1 // File: rg_etc1.cpp - Fast, high quality ETC1 block packer/unpacker - Rich Geldreich <richgel99@gmail.com> 2 // Please see ZLIB license at the end of rg_etc1.h. 3 // 4 // For more information Ericsson Texture Compression (ETC/ETC1), see: 5 // http://www.khronos.org/registry/gles/extensions/OES/OES_compressed_ETC1_RGB8_texture.txt 6 // 7 // v1.04 - 5/15/14 - Fix signed vs. unsigned subtraction problem (noticed when compiled with gcc) in pack_etc1_block_init(). 8 // This issue would cause an assert when this func. was called in debug. (Note this module was developed/testing with MSVC, 9 // I still need to test it throughly when compiled with gcc.) 10 // 11 // v1.03 - 5/12/13 - Initial public release 12 #include "rg_etc1.h" 13 14 #include <stdlib.h> 15 #include <memory.h> 16 #include <assert.h> 17 //#include <stdio.h> 18 #include <math.h> 19 20 #pragma warning (disable: 4201) // nonstandard extension used : nameless struct/union 21 22 #if defined(_DEBUG) || defined(DEBUG) 23 #define RG_ETC1_BUILD_DEBUG 24 #endif 25 26 #define RG_ETC1_ASSERT assert 27 28 namespace rg_etc1 29 { 30 typedef unsigned char uint8; 31 typedef unsigned short uint16; 32 typedef unsigned int uint; 33 typedef unsigned int uint32; 34 typedef long long int64; 35 typedef unsigned long long uint64; 36 37 const uint32 cUINT32_MAX = 0xFFFFFFFFU; 38 const uint64 cUINT64_MAX = 0xFFFFFFFFFFFFFFFFULL; //0xFFFFFFFFFFFFFFFFui64; 39 minimum(T a,T b)40 template<typename T> inline T minimum(T a, T b) { return (a < b) ? a : b; } minimum(T a,T b,T c)41 template<typename T> inline T minimum(T a, T b, T c) { return minimum(minimum(a, b), c); } maximum(T a,T b)42 template<typename T> inline T maximum(T a, T b) { return (a > b) ? a : b; } maximum(T a,T b,T c)43 template<typename T> inline T maximum(T a, T b, T c) { return maximum(maximum(a, b), c); } clamp(T value,T low,T high)44 template<typename T> inline T clamp(T value, T low, T high) { return (value < low) ? low : ((value > high) ? high : value); } square(T value)45 template<typename T> inline T square(T value) { return value * value; } zero_object(T & obj)46 template<typename T> inline void zero_object(T& obj) { memset((void*)&obj, 0, sizeof(obj)); } zero_this(T * pObj)47 template<typename T> inline void zero_this(T* pObj) { memset((void*)pObj, 0, sizeof(*pObj)); } 48 49 template<class T, size_t N> T decay_array_to_subtype(T (&a)[N]); 50 51 #define RG_ETC1_ARRAY_SIZE(X) (sizeof(X) / sizeof(decay_array_to_subtype(X))) 52 53 enum eNoClamp { cNoClamp }; 54 55 struct color_quad_u8 56 { clamprg_etc1::color_quad_u857 static inline int clamp(int v) { if (v & 0xFFFFFF00U) v = (~(static_cast<int>(v) >> 31)) & 0xFF; return v; } 58 59 struct component_traits { enum { cSigned = false, cFloat = false, cMin = 0U, cMax = 255U }; }; 60 61 public: 62 typedef unsigned char component_t; 63 typedef int parameter_t; 64 65 enum { cNumComps = 4 }; 66 67 union 68 { 69 struct 70 { 71 component_t r; 72 component_t g; 73 component_t b; 74 component_t a; 75 }; 76 77 component_t c[cNumComps]; 78 79 uint32 m_u32; 80 }; 81 color_quad_u8rg_etc1::color_quad_u882 inline color_quad_u8() 83 { 84 } 85 color_quad_u8rg_etc1::color_quad_u886 inline color_quad_u8(const color_quad_u8& other) : m_u32(other.m_u32) 87 { 88 } 89 color_quad_u8rg_etc1::color_quad_u890 explicit inline color_quad_u8(parameter_t y, parameter_t alpha = component_traits::cMax) 91 { 92 set(y, alpha); 93 } 94 color_quad_u8rg_etc1::color_quad_u895 inline color_quad_u8(parameter_t red, parameter_t green, parameter_t blue, parameter_t alpha = component_traits::cMax) 96 { 97 set(red, green, blue, alpha); 98 } 99 color_quad_u8rg_etc1::color_quad_u8100 explicit inline color_quad_u8(eNoClamp, parameter_t y, parameter_t alpha = component_traits::cMax) 101 { 102 set_noclamp_y_alpha(y, alpha); 103 } 104 color_quad_u8rg_etc1::color_quad_u8105 inline color_quad_u8(eNoClamp, parameter_t red, parameter_t green, parameter_t blue, parameter_t alpha = component_traits::cMax) 106 { 107 set_noclamp_rgba(red, green, blue, alpha); 108 } 109 clearrg_etc1::color_quad_u8110 inline void clear() 111 { 112 m_u32 = 0; 113 } 114 operator =rg_etc1::color_quad_u8115 inline color_quad_u8& operator= (const color_quad_u8& other) 116 { 117 m_u32 = other.m_u32; 118 return *this; 119 } 120 set_rgbrg_etc1::color_quad_u8121 inline color_quad_u8& set_rgb(const color_quad_u8& other) 122 { 123 r = other.r; 124 g = other.g; 125 b = other.b; 126 return *this; 127 } 128 operator =rg_etc1::color_quad_u8129 inline color_quad_u8& operator= (parameter_t y) 130 { 131 set(y, component_traits::cMax); 132 return *this; 133 } 134 setrg_etc1::color_quad_u8135 inline color_quad_u8& set(parameter_t y, parameter_t alpha = component_traits::cMax) 136 { 137 y = clamp(y); 138 alpha = clamp(alpha); 139 r = static_cast<component_t>(y); 140 g = static_cast<component_t>(y); 141 b = static_cast<component_t>(y); 142 a = static_cast<component_t>(alpha); 143 return *this; 144 } 145 set_noclamp_y_alpharg_etc1::color_quad_u8146 inline color_quad_u8& set_noclamp_y_alpha(parameter_t y, parameter_t alpha = component_traits::cMax) 147 { 148 RG_ETC1_ASSERT( (y >= component_traits::cMin) && (y <= component_traits::cMax) ); 149 RG_ETC1_ASSERT( (alpha >= component_traits::cMin) && (alpha <= component_traits::cMax) ); 150 151 r = static_cast<component_t>(y); 152 g = static_cast<component_t>(y); 153 b = static_cast<component_t>(y); 154 a = static_cast<component_t>(alpha); 155 return *this; 156 } 157 setrg_etc1::color_quad_u8158 inline color_quad_u8& set(parameter_t red, parameter_t green, parameter_t blue, parameter_t alpha = component_traits::cMax) 159 { 160 r = static_cast<component_t>(clamp(red)); 161 g = static_cast<component_t>(clamp(green)); 162 b = static_cast<component_t>(clamp(blue)); 163 a = static_cast<component_t>(clamp(alpha)); 164 return *this; 165 } 166 set_noclamp_rgbarg_etc1::color_quad_u8167 inline color_quad_u8& set_noclamp_rgba(parameter_t red, parameter_t green, parameter_t blue, parameter_t alpha) 168 { 169 RG_ETC1_ASSERT( (red >= component_traits::cMin) && (red <= component_traits::cMax) ); 170 RG_ETC1_ASSERT( (green >= component_traits::cMin) && (green <= component_traits::cMax) ); 171 RG_ETC1_ASSERT( (blue >= component_traits::cMin) && (blue <= component_traits::cMax) ); 172 RG_ETC1_ASSERT( (alpha >= component_traits::cMin) && (alpha <= component_traits::cMax) ); 173 174 r = static_cast<component_t>(red); 175 g = static_cast<component_t>(green); 176 b = static_cast<component_t>(blue); 177 a = static_cast<component_t>(alpha); 178 return *this; 179 } 180 set_noclamp_rgbrg_etc1::color_quad_u8181 inline color_quad_u8& set_noclamp_rgb(parameter_t red, parameter_t green, parameter_t blue) 182 { 183 RG_ETC1_ASSERT( (red >= component_traits::cMin) && (red <= component_traits::cMax) ); 184 RG_ETC1_ASSERT( (green >= component_traits::cMin) && (green <= component_traits::cMax) ); 185 RG_ETC1_ASSERT( (blue >= component_traits::cMin) && (blue <= component_traits::cMax) ); 186 187 r = static_cast<component_t>(red); 188 g = static_cast<component_t>(green); 189 b = static_cast<component_t>(blue); 190 return *this; 191 } 192 get_min_comprg_etc1::color_quad_u8193 static inline parameter_t get_min_comp() { return component_traits::cMin; } get_max_comprg_etc1::color_quad_u8194 static inline parameter_t get_max_comp() { return component_traits::cMax; } get_comps_are_signedrg_etc1::color_quad_u8195 static inline bool get_comps_are_signed() { return component_traits::cSigned; } 196 operator []rg_etc1::color_quad_u8197 inline component_t operator[] (uint i) const { RG_ETC1_ASSERT(i < cNumComps); return c[i]; } operator []rg_etc1::color_quad_u8198 inline component_t& operator[] (uint i) { RG_ETC1_ASSERT(i < cNumComps); return c[i]; } 199 set_componentrg_etc1::color_quad_u8200 inline color_quad_u8& set_component(uint i, parameter_t f) 