/*!
\brief Implementation of the Texture Decompression functions.
\file PVRCore/texture/PVRTDecompress.cpp
\author PowerVR by Imagination, Developer Technology Team
\copyright Copyright (c) Imagination Technologies Limited.
*/
//!\cond NO_DOXYGEN

#include <cstdlib>
#include <cstdio>
#include <climits>
#include <cmath>
#include <algorithm>
#include <cstring>
#include "PVRTDecompress.h"
#include <cassert>
#include <vector>

// rom-properties: Use librpbyteswap's byteorder macros.
// NOTE: Not able to detect built-in byteswapping intrinsics here.
#include "../../src/librpbyteswap/byteorder.h"
#define __swab32(x) \
	((uint32_t)((((uint32_t)x) << 24) | (((uint32_t)x) >> 24) | \
		((((uint32_t)x) & 0x0000FF00UL) << 8) | \
		((((uint32_t)x) & 0x00FF0000UL) >> 8)))
#if SYS_BYTEORDER == SYS_LIL_ENDIAN
#  define le32_to_cpu(x) (x)
#  define cpu_to_le32(x) (x)
#else /* SYS_BYTEORDER == SYS_BIG_ENDIAN */
#  define le32_to_cpu(x) __swab32(x)
#  define cpu_to_le32(x) __swab32(x)
#endif

namespace pvr {
struct Pixel32
{
#if SYS_BYTEORDER == SYS_LIL_ENDIAN
#  ifdef PVRTC_SWAP_R_B_CHANNELS
	uint8_t blue, green, red, alpha;
#  else /* !PVRTC_SWAP_R_B_CHANNELS */
	uint8_t red, green, blue, alpha;
#  endif /* PVRTC_SWAP_R_B_CHANNELS */
#else /* SYS_BYTEORDER == SYS_BIG_ENDIAN */
#  ifdef PVRTC_SWAP_R_B_CHANNELS
	uint8_t alpha, red, green, blue;
#  else /* !PVRTC_SWAP_R_B_CHANNELS */
	uint8_t alpha, blue, green, red;
#  endif /* PVRTC_SWAP_R_B_CHANNELS */
#endif
};

struct Pixel128S
{
	int32_t red, green, blue, alpha;
};

struct PVRTCWord
{
	uint32_t modulationData;
	uint32_t colorData;
};

struct PVRTCWordIndices
{
	int P[2], Q[2], R[2], S[2];
};

template<bool PVRTCII>
static Pixel32 getColorA(uint32_t colorData)
{
	Pixel32 color;

	// Opaque Color Mode - RGB 554
	const uint32_t opaque_flag = (PVRTCII ? 0x80000000 : 0x8000);
	if ((colorData & opaque_flag) != 0)
	{
		color.red = static_cast<uint8_t>((colorData & 0x7c00) >> 10); // 5->5 bits
		color.green = static_cast<uint8_t>((colorData & 0x3e0) >> 5); // 5->5 bits
		color.blue = static_cast<uint8_t>(colorData & 0x1e) | ((colorData & 0x1e) >> 4); // 4->5 bits
		color.alpha = static_cast<uint8_t>(0xf); // 0->4 bits
	}
	// Transparent Color Mode - ARGB 3443
	else
	{
		color.red = static_cast<uint8_t>((colorData & 0xf00) >> 7) | ((colorData & 0xf00) >> 11); // 4->5 bits
		color.green = static_cast<uint8_t>((colorData & 0xf0) >> 3) | ((colorData & 0xf0) >> 7); // 4->5 bits
		color.blue = static_cast<uint8_t>((colorData & 0xe) << 1) | ((colorData & 0xe) >> 2); // 3->5 bits
		color.alpha = static_cast<uint8_t>((colorData & 0x7000) >> 11); // 3->4 bits - note 0 at right
	}

	return color;
}

template<bool PVRTCII>
static Pixel32 getColorB(uint32_t colorData)
{
	Pixel32 color;

