I didn't apply the changes to a few macro-heavy files as Visual Studio removes all indentation from macros, whereas the indentation can be handy to see nesting.
335 lines
16 KiB
C++
335 lines
16 KiB
C++
/*******************************************************************************
|
|
* The MIT License (MIT)
|
|
*
|
|
* Copyright (c) 2015 David Williams and Matthew Williams
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
* in the Software without restriction, including without limitation the rights
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
* copies of the Software, and to permit persons to whom the Software is
|
|
* furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be included in all
|
|
* copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
* SOFTWARE.
|
|
*******************************************************************************/
|
|
|
|
#ifndef __PolyVox_PagedVolume_H__
|
|
#define __PolyVox_PagedVolume_H__
|
|
|
|
#include "BaseVolume.h"
|
|
#include "Region.h"
|
|
#include "Vector.h"
|
|
|
|
#include <limits>
|
|
#include <cstdlib> //For abort()
|
|
#include <cstring> //For memcpy
|
|
#include <unordered_map>
|
|
#include <list>
|
|
#include <map>
|
|
#include <memory>
|
|
#include <stdexcept> //For invalid_argument
|
|
#include <vector>
|
|
|
|
namespace PolyVox
|
|
{
|
|
/// The PagedVolume class provides a memory efficient method of storing voxel data while also allowing fast access and modification.
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
/// A PagedVolume is essentially a 3D array in which each element (or <i>voxel</i>) is identified by a three dimensional (x,y,z) coordinate.
|
|
/// We use the PagedVolume class to store our data in an efficient way, and it is the input to many of the algorithms (such as the surface
|
|
/// extractors) which form the heart of PolyVox. The PagedVolume class is templatised so that different types of data can be stored within each voxel.
|
|
///
|
|
/// Basic usage
|
|
/// -----------
|
|
///
|
|
/// The following code snippet shows how to construct a volume and demonstrates basic usage:
|
|
///
|
|
/// \code
|
|
/// PagedVolume<int> volume(Region(Vector3DInt32(0,0,0), Vector3DInt32(63,127,255)));
|
|
/// volume.setVoxel(15, 90, 42, int(5));
|
|
/// std::cout << "Voxel at (15, 90, 42) has value: " << volume.getVoxel(15, 90, 42) << std::endl;
|
|
/// std::cout << "Width = " << volume.getWidth() << ", Height = " << volume.getHeight() << ", Depth = " << volume.getDepth() << std::endl;
|
|
/// \endcode
|
|
///
|
|
/// The PagedVolume constructor takes a Region as a parameter. This specifies the valid range of voxels which can be held in the volume, so in this
|
|
/// particular case the valid voxel positions are (0,0,0) to (63, 127, 255). The result of attempts to access voxels outside this range will result
|
|
/// are defined by the WrapMode). PolyVox also has support for near infinite volumes which will be discussed later.
|
|
///
|
|
/// Access to individual voxels is provided via the setVoxel() and getVoxel() member functions. Advanced users may also be interested in
|
|
/// the Sampler nested class for faster read-only access to a large number of voxels.
|
|
///
|
|
/// Lastly the example prints out some properties of the PagedVolume. Note that the dimentsions getWidth(), getHeight(), and getDepth() are inclusive, such
|
|
/// that the width is 64 when the range of valid x coordinates goes from 0 to 63.
|
|
///
|
|
/// Data Representaion
|
|
/// ------------------
|
|
/// If stored carelessly, volume data can take up a huge amount of memory. For example, a volume of dimensions 1024x1024x1024 with
|
|
/// 1 byte per voxel will require 1GB of memory if stored in an uncompressed form. Natuarally our PagedVolume class is much more efficient
|
|
/// than this and it is worth understanding (at least at a high level) the approach which is used.
|
|
///
|
|
/// Essentially, the PagedVolume class stores its data as a collection of chunks. Each of these chunk is much smaller than the whole volume,
|
|
/// for example a typical size might be 32x32x32 voxels (though is is configurable by the user). In this case, a 256x512x1024 volume
|
|
/// would contain 8x16x32 = 4096 chunks. Typically these chunks do not need to all be in memory all the time, and the Pager class can
|
|
/// be used to control how they are loaded and unloaded. This mechanism allows a
|
|
/// potentially unlimited amount of data to be loaded, provided the user is able to take responsibility for storing any data which PolyVox
|
|
/// cannot fit in memory, and then returning it back to PolyVox on demand. For example, the user might choose to temporarily store this data
|
|
/// on disk or stream it to a remote database.
|
|
///
|
|
/// Essentially you are providing an extension to the PagedVolume class - a way for data to be stored once PolyVox has run out of memory for it. Note
|
|
/// that you don't actually have to do anything with the data - you could simply decide that once it gets removed from memory it doesn't matter
|
|
/// anymore.
