Merge branch 'master' of https://github.com/gkernel/cuda_lab

submission/ex7/3_shared_memory+constant_kernel/Makefile

+main: main.cu aux.cu aux.h Makefile
+	nvcc -o main main.cu aux.cu --ptxas-options=-v --use_fast_math --compiler-options -Wall -lopencv_highgui -lopencv_core
+

submission/ex7/3_shared_memory+constant_kernel/aux.cu

+// ###
+// ###
+// ### Practical Course: GPU Programming in Computer Vision
+// ###
+// ###
+// ### Technical University Munich, Computer Vision Group
+// ### Winter Semester 2013/2014, March 3 - April 4
+// ###
+// ###
+// ### Evgeny Strekalovskiy, Maria Klodt, Jan Stuehmer, Mohamed Souiai
+// ###
+// ###
+// ###
+// ### THIS FILE IS SUPPOSED TO REMAIN UNCHANGED
+// ###
+// ###
+
+
+#include "aux.h"
+#include <cstdlib>
+#include <iostream>
+using std::stringstream;
+using std::cerr;
+using std::cout;
+using std::endl;
+using std::string;
+
+
+
+
+// parameter processing: template specialization for T=bool
+template<>
+bool getParam<bool>(std::string param, bool &var, int argc, char **argv)
+{
+    const char *c_param = param.c_str();
+    for(int i=argc-1; i>=1; i--)
+    {
+        if (argv[i][0]!='-') continue;
+        if (strcmp(argv[i]+1, c_param)==0)
+        {
+            if (!(i+1<argc) || argv[i+1][0]=='-') { var = true; return true; }
+            std::stringstream ss;
+            ss << argv[i+1];
+            ss >> var;
+            return (bool)ss;
+        }
+    }
+    return false;
+}
+
+
+
+
+// opencv helpers
+void convert_layered_to_interleaved(float *aOut, const float *aIn, int w, int h, int nc)
+{
+    if (nc==1) { memcpy(aOut, aIn, w*h*sizeof(float)); return; }
+    size_t nOmega = (size_t)w*h;
+    for (int y=0; y<h; y++)
+    {
+        for (int x=0; x<w; x++)
+        {
+            for (int c=0; c<nc; c++)
+            {
+                aOut[(nc-1-c) + nc*(x + (size_t)w*y)] = aIn[x + (size_t)w*y + nOmega*c];
+            }
+        }
+    }
+}
+void convert_layered_to_mat(cv::Mat &mOut, const float *aIn)
+{
+    convert_layered_to_interleaved((float*)mOut.data, aIn, mOut.cols, mOut.rows, mOut.channels());
+}
+
+
+void convert_interleaved_to_layered(float *aOut, const float *aIn, int w, int h, int nc)
+{
+    if (nc==1) { memcpy(aOut, aIn, w*h*sizeof(float)); return; }
+    size_t nOmega = (size_t)w*h;
+    for (int y=0; y<h; y++)
+    {
+        for (int x=0; x<w; x++)
+        {
+            for (int c=0; c<nc; c++)
+            {
+                aOut[x + (size_t)w*y + nOmega*c] = aIn[(nc-1-c) + nc*(x + (size_t)w*y)];
+            }
+        }
+    }
+}
+void convert_mat_to_layered(float *aOut, const cv::Mat &mIn)
+{
+    convert_interleaved_to_layered(aOut, (float*)mIn.data, mIn.cols, mIn.rows, mIn.channels());
+}
+
+
+
+void showImage(string title, const cv::Mat &mat, int x, int y)
+{
+    const char *wTitle = title.c_str();
+    cv::namedWindow(wTitle, CV_WINDOW_AUTOSIZE);
+    cvMoveWindow(wTitle, x, y);
+    cv::imshow(wTitle, mat);
+}
+
+
+
+
+// adding Gaussian noise
+float noise(float sigma)
+{
+    float x1 = (float)rand()/RAND_MAX;
+    float x2 = (float)rand()/RAND_MAX;
+    return sigma * sqrtf(-2*log(std::max(x1,0.000001f)))*cosf(2*M_PI*x2);
+}
+void addNoise(cv::Mat &m, float sigma)
+{
+    float *data = (float*)m.data;
+    int w = m.cols;
+    int h = m.rows;
+    int nc = m.channels();
+    size_t n = (size_t)w*h*nc;
+    for(size_t i=0; i<n; i++)
+    {
+        data[i] += noise(sigma);
+    }
+}
+
+
+
+
+// cuda error checking
+string prev_file = "";
+int prev_line = 0;
+void cuda_check(string file, int line)
+{
+    cudaError_t e = cudaGetLastError();
+    if (e != cudaSuccess)
+    {
+        cout << endl << file << ", line " << line << ": " << cudaGetErrorString(e) << " (" << e << ")" << endl;
+        if (prev_line>0) cout << "Previous CUDA call:" << endl << prev_file << ", line " << prev_line << endl;
+        exit(1);
+    }
+    prev_file = file;
+    prev_line = line;
+}

