From d3dc4e399ca64e8d6c07d8e73640d2d87f3724fc Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 22:20:12 -0300 Subject: [PATCH 01/13] astc: Use 'enum class' instead of 'enum' for EIntegerEncoding --- src/video_core/textures/astc.cpp | 50 ++++++++++++++++---------------- 1 file changed, 25 insertions(+), 25 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 33bd318651..af5a6c4cea 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -144,11 +144,11 @@ private: const IntType& m_Bits; }; -enum EIntegerEncoding { eIntegerEncoding_JustBits, eIntegerEncoding_Quint, eIntegerEncoding_Trit }; +enum class IntegerEncoding { JustBits, Quint, Trit }; class IntegerEncodedValue { private: - const EIntegerEncoding m_Encoding; + const IntegerEncoding m_Encoding; const uint32_t m_NumBits; uint32_t m_BitValue; union { @@ -164,10 +164,10 @@ public: return *this; } - IntegerEncodedValue(EIntegerEncoding encoding, uint32_t numBits) + IntegerEncodedValue(IntegerEncoding encoding, uint32_t numBits) : m_Encoding(encoding), m_NumBits(numBits) {} - EIntegerEncoding GetEncoding() const { + IntegerEncoding GetEncoding() const { return m_Encoding; } uint32_t BaseBitLength() const { @@ -202,9 +202,9 @@ public: // Returns the number of bits required to encode nVals values. uint32_t GetBitLength(uint32_t nVals) const { uint32_t totalBits = m_NumBits * nVals; - if (m_Encoding == eIntegerEncoding_Trit) { + if (m_Encoding == IntegerEncoding::Trit) { totalBits += (nVals * 8 + 4) / 5; - } else if (m_Encoding == eIntegerEncoding_Quint) { + } else if (m_Encoding == IntegerEncoding::Quint) { totalBits += (nVals * 7 + 2) / 3; } return totalBits; @@ -227,24 +227,24 @@ public: // Is maxVal a power of two? if (!(check & (check - 1))) { - return IntegerEncodedValue(eIntegerEncoding_JustBits, Popcnt(maxVal)); + return IntegerEncodedValue(IntegerEncoding::JustBits, Popcnt(maxVal)); } // Is maxVal of the type 3*2^n - 1? if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) { - return IntegerEncodedValue(eIntegerEncoding_Trit, Popcnt(check / 3 - 1)); + return IntegerEncodedValue(IntegerEncoding::Trit, Popcnt(check / 3 - 1)); } // Is maxVal of the type 5*2^n - 1? if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) { - return IntegerEncodedValue(eIntegerEncoding_Quint, Popcnt(check / 5 - 1)); + return IntegerEncodedValue(IntegerEncoding::Quint, Popcnt(check / 5 - 1)); } // Apparently it can't be represented with a bounded integer sequence... // just iterate. maxVal--; } - return IntegerEncodedValue(eIntegerEncoding_JustBits, 0); + return IntegerEncodedValue(IntegerEncoding::JustBits, 0); } // Fills result with the values that are encoded in the given @@ -259,17 +259,17 @@ public: uint32_t nValsDecoded = 0; while (nValsDecoded < nValues) { switch (val.GetEncoding()) { - case eIntegerEncoding_Quint: + case IntegerEncoding::Quint: DecodeQuintBlock(bits, result, val.BaseBitLength()); nValsDecoded += 3; break; - case eIntegerEncoding_Trit: + case IntegerEncoding::Trit: DecodeTritBlock(bits, result, val.BaseBitLength()); nValsDecoded += 5; break; - case eIntegerEncoding_JustBits: + case IntegerEncoding::JustBits: val.SetBitValue(bits.ReadBits(val.BaseBitLength())); result.push_back(val); nValsDecoded++; @@ -332,7 +332,7 @@ private: } for (uint32_t i = 0; i < 5; i++) { - IntegerEncodedValue val(eIntegerEncoding_Trit, nBitsPerValue); + IntegerEncodedValue val(IntegerEncoding::Trit, nBitsPerValue); val.SetBitValue(m[i]); val.SetTritValue(t[i]); result.push_back(val); @@ -380,7 +380,7 @@ private: } for (uint32_t i = 0; i < 3; i++) { - IntegerEncodedValue val(eIntegerEncoding_Quint, nBitsPerValue); + IntegerEncodedValue val(IntegerEncoding::Quint, nBitsPerValue); val.m_BitValue = m[i]; val.m_QuintValue = q[i]; result.push_back(val); @@ -859,12 +859,12 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode switch (val.GetEncoding()) { // Replicate bits - case eIntegerEncoding_JustBits: + case IntegerEncoding::JustBits: out[outIdx++] = Replicate(bitval, bitlen, 8); break; // Use algorithm in C.2.13 - case eIntegerEncoding_Trit: { + case IntegerEncoding::Trit: { D = val.GetTritValue(); @@ -912,10 +912,10 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode assert(!"Unsupported trit encoding for color values!"); break; } // switch(bitlen) - } // case eIntegerEncoding_Trit + } // case IntegerEncoding::Trit break; - case eIntegerEncoding_Quint: { + case IntegerEncoding::Quint: { D = val.GetQuintValue(); @@ -956,11 +956,11 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode assert(!"Unsupported quint encoding for color values!"); break; } // switch(bitlen) - } // case eIntegerEncoding_Quint + } // case IntegerEncoding::Quint break; } // switch(val.GetEncoding()) - if (val.GetEncoding() != eIntegerEncoding_JustBits) { + if (val.GetEncoding() != IntegerEncoding::JustBits) { uint32_t T = D * C + B; T ^= A; T = (A & 0x80) | (T >> 2); @@ -983,11 +983,11 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { uint32_t result = 0; switch (val.GetEncoding()) { - case eIntegerEncoding_JustBits: + case IntegerEncoding::JustBits: result = Replicate(bitval, bitlen, 6); break; - case eIntegerEncoding_Trit: { + case IntegerEncoding::Trit: { D = val.GetTritValue(); assert(D < 3); @@ -1019,7 +1019,7 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { } } break; - case eIntegerEncoding_Quint: { + case IntegerEncoding::Quint: { D = val.GetQuintValue(); assert(D < 5); @@ -1046,7 +1046,7 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { } break; } - if (val.GetEncoding() != eIntegerEncoding_JustBits && bitlen > 0) { + if (val.GetEncoding() != IntegerEncoding::JustBits && bitlen > 0) { // Decode the value... result = D * C + B; result ^= A; From 731a9a322e5b0623a5700cccc52307163c51c564 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 22:22:27 -0300 Subject: [PATCH 02/13] astc: Use common types instead of stdint.h integer types --- src/video_core/textures/astc.cpp | 560 +++++++++++++++---------------- 1 file changed, 279 insertions(+), 281 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index af5a6c4cea..dcfab4dad6 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -21,6 +21,8 @@ #include #include +#include "common/common_types.h" + #include "video_core/textures/astc.h" class InputBitStream { @@ -123,20 +125,20 @@ public: Bits(const Bits&) = delete; Bits& operator=(const Bits&) = delete; - uint8_t operator[](uint32_t bitPos) const { - return static_cast((m_Bits >> bitPos) & 1); + u8 operator[](u32 bitPos) const { + return static_cast((m_Bits >> bitPos) & 1); } - IntType operator()(uint32_t start, uint32_t end) const { + IntType operator()(u32 start, u32 end) const { if (start == end) { return (*this)[start]; } else if (start > end) { - uint32_t t = start; + u32 t = start; start = end; end = t; } - uint64_t mask = (1 << (end - start + 1)) - 1; + u64 mask = (1 << (end - start + 1)) - 1; return (m_Bits >> start) & static_cast(mask); } @@ -149,11 +151,11 @@ enum class IntegerEncoding { JustBits, Quint, Trit }; class IntegerEncodedValue { private: const IntegerEncoding m_Encoding; - const uint32_t m_NumBits; - uint32_t m_BitValue; + const u32 m_NumBits; + u32 m_BitValue; union { - uint32_t m_QuintValue; - uint32_t m_TritValue; + u32 m_QuintValue; + u32 m_TritValue; }; public: @@ -164,34 +166,34 @@ public: return *this; } - IntegerEncodedValue(IntegerEncoding encoding, uint32_t numBits) + IntegerEncodedValue(IntegerEncoding encoding, u32 numBits) : m_Encoding(encoding), m_NumBits(numBits) {} IntegerEncoding GetEncoding() const { return m_Encoding; } - uint32_t BaseBitLength() const { + u32 BaseBitLength() const { return m_NumBits; } - uint32_t GetBitValue() const { + u32 GetBitValue() const { return m_BitValue; } - void SetBitValue(uint32_t val) { + void SetBitValue(u32 val) { m_BitValue = val; } - uint32_t GetTritValue() const { + u32 GetTritValue() const { return m_TritValue; } - void SetTritValue(uint32_t val) { + void SetTritValue(u32 val) { m_TritValue = val; } - uint32_t GetQuintValue() const { + u32 GetQuintValue() const { return m_QuintValue; } - void SetQuintValue(uint32_t val) { + void SetQuintValue(u32 val) { m_QuintValue = val; } @@ -200,8 +202,8 @@ public: } // Returns the number of bits required to encode nVals values. - uint32_t GetBitLength(uint32_t nVals) const { - uint32_t totalBits = m_NumBits * nVals; + u32 GetBitLength(u32 nVals) const { + u32 totalBits = m_NumBits * nVals; if (m_Encoding == IntegerEncoding::Trit) { totalBits += (nVals * 8 + 4) / 5; } else if (m_Encoding == IntegerEncoding::Quint) { @@ -211,8 +213,8 @@ public: } // Count the number of bits set in a number. - static inline uint32_t Popcnt(uint32_t n) { - uint32_t c; + static inline u32 Popcnt(u32 n) { + u32 c; for (c = 0; n; c++) { n &= n - 1; } @@ -221,9 +223,9 @@ public: // Returns a new instance of this struct that corresponds to the // can take no more than maxval values - static IntegerEncodedValue CreateEncoding(uint32_t maxVal) { + static IntegerEncodedValue CreateEncoding(u32 maxVal) { while (maxVal > 0) { - uint32_t check = maxVal + 1; + u32 check = maxVal + 1; // Is maxVal a power of two? if (!(check & (check - 1))) { @@ -251,12 +253,12 @@ public: // bitstream. We must know beforehand what the maximum possible // value is, and how many values we're decoding. static void DecodeIntegerSequence(std::vector& result, - InputBitStream& bits, uint32_t maxRange, uint32_t nValues) { + InputBitStream& bits, u32 maxRange, u32 nValues) { // Determine encoding parameters IntegerEncodedValue val = IntegerEncodedValue::CreateEncoding(maxRange); // Start decoding - uint32_t nValsDecoded = 0; + u32 nValsDecoded = 0; while (nValsDecoded < nValues) { switch (val.GetEncoding()) { case IntegerEncoding::Quint: @@ -280,11 +282,11 @@ public: private: static void DecodeTritBlock(InputBitStream& bits, std::vector& result, - uint32_t nBitsPerValue) { + u32 nBitsPerValue) { // Implement the algorithm in section C.2.12 - uint32_t m[5]; - uint32_t t[5]; - uint32_t T; + u32 m[5]; + u32 t[5]; + u32 T; // Read the trit encoded block according to // table C.2.14 @@ -299,9 +301,9 @@ private: m[4] = bits.ReadBits(nBitsPerValue); T |= bits.ReadBit() << 7; - uint32_t C = 0; + u32 C = 0; - Bits Tb(T); + Bits Tb(T); if (Tb(2, 4) == 7) { C = (Tb(5, 7) << 2) | Tb(0, 1); t[4] = t[3] = 2; @@ -316,7 +318,7 @@ private: } } - Bits Cb(C); + Bits Cb(C); if (Cb(0, 1) == 3) { t[2] = 2; t[1] = Cb[4]; @@ -331,7 +333,7 @@ private: t[0] = (Cb[1] << 1) | (Cb[0] & ~Cb[1]); } - for (uint32_t i = 0; i < 5; i++) { + for (u32 i = 0; i < 5; i++) { IntegerEncodedValue val(IntegerEncoding::Trit, nBitsPerValue); val.SetBitValue(m[i]); val.SetTritValue(t[i]); @@ -340,11 +342,11 @@ private: } static void DecodeQuintBlock(InputBitStream& bits, std::vector& result, - uint32_t nBitsPerValue) { + u32 nBitsPerValue) { // Implement the algorithm in section C.2.12 - uint32_t m[3]; - uint32_t