	// Opaque Color Mode - RGB 555
	if (colorData & 0x80000000)
	{
		color.red = static_cast<uint8_t>((colorData & 0x7c000000) >> 26); // 5->5 bits
		color.green = static_cast<uint8_t>((colorData & 0x3e00000) >> 21); // 5->5 bits
		color.blue = static_cast<uint8_t>((colorData & 0x1f0000) >> 16); // 5->5 bits
		color.alpha = static_cast<uint8_t>(0xf); // 0 bits
	}
	// Transparent Color Mode - ARGB 3444
	else
	{
		color.red = static_cast<uint8_t>(((colorData & 0xf000000) >> 23) | ((colorData & 0xf000000) >> 27)); // 4->5 bits
		color.green = static_cast<uint8_t>(((colorData & 0xf00000) >> 19) | ((colorData & 0xf00000) >> 23)); // 4->5 bits
		color.blue = static_cast<uint8_t>(((colorData & 0xf0000) >> 15) | ((colorData & 0xf0000) >> 19)); // 4->5 bits
		color.alpha = static_cast<uint8_t>((colorData & 0x70000000) >> 27); // 3->4 bits - note 0 at right
		if (PVRTCII) {
			// PVRTC-II sets the low alpha bit of Color B to 1, not 0.
			color.alpha |= 1;
		}
	}

	return color;
}

static void interpolateColors(Pixel32 P, Pixel32 Q, Pixel32 R, Pixel32 S, Pixel128S* pPixel, uint8_t bpp)
{
	uint32_t wordWidth = 4;
	uint32_t wordHeight = 4;
	if (bpp == 2) { wordWidth = 8; }

	// Convert to int 32.
	Pixel128S hP = { static_cast<int32_t>(P.red), static_cast<int32_t>(P.green), static_cast<int32_t>(P.blue), static_cast<int32_t>(P.alpha) };
	Pixel128S hQ = { static_cast<int32_t>(Q.red), static_cast<int32_t>(Q.green), static_cast<int32_t>(Q.blue), static_cast<int32_t>(Q.alpha) };
	Pixel128S hR = { static_cast<int32_t>(R.red), static_cast<int32_t>(R.green), static_cast<int32_t>(R.blue), static_cast<int32_t>(R.alpha) };
	Pixel128S hS = { static_cast<int32_t>(S.red), static_cast<int32_t>(S.green), static_cast<int32_t>(S.blue), static_cast<int32_t>(S.alpha) };

	// Get vectors.
	Pixel128S QminusP = { hQ.red - hP.red, hQ.green - hP.green, hQ.blue - hP.blue, hQ.alpha - hP.alpha };
	Pixel128S SminusR = { hS.red - hR.red, hS.green - hR.green, hS.blue - hR.blue, hS.alpha - hR.alpha };

	// Multiply colors.
	hP.red *= wordWidth;
	hP.green *= wordWidth;
	hP.blue *= wordWidth;
	hP.alpha *= wordWidth;
	hR.red *= wordWidth;
	hR.green *= wordWidth;
	hR.blue *= wordWidth;
	hR.alpha *= wordWidth;

	if (bpp == 2)
	{
		// Loop through pixels to achieve results.
		for (uint32_t x = 0; x < wordWidth; x++)
		{
			Pixel128S result = { 4 * hP.red, 4 * hP.green, 4 * hP.blue, 4 * hP.alpha };
			Pixel128S dY = { hR.red - hP.red, hR.green - hP.green, hR.blue - hP.blue, hR.alpha - hP.alpha };

			for (uint32_t y = 0; y < wordHeight; y++)
			{
				pPixel[y * wordWidth + x].red = static_cast<int32_t>((result.red >> 7) + (result.red >> 2));
				pPixel[y * wordWidth + x].green = static_cast<int32_t>((result.green >> 7) + (result.green >> 2));
				pPixel[y * wordWidth + x].blue = static_cast<int32_t>((result.blue >> 7) + (result.blue >> 2));
				pPixel[y * wordWidth + x].alpha = static_cast<int32_t>((result.alpha >> 5) + (result.alpha >> 1));

				result.red += dY.red;
				result.green += dY.green;
				result.blue += dY.blue;
				result.alpha += dY.alpha;
			}

			hP.red += QminusP.red;
			hP.green += QminusP.green;
			hP.blue += QminusP.blue;
			hP.alpha += QminusP.alpha;

