BansheeEngine/BansheeCore/Source/BsPixelUtil.cpp

//__________________________ Banshee Project - A modern game development toolkit _________________________________//
//_____________________________________ www.banshee-project.com __________________________________________________//
//________________________ Copyright (c) 2014 Marko Pintera. All rights reserved. ________________________________//
#include "BsPixelUtil.h"
#include "BsBitwise.h"
#include "BsColor.h"
#include "BsMath.h"
#include "BsException.h"
#include "nvtt/nvtt.h"

namespace BansheeEngine 
{
	/**
	 * @brief	Performs pixel data resampling using the point filter (nearest neighbor).
	 *			Does not perform format conversions.
	 *
	 * @tparam elementSize	Size of a single pixel in bytes.
	 */
	template<UINT32 elementSize> struct NearestResampler 
	{
		static void scale(const PixelData& source, const PixelData& dest) 
		{
			UINT8* sourceData = source.getData();
			UINT8* destPtr = dest.getData();

			// Get steps for traversing source data in 16/48 fixed point format
			UINT64 stepX = ((UINT64)source.getWidth() << 48) / dest.getWidth();
			UINT64 stepY = ((UINT64)source.getHeight() << 48) / dest.getHeight();
			UINT64 stepZ = ((UINT64)source.getDepth() << 48) / dest.getDepth();

			UINT64 curZ = (stepZ >> 1) - 1; // Offset half a pixel to start at pixel center
			for (UINT32 z = dest.getFront(); z < dest.getBack(); z++, curZ += stepZ) 
			{
				UINT32 offsetZ = (UINT32)(curZ >> 48) * source.getSlicePitch();

				UINT64 curY = (stepY >> 1) - 1; // Offset half a pixel to start at pixel center
				for (UINT32 y = dest.getTop(); y < dest.getBottom(); y++, curY += stepY) 
				{
					UINT32 offsetY = (UINT32)(curY >> 48) * source.getRowPitch();

					UINT64 curX = (stepX >> 1) - 1; // Offset half a pixel to start at pixel center
					for (UINT32 x = dest.getLeft(); x < dest.getRight(); x++, curX += stepX) 
					{
						UINT32 offsetX = (UINT32)(curX >> 48);
						UINT32 offsetBytes = elementSize*(offsetX + offsetY + offsetZ);

						UINT8* curSourcePtr = sourceData + offsetBytes;
							
						memcpy(destPtr, curSourcePtr, elementSize);
						destPtr += elementSize;
					}

					destPtr += elementSize*dest.getRowSkip();
				}

				destPtr += elementSize*dest.getSliceSkip();
			}
		}
	};

	/**
	 * @brief	Performs pixel data resampling using the box filter (linear).
	 *			Performs format conversions.
	 */
	struct LinearResampler 
	{
		static void scale(const PixelData& source, const PixelData& dest) 
		{
			UINT32 sourceElemSize = PixelUtil::getNumElemBytes(source.getFormat());
			UINT32 destElemSize = PixelUtil::getNumElemBytes(dest.getFormat());

			UINT8* sourceData = source.getData();
			UINT8* destPtr = dest.getData();

			// Get steps for traversing source data in 16/48 fixed point precision format
			UINT64 stepX = ((UINT64)source.getWidth() << 48) / dest.getWidth();
			UINT64 stepY = ((UINT64)source.getHeight() << 48) / dest.getHeight();
			UINT64 stepZ = ((UINT64)source.getDepth() << 48) / dest.getDepth();

			// Contains 16/16 fixed point precision format. Most significant
			// 16 bits will contain the coordinate in the source image, and the
			// least significant 16 bits will contain the fractional part of the coordinate
			// that will be used for determining the blend amount.
			UINT32 temp = 0;

			UINT64 curZ = (stepZ >> 1) - 1; // Offset half a pixel to start at pixel center
			for (UINT32 z = dest.getFront(); z < dest.getBack(); z++, curZ += stepZ) 
			{
				temp = UINT32(curZ >> 32);
				temp = (temp > 0x8000)? temp - 0x8000 : 0;
				UINT32 sampleCoordZ1 = temp >> 16;
				UINT32 sampleCoordZ2 = std::min(sampleCoordZ1 + 1, (UINT32)source.getDepth() - 1);
				float sampleWeightZ = (temp & 0xFFFF) / 65536.0f; 

				UINT64 curY = (stepY >> 1) - 1; // Offset half a pixel to start at pixel center
				for (UINT32 y = dest.getTop(); y < dest.getBottom(); y++, curY += stepY) 
				{
					temp = (UINT32)(curY >> 32);
					temp = (temp > 0x8000)? temp - 0x8000 : 0;
					UINT32 sampleCoordY1 = temp >> 16;
					UINT32 sampleCoordY2 = std::min(sampleCoordY1 + 1, (UINT32)source.getHeight() - 1);
					float sampleWeightY = (temp & 0xFFFF) / 65536.0f;

					UINT64 curX = (stepX >> 1) - 1; // Offset half a pixel to start at pixel center
					for (UINT32 x = dest.getLeft(); x < dest.getRight(); x++, curX += stepX) 
					{
						temp = (UINT32)(curX >> 32);
						temp = (temp > 0x8000)? temp - 0x8000 : 0;
						UINT32 sampleCoordX1 = temp >> 16;
						UINT32 sampleCoordX2 = std::min(sampleCoordX1 + 1, (UINT32)source.getWidth() - 1);
						float sampleWeightX = (temp & 0xFFFF) / 65536.0f;

						Color x1y1z1, x2y1z1, x1y2z1, x2y2z1;
						Color x1y1z2, x2y1z2, x1y2z2, x2y2z2;

#define GETSOURCEDATA(x, y, z) sourceData + sourceElemSize*((x)+(y)*source.getRowPitch() + (z)*source.getSlicePitch())

						PixelUtil::unpackColor(&x1y1z1, source.getFormat(), GETSOURCEDATA(sampleCoordX1, sampleCoordY1, sampleCoordZ1));
						PixelUtil::unpackColor(&x2y1z1, source.getFormat(), GETSOURCEDATA(sampleCoordX2, sampleCoordY1, sampleCoordZ1));
						PixelUtil::unpackColor(&x1y2z1, source.getFormat(), GETSOURCEDATA(sampleCoordX1, sampleCoordY2, sampleCoordZ1));
						PixelUtil::unpackColor(&x2y2z1, source.getFormat(), GETSOURCEDATA(sampleCoordX2, sampleCoordY2, sampleCoordZ1));
						PixelUtil::unpackColor(&x1y1z2, source.getFormat(), GETSOURCEDATA(sampleCoordX1, sampleCoordY1, sampleCoordZ2));
						PixelUtil::unpackColor(&x2y1z2, source.getFormat(), GETSOURCEDATA(sampleCoordX2, sampleCoordY1, sampleCoordZ2));
						PixelUtil::unpackColor(&x1y2z2, source.getFormat(), GETSOURCEDATA(sampleCoordX1, sampleCoordY2, sampleCoordZ2));
						PixelUtil::unpackColor(&x2y2z2, source.getFormat(), GETSOURCEDATA(sampleCoordX2, sampleCoordY2, sampleCoordZ2));
#undef GETSOURCEDATA

						Color accum =
							x1y1z1 * ((1.0f - sampleWeightX)*(1.0f - sampleWeightY)*(1.0f - sampleWeightZ)) +
							x2y1z1 * (        sampleWeightX *(1.0f - sampleWeightY)*(1.0f - sampleWeightZ)) +
							x1y2z1 * ((1.0f - sampleWeightX)*        sampleWeightY *(1.0f - sampleWeightZ)) +
							x2y2z1 * (        sampleWeightX *        sampleWeightY *(1.0f - sampleWeightZ)) +
							x1y1z2 * ((1.0f - sampleWeightX)*(1.0f - sampleWeightY)*        sampleWeightZ ) +
							x2y1z2 * (        sampleWeightX *(1.0f - sampleWeightY)*        sampleWeightZ ) +
							x1y2z2 * ((1.0f - sampleWeightX)*        sampleWeightY *        sampleWeightZ ) +
							x2y2z2 * (        sampleWeightX *        sampleWeightY *        sampleWeightZ );