q[3]; - uint32_t Q; + u32 m[3]; + u32 q[3]; + u32 Q; // Read the trit encoded block according to // table C.2.15 @@ -355,12 +357,12 @@ private: m[2] = bits.ReadBits(nBitsPerValue); Q |= bits.ReadBits(2) << 5; - Bits Qb(Q); + Bits Qb(Q); if (Qb(1, 2) == 3 && Qb(5, 6) == 0) { q[0] = q[1] = 4; q[2] = (Qb[0] << 2) | ((Qb[4] & ~Qb[0]) << 1) | (Qb[3] & ~Qb[0]); } else { - uint32_t C = 0; + u32 C = 0; if (Qb(1, 2) == 3) { q[2] = 4; C = (Qb(3, 4) << 3) | ((~Qb(5, 6) & 3) << 1) | Qb[0]; @@ -369,7 +371,7 @@ private: C = Qb(0, 4); } - Bits Cb(C); + Bits Cb(C); if (Cb(0, 2) == 5) { q[1] = 4; q[0] = Cb(3, 4); @@ -379,7 +381,7 @@ private: } } - for (uint32_t i = 0; i < 3; i++) { + for (u32 i = 0; i < 3; i++) { IntegerEncodedValue val(IntegerEncoding::Quint, nBitsPerValue); val.m_BitValue = m[i]; val.m_QuintValue = q[i]; @@ -391,17 +393,17 @@ private: namespace ASTCC { struct TexelWeightParams { - uint32_t m_Width = 0; - uint32_t m_Height = 0; + u32 m_Width = 0; + u32 m_Height = 0; bool m_bDualPlane = false; - uint32_t m_MaxWeight = 0; + u32 m_MaxWeight = 0; bool m_bError = false; bool m_bVoidExtentLDR = false; bool m_bVoidExtentHDR = false; - uint32_t GetPackedBitSize() const { + u32 GetPackedBitSize() const { // How many indices do we have? - uint32_t nIdxs = m_Height * m_Width; + u32 nIdxs = m_Height * m_Width; if (m_bDualPlane) { nIdxs *= 2; } @@ -409,8 +411,8 @@ struct TexelWeightParams { return IntegerEncodedValue::CreateEncoding(m_MaxWeight).GetBitLength(nIdxs); } - uint32_t GetNumWeightValues() const { - uint32_t ret = m_Width * m_Height; + u32 GetNumWeightValues() const { + u32 ret = m_Width * m_Height; if (m_bDualPlane) { ret *= 2; } @@ -422,7 +424,7 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { TexelWeightParams params; // Read the entire block mode all at once - uint16_t modeBits = static_cast(strm.ReadBits(11)); + u16 modeBits = static_cast(strm.ReadBits(11)); // Does this match the void extent block mode? if ((modeBits & 0x01FF) == 0x1FC) { @@ -457,7 +459,7 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { // of the block mode. Layout is determined by a number // between 0 and 9 corresponding to table C.2.8 of the // ASTC spec. - uint32_t layout = 0; + u32 layout = 0; if ((modeBits & 0x1) || (modeBits & 0x2)) { // layout is in [0-4] @@ -509,7 +511,7 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { assert(layout < 10); // Determine R - uint32_t R = !!(modeBits & 0x10); + u32 R = !!(modeBits & 0x10); if (layout < 5) { R |= (modeBits & 0x3) << 1; } else { @@ -520,54 +522,54 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { // Determine width & height switch (layout) { case 0: { - uint32_t A = (modeBits >> 5) & 0x3; - uint32_t B = (modeBits >> 7) & 0x3; + u32 A = (modeBits >> 5) & 0x3; + u32 B = (modeBits >> 7) & 0x3; params.m_Width = B + 4; params.m_Height = A + 2; break; } case 1: { - uint32_t A = (modeBits >> 5) & 0x3; - uint32_t B = (modeBits >> 7) & 0x3; + u32 A = (modeBits >> 5) & 0x3; + u32 B = (modeBits >> 7) & 0x3; params.m_Width = B + 8; params.m_Height = A + 2; break; } case 2: { - uint32_t A = (modeBits >> 5) & 0x3; - uint32_t B = (modeBits >> 7) & 0x3; + u32 A = (modeBits >> 5) & 0x3; + u32 B = (modeBits >> 7) & 0x3; params.m_Width = A + 2; params.m_Height = B + 8; break; } case 3: { - uint32_t A = (modeBits >> 5) & 0x3; - uint32_t B = (modeBits >> 7) & 0x1; + u32 A = (modeBits >> 5) & 0x3; + u32 B = (modeBits >> 7) & 0x1; params.m_Width = A + 2; params.m_Height = B + 6; break; } case 4: { - uint32_t A = (modeBits >> 5) & 0x3; - uint32_t B = (modeBits >> 7) & 0x1; + u32 A = (modeBits >> 5) & 0x3; + u32 B = (modeBits >> 7) & 0x1; params.m_Width = B + 2; params.m_Height = A + 2; break; } case 5: { - uint32_t A = (modeBits >> 5) & 0x3; + u32 A = (modeBits >> 5) & 0x3; params.m_Width = 12; params.m_Height = A + 2; break; } case 6: { - uint32_t A = (modeBits >> 5) & 0x3; + u32 A = (modeBits >> 5) & 0x3; params.m_Width = A + 2; params.m_Height = 12; break; @@ -586,8 +588,8 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { } case 9: { - uint32_t A = (modeBits >> 5) & 0x3; - uint32_t B = (modeBits >> 9) & 0x3; + u32 A = (modeBits >> 5) & 0x3; + u32 B = (modeBits >> 9) & 0x3; params.m_Width = A + 6; params.m_Height = B + 6; break; @@ -605,10 +607,10 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { bool H = (layout != 9) && (modeBits & 0x200); if (H) { - const uint32_t maxWeights[6] = {9, 11, 15, 19, 23, 31}; + const u32 maxWeights[6] = {9, 11, 15, 19, 23, 31}; params.m_MaxWeight = maxWeights[R - 2]; } else { - const uint32_t maxWeights[6] = {1, 2, 3, 4, 5, 7}; + const u32 maxWeights[6] = {1, 2, 3, 4, 5, 7}; params.m_MaxWeight = maxWeights[R - 2]; } @@ -617,32 +619,32 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { return params; } -static void FillVoidExtentLDR(InputBitStream& strm, uint32_t* const outBuf, uint32_t blockWidth, - uint32_t blockHeight) { +static void FillVoidExtentLDR(InputBitStream& strm, u32* const outBuf, u32 blockWidth, + u32 blockHeight) { // Don't actually care about the void extent, just read the bits... for (int i = 0; i < 4; ++i) { strm.ReadBits(13); } // Decode the RGBA components and renormalize them to the range [0, 255] - uint16_t r = static_cast(strm.ReadBits(16)); - uint16_t g = static_cast(strm.ReadBits(16)); - uint16_t b = static_cast(strm.ReadBits(16)); - uint16_t a = static_cast(strm.ReadBits(16)); + u16 r = static_cast(strm.ReadBits(16)); + u16 g = static_cast(strm.ReadBits(16)); + u16 b = static_cast(strm.ReadBits(16)); + u16 a = static_cast(strm.ReadBits(16)); - uint32_t rgba = (r >> 8) | (g & 0xFF00) | (static_cast(b) & 0xFF00) << 8 | - (static_cast(a) & 0xFF00) << 16; + u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast(b) & 0xFF00) << 8 | + (static_cast(a) & 0xFF00) << 16; - for (uint32_t j = 0; j < blockHeight; j++) { - for (uint32_t i = 0; i < blockWidth; i++) { + for (u32 j = 0; j < blockHeight; j++) { + for (u32 i = 0; i < blockWidth; i++) { outBuf[j * blockWidth + i] = rgba; } } } -static void FillError(uint32_t* outBuf, uint32_t blockWidth, uint32_t blockHeight) { - for (uint32_t j = 0; j < blockHeight; j++) { - for (uint32_t i = 0; i < blockWidth; i++) { +static void FillError(u32* outBuf, u32 blockWidth, u32 blockHeight) { + for (u32 j = 0; j < blockHeight; j++) { + for (u32 i = 0; i < blockWidth; i++) { outBuf[j * blockWidth + i] = 0xFFFF00FF; } } @@ -651,18 +653,18 @@ static void FillError(uint32_t* outBuf, uint32_t blockWidth, uint32_t blockHeigh // Replicates low numBits such that [(toBit - 1):(toBit - 1 - fromBit)] // is the same as [(numBits - 1):0] and repeats all the way down. template -static IntType Replicate(const IntType& val, uint32_t numBits, uint32_t toBit) { +static IntType Replicate(const IntType& val, u32 numBits, u32 toBit) { if (numBits == 0) return 0; if (toBit == 0) return 0; IntType v = val & static_cast((1 << numBits) - 1); IntType res = v; - uint32_t reslen = numBits; + u32 reslen = numBits; while (reslen < toBit) { - uint32_t comp = 0; + u32 comp = 0; if (numBits > toBit - reslen) { - uint32_t newshift = toBit - reslen; + u32 newshift = toBit - reslen; comp = numBits - newshift; numBits = newshift; } @@ -675,14 +677,14 @@ static IntType Replicate(const IntType& val, uint32_t numBits, uint32_t toBit) { class Pixel { protected: - using ChannelType = int16_t; - uint8_t m_BitDepth[4] = {8, 8, 8, 8}; - int16_t color[4] = {}; + using ChannelType = s16; + u8 m_BitDepth[4] = {8, 8, 8, 8}; + s16 color[4] = {}; public: Pixel() = default; - Pixel(uint32_t a, uint32_t r, uint32_t g, uint32_t b, unsigned bitDepth = 8) - : m_BitDepth{uint8_t(bitDepth), uint8_t(bitDepth), uint8_t(bitDepth), uint8_t(bitDepth)}, + Pixel(u32 a, u32 r, u32 g, u32 b, unsigned bitDepth = 8) + : m_BitDepth{u8(bitDepth), u8(bitDepth), u8(bitDepth), u8(bitDepth)}, color{static_cast(a), static_cast(r), static_cast(g), static_cast(b)} {} @@ -691,22 +693,22 @@ public: // significant bits when going from larger to smaller bit depth // or by repeating the most significant bits when going from // smaller to larger bit depths. - void ChangeBitDepth(const uint8_t (&depth)[4]) { - for (uint32_t i = 0; i < 4; i++) { + void ChangeBitDepth(const u8 (&depth)[4]) { + for (u32 i = 0; i < 4; i++) { Component(i) = ChangeBitDepth(Component(i), m_BitDepth[i], depth[i]); m_BitDepth[i] = depth[i]; } } template - static float ConvertChannelToFloat(IntType channel, uint8_t bitDepth) { + static float ConvertChannelToFloat(IntType channel, u8 bitDepth) { float denominator = static_cast((1 << bitDepth) - 1); return static_cast(channel) / denominator; } // Changes the bit depth of a single component. See the comment // above for how we do this. - static ChannelType ChangeBitDepth(Pixel::ChannelType val, uint8_t oldDepth, uint8_t newDepth) { + static ChannelType ChangeBitDepth(Pixel::ChannelType val, u8 oldDepth, u8 newDepth) { assert(newDepth <= 8); assert(oldDepth <= 8); @@ -722,12 +724,11 @@ public: if (newDepth == 0) { return 0xFF; } else { - uint8_t bitsWasted = static_cast(oldDepth - newDepth); - uint16_t v = static_cast(val); - v = static_cast((v + (1 << (bitsWasted - 1))) >> bitsWasted); - v = ::std::min(::std::max(0, v), - static_cast((1 << newDepth) - 1)); - return static_cast(v); + u8 bitsWasted = static_cast(oldDepth - newDepth); + u16 v = static_cast(val); + v = static_cast((v + (1 << (bitsWasted - 1))) >> bitsWasted); + v = ::std::min(::std::max(0, v), static_cast((1 << newDepth) - 1)); + return static_cast(v); } } @@ -759,14 +760,14 @@ public: ChannelType& B() { return color[3]; } - const ChannelType& Component(uint32_t idx) const { + const ChannelType& Component(u32 idx) const { return color[idx]; } - ChannelType& Component(uint32_t idx) { + ChannelType& Component(u32 idx) { return color[idx]; } - void GetBitDepth(uint8_t (&outDepth)[4]) const { + void GetBitDepth(u8 (&outDepth)[4]) const { for (int i = 0; i < 4; i++) { outDepth[i] = m_BitDepth[i]; } @@ -776,12 +777,12 @@ public: // and then pack each channel into an R8G8B8A8 32-bit integer. We assume // that the architecture is little-endian, so the alpha channel will end // up in the most-significant byte. - uint32_t Pack() const { + u32 Pack() const { Pixel eightBit(*this); - const uint8_t eightBitDepth[4] = {8, 8, 8, 8}; + const u8 eightBitDepth[4] = {8, 8, 8, 8}; eightBit.ChangeBitDepth(eightBitDepth); - uint32_t r = 0; + u32 r = 0; r |= eightBit.A(); r <<= 8; r |= eightBit.B(); @@ -794,7 +795,7 @@ public: // Clamps the pixel to the range [0,255] void ClampByte() { - for (uint32_t i = 0; i < 4; i++) { + for (u32 i = 0; i < 4; i++) { color[i] = (color[i] < 0) ? 0 : ((color[i] > 255) ? 