submission/ex7/3_shared_memory+constant_kernel/aux.h

+// ###
+// ###
+// ### Practical Course: GPU Programming in Computer Vision
+// ###
+// ###
+// ### Technical University Munich, Computer Vision Group
+// ### Winter Semester 2013/2014, March 3 - April 4
+// ###
+// ###
+// ### Evgeny Strekalovskiy, Maria Klodt, Jan Stuehmer, Mohamed Souiai
+// ###
+// ###
+// ###
+// ### THIS FILE IS SUPPOSED TO REMAIN UNCHANGED
+// ###
+// ###
+
+
+#ifndef AUX_H
+#define AUX_H
+
+#include <cuda_runtime.h>
+#include <ctime>
+#include <opencv2/highgui/highgui.hpp>
+#include <opencv2/imgproc/imgproc.hpp>
+#include <string>
+#include <sstream>
+
+
+
+
+// parameter processing
+template<typename T>
+bool getParam(std::string param, T &var, int argc, char **argv)
+{
+    const char *c_param = param.c_str();
+    for(int i=argc-1; i>=1; i--)
+    {
+        if (argv[i][0]!='-') continue;
+        if (strcmp(argv[i]+1, c_param)==0)
+        {
+            if (!(i+1<argc)) continue;
+            std::stringstream ss;
+            ss << argv[i+1];
+            ss >> var;
+            return (bool)ss;
+        }
+    }
+    return false;
+}
+
+
+
+
+// opencv helpers
+void convert_mat_to_layered(float *aOut, const cv::Mat &mIn);
+void convert_layered_to_mat(cv::Mat &mOut, const float *aIn);
+void showImage(std::string title, const cv::Mat &mat, int x, int y);
+
+
+
+
+// adding Gaussian noise
+void addNoise(cv::Mat &m, float sigma);
+
+
+
+
+// measuring time
+class Timer
+{
+    public:
+	Timer() : tStart(0), running(false), sec(0.f)
+	{
+	}
+	void start()
+	{
+		tStart = clock();
+		running = true;
+	}
+	void end()
+	{
+		if (!running) { sec = 0; return; }
+        cudaDeviceSynchronize();
+		clock_t tEnd = clock();
+		sec = (float)(tEnd - tStart) / CLOCKS_PER_SEC;
+		running = false;
+	}
+	float get()
+	{
+		if (running) end();
+		return sec;
+	}
+    private:
+	clock_t tStart;
+	bool running;
+	float sec;
+};
+
+
+
+
+// cuda error checking
+#define CUDA_CHECK cuda_check(__FILE__,__LINE__)
+void cuda_check(std::string file, int line);
+
+
+
+#endif  // AUX_H

