godot/thirdparty/cvtt/ConvectionKernels_ETC.h
K. S. Ernest (iFire) Lee 419b342a9a Faster CVTT by reducing quality.
Make BC6 and BC7 CVTT faster while still having better quality than DXT5.
2022-02-04 15:15:26 -08:00

127 lines
8.4 KiB
C++

#pragma once
#ifndef __CVTT_CONVECTIONKERNELS_ETC_H__
#define __CVTT_CONVECTIONKERNELS_ETC_H__
#include "ConvectionKernels.h"
#include "ConvectionKernels_ParallelMath.h"
namespace cvtt
{
struct Options;
namespace Internal
{
class ETCComputer
{
public:
static void CompressETC1Block(uint8_t *outputBuffer, const PixelBlockU8 *inputBlocks, ETC1CompressionData *compressionData, const Options &options);
static void CompressETC2Block(uint8_t *outputBuffer, const PixelBlockU8 *inputBlocks, ETC2CompressionData *compressionData, const Options &options, bool punchthroughAlpha);
static void CompressETC2AlphaBlock(uint8_t *outputBuffer, const PixelBlockU8 *inputBlocks, const Options &options);
static void CompressEACBlock(uint8_t *outputBuffer, const PixelBlockScalarS16 *inputBlocks, bool isSigned, const Options &options);
static ETC2CompressionData *AllocETC2Data(cvtt::Kernels::allocFunc_t allocFunc, void *context, const cvtt::Options &options);
static void ReleaseETC2Data(ETC2CompressionData *compressionData, cvtt::Kernels::freeFunc_t freeFunc);
static ETC1CompressionData *AllocETC1Data(cvtt::Kernels::allocFunc_t allocFunc, void *context);
static void ReleaseETC1Data(ETC1CompressionData *compressionData, cvtt::Kernels::freeFunc_t freeFunc);
private:
typedef ParallelMath::Float MFloat;
typedef ParallelMath::SInt16 MSInt16;
typedef ParallelMath::UInt15 MUInt15;
typedef ParallelMath::UInt16 MUInt16;
typedef ParallelMath::SInt32 MSInt32;
typedef ParallelMath::UInt31 MUInt31;
struct DifferentialResolveStorage
{
static const unsigned int MaxAttemptsPerSector = 57 + 81 + 81 + 81 + 81 + 81 + 81 + 81;
MUInt15 diffNumAttempts[2];
MFloat diffErrors[2][MaxAttemptsPerSector];
MUInt16 diffSelectors[2][MaxAttemptsPerSector];
MUInt15 diffColors[2][MaxAttemptsPerSector];
MUInt15 diffTables[2][MaxAttemptsPerSector];
uint16_t attemptSortIndexes[2][MaxAttemptsPerSector];
};
struct HModeEval
{
MFloat errors[62][16];
MUInt16 signBits[62];
MUInt15 uniqueQuantizedColors[62];
MUInt15 numUniqueColors[2];
};
struct ETC1CompressionDataInternal : public cvtt::ETC1CompressionData
{
explicit ETC1CompressionDataInternal(void *context)
: m_context(context)
{
}
DifferentialResolveStorage m_drs;
void *m_context;
};
struct ETC2CompressionDataInternal : public cvtt::ETC2CompressionData
{
explicit ETC2CompressionDataInternal(void *context, const cvtt::Options &options);
HModeEval m_h;
DifferentialResolveStorage m_drs;
void *m_context;
float m_chromaSideAxis0[3];
float m_chromaSideAxis1[3];
};
static MFloat ComputeErrorUniform(const MUInt15 pixelA[3], const MUInt15 pixelB[3]);
static MFloat ComputeErrorWeighted(const MUInt15 reconstructed[3], const MFloat pixelB[3], const Options options);
static MFloat ComputeErrorFakeBT709(const MUInt15 reconstructed[3], const MFloat pixelB[3]);
static void TestHalfBlock(MFloat &outError, MUInt16 &outSelectors, MUInt15 quantizedPackedColor, const MUInt15 pixels[8][3], const MFloat preWeightedPixels[8][3], const MSInt16 modifiers[4], bool isDifferential, const Options &options);
static void TestHalfBlockPunchthrough(MFloat &outError, MUInt16 &outSelectors, MUInt15 quantizedPackedColor, const MUInt15 pixels[8][3], const MFloat preWeightedPixels[8][3], const ParallelMath::Int16CompFlag isTransparent[8], const MUInt15 modifier, const Options &options);
static void FindBestDifferentialCombination(int flip, int d, const ParallelMath::Int16CompFlag canIgnoreSector[2], ParallelMath::Int16CompFlag& bestIsThisMode, MFloat& bestTotalError, MUInt15& bestFlip, MUInt15& bestD, MUInt15 bestColors[2], MUInt16 bestSelectors[2], MUInt15 bestTables[2], DifferentialResolveStorage &drs);