255 : color[i]); } } @@ -804,20 +805,20 @@ public: } }; -static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* modes, - const uint32_t nPartitions, const uint32_t nBitsForColorData) { +static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nPartitions, + const u32 nBitsForColorData) { // First figure out how many color values we have - uint32_t nValues = 0; - for (uint32_t i = 0; i < nPartitions; i++) { + u32 nValues = 0; + for (u32 i = 0; i < nPartitions; i++) { nValues += ((modes[i] >> 2) + 1) << 1; } // Then based on the number of values and the remaining number of bits, // figure out the max value for each of them... - uint32_t range = 256; + u32 range = 256; while (--range > 0) { IntegerEncodedValue val = IntegerEncodedValue::CreateEncoding(range); - uint32_t bitLength = val.GetBitLength(nValues); + u32 bitLength = val.GetBitLength(nValues); if (bitLength <= nBitsForColorData) { // Find the smallest possible range that matches the given encoding while (--range > 0) { @@ -840,7 +841,7 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode // Once we have the decoded values, we need to dequantize them to the 0-255 range // This procedure is outlined in ASTC spec C.2.13 - uint32_t outIdx = 0; + u32 outIdx = 0; for (auto itr = decodedColorValues.begin(); itr != decodedColorValues.end(); ++itr) { // Have we already decoded all that we need? if (outIdx >= nValues) { @@ -848,12 +849,12 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode } const IntegerEncodedValue& val = *itr; - uint32_t bitlen = val.BaseBitLength(); - uint32_t bitval = val.GetBitValue(); + u32 bitlen = val.BaseBitLength(); + u32 bitval = val.GetBitValue(); assert(bitlen >= 1); - uint32_t A = 0, B = 0, C = 0, D = 0; + u32 A = 0, B = 0, C = 0, D = 0; // A is just the lsb replicated 9 times. A = Replicate(bitval & 1, 1, 9); @@ -876,35 +877,35 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode case 2: { C = 93; // B = b000b0bb0 - uint32_t b = (bitval >> 1) & 1; + u32 b = (bitval >> 1) & 1; B = (b << 8) | (b << 4) | (b << 2) | (b << 1); } break; case 3: { C = 44; // B = cb000cbcb - uint32_t cb = (bitval >> 1) & 3; + u32 cb = (bitval >> 1) & 3; B = (cb << 7) | (cb << 2) | cb; } break; case 4: { C = 22; // B = dcb000dcb - uint32_t dcb = (bitval >> 1) & 7; + u32 dcb = (bitval >> 1) & 7; B = (dcb << 6) | dcb; } break; case 5: { C = 11; // B = edcb000ed - uint32_t edcb = (bitval >> 1) & 0xF; + u32 edcb = (bitval >> 1) & 0xF; B = (edcb << 5) | (edcb >> 2); } break; case 6: { C = 5; // B = fedcb000f - uint32_t fedcb = (bitval >> 1) & 0x1F; + u32 fedcb = (bitval >> 1) & 0x1F; B = (fedcb << 4) | (fedcb >> 4); } break; @@ -927,28 +928,28 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode case 2: { C = 54; // B = b0000bb00 - uint32_t b = (bitval >> 1) & 1; + u32 b = (bitval >> 1) & 1; B = (b << 8) | (b << 3) | (b << 2); } break; case 3: { C = 26; // B = cb0000cbc - uint32_t cb = (bitval >> 1) & 3; + u32 cb = (bitval >> 1) & 3; B = (cb << 7) | (cb << 1) | (cb >> 1); } break; case 4: { C = 13; // B = dcb0000dc - uint32_t dcb = (bitval >> 1) & 7; + u32 dcb = (bitval >> 1) & 7; B = (dcb << 6) | (dcb >> 1); } break; case 5: { C = 6; // B = edcb0000e - uint32_t edcb = (bitval >> 1) & 0xF; + u32 edcb = (bitval >> 1) & 0xF; B = (edcb << 5) | (edcb >> 3); } break; @@ -961,7 +962,7 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode } // switch(val.GetEncoding()) if (val.GetEncoding() != IntegerEncoding::JustBits) { - uint32_t T = D * C + B; + u32 T = D * C + B; T ^= A; T = (A & 0x80) | (T >> 2); out[outIdx++] = T; @@ -969,19 +970,19 @@ static void DecodeColorValues(uint32_t* out, uint8_t* data, const uint32_t* mode } // Make sure that each of our values is in the proper range... - for (uint32_t i = 0; i < nValues; i++) { + for (u32 i = 0; i < nValues; i++) { assert(out[i] <= 255); } } -static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { - uint32_t bitval = val.GetBitValue(); - uint32_t bitlen = val.BaseBitLength(); +static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) { + u32 bitval = val.GetBitValue(); + u32 bitlen = val.BaseBitLength(); - uint32_t A = Replicate(bitval & 1, 1, 7); - uint32_t B = 0, C = 0, D = 0; + u32 A = Replicate(bitval & 1, 1, 7); + u32 B = 0, C = 0, D = 0; - uint32_t result = 0; + u32 result = 0; switch (val.GetEncoding()) { case IntegerEncoding::JustBits: result = Replicate(bitval, bitlen, 6); @@ -993,7 +994,7 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { switch (bitlen) { case 0: { - uint32_t results[3] = {0, 32, 63}; + u32 results[3] = {0, 32, 63}; result = results[D]; } break; @@ -1003,13 +1004,13 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { case 2: { C = 23; - uint32_t b = (bitval >> 1) & 1; + u32 b = (bitval >> 1) & 1; B = (b << 6) | (b << 2) | b; } break; case 3: { C = 11; - uint32_t cb = (bitval >> 1) & 3; + u32 cb = (bitval >> 1) & 3; B = (cb << 5) | cb; } break; @@ -1025,7 +1026,7 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { switch (bitlen) { case 0: { - uint32_t results[5] = {0, 16, 32, 47, 63}; + u32 results[5] = {0, 16, 32, 47, 63}; result = results[D]; } break; @@ -1035,7 +1036,7 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { case 2: { C = 13; - uint32_t b = (bitval >> 1) & 1; + u32 b = (bitval >> 1) & 1; B = (b << 6) | (b << 1); } break; @@ -1063,12 +1064,11 @@ static uint32_t UnquantizeTexelWeight(const IntegerEncodedValue& val) { return result; } -static void UnquantizeTexelWeights(uint32_t out[2][144], - const std::vector& weights, - const TexelWeightParams& params, const uint32_t blockWidth, - const uint32_t blockHeight) { - uint32_t weightIdx = 0; - uint32_t unquantized[2][144]; +static void UnquantizeTexelWeights(u32 out[2][144], const std::vector& weights, + const TexelWeightParams& params, const u32 blockWidth, + const u32 blockHeight) { + u32 weightIdx = 0; + u32 unquantized[2][144]; for (auto itr = weights.begin(); itr != weights.end(); ++itr) { unquantized[0][weightIdx] = UnquantizeTexelWeight(*itr); @@ -1086,34 +1086,34 @@ static void UnquantizeTexelWeights(uint32_t out[2][144], } // Do infill if necessary (Section C.2.18) ... - uint32_t Ds = (1024 + (blockWidth / 2)) / (blockWidth - 1); - uint32_t Dt = (1024 + (blockHeight / 2)) / (blockHeight - 1); + u32 Ds = (1024 + (blockWidth / 2)) / (blockWidth - 1); + u32 Dt = (1024 + (blockHeight / 2)) / (blockHeight - 1); - const uint32_t kPlaneScale = params.m_bDualPlane ? 2U : 1U; - for (uint32_t plane = 0; plane < kPlaneScale; plane++) - for (uint32_t t = 0; t < blockHeight; t++) - for (uint32_t s = 0; s < blockWidth; s++) { - uint32_t cs = Ds * s; - uint32_t ct = Dt * t; + const u32 kPlaneScale = params.m_bDualPlane ? 2U : 1U; + for (u32 plane = 0; plane < kPlaneScale; plane++) + for (u32 t = 0; t < blockHeight; t++) + for (u32 s = 0; s < blockWidth; s++) { + u32 cs = Ds * s; + u32 ct = Dt * t; - uint32_t gs = (cs * (params.m_Width - 1) + 32) >> 6; - uint32_t gt = (ct * (params.m_Height - 1) + 32) >> 6; + u32 gs = (cs * (params.m_Width - 1) + 32) >> 6; + u32 gt = (ct * (params.m_Height - 1) + 32) >> 6; - uint32_t js = gs >> 4; - uint32_t fs = gs & 0xF; + u32 js = gs >> 4; + u32 fs = gs & 0xF; - uint32_t jt = gt >> 4; - uint32_t ft = gt & 0x0F; + u32 jt = gt >> 4; + u32 ft = gt & 0x0F; - uint32_t w11 = (fs * ft + 8) >> 4; - uint32_t w10 = ft - w11; - uint32_t w01 = fs - w11; - uint32_t w00 = 16 - fs - ft + w11; + u32 w11 = (fs * ft + 8) >> 4; + u32 w10 = ft - w11; + u32 w01 = fs - w11; + u32 w00 = 16 - fs - ft + w11; - uint32_t v0 = js + jt * params.m_Width; + u32 v0 = js + jt * params.m_Width; #define FIND_TEXEL(tidx, bidx) \ - uint32_t p##bidx = 0; \ + u32 p##bidx = 0; \ do { \ if ((tidx) < (params.m_Width * params.m_Height)) { \ p##bidx = unquantized[plane][(tidx)]; \ @@ -1133,7 +1133,7 @@ static void UnquantizeTexelWeights(uint32_t out[2][144], } // Transfers a bit as described in C.2.14 -static inline void BitTransferSigned(int32_t& a, int32_t& b) { +static inline void BitTransferSigned(s32& a, s32& b) { b >>= 1; b |= a & 0x80; a >>= 1; @@ -1144,14 +1144,14 @@ static inline void BitTransferSigned(int32_t& a, int32_t& b) { // Adds more precision to the blue channel as described // in C.2.14 -static inline Pixel BlueContract(int32_t a, int32_t r, int32_t g, int32_t b) { - return Pixel(static_cast(a), static_cast((r + b) >> 1), - static_cast((g + b) >> 1), static_cast(b)); +static inline Pixel BlueContract(s32 a, s32 r, s32 g, s32 b) { + return Pixel(static_cast(a), static_cast((r + b) >> 1), + static_cast((g + b) >> 1), static_cast(b)); } // Partition selection functions as specified in // C.2.21 -static inline uint32_t hash52(uint32_t p) { +static inline u32 hash52(u32 p) { p ^= p >> 15; p -= p << 17; p += p << 7; @@ -1165,8 +1165,7 @@ static inline uint32_t hash52(uint32_t p) { return p; } -static uint32_t SelectPartition(int32_t seed, int32_t x, int32_t y, int32_t z, - int32_t partitionCount, int32_t smallBlock) { +static u32 SelectPartition(s32 seed, s32 x, s32 y, s32 z, s32 partitionCount, s32 smallBlock) { if (1 == partitionCount) return 0; @@ -1178,34 +1177,34 @@ static uint32_t SelectPartition(int32_t seed, int32_t x, int32_t y, int32_t z, seed += (partitionCount - 1) * 1024; - uint32_t rnum = hash52(static_cast(seed)); - uint8_t seed1 = static_cast(rnum & 0xF); - uint8_t seed2 = static_cast((rnum >> 4) & 0xF); - uint8_t seed3 = static_cast((rnum >> 8) & 0xF); - uint8_t seed4 = static_cast((rnum >> 12) & 0xF); - uint8_t seed5 = static_cast((rnum >> 16) & 0xF); - uint8_t seed6 = static_cast((rnum >> 20) & 0xF); - uint8_t seed7 = static_cast((rnum >> 24) & 0xF); - uint8_t seed8 = static_cast((rnum >> 28) & 0xF); - uint8_t seed9 = static_cast((rnum >> 18) & 0xF); - uint8_t seed10 = static_cast((rnum >> 22) & 0xF); - uint8_t seed11 = static_cast((rnum >> 26) & 0xF); - uint8_t seed12 = static_cast(((rnum >> 30) | (rnum << 2)) & 0xF); + u32 rnum = hash52(static_cast(seed)); + u8 seed1 = static_cast(rnum & 0xF); + u8 seed2 = static_cast((rnum >> 4) & 0xF); + u8 seed3 = static_cast((rnum >> 8) & 0xF); + u8 seed4 = static_cast((rnum >> 12) & 0xF); + u8 seed5 = static_cast((rnum >> 16) & 0xF); + u8 seed6 = static_cast((rnum >> 20) & 0xF); + u8 seed7 = static_cast((rnum >> 24) & 0xF); + u8 seed8 = static_cast((rnum >> 28) & 0xF); + u8 seed9 = static_cast((rnum >> 18) & 0xF); + u8 seed10 = static_cast((rnum >> 22) & 0xF); + u8 seed11 = static_cast((rnum >> 26) & 0xF); + u8 seed12 = static_cast(((rnum >> 30) | (rnum << 2)) & 0xF); - seed1 = static_cast(seed1 * seed1); - seed2 = static_cast(seed2 * seed2); - seed3 = static_cast(seed3 * seed3); - seed4 = static_cast(seed4 * seed4); - seed5 = static_cast(seed5 * seed5); - seed6 = static_cast(seed6 * seed6); - seed7 = static_cast(seed7 * seed7); - seed8 = static_cast(seed8 * seed8); - seed9 = static_cast(seed9 * seed9); - seed10 = static_cast(seed10 * seed10); - seed11 = static_cast(seed11 * seed11); - seed12 = static_cast(seed12 * seed12); + seed1 = static_cast(seed1 * seed1); + seed2 = static_cast(seed2 * seed2); + seed3 = static_cast(seed3 * seed3); + seed4 = static_cast(seed4 * seed4); + seed5 = static_cast(seed5 * seed5); + seed6 = static_cast(seed6 * seed6); + seed7 = static_cast(seed7 * seed7); + seed8 = static_cast(seed8 * seed8); + seed9 = static_cast(seed9 * seed9); + seed10 = static_cast(seed10 * seed10); + seed11 = static_cast(seed11 * seed11); + seed12 = static_cast(seed12 * seed12); - int32_t sh1, sh2, sh3; + s32 sh1, sh2, sh3; if (seed & 1) { sh1 = (seed & 2) ? 