|
|
///
|
|
/// Cache-aware traversal
|
|
/// ---------------------
|
|
/// *NOTE: This needs updating for PagedVolume rather than the old LargeVolume*
|
|
/// You might be suprised at just how many cache misses can occur when you traverse the volume in a naive manner. Consider a 1024x1024x1024 volume
|
|
/// with chunks of size 32x32x32. And imagine you iterate over this volume with a simple three-level for loop which iterates over x, the y, then z.
|
|
/// If you start at position (0,0,0) then ny the time you reach position (1023,0,0) you have touched 1024 voxels along one edge of the volume and
|
|
/// have pulled 32 chunks into the cache. By the time you reach (1023,1023,0) you have hit 1024x1024 voxels and pulled 32x32 chunks into the cache.
|
|
/// You are now ready to touch voxel (0,0,1) which is right next to where you started, but unless your cache is at least 32x32 chunks large then this
|
|
/// initial chunk has already been cleared from the cache.
|
|
///
|
|
/// Ensuring you have a large enough cache size can obviously help the above situation, but you might also consider iterating over the voxels in a
|
|
/// different order. For example, if you replace your three-level loop with a six-level loop then you can first process all the voxels between (0,0,0)
|
|
/// and (31,31,31), then process all the voxels between (32,0,0) and (63,0,0), and so forth. Using this approach you will have no cache misses even
|
|
/// is your cache size is only one. Of course the logic is more complex, but writing code in such a cache-aware manner may be beneficial in some situations.
|
|
///
|
|
/// Threading
|
|
/// ---------
|
|
/// The PagedVolume class does not make any guarentees about thread safety. You should ensure that all accesses are performed from the same thread.
|
|
/// This is true even if you are only reading data from the volume, as concurrently reading from different threads can invalidate the contents
|
|
/// of the chunk cache (amoung other problems).
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
template <typename VoxelType>
|
|
class PagedVolume : public BaseVolume<VoxelType>
|
|
{
|
|
public:
|
|
/// The PagedVolume stores it data as a set of Chunk instances which can be loaded and unloaded as memory requirements dictate.
|
|
class Chunk;
|
|
/// The Pager class is responsible for the loading and unloading of Chunks, and can be overridden by the user.
|
|
class Pager;
|
|
|
|
class Chunk
|
|
{
|
|
friend class PagedVolume;
|
|
|
|
public:
|
|
Chunk(Vector3DInt32 v3dPosition, uint16_t uSideLength, Pager* pPager = nullptr);
|
|
~Chunk();
|
|
|
|
VoxelType* getData(void) const;
|
|
uint32_t getDataSizeInBytes(void) const;
|
|
|
|
VoxelType getVoxel(uint32_t uXPos, uint32_t uYPos, uint32_t uZPos) const;
|
|
VoxelType getVoxel(const Vector3DUint16& v3dPos) const;
|
|
|
|
void setVoxel(uint32_t uXPos, uint32_t uYPos, uint32_t uZPos, VoxelType tValue);
|
|
void setVoxel(const Vector3DUint16& v3dPos, VoxelType tValue);
|
|
|
|
void changeLinearOrderingToMorton(void);
|
|
void changeMortonOrderingToLinear(void);
|
|
|
|
private:
|
|
/// Private copy constructor to prevent accisdental copying
|
|
Chunk(const Chunk& /*rhs*/) {};
|
|
|
|
/// Private assignment operator to prevent accisdental copying
|
|
Chunk& operator=(const Chunk& /*rhs*/) {};
|
|
|
|
// This is updated by the PagedVolume and used to discard the least recently used chunks.
|
|
uint32_t m_uChunkLastAccessed;
|
|
|
|
// This is so we can tell whether a uncompressed chunk has to be recompressed and whether
|
|
// a compressed chunk has to be paged back to disk, or whether they can just be discarded.
|
|
bool m_bDataModified;
|
|
|
|
uint32_t calculateSizeInBytes(void);
|
|
static uint32_t calculateSizeInBytes(uint32_t uSideLength);
|
|
|
|
VoxelType* m_tData;
|
|
uint16_t m_uSideLength;
|
|
uint8_t m_uSideLengthPower;
|
|
Pager* m_pPager;
|
|
|
|
// Note: Do we really need to store this position here as well as in the block maps?
|
|
Vector3DInt32 m_v3dChunkSpacePosition;
|
|
};
|
|
|
|
/**
|
|
* Users can override this class and provide an instance of the derived class to the PagedVolume constructor. This derived class
|
|
* could then perform tasks such as compression and decompression of the data, and read/writing it to a file, database, network,
|
|
* or other storage as appropriate. See FilePager for a simple example of such a derived class.