						PixelUtil::packColor(accum, dest.getFormat(), destPtr);

						destPtr += destElemSize;
					}

					destPtr += destElemSize * dest.getRowSkip();
				}

				destPtr += destElemSize * dest.getSliceSkip();
			}
		}
	};


	/**
	 * @brief	Performs pixel data resampling using the box filter (linear).
	 *			Only handles float RGB or RGBA pixel data (32 bits per channel).
	 */
	struct LinearResampler_Float32 
	{
		static void scale(const PixelData& source, const PixelData& dest) 
		{
			UINT32 numSourceChannels = PixelUtil::getNumElemBytes(source.getFormat()) / sizeof(float);
			UINT32 numDestChannels = PixelUtil::getNumElemBytes(dest.getFormat()) / sizeof(float);

			float* sourceData = (float*)source.getData();
			float* destPtr = (float*)dest.getData();

			// Get steps for traversing source data in 16/48 fixed point precision format
			UINT64 stepX = ((UINT64)source.getWidth() << 48) / dest.getWidth();
			UINT64 stepY = ((UINT64)source.getHeight() << 48) / dest.getHeight();
			UINT64 stepZ = ((UINT64)source.getDepth() << 48) / dest.getDepth();

			// Contains 16/16 fixed point precision format. Most significant
			// 16 bits will contain the coordinate in the source image, and the
			// least significant 16 bits will contain the fractional part of the coordinate
			// that will be used for determining the blend amount.
			UINT32 temp = 0;

			UINT64 curZ = (stepZ >> 1) - 1; // Offset half a pixel to start at pixel center
			for (UINT32 z = dest.getFront(); z < dest.getBack(); z++, curZ += stepZ) 
			{
				temp = (UINT32)(curZ >> 32);
				temp = (temp > 0x8000)? temp - 0x8000 : 0;
				UINT32 sampleCoordZ1 = temp >> 16;
				UINT32 sampleCoordZ2 = std::min(sampleCoordZ1 + 1, (UINT32)source.getDepth() - 1);
				float sampleWeightZ = (temp & 0xFFFF) / 65536.0f;

				UINT64 curY = (stepY >> 1) - 1; // Offset half a pixel to start at pixel center
				for (UINT32 y = dest.getTop(); y < dest.getBottom(); y++, curY += stepY) 
				{
					temp = (UINT32)(curY >> 32);
					temp = (temp > 0x8000)? temp - 0x8000 : 0;
					UINT32 sampleCoordY1 = temp >> 16;
					UINT32 sampleCoordY2 = std::min(sampleCoordY1 + 1, (UINT32)source.getHeight() - 1);
					float sampleWeightY = (temp & 0xFFFF) / 65536.0f;

					UINT64 curX = (stepX >> 1) - 1; // Offset half a pixel to start at pixel center
					for (UINT32 x = dest.getLeft(); x < dest.getRight(); x++, curX += stepX) 
					{
						temp = (UINT32)(curX >> 32);
						temp = (temp > 0x8000)? temp - 0x8000 : 0;
						UINT32 sampleCoordX1 = temp >> 16;
						UINT32 sampleCoordX2 = std::min(sampleCoordX1 + 1, (UINT32)source.getWidth() - 1);
						float sampleWeightX = (temp & 0xFFFF) / 65536.0f;

						// process R,G,B,A simultaneously for cache coherence?
						float accum[4] = { 0.0f, 0.0f, 0.0f, 0.0f };


#define ACCUM3(x,y,z,factor) \
						{ float f = factor; \
						UINT32 offset = (x + y*source.getRowPitch() + z*source.getSlicePitch())*numSourceChannels; \
						accum[0] += sourceData[offset + 0] * f; accum[1] += sourceData[offset + 1] * f; \
						accum[2] += sourceData[offset + 2] * f; }

#define ACCUM4(x,y,z,factor) \
						{ float f = factor; \
						UINT32 offset = (x + y*source.getRowPitch() + z*source.getSlicePitch())*numSourceChannels; \
						accum[0] += sourceData[offset + 0] * f; accum[1] += sourceData[offset + 1] * f; \
						accum[2] += sourceData[offset + 2] * f; accum[3] += sourceData[offset + 3] * f; }

						if (numSourceChannels == 3 || numDestChannels == 3)
						{
							// RGB
							ACCUM3(sampleCoordX1, sampleCoordY1, sampleCoordZ1, (1.0f - sampleWeightX) * (1.0f - sampleWeightY) * (1.0f - sampleWeightZ));
							ACCUM3(sampleCoordX2, sampleCoordY1, sampleCoordZ1, sampleWeightX		   * (1.0f - sampleWeightY) * (1.0f - sampleWeightZ));
							ACCUM3(sampleCoordX1, sampleCoordY2, sampleCoordZ1, (1.0f - sampleWeightX) * sampleWeightY			* (1.0f - sampleWeightZ));
							ACCUM3(sampleCoordX2, sampleCoordY2, sampleCoordZ1, sampleWeightX		   * sampleWeightY		    * (1.0f - sampleWeightZ));
							ACCUM3(sampleCoordX1, sampleCoordY1, sampleCoordZ2, (1.0f - sampleWeightX) * (1.0f - sampleWeightY) * sampleWeightZ);
							ACCUM3(sampleCoordX2, sampleCoordY1, sampleCoordZ2, sampleWeightX		   * (1.0f - sampleWeightY) * sampleWeightZ);
							ACCUM3(sampleCoordX1, sampleCoordY2, sampleCoordZ2, (1.0f - sampleWeightX) * sampleWeightY			* sampleWeightZ);
							ACCUM3(sampleCoordX2, sampleCoordY2, sampleCoordZ2, sampleWeightX		   * sampleWeightY			* sampleWeightZ);
							accum[3] = 1.0f;
						}
						else 
						{
							// RGBA
							ACCUM4(sampleCoordX1, sampleCoordY1, sampleCoordZ1, (1.0f - sampleWeightX) * (1.0f - sampleWeightY) * (1.0f - sampleWeightZ));
							ACCUM4(sampleCoordX2, sampleCoordY1, sampleCoordZ1, sampleWeightX		   * (1.0f - sampleWeightY) * (1.0f - sampleWeightZ));
							ACCUM4(sampleCoordX1, sampleCoordY2, sampleCoordZ1, (1.0f - sampleWeightX) * sampleWeightY			* (1.0f - sampleWeightZ));
							ACCUM4(sampleCoordX2, sampleCoordY2, sampleCoordZ1, sampleWeightX		   * sampleWeightY			* (1.0f - sampleWeightZ));
							ACCUM4(sampleCoordX1, sampleCoordY1, sampleCoordZ2, (1.0f - sampleWeightX) * (1.0f - sampleWeightY) * sampleWeightZ);
							ACCUM4(sampleCoordX2, sampleCoordY1, sampleCoordZ2, sampleWeightX		   * (1.0f - sampleWeightY) * sampleWeightZ);
							ACCUM4(sampleCoordX1, sampleCoordY2, sampleCoordZ2, (1.0f - sampleWeightX) * sampleWeightY			* sampleWeightZ);
							ACCUM4(sampleCoordX2, sampleCoordY2, sampleCoordZ2, sampleWeightX		   * sampleWeightY			* sampleWeightZ);
						}

						memcpy(destPtr, accum, sizeof(float)*numDestChannels);

#undef ACCUM3
#undef ACCUM4

						destPtr += numDestChannels;
					}

					destPtr += numDestChannels*dest.getRowSkip();
				}

				destPtr += numDestChannels*dest.getSliceSkip();
			}
		}
	};



	// byte linear resampler, does not do any format conversions.
	// only handles pixel formats that use 1 byte per color channel.
	// 2D only; punts 3D pixelboxes to default LinearResampler (slow).
	// templated on bytes-per-pixel to allow compiler optimizations, such
	// as unrolling loops and replacing multiplies with bitshifts