4 : 5; sh2 = (partitionCount == 3) ? 6 : 5; @@ -1215,23 +1214,23 @@ static uint32_t SelectPartition(int32_t seed, int32_t x, int32_t y, int32_t z, } sh3 = (seed & 0x10) ? sh1 : sh2; - seed1 = static_cast(seed1 >> sh1); - seed2 = static_cast(seed2 >> sh2); - seed3 = static_cast(seed3 >> sh1); - seed4 = static_cast(seed4 >> sh2); - seed5 = static_cast(seed5 >> sh1); - seed6 = static_cast(seed6 >> sh2); - seed7 = static_cast(seed7 >> sh1); - seed8 = static_cast(seed8 >> sh2); - seed9 = static_cast(seed9 >> sh3); - seed10 = static_cast(seed10 >> sh3); - seed11 = static_cast(seed11 >> sh3); - seed12 = static_cast(seed12 >> sh3); + seed1 = static_cast(seed1 >> sh1); + seed2 = static_cast(seed2 >> sh2); + seed3 = static_cast(seed3 >> sh1); + seed4 = static_cast(seed4 >> sh2); + seed5 = static_cast(seed5 >> sh1); + seed6 = static_cast(seed6 >> sh2); + seed7 = static_cast(seed7 >> sh1); + seed8 = static_cast(seed8 >> sh2); + seed9 = static_cast(seed9 >> sh3); + seed10 = static_cast(seed10 >> sh3); + seed11 = static_cast(seed11 >> sh3); + seed12 = static_cast(seed12 >> sh3); - int32_t a = seed1 * x + seed2 * y + seed11 * z + (rnum >> 14); - int32_t b = seed3 * x + seed4 * y + seed12 * z + (rnum >> 10); - int32_t c = seed5 * x + seed6 * y + seed9 * z + (rnum >> 6); - int32_t d = seed7 * x + seed8 * y + seed10 * z + (rnum >> 2); + s32 a = seed1 * x + seed2 * y + seed11 * z + (rnum >> 14); + s32 b = seed3 * x + seed4 * y + seed12 * z + (rnum >> 10); + s32 c = seed5 * x + seed6 * y + seed9 * z + (rnum >> 6); + s32 d = seed7 * x + seed8 * y + seed10 * z + (rnum >> 2); a &= 0x3F; b &= 0x3F; @@ -1252,24 +1251,23 @@ static uint32_t SelectPartition(int32_t seed, int32_t x, int32_t y, int32_t z, return 3; } -static inline uint32_t Select2DPartition(int32_t seed, int32_t x, int32_t y, int32_t partitionCount, - int32_t smallBlock) { +static inline u32 Select2DPartition(s32 seed, s32 x, s32 y, s32 partitionCount, s32 smallBlock) { return SelectPartition(seed, x, y, 0, partitionCount, smallBlock); } // Section C.2.14 -static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const uint32_t*& colorValues, - uint32_t colorEndpointMode) { +static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues, + u32 colorEndpointMode) { #define READ_UINT_VALUES(N) \ - uint32_t v[N]; \ - for (uint32_t i = 0; i < N; i++) { \ + u32 v[N]; \ + for (u32 i = 0; i < N; i++) { \ v[i] = *(colorValues++); \ } #define READ_INT_VALUES(N) \ - int32_t v[N]; \ - for (uint32_t i = 0; i < N; i++) { \ - v[i] = static_cast(*(colorValues++)); \ + s32 v[N]; \ + for (u32 i = 0; i < N; i++) { \ + v[i] = static_cast(*(colorValues++)); \ } switch (colorEndpointMode) { @@ -1281,8 +1279,8 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const uint32_t*& colorValue case 1: { READ_UINT_VALUES(2) - uint32_t L0 = (v[0] >> 2) | (v[1] & 0xC0); - uint32_t L1 = std::max(L0 + (v[1] & 0x3F), 0xFFU); + u32 L0 = (v[0] >> 2) | (v[1] & 0xC0); + u32 L1 = std::max(L0 + (v[1] & 0x3F), 0xFFU); ep1 = Pixel(0xFF, L0, L0, L0); ep2 = Pixel(0xFF, L1, L1, L1); } break; @@ -1379,8 +1377,8 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const uint32_t*& colorValue #undef READ_INT_VALUES } -static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, - const uint32_t blockHeight, uint32_t* outBuf) { +static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 blockHeight, + u32* outBuf) { InputBitStream strm(inBuf); TexelWeightParams weightParams = DecodeBlockInfo(strm); @@ -1415,7 +1413,7 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, } // Read num partitions - uint32_t nPartitions = strm.ReadBits(2) + 1; + u32 nPartitions = strm.ReadBits(2) + 1; assert(nPartitions <= 4); if (nPartitions == 4 && weightParams.m_bDualPlane) { @@ -1428,17 +1426,17 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, // each partition. // Determine partitions, partition index, and color endpoint modes - int32_t planeIdx = -1; - uint32_t partitionIndex; - uint32_t colorEndpointMode[4] = {0, 0, 0, 0}; + s32 planeIdx = -1; + u32 partitionIndex; + u32 colorEndpointMode[4] = {0, 0, 0, 0}; // Define color data. - uint8_t colorEndpointData[16]; + u8 colorEndpointData[16]; memset(colorEndpointData, 0, sizeof(colorEndpointData)); OutputBitStream colorEndpointStream(colorEndpointData, 16 * 8, 0); // Read extra config data... - uint32_t baseCEM = 0; + u32 baseCEM = 0; if (nPartitions == 1) { colorEndpointMode[0] = strm.ReadBits(4); partitionIndex = 0; @@ -1446,14 +1444,14 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, partitionIndex = strm.ReadBits(10); baseCEM = strm.ReadBits(6); } - uint32_t baseMode = (baseCEM & 3); + u32 baseMode = (baseCEM & 3); // Remaining bits are color endpoint data... - uint32_t nWeightBits = weightParams.GetPackedBitSize(); - int32_t remainingBits = 128 - nWeightBits - strm.GetBitsRead(); + u32 nWeightBits = weightParams.GetPackedBitSize(); + s32 remainingBits = 128 - nWeightBits - strm.GetBitsRead(); // Consider extra bits prior to texel data... - uint32_t extraCEMbits = 0; + u32 extraCEMbits = 0; if (baseMode) { switch (nPartitions) { case 2: @@ -1473,17 +1471,17 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, remainingBits -= extraCEMbits; // Do we have a dual plane situation? - uint32_t planeSelectorBits = 0; + u32 planeSelectorBits = 0; if (weightParams.m_bDualPlane) { planeSelectorBits = 2; } remainingBits -= planeSelectorBits; // Read color data... - uint32_t colorDataBits = remainingBits; + u32 colorDataBits = remainingBits; while (remainingBits > 0) { - uint32_t nb = std::min(remainingBits, 8); - uint32_t b = strm.ReadBits(nb); + u32 nb = std::min(remainingBits, 8); + u32 b = strm.ReadBits(nb); colorEndpointStream.WriteBits(b, nb); remainingBits -= 8; } @@ -1493,24 +1491,24 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, // Read the rest of the CEM if (baseMode) { - uint32_t extraCEM = strm.ReadBits(extraCEMbits); - uint32_t CEM = (extraCEM << 6) | baseCEM; + u32 extraCEM = strm.ReadBits(extraCEMbits); + u32 CEM = (extraCEM << 6) | baseCEM; CEM >>= 2; bool C[4] = {0}; - for (uint32_t i = 0; i < nPartitions; i++) { + for (u32 i = 0; i < nPartitions; i++) { C[i] = CEM & 1; CEM >>= 1; } - uint8_t M[4] = {0}; - for (uint32_t i = 0; i < nPartitions; i++) { + u8 M[4] = {0}; + for (u32 i = 0; i < nPartitions; i++) { M[i] = CEM & 3; CEM >>= 2; assert(M[i] <= 3); } - for (uint32_t i = 0; i < nPartitions; i++) { + for (u32 i = 0; i < nPartitions; i++) { colorEndpointMode[i] = baseMode; if (!(C[i])) colorEndpointMode[i] -= 1; @@ -1518,35 +1516,35 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, colorEndpointMode[i] |= M[i]; } } else if (nPartitions > 1) { - uint32_t CEM = baseCEM >> 2; - for (uint32_t i = 0; i < nPartitions; i++) { + u32 CEM = baseCEM >> 2; + for (u32 i = 0; i < nPartitions; i++) { colorEndpointMode[i] = CEM; } } // Make sure everything up till here is sane. - for (uint32_t i = 0; i < nPartitions; i++) { + for (u32 i = 0; i < nPartitions; i++) { assert(colorEndpointMode[i] < 16); } assert(strm.GetBitsRead() + weightParams.GetPackedBitSize() == 128); // Decode both color data and texel weight data - uint32_t colorValues[32]; // Four values, two endpoints, four maximum paritions + u32 colorValues[32]; // Four values, two endpoints, four maximum paritions DecodeColorValues(colorValues, colorEndpointData, colorEndpointMode, nPartitions, colorDataBits); Pixel endpoints[4][2]; - const uint32_t* colorValuesPtr = colorValues; - for (uint32_t i = 0; i < nPartitions; i++) { + const u32* colorValuesPtr = colorValues; + for (u32 i = 0; i < nPartitions; i++) { ComputeEndpoints(endpoints[i][0], endpoints[i][1], colorValuesPtr, colorEndpointMode[i]); } // Read the texel weight data.. - uint8_t texelWeightData[16]; + u8 texelWeightData[16]; memcpy(texelWeightData, inBuf, sizeof(texelWeightData)); // Reverse everything - for (uint32_t i = 0; i < 8; i++) { + for (u32 i = 0; i < 8; i++) { // Taken from http://graphics.stanford.edu/~seander/bithacks.html#ReverseByteWith64Bits #define REVERSE_BYTE(b) (((b)*0x80200802ULL) & 0x0884422110ULL) * 0x0101010101ULL >> 32 unsigned char a = static_cast(REVERSE_BYTE(texelWeightData[i])); @@ -1558,10 +1556,10 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, } // Make sure that higher non-texel bits are set to zero - const uint32_t clearByteStart = (weightParams.GetPackedBitSize() >> 3) + 1; + const u32 clearByteStart = (weightParams.GetPackedBitSize() >> 3) + 1; texelWeightData[clearByteStart - 1] = texelWeightData[clearByteStart - 1] & - static_cast((1 << (weightParams.GetPackedBitSize() % 8)) - 1); + static_cast((1 << (weightParams.GetPackedBitSize() % 8)) - 1); memset(texelWeightData + clearByteStart, 0, 16 - clearByteStart); std::vector texelWeightValues; @@ -1572,36 +1570,36 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, weightParams.GetNumWeightValues()); // Blocks can be at most 12x12, so we can have as many as 144 weights - uint32_t weights[2][144]; + u32 weights[2][144]; UnquantizeTexelWeights(weights, texelWeightValues, weightParams, blockWidth, blockHeight); // Now that we have endpoints and weights, we can interpolate and generate // the proper decoding... - for (uint32_t j = 0; j < blockHeight; j++) - for (uint32_t i = 0; i < blockWidth; i++) { - uint32_t partition = Select2DPartition(partitionIndex, i, j, nPartitions, - (blockHeight * blockWidth) < 32); + for (u32 j = 0; j < blockHeight; j++) + for (u32 i = 0; i < blockWidth; i++) { + u32 partition = Select2DPartition(partitionIndex, i, j, nPartitions, + (blockHeight * blockWidth) < 32); assert(partition < nPartitions); Pixel p; - for (uint32_t c = 0; c < 4; c++) { - uint32_t C0 = endpoints[partition][0].Component(c); + for (u32 c = 0; c < 4; c++) { + u32 C0 = endpoints[partition][0].Component(c); C0 = Replicate(C0, 8, 16); - uint32_t C1 = endpoints[partition][1].Component(c); + u32 C1 = endpoints[partition][1].Component(c); C1 = Replicate(C1, 8, 16); - uint32_t plane = 0; + u32 plane = 0; if (weightParams.m_bDualPlane && (((planeIdx + 1) & 3) == c)) { plane = 1; } - uint32_t weight = weights[plane][j * blockWidth + i]; - uint32_t C = (C0 * (64 - weight) + C1 * weight + 32) / 64; + u32 weight = weights[plane][j * blockWidth + i]; + u32 C = (C0 * (64 - weight) + C1 * weight + 32) / 64; if (C == 65535) { p.Component(c) = 255; } else { double Cf = static_cast(C); - p.Component(c) = static_cast(255.0 * (Cf / 65536.0) + 0.5); + p.Component(c) = static_cast(255.0 * (Cf / 65536.0) + 0.5); } } @@ -1613,26 +1611,26 @@ static void DecompressBlock(const uint8_t