Added initial stup for Cuda Confidence Maps Pipeline.

modules/cudaconfidencemaps/core/cm.cpp

+#include <chrono>
+#include <iostream>
+#include <iomanip>
+
+#include "cm.h"
+#include "cm_cusp.h"
+
+void measureExecution(std::string message, const std::function<void()> &function)
+{
+	std::cout << "started:  " << message << std::endl;
+
+	auto start = std::chrono::high_resolution_clock::now();
+
+	function();
+
+	auto stop = std::chrono::high_resolution_clock::now();
+
+	float milliseconds = std::chrono::duration_cast<std::chrono::microseconds>(stop-start).count() / 1000.0f;
+	std::cout << "finished: " << std::left << std::setw(35) << message << "["
+	          << milliseconds << "ms]" << std::endl;
+}
+
+std::unique_ptr<ConfidenceMapSolver> createCUSPSolver(int width, int height, float gradientScaling,
+                                                      float alpha, float beta, float gamma)
+{
+	return std::make_unique<CuspConfidenceMapSolver>(width, height, gradientScaling, alpha, beta, gamma);
+}
\ No newline at end of file

modules/cudaconfidencemaps/core/cm.h

+#pragma once
+
+typedef unsigned char uint8;
+
+#include <memory>
+#include <vector>
+#include <functional>
+
+void measureExecution(std::string message, const std::function<void()> &function);
+
+class ConfidenceMapSolver
+{
+public:
+	virtual ~ConfidenceMapSolver() {}
+	virtual void createSystem(const uint8* image, int width, int height) = 0;
+	virtual void setInitialSolution(const std::vector<float> &val) = 0;
+	virtual void solve() = 0;
+
+	virtual const float* getSolution(int &width, int &height) const = 0;
+};
+
+std::unique_ptr<ConfidenceMapSolver> createCUSPSolver(int width, int height, float gradientScaling,
+                                                      float alpha, float beta, float gamma);
+/*std::unique_ptr<ConfidenceMapSolver> createEIGENSolver(int width, int height, float gradientScaling,
+                                                       float alpha, float beta, float gamma);*/
\ No newline at end of file

modules/cudaconfidencemaps/core/cm_cusp.cpp

+#include "cm_cusp.h"
+#include "cm_cusp_cuda_exports.h"
+
+CuspConfidenceMapSolver::CuspConfidenceMapSolver(int width, int height, float gradientScaling,
+	                                             float alpha, float beta, float gamma)
+	: width(width), height(height), gradientScaling(gradientScaling),
+	  alpha(alpha), beta(beta), gamma(gamma), gpuData(CUSP_CM_createGpuData(width, height)) { }
+
+CuspConfidenceMapSolver::~CuspConfidenceMapSolver()
+{
+	CUSP_CM_destroyGpuData(gpuData);
+}
+
+void CuspConfidenceMapSolver::createSystem(const uint8* image, int width, int height)
+{
+	if (width != this->width || height != this->height) {
+		CUSP_CM_resizeGpuData(*gpuData, width, height);
+	}
+
+	this->width = width;
+	this->height = height;
+
+	CUSP_CM_buildEquationSystem(*gpuData, image, width, height, alpha, beta, gamma, gradientScaling);
+}
+
+void CuspConfidenceMapSolver::setInitialSolution(const std::vector<float> &val)
+{
+	CUSP_CM_setInitialSolution(*gpuData, val);
+}
+
+void CuspConfidenceMapSolver::solve()
+{
+	measureExecution(" . uploading system", [=]() {
+		CUSP_CM_uploadSystem(*gpuData);
+	});
+	measureExecution(" . solving system", [=]() {
+		CUSP_CM_solveSystem(*gpuData, 100, 1e-20);
+	});
+	measureExecution(" . downloading solution", [=]() {
+		CUSP_CM_downloadSolution(*gpuData);
+	});
+}
+
+const float* CuspConfidenceMapSolver::getSolution(int &width, int &height) const
+{
+	const float *ptr = CUSP_CM_getSolutionPtr(*gpuData);
+	width = width; height = height; // FIXME: height and width are not actually checked against the size of the buffer...
+	return ptr;
+}