			hR.red += SminusR.red;
			hR.green += SminusR.green;
			hR.blue += SminusR.blue;
			hR.alpha += SminusR.alpha;
		}
	}
	else
	{
		// Loop through pixels to achieve results.
		for (uint32_t y = 0; y < wordHeight; y++)
		{
			Pixel128S result = { 4 * hP.red, 4 * hP.green, 4 * hP.blue, 4 * hP.alpha };
			Pixel128S dY = { hR.red - hP.red, hR.green - hP.green, hR.blue - hP.blue, hR.alpha - hP.alpha };

			for (uint32_t x = 0; x < wordWidth; x++)
			{
				pPixel[y * wordWidth + x].red = static_cast<int32_t>((result.red >> 6) + (result.red >> 1));
				pPixel[y * wordWidth + x].green = static_cast<int32_t>((result.green >> 6) + (result.green >> 1));
				pPixel[y * wordWidth + x].blue = static_cast<int32_t>((result.blue >> 6) + (result.blue >> 1));
				pPixel[y * wordWidth + x].alpha = static_cast<int32_t>((result.alpha >> 4) + (result.alpha));

				result.red += dY.red;
				result.green += dY.green;
				result.blue += dY.blue;
				result.alpha += dY.alpha;
			}

			hP.red += QminusP.red;
			hP.green += QminusP.green;
			hP.blue += QminusP.blue;
			hP.alpha += QminusP.alpha;

			hR.red += SminusR.red;
			hR.green += SminusR.green;
			hR.blue += SminusR.blue;
			hR.alpha += SminusR.alpha;
		}
	}
}

static void unpackModulations(const PVRTCWord& word, int32_t offsetX, int32_t offsetY, int32_t modulationValues[16][8], int32_t modulationModes[16][8], uint8_t bpp)
{
	uint32_t WordModMode = word.colorData & 0x1;
	uint32_t ModulationBits = word.modulationData;

	// Unpack differently depending on 2bpp or 4bpp modes.
	if (bpp == 2)
	{
		if (WordModMode)
		{
			// determine which of the three modes are in use:

			// If this is the either the H-only or V-only interpolation mode...
			if (ModulationBits & 0x1)
			{
				// look at the "LSB" for the "centre" (V=2,H=4) texel. Its LSB is now
				// actually used to indicate whether it's the H-only mode or the V-only...

				// The centre texel data is the at (y==2, x==4) and so its LSB is at bit 20.
				if (ModulationBits & (0x1 << 20))
				{
					// This is the V-only mode
					WordModMode = 3;
				}
				else
				{
					// This is the H-only mode
					WordModMode = 2;
				}

				// Create an extra bit for the centre pixel so that it looks like
				// we have 2 actual bits for this texel. It makes later coding much easier.
				if (ModulationBits & (0x1 << 21))
				{
					// set it to produce code for 1.0
					ModulationBits |= (0x1 << 20);
				}
				else
				{
					// clear it to produce 0.0 code
					ModulationBits &= ~(0x1 << 20);
				}
			} // end if H-Only or V-Only interpolation mode was chosen

			if (ModulationBits & 0x2) { ModulationBits |= 0x1; /*set it*/ }
			else
			{
				ModulationBits &= ~0x1; /*clear it*/
			}

			// run through all the pixels in the block. Note we can now treat all the
			// "stored" values as if they have 2bits (even when they didn't!)
			for (uint8_t y = 0; y < 4; y++)
			{
				for (uint8_t x = 0; x < 8; x++)
				{
					modulationModes[static_cast<uint32_t>(x + offsetX)][static_cast<uint32_t>(y + offsetY)] = WordModMode;

					// if this is a stored value...
					if (((x ^ y) & 1) == 0) {modulationValues[static_cast<uint32_t>(x + offsetX)][static_cast<uint32_t>(y + offsetY)] = ModulationBits & 3;
						ModulationBits >>= 2;
					}
				}
			} // end for y
		}
		// else if direct encoded 2bit mode - i.e. 1 mode bit per pixel
		else
		{
			for (uint8_t y = 0; y < 4; y++)
			{
				for (uint8_t x = 0; x < 8; x++)
				{
					modulationModes[static_cast<uint32_t>(x + offsetX)][static_cast<uint32_t>(y + offsetY)] = WordModMode;