	/**
	 * @brief	Performs pixel data resampling using the box filter (linear).
	 *			Only handles pixel formats with one byte per channel. Does
	 *			not perform format conversion.
	 *
	 * @tparam	channels	Number of channels in the pixel format.
	 */
	template<UINT32 channels> struct LinearResampler_Byte 
	{
		static void scale(const PixelData& source, const PixelData& dest) 
		{
			// Only optimized for 2D
			if (source.getDepth() > 1 || dest.getDepth() > 1) 
			{
				LinearResampler::scale(source, dest);
				return;
			}

			UINT8* sourceData = (UINT8*)source.getData();
			UINT8* destPtr = (UINT8*)dest.getData();

			// Get steps for traversing source data in 16/48 fixed point precision format
			UINT64 stepX = ((UINT64)source.getWidth() << 48) / dest.getWidth();
			UINT64 stepY = ((UINT64)source.getHeight() << 48) / dest.getHeight();

			// Contains 16/16 fixed point precision format. Most significant
			// 16 bits will contain the coordinate in the source image, and the
			// least significant 16 bits will contain the fractional part of the coordinate
			// that will be used for determining the blend amount.
			UINT32 temp;

			UINT64 curY = (stepY >> 1) - 1; // Offset half a pixel to start at pixel center
			for (UINT32 y = dest.getTop(); y < dest.getBottom(); y++, curY += stepY)
			{
				temp = (UINT32)(curY >> 36);
				temp = (temp > 0x800)? temp - 0x800: 0;
				UINT32 sampleWeightY = temp & 0xFFF;
				UINT32 sampleCoordY1 = temp >> 12;
				UINT32 sampleCoordY2 = std::min(sampleCoordY1 + 1, (UINT32)source.getBottom() - source.getTop() - 1);

				UINT32 sampleY1Offset = sampleCoordY1 * source.getRowPitch();
				UINT32 sampleY2Offset = sampleCoordY2 * source.getRowPitch();

				UINT64 curX = (stepX >> 1) - 1; // Offset half a pixel to start at pixel center
				for (UINT32 x = dest.getLeft(); x < dest.getRight(); x++, curX += stepX)
				{
					temp = (UINT32)(curX >> 36);
					temp = (temp > 0x800)? temp - 0x800 : 0;
					UINT32 sampleWeightX = temp & 0xFFF;
					UINT32 sampleCoordX1 = temp >> 12;
					UINT32 sampleCoordX2 = std::min(sampleCoordX1 + 1, (UINT32)source.getRight() - source.getLeft() - 1);

					UINT32 sxfsyf = sampleWeightX*sampleWeightY;
					for (UINT32 k = 0; k < channels; k++) 
					{
						UINT32 accum =
							sourceData[(sampleCoordX1 + sampleY1Offset)*channels+k]*(0x1000000-(sampleWeightX<<12)-(sampleWeightY<<12)+sxfsyf) +
							sourceData[(sampleCoordX2 + sampleY1Offset)*channels+k]*((sampleWeightX<<12)-sxfsyf) +
							sourceData[(sampleCoordX1 + sampleY2Offset)*channels+k]*((sampleWeightY<<12)-sxfsyf) +
							sourceData[(sampleCoordX2 + sampleY2Offset)*channels+k]*sxfsyf;

						// Round up to byte size
						*destPtr = (UINT8)((accum + 0x800000) >> 24);
						destPtr++;
					}
				}
				destPtr += channels*dest.getRowSkip();
			}
		}
	};

	/**
	 * @brief	Data describing a pixel format.
	 */
    struct PixelFormatDescription 
	{
		const char* name; /**< Name of the format. */
		UINT8 elemBytes; /**< Number of bytes one element (color value) uses. */
		UINT32 flags; /**< PixelFormatFlags set by the pixel format. */
        PixelComponentType componentType; /**< Data type of a single element of the format. */
		UINT8 componentCount; /**< Number of elements in the format. */

		UINT8 rbits, gbits, bbits, abits; /**< Number of bits per element in the format. */

        UINT32 rmask, gmask, bmask, amask; /**< Masks used by packers/unpackers. */
		UINT8 rshift, gshift, bshift, ashift; /**< Shifts used by packers/unpackers. */
    };