|
|
*/
|
|
class Pager
|
|
{
|
|
public:
|
|
/// Constructor
|
|
Pager() {};
|
|
/// Destructor
|
|
virtual ~Pager() {};
|
|
|
|
virtual void pageIn(const Region& region, Chunk* pChunk) = 0;
|
|
virtual void pageOut(const Region& region, Chunk* pChunk) = 0;
|
|
};
|
|
|
|
//There seems to be some descrepency between Visual Studio and GCC about how the following class should be declared.
|
|
//There is a work around (see also See http://goo.gl/qu1wn) given below which appears to work on VS2010 and GCC, but
|
|
//which seems to cause internal compiler errors on VS2008 when building with the /Gm 'Enable Minimal Rebuild' compiler
|
|
//option. For now it seems best to 'fix' it with the preprocessor insstead, but maybe the workaround can be reinstated
|
|
//in the future
|
|
//typedef Volume<VoxelType> VolumeOfVoxelType; //Workaround for GCC/VS2010 differences.
|
|
//class Sampler : public VolumeOfVoxelType::template Sampler< PagedVolume<VoxelType> >
|
|
#ifndef SWIG
|
|
#if defined(_MSC_VER)
|
|
class Sampler : public BaseVolume<VoxelType>::Sampler< PagedVolume<VoxelType> > //This line works on VS2010
|
|
#else
|
|
class Sampler : public BaseVolume<VoxelType>::template Sampler< PagedVolume<VoxelType> > //This line works on GCC
|
|
#endif
|
|
{
|
|
public:
|
|
Sampler(PagedVolume<VoxelType>* volume);
|
|
~Sampler();
|
|
|
|
inline VoxelType getVoxel(void) const;
|
|
|
|
void setPosition(const Vector3DInt32& v3dNewPos);
|
|
void setPosition(int32_t xPos, int32_t yPos, int32_t zPos);
|
|
inline bool setVoxel(VoxelType tValue);
|
|
|
|
void movePositiveX(void);
|
|
void movePositiveY(void);
|
|
void movePositiveZ(void);
|
|
|
|
void moveNegativeX(void);
|
|
void moveNegativeY(void);
|
|
void moveNegativeZ(void);
|
|
|
|
inline VoxelType peekVoxel1nx1ny1nz(void) const;
|
|
inline VoxelType peekVoxel1nx1ny0pz(void) const;
|
|
inline VoxelType peekVoxel1nx1ny1pz(void) const;
|
|
inline VoxelType peekVoxel1nx0py1nz(void) const;
|
|
inline VoxelType peekVoxel1nx0py0pz(void) const;
|
|
inline VoxelType peekVoxel1nx0py1pz(void) const;
|
|
inline VoxelType peekVoxel1nx1py1nz(void) const;
|
|
inline VoxelType peekVoxel1nx1py0pz(void) const;
|
|
inline VoxelType peekVoxel1nx1py1pz(void) const;
|
|
|
|
inline VoxelType peekVoxel0px1ny1nz(void) const;
|
|
inline VoxelType peekVoxel0px1ny0pz(void) const;
|
|
inline VoxelType peekVoxel0px1ny1pz(void) const;
|
|
inline VoxelType peekVoxel0px0py1nz(void) const;
|
|
inline VoxelType peekVoxel0px0py0pz(void) const;
|
|
inline VoxelType peekVoxel0px0py1pz(void) const;
|
|
inline VoxelType peekVoxel0px1py1nz(void) const;
|
|
inline VoxelType peekVoxel0px1py0pz(void) const;
|
|
inline VoxelType peekVoxel0px1py1pz(void) const;
|
|
|
|
inline VoxelType peekVoxel1px1ny1nz(void) const;
|
|
inline VoxelType peekVoxel1px1ny0pz(void) const;
|
|
inline VoxelType peekVoxel1px1ny1pz(void) const;
|
|
inline VoxelType peekVoxel1px0py1nz(void) const;
|
|
inline VoxelType peekVoxel1px0py0pz(void) const;
|
|
inline VoxelType peekVoxel1px0py1pz(void) const;
|
|
inline VoxelType peekVoxel1px1py1nz(void) const;
|
|
inline VoxelType peekVoxel1px1py0pz(void) const;
|
|
inline VoxelType peekVoxel1px1py1pz(void) const;
|
|
|
|
private:
|
|
//Other current position information
|
|
VoxelType* mCurrentVoxel;
|
|
|
|
uint16_t m_uXPosInChunk;
|
|
uint16_t m_uYPosInChunk;
|
|
uint16_t m_uZPosInChunk;
|
|
|
|
// This should ideally be const, but that prevent automatic generation of an assignment operator (https://goo.gl/Sn7KpZ).