					/*
					// double the bits so 0=> 00, and 1=>11
					*/
					if (ModulationBits & 1) { modulationValues[static_cast<uint32_t>(x + offsetX)][static_cast<uint32_t>(y + offsetY)] = 0x3; }
					else
					{
						modulationValues[static_cast<uint32_t>(x + offsetX)][static_cast<uint32_t>(y + offsetY)] = 0x0;
					}
					ModulationBits >>= 1;
				}
			} // end for y
		}
	}
	else
	{
		// Much simpler than the 2bpp decompression, only two modes, so the n/8 values are set directly.
		// run through all the pixels in the word.
		if (WordModMode)
		{
			for (uint8_t y = 0; y < 4; y++)
			{
				for (uint8_t x = 0; x < 4; x++)
				{
					modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] = ModulationBits & 3;
					// if (modulationValues==0) {}. We don't need to check 0, 0 = 0/8.
					if (modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] == 1)
					{ modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] = 4; }
					else if (modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] == 2)
					{
						modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] = 14; //+10 tells the decompressor to punch through alpha.
					}
					else if (modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] == 3)
					{
						modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] = 8;
					}
					ModulationBits >>= 2;
				} // end for x
			} // end for y
		}
		else
		{
			for (uint8_t y = 0; y < 4; y++)
			{
				for (uint8_t x = 0; x < 4; x++)
				{
					modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] = ModulationBits & 3;
					modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] *= 3;
					if (modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] > 3)
					{ modulationValues[static_cast<uint32_t>(y + offsetY)][static_cast<uint32_t>(x + offsetX)] -= 1; }
					ModulationBits >>= 2;
				} // end for x
			} // end for y
		}
	}
}

static int32_t getModulationValues(int32_t modulationValues[16][8], int32_t modulationModes[16][8], uint32_t xPos, uint32_t yPos, uint8_t bpp)
{
	if (bpp == 2)
	{
		static const uint8_t RepVals0[4] = { 0, 3, 5, 8 };

		// extract the modulation value. If a simple encoding
		if (modulationModes[xPos][yPos] == 0) { return RepVals0[modulationValues[xPos][yPos]]; }
		else
		{
			// if this is a stored value
			if (((xPos ^ yPos) & 1) == 0) { return RepVals0[modulationValues[xPos][yPos]]; }

			// else average from the neighbours
			// if H&V interpolation...
			else if (modulationModes[xPos][yPos] == 1)
			{
				return (RepVals0[modulationValues[xPos][yPos - 1]] + RepVals0[modulationValues[xPos][yPos + 1]] + RepVals0[modulationValues[xPos - 1][yPos]] +
						   RepVals0[modulationValues[xPos + 1][yPos]] + 2) /
					4;
			}
			// else if H-Only
			else if (modulationModes[xPos][yPos] == 2)
			{
				return (RepVals0[modulationValues[xPos - 1][yPos]] + RepVals0[modulationValues[xPos + 1][yPos]] + 1) / 2;
			}
			// else it's V-Only
			else
			{
				return (RepVals0[modulationValues[xPos][yPos - 1]] + RepVals0[modulationValues[xPos][yPos + 1]] + 1) / 2;
			}
		}
	}
	else if (bpp == 4)
	{
		return modulationValues[xPos][yPos];
	}

	return 0;
}

template<bool PVRTCII>
static void pvrtcGetDecompressedPixels(const PVRTCWord& P, const PVRTCWord& Q, const PVRTCWord& R, const PVRTCWord& S, Pixel32* pColorData, uint8_t bpp)
{
	// 4bpp only needs 8*8 values, but 2bpp needs 16*8, so rather than wasting processor time we just statically allocate 16*8.
	int32_t modulationValues[16][8];
	// Only 2bpp needs this.
	int32_t modulationModes[16][8];
	// 4bpp only needs 16 values, but 2bpp needs 32, so rather than wasting processor time we just statically allocate 32.
	Pixel128S upscaledColorA[32];
	Pixel128S upscaledColorB[32];

	uint32_t wordWidth = 4;
	uint32_t wordHeight = 4;
	if (bpp == 2) { wordWidth = 8; }