	/**
	 * @brief	A list of all available pixel formats.
	 */
    PixelFormatDescription _pixelFormats[PF_COUNT] = {
        {"PF_UNKNOWN",
        /* Bytes per element */
        0,
        /* Flags */
        0,
        /* Component type and count */
        PCT_BYTE, 0,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
		//-----------------------------------------------------------------------
		{"PF_R8",
		/* Bytes per element */
		1,
		/* Flags */
		0,
		/* Component type and count */
		PCT_BYTE, 1,
		/* rbits, gbits, bbits, abits */
		8, 0, 0, 0,
		/* Masks and shifts */
		0x000000FF, 0, 0, 0, 
		0, 0, 0, 0
		},
		//-----------------------------------------------------------------------
		{"PF_R8G8",
		/* Bytes per element */
		2,
		/* Flags */
		0,
		/* Component type and count */
		PCT_BYTE, 2,
		/* rbits, gbits, bbits, abits */
		8, 8, 0, 0,
		/* Masks and shifts */
		0x000000FF, 0x0000FF00, 0, 0, 
		0, 8, 0, 0
		},
	//-----------------------------------------------------------------------
        {"PF_R8G8B8",
        /* Bytes per element */
        3,  // 24 bit integer -- special
        /* Flags */
        PFF_NATIVEENDIAN,
        /* Component type and count */
        PCT_BYTE, 3,
        /* rbits, gbits, bbits, abits */
        8, 8, 8, 0,
        /* Masks and shifts */
        0x000000FF, 0x0000FF00, 0x00FF0000, 0,
        0, 8, 16, 0
        },
	//-----------------------------------------------------------------------
        {"PF_B8G8R8",
        /* Bytes per element */
        3,  // 24 bit integer -- special
        /* Flags */
        PFF_NATIVEENDIAN,
        /* Component type and count */
        PCT_BYTE, 3,
        /* rbits, gbits, bbits, abits */
        8, 8, 8, 0,
        /* Masks and shifts */
        0x00FF0000, 0x0000FF00, 0x000000FF, 0,
        16, 8, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_A8R8G8B8",
        /* Bytes per element */
        4,
        /* Flags */
        PFF_HASALPHA | PFF_NATIVEENDIAN,
        /* Component type and count */
        PCT_BYTE, 4,
        /* rbits, gbits, bbits, abits */
        8, 8, 8, 8,
        /* Masks and shifts */
        0x0000FF00, 0x00FF0000, 0xFF000000, 0x000000FF,
        8, 16, 24, 0
        },
	//-----------------------------------------------------------------------
        {"PF_A8B8G8R8",
        /* Bytes per element */
        4,
        /* Flags */
        PFF_HASALPHA | PFF_NATIVEENDIAN,
        /* Component type and count */
        PCT_BYTE, 4,
        /* rbits, gbits, bbits, abits */
        8, 8, 8, 8,
        /* Masks and shifts */
        0xFF000000, 0x00FF0000, 0x0000FF00, 0x000000FF,
        24, 16, 8, 0,
        },
	//-----------------------------------------------------------------------
        {"PF_B8G8R8A8",
        /* Bytes per element */
        4,
        /* Flags */
        PFF_HASALPHA | PFF_NATIVEENDIAN,
        /* Component type and count */
        PCT_BYTE, 4,
        /* rbits, gbits, bbits, abits */
        8, 8, 8, 8,
        /* Masks and shifts */
        0x00FF0000, 0x0000FF00, 0x000000FF, 0xFF000000,
        16, 8, 0, 24
        },
	//-----------------------------------------------------------------------
		{"PF_R8G8B8A8",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_HASALPHA | PFF_NATIVEENDIAN,
		/* Component type and count */
		PCT_BYTE, 4,
		/* rbits, gbits, bbits, abits */
		8, 8, 8, 8,
		/* Masks and shifts */
		0x000000FF, 0x0000FF00, 0x00FF0000, 0xFF000000,
		0, 8, 16, 24
		},
	//-----------------------------------------------------------------------
		{"PF_X8R8G8B8",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_NATIVEENDIAN,
		/* Component type and count */
		PCT_BYTE, 3,
		/* rbits, gbits, bbits, abits */
		8, 8, 8, 0,
		/* Masks and shifts */
		0x0000FF00, 0x00FF0000, 0xFF000000, 0x000000FF,
		8, 16, 24, 0
		},
	//-----------------------------------------------------------------------
		{"PF_X8B8G8R8",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_NATIVEENDIAN,
		/* Component type and count */
		PCT_BYTE, 3,
		/* rbits, gbits, bbits, abits */
		8, 8, 8, 0,
		/* Masks and shifts */
		0xFF000000, 0x00FF0000, 0x0000FF00, 0x000000FF,
		24, 16, 8, 0
		},
	//-----------------------------------------------------------------------
		{"PF_R8G8B8X8",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_HASALPHA | PFF_NATIVEENDIAN,
		/* Component type and count */
		PCT_BYTE, 3,
		/* rbits, gbits, bbits, abits */
		8, 8, 8, 0,
		/* Masks and shifts */
		0x000000FF, 0x0000FF00, 0x00FF0000, 0xFF000000,
		0, 8, 16, 0
		},
	//-----------------------------------------------------------------------
		{"PF_B8G8R8X8",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_HASALPHA | PFF_NATIVEENDIAN,
		/* Component type and count */
		PCT_BYTE, 3,
		/* rbits, gbits, bbits, abits */
		8, 8, 8, 0,
		/* Masks and shifts */
		0x00FF0000, 0x0000FF00, 0x000000FF, 0xFF000000,
		16, 8, 0, 0
		},
	//-----------------------------------------------------------------------
        {"PF_BC1",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED | PFF_HASALPHA,
        /* Component type and count */
        PCT_BYTE, 3, // No alpha
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_BC1a",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED,
        /* Component type and count */
        PCT_BYTE, 3,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_BC2",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED | PFF_HASALPHA,
        /* Component type and count */
        PCT_BYTE, 4,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_BC3",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED | PFF_HASALPHA,
        /* Component type and count */
        PCT_BYTE, 4,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_BC4",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED,
        /* Component type and count */
        PCT_BYTE, 1,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_BC5",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED,
        /* Component type and count */
        PCT_BYTE, 2,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_BC6H",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED,
        /* Component type and count */
        PCT_FLOAT16, 3,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_BC7",
        /* Bytes per element */
        0,
        /* Flags */
        PFF_COMPRESSED | PFF_HASALPHA,
        /* Component type and count */
        PCT_BYTE, 4,
        /* rbits, gbits, bbits, abits */
        0, 0, 0, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
		{"PF_FLOAT16_R",
		/* Bytes per element */
		2,
		/* Flags */
		PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT16, 1,
		/* rbits, gbits, bbits, abits */
		16, 0, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		},
	//-----------------------------------------------------------------------
		{"PF_FLOAT16_RG",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT16, 2,
		/* rbits, gbits, bbits, abits */
		16, 16, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		},
	//-----------------------------------------------------------------------
        {"PF_FLOAT16_RGB",
        /* Bytes per element */
        6,
        /* Flags */
        PFF_FLOAT,
        /* Component type and count */
        PCT_FLOAT16, 3,
        /* rbits, gbits, bbits, abits */
        16, 16, 16, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_FLOAT16_RGBA",
        /* Bytes per element */
        8,
        /* Flags */
        PFF_FLOAT | PFF_HASALPHA,
        /* Component type and count */
        PCT_FLOAT16, 4,
        /* rbits, gbits, bbits, abits */
        16, 16, 16, 16,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
		{"PF_FLOAT32_R",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT32, 1,
		/* rbits, gbits, bbits, abits */
		32, 0, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		},
	//-----------------------------------------------------------------------
		{"PF_FLOAT32_RG",
		/* Bytes per element */
		8,
		/* Flags */
		PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT32, 2,
		/* rbits, gbits, bbits, abits */
		32, 32, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		},
	//-----------------------------------------------------------------------
        {"PF_FLOAT32_RGB",
        /* Bytes per element */
        12,
        /* Flags */
        PFF_FLOAT,
        /* Component type and count */
        PCT_FLOAT32, 3,
        /* rbits, gbits, bbits, abits */
        32, 32, 32, 0,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
        {"PF_FLOAT32_RGBA",
        /* Bytes per element */
        16,
        /* Flags */
        PFF_FLOAT | PFF_HASALPHA,
        /* Component type and count */
        PCT_FLOAT32, 4,
        /* rbits, gbits, bbits, abits */
        32, 32, 32, 32,
        /* Masks and shifts */
        0, 0, 0, 0, 0, 0, 0, 0
        },
	//-----------------------------------------------------------------------
		{"PF_D32_S8X24",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_DEPTH | PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT32, 1,
		/* rbits, gbits, bbits, abits */
		0, 0, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		}, 
	//-----------------------------------------------------------------------
		{"PF_D24_S8",
		/* Bytes per element */
		8,
		/* Flags */
		PFF_DEPTH | PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT32, 2,
		/* rbits, gbits, bbits, abits */
		0, 0, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		}, 
	//-----------------------------------------------------------------------
		{"PF_D32",
		/* Bytes per element */
		4,
		/* Flags */
		PFF_DEPTH | PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT32, 1,
		/* rbits, gbits, bbits, abits */
		0, 0, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		}, 
	//-----------------------------------------------------------------------
		{"PF_D16",
		/* Bytes per element */
		2,
		/* Flags */
		PFF_DEPTH | PFF_FLOAT,
		/* Component type and count */
		PCT_FLOAT16, 1,
		/* rbits, gbits, bbits, abits */
		0, 0, 0, 0,
		/* Masks and shifts */
		0, 0, 0, 0, 0, 0, 0, 0
		}, 
    };

    static inline const PixelFormatDescription &getDescriptionFor(const PixelFormat fmt)
    {
        const int ord = (int)fmt;
        assert(ord>=0 && ord<PF_COUNT);

        return _pixelFormats[ord];
    }

	/**
	 * @brief	Handles compression output from NVTT library for a single image.
	 */
	struct NVTTCompressOutputHandler : public nvtt::OutputHandler
	{
		NVTTCompressOutputHandler(UINT8* buffer, UINT32 sizeBytes)
			:buffer(buffer), bufferWritePos(buffer), bufferEnd(buffer + sizeBytes)
		{ }

		virtual void beginImage(int size, int width, int height, int depth, int face, int miplevel)
		{ }

		virtual bool writeData(const void* data, int size)
		{
			assert((bufferWritePos + size) <= bufferEnd);
			memcpy(bufferWritePos, data, size);
			bufferWritePos += size;

			return true;
		}

		UINT8* buffer;
		UINT8* bufferWritePos;
		UINT8* bufferEnd;
	};

	/**
	 * @brief	Handles output from NVTT library for a mip-map chain.
	 */
	struct NVTTMipmapOutputHandler : public nvtt::OutputHandler
	{
		NVTTMipmapOutputHandler(const Vector<PixelDataPtr>& buffers)
			:buffers(buffers), bufferWritePos(nullptr), bufferEnd(nullptr)
		{ }

		virtual void beginImage(int size, int width, int height, int depth, int face, int miplevel)
		{ 
			assert(miplevel >= 0 && miplevel < buffers.size());
			assert(size == buffers[miplevel]->getConsecutiveSize());

			activeBuffer = buffers[miplevel];

			bufferWritePos = activeBuffer->getData();
			bufferEnd = bufferWritePos + activeBuffer->getConsecutiveSize();
		}

		virtual bool writeData(const void* data, int size)
		{
			assert((bufferWritePos + size) <= bufferEnd);
			memcpy(bufferWritePos, data, size);
			bufferWritePos += size;

			return true;
		}

		Vector<PixelDataPtr> buffers;
		PixelDataPtr activeBuffer;