static ParallelMath::Int16CompFlag ETCDifferentialIsLegalForChannel(const MUInt15 &a, const MUInt15 &b);
static ParallelMath::Int16CompFlag ETCDifferentialIsLegal(const MUInt15 &a, const MUInt15 &b);
static bool ETCDifferentialIsLegalForChannelScalar(const uint16_t &a, const uint16_t &b);
static bool ETCDifferentialIsLegalScalar(const uint16_t &a, const uint16_t &b);
static void EncodeTMode(uint8_t *outputBuffer, MFloat &bestError, const ParallelMath::Int16CompFlag isIsolated[16], const MUInt15 pixels[16][3], const MFloat preWeightedPixels[16][3], const Options &options);
static void EncodeHMode(uint8_t *outputBuffer, MFloat &bestError, const ParallelMath::Int16CompFlag groupings[16], const MUInt15 pixels[16][3], HModeEval &he, const MFloat preWeightedPixels[16][3], const Options &options);
static void EncodeVirtualTModePunchthrough(uint8_t *outputBuffer, MFloat &bestError, const ParallelMath::Int16CompFlag isIsolated[16], const MUInt15 pixels[16][3], const MFloat preWeightedPixels[16][3], const ParallelMath::Int16CompFlag isTransparent[16], const ParallelMath::Int16CompFlag& anyTransparent, const ParallelMath::Int16CompFlag& allTransparent, const Options &options);
static MUInt15 DecodePlanarCoeff(const MUInt15 &coeff, int ch);
static void EncodePlanar(uint8_t *outputBuffer, MFloat &bestError, const MUInt15 pixels[16][3], const MFloat preWeightedPixels[16][3], const Options &options);
static void CompressETC1BlockInternal(MFloat &bestTotalError, uint8_t *outputBuffer, const MUInt15 pixels[16][3], const MFloat preWeightedPixels[16][3], DifferentialResolveStorage& compressionData, const Options &options, bool punchthrough);
static void CompressETC1PunchthroughBlockInternal(MFloat &bestTotalError, uint8_t *outputBuffer, const MUInt15 pixels[16][3], const MFloat preWeightedPixels[16][3], const ParallelMath::Int16CompFlag isTransparent[16], DifferentialResolveStorage& compressionData, const Options &options);
static void CompressETC2AlphaBlockInternal(uint8_t *outputBuffer, const MUInt15 pixels[16], bool is11Bit, bool isSigned, const Options &options);
static void ExtractBlocks(MUInt15 pixels[16][3], MFloat preWeightedPixels[16][3], const PixelBlockU8 *inputBlocks, const Options &options);
static void ResolveHalfBlockFakeBT709RoundingAccurate(MUInt15 quantized[3], const MUInt15 sectorCumulative[3], bool isDifferential);
static void ResolveHalfBlockFakeBT709RoundingFast(MUInt15 quantized[3], const MUInt15 sectorCumulative[3], bool isDifferential);
static void ResolveTHFakeBT709Rounding(MUInt15 quantized[3], const MUInt15 target[3], const MUInt15 &granularity);
static void ConvertToFakeBT709(MFloat yuv[3], const MUInt15 color[3]);
static void ConvertToFakeBT709(MFloat yuv[3], const MFloat color[3]);
static void ConvertToFakeBT709(MFloat yuv[3], const MFloat &r, const MFloat &g, const MFloat &b);
static void ConvertFromFakeBT709(MFloat rgb[3], const MFloat yuv[3]);
static void QuantizeETC2Alpha(int tableIndex, const MUInt15& value, const MUInt15& baseValue, const MUInt15& multiplier, bool is11Bit, bool isSigned, MUInt15& outIndexes, MUInt15& outQuantizedValues);
static void EmitTModeBlock(uint8_t *outputBuffer, const ParallelMath::ScalarUInt16 lineColor[3], const ParallelMath::ScalarUInt16 isolatedColor[3], int32_t packedSelectors, ParallelMath::ScalarUInt16 table, bool opaque);
static void EmitHModeBlock(uint8_t *outputBuffer, const ParallelMath::ScalarUInt16 blockColors[2], ParallelMath::ScalarUInt16 sectorBits, ParallelMath::ScalarUInt16 signBits, ParallelMath::ScalarUInt16 table, bool opaque);
static void EmitETC1Block(uint8_t *outputBuffer, int blockBestFlip, int blockBestD, const int blockBestColors[2][3], const int blockBestTables[2], const ParallelMath::ScalarUInt16 blockBestSelectors[2], bool transparent);
static const int g_flipTables[2][2][8];
};
}
}
#endif