	// Get the modulations from each word.
	unpackModulations(P, 0, 0, modulationValues, modulationModes, bpp);
	unpackModulations(Q, wordWidth, 0, modulationValues, modulationModes, bpp);
	unpackModulations(R, 0, wordHeight, modulationValues, modulationModes, bpp);
	unpackModulations(S, wordWidth, wordHeight, modulationValues, modulationModes, bpp);

	// Bilinear upscale image data from 2x2 -> 4x4
	interpolateColors(getColorA<PVRTCII>(P.colorData), getColorA<PVRTCII>(Q.colorData),
		getColorA<PVRTCII>(R.colorData), getColorA<PVRTCII>(S.colorData), upscaledColorA, bpp);
	interpolateColors(getColorB<PVRTCII>(P.colorData), getColorB<PVRTCII>(Q.colorData),
		getColorB<PVRTCII>(R.colorData), getColorB<PVRTCII>(S.colorData), upscaledColorB, bpp);

	for (uint32_t y = 0; y < wordHeight; y++)
	{
		for (uint32_t x = 0; x < wordWidth; x++)
		{
			int32_t mod = getModulationValues(modulationValues, modulationModes, x + wordWidth / 2, y + wordHeight / 2, bpp);
			bool punchthroughAlpha = false;
			if (mod > 10)
			{
				punchthroughAlpha = true;
				mod -= 10;
			}

			Pixel128S result;
			if (PVRTCII && punchthroughAlpha)
			{
				// PVRTC-II: Punch-through alpha sets the RGB values to 0.
				result.red = 0;
				result.green = 0;
				result.blue = 0;
				result.alpha = 0;
			}
			else
			{
				result.red = (upscaledColorA[y * wordWidth + x].red * (8 - mod) + upscaledColorB[y * wordWidth + x].red * mod) / 8;
				result.green = (upscaledColorA[y * wordWidth + x].green * (8 - mod) + upscaledColorB[y * wordWidth + x].green * mod) / 8;
				result.blue = (upscaledColorA[y * wordWidth + x].blue * (8 - mod) + upscaledColorB[y * wordWidth + x].blue * mod) / 8;
				if (punchthroughAlpha) { result.alpha = 0; }
				else
				{
					result.alpha = (upscaledColorA[y * wordWidth + x].alpha * (8 - mod) + upscaledColorB[y * wordWidth + x].alpha * mod) / 8;
				}
			}

			// Convert the 32bit precision Result to 8 bit per channel color.
			if (bpp == 2)
			{
				pColorData[y * wordWidth + x].red = static_cast<uint8_t>(result.red);
				pColorData[y * wordWidth + x].green = static_cast<uint8_t>(result.green);
				pColorData[y * wordWidth + x].blue = static_cast<uint8_t>(result.blue);
				pColorData[y * wordWidth + x].alpha = static_cast<uint8_t>(result.alpha);
			}
			else if (bpp == 4)
			{
				pColorData[y + x * wordHeight].red = static_cast<uint8_t>(result.red);
				pColorData[y + x * wordHeight].green = static_cast<uint8_t>(result.green);
				pColorData[y + x * wordHeight].blue = static_cast<uint8_t>(result.blue);
				pColorData[y + x * wordHeight].alpha = static_cast<uint8_t>(result.alpha);
			}
		}
	}
}

static uint32_t wrapWordIndex(uint32_t numWords, int word) { return ((word + numWords) % numWords); }

static bool isPowerOf2(uint32_t input)
{
	uint32_t minus1;

	if (!input) { return 0; }

	minus1 = input - 1;
	return ((input | minus1) == (input ^ minus1));
}

template<bool PVRTCII>
static uint32_t TwiddleUV(uint32_t XSize, uint32_t YSize, uint32_t XPos, uint32_t YPos)
{
	// Check the sizes are valid.
	assert(YPos < YSize);
	assert(XPos < XSize);
	assert(isPowerOf2(YSize));
	assert(isPowerOf2(XSize));

	if (PVRTCII) {
		// PVRTC-II uses linear order, not Morton order.
		return (YPos * XSize) + XPos;
	} else {
		// Initially assume X is the larger size.
		uint32_t MinDimension = XSize;
		uint32_t MaxValue = YPos;
		uint32_t Twiddled = 0;
		uint32_t SrcBitPos = 1;
		uint32_t DstBitPos = 1;
		int ShiftCount = 0;