		UINT8* bufferWritePos;
		UINT8* bufferEnd;
	};

	nvtt::Format toNVTTFormat(PixelFormat format)
	{
		switch (format)
		{
		case PF_BC1:
			return nvtt::Format_BC1;
		case PF_BC1a:
			return nvtt::Format_BC1a;
		case PF_BC2:
			return nvtt::Format_BC2;
		case PF_BC3:
			return nvtt::Format_BC3;
		case PF_BC4:
			return nvtt::Format_BC4;
		case PF_BC5:
			return nvtt::Format_BC5;
		}

		// Unsupported format
		return nvtt::Format_BC3;
	}

	nvtt::Quality toNVTTQuality(CompressionQuality quality)
	{
		switch (quality)
		{
		case CompressionQuality::Fastest:
			return nvtt::Quality_Fastest;
		case CompressionQuality::Highest:
			return nvtt::Quality_Highest;
		case CompressionQuality::Normal:
			return nvtt::Quality_Normal;
		case CompressionQuality::Production:
			return nvtt::Quality_Normal;
		}

		// Unknown quality level
		return nvtt::Quality_Normal;
	}

	nvtt::AlphaMode toNVTTAlphaMode(AlphaMode alphaMode)
	{
		switch (alphaMode)
		{
		case AlphaMode::None:
			return nvtt::AlphaMode_None;
		case AlphaMode::Premultiplied:
			return nvtt::AlphaMode_Premultiplied;
		case AlphaMode::Transparency:
			return nvtt::AlphaMode_Transparency;
		}

		// Unknown alpha mode
		return nvtt::AlphaMode_None;
	}

	nvtt::WrapMode toNVTTWrapMode(MipMapWrapMode wrapMode)
	{
		switch (wrapMode)
		{
		case MipMapWrapMode::Clamp:
			return nvtt::WrapMode_Clamp;
		case MipMapWrapMode::Mirror:
			return nvtt::WrapMode_Mirror;
		case MipMapWrapMode::Repeat:
			return nvtt::WrapMode_Repeat;
		}

		// Unknown alpha mode
		return nvtt::WrapMode_Mirror;
	}

    UINT32 PixelUtil::getNumElemBytes(PixelFormat format)
    {
        return getDescriptionFor(format).elemBytes;
    }

	UINT32 PixelUtil::getMemorySize(UINT32 width, UINT32 height, UINT32 depth, PixelFormat format)
	{
		if(isCompressed(format))
		{
			switch(format)
			{
				// BC formats work by dividing the image into 4x4 blocks, then encoding each
				// 4x4 block with a certain number of bytes. 
				case PF_BC1:
				case PF_BC1a:
				case PF_BC4:
					return ((width+3)/4)*((height+3)/4)*8 * depth;
				case PF_BC2:
				case PF_BC3:
				case PF_BC5:
				case PF_BC6H:
				case PF_BC7:
					return ((width+3)/4)*((height+3)/4)*16 * depth;

				default:
					BS_EXCEPT(InvalidParametersException, "Invalid compressed pixel format");
			}
		}
		else
		{
			return width*height*depth*getNumElemBytes(format);
		}
	}

    UINT32 PixelUtil::getNumElemBits(PixelFormat format)
    {
        return getDescriptionFor(format).elemBytes * 8;
    }

    UINT32 PixelUtil::getFlags(PixelFormat format)
    {
        return getDescriptionFor(format).flags;
    }

    bool PixelUtil::hasAlpha(PixelFormat format)
    {
        return (PixelUtil::getFlags(format) & PFF_HASALPHA) > 0;
    }

    bool PixelUtil::isFloatingPoint(PixelFormat format)
    {
        return (PixelUtil::getFlags(format) & PFF_FLOAT) > 0;
    }

    bool PixelUtil::isCompressed(PixelFormat format)
    {
        return (PixelUtil::getFlags(format) & PFF_COMPRESSED) > 0;
    }

    bool PixelUtil::isDepth(PixelFormat format)
    {
        return (PixelUtil::getFlags(format) & PFF_DEPTH) > 0;
    }

    bool PixelUtil::isNativeEndian(PixelFormat format)
    {
        return (PixelUtil::getFlags(format) & PFF_NATIVEENDIAN) > 0;
    }

	bool PixelUtil::isValidExtent(UINT32 width, UINT32 height, UINT32 depth, PixelFormat format)
	{
		if(isCompressed(format))
		{
			switch(format)
			{
				case PF_BC1:
				case PF_BC2:
				case PF_BC1a:
				case PF_BC3:
				case PF_BC4:
				case PF_BC5:
				case PF_BC6H:
				case PF_BC7:
					return ((width & 3) == 0 && (height & 3) == 0 && depth == 1);
				default:
					return true;
			}
		}
		else
		{
			return true;
		}
	}

    void PixelUtil::getBitDepths(PixelFormat format, int rgba[4])
    {
        const PixelFormatDescription& des = getDescriptionFor(format);
        rgba[0] = des.rbits;
        rgba[1] = des.gbits;
        rgba[2] = des.bbits;
        rgba[3] = des.abits;
    }

	void PixelUtil::getBitMasks(PixelFormat format, UINT32 rgba[4])
    {
        const PixelFormatDescription& des = getDescriptionFor(format);
        rgba[0] = des.rmask;
        rgba[1] = des.gmask;
        rgba[2] = des.bmask;
        rgba[3] = des.amask;
    }

	void PixelUtil::getBitShifts(PixelFormat format, UINT8 rgba[4])
	{
		const PixelFormatDescription& des = getDescriptionFor(format);
		rgba[0] = des.rshift;
		rgba[1] = des.gshift;
		rgba[2] = des.bshift;
		rgba[3] = des.ashift;
	}

    String PixelUtil::getFormatName(PixelFormat srcformat)
    {
        return getDescriptionFor(srcformat).name;
    }

    bool PixelUtil::isAccessible(PixelFormat srcformat)
    {
        if (srcformat == PF_UNKNOWN)
            return false;

        UINT32 flags = getFlags(srcformat);
        return !((flags & PFF_COMPRESSED) || (flags & PFF_DEPTH));
    }

    PixelComponentType PixelUtil::getElementType(PixelFormat format)
    {
		const PixelFormatDescription& des = getDescriptionFor(format);
        return des.componentType;
    }

	UINT32 PixelUtil::getNumElements(PixelFormat format)
    {
		const PixelFormatDescription& des = getDescriptionFor(format);
        return des.componentCount;
    }

	UINT32 PixelUtil::getMaxMipmaps(UINT32 width, UINT32 height, UINT32 depth, PixelFormat format)
	{
		UINT32 count = 0;
        if((width > 0) && (height > 0))
        {
            do {
                if(width>1)		width = width/2;
                if(height>1)	height = height/2;
                if(depth>1)		depth = depth/2;
                    
                count ++;
            } while(!(width == 1 && height == 1 && depth == 1));
        }

		return count;
	}

    void PixelUtil::packColor(const Color& color, PixelFormat format, void* dest)
    {
        packColor(color.r, color.g, color.b, color.a, format, dest);
    }

    void PixelUtil::packColor(UINT8 r, UINT8 g, UINT8 b, UINT8 a, PixelFormat format,  void* dest)
    {
        const PixelFormatDescription &des = getDescriptionFor(format);

        if(des.flags & PFF_NATIVEENDIAN) 
		{
            // Shortcut for integer formats packing
            UINT32 value = ((Bitwise::fixedToFixed(r, 8, des.rbits)<<des.rshift) & des.rmask) |
                ((Bitwise::fixedToFixed(g, 8, des.gbits)<<des.gshift) & des.gmask) |
                ((Bitwise::fixedToFixed(b, 8, des.bbits)<<des.bshift) & des.bmask) |
                ((Bitwise::fixedToFixed(a, 8, des.abits)<<des.ashift) & des.amask);