modules/cudaconfidencemaps/core/cm_cusp.cu

+#include <vector>
+#include <cmath>
+#include <cusp/dia_matrix.h>
+#include <cusp/krylov/cg.h>
+#include <cusp/precond/diagonal.h>
+
+#include "cm_cusp_cuda_exports.h"
+
+typedef cusp::dia_matrix<int, float, cusp::host_memory> CuspHostDiaMatrix;
+typedef cusp::dia_matrix<int, float, cusp::device_memory> CuspDeviceDiaMatrix;
+typedef cusp::array1d<float, cusp::host_memory> CuspHostVector;
+typedef cusp::array1d<float, cusp::device_memory> CuspDeviceVector;
+
+class CuspGPUData {
+public:
+	CuspGPUData(int width, int height)
+		: L_h(width*height, width*height, width*height*9, 9),
+	      b_h(width*height), x_h(width*height),
+		  L_d(width*height, width*height, width*height*9, 9),
+	      b_d(width*height), x_d(width*height) {};
+
+	void resize(int width, int height)
+	{
+		this->L_h.resize(width*height, width*height, width*height*9, 9);
+		this->b_h.resize(width*height);
+		this->x_h.resize(width*height);
+	}
+
+	CuspHostDiaMatrix L_h;
+	CuspHostVector b_h;
+	CuspHostVector x_h;
+
+	CuspDeviceDiaMatrix L_d;
+	CuspDeviceVector b_d;
+	CuspDeviceVector x_d;
+};
+
+CuspGPUData* CUSP_CM_createGpuData(int width, int height)
+{
+	CuspGPUData *data = new CuspGPUData(width, height);
+	return data;
+}
+
+void CUSP_CM_destroyGpuData(CuspGPUData *data)
+{
+	delete data;
+}
+
+void CUSP_CM_resizeGpuData(CuspGPUData& data, int width, int height)
+{
+	data.resize(width, height);
+}
+
+void CUSP_CM_setInitialSolution(CuspGPUData& data, const std::vector<float>& values)
+{
+	for (int i = 0; i < data.x_h.size() && i < values.size(); ++i) {
+		data.x_h[i] = values[i];
+	}
+}
+
+void CUSP_CM_uploadSystem(CuspGPUData &data)
+{
+	// Check for NaN values
+	/*
+	for (int r = 0; r < data.x_h.size(); ++r) {
+		for (int d = 0; d < 9; ++d) {
+			if (std::isnan(data.L_h.values(r, d))) {
+				std::cout << "NaN detected in A: " << r << std::endl;
+				break;
+			}
+		}
+
+		if (std::isnan(data.x_h[r])) {
+			std::cout << "NaN detected in x: " << r << std::endl;
+		}
+
+		if (std::isnan(data.b_h[r])) {
+			std::cout << "NaN detected in b: " << r << std::endl;
+		}
+	}*/
+
+	data.x_d = data.x_h;
+	data.b_d = data.b_h;
+	data.L_d = data.L_h;
+}
+
+void CUSP_CM_downloadSolution(CuspGPUData &data)
+{
+	data.x_h = data.x_d;
+	// FIXME: Should not be needed... (Clamp solution)
+	for (int i = 0; i < data.x_h.size(); ++i) {
+		data.x_h[i] = min(1.0f, data.x_h[i]);
+	}
+}
+
+void CUSP_CM_solveSystem(CuspGPUData& data, int iterations, float precision)
+{
+	cusp::default_monitor<float> monitor(data.b_d, iterations, precision);
+	//cusp::precond::bridson_ainv<float, cusp::device_memory> M(A, 1);
+	cusp::precond::diagonal<float, cusp::device_memory> M(data.L_d);
+	//cusp::krylov::bicgstab(A, x, b, monitor);
+	cusp::krylov::cg(data.L_d, data.x_d, data.b_d, monitor, M);
+}
+
+const float* CUSP_CM_getSolutionPtr(const CuspGPUData& data)
+{
+	//const_cast<CuspGPUData&>(data).x_h = const_cast<CuspGPUData&>(data).x_d;
+	return &data.x_h[0];
+}
+
+
+
+
+struct ComputeLaplacianData
+{
+	float alpha, beta, gamma;
+	float gradientScaling;
+	const unsigned char *image;
+	int width, height;
+	int centralDiagonal;
+	int offsets[9];
+	float gammaList[9];
+};
+
+
+static inline float _getAttenuation(int y, int height, float alpha) {
+	return (1 - exp(-alpha * ((float)y / (float)(height-1))));
+}
+
+static inline float _getGradient(const ComputeLaplacianData &data, int idx, int offset)
+{