		// If Y is the larger dimension - switch the min/max values.
		if (YSize < XSize)
		{
			MinDimension = YSize;
			MaxValue = XPos;
		}

		// Step through all the bits in the "minimum" dimension
		while (SrcBitPos < MinDimension)
		{
			if (YPos & SrcBitPos) { Twiddled |= DstBitPos; }

			if (XPos & SrcBitPos) { Twiddled |= (DstBitPos << 1); }

			SrcBitPos <<= 1;
			DstBitPos <<= 2;
			ShiftCount += 1;
		}

		// Prepend any unused bits
		MaxValue >>= ShiftCount;
		Twiddled |= (MaxValue << (2 * ShiftCount));

		return Twiddled;
	}
}

static void mapDecompressedData(Pixel32* pOutput, uint32_t width, const Pixel32* pWord, const PVRTCWordIndices& words, uint8_t bpp)
{
	uint32_t wordWidth = 4;
	uint32_t wordHeight = 4;
	if (bpp == 2) { wordWidth = 8; }

	for (uint32_t y = 0; y < wordHeight / 2; y++)
	{
		for (uint32_t x = 0; x < wordWidth / 2; x++)
		{
			pOutput[(((words.P[1] * wordHeight) + y + wordHeight / 2) * width + words.P[0] * wordWidth + x + wordWidth / 2)] = pWord[y * wordWidth + x]; // map P

			pOutput[(((words.Q[1] * wordHeight) + y + wordHeight / 2) * width + words.Q[0] * wordWidth + x)] = pWord[y * wordWidth + x + wordWidth / 2]; // map Q

			pOutput[(((words.R[1] * wordHeight) + y) * width + words.R[0] * wordWidth + x + wordWidth / 2)] = pWord[(y + wordHeight / 2) * wordWidth + x]; // map R

			pOutput[(((words.S[1] * wordHeight) + y) * width + words.S[0] * wordWidth + x)] = pWord[(y + wordHeight / 2) * wordWidth + x + wordWidth / 2]; // map S
		}
	}
}
template<bool PVRTCII>
static uint32_t pvrtcDecompress(uint8_t* pCompressedData, Pixel32* pDecompressedData, uint32_t width, uint32_t height, uint8_t bpp)
{
	uint32_t wordWidth = 4;
	uint32_t wordHeight = 4;
	if (bpp == 2) { wordWidth = 8; }

	uint32_t* pWordMembers = (uint32_t*)pCompressedData;
	Pixel32* pOutData = pDecompressedData;

	// Calculate number of words
	int i32NumXWords = static_cast<int>(width / wordWidth);
	int i32NumYWords = static_cast<int>(height / wordHeight);

	// Structs used for decompression
	PVRTCWordIndices indices;
	std::vector<Pixel32> pPixels(wordWidth * wordHeight * sizeof(Pixel32));

	// For each row of words
	for (int32_t wordY = -1; wordY < i32NumYWords - 1; wordY++)
	{
		// for each column of words
		for (int32_t wordX = -1; wordX < i32NumXWords - 1; wordX++)
		{
			indices.P[0] = static_cast<int>(wrapWordIndex(i32NumXWords, wordX));
			indices.P[1] = static_cast<int>(wrapWordIndex(i32NumYWords, wordY));
			indices.Q[0] = static_cast<int>(wrapWordIndex(i32NumXWords, wordX + 1));
			indices.Q[1] = static_cast<int>(wrapWordIndex(i32NumYWords, wordY));
			indices.R[0] = static_cast<int>(wrapWordIndex(i32NumXWords, wordX));
			indices.R[1] = static_cast<int>(wrapWordIndex(i32NumYWords, wordY + 1));
			indices.S[0] = static_cast<int>(wrapWordIndex(i32NumXWords, wordX + 1));
			indices.S[1] = static_cast<int>(wrapWordIndex(i32NumYWords, wordY + 1));