|
|
// We could provide one manually, but it's currently unused so there is no real test for if it works. I'm putting
|
|
// together a new release at the moment so I'd rathern not make 'risky' changes.
|
|
uint16_t m_uChunkSideLengthMinusOne;
|
|
};
|
|
|
|
#endif // SWIG
|
|
|
|
public:
|
|
/// Constructor for creating a fixed size volume.
|
|
PagedVolume(Pager* pPager, uint32_t uTargetMemoryUsageInBytes = 256 * 1024 * 1024, uint16_t uChunkSideLength = 32);
|
|
/// Destructor
|
|
~PagedVolume();
|
|
|
|
/// Gets a voxel at the position given by <tt>x,y,z</tt> coordinates
|
|
VoxelType getVoxel(int32_t uXPos, int32_t uYPos, int32_t uZPos) const;
|
|
/// Gets a voxel at the position given by a 3D vector
|
|
VoxelType getVoxel(const Vector3DInt32& v3dPos) const;
|
|
|
|
/// Sets the voxel at the position given by <tt>x,y,z</tt> coordinates
|
|
void setVoxel(int32_t uXPos, int32_t uYPos, int32_t uZPos, VoxelType tValue);
|
|
/// Sets the voxel at the position given by a 3D vector
|
|
void setVoxel(const Vector3DInt32& v3dPos, VoxelType tValue);
|
|
|
|
/// Tries to ensure that the voxels within the specified Region are loaded into memory.
|
|
void prefetch(Region regPrefetch);
|
|
/// Ensures that any voxels within the specified Region are removed from memory.
|
|
//void flush(Region regFlush);
|
|
/// Removes all voxels from memory
|
|
void flushAll();
|
|
|
|
/// Calculates approximatly how many bytes of memory the volume is currently using.
|
|
uint32_t calculateSizeInBytes(void);
|
|
|
|
protected:
|
|
/// Copy constructor
|
|
PagedVolume(const PagedVolume& rhs);
|
|
|
|
/// Assignment operator
|
|
PagedVolume& operator=(const PagedVolume& rhs);
|
|
|
|
private:
|
|
bool canReuseLastAccessedChunk(int32_t iChunkX, int32_t iChunkY, int32_t iChunkZ) const;
|
|
Chunk* getChunk(int32_t uChunkX, int32_t uChunkY, int32_t uChunkZ) const;
|
|
|
|
// Storing these properties individually has proved to be faster than keeping
|
|
// them in a Vector3DInt32 as it avoids constructions and comparison overheads.
|
|
// They are also at the start of the class in the hope that they will be pulled
|
|
// into cache - I've got no idea if this actually makes a difference.
|
|
mutable int32_t m_v3dLastAccessedChunkX = 0;
|
|
mutable int32_t m_v3dLastAccessedChunkY = 0;
|
|
mutable int32_t m_v3dLastAccessedChunkZ = 0;
|
|
mutable Chunk* m_pLastAccessedChunk = nullptr;
|
|
|
|
mutable uint32_t m_uTimestamper = 0;
|
|
|
|
uint32_t m_uChunkCountLimit = 0;
|
|
|
|
// Chunks are stored in the following array which is used as a hash-table. Conventional wisdom is that such a hash-table
|
|
// should not be more than half full to avoid conflicts, and a practical chunk size seems to be 64^3. With this configuration
|
|
// there can be up to 32768*64^3 = 8 gigavoxels (with each voxel perhaps being many bytes). This should effectively make use
|
|
// of even high end machines. Of course, the user can choose to limit the memory usage in which case much less of the chunk
|
|
// array will actually be used. None-the-less, we have chosen to use a fixed size array (rather than a vector) as it appears to
|
|
// be slightly faster (probably due to the extra pointer indirection in a vector?) and the actual size of this array should
|
|
// just be 1Mb or so.
|
|
static const uint32_t uChunkArraySize = 65536;
|
|
mutable std::unique_ptr< Chunk > m_arrayChunks[uChunkArraySize];
|
|
|
|
// The size of the chunks
|
|
uint16_t m_uChunkSideLength;
|
|
uint8_t m_uChunkSideLengthPower;
|
|
int32_t m_iChunkMask;
|
|
|
|
Pager* m_pPager = nullptr;
|
|
};
|
|
}
|
|
|
|
#include "PagedVolume.inl"
|
|
#include "PagedVolumeChunk.inl"
|
|
#include "PagedVolumeSampler.inl"
|
|
|
|
#endif //__PolyVox_PagedVolume_H__
|