inBuf[16], const uint32_t blockWidth, namespace Tegra::Texture::ASTC { -std::vector Decompress(const uint8_t* data, uint32_t width, uint32_t height, - uint32_t depth, uint32_t block_width, uint32_t block_height) { - uint32_t blockIdx = 0; +std::vector Decompress(const u8* data, u32 width, u32 height, u32 depth, u32 block_width, + u32 block_height) { + u32 blockIdx = 0; std::size_t depth_offset = 0; - std::vector outData(height * width * depth * 4); - for (uint32_t k = 0; k < depth; k++) { - for (uint32_t j = 0; j < height; j += block_height) { - for (uint32_t i = 0; i < width; i += block_width) { + std::vector outData(height * width * depth * 4); + for (u32 k = 0; k < depth; k++) { + for (u32 j = 0; j < height; j += block_height) { + for (u32 i = 0; i < width; i += block_width) { - const uint8_t* blockPtr = data + blockIdx * 16; + const u8* blockPtr = data + blockIdx * 16; // Blocks can be at most 12x12 - uint32_t uncompData[144]; + u32 uncompData[144]; ASTCC::DecompressBlock(blockPtr, block_width, block_height, uncompData); - uint32_t decompWidth = std::min(block_width, width - i); - uint32_t decompHeight = std::min(block_height, height - j); + u32 decompWidth = std::min(block_width, width - i); + u32 decompHeight = std::min(block_height, height - j); - uint8_t* outRow = depth_offset + outData.data() + (j * width + i) * 4; - for (uint32_t jj = 0; jj < decompHeight; jj++) { + u8* outRow = depth_offset + outData.data() + (j * width + i) * 4; + for (u32 jj = 0; jj < decompHeight; jj++) { memcpy(outRow + jj * width * 4, uncompData + jj * block_width, decompWidth * 4); } From 835a3d09c67eb5a028e5594f64490135b7b423a4 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 22:26:48 -0300 Subject: [PATCH 03/13] astc: Move Popcnt to an anonymous namespace and make it constexpr --- src/video_core/textures/astc.cpp | 22 +++++++++++++--------- 1 file changed, 13 insertions(+), 9 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index dcfab4dad6..aba47a0e81 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -25,6 +25,19 @@ #include "video_core/textures/astc.h" +namespace { + +/// Count the number of bits set in a number. +constexpr u32 Popcnt(u32 n) { + u32 c = 0; + for (; n; c++) { + n &= n - 1; + } + return c; +} + +} // Anonymous namespace + class InputBitStream { public: explicit InputBitStream(const unsigned char* ptr, int start_offset = 0) @@ -212,15 +225,6 @@ public: return totalBits; } - // Count the number of bits set in a number. - static inline u32 Popcnt(u32 n) { - u32 c; - for (c = 0; n; c++) { - n &= n - 1; - } - return c; - } - // Returns a new instance of this struct that corresponds to the // can take no more than maxval values static IntegerEncodedValue CreateEncoding(u32 maxVal) { From e7d97605e8f0337406d3fcabca84cb32daf78950 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 22:28:51 -0300 Subject: [PATCH 04/13] astc: Rename C types to common_types --- src/video_core/textures/astc.cpp | 157 +++++++++++++++---------------- 1 file changed, 78 insertions(+), 79 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index aba47a0e81..7d5796794e 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -17,7 +17,6 @@ #include #include -#include #include #include @@ -40,18 +39,18 @@ constexpr u32 Popcnt(u32 n) { class InputBitStream { public: - explicit InputBitStream(const unsigned char* ptr, int start_offset = 0) + explicit InputBitStream(const u8* ptr, s32 start_offset = 0) : m_CurByte(ptr), m_NextBit(start_offset % 8) {} ~InputBitStream() = default; - int GetBitsRead() const { + s32 GetBitsRead() const { return m_BitsRead; } - int ReadBit() { + s32 ReadBit() { - int bit = *m_CurByte >> m_NextBit++; + s32 bit = *m_CurByte >> m_NextBit++; while (m_NextBit >= 8) { m_NextBit -= 8; m_CurByte++; @@ -61,57 +60,57 @@ public: return bit & 1; } - unsigned int ReadBits(unsigned int nBits) { - unsigned int ret = 0; - for (unsigned int i = 0; i < nBits; i++) { + u32 ReadBits(u32 nBits) { + u32 ret = 0; + for (u32 i = 0; i < nBits; i++) { ret |= (ReadBit() & 1) << i; } return ret; } private: - const unsigned char* m_CurByte; - int m_NextBit = 0; - int m_BitsRead = 0; + const u8* m_CurByte; + s32 m_NextBit = 0; + s32 m_BitsRead = 0; }; class OutputBitStream { public: - explicit OutputBitStream(unsigned char* ptr, int nBits = 0, int start_offset = 0) + explicit OutputBitStream(u8* ptr, s32 nBits = 0, s32 start_offset = 0) : m_NumBits(nBits), m_CurByte(ptr), m_NextBit(start_offset % 8) {} ~OutputBitStream() = default; - int GetBitsWritten() const { + s32 GetBitsWritten() const { return m_BitsWritten; } - void WriteBitsR(unsigned int val, unsigned int nBits) { - for (unsigned int i = 0; i < nBits; i++) { + void WriteBitsR(u32 val, u32 nBits) { + for (u32 i = 0; i < nBits; i++) { WriteBit((val >> (nBits - i - 1)) & 1); } } - void WriteBits(unsigned int val, unsigned int nBits) { - for (unsigned int i = 0; i < nBits; i++) { + void WriteBits(u32 val, u32 nBits) { + for (u32 i = 0; i < nBits; i++) { WriteBit((val >> i) & 1); } } private: - void WriteBit(int b) { + void WriteBit(s32 b) { if (done) return; - const unsigned int mask = 1 << m_NextBit++; + const u32 mask = 1 << m_NextBit++; // clear the bit - *m_CurByte &= static_cast(~mask); + *m_CurByte &= static_cast(~mask); // Write the bit, if necessary if (b) - *m_CurByte |= static_cast(mask); + *m_CurByte |= static_cast(mask); // Next byte? if (m_NextBit >= 8) { @@ -122,10 +121,10 @@ private: done = done || ++m_BitsWritten >= m_NumBits; } - int m_BitsWritten = 0; - const int m_NumBits; - unsigned char* m_CurByte; - int m_NextBit = 0; + s32 m_BitsWritten = 0; + const s32 m_NumBits; + u8* m_CurByte; + s32 m_NextBit = 0; bool done = false; }; @@ -159,7 +158,7 @@ private: const IntType& m_Bits; }; -enum class IntegerEncoding { JustBits, Quint, Trit }; +enum class IntegerEncoding { JustBits, Qus32, Trit }; class IntegerEncodedValue { private: @@ -167,7 +166,7 @@ private: const u32 m_NumBits; u32 m_BitValue; union { - u32 m_QuintValue; + u32 m_Qus32Value; u32 m_TritValue; }; @@ -203,11 +202,11 @@ public: m_TritValue = val; } - u32 GetQuintValue() const { - return m_QuintValue; + u32 GetQus32Value() const { + return m_Qus32Value; } - void SetQuintValue(u32 val) { - m_QuintValue = val; + void SetQus32Value(u32 val) { + m_Qus32Value = val; } bool MatchesEncoding(const IntegerEncodedValue& other) const { @@ -219,7 +218,7 @@ public: u32 totalBits = m_NumBits * nVals; if (m_Encoding == IntegerEncoding::Trit) { totalBits += (nVals * 8 + 4) / 5; - } else if (m_Encoding == IntegerEncoding::Quint) { + } else if (m_Encoding == IntegerEncoding::Qus32) { totalBits += (nVals * 7 + 2) / 3; } return totalBits; @@ -243,10 +242,10 @@ public: // Is maxVal of the type 5*2^n - 1? if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) { - return IntegerEncodedValue(IntegerEncoding::Quint, Popcnt(check / 5 - 1)); + return IntegerEncodedValue(IntegerEncoding::Qus32, Popcnt(check / 5 - 1)); } - // Apparently it can't be represented with a bounded integer sequence... + // Apparently it can't be represented with a bounded s32eger sequence... // just iterate. maxVal--; } @@ -265,8 +264,8 @@ public: u32 nValsDecoded = 0; while (nValsDecoded < nValues) { switch (val.GetEncoding()) { - case IntegerEncoding::Quint: - DecodeQuintBlock(bits, result, val.BaseBitLength()); + case IntegerEncoding::Qus32: + DecodeQus32Block(bits, result, val.BaseBitLength()); nValsDecoded += 3; break; @@ -345,7 +344,7 @@ private: } } - static void DecodeQuintBlock(InputBitStream& bits, std::vector& result, + static void DecodeQus32Block(InputBitStream& bits, std::vector& result, u32 nBitsPerValue) { // Implement the algorithm in section C.2.12 u32 m[3]; @@ -386,9 +385,9 @@ private: } for (u32 i = 0; i < 3; i++) { - IntegerEncodedValue val(IntegerEncoding::Quint, nBitsPerValue); + IntegerEncodedValue val(IntegerEncoding::Qus32, nBitsPerValue); val.m_BitValue = m[i]; - val.m_QuintValue = q[i]; + val.m_Qus32Value = q[i]; result.push_back(val); } } @@ -626,7 +625,7 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { static void FillVoidExtentLDR(InputBitStream& strm, u32* const outBuf, u32 blockWidth, u32 blockHeight) { // Don't actually care about the void extent, just read the bits... - for (int i = 0; i < 4; ++i) { + for (s32 i = 0; i < 4; ++i) { strm.ReadBits(13); } @@ -687,7 +686,7 @@ protected: public: Pixel() = default; - Pixel(u32 a, u32 r, u32 g, u32 b, unsigned bitDepth = 8) + Pixel(u32 a, u32 r, u32 g, u32 b, u32 bitDepth = 8) : m_BitDepth{u8(bitDepth), u8(bitDepth), u8(bitDepth), u8(bitDepth)}, color{static_cast(a), static_cast(r), static_cast(g), static_cast(b)} {} @@ -772,13 +771,13 @@ public: } void GetBitDepth(u8 (&outDepth)[4]) const { - for (int i = 0; i < 4; i++) { + for (s32 i = 0; i < 4; i++) { outDepth[i] = m_BitDepth[i]; } } // Take all of the components, transform them to their 8-bit variants, - // and then pack each channel into an R8G8B8A8 32-bit integer. We assume + // and then pack each channel s32o an R8G8B8A8 32-bit s32eger. We assume // that the architecture is little-endian, so the alpha channel will end // up in the most-significant byte. u32 Pack() const { @@ -838,7 +837,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP } } - // We now have enough to decode our integer sequence. + // We now have enough to decode our s32eger sequence. std::vector decodedColorValues; InputBitStream colorStream(data); IntegerEncodedValue::DecodeIntegerSequence(decodedColorValues, colorStream, range, nValues); @@ -920,9 +919,9 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP } // case IntegerEncoding::Trit break; - case IntegerEncoding::Quint: { + case IntegerEncoding::Qus32: { - D = val.GetQuintValue(); + D = val.GetQus32Value(); switch (bitlen) { case 1: { @@ -958,10 +957,10 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP } break; default: - assert(!"Unsupported quint encoding for color values!"); + assert(!"Unsupported qus32 encoding for color values!"); break; } // switch(bitlen) - } // case IntegerEncoding::Quint + } // case IntegerEncoding::Qus32 break; } // switch(val.GetEncoding()) @@ -1024,8 +1023,8 @@ static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) { } } break; - case IntegerEncoding::Quint: { - D = val.GetQuintValue(); + case IntegerEncoding::Qus32: { + D = val.GetQus32Value(); assert(D < 5); switch (bitlen) { @@ -1045,7 +1044,7 @@ static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) { } break; default: - assert(!"Invalid quint encoding for texel weight"); + assert(!"Invalid qus32 encoding for texel weight"); break; } } break; @@ -1260,8 +1259,8 @@ static inline u32 Select2DPartition(s32 seed, s32 x, s32 y, s32 partitionCount, } // Section C.2.14 -static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues, - u32 colorEndpointMode) { +static void ComputeEndpos32s(Pixel& ep1, Pixel& ep2, const u32*& colorValues, + u32 