201 { 202 RG_ETC1_ASSERT(i < cNumComps); 203 204 c[i] = static_cast<component_t>(clamp(f)); 205 206 return *this; 207 } 208 set_grayscalerg_etc1::color_quad_u8209 inline color_quad_u8& set_grayscale(parameter_t l) 210 { 211 component_t x = static_cast<component_t>(clamp(l)); 212 c[0] = x; 213 c[1] = x; 214 c[2] = x; 215 return *this; 216 } 217 clamprg_etc1::color_quad_u8218 inline color_quad_u8& clamp(const color_quad_u8& l, const color_quad_u8& h) 219 { 220 for (uint i = 0; i < cNumComps; i++) 221 c[i] = static_cast<component_t>(rg_etc1::clamp<parameter_t>(c[i], l[i], h[i])); 222 return *this; 223 } 224 clamprg_etc1::color_quad_u8225 inline color_quad_u8& clamp(parameter_t l, parameter_t h) 226 { 227 for (uint i = 0; i < cNumComps; i++) 228 c[i] = static_cast<component_t>(rg_etc1::clamp<parameter_t>(c[i], l, h)); 229 return *this; 230 } 231 232 // Returns CCIR 601 luma (consistent with color_utils::RGB_To_Y). get_lumarg_etc1::color_quad_u8233 inline parameter_t get_luma() const 234 { 235 return static_cast<parameter_t>((19595U * r + 38470U * g + 7471U * b + 32768U) >> 16U); 236 } 237 238 // Returns REC 709 luma. get_luma_rec709rg_etc1::color_quad_u8239 inline parameter_t get_luma_rec709() const 240 { 241 return static_cast<parameter_t>((13938U * r + 46869U * g + 4729U * b + 32768U) >> 16U); 242 } 243 squared_distance_rgbrg_etc1::color_quad_u8244 inline uint squared_distance_rgb(const color_quad_u8& c) const 245 { 246 return rg_etc1::square(r - c.r) + rg_etc1::square(g - c.g) + rg_etc1::square(b - c.b); 247 } 248 squared_distance_rgbarg_etc1::color_quad_u8249 inline uint squared_distance_rgba(const color_quad_u8& c) const 250 { 251 return rg_etc1::square(r - c.r) + rg_etc1::square(g - c.g) + rg_etc1::square(b - c.b) + rg_etc1::square(a - c.a); 252 } 253 rgb_equalsrg_etc1::color_quad_u8254 inline bool rgb_equals(const color_quad_u8& rhs) const 255 { 256 return (r == rhs.r) && (g == rhs.g) && (b == rhs.b); 257 } 258 operator ==rg_etc1::color_quad_u8259 inline bool operator== (const color_quad_u8& rhs) const 260 { 261 return m_u32 == rhs.m_u32; 262 } 263 operator +=rg_etc1::color_quad_u8264 color_quad_u8& operator+= (const color_quad_u8& other) 265 { 266 for (uint i = 0; i < 4; i++) 267 c[i] = static_cast<component_t>(clamp(c[i] + other.c[i])); 268 return *this; 269 } 270 operator -=rg_etc1::color_quad_u8271 color_quad_u8& operator-= (const color_quad_u8& other) 272 { 273 for (uint i = 0; i < 4; i++) 274 c[i] = static_cast<component_t>(clamp(c[i] - other.c[i])); 275 return *this; 276 } 277 operator +(const color_quad_u8 & lhs,const color_quad_u8 & rhs)278 friend color_quad_u8 operator+ (const color_quad_u8& lhs, const color_quad_u8& rhs) 279 { 280 color_quad_u8 result(lhs); 281 result += rhs; 282 return result; 283 } 284 operator -(const color_quad_u8 & lhs,const color_quad_u8 & rhs)285 friend color_quad_u8 operator- (const color_quad_u8& lhs, const color_quad_u8& rhs) 286 { 287 color_quad_u8 result(lhs); 288 result -= rhs; 289 return result; 290 } 291 }; // class color_quad_u8 292 293 struct vec3F 294 { 295 float m_s[3]; 296 vec3Frg_etc1::vec3F297 inline vec3F() { } vec3Frg_etc1::vec3F298 inline vec3F(float s) { m_s[0] = s; m_s[1] = s; m_s[2] = s; } vec3Frg_etc1::vec3F299 inline vec3F(float x, float y, float z) { m_s[0] = x; m_s[1] = y; m_s[2] = z; } 300 operator []rg_etc1::vec3F301 inline float operator[] (uint i) const { RG_ETC1_ASSERT(i < 3); return m_s[i]; } 302 operator +=rg_etc1::vec3F303 inline vec3F& operator += (const vec3F& other) { for (uint i = 0; i < 3; i++) m_s[i] += other.m_s[i]; return *this; } 304 operator *=rg_etc1::vec3F305 inline vec3F& operator *= (float s) { for (uint i = 0; i < 3; i++) m_s[i] *= s; return *this; } 306 }; 307 308 enum etc_constants 309 { 310 cETC1BytesPerBlock = 8U, 311 312 cETC1SelectorBits = 2U, 313 cETC1SelectorValues = 1U << cETC1SelectorBits, 314 cETC1SelectorMask = cETC1SelectorValues - 1U, 315 316 cETC1BlockShift = 2U, 317 cETC1BlockSize = 1U << cETC1BlockShift, 318 319 cETC1LSBSelectorIndicesBitOffset = 0, 320 cETC1MSBSelectorIndicesBitOffset = 16, 321 322 cETC1FlipBitOffset = 32, 323 cETC1DiffBitOffset = 33, 324 325 cETC1IntenModifierNumBits = 3, 326 cETC1IntenModifierValues = 1 << cETC1IntenModifierNumBits, 327 cETC1RightIntenModifierTableBitOffset = 34, 328 cETC1LeftIntenModifierTableBitOffset = 37, 329 330 // Base+Delta encoding (5 bit bases, 3 bit delta) 331 cETC1BaseColorCompNumBits = 5, 332 cETC1BaseColorCompMax = 1 << cETC1BaseColorCompNumBits, 333 334 cETC1DeltaColorCompNumBits = 3, 335 cETC1DeltaColorComp = 1 << cETC1DeltaColorCompNumBits, 336 cETC1DeltaColorCompMax = 1 << cETC1DeltaColorCompNumBits, 337 338 cETC1BaseColor5RBitOffset = 59, 339 cETC1BaseColor5GBitOffset = 51, 340 cETC1BaseColor5BBitOffset = 43, 341 342 cETC1DeltaColor3RBitOffset = 56, 343 cETC1DeltaColor3GBitOffset = 48, 344 cETC1DeltaColor3BBitOffset = 40, 345 346 // Absolute (non-delta) encoding (two 4-bit per component bases) 347 cETC1AbsColorCompNumBits = 4, 348 cETC1AbsColorCompMax = 1 << cETC1AbsColorCompNumBits, 349 350 cETC1AbsColor4R1BitOffset = 60, 351 cETC1AbsColor4G1BitOffset = 52, 352 cETC1AbsColor4B1BitOffset = 44, 353 354 cETC1AbsColor4R2BitOffset = 56, 355 cETC1AbsColor4G2BitOffset = 48, 356 cETC1AbsColor4B2BitOffset = 40, 357 358 cETC1ColorDeltaMin = -4, 359 cETC1ColorDeltaMax = 3, 360 361 // Delta3: 362 // 0 1 2 3 4 5 6 7 363 // 000 001 010 011 100 101 110 111 364 // 0 1 2 3 -4 -3 -2 -1 365 }; 366 367 static uint8 g_quant5_tab[256+16]; 368 369 static const int g_etc1_inten_tables[cETC1IntenModifierValues][cETC1SelectorValues] = 370 { 371 { -8, -2, 2, 8 }, { -17, -5, 5, 17 }, { -29, -9, 9, 29 }, { -42, -13, 13, 42 }, 372 { -60, -18, 18, 60 }, { -80, -24, 24, 80 }, { -106, -33, 33, 106 }, { -183, -47, 47, 183 } 373 }; 374 375 static const uint8 g_etc1_to_selector_index[cETC1SelectorValues] = { 2, 3, 1, 0 }; 376 static const uint8 g_selector_index_to_etc1[cETC1SelectorValues] = { 3, 2, 0, 1 }; 377 378 // Given an ETC1 diff/inten_table/selector, and an 8-bit desired color, this table encodes the best packed_color in the low byte, and the abs error in the high byte. 379 static uint16 g_etc1_inverse_lookup[2*8*4][256]; // [diff/inten_table/selector][desired_color] 380 381 // g_color8_to_etc_block_config[color][table_index] = Supplies for each 8-bit color value a list of packed ETC1 diff/intensity table/selectors/packed_colors that map to that color. 382 // To pack: diff | (inten << 1) | (selector << 4) | (packed_c << 8) 383 static const uint16 g_color8_to_etc_block_config_0_255[2][33] = 384 { 385 { 0x0000, 0x0010, 0x0002, 0x0012, 0x0004, 0x0014, 0x0006, 0x0016, 0x0008, 0x0018, 0x000A, 0x001A, 0x000C, 0x001C, 0x000E, 0x001E, 386 0x0001, 0x0011, 0x0003, 0x0013, 0x0005, 0x0015, 0x0007, 0x0017, 0x0009, 0x0019, 0x000B, 0x001B, 0x000D, 0x001D, 0x000F, 0x001F, 0xFFFF }, 387 { 0x0F20, 0x0F30, 0x0E32, 0x0F22, 0x0E34, 0x0F24, 0x0D36, 0x0F26, 0x0C38, 0x0E28, 0x0B3A, 0x0E2A, 0x093C, 0x0E2C, 0x053E, 0x0D2E, 388 0x1E31, 0x1F21, 0x1D33, 0x1F23, 0x1C35, 0x1E25, 0x1A37, 0x1E27, 0x1839, 0x1D29, 0x163B, 0x1C2B, 0x133D, 0x1B2D, 0x093F, 0x1A2F, 0xFFFF }, 389 }; 390 391 // Really only [254][11]. 