            // And write to memory
            Bitwise::intWrite(dest, des.elemBytes, value);
        } 
		else 
		{
            // Convert to float
            packColor((float)r/255.0f,(float)g/255.0f,(float)b/255.0f,(float)a/255.0f, format, dest);
        }
    }

    void PixelUtil::packColor(float r, float g, float b, float a, const PixelFormat format, void* dest)
    {
        const PixelFormatDescription& des = getDescriptionFor(format);

        if(des.flags & PFF_NATIVEENDIAN) 
		{
            // Do the packing
            const unsigned int value = ((Bitwise::floatToFixed(r, des.rbits)<<des.rshift) & des.rmask) |
                ((Bitwise::floatToFixed(g, des.gbits)<<des.gshift) & des.gmask) |
                ((Bitwise::floatToFixed(b, des.bbits)<<des.bshift) & des.bmask) |
                ((Bitwise::floatToFixed(a, des.abits)<<des.ashift) & des.amask);

            // And write to memory
            Bitwise::intWrite(dest, des.elemBytes, value);
        }
		else 
		{
            switch(format)
            {
            case PF_FLOAT32_R:
                ((float*)dest)[0] = r;
                break;
			case PF_FLOAT32_RG:
				((float*)dest)[0] = r;
				((float*)dest)[1] = g;
				break;
            case PF_FLOAT32_RGB:
                ((float*)dest)[0] = r;
                ((float*)dest)[1] = g;
                ((float*)dest)[2] = b;
                break;
            case PF_FLOAT32_RGBA:
                ((float*)dest)[0] = r;
                ((float*)dest)[1] = g;
                ((float*)dest)[2] = b;
                ((float*)dest)[3] = a;
                break;
            case PF_FLOAT16_R:
                ((UINT16*)dest)[0] = Bitwise::floatToHalf(r);
                break;
			case PF_FLOAT16_RG:
				((UINT16*)dest)[0] = Bitwise::floatToHalf(r);
				((UINT16*)dest)[1] = Bitwise::floatToHalf(g);
				break;
            case PF_FLOAT16_RGB:
                ((UINT16*)dest)[0] = Bitwise::floatToHalf(r);
                ((UINT16*)dest)[1] = Bitwise::floatToHalf(g);
                ((UINT16*)dest)[2] = Bitwise::floatToHalf(b);
                break;
            case PF_FLOAT16_RGBA:
                ((UINT16*)dest)[0] = Bitwise::floatToHalf(r);
                ((UINT16*)dest)[1] = Bitwise::floatToHalf(g);
                ((UINT16*)dest)[2] = Bitwise::floatToHalf(b);
                ((UINT16*)dest)[3] = Bitwise::floatToHalf(a);
                break;
			case PF_R8G8:
				((UINT8*)dest)[0] = (UINT8)Bitwise::floatToFixed(r, 8);
                ((UINT8*)dest)[1] = (UINT8)Bitwise::floatToFixed(g, 8);
				break;
			case PF_R8:
				((UINT8*)dest)[0] = (UINT8)Bitwise::floatToFixed(r, 8);
				break;
            default:
                BS_EXCEPT(NotImplementedException, "Pack to " + getFormatName(format) + " not implemented");
                break;
            }
        }
    }
    void PixelUtil::unpackColor(Color* color, PixelFormat format, const void* src)
    {
		unpackColor(&color->r, &color->g, &color->b, &color->a, format, src);
    }

    void PixelUtil::unpackColor(UINT8* r, UINT8* g, UINT8* b, UINT8* a, PixelFormat format, const void* src)
    {
        const PixelFormatDescription &des = getDescriptionFor(format);

        if(des.flags & PFF_NATIVEENDIAN) 
		{
            // Shortcut for integer formats unpacking
            const UINT32 value = Bitwise::intRead(src, des.elemBytes);

            *r = (UINT8)Bitwise::fixedToFixed((value & des.rmask)>>des.rshift, des.rbits, 8);
            *g = (UINT8)Bitwise::fixedToFixed((value & des.gmask)>>des.gshift, des.gbits, 8);
            *b = (UINT8)Bitwise::fixedToFixed((value & des.bmask)>>des.bshift, des.bbits, 8);

            if(des.flags & PFF_HASALPHA)
            {
                *a = (UINT8)Bitwise::fixedToFixed((value & des.amask)>>des.ashift, des.abits, 8);
            }
            else
            {
                *a = 255; // No alpha, default a component to full
            }
        } 
		else 
		{
            // Do the operation with the more generic floating point
            float rr, gg, bb, aa;
            unpackColor(&rr,&gg,&bb,&aa, format, src);

            *r = (UINT8)Bitwise::floatToFixed(rr, 8);
            *g = (UINT8)Bitwise::floatToFixed(gg, 8);
            *b = (UINT8)Bitwise::floatToFixed(bb, 8);
            *a = (UINT8)Bitwise::floatToFixed(aa, 8);
        }
    }

    void PixelUtil::unpackColor(float* r, float* g, float* b, float* a, PixelFormat format, const void* src)
    {
        const PixelFormatDescription &des = getDescriptionFor(format);

        if(des.flags & PFF_NATIVEENDIAN) 
		{
            // Shortcut for integer formats unpacking
            const unsigned int value = Bitwise::intRead(src, des.elemBytes);

			*r = Bitwise::fixedToFloat((value & des.rmask)>>des.rshift, des.rbits);
			*g = Bitwise::fixedToFloat((value & des.gmask)>>des.gshift, des.gbits);
			*b = Bitwise::fixedToFloat((value & des.bmask)>>des.bshift, des.bbits);

            if(des.flags & PFF_HASALPHA)
            {
                *a = Bitwise::fixedToFloat((value & des.amask)>>des.ashift, des.abits);
            }
            else
            {
                *a = 1.0f; // No alpha, default a component to full
            }
        } 
		else 
		{
            switch(format)
            {
            case PF_FLOAT32_R:
                *r = *g = *b = ((float*)src)[0];
                *a = 1.0f;
                break;
			case PF_FLOAT32_RG:
				*r = ((float*)src)[0];
				*g = *b = ((float*)src)[1];
				*a = 1.0f;
				break;
            case PF_FLOAT32_RGB:
                *r = ((float*)src)[0];
                *g = ((float*)src)[1];
                *b = ((float*)src)[2];
                *a = 1.0f;
                break;
            case PF_FLOAT32_RGBA:
                *r = ((float*)src)[0];
                *g = ((float*)src)[1];
                *b = ((float*)src)[2];
                *a = ((float*)src)[3];
                break;
            case PF_FLOAT16_R:
                *r = *g = *b = Bitwise::halfToFloat(((UINT16*)src)[0]);
                *a = 1.0f;
                break;
			case PF_FLOAT16_RG:
				*r = Bitwise::halfToFloat(((UINT16*)src)[0]);
				*g = *b = Bitwise::halfToFloat(((UINT16*)src)[1]);
				*a = 1.0f;
				break;
            case PF_FLOAT16_RGB:
                *r = Bitwise::halfToFloat(((UINT16*)src)[0]);
                *g = Bitwise::halfToFloat(((UINT16*)src)[1]);
                *b = Bitwise::halfToFloat(((UINT16*)src)[2]);
                *a = 1.0f;
                break;
            case PF_FLOAT16_RGBA:
                *r = Bitwise::halfToFloat(((UINT16*)src)[0]);
                *g = Bitwise::halfToFloat(((UINT16*)src)[1]);
                *b = Bitwise::halfToFloat(((UINT16*)src)[2]);
                *a = Bitwise::halfToFloat(((UINT16*)src)[3]);
                break;
			case PF_R8G8:
				*r = Bitwise::fixedToFloat(((UINT8*)src)[0], 8);
				*g = Bitwise::fixedToFloat(((UINT8*)src)[1], 8);
				*b = 0.0f;
				*a = 1.0f;
				break;
			case PF_R8:
				*r = Bitwise::fixedToFloat(((UINT8*)src)[0], 8);
				*g = 0.0f;
				*b = 0.0f;
				*a = 1.0f;
				break;
            default:
                BS_EXCEPT(NotImplementedException, "Unpack from " + getFormatName(format) + " not implemented");
                break;
            }
        }
    }

    void PixelUtil::bulkPixelConversion(const PixelData &src, PixelData &dst)
    {
        assert(src.getWidth() == dst.getWidth() &&
			   src.getHeight() == dst.getHeight() &&
			   src.getDepth() == dst.getDepth());