submission/ex7/3_texture+constant_kernel/Makefile

+all: main
+
+main: main.cu aux.cu aux.h Makefile
+	nvcc -o main main.cu aux.cu --ptxas-options=-v --use_fast_math --compiler-options -Wall -lopencv_highgui -lopencv_core
+

submission/ex7/3_texture+constant_kernel/aux.cu

+// ###
+// ###
+// ### Practical Course: GPU Programming in Computer Vision
+// ###
+// ###
+// ### Technical University Munich, Computer Vision Group
+// ### Winter Semester 2013/2014, March 3 - April 4
+// ###
+// ###
+// ### Evgeny Strekalovskiy, Maria Klodt, Jan Stuehmer, Mohamed Souiai
+// ###
+// ###
+// ###
+// ### THIS FILE IS SUPPOSED TO REMAIN UNCHANGED
+// ###
+// ###
+
+
+#include "aux.h"
+#include <cstdlib>
+#include <iostream>
+using std::stringstream;
+using std::cerr;
+using std::cout;
+using std::endl;
+using std::string;
+
+
+
+
+// parameter processing: template specialization for T=bool
+template<>
+bool getParam<bool>(std::string param, bool &var, int argc, char **argv)
+{
+    const char *c_param = param.c_str();
+    for(int i=argc-1; i>=1; i--)
+    {
+        if (argv[i][0]!='-') continue;
+        if (strcmp(argv[i]+1, c_param)==0)
+        {
+            if (!(i+1<argc) || argv[i+1][0]=='-') { var = true; return true; }
+            std::stringstream ss;
+            ss << argv[i+1];
+            ss >> var;
+            return (bool)ss;
+        }
+    }
+    return false;
+}
+
+
+
+
+// opencv helpers
+void convert_layered_to_interleaved(float *aOut, const float *aIn, int w, int h, int nc)
+{
+    if (nc==1) { memcpy(aOut, aIn, w*h*sizeof(float)); return; }
+    size_t nOmega = (size_t)w*h;
+    for (int y=0; y<h; y++)
+    {
+        for (int x=0; x<w; x++)
+        {
+            for (int c=0; c<nc; c++)
+            {
+                aOut[(nc-1-c) + nc*(x + (size_t)w*y)] = aIn[x + (size_t)w*y + nOmega*c];
+            }
+        }
+    }
+}
+void convert_layered_to_mat(cv::Mat &mOut, const float *aIn)
+{
+    convert_layered_to_interleaved((float*)mOut.data, aIn, mOut.cols, mOut.rows, mOut.channels());
+}
+
+
+void convert_interleaved_to_layered(float *aOut, const float *aIn, int w, int h, int nc)
+{
+    if (nc==1) { memcpy(aOut, aIn, w*h*sizeof(float)); return; }
+    size_t nOmega = (size_t)w*h;
+    for (int y=0; y<h; y++)
+    {
+        for (int x=0; x<w; x++)
+        {
+            for (int c=0; c<nc; c++)
+            {
+                aOut[x + (size_t)w*y + nOmega*c] = aIn[(nc-1-c) + nc*(x + (size_t)w*y)];
+            }
+        }
+    }
+}
+void convert_mat_to_layered(float *aOut, const cv::Mat &mIn)
+{
+    convert_interleaved_to_layered(aOut, (float*)mIn.data, mIn.cols, mIn.rows, mIn.channels());
+}
+
+
+
+void showImage(string title, const cv::Mat &mat, int x, int y)
+{
+    const char *wTitle = title.c_str();
+    cv::namedWindow(wTitle, CV_WINDOW_AUTOSIZE);
+    cvMoveWindow(wTitle, x, y);
+    cv::imshow(wTitle, mat);
+}
+
+
+
+
+// adding Gaussian noise
+float noise(float sigma)
+{
+    float x1 = (float)rand()/RAND_MAX;
+    float x2 = (float)rand()/RAND_MAX;
+    return sigma * sqrtf(-2*log(std::max(x1,0.000001f)))*cosf(2*M_PI*x2);
+}
+void addNoise(cv::Mat &m, float sigma)
+{
+    float *data = (float*)m.data;
+    int w = m.cols;
+    int h = m.rows;
+    int nc = m.channels();
+    size_t n = (size_t)w*h*nc;
+    for(size_t i=0; i<n; i++)
+    {
+        data[i] += noise(sigma);
+    }
+}
+
+
+
+
+// cuda error checking
+string prev_file = "";
+int prev_line = 0;
+void cuda_check(string file, int line)
+{
+    cudaError_t e = cudaGetLastError();
+    if (e != cudaSuccess)
+    {
+        cout << endl << file << ", line " << line << ": " << cudaGetErrorString(e) << " (" << e << ")" << endl;
+        if (prev_line>0) cout << "Previous CUDA call:" << endl << prev_file << ", line " << prev_line << endl;
+        exit(1);
+    }
+    prev_file = file;
+    prev_line = line;
+}