392 static const uint16 g_color8_to_etc_block_config_1_to_254[254][12] = 393 { 394 { 0x021C, 0x0D0D, 0xFFFF }, { 0x0020, 0x0021, 0x0A0B, 0x061F, 0xFFFF }, { 0x0113, 0x0217, 0xFFFF }, { 0x0116, 0x031E, 395 0x0B0E, 0x0405, 0xFFFF }, { 0x0022, 0x0204, 0x050A, 0x0023, 0xFFFF }, { 0x0111, 0x0319, 0x0809, 0x170F, 0xFFFF }, { 396 0x0303, 0x0215, 0x0607, 0xFFFF }, { 0x0030, 0x0114, 0x0408, 0x0031, 0x0201, 0x051D, 0xFFFF }, { 0x0100, 0x0024, 0x0306, 397 0x0025, 0x041B, 0x0E0D, 0xFFFF }, { 0x021A, 0x0121, 0x0B0B, 0x071F, 0xFFFF }, { 0x0213, 0x0317, 0xFFFF }, { 0x0112, 398 0x0505, 0xFFFF }, { 0x0026, 0x070C, 0x0123, 0x0027, 0xFFFF }, { 0x0211, 0x0909, 0xFFFF }, { 0x0110, 0x0315, 0x0707, 399 0x0419, 0x180F, 0xFFFF }, { 0x0218, 0x0131, 0x0301, 0x0403, 0x061D, 0xFFFF }, { 0x0032, 0x0202, 0x0033, 0x0125, 0x051B, 400 0x0F0D, 0xFFFF }, { 0x0028, 0x031C, 0x0221, 0x0029, 0xFFFF }, { 0x0120, 0x0313, 0x0C0B, 0x081F, 0xFFFF }, { 0x0605, 401 0x0417, 0xFFFF }, { 0x0216, 0x041E, 0x0C0E, 0x0223, 0x0127, 0xFFFF }, { 0x0122, 0x0304, 0x060A, 0x0311, 0x0A09, 0xFFFF 402 }, { 0x0519, 0x190F, 0xFFFF }, { 0x002A, 0x0231, 0x0503, 0x0415, 0x0807, 0x002B, 0x071D, 0xFFFF }, { 0x0130, 0x0214, 403 0x0508, 0x0401, 0x0133, 0x0225, 0x061B, 0xFFFF }, { 0x0200, 0x0124, 0x0406, 0x0321, 0x0129, 0x100D, 0xFFFF }, { 0x031A, 404 0x0D0B, 0x091F, 0xFFFF }, { 0x0413, 0x0705, 0x0517, 0xFFFF }, { 0x0212, 0x0034, 0x0323, 0x0035, 0x0227, 0xFFFF }, { 405 0x0126, 0x080C, 0x0B09, 0xFFFF }, { 0x0411, 0x0619, 0x1A0F, 0xFFFF }, { 0x0210, 0x0331, 0x0603, 0x0515, 0x0907, 0x012B, 406 0xFFFF }, { 0x0318, 0x002C, 0x0501, 0x0233, 0x0325, 0x071B, 0x002D, 0x081D, 0xFFFF }, { 0x0132, 0x0302, 0x0229, 0x110D, 407 0xFFFF }, { 0x0128, 0x041C, 0x0421, 0x0E0B, 0x0A1F, 0xFFFF }, { 0x0220, 0x0513, 0x0617, 0xFFFF }, { 0x0135, 0x0805, 408 0x0327, 0xFFFF }, { 0x0316, 0x051E, 0x0D0E, 0x0423, 0xFFFF }, { 0x0222, 0x0404, 0x070A, 0x0511, 0x0719, 0x0C09, 0x1B0F, 409 0xFFFF }, { 0x0703, 0x0615, 0x0A07, 0x022B, 0xFFFF }, { 0x012A, 0x0431, 0x0601, 0x0333, 0x012D, 0x091D, 0xFFFF }, { 410 0x0230, 0x0314, 0x0036, 0x0608, 0x0425, 0x0037, 0x0329, 0x081B, 0x120D, 0xFFFF }, { 0x0300, 0x0224, 0x0506, 0x0521, 411 0x0F0B, 0x0B1F, 0xFFFF }, { 0x041A, 0x0613, 0x0717, 0xFFFF }, { 0x0235, 0x0905, 0xFFFF }, { 0x0312, 0x0134, 0x0523, 412 0x0427, 0xFFFF }, { 0x0226, 0x090C, 0x002E, 0x0611, 0x0D09, 0x002F, 0xFFFF }, { 0x0715, 0x0B07, 0x0819, 0x032B, 0x1C0F, 413 0xFFFF }, { 0x0310, 0x0531, 0x0701, 0x0803, 0x022D, 0x0A1D, 0xFFFF }, { 0x0418, 0x012C, 0x0433, 0x0525, 0x0137, 0x091B, 414 0x130D, 0xFFFF }, { 0x0232, 0x0402, 0x0621, 0x0429, 0xFFFF }, { 0x0228, 0x051C, 0x0713, 0x100B, 0x0C1F, 0xFFFF }, { 415 0x0320, 0x0335, 0x0A05, 0x0817, 0xFFFF }, { 0x0623, 0x0527, 0xFFFF }, { 0x0416, 0x061E, 0x0E0E, 0x0711, 0x0E09, 0x012F, 416 0xFFFF }, { 0x0322, 0x0504, 0x080A, 0x0919, 0x1D0F, 0xFFFF }, { 0x0631, 0x0903, 0x0815, 0x0C07, 0x042B, 0x032D, 0x0B1D, 417 0xFFFF }, { 0x022A, 0x0801, 0x0533, 0x0625, 0x0237, 0x0A1B, 0xFFFF }, { 0x0330, 0x0414, 0x0136, 0x0708, 0x0721, 0x0529, 418 0x140D, 0xFFFF }, { 0x0400, 0x0324, 0x0606, 0x0038, 0x0039, 0x110B, 0x0D1F, 0xFFFF }, { 0x051A, 0x0813, 0x0B05, 0x0917, 419 0xFFFF }, { 0x0723, 0x0435, 0x0627, 0xFFFF }, { 0x0412, 0x0234, 0x0F09, 0x022F, 0xFFFF }, { 0x0326, 0x0A0C, 0x012E, 420 0x0811, 0x0A19, 0x1E0F, 0xFFFF }, { 0x0731, 0x0A03, 0x0915, 0x0D07, 0x052B, 0xFFFF }, { 0x0410, 0x0901, 0x0633, 0x0725, 421 0x0337, 0x0B1B, 0x042D, 0x0C1D, 0xFFFF }, { 0x0518, 0x022C, 0x0629, 0x150D, 0xFFFF }, { 0x0332, 0x0502, 0x0821, 0x0139, 422 0x120B, 0x0E1F, 0xFFFF }, { 0x0328, 0x061C, 0x0913, 0x0A17, 0xFFFF }, { 0x0420, 0x0535, 0x0C05, 0x0727, 0xFFFF }, { 423 0x0823, 0x032F, 0xFFFF }, { 0x0516, 0x071E, 0x0F0E, 0x0911, 0x0B19, 0x1009, 0x1F0F, 0xFFFF }, { 0x0422, 0x0604, 0x090A, 424 0x0B03, 0x0A15, 0x0E07, 0x062B, 0xFFFF }, { 0x0831, 0x0A01, 0x0733, 0x052D, 0x0D1D, 0xFFFF }, { 0x032A, 0x0825, 0x0437, 425 0x0729, 0x0C1B, 0x160D, 0xFFFF }, { 0x0430, 0x0514, 0x0236, 0x0808, 0x0921, 0x0239, 0x130B, 0x0F1F, 0xFFFF }, { 0x0500, 426 0x0424, 0x0706, 0x0138, 0x0A13, 0x0B17, 0xFFFF }, { 0x061A, 0x0635, 0x0D05, 0xFFFF }, { 0x0923, 0x0827, 0xFFFF }, { 427 0x0512, 0x0334, 0x003A, 0x0A11, 0x1109, 0x003B, 0x042F, 0xFFFF }, { 0x0426, 0x0B0C, 0x022E, 0x0B15, 0x0F07, 0x0C19, 428 0x072B, 0xFFFF }, { 0x0931, 0x0B01, 0x0C03, 0x062D, 0x0E1D, 0xFFFF }, { 0x0510, 0x0833, 0x0925, 0x0537, 0x0D1B, 0x170D, 429 0xFFFF }, { 0x0618, 0x032C, 0x0A21, 0x0339, 0x0829, 0xFFFF }, { 0x0432, 0x0602, 0x0B13, 0x140B, 0x101F, 0xFFFF }, { 430 0x0428, 0x071C, 0x0735, 0x0E05, 0x0C17, 0xFFFF }, { 0x0520, 0x0A23, 0x0927, 0xFFFF }, { 0x0B11, 0x1209, 0x013B, 0x052F, 431 0xFFFF }, { 0x0616, 0x081E, 0x0D19, 0xFFFF }, { 0x0522, 0x0704, 0x0A0A, 0x0A31, 0x0D03, 0x0C15, 0x1007, 0x082B, 0x072D, 432 0x0F1D, 0xFFFF }, { 0x0C01, 0x0933, 0x0A25, 0x0637, 0x0E1B, 0xFFFF }, { 0x042A, 0x0B21, 0x0929, 0x180D, 0xFFFF }, { 433 0x0530, 0x0614, 0x0336, 0x0908, 0x0439, 0x150B, 0x111F, 0xFFFF }, { 0x0600, 0x0524, 0x0806, 0x0238, 0x0C13, 0x0F05, 434 0x0D17, 0xFFFF }, { 0x071A, 0x0B23, 0x0835, 0x0A27, 0xFFFF }, { 0x1309, 0x023B, 0x062F, 0xFFFF }, { 0x0612, 0x0434, 435 0x013A, 0x0C11, 0x0E19, 0xFFFF }, { 0x0526, 0x0C0C, 0x032E, 0x0B31, 0x0E03, 0x0D15, 0x1107, 0x092B, 0xFFFF }, { 0x0D01, 436 0x0A33, 0x0B25, 0x0737, 0x0F1B, 0x082D, 0x101D, 0xFFFF }, { 0x0610, 0x0A29, 0x190D, 0xFFFF }, { 0x0718, 0x042C, 0x0C21, 437 0x0539, 0x160B, 0x121F, 0xFFFF }, { 0x0532, 0x0702, 0x0D13, 0x0E17, 0xFFFF }, { 0x0528, 0x081C, 0x0935, 0x1005, 0x0B27, 438 0xFFFF }, { 0x0620, 0x0C23, 0x033B, 0x072F, 0xFFFF }, { 0x0D11, 0x0F19, 0x1409, 0xFFFF }, { 0x0716, 0x003C, 0x091E, 439 0x0F03, 0x0E15, 0x1207, 0x0A2B, 0x003D, 0xFFFF }, { 0x0622, 0x0804, 0x0B0A, 0x0C31, 0x0E01, 0x0B33, 0x092D, 0x111D, 440 0xFFFF }, { 0x0C25, 0x0837, 0x0B29, 0x101B, 0x1A0D, 0xFFFF }, { 0x052A, 0x0D21, 0x0639, 0x170B, 0x131F, 0xFFFF }, { 441 0x0630, 0x0714, 0x0436, 0x0A08, 0x0E13, 0x0F17, 0xFFFF }, { 0x0700, 0x0624, 0x0906, 0x0338, 0x0A35, 0x1105, 0xFFFF }, { 442 0x081A, 0x0D23, 0x0C27, 0xFFFF }, { 0x0E11, 0x1509, 0x043B, 0x082F, 0xFFFF }, { 0x0712, 0x0534, 0x023A, 0x0F15, 0x1307, 443 0x1019, 0x0B2B, 0x013D, 0xFFFF }, { 0x0626, 0x0D0C, 0x042E, 0x0D31, 0x0F01, 0x1003, 0x0A2D, 0x121D, 0xFFFF }, { 0x0C33, 444 0x0D25, 0x0937, 0x111B, 0x1B0D, 0xFFFF }, { 0x0710, 0x0E21, 0x0739, 0x0C29, 0xFFFF }, { 0x0818, 0x052C, 0x0F13, 0x180B, 445 0x141F, 0xFFFF }, { 0x0632, 0x0802, 0x0B35, 0x1205, 0x1017, 0xFFFF }, { 0x0628, 0x091C, 0x0E23, 0x0D27, 0xFFFF }, { 446 0x0720, 0x0F11, 0x1609, 0x053B, 0x092F, 0xFFFF }, { 0x1119, 0x023D, 0xFFFF }, { 0x0816, 0x013C, 0x0A1E, 0x0E31, 0x1103, 447 0x1015, 0x1407, 0x0C2B, 0x0B2D, 0x131D, 0xFFFF }, { 0x0722, 0x0904, 0x0C0A, 0x1001, 0x0D33, 0x0E25, 0x0A37, 0x121B, 448 0xFFFF }, { 0x0F21, 0x0D29, 0x1C0D, 0xFFFF }, { 0x062A, 0x0839, 0x190B, 0x151F, 0xFFFF }, { 0x0730, 0x0814, 0x0536, 449 0x0B08, 0x1013, 0x1305, 0x1117, 0xFFFF }, { 0x0800, 0x0724, 0x0A06, 0x0438, 0x0F23, 0x0C35, 0x0E27, 0xFFFF }, { 0x091A, 450 0x1709, 0x063B, 0x0A2F, 0xFFFF }, { 0x1011, 0x1219, 0x033D, 0xFFFF }, { 0x0812, 0x0634, 0x033A, 0x0F31, 0x1203, 0x1115, 451 0x1507, 0x0D2B, 0xFFFF }, { 0x0726, 0x0E0C, 0x052E, 0x1101, 0x0E33, 0x0F25, 0x0B37, 0x131B, 0x0C2D, 0x141D, 0xFFFF }, { 452 0x0E29, 0x1D0D, 0xFFFF }, { 0x0810, 0x1021, 0x0939, 0x1A0B, 0x161F, 0xFFFF }, { 0x0918, 0x062C, 0x1113, 0x1217, 0xFFFF 453 }, { 0x0732, 0x0902, 0x0D35, 0x1405, 0x0F27, 0xFFFF }, { 0x0728, 0x0A1C, 0x1023, 0x073B, 0x0B2F, 0xFFFF }, { 0x0820, 454 0x1111, 0x1319, 0x1809, 0xFFFF }, { 0x1303, 0x1215, 0x1607, 0x0E2B, 0x043D, 0xFFFF }, { 0x0916, 0x023C, 0x0B1E, 0x1031, 455 0x1201, 0x0F33, 0x0D2D, 0x151D, 0xFFFF }, { 0x0822, 0x0A04, 0x0D0A, 0x1025, 0x0C37, 0x0F29, 0x141B, 0x1E0D, 0xFFFF }, { 456 0x1121, 0x0A39, 0x1B0B, 0x171F, 0xFFFF }, { 0x072A, 0x1213, 0x1317, 0xFFFF }, { 0x0830, 0x0914, 0x0636, 0x0C08, 0x0E35, 457 0x1505, 0xFFFF }, { 0x0900, 0x0824, 0x0B06, 0x0538, 0x1123, 0x1027, 0xFFFF }, { 0x0A1A, 0x1211, 0x1909, 0x083B, 0x0C2F, 458 0xFFFF }, { 0x1315, 0x1707, 0x1419, 0x0F2B, 0x053D, 0xFFFF }, { 0x0912, 0x0734, 0x043A, 0x1131, 0x1301, 0x1403, 0x0E2D, 459 0x161D, 0xFFFF }, { 0x0826, 0x0F0C, 0x062E, 0x1033, 0x1125, 0x0D37, 0x151B, 0x1F0D, 0xFFFF }, { 0x1221, 0x0B39, 0x1029, 460 0xFFFF }, { 0x0910, 0x1313, 0x1C0B, 0x181F, 0xFFFF }, { 0x0A18, 0x072C, 0x0F35, 0x1605, 0x1417, 0xFFFF }, { 0x0832, 461 0x0A02, 0x1223, 0x1127, 0xFFFF }, { 0x0828, 0x0B1C, 0x1311, 0x1A09, 0x093B, 0x0D2F, 0xFFFF }, { 0x0920, 0x1519, 0x063D, 462 0xFFFF }, { 0x1231, 0x1503, 0x1415, 0x1807, 0x102B, 0x0F2D, 0x171D, 0xFFFF }, { 0x0A16, 0x033C, 0x0C1E, 0x1401, 0x1133, 463 0x1225, 