From 6f0b2b195e4fde12fc6316f837f63c7db6174dc6 Mon Sep 17 00:00:00 2001
From: Zhang Ting <zhangting_2017@163.com>
Date: Fri, 1 Nov 2019 23:00:51 +0800
Subject: [PATCH] [cherry-pick] fix the bug of conv_transpose cudnn kernel,
 test=release/1.6 (#20958) (#20974)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

fix the bug of conv_transpose cudnn kernel：cherry-pick #20958
---
 paddle/fluid/operators/conv_transpose_cudnn_op.cu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/paddle/fluid/operators/conv_transpose_cudnn_op.cu b/paddle/fluid/operators/conv_transpose_cudnn_op.cu
index 15e8f38312..d6c7f964c6 100644
--- a/paddle/fluid/operators/conv_transpose_cudnn_op.cu
+++ b/paddle/fluid/operators/conv_transpose_cudnn_op.cu
@@ -72,7 +72,7 @@ class CUDNNConvTransposeOpKernel : public framework::OpKernel<T> {
     const T* filter_data = filter->data<T>();
     const std::string data_layout_str = ctx.Attr<std::string>("data_format");
     const paddle::operators::DataLayout data_layout =
-        (data_layout_str == "NCHW" ? DataLayout::kNCHW : DataLayout::kNHWC);
+        (data_layout_str != "NHWC" ? DataLayout::kNCHW : DataLayout::kNHWC);
 
     // if channel_last, transpose to channel_first
     Tensor input_transpose;
-- 
GitLab