+	const unsigned char *image = data.image;
+	int y1 = idx / data.width;
+	int y2 = (idx+offset) / data.width;
+
+	float a1 = _getAttenuation(y1, data.height, data.alpha);
+	float a2 = _getAttenuation(y2, data.height, data.alpha);
+
+	return abs(image[idx]*a1/255.0f - image[idx+offset]*a2/255.0f);
+}
+
+static inline float _calculateWeight(float gradient, float beta, float gamma, float scaling)
+{
+	return exp(-beta * (gradient*scaling + gamma));
+}
+
+static inline float _getWeight(const ComputeLaplacianData &data, int x, int y, int diagonal)
+{
+	float gradient = _getGradient(data, y * data.width + x, data.offsets[diagonal]);
+	float weight = _calculateWeight(gradient, data.beta, data.gammaList[diagonal], data.gradientScaling);
+	return weight + 1e-4;
+}
+
+void CUSP_CM_buildEquationSystem(CuspGPUData &gpudata, const unsigned char* image, int width, int height,
+                                 float alpha, float beta, float gamma,
+                                 float gradientScaling)
+{
+	// Gather all of the options together
+	ComputeLaplacianData data;
+	data.alpha = alpha; data.beta = beta; data.gamma = gamma;
+	data.gradientScaling = 1.0f / 0.1764f;
+	data.image = image;
+	data.width = width; data.height = height;
+	data.centralDiagonal = 4;
+	int offsets[9] = {-width-1, -width, -width+1, -1, 0, 1, width-1, width, width+1};
+	float gammaList[9] = {sqrt(2.0f)*gamma, 0.0f, sqrt(2.0f)*gamma, gamma, 0.0f, gamma, sqrt(2.0f)*gamma, 0.0f, sqrt(2.0f)*gamma};
+	for (int i = 0; i < 9; ++i) {
+		data.offsets[i] = offsets[i];
+		data.gammaList[i] = gammaList[i];
+	}
+
+	// Prepare equation system data structure
+	for (int i = 0; i < 9; ++i) {
+		gpudata.L_h.diagonal_offsets[i] = data.offsets[i];
+	}
+
+	// Reset B
+	for (int x = 0; x < width * height; ++x) {
+		gpudata.b_h[x] = 0.0f;
+	}
+
+	// Fill in first row of b, which sholud be one
+	for (int x = 0; x < width; ++x) {
+		gpudata.b_h[x] = 1.0f;
+	}
+
+	// Fill in rows of A corresponding to first and last row of the image
+	for (int x = 0; x < width; ++x) {
+		for (int d = 0; d < 9; ++d) {
+			gpudata.L_h.values(x, d) = (d == data.centralDiagonal ? 1.0f : 0.0f);
+			gpudata.L_h.values(width*(height-1) + x, d) = (d == data.centralDiagonal ? 1.0f : 0.0f);
+		}
+	}
+
+	// Fill in the rest of the matrix
+	for (int y = 1; y < height-1; ++y) {
+		for (int x = 0; x < width; ++x) {
+			int idx = y * width + x;
+
+			// Filter off out-of-bounds edges
+			unsigned short filter = 495; // 111 101 111
+
+			if (x == 0)        filter &= 203; // 011 001 011
+			if (x == width-1)  filter &= 422; // 110 100 110
+			if (y == 1)        filter &=  47; // 000 101 111
+			if (y == height-2) filter &= 488; // 111 101 000
+
+			float valueSum = 0.0f;
+
+			for (int d = 0; d < 9; ++d) {
+				gpudata.L_h.values(idx, d) = 0;
+
+				float value = 0.0f;
+				
+				if (((256>>d) & filter) != 0) {
+					value = _getWeight(data, x, y, d);
+					gpudata.L_h.values(idx, d) = -value;
+				} else if(y == 1 || y == height - 2) {
+					unsigned short filter2 = 495; // 111 101 11
+					if (x == 0)        filter2 &= 203; // 011 001 011
+					if (x == width-1)  filter2 &= 422; // 110 100 110
+					if (((256>>d) & filter2) != 0) {
+						value = _getWeight(data, x, y, d);
+
+						if (y == 1) gpudata.b_h[idx] += value;
+					}
+				}
+
+				valueSum += value;
+			}
+
+			gpudata.L_h.values(idx, data.centralDiagonal) = valueSum;
+		}
+	}
+}
\ No newline at end of file