0x0E37, 0x161B, 0xFFFF }, { 0x0922, 0x0B04, 0x0E0A, 0x1321, 0x1129, 0xFFFF }, { 0x0C39, 0x1D0B, 0x191F, 0xFFFF 464 }, { 0x082A, 0x1413, 0x1705, 0x1517, 0xFFFF }, { 0x0930, 0x0A14, 0x0736, 0x0D08, 0x1323, 0x1035, 0x1227, 0xFFFF }, { 465 0x0A00, 0x0924, 0x0C06, 0x0638, 0x1B09, 0x0A3B, 0x0E2F, 0xFFFF }, { 0x0B1A, 0x1411, 0x1619, 0x073D, 0xFFFF }, { 0x1331, 466 0x1603, 0x1515, 0x1907, 0x112B, 0xFFFF }, { 0x0A12, 0x0834, 0x053A, 0x1501, 0x1233, 0x1325, 0x0F37, 0x171B, 0x102D, 467 0x181D, 0xFFFF }, { 0x0926, 0x072E, 0x1229, 0xFFFF }, { 0x1421, 0x0D39, 0x1E0B, 0x1A1F, 0xFFFF }, { 0x0A10, 0x1513, 468 0x1617, 0xFFFF }, { 0x0B18, 0x082C, 0x1135, 0x1805, 0x1327, 0xFFFF }, { 0x0932, 0x0B02, 0x1423, 0x0B3B, 0x0F2F, 0xFFFF 469 }, { 0x0928, 0x0C1C, 0x1511, 0x1719, 0x1C09, 0xFFFF }, { 0x0A20, 0x1703, 0x1615, 0x1A07, 0x122B, 0x083D, 0xFFFF }, { 470 0x1431, 0x1601, 0x1333, 0x112D, 0x191D, 0xFFFF }, { 0x0B16, 0x043C, 0x0D1E, 0x1425, 0x1037, 0x1329, 0x181B, 0xFFFF }, { 471 0x0A22, 0x0C04, 0x0F0A, 0x1521, 0x0E39, 0x1F0B, 0x1B1F, 0xFFFF }, { 0x1613, 0x1717, 0xFFFF }, { 0x092A, 0x1235, 0x1905, 472 0xFFFF }, { 0x0A30, 0x0B14, 0x0836, 0x0E08, 0x1523, 0x1427, 0xFFFF }, { 0x0B00, 0x0A24, 0x0D06, 0x0738, 0x1611, 0x1D09, 473 0x0C3B, 0x102F, 0xFFFF }, { 0x0C1A, 0x1715, 0x1B07, 0x1819, 0x132B, 0x093D, 0xFFFF }, { 0x1531, 0x1701, 0x1803, 0x122D, 474 0x1A1D, 0xFFFF }, { 0x0B12, 0x0934, 0x063A, 0x1433, 0x1525, 0x1137, 0x191B, 0xFFFF }, { 0x0A26, 0x003E, 0x082E, 0x1621, 475 0x0F39, 0x1429, 0x003F, 0xFFFF }, { 0x1713, 0x1C1F, 0xFFFF }, { 0x0B10, 0x1335, 0x1A05, 0x1817, 0xFFFF }, { 0x0C18, 476 0x092C, 0x1623, 0x1527, 0xFFFF }, { 0x0A32, 0x0C02, 0x1711, 0x1E09, 0x0D3B, 0x112F, 0xFFFF }, { 0x0A28, 0x0D1C, 0x1919, 477 0x0A3D, 0xFFFF }, { 0x0B20, 0x1631, 0x1903, 0x1815, 0x1C07, 0x142B, 0x132D, 0x1B1D, 0xFFFF }, { 0x1801, 0x1533, 0x1625, 478 0x1237, 0x1A1B, 0xFFFF }, { 0x0C16, 0x053C, 0x0E1E, 0x1721, 0x1529, 0x013F, 0xFFFF }, { 0x0B22, 0x0D04, 0x1039, 0x1D1F, 479 0xFFFF }, { 0x1813, 0x1B05, 0x1917, 0xFFFF }, { 0x0A2A, 0x1723, 0x1435, 0x1627, 0xFFFF }, { 0x0B30, 0x0C14, 0x0936, 480 0x0F08, 0x1F09, 0x0E3B, 0x122F, 0xFFFF }, { 0x0C00, 0x0B24, 0x0E06, 0x0838, 0x1811, 0x1A19, 0x0B3D, 0xFFFF }, { 0x0D1A, 481 0x1731, 0x1A03, 0x1915, 0x1D07, 0x152B, 0xFFFF }, { 0x1901, 0x1633, 0x1725, 0x1337, 0x1B1B, 0x142D, 0x1C1D, 0xFFFF }, { 482 0x0C12, 0x0A34, 0x073A, 0x1629, 0x023F, 0xFFFF }, { 0x0B26, 0x013E, 0x092E, 0x1821, 0x1139, 0x1E1F, 0xFFFF }, { 0x1913, 483 0x1A17, 0xFFFF }, { 0x0C10, 0x1535, 0x1C05, 0x1727, 0xFFFF }, { 0x0D18, 0x0A2C, 0x1823, 0x0F3B, 0x132F, 0xFFFF }, { 484 0x0B32, 0x0D02, 0x1911, 0x1B19, 0xFFFF }, { 0x0B28, 0x0E1C, 0x1B03, 0x1A15, 0x1E07, 0x162B, 0x0C3D, 0xFFFF }, { 0x0C20, 485 0x1831, 0x1A01, 0x1733, 0x152D, 0x1D1D, 0xFFFF }, { 0x1825, 0x1437, 0x1729, 0x1C1B, 0x033F, 0xFFFF }, { 0x0D16, 0x063C, 486 0x0F1E, 0x1921, 0x1239, 0x1F1F, 0xFFFF }, { 0x0C22, 0x0E04, 0x1A13, 0x1B17, 0xFFFF }, { 0x1635, 0x1D05, 0xFFFF }, { 487 0x0B2A, 0x1923, 0x1827, 0xFFFF }, { 0x0C30, 0x0D14, 0x0A36, 0x1A11, 0x103B, 0x142F, 0xFFFF }, { 0x0D00, 0x0C24, 0x0F06, 488 0x0938, 0x1B15, 0x1F07, 0x1C19, 0x172B, 0x0D3D, 0xFFFF }, { 0x0E1A, 0x1931, 0x1B01, 0x1C03, 0x162D, 0x1E1D, 0xFFFF }, { 489 0x1833, 0x1925, 0x1537, 0x1D1B, 0xFFFF }, { 0x0D12, 0x0B34, 0x083A, 0x1A21, 0x1339, 0x1829, 0x043F, 0xFFFF }, { 0x0C26, 490 0x023E, 0x0A2E, 0x1B13, 0xFFFF }, { 0x1735, 0x1E05, 0x1C17, 0xFFFF }, { 0x0D10, 0x1A23, 0x1927, 0xFFFF }, { 0x0E18, 491 0x0B2C, 0x1B11, 0x113B, 0x152F, 0xFFFF }, { 0x0C32, 0x0E02, 0x1D19, 0x0E3D, 0xFFFF }, { 0x0C28, 0x0F1C, 0x1A31, 0x1D03, 492 0x1C15, 0x182B, 0x172D, 0x1F1D, 0xFFFF }, { 0x0D20, 0x1C01, 0x1933, 0x1A25, 0x1637, 0x1E1B, 0xFFFF }, { 0x1B21, 0x1929, 493 0x053F, 0xFFFF }, { 0x0E16, 0x073C, 0x1439, 0xFFFF }, { 0x0D22, 0x0F04, 0x1C13, 0x1F05, 0x1D17, 0xFFFF }, { 0x1B23, 494 0x1835, 0x1A27, 0xFFFF }, { 0x0C2A, 0x123B, 0x162F, 0xFFFF }, { 0x0D30, 0x0E14, 0x0B36, 0x1C11, 0x1E19, 0x0F3D, 0xFFFF 495 }, { 0x0E00, 0x0D24, 0x0A38, 0x1B31, 0x1E03, 0x1D15, 0x192B, 0xFFFF }, { 0x0F1A, 0x1D01, 0x1A33, 0x1B25, 0x1737, 0x1F1B, 496 0x182D, 0xFFFF }, { 0x1A29, 0x063F, 0xFFFF }, { 0x0E12, 0x0C34, 0x093A, 0x1C21, 0x1539, 0xFFFF }, { 0x0D26, 0x033E, 497 0x0B2E, 0x1D13, 0x1E17, 0xFFFF }, { 0x1935, 0x1B27, 0xFFFF }, { 0x0E10, 0x1C23, 0x133B, 0x172F, 0xFFFF }, { 0x0F18, 498 0x0C2C, 0x1D11, 0x1F19, 0xFFFF }, { 0x0D32, 0x0F02, 0x1F03, 0x1E15, 0x1A2B, 0x103D, 0xFFFF }, { 0x0D28, 0x1C31, 0x1E01, 499 0x1B33, 0x192D, 0xFFFF }, { 0x0E20, 0x1C25, 0x1837, 0x1B29, 0x073F, 0xFFFF }, { 0x1D21, 0x1639, 0xFFFF }, { 0x0F16, 500 0x083C, 0x1E13, 0x1F17, 0xFFFF }, { 0x0E22, 0x1A35, 0xFFFF }, { 0x1D23, 0x1C27, 0xFFFF }, { 0x0D2A, 0x1E11, 0x143B, 501 0x182F, 0xFFFF }, { 0x0E30, 0x0F14, 0x0C36, 0x1F15, 0x1B2B, 0x113D, 0xFFFF }, { 0x0F00, 0x0E24, 0x0B38, 0x1D31, 0x1F01, 502 0x1A2D, 0xFFFF }, { 0x1C33, 0x1D25, 0x1937, 0xFFFF }, { 0x1E21, 0x1739, 0x1C29, 0x083F, 0xFFFF }, { 0x0F12, 0x0D34, 503 0x0A3A, 0x1F13, 0xFFFF }, { 0x0E26, 0x043E, 0x0C2E, 0x1B35, 0xFFFF }, { 0x1E23, 0x1D27, 0xFFFF }, { 0x0F10, 0x1F11, 504 0x153B, 0x192F, 0xFFFF }, { 0x0D2C, 0x123D, 0xFFFF }, 505 }; 506 507 struct etc1_block 508 { 509 // big endian uint64: 510 // bit ofs: 56 48 40 32 24 16 8 0 511 // byte ofs: b0, b1, b2, b3, b4, b5, b6, b7 512 union 513 { 514 uint64 m_uint64; 515 uint8 m_bytes[8]; 516 }; 517 518 uint8 m_low_color[2]; 519 uint8 m_high_color[2]; 520 521 enum { cNumSelectorBytes = 4 }; 522 uint8 m_selectors[cNumSelectorBytes]; 523 clearrg_etc1::etc1_block524 inline void clear() 525 { 526 zero_this(this); 527 } 528 get_byte_bitsrg_etc1::etc1_block529 inline uint get_byte_bits(uint ofs, uint num) const 530 { 531 RG_ETC1_ASSERT((ofs + num) <= 64U); 532 RG_ETC1_ASSERT(num && (num <= 8U)); 533 RG_ETC1_ASSERT((ofs >> 3) == ((ofs + num - 1) >> 3)); 534 const uint byte_ofs = 7 - (ofs >> 3); 535 const uint byte_bit_ofs = ofs & 7; 536 return (m_bytes[byte_ofs] >> byte_bit_ofs) & ((1 << num) - 1); 537 } 538 set_byte_bitsrg_etc1::etc1_block539 inline void set_byte_bits(uint ofs, uint num, uint bits) 540 { 541 RG_ETC1_ASSERT((ofs + num) <= 64U); 542 RG_ETC1_ASSERT(num && (num < 32U)); 543 RG_ETC1_ASSERT((ofs >> 3) == ((ofs + num - 1) >> 3)); 544 RG_ETC1_ASSERT(bits < (1U << num)); 545 const uint byte_ofs = 7 - (ofs >> 3); 546 const uint byte_bit_ofs = ofs & 7; 547 const uint mask = (1 << num) - 1; 548 m_bytes[byte_ofs] &= ~(mask << byte_bit_ofs); 549 m_bytes[byte_ofs] |= (bits << byte_bit_ofs); 550 } 551 552 // false = left/right subblocks 553 // true = upper/lower subblocks get_flip_bitrg_etc1::etc1_block554 inline bool get_flip_bit() const 555 { 556 return (m_bytes[3] & 1) != 0; 557 } 558 set_flip_bitrg_etc1::etc1_block559 inline void set_flip_bit(bool flip) 560 { 561 m_bytes[3] &= ~1; 562 m_bytes[3] |= static_cast<uint8>(flip); 563 } 564 get_diff_bitrg_etc1::etc1_block565 inline bool get_diff_bit() const 566 { 567 return (m_bytes[3] & 2) != 0; 568 } 569 set_diff_bitrg_etc1::etc1_block570 inline void set_diff_bit(bool diff) 571 { 572 m_bytes[3] &= ~2; 573 m_bytes[3] |= (static_cast<uint>(diff) << 1); 574 } 575 576 // Returns intensity modifier table (0-7) used by subblock subblock_id. 577 // subblock_id=0 left/top (CW 1), 1=right/bottom (CW 2) get_inten_tablerg_etc1::etc1_block578 inline uint get_inten_table(uint subblock_id) const 579 { 580 RG_ETC1_ASSERT(subblock_id < 2); 581 const uint ofs = subblock_id ? 2 : 5; 582 return (m_bytes[3] >> ofs) & 7; 583 } 584 585 // Sets intensity modifier table (0-7) used by subblock subblock_id (0 or 1) set_inten_tablerg_etc1::etc1_block586 inline void set_inten_table(uint subblock_id, uint t) 587 { 588 RG_ETC1_ASSERT(subblock_id < 2); 589 RG_ETC1_ASSERT(t < 8); 590 const uint ofs = subblock_id ? 