colorEndpos32Mode) { #define READ_UINT_VALUES(N) \ u32 v[N]; \ for (u32 i = 0; i < N; i++) { \ @@ -1274,7 +1273,7 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues, v[i] = static_cast(*(colorValues++)); \ } - switch (colorEndpointMode) { + switch (colorEndpos32Mode) { case 0: { READ_UINT_VALUES(2) ep1 = Pixel(0xFF, v[0], v[0], v[0]); @@ -1373,7 +1372,7 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues, } break; default: - assert(!"Unsupported color endpoint mode (is it HDR?)"); + assert(!"Unsupported color endpos32 mode (is it HDR?)"); break; } @@ -1426,23 +1425,23 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 return; } - // Based on the number of partitions, read the color endpoint mode for + // Based on the number of partitions, read the color endpos32 mode for // each partition. - // Determine partitions, partition index, and color endpoint modes + // Determine partitions, partition index, and color endpos32 modes s32 planeIdx = -1; u32 partitionIndex; - u32 colorEndpointMode[4] = {0, 0, 0, 0}; + u32 colorEndpos32Mode[4] = {0, 0, 0, 0}; // Define color data. - u8 colorEndpointData[16]; - memset(colorEndpointData, 0, sizeof(colorEndpointData)); - OutputBitStream colorEndpointStream(colorEndpointData, 16 * 8, 0); + u8 colorEndpos32Data[16]; + memset(colorEndpos32Data, 0, sizeof(colorEndpos32Data)); + OutputBitStream colorEndpos32Stream(colorEndpos32Data, 16 * 8, 0); // Read extra config data... u32 baseCEM = 0; if (nPartitions == 1) { - colorEndpointMode[0] = strm.ReadBits(4); + colorEndpos32Mode[0] = strm.ReadBits(4); partitionIndex = 0; } else { partitionIndex = strm.ReadBits(10); @@ -1450,7 +1449,7 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 } u32 baseMode = (baseCEM & 3); - // Remaining bits are color endpoint data... + // Remaining bits are color endpos32 data... u32 nWeightBits = weightParams.GetPackedBitSize(); s32 remainingBits = 128 - nWeightBits - strm.GetBitsRead(); @@ -1486,7 +1485,7 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 while (remainingBits > 0) { u32 nb = std::min(remainingBits, 8); u32 b = strm.ReadBits(nb); - colorEndpointStream.WriteBits(b, nb); + colorEndpos32Stream.WriteBits(b, nb); remainingBits -= 8; } @@ -1513,34 +1512,34 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 } for (u32 i = 0; i < nPartitions; i++) { - colorEndpointMode[i] = baseMode; + colorEndpos32Mode[i] = baseMode; if (!(C[i])) - colorEndpointMode[i] -= 1; - colorEndpointMode[i] <<= 2; - colorEndpointMode[i] |= M[i]; + colorEndpos32Mode[i] -= 1; + colorEndpos32Mode[i] <<= 2; + colorEndpos32Mode[i] |= M[i]; } } else if (nPartitions > 1) { u32 CEM = baseCEM >> 2; for (u32 i = 0; i < nPartitions; i++) { - colorEndpointMode[i] = CEM; + colorEndpos32Mode[i] = CEM; } } // Make sure everything up till here is sane. for (u32 i = 0; i < nPartitions; i++) { - assert(colorEndpointMode[i] < 16); + assert(colorEndpos32Mode[i] < 16); } assert(strm.GetBitsRead() + weightParams.GetPackedBitSize() == 128); // Decode both color data and texel weight data - u32 colorValues[32]; // Four values, two endpoints, four maximum paritions - DecodeColorValues(colorValues, colorEndpointData, colorEndpointMode, nPartitions, + u32 colorValues[32]; // Four values, two endpos32s, four maximum paritions + DecodeColorValues(colorValues, colorEndpos32Data, colorEndpos32Mode, nPartitions, colorDataBits); - Pixel endpoints[4][2]; + Pixel endpos32s[4][2]; const u32* colorValuesPtr = colorValues; for (u32 i = 0; i < nPartitions; i++) { - ComputeEndpoints(endpoints[i][0], endpoints[i][1], colorValuesPtr, colorEndpointMode[i]); + ComputeEndpos32s(endpos32s[i][0], endpos32s[i][1], colorValuesPtr, colorEndpos32Mode[i]); } // Read the texel weight data.. @@ -1551,8 +1550,8 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 for (u32 i = 0; i < 8; i++) { // Taken from http://graphics.stanford.edu/~seander/bithacks.html#ReverseByteWith64Bits #define REVERSE_BYTE(b) (((b)*0x80200802ULL) & 0x0884422110ULL) * 0x0101010101ULL >> 32 - unsigned char a = static_cast(REVERSE_BYTE(texelWeightData[i])); - unsigned char b = static_cast(REVERSE_BYTE(texelWeightData[15 - i])); + u8 a = static_cast(REVERSE_BYTE(texelWeightData[i])); + u8 b = static_cast(REVERSE_BYTE(texelWeightData[15 - i])); #undef REVERSE_BYTE texelWeightData[i] = b; @@ -1577,7 +1576,7 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 u32 weights[2][144]; UnquantizeTexelWeights(weights, texelWeightValues, weightParams, blockWidth, blockHeight); - // Now that we have endpoints and weights, we can interpolate and generate + // Now that we have endpos32s and weights, we can s32erpolate and generate // the proper decoding... for (u32 j = 0; j < blockHeight; j++) for (u32 i = 0; i < blockWidth; i++) { @@ -1587,9 +1586,9 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 Pixel p; for (u32 c = 0; c < 4; c++) { - u32 C0 = endpoints[partition][0].Component(c); + u32 C0 = endpos32s[partition][0].Component(c); C0 = Replicate(C0, 8, 16); - u32 C1 = endpoints[partition][1].Component(c); + u32 C1 = endpos32s[partition][1].Component(c); C1 = Replicate(C1, 8, 16); u32 plane = 0; From 5ed377b989b01141a65f02a0ae0a9ed52ce43346 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 22:30:31 -0300 Subject: [PATCH 05/13] astc: Make IntegerEncodedValue trivially copyable --- src/video_core/textures/astc.cpp | 11 ++--------- 1 file changed, 2 insertions(+), 9 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 7d5796794e..02cbad1ea8 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -162,8 +162,8 @@ enum class IntegerEncoding { JustBits, Qus32, Trit }; class IntegerEncodedValue { private: - const IntegerEncoding m_Encoding; - const u32 m_NumBits; + IntegerEncoding m_Encoding; + u32 m_NumBits; u32 m_BitValue; union { u32 m_Qus32Value; @@ -171,13 +171,6 @@ private: }; public: - // Jank, but we're not doing any heavy lifting in this class, so it's - // probably OK. It allows us to use these in std::vectors... - IntegerEncodedValue& operator=(const IntegerEncodedValue& other) { - new (this) IntegerEncodedValue(other); - return *this; - } - IntegerEncodedValue(IntegerEncoding encoding, u32 numBits) : m_Encoding(encoding), m_NumBits(numBits) {} From 70a31eda62eba7f4e3a700a356dec08478efb5ef Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 22:36:45 -0300 Subject: [PATCH 06/13] astc: Make IntegerEncodedValue constructor constexpr --- src/video_core/textures/astc.cpp | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 02cbad1ea8..f4513998c2 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -162,16 +162,17 @@ enum class IntegerEncoding { JustBits, Qus32, Trit }; class IntegerEncodedValue { private: - IntegerEncoding m_Encoding; - u32 m_NumBits; - u32 m_BitValue; + IntegerEncoding m_Encoding{}; + u32 m_NumBits = 0; + u32 m_BitValue = 0; union { - u32 m_Qus32Value; + u32 m_Qus32Value = 0; u32 m_TritValue; }; public: - IntegerEncodedValue(IntegerEncoding encoding, u32 numBits) + constexpr IntegerEncodedValue() = default; + constexpr IntegerEncodedValue(IntegerEncoding encoding, u32 numBits) : m_Encoding(encoding), m_NumBits(numBits) {} IntegerEncoding GetEncoding() const { From e183820956b528f64bb5d6e05052752bf437d3a1 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 22:49:28 -0300 Subject: [PATCH 07/13] astc: Make IntegerEncodedValue a trivial structure --- src/video_core/textures/astc.cpp | 397 ++++++++++++++----------------- 1 file changed, 181 insertions(+), 216 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index f4513998c2..2d948ee624 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -160,232 +160,198 @@ private: enum class IntegerEncoding { JustBits, Qus32, Trit }; -class IntegerEncodedValue { -private: - IntegerEncoding m_Encoding{}; - u32 m_NumBits = 0; - u32 m_BitValue = 0; - union { - u32 m_Qus32Value = 0; - u32 m_TritValue; - }; +struct IntegerEncodedValue { + constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_) + : encoding{encoding_}, num_bits{num_bits_} {} -public: - constexpr IntegerEncodedValue() = default; - constexpr IntegerEncodedValue(IntegerEncoding encoding, u32 numBits) - : m_Encoding(encoding), m_NumBits(numBits) {} - - IntegerEncoding GetEncoding() const { - return m_Encoding; - } - u32 BaseBitLength() const { - return m_NumBits; - } - - u32 GetBitValue() const { - return m_BitValue; - } - void SetBitValue(u32 val) { - m_BitValue = val; - } - - u32 GetTritValue() const { - return m_TritValue; - } - void SetTritValue(u32 val) { - m_TritValue = val; - } - - u32 GetQus32Value() const { - return m_Qus32Value; - } - void SetQus32Value(u32 val) { - m_Qus32Value = val; - } - - bool MatchesEncoding(const IntegerEncodedValue& other) const { - return m_Encoding == other.m_Encoding && m_NumBits == other.m_NumBits; + constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const { + return encoding == other.encoding && num_bits == other.num_bits; } // Returns the number of bits required to encode nVals values. u32 GetBitLength(u32 nVals) const { - u32 totalBits = m_NumBits * nVals; - if (m_Encoding == IntegerEncoding::Trit) { + u32 totalBits = num_bits * nVals; + if (encoding == IntegerEncoding::Trit) { totalBits += (nVals * 8 + 4) / 5; - } else if (m_Encoding == IntegerEncoding::Qus32) { + } else if (encoding == IntegerEncoding::Qus32) { totalBits += (nVals * 7 + 2) / 3; } return totalBits; } - // Returns a new instance of this struct that corresponds to the - // can take no more than maxval values - static IntegerEncodedValue CreateEncoding(u32 maxVal) { - while (maxVal > 0) { - u32 check = maxVal + 1; + IntegerEncoding encoding; + u32 num_bits; + u32 bit_value = 0; + union { + u32 qus32_value = 0; + u32 trit_value; + }; +}; - // Is maxVal a power of two? - if (!