modules/cudaconfidencemaps/core/cm_cusp.h

+#pragma once
+
+#include "cm.h"
+
+class CuspGPUData;
+
+/* Laplacian construction parameters */
+class CuspConfidenceMapSolver : public ConfidenceMapSolver
+{
+public:
+	CuspConfidenceMapSolver(int width, int height, float gradientScaling,
+	                        float alpha, float beta, float gamma);
+
+	virtual ~CuspConfidenceMapSolver();
+
+	virtual void createSystem(const uint8* image, int width, int height);
+	virtual void setInitialSolution(const std::vector<float> &val);
+	virtual void solve();
+
+	virtual const float* getSolution(int &width, int &height) const;
+
+private:
+	// Input image data
+	int width, height;
+
+	// Solver parameters
+	float alpha, beta, gamma;
+	float gradientScaling;
+
+	// Matrices and Vectors
+	CuspGPUData* gpuData;
+};
\ No newline at end of file

modules/cudaconfidencemaps/core/cm_cusp_build_system.cu

+
+/*<
+
+__global__ void k_buildCMSystem(cudaTextureObject_t image, int width, int height,
+                                float alpha, float beta, float gamma,
+                                float *out_weights)
+{
+	int x = blockIdx.x * blockDim.x + threadIdx.x;
+	int y = blockIdx.y * blockDim.y + threadIdx.y;
+	int idx = y * width + x;
+
+	if (x >= width || y >= height) return;
+
+	float attenuation_center = exp(-alpha * (float)y / height);
+	float attenuation_bottom = exp(-alpha * (float)(y+1) / height);
+
+	float c_center = tex2D(image, x, y)     * attenuation_center;
+	float c_right  = tex2D(image, x+1, y)   * attenuation_center;
+	float c_bottom = tex2D(image, x, y+1)   * attenuation_bottom;
+	float c_se     = tex2D(image, x+1, y+1) * attenuation_bottom;
+	float c_sw     = tex2D(image, x-1, y+1) * attenuation_bottom;
+
+	float w_right  = exp(-beta * (abs(c_center - c_right) + gamma));
+	float w_bottom = exp(-beta * (abs(c_center - c_bottom)));
+	float w_se     = exp(-beta * (abs(c_center - c_se) + sqrt(2.0f) * gamma));
+	float w_sw     = exp(-beta * (abs(c_center - c_sw) + sqrt(2.0f) * gamma));
+
+	out_weights[idx*4 + 0] = w_right;
+	out_weights[idx*4 + 1] = w_sw;
+	out_weights[idx*4 + 2] = w_bottom;
+	out_weights[idx*4 + 3] = w_se;
+}
+
+
+int main()
+{
+	// Create texture object
+	cudaResourceDesc resDesc;
+	memset(&resDesc, 0, sizeof(resDesc));
+	resDesc.resType = cudaResourceTypePitch2D
+
+
+	return 0;
+}*/
\ No newline at end of file

modules/cudaconfidencemaps/core/cm_cusp_cuda_exports.h

+#pragma once
+
+#include <vector>
+
+class CuspGPUData;
+
+CuspGPUData* CUSP_CM_createGpuData(int width, int height);
+void CUSP_CM_destroyGpuData(CuspGPUData *data);
+void CUSP_CM_resizeGpuData(CuspGPUData& data, int width, int height);
+void CUSP_CM_buildEquationSystem(CuspGPUData& data, const unsigned char* image, int width, int height,
+                                 float alpha, float beta, float gamma,
+                                 float gradientScaling);
+void CUSP_CM_setInitialSolution(CuspGPUData& data, const std::vector<float>& values);
+void CUSP_CM_uploadSystem(CuspGPUData &data);
+void CUSP_CM_downloadSolution(CuspGPUData &data);
+void CUSP_CM_solveSystem(CuspGPUData& data, int iterations, float precision);
+const float* CUSP_CM_getSolutionPtr(const CuspGPUData& data);
\ No newline at end of file

modules/cudaconfidencemaps/cudaconfidencemaps.cmake

+# CMake file for vis module
+
+IF(${ModuleEnabled})
+	# Find CUDA
+    FIND_PACKAGE(CUDA REQUIRED)
+
+    if(CUDA_FOUND)
+		# Source files:
+		FILE(GLOB ThisModSources RELATIVE ${ModulesDir}
+			modules/cudaconfidencemaps/pipelines/*.cpp
+			modules/cudaconfidencemaps/processors/*.cpp
+		)
+
+		# Header files (including GLSL files so that they'll appear in VS projects)
+		FILE(GLOB ThisModHeaders RELATIVE ${ModulesDir}
+			modules/cudaconfidencemaps/glsl/*.frag
+			modules/cudaconfidencemaps/glsl/*.geom
+			modules/cudaconfidencemaps/glsl/*.vert
+			modules/cudaconfidencemaps/pipelines/*.h
+			modules/cudaconfidencemaps/processors/*.h
+		)
+
+		SET(ThisModShaderDirectories "modules/cudaconfidencemaps/glsl")
+		SET(ThisModDependencies base io)
+	else()
+		MESSAGE(FATAL_ERROR "Could not find CUDA SDK.")
+	endif()
+ENDIF(${ModuleEnabled})
+
+SET(ThisModStatus EXPERIMENTAL)
+SET(ThisModExternalDependencies TRUE)