submission/ex7/3_texture+constant_kernel/aux.h

+// ###
+// ###
+// ### Practical Course: GPU Programming in Computer Vision
+// ###
+// ###
+// ### Technical University Munich, Computer Vision Group
+// ### Winter Semester 2013/2014, March 3 - April 4
+// ###
+// ###
+// ### Evgeny Strekalovskiy, Maria Klodt, Jan Stuehmer, Mohamed Souiai
+// ###
+// ###
+// ###
+// ### THIS FILE IS SUPPOSED TO REMAIN UNCHANGED
+// ###
+// ###
+
+
+#ifndef AUX_H
+#define AUX_H
+
+#include <cuda_runtime.h>
+#include <ctime>
+#include <opencv2/highgui/highgui.hpp>
+#include <opencv2/imgproc/imgproc.hpp>
+#include <string>
+#include <sstream>
+
+
+
+
+// parameter processing
+template<typename T>
+bool getParam(std::string param, T &var, int argc, char **argv)
+{
+    const char *c_param = param.c_str();
+    for(int i=argc-1; i>=1; i--)
+    {
+        if (argv[i][0]!='-') continue;
+        if (strcmp(argv[i]+1, c_param)==0)
+        {
+            if (!(i+1<argc)) continue;
+            std::stringstream ss;
+            ss << argv[i+1];
+            ss >> var;
+            return (bool)ss;
+        }
+    }
+    return false;
+}
+
+
+
+
+// opencv helpers
+void convert_mat_to_layered(float *aOut, const cv::Mat &mIn);
+void convert_layered_to_mat(cv::Mat &mOut, const float *aIn);
+void showImage(std::string title, const cv::Mat &mat, int x, int y);
+
+
+
+
+// adding Gaussian noise
+void addNoise(cv::Mat &m, float sigma);
+
+
+
+
+// measuring time
+class Timer
+{
+    public:
+	Timer() : tStart(0), running(false), sec(0.f)
+	{
+	}
+	void start()
+	{
+		tStart = clock();
+		running = true;
+	}
+	void end()
+	{
+		if (!running) { sec = 0; return; }
+        cudaDeviceSynchronize();
+		clock_t tEnd = clock();
+		sec = (float)(tEnd - tStart) / CLOCKS_PER_SEC;
+		running = false;
+	}
+	float get()
+	{
+		if (running) end();
+		return sec;
+	}
+    private:
+	clock_t tStart;
+	bool running;
+	float sec;
+};
+
+
+
+
+// cuda error checking
+#define CUDA_CHECK cuda_check(__FILE__,__LINE__)
+void cuda_check(std::string file, int line);
+
+
+
+#endif  // AUX_H

submission/ex7/todo

-Gaurav - 
-global memory + constant kernel (with miklos's base code)
-texture + constant kernel (with existing texture code)