		// Check for compressed formats, we don't support decompression
		if(PixelUtil::isCompressed(src.getFormat()))
		{
			if(src.getFormat() == dst.getFormat())
			{
				memcpy(dst.getData(), src.getData(), src.getConsecutiveSize());
				return;
			}
			else
			{
				BS_EXCEPT(NotImplementedException, "This method can not be used to compress or decompress images");
			}
		}

		// Check for compression
		if (PixelUtil::isCompressed(dst.getFormat()))
		{
			if (src.getFormat() == dst.getFormat())
			{
				memcpy(dst.getData(), src.getData(), src.getConsecutiveSize());
				return;
			}
			else
			{
				CompressionOptions co;
				co.format = dst.getFormat();
				compress(src, dst, co);

				return;
			}
		}

        // The easy case
        if(src.getFormat() == dst.getFormat()) 
		{
            // Everything consecutive?
            if(src.isConsecutive() && dst.isConsecutive())
            {
				memcpy(dst.getData(), src.getData(), src.getConsecutiveSize());
                return;
            }

			const UINT32 srcPixelSize = PixelUtil::getNumElemBytes(src.getFormat());
			const UINT32 dstPixelSize = PixelUtil::getNumElemBytes(dst.getFormat());
            UINT8 *srcptr = static_cast<UINT8*>(src.getData())
                + (src.getLeft() + src.getTop() * src.getRowPitch() + src.getFront() * src.getSlicePitch()) * srcPixelSize;
            UINT8 *dstptr = static_cast<UINT8*>(dst.getData())
				+ (dst.getLeft() + dst.getTop() * dst.getRowPitch() + dst.getFront() * dst.getSlicePitch()) * dstPixelSize;

            // Calculate pitches+skips in bytes
			const UINT32 srcRowPitchBytes = src.getRowPitch()*srcPixelSize;
			const UINT32 srcSliceSkipBytes = src.getSliceSkip()*srcPixelSize;

			const UINT32 dstRowPitchBytes = dst.getRowPitch()*dstPixelSize;
			const UINT32 dstSliceSkipBytes = dst.getSliceSkip()*dstPixelSize;

            // Otherwise, copy per row
			const UINT32 rowSize = src.getWidth()*srcPixelSize;
			for (UINT32 z = src.getFront(); z < src.getBack(); z++)
            {
                for(UINT32 y = src.getTop(); y < src.getBottom(); y++)
                {
					memcpy(dstptr, srcptr, rowSize);

                    srcptr += srcRowPitchBytes;
                    dstptr += dstRowPitchBytes;
                }

                srcptr += srcSliceSkipBytes;
                dstptr += dstSliceSkipBytes;
            }

            return;
        }

		// Converting to PF_X8R8G8B8 is exactly the same as converting to
		// PF_A8R8G8B8. (same with PF_X8B8G8R8 and PF_A8B8G8R8)
		if(dst.getFormat() == PF_X8R8G8B8 || dst.getFormat() == PF_X8B8G8R8)
		{
			// Do the same conversion, with PF_A8R8G8B8, which has a lot of
			// optimized conversions
			PixelFormat tempFormat = dst.getFormat() == PF_X8R8G8B8?PF_A8R8G8B8:PF_A8B8G8R8;
			PixelData tempdst(dst.getWidth(), dst.getHeight(), dst.getDepth(), tempFormat);
			bulkPixelConversion(src, tempdst);
			return;
		}

		// Converting from PF_X8R8G8B8 is exactly the same as converting from
		// PF_A8R8G8B8, given that the destination format does not have alpha.
		if((src.getFormat() == PF_X8R8G8B8 || src.getFormat() == PF_X8B8G8R8) && !hasAlpha(dst.getFormat()))
		{
			// Do the same conversion, with PF_A8R8G8B8, which has a lot of
			// optimized conversions
			PixelFormat tempFormat = src.getFormat()==PF_X8R8G8B8?PF_A8R8G8B8:PF_A8B8G8R8;
			PixelData tempsrc(src.getWidth(), src.getHeight(), src.getDepth(), tempFormat);
			tempsrc.setExternalBuffer(src.getData());
			bulkPixelConversion(tempsrc, dst);
			return;
		}

		const UINT32 srcPixelSize = PixelUtil::getNumElemBytes(src.getFormat());
		const UINT32 dstPixelSize = PixelUtil::getNumElemBytes(dst.getFormat());
        UINT8 *srcptr = static_cast<UINT8*>(src.getData())
            + (src.getLeft() + src.getTop() * src.getRowPitch() + src.getFront() * src.getSlicePitch()) * srcPixelSize;
        UINT8 *dstptr = static_cast<UINT8*>(dst.getData())
            + (dst.getLeft() + dst.getTop() * dst.getRowPitch() + dst.getFront() * dst.getSlicePitch()) * dstPixelSize;
		
        // Calculate pitches+skips in bytes
		const UINT32 srcRowSkipBytes = src.getRowSkip()*srcPixelSize;
		const UINT32 srcSliceSkipBytes = src.getSliceSkip()*srcPixelSize;
		const UINT32 dstRowSkipBytes = dst.getRowSkip()*dstPixelSize;
		const UINT32 dstSliceSkipBytes = dst.getSliceSkip()*dstPixelSize;

        // The brute force fallback
        float r,g,b,a;
		for (UINT32 z = src.getFront(); z<src.getBack(); z++)
		{
			for (UINT32 y = src.getTop(); y < src.getBottom(); y++)
            {
				for (UINT32 x = src.getLeft(); x<src.getRight(); x++)
                {
                    unpackColor(&r, &g, &b, &a, src.getFormat(), srcptr);
                    packColor(r, g, b, a, dst.getFormat(), dstptr);

                    srcptr += srcPixelSize;
                    dstptr += dstPixelSize;
                }

                srcptr += srcRowSkipBytes;
                dstptr += dstRowSkipBytes;
            }

            srcptr += srcSliceSkipBytes;
            dstptr += dstSliceSkipBytes;
        }
    }

	void PixelUtil::scale(const PixelData& src, PixelData& scaled, Filter filter)
	{
		assert(PixelUtil::isAccessible(src.getFormat()));
		assert(PixelUtil::isAccessible(scaled.getFormat()));

		PixelData temp;
		switch (filter) 
		{
		default:
		case FILTER_NEAREST:
			if(src.getFormat() == scaled.getFormat()) 
			{
				// No intermediate buffer needed
				temp = scaled;
			}
			else
			{
				// Allocate temporary buffer of destination size in source format 
				temp = PixelData(scaled.getWidth(), scaled.getHeight(), scaled.getDepth(), src.getFormat());
				temp.allocateInternalBuffer();
			}

			// No conversion
			switch (PixelUtil::getNumElemBytes(src.getFormat())) 
			{
			case 1: NearestResampler<1>::scale(src, temp); break;
			case 2: NearestResampler<2>::scale(src, temp); break;
			case 3: NearestResampler<3>::scale(src, temp); break;
			case 4: NearestResampler<4>::scale(src, temp); break;
			case 6: NearestResampler<6>::scale(src, temp); break;
			case 8: NearestResampler<8>::scale(src, temp); break;
			case 12: NearestResampler<12>::scale(src, temp); break;
			case 16: NearestResampler<16>::scale(src, temp); break;
			default:
				// Never reached
				assert(false);
			}

			if(temp.getData() != scaled.getData())
			{
				// Blit temp buffer
				PixelUtil::bulkPixelConversion(temp, scaled);

				temp.freeInternalBuffer();
			}

			break;

		case FILTER_LINEAR:
			switch (src.getFormat()) 
			{
			case PF_R8G8:
			case PF_R8G8B8: case PF_B8G8R8:
			case PF_R8G8B8A8: case PF_B8G8R8A8:
			case PF_A8B8G8R8: case PF_A8R8G8B8:
			case PF_X8B8G8R8: case PF_X8R8G8B8:
				if(src.getFormat() == scaled.getFormat()) 
				{
					// No intermediate buffer needed
					temp = scaled;
				}
				else
				{
					// Allocate temp buffer of destination size in source format 
					temp = PixelData(scaled.getWidth(), scaled.getHeight(), scaled.getDepth(), src.getFormat());
					temp.allocateInternalBuffer();
				}