			// Work out the offsets into the twiddle structs, multiply by two as there are two members per word.
			uint32_t WordOffsets[4] = {
				TwiddleUV<PVRTCII>(i32NumXWords, i32NumYWords, indices.P[0], indices.P[1]) * 2,
				TwiddleUV<PVRTCII>(i32NumXWords, i32NumYWords, indices.Q[0], indices.Q[1]) * 2,
				TwiddleUV<PVRTCII>(i32NumXWords, i32NumYWords, indices.R[0], indices.R[1]) * 2,
				TwiddleUV<PVRTCII>(i32NumXWords, i32NumYWords, indices.S[0], indices.S[1]) * 2,
			};

			// Access individual elements to fill out PVRTCWord
			PVRTCWord P, Q, R, S;
			P.colorData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[0] + 1]));
			P.modulationData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[0]]));
			Q.colorData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[1] + 1]));
			Q.modulationData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[1]]));
			R.colorData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[2] + 1]));
			R.modulationData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[2]]));
			S.colorData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[3] + 1]));
			S.modulationData = static_cast<uint32_t>(le32_to_cpu(pWordMembers[WordOffsets[3]]));

			// assemble 4 words into struct to get decompressed pixels from
			pvrtcGetDecompressedPixels<PVRTCII>(P, Q, R, S, pPixels.data(), bpp);
			mapDecompressedData(pOutData, width, pPixels.data(), indices, bpp);

		} // for each word
	} // for each row of words

	// Return the data size
	return width * height / static_cast<uint32_t>((wordWidth / 2));
}

template<bool PVRTCII>
static uint32_t PVRTDecompressPVRTC_int(const void* pCompressedData, uint32_t Do2bitMode, uint32_t XDim, uint32_t YDim, uint8_t* pResultImage)
{
	// Cast the output buffer to a Pixel32 pointer.
	Pixel32* pDecompressedData = (Pixel32*)pResultImage;

	// Check the X and Y values are at least the minimum size.
	uint32_t XTrueDim = std::max(XDim, ((Do2bitMode == 1u) ? 16u : 8u));
	uint32_t YTrueDim = std::max(YDim, 8u);

	// If the dimensions aren't correct, we need to create a new buffer instead of just using the provided one, as the buffer will overrun otherwise.
	// rom-properties: make sure we don't hit this case
	assert(XTrueDim == XDim);
	assert(YTrueDim == YDim);
	if (XTrueDim != XDim || YTrueDim != YDim) { pDecompressedData = new Pixel32[XTrueDim * YTrueDim]; }

	// Decompress the surface.
	uint32_t retval = pvrtcDecompress<PVRTCII>((uint8_t*)pCompressedData,
		pDecompressedData, XTrueDim, YTrueDim, uint8_t(Do2bitMode == 1 ? 2 : 4));

	// If the dimensions were too small, then copy the new buffer back into the output buffer.
	if (XTrueDim != XDim || YTrueDim != YDim)
	{
		// Loop through all the required pixels.
		for (uint32_t x = 0; x < XDim; ++x)
		{
			for (uint32_t y = 0; y < YDim; ++y) { ((Pixel32*)pResultImage)[x + y * XDim] = pDecompressedData[x + y * XTrueDim]; }
		}

		// Free the temporary buffer.
		delete[] pDecompressedData;
	}
	return retval;
}

uint32_t PVRTDecompressPVRTC(const void* pCompressedData, uint32_t Do2bitMode, uint32_t XDim, uint32_t YDim, uint8_t* pResultImage)
{
	return PVRTDecompressPVRTC_int<false>(pCompressedData, Do2bitMode, XDim, YDim, pResultImage);
}

uint32_t PVRTDecompressPVRTCII(const void* pCompressedData, uint32_t Do2bitMode, uint32_t XDim, uint32_t YDim, uint8_t* pResultImage)
{
	return PVRTDecompressPVRTC_int<true>(pCompressedData, Do2bitMode, XDim, YDim, pResultImage);
}	
} // namespace pvr
//!\endcond