1#pragma once
2#ifndef __CVTT_CONVECTIONKERNELS_ETC_H__
3#define __CVTT_CONVECTIONKERNELS_ETC_H__
4
5#include "ConvectionKernels.h"
6#include "ConvectionKernels_ParallelMath.h"
7
8namespace cvtt
9{
10 struct Options;
11
12 namespace Internal
13 {
14 class ETCComputer
15 {
16 public:
17 static void CompressETC1Block(uint8_t *outputBuffer, const PixelBlockU8 *inputBlocks, ETC1CompressionData *compressionData, const Options &options);
18 static void CompressETC2Block(uint8_t *outputBuffer, const PixelBlockU8 *inputBlocks, ETC2CompressionData *compressionData, const Options &options, bool punchthroughAlpha);
19 static void CompressETC2AlphaBlock(uint8_t *outputBuffer, const PixelBlockU8 *inputBlocks, const Options &options);
20 static void CompressEACBlock(uint8_t *outputBuffer, const PixelBlockScalarS16 *inputBlocks, bool isSigned, const Options &options);
21
22 static ETC2CompressionData *AllocETC2Data(cvtt::Kernels::allocFunc_t allocFunc, void *context, const cvtt::Options &options);
23 static void ReleaseETC2Data(ETC2CompressionData *compressionData, cvtt::Kernels::freeFunc_t freeFunc);
24
25 static ETC1CompressionData *AllocETC1Data(cvtt::Kernels::allocFunc_t allocFunc, void *context);
26 static void ReleaseETC1Data(ETC1CompressionData *compressionData, cvtt::Kernels::freeFunc_t freeFunc);
27
28 private:
29 typedef ParallelMath::Float MFloat;
30 typedef ParallelMath::SInt16 MSInt16;
31 typedef ParallelMath::UInt15 MUInt15;
32 typedef ParallelMath::UInt16 MUInt16;
33 typedef ParallelMath::SInt32 MSInt32;
34 typedef ParallelMath::UInt31 MUInt31;
35
36 struct DifferentialResolveStorage
37 {
38 static const unsigned int MaxAttemptsPerSector = 57 + 81 + 81 + 81 + 81 + 81 + 81 + 81;
39
40 MUInt15 diffNumAttempts[2];
41 MFloat diffErrors[2][MaxAttemptsPerSector];
42 MUInt16 diffSelectors[2][MaxAttemptsPerSector];
43 MUInt15 diffColors[2][MaxAttemptsPerSector];
44 MUInt15 diffTables[2][MaxAttemptsPerSector];
45
46 uint16_t attemptSortIndexes[2][MaxAttemptsPerSector];
47 };
48
49 struct HModeEval
50 {
51 MFloat errors[62][16];
52 MUInt16 signBits[62];
53 MUInt15 uniqueQuantizedColors[62];
54 MUInt15 numUniqueColors[2];
55 };
56
57 struct ETC1CompressionDataInternal : public cvtt::ETC1CompressionData
58 {
59 explicit ETC1CompressionDataInternal(void *context)
60 : m_context(context)
61 {
62 }
63
64 DifferentialResolveStorage m_drs;
65 void *m_context;
66 };
67
68 struct ETC2CompressionDataInternal : public cvtt::ETC2CompressionData
69 {
70 explicit ETC2CompressionDataInternal(void *context, const cvtt::Options &options);
71
72 HModeEval m_h;
73 DifferentialResolveStorage m_drs;
74
75 void *m_context;
76 float m_chromaSideAxis0[3];
77 float m_chromaSideAxis1[3];
78 };
79
80 static MFloat ComputeErrorUniform(const MUInt15 pixelA[3], const MUInt15 pixelB[3]);
81 static MFloat ComputeErrorWeighted(const MUInt15 reconstructed[3], const MFloat pixelB[3], const Options options);
82 static MFloat ComputeErrorFakeBT709(const MUInt15 reconstructed[3], const MFloat pixelB[3]);
83
84 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);
85 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);
86 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);
87
88 static ParallelMath::Int16CompFlag ETCDifferentialIsLegalForChannel(const MUInt15 &a, const MUInt15 &b);
89 static ParallelMath::Int16CompFlag ETCDifferentialIsLegal(const MUInt15 &a, const MUInt15 &b);
90 static bool ETCDifferentialIsLegalForChannelScalar(const uint16_t &a, const uint16_t &b);
91 static bool ETCDifferentialIsLegalScalar(const uint16_t &a, const uint16_t &b);
92
93 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);
94 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);
95
96 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);
97
98 static MUInt15 DecodePlanarCoeff(const MUInt15 &coeff, int ch);
99 static void EncodePlanar(uint8_t *outputBuffer, MFloat &bestError, const MUInt15 pixels[16][3], const MFloat preWeightedPixels[16][3], const Options &options);
100
101 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);
102 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);
103 static void CompressETC2AlphaBlockInternal(uint8_t *outputBuffer, const MUInt15 pixels[16], bool is11Bit, bool isSigned, const Options &options);
104
105 static void ExtractBlocks(MUInt15 pixels[16][3], MFloat preWeightedPixels[16][3], const PixelBlockU8 *inputBlocks, const Options &options);
106
107 static void ResolveHalfBlockFakeBT709RoundingAccurate(MUInt15 quantized[3], const MUInt15 sectorCumulative[3], bool isDifferential);
108 static void ResolveHalfBlockFakeBT709RoundingFast(MUInt15 quantized[3], const MUInt15 sectorCumulative[3], bool isDifferential);
109 static void ResolveTHFakeBT709Rounding(MUInt15 quantized[3], const MUInt15 target[3], const MUInt15 &granularity);
110 static void ConvertToFakeBT709(MFloat yuv[3], const MUInt15 color[3]);
111 static void ConvertToFakeBT709(MFloat yuv[3], const MFloat color[3]);
112 static void ConvertToFakeBT709(MFloat yuv[3], const MFloat &r, const MFloat &g, const MFloat &b);
113 static void ConvertFromFakeBT709(MFloat rgb[3], const MFloat yuv[3]);
114
115 static void QuantizeETC2Alpha(int tableIndex, const MUInt15& value, const MUInt15& baseValue, const MUInt15& multiplier, bool is11Bit, bool isSigned, MUInt15& outIndexes, MUInt15& outQuantizedValues);
116
117 static void EmitTModeBlock(uint8_t *outputBuffer, const ParallelMath::ScalarUInt16 lineColor[3], const ParallelMath::ScalarUInt16 isolatedColor[3], int32_t packedSelectors, ParallelMath::ScalarUInt16 table, bool opaque);
118 static void EmitHModeBlock(uint8_t *outputBuffer, const ParallelMath::ScalarUInt16 blockColors[2], ParallelMath::ScalarUInt16 sectorBits, ParallelMath::ScalarUInt16 signBits, ParallelMath::ScalarUInt16 table, bool opaque);
119 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);
120
121 static const int g_flipTables[2][2][8];
122 };
123 }
124}
125
126#endif
127