From 2594f59d0d20460a3ddd9aa3737e725f8ac3ed8a Mon Sep 17 00:00:00 2001
From: Francesco Gatti <gattifrancesco@hotmail.it>
Date: Wed, 30 Oct 2019 16:45:46 +0100
Subject: [PATCH] conv2d ok, but deconv ha different dim with tensorrt

---
 include/tkDNN/Layer.h        |  2 +-
 include/tkDNN/NetworkRT.h    |  1 -
 src/Conv2d.cpp               | 12 ++++++------
 tests/simple/test_simple.cpp |  2 ++
 4 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/include/tkDNN/Layer.h b/include/tkDNN/Layer.h
index 6015235..31dc4d2 100644
--- a/include/tkDNN/Layer.h
+++ b/include/tkDNN/Layer.h
@@ -166,7 +166,7 @@ public:
 protected:
     cudnnFilterDescriptor_t filterDesc;
     cudnnConvolutionDescriptor_t convDesc;
-    cudnnConvolutionFwdAlgo_t     fwAlgo;
+    cudnnConvolutionFwdAlgo_t     algo;
     cudnnConvolutionBwdDataAlgo_t bwAlgo;
     cudnnTensorDescriptor_t biasTensorDesc;
 
diff --git a/include/tkDNN/NetworkRT.h b/include/tkDNN/NetworkRT.h
index 97dd478..d30da03 100644
--- a/include/tkDNN/NetworkRT.h
+++ b/include/tkDNN/NetworkRT.h
@@ -75,7 +75,6 @@ public:
 
     nvinfer1::ILayer* convert_layer(nvinfer1::ITensor *input, Layer *l);
     nvinfer1::ILayer* convert_layer(nvinfer1::ITensor *input, Conv2d *l);
-    nvinfer1::ILayer* convert_layer(nvinfer1::ITensor *input, DeConv2d *l);
     nvinfer1::ILayer* convert_layer(nvinfer1::ITensor *input, Activation *l);
     nvinfer1::ILayer* convert_layer(nvinfer1::ITensor *input, Dense *l);
     nvinfer1::ILayer* convert_layer(nvinfer1::ITensor *input, Pooling *l);
diff --git a/src/Conv2d.cpp b/src/Conv2d.cpp
index da1a4a2..cc4fbd5 100644
--- a/src/Conv2d.cpp
+++ b/src/Conv2d.cpp
@@ -73,10 +73,10 @@ void Conv2d::initCUDNN(bool back) {
     } else {
         checkCUDNN( cudnnGetConvolutionForwardAlgorithm(net->cudnnHandle,
                                                         srcTensor, filterDesc, convDesc, dstTensor,
-                                                        CUDNN_CONVOLUTION_FWD_PREFER_FASTEST, 0, &fwAlgo) );
+                                                        CUDNN_CONVOLUTION_FWD_PREFER_FASTEST, 0, &algo) );
         checkCUDNN(cudnnGetConvolutionForwardWorkspaceSize(net->cudnnHandle,
                                                            srcTensor, filterDesc, convDesc, dstTensor,
-                                                           fwAlgo, &ws_sizeInBytes));
+                                                           algo, &ws_sizeInBytes));
     }
 }
 
@@ -93,14 +93,14 @@ void Conv2d::inferCUDNN(dnnType* srcData, bool back) {
     } else {
         checkCUDNN(cudnnConvolutionForward(net->cudnnHandle,
                                            &alpha, srcTensorDesc, srcData, filterDesc,
-                                           data_d, convDesc, fwAlgo, workSpace, ws_sizeInBytes,
+                                           data_d, convDesc, algo, workSpace, ws_sizeInBytes,
                                            &beta, dstTensorDesc, dstData));
     }
 
     if(!batchnorm) {
         // bias
         alpha = dnnType(1);
-        beta  = dnnType(0);
+        beta  = dnnType(1);
         checkCUDNN( cudnnAddTensor(net->cudnnHandle,
                                    &alpha, biasTensorDesc, bias_d,
                                    &beta, dstTensorDesc, dstData) );
@@ -116,7 +116,7 @@ void Conv2d::inferCUDNN(dnnType* srcData, bool back) {
     }
 }
 
-Conv2d::Conv2d( Network *net, int out_ch, int kernelH, int kernelW, 
+Conv2d::Conv2d( Network *net, int out_ch, int kernelH, int kernelW,
                 int strideH, int strideW, int paddingH, int paddingW,
                 std::string fname_weights, bool batchnorm, bool deConv) :
     
@@ -156,7 +156,7 @@ Conv2d::Conv2d( Network *net, int out_ch, int kernelH, int kernelW,
 }
 
 Conv2d::~Conv2d() {
-    
+
     checkCUDNN( cudnnDestroyFilterDescriptor(filterDesc) );
     checkCUDNN( cudnnDestroyConvolutionDescriptor(convDesc) );
     checkCUDNN( cudnnDestroyTensorDescriptor(biasTensorDesc) );
diff --git a/tests/simple/test_simple.cpp b/tests/simple/test_simple.cpp
index d331b96..5634e4a 100644
--- a/tests/simple/test_simple.cpp
+++ b/tests/simple/test_simple.cpp
@@ -19,6 +19,8 @@ int main() {
     tk::dnn::Dense      l5(&net, 4, d2_bin);
     tk::dnn::Activation l6(&net, CUDNN_ACTIVATION_RELU);
 
+    net.print();
+
     // Load input
     dnnType *data;
     dnnType *input_h;