From eef9f8e5bb9fdbe22c6301a56f12e315f9ac64d9 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Wed, 23 May 2018 15:35:08 +0000
Subject: [PATCH] Minor fixes
---
src/batchnorm_layer.c | 42 +++++++++++++++++++++---------------------
1 files changed, 21 insertions(+), 21 deletions(-)
diff --git a/src/batchnorm_layer.c b/src/batchnorm_layer.c
index 4443291..d35d9d2 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
@@ -189,19 +189,19 @@
CUDNN_BATCHNORM_SPATIAL,
&one,
&zero,
- l.dstTensorDesc,
- l.x_gpu,
- l.dstTensorDesc,
- l.output_gpu,
+ l.normDstTensorDesc,
+ l.x_gpu, // input
+ l.normDstTensorDesc,
+ l.output_gpu, // output
l.normTensorDesc,
l.scales_gpu,
l.biases_gpu,
.01,
- l.rolling_mean_gpu,
- l.rolling_variance_gpu,
+ l.rolling_mean_gpu, // output (should be FP32)
+ l.rolling_variance_gpu, // output (should be FP32)
.00001,
- l.mean_gpu,
- l.variance_gpu);
+ l.mean_gpu, // output (should be FP32)
+ l.variance_gpu); // output (should be FP32)
#else
fast_mean_gpu(l.output_gpu, l.batch, l.out_c, l.out_h*l.out_w, l.mean_gpu);
fast_variance_gpu(l.output_gpu, l.mean_gpu, l.batch, l.out_c, l.out_h*l.out_w, l.variance_gpu);
@@ -242,19 +242,19 @@
&zero,
&one,
&one,
- l.dstTensorDesc,
- l.x_gpu,
- l.dstTensorDesc,
- l.delta_gpu,
- l.dstTensorDesc,
- l.x_norm_gpu,
+ l.normDstTensorDesc,
+ l.x_gpu, // input
+ l.normDstTensorDesc,
+ l.delta_gpu, // input
+ l.normDstTensorDesc,
+ l.x_norm_gpu, // output
l.normTensorDesc,
- l.scales_gpu,
- l.scale_updates_gpu,
- l.bias_updates_gpu,
+ l.scales_gpu, // output (should be FP32)
+ l.scale_updates_gpu, // output (should be FP32)
+ l.bias_updates_gpu, // output (should be FP32)
.00001,
- l.mean_gpu,
- l.variance_gpu);
+ l.mean_gpu, // input (should be FP32)
+ l.variance_gpu); // input (should be FP32)
copy_ongpu(l.outputs*l.batch, l.x_norm_gpu, 1, l.delta_gpu, 1);
#else
backward_bias_gpu(l.bias_updates_gpu, l.delta_gpu, l.batch, l.out_c, l.out_w*l.out_h);
--
Gitblit v1.10.0