/*=========================================================================

Program:   Visualization Toolkit
Module:    vtkOpenGLHAVSVolumeMapper.h

Copyright (c) Ken Martin, Will Schroeder, Bill Lorensen
All rights reserved.
See Copyright.txt or http://www.kitware.com/Copyright.htm for details.

This software is distributed WITHOUT ANY WARRANTY; without even
the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
PURPOSE.  See the above copyright notice for more information.

=========================================================================*/

/* Copyright 2005, 2006 by University of Utah. */

// .NAME vtkOpenGLHAVSVolumeMapper - Hardware-Assisted
// Visibility Sorting unstructured grid mapper, OpenGL implementation
//
// .SECTION Description
//
// vtkHAVSVolumeMapper is a class that renders polygonal data
// (represented as an unstructured grid) using the Hardware-Assisted
// Visibility Sorting (HAVS) algorithm.  First the unique triangles are sorted
// in object space, then they are sorted in image space using a fixed size
// A-buffer implemented on the GPU called the k-buffer.  The HAVS algorithm
// excels at rendering large datasets quickly.  The trade-off is that the
// algorithm may produce some rendering artifacts due to an insufficient k
// size (currently 2 or 6 is supported) or read/write race conditions.
// 
// A built in level-of-detail (LOD) approach samples the geometry using one of
// two heuristics (field or area).  If LOD is enabled, the amount of geometry
// that is sampled and rendered changes dynamically to stay within the target
// frame rate.  The field sampling method generally works best for datasets
// with cell sizes that don't vary much in size.  On the contrary, the area
// sampling approach gives better approximations when the volume has a lot of
// variation in cell size.
//
// The HAVS algorithm uses several advanced features on graphics hardware.
// The k-buffer sorting network is implemented using framebuffer objects
// (FBOs) with multiple render targets (MRTs).  Therefore, only cards that
// support these features can run the algorithm (at least an ATI 9500 or an
// NVidia NV40 (6600)).
//
// .SECTION Notes
//
// Several issues had to be addressed to get the HAVS algorithm working within
// the vtk framework.  These additions forced the code to forsake speed for
// the sake of compliance and robustness.
//
// The HAVS algorithm operates on the triangles that compose the mesh.
// Therefore, before rendering, the cells are decomposed into unique triangles
// and stored on the GPU for efficient rendering.  The use of GPU data
// structures is only recommended if the entire geometry can fit in graphics
// memory.  Otherwise this feature should be disabled. 
//
// Another new feature is the handling of mixed data types (eg., polygonal
// data with volume data).  This is handled by reading the z-buffer from the
// current window and copying it into the framebuffer object for off-screen
// rendering.  The depth test is then enabled so that the volume only appears
// over the opaque geometry.  Finally, the results of the off-screen rendering
// are blended into the framebuffer as a transparent, view-aligned texture. 
// 
// Instead of using a preintegrated 3D lookup table for storing the ray
// integral, this implementation uses partial pre-integration.  This improves
// the performance of dynamic transfer function updates by avoiding a costly
// preprocess of the table.
//
// A final change to the original algorithm is the handling of non-convexities
// in the mesh.  Due to read/write hazards that may create undesired artifacts
// with non-convexities when using a inside/outside toggle in the fragment
// program, another approach was employed.  To handle non-convexities, the
// fragment shader determines if a ray-gap is larger than the max cell size
// and kill the fragment if so.  This approximation performs rather well in
// practice but may miss small non-convexities.
// 
// For more information on the HAVS algorithm see:
//
//  "Hardware-Assisted Visibility Sorting for Unstructured Volume
// Rendering" by S. P. Callahan, M. Ikits, J. L. D. Comba, and C. T. Silva, 
// IEEE Transactions of Visualization and Computer Graphics; May/June 2005.
//
// For more information on the Level-of-Detail algorithm, see:
//
// "Interactive Rendering of Large Unstructured Grids Using Dynamic
// Level-of-Detail" by S. P. Callahan, J. L. D. Comba, P. Shirley, and
// C. T. Silva, Proceedings of IEEE Visualization '05, Oct. 2005.
//
// .SECTION Acknowledgments
//
// This code was developed by Steven P. Callahan under the supervision
// of Prof. Claudio T. Silva. The code also contains contributions
// from Milan Ikits, Linh Ha, Huy T. Vo, Carlos E. Scheidegger, and 
// Joao L. D. Comba.  
//
// The work was supported by grants, contracts, and gifts from the
// National Science Foundation, the Department of Energy, the Army
// Research Office, and IBM.
//
// The port of HAVS to VTK and ParaView has been primarily supported
// by Sandia National Labs.
//

