. Manually selected algo). Also fixed time measure on Linux for multi-threading.
---
src/batchnorm_layer.c | 14 +++++++-------
1 files changed, 7 insertions(+), 7 deletions(-)
diff --git a/src/batchnorm_layer.c b/src/batchnorm_layer.c
index 4443291..883ab34 100644
--- a/src/batchnorm_layer.c
+++ b/src/batchnorm_layer.c
@@ -54,8 +54,8 @@
layer.x_norm_gpu = cuda_make_array(layer.output, layer.batch*layer.outputs);
#ifdef CUDNN
cudnnCreateTensorDescriptor(&layer.normTensorDesc);
- cudnnCreateTensorDescriptor(&layer.dstTensorDesc);
- cudnnSetTensor4dDescriptor(layer.dstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, layer.batch, layer.out_c, layer.out_h, layer.out_w);
+ cudnnCreateTensorDescriptor(&layer.normDstTensorDesc);
+ cudnnSetTensor4dDescriptor(layer.normDstTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, layer.batch, layer.out_c, layer.out_h, layer.out_w);
cudnnSetTensor4dDescriptor(layer.normTensorDesc, CUDNN_TENSOR_NCHW, CUDNN_DATA_FLOAT, 1, layer.out_c, 1, 1);
#endif
#endif
@@ -189,9 +189,9 @@
CUDNN_BATCHNORM_SPATIAL,
&one,
&zero,
- l.dstTensorDesc,
+ l.normDstTensorDesc,
l.x_gpu,
- l.dstTensorDesc,
+ l.normDstTensorDesc,
l.output_gpu,
l.normTensorDesc,
l.scales_gpu,
@@ -242,11 +242,11 @@
&zero,
&one,
&one,
- l.dstTensorDesc,
+ l.normDstTensorDesc,
l.x_gpu,
- l.dstTensorDesc,
+ l.normDstTensorDesc,
l.delta_gpu,
- l.dstTensorDesc,
+ l.normDstTensorDesc,
l.x_norm_gpu,
l.normTensorDesc,
l.scales_gpu,
--
Gitblit v1.10.0