2 : 5; 591 m_bytes[3] &= ~(7 << ofs); 592 m_bytes[3] |= (t << ofs); 593 } 594 595 // Returned selector value ranges from 0-3 and is a direct index into g_etc1_inten_tables. get_selectorrg_etc1::etc1_block596 inline uint get_selector(uint x, uint y) const 597 { 598 RG_ETC1_ASSERT((x | y) < 4); 599 600 const uint bit_index = x * 4 + y; 601 const uint byte_bit_ofs = bit_index & 7; 602 const uint8 *p = &m_bytes[7 - (bit_index >> 3)]; 603 const uint lsb = (p[0] >> byte_bit_ofs) & 1; 604 const uint msb = (p[-2] >> byte_bit_ofs) & 1; 605 const uint val = lsb | (msb << 1); 606 607 return g_etc1_to_selector_index[val]; 608 } 609 610 // Selector "val" ranges from 0-3 and is a direct index into g_etc1_inten_tables. set_selectorrg_etc1::etc1_block611 inline void set_selector(uint x, uint y, uint val) 612 { 613 RG_ETC1_ASSERT((x | y | val) < 4); 614 const uint bit_index = x * 4 + y; 615 616 uint8 *p = &m_bytes[7 - (bit_index >> 3)]; 617 618 const uint byte_bit_ofs = bit_index & 7; 619 const uint mask = 1 << byte_bit_ofs; 620 621 const uint etc1_val = g_selector_index_to_etc1[val]; 622 623 const uint lsb = etc1_val & 1; 624 const uint msb = etc1_val >> 1; 625 626 p[0] &= ~mask; 627 p[0] |= (lsb << byte_bit_ofs); 628 629 p[-2] &= ~mask; 630 p[-2] |= (msb << byte_bit_ofs); 631 } 632 set_base4_colorrg_etc1::etc1_block633 inline void set_base4_color(uint idx, uint16 c) 634 { 635 if (idx) 636 { 637 set_byte_bits(cETC1AbsColor4R2BitOffset, 4, (c >> 8) & 15); 638 set_byte_bits(cETC1AbsColor4G2BitOffset, 4, (c >> 4) & 15); 639 set_byte_bits(cETC1AbsColor4B2BitOffset, 4, c & 15); 640 } 641 else 642 { 643 set_byte_bits(cETC1AbsColor4R1BitOffset, 4, (c >> 8) & 15); 644 set_byte_bits(cETC1AbsColor4G1BitOffset, 4, (c >> 4) & 15); 645 set_byte_bits(cETC1AbsColor4B1BitOffset, 4, c & 15); 646 } 647 } 648 get_base4_colorrg_etc1::etc1_block649 inline uint16 get_base4_color(uint idx) const 650 { 651 uint r, g, b; 652 if (idx) 653 { 654 r = get_byte_bits(cETC1AbsColor4R2BitOffset, 4); 655 g = get_byte_bits(cETC1AbsColor4G2BitOffset, 4); 656 b = get_byte_bits(cETC1AbsColor4B2BitOffset, 4); 657 } 658 else 659 { 660 r = get_byte_bits(cETC1AbsColor4R1BitOffset, 4); 661 g = get_byte_bits(cETC1AbsColor4G1BitOffset, 4); 662 b = get_byte_bits(cETC1AbsColor4B1BitOffset, 4); 663 } 664 return static_cast<uint16>(b | (g << 4U) | (r << 8U)); 665 } 666 set_base5_colorrg_etc1::etc1_block667 inline void set_base5_color(uint16 c) 668 { 669 set_byte_bits(cETC1BaseColor5RBitOffset, 5, (c >> 10) & 31); 670 set_byte_bits(cETC1BaseColor5GBitOffset, 5, (c >> 5) & 31); 671 set_byte_bits(cETC1BaseColor5BBitOffset, 5, c & 31); 672 } 673 get_base5_colorrg_etc1::etc1_block674 inline uint16 get_base5_color() const 675 { 676 const uint r = get_byte_bits(cETC1BaseColor5RBitOffset, 5); 677 const uint g = get_byte_bits(cETC1BaseColor5GBitOffset, 5); 678 const uint b = get_byte_bits(cETC1BaseColor5BBitOffset, 5); 679 return static_cast<uint16>(b | (g << 5U) | (r << 10U)); 680 } 681 set_delta3_colorrg_etc1::etc1_block682 void set_delta3_color(uint16 c) 683 { 684 set_byte_bits(cETC1DeltaColor3RBitOffset, 3, (c >> 6) & 7); 685 set_byte_bits(cETC1DeltaColor3GBitOffset, 3, (c >> 3) & 7); 686 set_byte_bits(cETC1DeltaColor3BBitOffset, 3, c & 7); 687 } 688 get_delta3_colorrg_etc1::etc1_block689 inline uint16 get_delta3_color() const 690 { 691 const uint r = get_byte_bits(cETC1DeltaColor3RBitOffset, 3); 692 const uint g = get_byte_bits(cETC1DeltaColor3GBitOffset, 3); 693 const uint b = get_byte_bits(cETC1DeltaColor3BBitOffset, 3); 694 return static_cast<uint16>(b | (g << 3U) | (r << 6U)); 695 } 696 697 // Base color 5 698 static uint16 pack_color5(const color_quad_u8& color, bool scaled, uint bias = 127U); 699 static uint16 pack_color5(uint r, uint g, uint b, bool scaled, uint bias = 127U); 700 701 static color_quad_u8 unpack_color5(uint16 packed_color5, bool scaled, uint alpha = 255U); 702 static void unpack_color5(uint& r, uint& g, uint& b, uint16 packed_color, bool scaled); 703 704 static bool unpack_color5(color_quad_u8& result, uint16 packed_color5, uint16 packed_delta3, bool scaled, uint alpha = 255U); 705 static bool unpack_color5(uint& r, uint& g, uint& b, uint16 packed_color5, uint16 packed_delta3, bool scaled, uint alpha = 255U); 706 707 // Delta color 3 708 // Inputs range from -4 to 3 (cETC1ColorDeltaMin to cETC1ColorDeltaMax) 709 static uint16 pack_delta3(int r, int g, int b); 710 711 // Results range from -4 to 3 (cETC1ColorDeltaMin to cETC1ColorDeltaMax) 712 static void unpack_delta3(int& r, int& g, int& b, uint16 packed_delta3); 713 714 // Abs color 4 715 static uint16 pack_color4(const color_quad_u8& color, bool scaled, uint bias = 127U); 716 static uint16 pack_color4(uint r, uint g, uint b, bool scaled, uint bias = 127U); 717 718 static color_quad_u8 unpack_color4(uint16 packed_color4, bool scaled, uint alpha = 255U); 719 static void unpack_color4(uint& r, uint& g, uint& b, uint16 packed_color4, bool scaled); 720 721 // subblock colors 722 static void get_diff_subblock_colors(color_quad_u8* pDst, uint16 packed_color5, uint table_idx); 723 static bool get_diff_subblock_colors(color_quad_u8* pDst, uint16 packed_color5, uint16 packed_delta3, uint table_idx); 724 static void get_abs_subblock_colors(color_quad_u8* pDst, uint16 packed_color4, uint table_idx); 725 unscaled_to_scaled_colorrg_etc1::etc1_block726 static inline void unscaled_to_scaled_color(color_quad_u8& dst, const color_quad_u8& src, bool color4) 727 { 728 if (color4) 729 { 730 dst.r = src.r | (src.r << 4); 731 dst.g = src.g | (src.g << 4); 732 dst.b = src.b | (src.b << 4); 733 } 734 else 735 { 736 dst.r = (src.r >> 2) | (src.r << 3); 737 dst.g = (src.g >> 2) | (src.g << 3); 738 dst.b = (src.b >> 2) | (src.b << 3); 739 } 740 dst.a = src.a; 741 } 742 }; 743 744 // Returns pointer to sorted array. 745 template<typename T, typename Q> 746 T* indirect_radix_sort(uint num_indices, T* pIndices0, T* pIndices1, const Q* pKeys, uint key_ofs, uint key_size, bool init_indices) 747 { 748 RG_ETC1_ASSERT((key_ofs >= 0) && (key_ofs < sizeof(T))); 749 RG_ETC1_ASSERT((key_size >= 1) && (key_size <= 4)); 750 751 if (init_indices) 752 { 753 T* p = pIndices0; 754 T* q = pIndices0 + (num_indices >> 1) * 2; 755 uint i; 756 for (i = 0; p != q; p += 2, i += 2) 757 { 758 p[0] = static_cast<T>(i); 759 p[1] = static_cast<T>(i + 1); 760 } 761 762 if (num_indices & 1) 763 *p = static_cast<T>(i); 764 } 765 766 uint hist[256 * 4]; 767 768 memset(hist, 0, sizeof(hist[0]) * 256 * key_size); 769 770 #define RG_ETC1_GET_KEY(p) (*(const uint*)((const uint8*)(pKeys + *(p)) + key_ofs)) 771 #define RG_ETC1_GET_KEY_FROM_INDEX(i) (*(const uint*)((const uint8*)(pKeys + (i)) + key_ofs)) 772 773 if (key_size == 4) 774 { 775 T* p = pIndices0; 776 T* q = pIndices0 + num_indices; 777 for ( ; p != q; p++) 778 { 779 const uint key = RG_ETC1_GET_KEY(p); 780 781 hist[ key & 0xFF]++; 782 hist[256 + ((key >> 8) & 0xFF)]++; 783 hist[512 + ((key >> 16) & 0xFF)]++; 784 hist[768 + ((key >> 24) & 0xFF)]++; 785 } 786 } 787 else if (key_size == 3) 788 { 789 T* p = pIndices0; 790 T* q = pIndices0 + num_indices; 791 for ( ; p != q; p++) 792 { 793 const uint key = RG_ETC1_GET_KEY(p); 794 795 hist[ key & 0xFF]++; 796 hist[256 + ((key >> 8) & 0xFF)]++; 797 hist[512 + ((key >> 16) & 0xFF)]++; 798 } 799 } 800 else if (key_size == 2) 801 { 802 T* p = pIndices0; 803 T* q = pIndices0 + (num_indices >> 1) * 2; 804 805 for ( ; p != q; p += 2) 806 { 807 const uint key0 = RG_ETC1_GET_KEY(p); 808 const uint key1 = RG_ETC1_GET_KEY(p+1); 809 810 hist[ key0 & 0xFF]++; 811 hist[256 + ((key0 >> 8) & 0xFF)]++; 812 813 hist[ key1 & 0xFF]++; 814 hist[256 + ((key1 >> 8) & 0xFF)]++; 815 } 816 817 if (num_indices & 1) 818 { 819 const uint key = RG_ETC1_GET_KEY(p); 820 821 hist[ key & 0xFF]++; 822 hist[256 + ((key >> 8) & 0xFF)]++; 823 } 824 } 825 else 826 { 827 RG_ETC1_ASSERT(key_size == 1); 828 if (key_size != 1) 829 return NULL; 830 831 T* p = pIndices0; 832 T* q = pIndices0 + (num_indices >> 1) * 2; 833 834 for ( ; p != q; p += 2) 835 { 836 const