1 | #pragma once |
2 | |
3 | #include "ConvectionKernels_ParallelMath.h" |
4 | |
5 | |
6 | namespace cvtt |
7 | { |
8 | namespace Tables |
9 | { |
10 | namespace BC7SC |
11 | { |
12 | struct Table; |
13 | } |
14 | } |
15 | |
16 | namespace Internal |
17 | { |
18 | namespace BC67 |
19 | { |
20 | struct WorkInfo; |
21 | } |
22 | |
23 | template<int TVectorSize> |
24 | class IndexSelectorHDR; |
25 | } |
26 | |
27 | struct PixelBlockU8; |
28 | } |
29 | |
30 | namespace cvtt |
31 | { |
32 | namespace Internal |
33 | { |
34 | class BC7Computer |
35 | { |
36 | public: |
37 | static void Pack(uint32_t flags, const PixelBlockU8* inputs, uint8_t* packedBlocks, const float channelWeights[4], const BC7EncodingPlan &encodingPlan, int numRefineRounds); |
38 | static void UnpackOne(PixelBlockU8 &output, const uint8_t* packedBlock); |
39 | |
40 | private: |
41 | static const int MaxTweakRounds = 4; |
42 | |
43 | typedef ParallelMath::SInt16 MSInt16; |
44 | typedef ParallelMath::UInt15 MUInt15; |
45 | typedef ParallelMath::UInt16 MUInt16; |
46 | typedef ParallelMath::SInt32 MSInt32; |
47 | typedef ParallelMath::Float MFloat; |
48 | |
49 | static void TweakAlpha(const MUInt15 original[2], int tweak, int range, MUInt15 result[2]); |
50 | static void Quantize(MUInt15* color, int bits, int channels); |
51 | static void QuantizeP(MUInt15* color, int bits, uint16_t p, int channels); |
52 | static void Unquantize(MUInt15* color, int bits, int channels); |
53 | static void CompressEndpoints0(MUInt15 ep[2][4], uint16_t p[2]); |
54 | static void CompressEndpoints1(MUInt15 ep[2][4], uint16_t p); |
55 | static void CompressEndpoints2(MUInt15 ep[2][4]); |
56 | static void CompressEndpoints3(MUInt15 ep[2][4], uint16_t p[2]); |
57 | static void CompressEndpoints4(MUInt15 epRGB[2][3], MUInt15 epA[2]); |
58 | static void CompressEndpoints5(MUInt15 epRGB[2][3], MUInt15 epA[2]); |
59 | static void CompressEndpoints6(MUInt15 ep[2][4], uint16_t p[2]); |
60 | static void CompressEndpoints7(MUInt15 ep[2][4], uint16_t p[2]); |
61 | static void TrySingleColorRGBAMultiTable(uint32_t flags, const MUInt15 pixels[16][4], const MFloat average[4], int numRealChannels, const uint8_t *fragmentStart, int shapeLength, const MFloat &staticAlphaError, const ParallelMath::Int16CompFlag punchThroughInvalid[4], MFloat& shapeBestError, MUInt15 shapeBestEP[2][4], MUInt15 *fragmentBestIndexes, const float *channelWeightsSq, const cvtt::Tables::BC7SC::Table*const* tables, int numTables, const ParallelMath::RoundTowardNearestForScope *rtn); |
62 | static void TrySinglePlane(uint32_t flags, const MUInt15 pixels[16][4], const MFloat floatPixels[16][4], const float channelWeights[4], const BC7EncodingPlan &encodingPlan, int numRefineRounds, BC67::WorkInfo& work, const ParallelMath::RoundTowardNearestForScope *rtn); |
63 | static void TryDualPlane(uint32_t flags, const MUInt15 pixels[16][4], const MFloat floatPixels[16][4], const float channelWeights[4], const BC7EncodingPlan &encodingPlan, int numRefineRounds, BC67::WorkInfo& work, const ParallelMath::RoundTowardNearestForScope *rtn); |
64 | |
65 | template<class T> |
66 | static void Swap(T& a, T& b); |
67 | }; |
68 | |
69 | |
70 | class BC6HComputer |
71 | { |
72 | public: |
73 | static void Pack(uint32_t flags, const PixelBlockF16* inputs, uint8_t* packedBlocks, const float channelWeights[4], bool isSigned, int numTweakRounds, int numRefineRounds); |
74 | static void UnpackOne(PixelBlockF16 &output, const uint8_t *pBC, bool isSigned); |
75 | |
76 | private: |
77 | typedef ParallelMath::Float MFloat; |
78 | typedef ParallelMath::SInt16 MSInt16; |
79 | typedef ParallelMath::UInt16 MUInt16; |
80 | typedef ParallelMath::UInt15 MUInt15; |
81 | typedef ParallelMath::AInt16 MAInt16; |
82 | typedef ParallelMath::SInt32 MSInt32; |
83 | typedef ParallelMath::UInt31 MUInt31; |
84 | |
85 | static const int MaxTweakRounds = 4; |
86 | static const int MaxRefineRounds = 3; |
87 | |
88 | static MSInt16 QuantizeSingleEndpointElementSigned(const MSInt16 &elem2CL, int precision, const ParallelMath::RoundUpForScope* ru); |
89 | static MUInt15 QuantizeSingleEndpointElementUnsigned(const MUInt15 &elem, int precision, const ParallelMath::RoundUpForScope* ru); |
90 | static void UnquantizeSingleEndpointElementSigned(const MSInt16 &comp, int precision, MSInt16 &outUnquantized, MSInt16 &outUnquantizedFinished2CL); |
91 | static void UnquantizeSingleEndpointElementUnsigned(const MUInt15 &comp, int precision, MUInt16 &outUnquantized, MUInt16 &outUnquantizedFinished); |
92 | static void QuantizeEndpointsSigned(const MSInt16 endPoints[2][3], const MFloat floatPixelsColorSpace[16][3], const MFloat floatPixelsLinearWeighted[16][3], MAInt16 quantizedEndPoints[2][3], MUInt15 indexes[16], IndexSelectorHDR<3> &indexSelector, int fixupIndex, int precision, int indexRange, const float *channelWeights, bool fastIndexing, const ParallelMath::RoundTowardNearestForScope *rtn); |
93 | static void QuantizeEndpointsUnsigned(const MSInt16 endPoints[2][3], const MFloat floatPixelsColorSpace[16][3], const MFloat floatPixelsLinearWeighted[16][3], MAInt16 quantizedEndPoints[2][3], MUInt15 indexes[16], IndexSelectorHDR<3> &indexSelector, int fixupIndex, int precision, int indexRange, const float *channelWeights, bool fastIndexing, const ParallelMath::RoundTowardNearestForScope *rtn); |
94 | static void EvaluatePartitionedLegality(const MAInt16 ep0[2][3], const MAInt16 ep1[2][3], int aPrec, const int bPrec[3], bool isTransformed, MAInt16 outEncodedEPs[2][2][3], ParallelMath::Int16CompFlag& outIsLegal); |
95 | static void EvaluateSingleLegality(const MAInt16 ep[2][3], int aPrec, const int bPrec[3], bool isTransformed, MAInt16 outEncodedEPs[2][3], ParallelMath::Int16CompFlag& outIsLegal); |
96 | static void SignExtendSingle(int &v, int bits); |
97 | }; |
98 | } |
99 | } |
100 | |