(check & (check - 1))) { - return IntegerEncodedValue(IntegerEncoding::JustBits, Popcnt(maxVal)); - } +static void DecodeTritBlock(InputBitStream& bits, std::vector& result, + u32 nBitsPerValue) { + // Implement the algorithm in section C.2.12 + u32 m[5]; + u32 t[5]; + u32 T; - // Is maxVal of the type 3*2^n - 1? - if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) { - return IntegerEncodedValue(IntegerEncoding::Trit, Popcnt(check / 3 - 1)); - } + // Read the trit encoded block according to + // table C.2.14 + m[0] = bits.ReadBits(nBitsPerValue); + T = bits.ReadBits(2); + m[1] = bits.ReadBits(nBitsPerValue); + T |= bits.ReadBits(2) << 2; + m[2] = bits.ReadBits(nBitsPerValue); + T |= bits.ReadBit() << 4; + m[3] = bits.ReadBits(nBitsPerValue); + T |= bits.ReadBits(2) << 5; + m[4] = bits.ReadBits(nBitsPerValue); + T |= bits.ReadBit() << 7; - // Is maxVal of the type 5*2^n - 1? - if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) { - return IntegerEncodedValue(IntegerEncoding::Qus32, Popcnt(check / 5 - 1)); - } + u32 C = 0; - // Apparently it can't be represented with a bounded s32eger sequence... - // just iterate. - maxVal--; - } - return IntegerEncodedValue(IntegerEncoding::JustBits, 0); - } - - // Fills result with the values that are encoded in the given - // bitstream. We must know beforehand what the maximum possible - // value is, and how many values we're decoding. - static void DecodeIntegerSequence(std::vector& result, - InputBitStream& bits, u32 maxRange, u32 nValues) { - // Determine encoding parameters - IntegerEncodedValue val = IntegerEncodedValue::CreateEncoding(maxRange); - - // Start decoding - u32 nValsDecoded = 0; - while (nValsDecoded < nValues) { - switch (val.GetEncoding()) { - case IntegerEncoding::Qus32: - DecodeQus32Block(bits, result, val.BaseBitLength()); - nValsDecoded += 3; - break; - - case IntegerEncoding::Trit: - DecodeTritBlock(bits, result, val.BaseBitLength()); - nValsDecoded += 5; - break; - - case IntegerEncoding::JustBits: - val.SetBitValue(bits.ReadBits(val.BaseBitLength())); - result.push_back(val); - nValsDecoded++; - break; - } - } - } - -private: - static void DecodeTritBlock(InputBitStream& bits, std::vector& result, - u32 nBitsPerValue) { - // Implement the algorithm in section C.2.12 - u32 m[5]; - u32 t[5]; - u32 T; - - // Read the trit encoded block according to - // table C.2.14 - m[0] = bits.ReadBits(nBitsPerValue); - T = bits.ReadBits(2); - m[1] = bits.ReadBits(nBitsPerValue); - T |= bits.ReadBits(2) << 2; - m[2] = bits.ReadBits(nBitsPerValue); - T |= bits.ReadBit() << 4; - m[3] = bits.ReadBits(nBitsPerValue); - T |= bits.ReadBits(2) << 5; - m[4] = bits.ReadBits(nBitsPerValue); - T |= bits.ReadBit() << 7; - - u32 C = 0; - - Bits Tb(T); - if (Tb(2, 4) == 7) { - C = (Tb(5, 7) << 2) | Tb(0, 1); - t[4] = t[3] = 2; + Bits Tb(T); + if (Tb(2, 4) == 7) { + C = (Tb(5, 7) << 2) | Tb(0, 1); + t[4] = t[3] = 2; + } else { + C = Tb(0, 4); + if (Tb(5, 6) == 3) { + t[4] = 2; + t[3] = Tb[7]; } else { - C = Tb(0, 4); - if (Tb(5, 6) == 3) { - t[4] = 2; - t[3] = Tb[7]; - } else { - t[4] = Tb[7]; - t[3] = Tb(5, 6); - } + t[4] = Tb[7]; + t[3] = Tb(5, 6); + } + } + + Bits Cb(C); + if (Cb(0, 1) == 3) { + t[2] = 2; + t[1] = Cb[4]; + t[0] = (Cb[3] << 1) | (Cb[2] & ~Cb[3]); + } else if (Cb(2, 3) == 3) { + t[2] = 2; + t[1] = 2; + t[0] = Cb(0, 1); + } else { + t[2] = Cb[4]; + t[1] = Cb(2, 3); + t[0] = (Cb[1] << 1) | (Cb[0] & ~Cb[1]); + } + + for (std::size_t i = 0; i < 5; ++i) { + IntegerEncodedValue& val = result.emplace_back(IntegerEncoding::Trit, nBitsPerValue); + val.bit_value = m[i]; + val.trit_value = t[i]; + } +} + +static void DecodeQus32Block(InputBitStream& bits, std::vector& result, + u32 nBitsPerValue) { + // Implement the algorithm in section C.2.12 + u32 m[3]; + u32 q[3]; + u32 Q; + + // Read the trit encoded block according to + // table C.2.15 + m[0] = bits.ReadBits(nBitsPerValue); + Q = bits.ReadBits(3); + m[1] = bits.ReadBits(nBitsPerValue); + Q |= bits.ReadBits(2) << 3; + m[2] = bits.ReadBits(nBitsPerValue); + Q |= bits.ReadBits(2) << 5; + + Bits Qb(Q); + if (Qb(1, 2) == 3 && Qb(5, 6) == 0) { + q[0] = q[1] = 4; + q[2] = (Qb[0] << 2) | ((Qb[4] & ~Qb[0]) << 1) | (Qb[3] & ~Qb[0]); + } else { + u32 C = 0; + if (Qb(1, 2) == 3) { + q[2] = 4; + C = (Qb(3, 4) << 3) | ((~Qb(5, 6) & 3) << 1) | Qb[0]; + } else { + q[2] = Qb(5, 6); + C = Qb(0, 4); } Bits Cb(C); - if (Cb(0, 1) == 3) { - t[2] = 2; - t[1] = Cb[4]; - t[0] = (Cb[3] << 1) | (Cb[2] & ~Cb[3]); - } else if (Cb(2, 3) == 3) { - t[2] = 2; - t[1] = 2; - t[0] = Cb(0, 1); + if (Cb(0, 2) == 5) { + q[1] = 4; + q[0] = Cb(3, 4); } else { - t[2] = Cb[4]; - t[1] = Cb(2, 3); - t[0] = (Cb[1] << 1) | (Cb[0] & ~Cb[1]); - } - - for (u32 i = 0; i < 5; i++) { - IntegerEncodedValue val(IntegerEncoding::Trit, nBitsPerValue); - val.SetBitValue(m[i]); - val.SetTritValue(t[i]); - result.push_back(val); + q[1] = Cb(3, 4); + q[0] = Cb(0, 2); } } - static void DecodeQus32Block(InputBitStream& bits, std::vector& result, - u32 nBitsPerValue) { - // Implement the algorithm in section C.2.12 - u32 m[3]; - u32 q[3]; - u32 Q; + for (std::size_t i = 0; i < 3; ++i) { + IntegerEncodedValue& val = result.emplace_back(IntegerEncoding::Qus32, nBitsPerValue); + val.bit_value = m[i]; + val.qus32_value = q[i]; + } +} - // Read the trit encoded block according to - // table C.2.15 - m[0] = bits.ReadBits(nBitsPerValue); - Q = bits.ReadBits(3); - m[1] = bits.ReadBits(nBitsPerValue); - Q |= bits.ReadBits(2) << 3; - m[2] = bits.ReadBits(nBitsPerValue); - Q |= bits.ReadBits(2) << 5; +// Returns a new instance of this struct that corresponds to the +// can take no more than maxval values +static IntegerEncodedValue CreateEncoding(u32 maxVal) { + while (maxVal > 0) { + u32 check = maxVal + 1; - Bits Qb(Q); - if (Qb(1, 2) == 3 && Qb(5, 6) == 0) { - q[0] = q[1] = 4; - q[2] = (Qb[0] << 2) | ((Qb[4] & ~Qb[0]) << 1) | (Qb[3] & ~Qb[0]); - } else { - u32 C = 0; - if (Qb(1, 2) == 3) { - q[2] = 4; - C = (Qb(3, 4) << 3) | ((~Qb(5, 6) & 3) << 1) | Qb[0]; - } else { - q[2] = Qb(5, 6); - C = Qb(0, 4); - } - - Bits Cb(C); - if (Cb(0, 2) == 5) { - q[1] = 4; - q[0] = Cb(3, 4); - } else { - q[1] = Cb(3, 4); - q[0] = Cb(0, 2); - } + // Is maxVal a power of two? + if (!(check & (check - 1))) { + return IntegerEncodedValue(IntegerEncoding::JustBits, Popcnt(maxVal)); } - for (u32 i = 0; i < 3; i++) { - IntegerEncodedValue val(IntegerEncoding::Qus32, nBitsPerValue); - val.m_BitValue = m[i]; - val.m_Qus32Value = q[i]; + // Is maxVal of the type 3*2^n - 1? + if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) { + return IntegerEncodedValue(IntegerEncoding::Trit, Popcnt(check / 3 - 1)); + } + + // Is maxVal of the type 5*2^n - 1? + if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) { + return IntegerEncodedValue(IntegerEncoding::Qus32, Popcnt(check / 5 - 1)); + } + + // Apparently it can't be represented with a bounded s32eger sequence... + // just iterate. + maxVal--; + } + return IntegerEncodedValue(IntegerEncoding::JustBits, 0); +} + +// Fills result with the values that are encoded in the given +// bitstream. We must know beforehand what the maximum possible +// value is, and how many values we're decoding. +static void DecodeIntegerSequence(std::vector& result, InputBitStream& bits, + u32 maxRange, u32 nValues) { + // Determine encoding parameters + IntegerEncodedValue val = CreateEncoding(maxRange); + + // Start decoding + u32 nValsDecoded = 0; + while (nValsDecoded < nValues) { + switch (val.encoding) { + case IntegerEncoding::Qus32: + DecodeQus32Block(bits, result, val.num_bits); + nValsDecoded += 3; + break; + + case IntegerEncoding::Trit: + DecodeTritBlock(bits, result, val.num_bits); + nValsDecoded += 5; + break; + + case IntegerEncoding::JustBits: + val.bit_value = bits.ReadBits(val.num_bits); result.push_back(val); + nValsDecoded++; + break; } } -}; +} namespace ASTCC { @@ -405,7 +371,7 @@ struct TexelWeightParams { nIdxs *= 2; } - return IntegerEncodedValue::CreateEncoding(m_MaxWeight).GetBitLength(nIdxs); + return CreateEncoding(m_MaxWeight).GetBitLength(nIdxs); } u32 GetNumWeightValues() const { @@ -814,12 +780,12 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP // figure out the max value for each of them... u32 range = 256; while (--range > 0) { - IntegerEncodedValue val = IntegerEncodedValue::CreateEncoding(range); + IntegerEncodedValue val = CreateEncoding(range); u32 bitLength = val.GetBitLength(nValues); if (bitLength <= nBitsForColorData) { // Find the smallest possible range that matches the given encoding while (--range > 0) { - IntegerEncodedValue newval = IntegerEncodedValue::CreateEncoding(range); + IntegerEncodedValue newval = CreateEncoding(range); if (!newval.MatchesEncoding(val)) { break; } @@ -834,7 +800,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP // We now have enough to decode our s32eger sequence. std::vector decodedColorValues; InputBitStream colorStream(data); - IntegerEncodedValue::DecodeIntegerSequence(decodedColorValues, colorStream, range, nValues); + DecodeIntegerSequence(decodedColorValues, colorStream, range, nValues); // Once we have the decoded values, we need to dequantize them to the 0-255 range // This procedure is outlined in ASTC spec C.2.13 @@ -846,8 +812,8 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP } const IntegerEncodedValue& val = *itr; - u32 bitlen = val.BaseBitLength(); - u32 bitval = val.GetBitValue(); + u32 bitlen = val.num_bits; + u32 bitval = val.bit_value; assert(bitlen >= 1); @@ -855,7 +821,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP // A is just the lsb replicated 9 times. A = Replicate(bitval & 1, 1, 9); - switch (val.GetEncoding()) { + switch (val.encoding) { // Replicate bits case IntegerEncoding::JustBits: out[outIdx++] = Replicate(bitval, bitlen, 8); @@ -864,7 +830,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP // Use algorithm in C.2.13 case IntegerEncoding::Trit: { - D = val.GetTritValue(); + D = val.trit_value; switch (bitlen) { case 1: { @@ -915,7 +881,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP case IntegerEncoding::Qus32: { - D = val.GetQus32Value(); + D = val.qus32_value; switch (bitlen) { case 1: { @@ -956,9 +922,9 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP } // switch(bitlen) } // case IntegerEncoding::Qus32 break; - } // switch(val.GetEncoding()) + } // switch(val.encoding) - if (val.GetEncoding() != IntegerEncoding::JustBits) { + if (val.encoding != IntegerEncoding::JustBits) { u32 T = D * C + B; T ^= A; T = (A & 0x80) | (T >> 2); @@ -973,20 +939,20 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP } static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) { - u32 bitval = val.GetBitValue(); - u32 bitlen = val.BaseBitLength(); + u32 bitval = val.bit_value; + u32 bitlen = val.num_bits; u32 A = Replicate(bitval & 1, 1, 7); u32 B = 0, C = 0, D = 0; u32 result = 0; - switch (val.GetEncoding()) { + switch (val.encoding) { case IntegerEncoding::JustBits: result = Replicate(bitval, bitlen, 6); break; case IntegerEncoding::Trit: { - D = val.GetTritValue(); + D = val.trit_value; assert(D < 3); switch (bitlen) { @@ -1018,7 +984,7 @@ static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) { } break; case IntegerEncoding::Qus32: { - D = val.GetQus32Value(); + D = val.qus32_value; assert(D < 5); switch (bitlen) { @@ -1044,7 +1010,7 @@ static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) { } break; } - if (val.GetEncoding() != IntegerEncoding::JustBits && bitlen > 0) { + if (val.encoding != IntegerEncoding::JustBits && bitlen > 0) { // Decode the value... result = D * C + B; result ^= A; @@ -1562,9 +1528,8 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 std::vector texelWeightValues; InputBitStream weightStream(texelWeightData); - IntegerEncodedValue::DecodeIntegerSequence(texelWeightValues, weightStream, - weightParams.m_MaxWeight, - weightParams.GetNumWeightValues()); + DecodeIntegerSequence(texelWeightValues, weightStream, weightParams.m_MaxWeight, + weightParams.GetNumWeightValues()); // Blocks