modules/cudaconfidencemaps/ext/cusplibrary-0.4.0/.gitignore

+syntax: glob
+
+# Ignore files left by editors
+*~
+*.bak
+*.swp
+
+# Ignore compiled files
+*.o
+*.obj
+*.exe
+*.dll
+*.pyc
+*.so
+*.pdb
+
+# Ignore CUDA auxiliary files
+*.linkinfo
+*.cubin
+
+# Other things to ignore
+*.vtu
+*.dblite
+nvcc_options_file.txt
+print_sm_version
+tester
+
+# Ignore build files
+#<AutoIgnore>
+examples/Solvers/gmres
+examples/Solvers/bicgstab
+examples/Solvers/cg_m
+examples/Solvers/cg
+examples/InputOutput/matrix_market
+examples/MatrixFormats/csr
+examples/MatrixFormats/ell
+examples/MatrixFormats/hyb
+examples/MatrixFormats/dia
+examples/MatrixFormats/coo
+examples/MatrixAssembly/unordered_triplets
+examples/Gallery/diffusion
+examples/Gallery/poisson
+examples/Monitors/convergence_monitor
+examples/Monitors/verbose_monitor
+examples/Monitors/default_monitor
+examples/Preconditioners/ainv
+examples/Preconditioners/smoothed_aggregation
+examples/Preconditioners/diagonal
+examples/Views/csr
+examples/Views/cg_raw
+examples/Views/array1d
+examples/Views/csr_raw
+examples/Views/array2d_raw
+examples/Algorithms/maximal_independent_set
+examples/Algorithms/transpose
+examples/Algorithms/multiply
+examples/Algorithms/blas
+examples/LinearOperator/stencil
+examples/Version/version
+#<AutoIgnore/>

modules/cudaconfidencemaps/ext/cusplibrary-0.4.0/CHANGELOG

+#######################################
+#            Cusp v0.4.0              #
+#######################################
+
+Summary
+  Cusp v0.4.0 provides support for CUDA 5.5 and Thrust v1.7.  Version
+  0.4.0 includes several new graph algorithms based on backend integration
+  with back40computing, a refactored smoothed aggregation solver allowing
+  user-defined functions for core components and experimental work concerning
+  sparse matrix visualization.
+
+New Features
+  array2d formats allow access to row and column elements regardless of orientation
+  gallery method to generate grids
+  OpenGL implementation of spy to view sparse matrices
+  Generalized multilevel class
+  Added Sparse * Dense multiplication routine
+
+  Algorithms
+     cusp::graph::bfs
+     cusp::graph::connected_components
+     cusp::graph::hilbert_curve
+     cusp::graph::maximum_flow
+     cusp::graph::max_flow_to_min_cut
+     cusp::graph::pseudo_peripheral_vertex
+     cusp::graph::symmetric_rcm
+     cusp::krylov::cr
+
+Bug Fixes
+  Resolved issues related to Thrust 1.6 & Thrust 1.7
+  Resolved issues related to CUDA 5.5
+  Silenced warnings concerning uninitialized memory in generalized SpMV
+
+#######################################
+#            Cusp v0.3.1              #
+#######################################
+
+Summary
+  Minor compatibility fixes for Thrust 1.6.x
+
+#######################################
+#            Cusp v0.3.0              #
+#######################################
+
+Summary
+  Cusp v0.3.0 provides support for CUDA 4.1 and Thrust v1.5.  Version
+  0.3.0 includes incremental improvements to MatrixMarket I/O support
+  and array format conversions.
+
+New Features
+  MatrixMarket I/O routines support complex matrices
+  array1d is convertible to and from all other formats (e.g. array1d<->array2d)
+
+Bug Fixes
+  Resize final output of CSR * CSR to account for excluded zeros
+
+Acknowledgements
+  Justin Luitjens and Simon Layton for diagnosing CSR SpMM bug
+  Kashif Rasul for Thrust v1.6 support patch
+  Filipe Maia and Steven Dalton for general support and maintenance
+
+#######################################
+#            Cusp v0.2.0              #