uint key0 = RG_ETC1_GET_KEY(p); 837 const uint key1 = RG_ETC1_GET_KEY(p+1); 838 839 hist[key0 & 0xFF]++; 840 hist[key1 & 0xFF]++; 841 } 842 843 if (num_indices & 1) 844 { 845 const uint key = RG_ETC1_GET_KEY(p); 846 847 hist[key & 0xFF]++; 848 } 849 } 850 851 T* pCur = pIndices0; 852 T* pNew = pIndices1; 853 854 for (uint pass = 0; pass < key_size; pass++) 855 { 856 const uint* pHist = &hist[pass << 8]; 857 858 uint offsets[256]; 859 860 uint cur_ofs = 0; 861 for (uint i = 0; i < 256; i += 2) 862 { 863 offsets[i] = cur_ofs; 864 cur_ofs += pHist[i]; 865 866 offsets[i+1] = cur_ofs; 867 cur_ofs += pHist[i+1]; 868 } 869 870 const uint pass_shift = pass << 3; 871 872 T* p = pCur; 873 T* q = pCur + (num_indices >> 1) * 2; 874 875 for ( ; p != q; p += 2) 876 { 877 uint index0 = p[0]; 878 uint index1 = p[1]; 879 880 uint c0 = (RG_ETC1_GET_KEY_FROM_INDEX(index0) >> pass_shift) & 0xFF; 881 uint c1 = (RG_ETC1_GET_KEY_FROM_INDEX(index1) >> pass_shift) & 0xFF; 882 883 if (c0 == c1) 884 { 885 uint dst_offset0 = offsets[c0]; 886 887 offsets[c0] = dst_offset0 + 2; 888 889 pNew[dst_offset0] = static_cast<T>(index0); 890 pNew[dst_offset0 + 1] = static_cast<T>(index1); 891 } 892 else 893 { 894 uint dst_offset0 = offsets[c0]++; 895 uint dst_offset1 = offsets[c1]++; 896 897 pNew[dst_offset0] = static_cast<T>(index0); 898 pNew[dst_offset1] = static_cast<T>(index1); 899 } 900 } 901 902 if (num_indices & 1) 903 { 904 uint index = *p; 905 uint c = (RG_ETC1_GET_KEY_FROM_INDEX(index) >> pass_shift) & 0xFF; 906 907 uint dst_offset = offsets[c]; 908 offsets[c] = dst_offset + 1; 909 910 pNew[dst_offset] = static_cast<T>(index); 911 } 912 913 T* t = pCur; 914 pCur = pNew; 915 pNew = t; 916 } 917 918 return pCur; 919 } 920 921 #undef RG_ETC1_GET_KEY 922 #undef RG_ETC1_GET_KEY_FROM_INDEX 923 pack_color5(const color_quad_u8 & color,bool scaled,uint bias)924 uint16 etc1_block::pack_color5(const color_quad_u8& color, bool scaled, uint bias) 925 { 926 return pack_color5(color.r, color.g, color.b, scaled, bias); 927 } 928 pack_color5(uint r,uint g,uint b,bool scaled,uint bias)929 uint16 etc1_block::pack_color5(uint r, uint g, uint b, bool scaled, uint bias) 930 { 931 if (scaled) 932 { 933 r = (r * 31U + bias) / 255U; 934 g = (g * 31U + bias) / 255U; 935 b = (b * 31U + bias) / 255U; 936 } 937 938 r = rg_etc1::minimum(r, 31U); 939 g = rg_etc1::minimum(g, 31U); 940 b = rg_etc1::minimum(b, 31U); 941 942 return static_cast<uint16>(b | (g << 5U) | (r << 10U)); 943 } 944 unpack_color5(uint16 packed_color5,bool scaled,uint alpha)945 color_quad_u8 etc1_block::unpack_color5(uint16 packed_color5, bool scaled, uint alpha) 946 { 947 uint b = packed_color5 & 31U; 948 uint g = (packed_color5 >> 5U) & 31U; 949 uint r = (packed_color5 >> 10U) & 31U; 950 951 if (scaled) 952 { 953 b = (b << 3U) | (b >> 2U); 954 g = (g << 3U) | (g >> 2U); 955 r = (r << 3U) | (r >> 2U); 956 } 957 958 return color_quad_u8(cNoClamp, r, g, b, rg_etc1::minimum(alpha, 255U)); 959 } 960 unpack_color5(uint & r,uint & g,uint & b,uint16 packed_color5,bool scaled)961 void etc1_block::unpack_color5(uint& r, uint& g, uint& b, uint16 packed_color5, bool scaled) 962 { 963 color_quad_u8 c(unpack_color5(packed_color5, scaled, 0)); 964 r = c.r; 965 g = c.g; 966 b = c.b; 967 } 968 unpack_color5(color_quad_u8 & result,uint16 packed_color5,uint16 packed_delta3,bool scaled,uint alpha)969 bool etc1_block::unpack_color5(color_quad_u8& result, uint16 packed_color5, uint16 packed_delta3, bool scaled, uint alpha) 970 { 971 int dc_r, dc_g, dc_b; 972 unpack_delta3(dc_r, dc_g, dc_b, packed_delta3); 973 974 int b = (packed_color5 & 31U) + dc_b; 975 int g = ((packed_color5 >> 5U) & 31U) + dc_g; 976 int r = ((packed_color5 >> 10U) & 31U) + dc_r; 977 978 bool success = true; 979 if (static_cast<uint>(r | g | b) > 31U) 980 { 981 success = false; 982 r = rg_etc1::clamp<int>(r, 0, 31); 983 g = rg_etc1::clamp<int>(g, 0, 31); 984 b = rg_etc1::clamp<int>(b, 0, 31); 985 } 986 987 if (scaled) 988 { 989 b = (b << 3U) | (b >> 2U); 990 g = (g << 3U) | (g >> 2U); 991 r = (r << 3U) | (r >> 2U); 992 } 993 994 result.set_noclamp_rgba(r, g, b, rg_etc1::minimum(alpha, 255U)); 995 return success; 996 } 997 unpack_color5(uint & r,uint & g,uint & b,uint16 packed_color5,uint16 packed_delta3,bool scaled,uint alpha)998 bool etc1_block::unpack_color5(uint& r, uint& g, uint& b, uint16 packed_color5, uint16 packed_delta3, bool scaled, uint alpha) 999 { 1000 color_quad_u8 result; 1001 const bool success = unpack_color5(result, packed_color5, packed_delta3, scaled, alpha); 1002 r = result.r; 1003 g = result.g; 1004 b = result.b; 1005 return success; 1006 } 1007 pack_delta3(int r,int g,int b)1008 uint16 etc1_block::pack_delta3(int r, int g, int b) 1009 { 1010 RG_ETC1_ASSERT((r >= cETC1ColorDeltaMin) && (r <= cETC1ColorDeltaMax)); 1011 RG_ETC1_ASSERT((g >= cETC1ColorDeltaMin) && (g <= cETC1ColorDeltaMax)); 1012 RG_ETC1_ASSERT((b >= cETC1ColorDeltaMin) && (b <= cETC1ColorDeltaMax)); 1013 if (r < 0) r += 8; 1014 if (g < 0) g += 8; 1015 if (b < 0) b += 8; 1016 return static_cast<uint16>(b | (g << 3) | (r << 6)); 1017 } 1018 unpack_delta3(int & r,int & g,int & b,uint16 packed_delta3)1019 void etc1_block::unpack_delta3(int& r, int& g, int& b, uint16 packed_delta3) 1020 { 1021 r = (packed_delta3 >> 6) & 7; 1022 g = (packed_delta3 >> 3) & 7; 1023 b = packed_delta3 & 7; 1024 if (r >= 4) r -= 8; 1025 if (g >= 4) g -= 8; 1026 if (b >= 4) b -= 8; 1027 } 1028 pack_color4(const color_quad_u8 & color,bool scaled,uint bias)1029 uint16 etc1_block::pack_color4(const color_quad_u8& color, bool scaled, uint bias) 1030 { 1031 return pack_color4(color.r, color.g, color.b, scaled, bias); 1032 } 1033 pack_color4(uint r,uint g,uint b,bool scaled,uint bias)1034 uint16 etc1_block::pack_color4(uint r, uint g, uint b, bool scaled, uint bias) 1035 { 1036 if (scaled) 1037 { 1038 r = (r * 15U + bias) / 255U; 1039 g = (g * 15U + bias) / 255U; 1040 b = (b * 15U + bias) / 255U; 1041 } 1042 1043 r = rg_etc1::minimum(r, 15U); 1044 g = rg_etc1::minimum(g, 15U); 1045 b = rg_etc1::minimum(b, 15U); 1046 1047 return static_cast<uint16>(b | (g << 4U) | (r << 8U)); 1048 } 1049 unpack_color4(uint16 packed_color4,bool scaled,uint alpha)1050 color_quad_u8 etc1_block::unpack_color4(uint16 packed_color4, bool scaled, uint alpha) 1051 { 1052 uint b = packed_color4 & 15U; 1053 uint g = (packed_color4 >> 4U) & 15U; 1054 uint r = (packed_color4 >> 8U) & 15U; 1055 1056 if (scaled) 1057 { 1058 b = (b << 4U) | b; 1059 g = (g << 4U) | g; 1060 r = (r << 4U) | r; 1061 } 1062 1063 return color_quad_u8(cNoClamp, r, g, b, rg_etc1::minimum(alpha, 255U)); 1064 } 1065 unpack_color4(uint & r,uint & g,uint & b,uint16 packed_color4,bool scaled)1066 void etc1_block::unpack_color4(uint& r, uint& g, uint& b, uint16 packed_color4, bool scaled) 1067 { 1068 color_quad_u8 c(unpack_color4(packed_color4, scaled, 0)); 1069 r = c.r; 1070 g = c.g; 1071 b = c.b; 1072 } 1073 get_diff_subblock_colors(color_quad_u8 * pDst,uint16 packed_color5,uint table_idx)1074 void etc1_block::get_diff_subblock_colors(color_quad_u8* pDst, uint16 packed_color5, uint table_idx) 1075 { 1076 RG_ETC1_ASSERT(table_idx < cETC1IntenModifierValues); 1077 const int *pInten_modifer_table = &g_etc1_inten_tables[table_idx][0]; 1078 1079 uint r, g, b; 1080 unpack_color5(r, g, b, packed_color5, true); 1081 1082 const int ir = static_cast<int>(r), ig = static_cast<int>(g), ib = static_cast<int>(b); 1083 1084 const int y0 = pInten_modifer_table[0]; 1085 pDst[0].set(ir + y0, ig + y0, ib + y0); 1086 1087 const int y1 = pInten_modifer_table[1]; 1088 pDst[1].set(ir + y1, ig + y1, ib + y1); 1089 1090 const int y2 = pInten_modifer_table[2]; 1091 pDst[2].set(ir + y2, ig + y2, ib + y2); 1092 1093 const int y3 = pInten_modifer_table[3]; 1094 pDst[3].set(ir + y3, ig + y3, ib + y3); 1095 } 1096 get_diff_subblock_colors(color_quad_u8 * pDst,uint16 