csresnext50-panet-spp works with TensorRT

Signed-off-by: Francesco Gatti <gattifrancesco@hotmail.it>

include/tkDNN/Layer.h

@@ -321,7 +321,7 @@ public:
     int winH, winW;
     int strideH, strideW;
     int paddingH, paddingW;
-    bool test;
+    bool maxpoolfixedsize;
     tkdnnPoolingMode_t pool_mode;
 
     Pooling(Network *net, int winH, int winW, 
@@ -474,7 +474,7 @@ public:
 
     dnnType *predictions;
 
-    static const int MAX_DETECTIONS = 4096;
+    static const int MAX_DETECTIONS = 1024;
     static Yolo::detection *allocateDetections(int nboxes, int classes);
     static void             mergeDetections(Yolo::detection *dets, int ndets, int classes);
 };

include/tkDNN/NetworkRT.h

@@ -36,6 +36,7 @@ using namespace nvinfer1;
 #include "pluginsRT/DeformableConvRT.h"
 #include "pluginsRT/FlattenConcatRT.h"
 #include "pluginsRT/ReshapeRT.h"
+#include "pluginsRT/MaxPoolingFixedSizeRT.h"
 
 class PluginFactory : IPluginFactory
 {

include/tkDNN/pluginsRT/MaxPoolingFixedSizeRT.h

@@ -0,0 +1,74 @@
+#include<cassert>
+#include "../kernels.h"
+
+class MaxPoolFixedSizeRT : public IPlugin {
+
+public:
+	MaxPoolFixedSizeRT(int c, int h, int w, int n, int strideH, int strideW, int winSize, int padding) {
+		this->c = c;	
+		this->h = h;
+		this->w = w;
+		this->n = n;
+		this->stride_H = strideH;
+		this->stride_W = strideW;
+		this->winSize = winSize;
+		this->padding = padding;
+	}
+
+	~MaxPoolFixedSizeRT(){
+	}
+
+	int getNbOutputs() const override {
+		return 1;
+	}
+
+	Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override {
+		return DimsCHW{this->c, this->h, this->w};
+	}
+
+	void configure(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, int maxBatchSize) override {        
+	}
+
+	int initialize() override {
+		return 0;
+	}
+
+	virtual void terminate() override {
+	}
+
+	virtual size_t getWorkspaceSize(int maxBatchSize) const override {
+		return 0;
+	}
+
+	virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override {
+
+		std::cout<<this->n<<"  "<<this->c<<"  "<<this->h<<"  "<<this->w<<"  "<<this->stride_H<<"  "<<this->stride_W<<"  "<<this->winSize<<"  "<<this->padding<<std::endl;
+		dnnType *srcData = (dnnType*)reinterpret_cast<const dnnType*>(inputs[0]);
+		dnnType *dstData = reinterpret_cast<dnnType*>(outputs[0]);
+		MaxPoolingForward(srcData, dstData, this->n, this->c, this->h, this->w, this->stride_H, this->stride_W, this->winSize, this->padding);
+		return 0;
+	}
+
+
+	virtual size_t getSerializationSize() override {
+		return 8*sizeof(int);
+	}
+
+	virtual void serialize(void* buffer) override {
+		char *buf = reinterpret_cast<char*>(buffer);
+
+		tk::dnn::writeBUF(buf, this->c);
+		tk::dnn::writeBUF(buf, this->h);
+		tk::dnn::writeBUF(buf, this->w);
+		tk::dnn::writeBUF(buf, this->n);
+		tk::dnn::writeBUF(buf, this->stride_H);
+		tk::dnn::writeBUF(buf, this->stride_W);
+		tk::dnn::writeBUF(buf, this->winSize);
+		tk::dnn::writeBUF(buf, this->padding);
+	}
+
+	int n, c, h, w;
+	int stride_H, stride_W;
+	int winSize;
+	int padding;
+};

include/tkDNN/pluginsRT/ShortcutRT.h

@@ -4,7 +4,10 @@
 class ShortcutRT : public IPlugin {
 
 public:
-	ShortcutRT() {
+	ShortcutRT(tk::dnn::dataDim_t bdim) {
+		this->bc = bdim.c;
+		this->bh = bdim.h;
+		this->bw = bdim.w;
 	}
 
 	~ShortcutRT(){
@@ -44,22 +47,27 @@ public:
 		dnnType *dstData = reinterpret_cast<dnnType*>(outputs[0]);
 
 		checkCuda( cudaMemcpyAsync(dstData, srcData, batchSize*c*h*w*sizeof(dnnType), cudaMemcpyDeviceToDevice, stream));
-		shortcutForward(srcDataBack, dstData, batchSize, c, h, w, 1, batchSize, c, h, w, 1, stream);
+		shortcutForward(srcDataBack, dstData, batchSize, c, h, w, 1, batchSize, bc, bh, bw, 1, stream);
 
 		return 0;
 	}
 
 
 	virtual size_t getSerializationSize() override {
-		return 3*sizeof(int);
+		return 6*sizeof(int);
 	}
 
 	virtual void serialize(void* buffer) override {
 		char *buf = reinterpret_cast<char*>(buffer);
+		tk::dnn::writeBUF(buf, bc);
+		tk::dnn::writeBUF(buf, bh);
+		tk::dnn::writeBUF(buf, bw);
 		tk::dnn::writeBUF(buf, c);
 		tk::dnn::writeBUF(buf, h);
 		tk::dnn::writeBUF(buf, w);
+		
 	}
 
 	int c, h, w;
+	int bc, bh, bw;
 };