				// No conversion
				switch (PixelUtil::getNumElemBytes(src.getFormat())) 
				{
				case 1: LinearResampler_Byte<1>::scale(src, temp); break;
				case 2: LinearResampler_Byte<2>::scale(src, temp); break;
				case 3: LinearResampler_Byte<3>::scale(src, temp); break;
				case 4: LinearResampler_Byte<4>::scale(src, temp); break;
				default:
					// Never reached
					assert(false);
				}

				if(temp.getData() != scaled.getData())
				{
					// Blit temp buffer
					PixelUtil::bulkPixelConversion(temp, scaled);
					temp.freeInternalBuffer();
				}

				break;
			case PF_FLOAT32_RGB:
			case PF_FLOAT32_RGBA:
				if (scaled.getFormat() == PF_FLOAT32_RGB || scaled.getFormat() == PF_FLOAT32_RGBA)
				{
					// float32 to float32, avoid unpack/repack overhead
					LinearResampler_Float32::scale(src, scaled);
					break;
				}
				// Else, fall through
			default:
				// Fallback case, slow but works
				LinearResampler::scale(src, scaled);
			}
			break;
		}
	}

	void PixelUtil::applyGamma(UINT8* buffer, float gamma, UINT32 size, UINT8 bpp)
	{
		if(gamma == 1.0f)
			return;

		UINT32 stride = bpp >> 3;

		for(size_t i = 0, j = size / stride; i < j; i++, buffer += stride)
		{
			float r = (float)buffer[0];
			float g = (float)buffer[1];
			float b = (float)buffer[2];

			r = r * gamma;
			g = g * gamma;
			b = b * gamma;

			float scale = 1.0f;
			float tmp = 0.0f;

			if(r > 255.0f && (tmp=(255.0f/r)) < scale)
				scale = tmp;

			if(g > 255.0f && (tmp=(255.0f/g)) < scale)
				scale = tmp;

			if(b > 255.0f && (tmp=(255.0f/b)) < scale)
				scale = tmp;

			r *= scale; 
			g *= scale; 
			b *= scale;

			buffer[0] = (UINT8)r;
			buffer[1] = (UINT8)g;
			buffer[2] = (UINT8)b;
		}
	}

	void PixelUtil::compress(const PixelData& src, PixelData& dst, const CompressionOptions& options)
	{
		if (!isCompressed(options.format))
			BS_EXCEPT(InvalidParametersException, "Wanted format is not a compressed format.");

		// Note: NVTT site has implementations for these two formats for when I decide to add them
		if (options.format == PF_BC6H || options.format == PF_BC7)
			BS_EXCEPT(InvalidParametersException, "Specified formats are not yet supported.");

		if (src.getDepth() != 1)
			BS_EXCEPT(InvalidParametersException, "3D textures are not supported.");

		PixelFormat pf = options.format;

		if (isCompressed(src.getFormat()))
			BS_EXCEPT(InvalidParametersException, "Source data cannot be compressed.");

		PixelData bgraData(src.getWidth(), src.getHeight(), 1, PF_B8G8R8A8);
		bgraData.allocateInternalBuffer();
		bulkPixelConversion(src, bgraData);

		nvtt::InputOptions io;
		io.setTextureLayout(nvtt::TextureType_2D, src.getWidth(), src.getHeight());
		io.setMipmapData(bgraData.getData(), src.getWidth(), src.getHeight());
		io.setMipmapGeneration(false);
		io.setAlphaMode(toNVTTAlphaMode(options.alphaMode));
		io.setNormalMap(options.isNormalMap);

		if (options.isSRGB)
			io.setGamma(2.2f, 2.2f);
		else
			io.setGamma(1.0f, 1.0f);

		nvtt::CompressionOptions co;
		co.setFormat(toNVTTFormat(options.format));
		co.setQuality(toNVTTQuality(options.quality));
		
		NVTTCompressOutputHandler outputHandler(dst.getData(), dst.getConsecutiveSize());

		nvtt::OutputOptions oo;
		oo.setOutputHeader(false);
		oo.setOutputHandler(&outputHandler);
		
		nvtt::Compressor compressor;
		if (!compressor.process(io, co, oo))
			BS_EXCEPT(InternalErrorException, "Compressing failed.");
	}

	Vector<PixelDataPtr> PixelUtil::genMipmaps(const PixelData& src, const MipMapGenOptions& options)
	{
		if (src.getDepth() != 1)
			BS_EXCEPT(InvalidParametersException, "3D textures are not supported.");

		// Note: Add support for floating point mips, no reason they shouldn't be supported other than
		// nvtt doesn't support them natively
		if (isCompressed(src.getFormat()) || isFloatingPoint(src.getFormat()))
			BS_EXCEPT(InvalidParametersException, "Source data cannot be compressed or in floating point format.");

		if (!Math::isPow2(src.getWidth()) || !Math::isPow2(src.getHeight()))
			BS_EXCEPT(InvalidParametersException, "Texture width & height must be powers of 2.");

		PixelData argbData(src.getWidth(), src.getHeight(), 1, PF_A8R8G8B8);
		argbData.allocateInternalBuffer();
		bulkPixelConversion(src, argbData);

		nvtt::InputOptions io;
		io.setTextureLayout(nvtt::TextureType_2D, src.getWidth(), src.getHeight());
		io.setMipmapData(argbData.getData(), src.getWidth(), src.getHeight());
		io.setMipmapGeneration(true);
		io.setNormalMap(options.isNormalMap);
		io.setNormalizeMipmaps(options.normalizeMipmaps);
		io.setWrapMode(toNVTTWrapMode(options.wrapMode));

		nvtt::CompressionOptions co;
		co.setFormat(nvtt::Format_RGBA);
		
		UINT32 numMips = getMaxMipmaps(src.getWidth(), src.getHeight(), 1, src.getFormat());

		Vector<PixelDataPtr> argbMipBuffers;

		// Note: This can be done more effectively without creating so many temp buffers
		// and working with the original formats directly, but it would complicate the code
		// too much at the moment.
		UINT32 curWidth = src.getWidth();
		UINT32 curHeight = src.getHeight();
		for (UINT32 i = 0; i < numMips; i++)
		{
			argbMipBuffers.push_back(bs_shared_ptr<PixelData>(curWidth, curHeight, 1, PF_A8R8G8B8));
			argbMipBuffers.back()->allocateInternalBuffer();

			if (curWidth > 1) 
				curWidth = curWidth / 2;

			if (curHeight > 1)
				curHeight = curHeight / 2;
		}

		argbMipBuffers.push_back(bs_shared_ptr<PixelData>(curWidth, curHeight, 1, PF_A8R8G8B8));
		argbMipBuffers.back()->allocateInternalBuffer();

		NVTTMipmapOutputHandler outputHandler(argbMipBuffers);

		nvtt::OutputOptions oo;
		oo.setOutputHeader(false);
		oo.setOutputHandler(&outputHandler);

		nvtt::Compressor compressor;
		if (!compressor.process(io, co, oo))
			BS_EXCEPT(InternalErrorException, "Mipmap generation failed.");

		argbData.freeInternalBuffer();

		Vector<PixelDataPtr> outputMipBuffers;
		for (UINT32 i = 0; i < (UINT32)argbMipBuffers.size(); i++)
		{
			PixelDataPtr argbBuffer = argbMipBuffers[i];
			PixelDataPtr outputBuffer = bs_shared_ptr<PixelData>(argbBuffer->getWidth(), argbBuffer->getHeight(), 1, src.getFormat());
			outputBuffer->allocateInternalBuffer();

			bulkPixelConversion(*argbBuffer, *outputBuffer);
			argbBuffer->freeInternalBuffer();

			outputMipBuffers.push_back(outputBuffer);
		}

		return outputMipBuffers;
	}
}