#ifndef __vtkOpenGLHAVSVolumeMapper_h
#define __vtkOpenGLHAVSVolumeMapper_h

#include "vtkHAVSVolumeMapper.h"

#include <vtkWeakPointer.h> // to cache the vtkRenderWindow
class vtkRenderer;
class vtkRenderWindow;


class VTK_VOLUMERENDERING_EXPORT vtkOpenGLHAVSVolumeMapper : public vtkHAVSVolumeMapper
{
public:
  static vtkOpenGLHAVSVolumeMapper *New();
  vtkTypeMacro(vtkOpenGLHAVSVolumeMapper,
                       vtkHAVSVolumeMapper);
  virtual void PrintSelf(ostream& os, vtkIndent indent);
  
  // Description:
  // Render the volume
  virtual void Render(vtkRenderer *ren, vtkVolume *vol);

  // Description
  // Release any graphics resources that are being consumed by this volume
  // renderer.
  virtual void ReleaseGraphicsResources(vtkWindow *);

  // Description:
  // Set/get whether or not the data structures should be stored on the GPU 
  // for better peformance.
  virtual void SetGPUDataStructures(bool);

  // Description:
  // Check hardware support for the HAVS algorithm.  Necessary
  // features include off-screen rendering, 32-bit fp textures, multiple
  // render targets, and framebuffer objects.
  // Subclasses must override this method to indicate if supported by Hardware.
  virtual bool SupportedByHardware(vtkRenderer *r);
protected:

  vtkOpenGLHAVSVolumeMapper();
  ~vtkOpenGLHAVSVolumeMapper();
  virtual int FillInputPortInformation(int port, vtkInformation* info);

//BTX
  virtual void Initialize(vtkRenderer *ren, vtkVolume *vol);
  virtual void InitializeLookupTables(vtkVolume *vol);
  void InitializeGPUDataStructures();
  void InitializeShaders();
  void DeleteShaders();
  void InitializeFramebufferObject();

  void RenderHAVS(vtkRenderer *ren);
  void SetupFBOZBuffer(int screenWidth, int screenHeight, float depthNear, float depthFar, 
                       float *zbuffer);
  void SetupFBOMRT();
  void DrawFBOInit(int screenWidth, int screenHeight, float depthNear, float depthFar);
  void DrawFBOGeometry();
  void DrawFBOFlush(int screenWidth, int screenHeight, float depthNear, float depthFar);
  void DrawBlend(int screenWidth, int screenHeight, float depthNear, float depthFar);

  void CheckOpenGLError(const char *str);

  // GPU
  unsigned int VBOVertexName;
  unsigned int VBOTexCoordName;
  unsigned int VBOVertexIndexName;
  unsigned int VertexProgram;
  unsigned int FragmentProgramBegin;
  unsigned int FragmentProgram;
  unsigned int FragmentProgramEnd;
  unsigned int FramebufferObject;
  int FramebufferObjectSize;
  unsigned int FramebufferTextures[4];
  unsigned int DepthTexture;

  // Lookup Tables
  unsigned int PsiTableTexture;
  unsigned int TransferFunctionTexture;

  vtkWeakPointer<vtkRenderWindow> RenderWindow;
//ETX

private:
  vtkOpenGLHAVSVolumeMapper(const vtkOpenGLHAVSVolumeMapper&);  // Not implemented.
  void operator=(const vtkOpenGLHAVSVolumeMapper&);  // Not implemented.
};

#endif