packed_color5,uint16 packed_delta3,uint table_idx)1097 bool etc1_block::get_diff_subblock_colors(color_quad_u8* pDst, uint16 packed_color5, uint16 packed_delta3, uint table_idx) 1098 { 1099 RG_ETC1_ASSERT(table_idx < cETC1IntenModifierValues); 1100 const int *pInten_modifer_table = &g_etc1_inten_tables[table_idx][0]; 1101 1102 uint r, g, b; 1103 bool success = unpack_color5(r, g, b, packed_color5, packed_delta3, true); 1104 1105 const int ir = static_cast<int>(r), ig = static_cast<int>(g), ib = static_cast<int>(b); 1106 1107 const int y0 = pInten_modifer_table[0]; 1108 pDst[0].set(ir + y0, ig + y0, ib + y0); 1109 1110 const int y1 = pInten_modifer_table[1]; 1111 pDst[1].set(ir + y1, ig + y1, ib + y1); 1112 1113 const int y2 = pInten_modifer_table[2]; 1114 pDst[2].set(ir + y2, ig + y2, ib + y2); 1115 1116 const int y3 = pInten_modifer_table[3]; 1117 pDst[3].set(ir + y3, ig + y3, ib + y3); 1118 1119 return success; 1120 } 1121 get_abs_subblock_colors(color_quad_u8 * pDst,uint16 packed_color4,uint table_idx)1122 void etc1_block::get_abs_subblock_colors(color_quad_u8* pDst, uint16 packed_color4, uint table_idx) 1123 { 1124 RG_ETC1_ASSERT(table_idx < cETC1IntenModifierValues); 1125 const int *pInten_modifer_table = &g_etc1_inten_tables[table_idx][0]; 1126 1127 uint r, g, b; 1128 unpack_color4(r, g, b, packed_color4, true); 1129 1130 const int ir = static_cast<int>(r), ig = static_cast<int>(g), ib = static_cast<int>(b); 1131 1132 const int y0 = pInten_modifer_table[0]; 1133 pDst[0].set(ir + y0, ig + y0, ib + y0); 1134 1135 const int y1 = pInten_modifer_table[1]; 1136 pDst[1].set(ir + y1, ig + y1, ib + y1); 1137 1138 const int y2 = pInten_modifer_table[2]; 1139 pDst[2].set(ir + y2, ig + y2, ib + y2); 1140 1141 const int y3 = pInten_modifer_table[3]; 1142 pDst[3].set(ir + y3, ig + y3, ib + y3); 1143 } 1144 unpack_etc1_block(const void * pETC1_block,unsigned int * pDst_pixels_rgba,bool preserve_alpha)1145 bool unpack_etc1_block(const void* pETC1_block, unsigned int* pDst_pixels_rgba, bool preserve_alpha) 1146 { 1147 color_quad_u8* pDst = reinterpret_cast<color_quad_u8*>(pDst_pixels_rgba); 1148 const etc1_block& block = *static_cast<const etc1_block*>(pETC1_block); 1149 1150 const bool diff_flag = block.get_diff_bit(); 1151 const bool flip_flag = block.get_flip_bit(); 1152 const uint table_index0 = block.get_inten_table(0); 1153 const uint table_index1 = block.get_inten_table(1); 1154 1155 color_quad_u8 subblock_colors0[4]; 1156 color_quad_u8 subblock_colors1[4]; 1157 bool success = true; 1158 1159 if (diff_flag) 1160 { 1161 const uint16 base_color5 = block.get_base5_color(); 1162 const uint16 delta_color3 = block.get_delta3_color(); 1163 etc1_block::get_diff_subblock_colors(subblock_colors0, base_color5, table_index0); 1164 1165 if (!etc1_block::get_diff_subblock_colors(subblock_colors1, base_color5, delta_color3, table_index1)) 1166 success = false; 1167 } 1168 else 1169 { 1170 const uint16 base_color4_0 = block.get_base4_color(0); 1171 etc1_block::get_abs_subblock_colors(subblock_colors0, base_color4_0, table_index0); 1172 1173 const uint16 base_color4_1 = block.get_base4_color(1); 1174 etc1_block::get_abs_subblock_colors(subblock_colors1, base_color4_1, table_index1); 1175 } 1176 1177 if (preserve_alpha) 1178 { 1179 if (flip_flag) 1180 { 1181 for (uint y = 0; y < 2; y++) 1182 { 1183 pDst[0].set_rgb(subblock_colors0[block.get_selector(0, y)]); 1184 pDst[1].set_rgb(subblock_colors0[block.get_selector(1, y)]); 1185 pDst[2].set_rgb(subblock_colors0[block.get_selector(2, y)]); 1186 pDst[3].set_rgb(subblock_colors0[block.get_selector(3, y)]); 1187 pDst += 4; 1188 } 1189 1190 for (uint y = 2; y < 4; y++) 1191 { 1192 pDst[0].set_rgb(subblock_colors1[block.get_selector(0, y)]); 1193 pDst[1].set_rgb(subblock_colors1[block.get_selector(1, y)]); 1194 pDst[2].set_rgb(subblock_colors1[block.get_selector(2, y)]); 1195 pDst[3].set_rgb(subblock_colors1[block.get_selector(3, y)]); 1196 pDst += 4; 1197 } 1198 } 1199 else 1200 { 1201 for (uint y = 0; y < 4; y++) 1202 { 1203 pDst[0].set_rgb(subblock_colors0[block.get_selector(0, y)]); 1204 pDst[1].set_rgb(subblock_colors0[block.get_selector(1, y)]); 1205 pDst[2].set_rgb(subblock_colors1[block.get_selector(2, y)]); 1206 pDst[3].set_rgb(subblock_colors1[block.get_selector(3, y)]); 1207 pDst += 4; 1208 } 1209 } 1210 } 1211 else 1212 { 1213 if (flip_flag) 1214 { 1215 // 0000 1216 // 0000 1217 // 1111 1218 // 1111 1219 for (uint y = 0; y < 2; y++) 1220 { 1221 pDst[0] = subblock_colors0[block.get_selector(0, y)]; 1222 pDst[1] = subblock_colors0[block.get_selector(1, y)]; 1223 pDst[2] = subblock_colors0[block.get_selector(2, y)]; 1224 pDst[3] = subblock_colors0[block.get_selector(3, y)]; 1225 pDst += 4; 1226 } 1227 1228 for (uint y = 2; y < 4; y++) 1229 { 1230 pDst[0] = subblock_colors1[block.get_selector(0, y)]; 1231 pDst[1] = subblock_colors1[block.get_selector(1, y)]; 1232 pDst[2] = subblock_colors1[block.get_selector(2, y)]; 1233 pDst[3] = subblock_colors1[block.get_selector(3, y)]; 1234 pDst += 4; 1235 } 1236 } 1237 else 1238 { 1239 // 0011 1240 // 0011 1241 // 0011 1242 // 0011 1243 for (uint y = 0; y < 4; y++) 1244 { 1245 pDst[0] = subblock_colors0[block.get_selector(0, y)]; 1246 pDst[1] = subblock_colors0[block.get_selector(1, y)]; 1247 pDst[2] = subblock_colors1[block.get_selector(2, y)]; 1248 pDst[3] = subblock_colors1[block.get_selector(3, y)]; 1249 pDst += 4; 1250 } 1251 } 1252 } 1253 1254 return success; 1255 } 1256 1257 struct etc1_solution_coordinates 1258 { etc1_solution_coordinatesrg_etc1::etc1_solution_coordinates1259 inline etc1_solution_coordinates() : 1260 m_unscaled_color(0, 0, 0, 0), 1261 m_inten_table(0), 1262 m_color4(false) 1263 { 1264 } 1265 etc1_solution_coordinatesrg_etc1::etc1_solution_coordinates1266 inline etc1_solution_coordinates(uint r, uint g, uint b, uint inten_table, bool color4) : 1267 m_unscaled_color(r, g, b, 255), 1268 m_inten_table(inten_table), 1269 m_color4(color4) 1270 { 1271 } 1272 etc1_solution_coordinatesrg_etc1::etc1_solution_coordinates1273 inline etc1_solution_coordinates(const color_quad_u8& c, uint inten_table, bool color4) : 1274 m_unscaled_color(c), 1275 m_inten_table(inten_table), 1276 m_color4(color4) 1277 { 1278 } 1279 etc1_solution_coordinatesrg_etc1::etc1_solution_coordinates1280 inline etc1_solution_coordinates(const etc1_solution_coordinates& other) 1281 { 1282 *this = other; 1283 } 1284 operator =rg_etc1::etc1_solution_coordinates1285 inline etc1_solution_coordinates& operator= (const etc1_solution_coordinates& rhs) 1286 { 1287 m_unscaled_color = rhs.m_unscaled_color; 1288 m_inten_table = rhs.m_inten_table; 1289 m_color4 = rhs.m_color4; 1290 return *this; 1291 } 1292 clearrg_etc1::etc1_solution_coordinates1293 inline void clear() 1294 { 1295 m_unscaled_color.clear(); 1296 m_inten_table = 0; 1297 m_color4 = false; 1298 } 1299 get_scaled_colorrg_etc1::etc1_solution_coordinates1300 inline color_quad_u8 get_scaled_color() const 1301 { 1302 int br, bg, bb; 1303 if (m_color4) 1304 { 1305 br = m_unscaled_color.r | (m_unscaled_color.r << 4); 1306 bg = m_unscaled_color.g | (m_unscaled_color.g << 4); 1307 bb = m_unscaled_color.b | (m_unscaled_color.b << 4); 1308 } 1309 else 1310 { 1311 br = (m_unscaled_color.r >> 2) | (m_unscaled_color.r << 3); 1312 bg = (m_unscaled_color.g >> 2) | (m_unscaled_color.g << 3); 1313 bb = (m_unscaled_color.b >> 2) | (m_unscaled_color.b << 3); 1314 } 1315 return color_quad_u8(br, bg, bb); 1316 } 1317 get_block_colorsrg_etc1::etc1_solution_coordinates1318 inline void get_block_colors(color_quad_u8* pBlock_colors) 1319 { 1320 int br, bg, bb; 1321 if (m_color4) 1322 { 1323 br = m_unscaled_color.r | (m_unscaled_color.r << 4); 1324 bg = m_unscaled_color.g | (m_unscaled_color.g << 4); 1325 bb = m_unscaled_color.b | (m_unscaled_color.b << 4); 1326 } 1327 else 1328 { 1329 br = (m_unscaled_color.r >> 