can be at most 12x12, so we can have as many as 144 weights u32 weights[2][144]; From 801fd04f75fbd5072139759c2a7aac4571b29885 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 23:40:02 -0300 Subject: [PATCH 08/13] astc: Create a LUT at compile time for encoding values --- src/video_core/textures/astc.cpp | 26 +++++++++++++++++++------- 1 file changed, 19 insertions(+), 7 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 2d948ee624..eb93b44bd1 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -161,6 +161,8 @@ private: enum class IntegerEncoding { JustBits, Qus32, Trit }; struct IntegerEncodedValue { + constexpr IntegerEncodedValue() = default; + constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_) : encoding{encoding_}, num_bits{num_bits_} {} @@ -179,8 +181,8 @@ struct IntegerEncodedValue { return totalBits; } - IntegerEncoding encoding; - u32 num_bits; + IntegerEncoding encoding{}; + u32 num_bits = 0; u32 bit_value = 0; union { u32 qus32_value = 0; @@ -296,7 +298,7 @@ static void DecodeQus32Block(InputBitStream& bits, std::vector 0) { u32 check = maxVal + 1; @@ -322,13 +324,23 @@ static IntegerEncodedValue CreateEncoding(u32 maxVal) { return IntegerEncodedValue(IntegerEncoding::JustBits, 0); } +static constexpr std::array MakeEncodedValues() { + std::array encodings{}; + for (std::size_t i = 0; i < encodings.size(); ++i) { + encodings[i] = CreateEncoding(static_cast(i)); + } + return encodings; +} + +static constexpr std::array EncodingsValues = MakeEncodedValues(); + // Fills result with the values that are encoded in the given // bitstream. We must know beforehand what the maximum possible // value is, and how many values we're decoding. static void DecodeIntegerSequence(std::vector& result, InputBitStream& bits, u32 maxRange, u32 nValues) { // Determine encoding parameters - IntegerEncodedValue val = CreateEncoding(maxRange); + IntegerEncodedValue val = EncodingsValues[maxRange]; // Start decoding u32 nValsDecoded = 0; @@ -371,7 +383,7 @@ struct TexelWeightParams { nIdxs *= 2; } - return CreateEncoding(m_MaxWeight).GetBitLength(nIdxs); + return EncodingsValues[m_MaxWeight].GetBitLength(nIdxs); } u32 GetNumWeightValues() const { @@ -780,12 +792,12 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP // figure out the max value for each of them... u32 range = 256; while (--range > 0) { - IntegerEncodedValue val = CreateEncoding(range); + IntegerEncodedValue val = EncodingsValues[range]; u32 bitLength = val.GetBitLength(nValues); if (bitLength <= nBitsForColorData) { // Find the smallest possible range that matches the given encoding while (--range > 0) { - IntegerEncodedValue newval = CreateEncoding(range); + IntegerEncodedValue newval = EncodingsValues[range]; if (!newval.MatchesEncoding(val)) { break; } From 3377b78ea783ced2088370d698114b79b4ed1aa7 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Fri, 13 Mar 2020 23:52:51 -0300 Subject: [PATCH 09/13] astc: Call std::vector::reserve on texelWeightValues to avoid reallocating --- src/video_core/textures/astc.cpp | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index eb93b44bd1..5c65844ddd 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -1538,6 +1538,8 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 memset(texelWeightData + clearByteStart, 0, 16 - clearByteStart); std::vector texelWeightValues; + texelWeightValues.reserve(64); + InputBitStream weightStream(texelWeightData); DecodeIntegerSequence(texelWeightValues, weightStream, weightParams.m_MaxWeight, From 134f3ff9b4e0ca0e9a5d41d14d5f5a938762b5fd Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Sat, 14 Mar 2020 00:09:56 -0300 Subject: [PATCH 10/13] astc: Call std::vector:reserve on decodedClolorValues to avoid reallocating --- src/video_core/textures/astc.cpp | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 5c65844ddd..c15548d348 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -811,6 +811,8 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP // We now have enough to decode our s32eger sequence. std::vector decodedColorValues; + decodedColorValues.reserve(32); + InputBitStream colorStream(data); DecodeIntegerSequence(decodedColorValues, colorStream, range, nValues); From d71d7d917ea97c0e00602a714868f48445c8cd50 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Sat, 14 Mar 2020 00:13:58 -0300 Subject: [PATCH 11/13] astc: Pass val in Replicate by copy --- src/video_core/textures/astc.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index c15548d348..0286708005 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -628,7 +628,7 @@ static void FillError(u32* outBuf, u32 blockWidth, u32 blockHeight) { // Replicates low numBits such that [(toBit - 1):(toBit - 1 - fromBit)] // is the same as [(numBits - 1):0] and repeats all the way down. template -static IntType Replicate(const IntType& val, u32 numBits, u32 toBit) { +static IntType Replicate(IntType val, u32 numBits, u32 toBit) { if (numBits == 0) return 0; if (toBit == 0) From 9b8fb3c7561e60a4df6b61526f566da6b135da3d Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Sat, 14 Mar 2020 00:44:36 -0300 Subject: [PATCH 12/13] astc: Minor changes to InputBitStream --- src/video_core/textures/astc.cpp | 62 +++++++++++++++++--------------- 1 file changed, 34 insertions(+), 28 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 0286708005..5fce8546c1 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -39,18 +39,15 @@ constexpr u32 Popcnt(u32 n) { class InputBitStream { public: - explicit InputBitStream(const u8* ptr, s32 start_offset = 0) + explicit InputBitStream(const u8* ptr, std::size_t start_offset = 0) : m_CurByte(ptr), m_NextBit(start_offset % 8) {} - ~InputBitStream() = default; - - s32 GetBitsRead() const { + std::size_t GetBitsRead() const { return m_BitsRead; } - s32 ReadBit() { - - s32 bit = *m_CurByte >> m_NextBit++; + u32 ReadBit() { + u32 bit = *m_CurByte >> m_NextBit++; while (m_NextBit >= 8) { m_NextBit -= 8; m_CurByte++; @@ -60,9 +57,18 @@ public: return bit & 1; } - u32 ReadBits(u32 nBits) { + u32 ReadBits(std::size_t nBits) { u32 ret = 0; - for (u32 i = 0; i < nBits; i++) { + for (std::size_t i = 0; i < nBits; ++i) { + ret |= (ReadBit() & 1) << i; + } + return ret; + } + + template + u32 ReadBits() { + u32 ret = 0; + for (std::size_t i = 0; i < nBits; ++i) { ret |= (ReadBit() & 1) << i; } return ret; @@ -70,8 +76,8 @@ public: private: const u8* m_CurByte; - s32 m_NextBit = 0; - s32 m_BitsRead = 0; + std::size_t m_NextBit = 0; + std::size_t m_BitsRead = 0; }; class OutputBitStream { @@ -200,13 +206,13 @@ static void DecodeTritBlock(InputBitStream& bits, std::vector(); m[1] = bits.ReadBits(nBitsPerValue); - T |= bits.ReadBits(2) << 2; + T |= bits.ReadBits<2>() << 2; m[2] = bits.ReadBits(nBitsPerValue); T |= bits.ReadBit() << 4; m[3] = bits.ReadBits(nBitsPerValue); - T |= bits.ReadBits(2) << 5; + T |= bits.ReadBits<2>() << 5; m[4] = bits.ReadBits(nBitsPerValue); T |= bits.ReadBit() << 7; @@ -259,11 +265,11 @@ static void DecodeQus32Block(InputBitStream& bits, std::vector(); m[1] = bits.ReadBits(nBitsPerValue); - Q |= bits.ReadBits(2) << 3; + Q |= bits.ReadBits<2>() << 3; m[2] = bits.ReadBits(nBitsPerValue); - Q |= bits.ReadBits(2) << 5; + Q |= bits.ReadBits<2>() << 5; Bits Qb(Q); if (Qb(1, 2) == 3 && Qb(5, 6) == 0) { @@ -399,7 +405,7 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) { TexelWeightParams params; // Read the entire block mode all at once - u16 modeBits = static_cast(strm.ReadBits(11)); + u16 modeBits = static_cast(strm.ReadBits<11>()); // Does this match the void extent block mode? if ((modeBits & 0x01FF) == 0x1FC) { @@ -598,14 +604,14 @@ static void FillVoidExtentLDR(InputBitStream& strm, u32* const outBuf, u32 block u32 blockHeight) { // Don't actually care about the void extent, just read the bits... for (s32 i = 0; i < 4; ++i) { - strm.ReadBits(13); + strm.ReadBits<13>(); } // Decode the RGBA components and renormalize them to the range [0, 255] - u16 r = static_cast(strm.ReadBits(16)); - u16 g = static_cast(strm.ReadBits(16)); - u16 b = static_cast(strm.ReadBits(16)); - u16 a = static_cast(strm.ReadBits(16)); + u16 r = static_cast(strm.ReadBits<16>()); + u16 g = static_cast(strm.ReadBits<16>()); + u16 b = static_cast(strm.ReadBits<16>()); + u16 a = static_cast(strm.ReadBits<16>()); u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast(b) & 0xFF00) << 8 | (static_cast(a) & 0xFF00) << 16; @@ -1390,7 +1396,7 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 } // Read num partitions - u32 nPartitions = strm.ReadBits(2) + 1; + u32 nPartitions = strm.ReadBits<2>() + 1; assert(nPartitions <= 4); if (nPartitions == 4 && weightParams.m_bDualPlane) { @@ -1415,17 +1421,17 @@ static void DecompressBlock(const u8 inBuf[16], const u32 blockWidth, const u32 // Read extra config data... u32 baseCEM = 0; if (nPartitions == 1) { - colorEndpos32Mode[0] = strm.ReadBits(4); + colorEndpos32Mode[0] = strm.ReadBits<4>(); partitionIndex = 0; } else { - partitionIndex = strm.ReadBits(10); - baseCEM = strm.ReadBits(6); + partitionIndex = strm.ReadBits<10>(); + baseCEM = strm.ReadBits<6>(); } u32 baseMode = (baseCEM & 3); // Remaining bits are color endpos32 data... u32 nWeightBits = weightParams.GetPackedBitSize(); - s32 remainingBits = 128 - nWeightBits - strm.GetBitsRead(); + s32 remainingBits = 128 - nWeightBits - static_cast(strm.GetBitsRead()); // Consider extra bits prior to texel data... u32 extraCEMbits = 0; From 42cb8f11241568fe55de40e731e3fe93b39df2d4 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Sat, 14 Mar 2020 01:05:20 -0300 Subject: [PATCH 13/13] astc: Fix typos from search and replace --- src/video_core/textures/astc.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/video_core/textures/astc.cpp b/src/video_core/textures/astc.cpp index 5fce8546c1..404708d922 100644 --- a/src/video_core/textures/astc.cpp +++ b/src/video_core/textures/astc.cpp @@ -323,7 +323,7 @@ static constexpr IntegerEncodedValue CreateEncoding(u32 maxVal) { return IntegerEncodedValue(IntegerEncoding::Qus32, Popcnt(check / 5 - 1)); } - // Apparently it can't be represented with a bounded s32eger sequence... + // Apparently it can't be represented with a bounded integer sequence... // just iterate. maxVal--; } @@ -755,7 +755,7 @@ public: } // Take all of the components, transform them to their 8-bit variants, - // and then pack each channel s32o an R8G8B8A8 32-bit s32eger. We assume + // and then pack each channel into an R8G8B8A8 32-bit integer. We assume // that the architecture is little-endian, so the alpha channel will end // up in the most-significant byte. u32 Pack() const { @@ -815,7 +815,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP } } - // We now have enough to decode our s32eger sequence. + // We now have enough to decode our integer sequence. std::vector decodedColorValues; decodedColorValues.reserve(32);