2) | (m_unscaled_color.r << 3); 1330 bg = (m_unscaled_color.g >> 2) | (m_unscaled_color.g << 3); 1331 bb = (m_unscaled_color.b >> 2) | (m_unscaled_color.b << 3); 1332 } 1333 const int* pInten_table = g_etc1_inten_tables[m_inten_table]; 1334 pBlock_colors[0].set(br + pInten_table[0], bg + pInten_table[0], bb + pInten_table[0]); 1335 pBlock_colors[1].set(br + pInten_table[1], bg + pInten_table[1], bb + pInten_table[1]); 1336 pBlock_colors[2].set(br + pInten_table[2], bg + pInten_table[2], bb + pInten_table[2]); 1337 pBlock_colors[3].set(br + pInten_table[3], bg + pInten_table[3], bb + pInten_table[3]); 1338 } 1339 1340 color_quad_u8 m_unscaled_color; 1341 uint m_inten_table; 1342 bool m_color4; 1343 }; 1344 1345 class etc1_optimizer 1346 { 1347 etc1_optimizer(const etc1_optimizer&); 1348 etc1_optimizer& operator= (const etc1_optimizer&); 1349 1350 public: etc1_optimizer()1351 etc1_optimizer() 1352 { 1353 clear(); 1354 } 1355 clear()1356 void clear() 1357 { 1358 m_pParams = NULL; 1359 m_pResult = NULL; 1360 m_pSorted_luma = NULL; 1361 m_pSorted_luma_indices = NULL; 1362 } 1363 1364 struct params : etc1_pack_params 1365 { paramsrg_etc1::etc1_optimizer::params1366 params() 1367 { 1368 clear(); 1369 } 1370 paramsrg_etc1::etc1_optimizer::params1371 params(const etc1_pack_params& base_params) : 1372 etc1_pack_params(base_params) 1373 { 1374 clear_optimizer_params(); 1375 } 1376 clearrg_etc1::etc1_optimizer::params1377 void clear() 1378 { 1379 etc1_pack_params::clear(); 1380 clear_optimizer_params(); 1381 } 1382 clear_optimizer_paramsrg_etc1::etc1_optimizer::params1383 void clear_optimizer_params() 1384 { 1385 m_num_src_pixels = 0; 1386 m_pSrc_pixels = 0; 1387 1388 m_use_color4 = false; 1389 static const int s_default_scan_delta[] = { 0 }; 1390 m_pScan_deltas = s_default_scan_delta; 1391 m_scan_delta_size = 1; 1392 1393 m_base_color5.clear(); 1394 m_constrain_against_base_color5 = false; 1395 } 1396 1397 uint m_num_src_pixels; 1398 const color_quad_u8* m_pSrc_pixels; 1399 1400 bool m_use_color4; 1401 const int* m_pScan_deltas; 1402 uint m_scan_delta_size; 1403 1404 color_quad_u8 m_base_color5; 1405 bool m_constrain_against_base_color5; 1406 }; 1407 1408 struct results 1409 { 1410 uint64 m_error; 1411 color_quad_u8 m_block_color_unscaled; 1412 uint m_block_inten_table; 1413 uint m_n; 1414 uint8* m_pSelectors; 1415 bool m_block_color4; 1416 operator =rg_etc1::etc1_optimizer::results1417 inline results& operator= (const results& rhs) 1418 { 1419 m_block_color_unscaled = rhs.m_block_color_unscaled; 1420 m_block_color4 = rhs.m_block_color4; 1421 m_block_inten_table = rhs.m_block_inten_table; 1422 m_error = rhs.m_error; 1423 RG_ETC1_ASSERT(m_n == rhs.m_n); 1424 memcpy(m_pSelectors, rhs.m_pSelectors, rhs.m_n); 1425 return *this; 1426 } 1427 }; 1428 1429 void init(const params& params, results& result); 1430 bool compute(); 1431 1432 private: 1433 struct potential_solution 1434 { potential_solutionrg_etc1::etc1_optimizer::potential_solution1435 potential_solution() : m_coords(), m_error(cUINT64_MAX), m_valid(false) 1436 { 1437 } 1438 1439 etc1_solution_coordinates m_coords; 1440 uint8 m_selectors[8]; 1441 uint64 m_error; 1442 bool m_valid; 1443 clearrg_etc1::etc1_optimizer::potential_solution1444 void clear() 1445 { 1446 m_coords.clear(); 1447 m_error = cUINT64_MAX; 1448 m_valid = false; 1449 } 1450 }; 1451 1452 const params* m_pParams; 1453 results* m_pResult; 1454 1455 int m_limit; 1456 1457 vec3F m_avg_color; 1458 int m_br, m_bg, m_bb; 1459 uint16 m_luma[8]; 1460 uint32 m_sorted_luma[2][8]; 1461 const uint32* m_pSorted_luma_indices; 1462 uint32* m_pSorted_luma; 1463 1464 uint8 m_selectors[8]; 1465 uint8 m_best_selectors[8]; 1466 1467 potential_solution m_best_solution; 1468 potential_solution m_trial_solution; 1469 uint8 m_temp_selectors[8]; 1470 1471 bool evaluate_solution(const etc1_solution_coordinates& coords, potential_solution& trial_solution, potential_solution* pBest_solution); 1472 bool evaluate_solution_fast(const etc1_solution_coordinates& coords, potential_solution& trial_solution, potential_solution* pBest_solution); 1473 }; 1474 compute()1475 bool etc1_optimizer::compute() 1476 { 1477 const uint n = m_pParams->m_num_src_pixels; 1478 const int scan_delta_size = m_pParams->m_scan_delta_size; 1479 1480 // Scan through a subset of the 3D lattice centered around the avg block color trying each 3D (555 or 444) lattice point as a potential block color. 1481 // Each time a better solution is found try to refine the current solution's block color based of the current selectors and intensity table index. 1482 for (int zdi = 0; zdi < scan_delta_size; zdi++) 1483 { 1484 const int zd = m_pParams->m_pScan_deltas[zdi]; 1485 const int mbb = m_bb + zd; 1486 if (mbb < 0) continue; else if (mbb > m_limit) break; 1487 1488 for (int ydi = 0; ydi < scan_delta_size; ydi++) 1489 { 1490 const int yd = m_pParams->m_pScan_deltas[ydi]; 1491 const int mbg = m_bg + yd; 1492 if (mbg < 0) continue; else if (mbg > m_limit) break; 1493 1494 for (int xdi = 0; xdi < scan_delta_size; xdi++) 1495 { 1496 const int xd = m_pParams->m_pScan_deltas[xdi]; 1497 const int mbr = m_br + xd; 1498 if (mbr < 0) continue; else if (mbr > m_limit) break; 1499 1500 etc1_solution_coordinates coords(mbr, mbg, mbb, 0, m_pParams->m_use_color4); 1501 if (m_pParams->m_quality == cHighQuality) 1502 { 1503 if (!evaluate_solution(coords, m_trial_solution, &m_best_solution)) 1504 continue; 1505 } 1506 else 1507 { 1508 if (!evaluate_solution_fast(coords, m_trial_solution, &m_best_solution)) 1509 continue; 1510 } 1511 1512 // Now we have the input block, the avg. color of the input pixels, a set of trial selector indices, and the block color+intensity index. 1513 // Now, for each component, attempt to refine the current solution by solving a simple linear equation. For example, for 4 colors: 1514 // The goal is: 1515 // pixel0 - (block_color+inten_table[selector0]) + pixel1 - (block_color+inten_table[selector1]) + pixel2 - (block_color+inten_table[selector2]) + pixel3 - (block_color+inten_table[selector3]) = 0 1516 // Rearranging this: 1517 // (pixel0 + pixel1 + pixel2 + pixel3) - (block_color+inten_table[selector0]) - (block_color+inten_table[selector1]) - (block_color+inten_table[selector2]) - (block_color+inten_table[selector3]) = 0 1518 // (pixel0 + pixel1 + pixel2 + pixel3) - block_color - inten_table[selector0] - block_color-inten_table[selector1] - block_color-inten_table[selector2] - block_color-inten_table[selector3] = 0 1519 // (pixel0 + pixel1 + pixel2 + pixel3) - 4*block_color - inten_table[selector0] - inten_table[selector1] - inten_table[selector2] - inten_table[selector3] = 0 1520 // (pixel0 + pixel1 + pixel2 + pixel3) - 4*block_color - (inten_table[selector0] + inten_table[selector1] + inten_table[selector2] + inten_table[selector3]) = 0 1521 // (pixel0 + pixel1 + pixel2 + pixel3)/4 - block_color - (inten_table[selector0] + inten_table[selector1] + inten_table[selector2] + inten_table[selector3])/4 = 0 1522 // block_color = (pixel0 + pixel1 + pixel2 + pixel3)/4 - (inten_table[selector0] + inten_table[selector1] + inten_table[selector2] + inten_table[selector3])/4 1523 // So what this means: 1524 // optimal_block_color = avg_input - avg_inten_delta 1525 // So the optimal block color can be computed by taking the average block color and subtracting the current average of the intensity delta. 1526 // Unfortunately, optimal_block_color must then be quantized to 555 or 444 so it's not always possible to improve matters using this formula. 1527 // Also, the above formula is for unclamped intensity deltas. The actual implementation takes into account clamping. 1528 1529 const uint max_refinement_trials = (m_pParams->m_quality == cLowQuality) ? 2 : (((xd | yd | zd) == 0) ? 4 : 2); 1530 for (uint refinement_trial = 0; refinement_trial < max_refinement_trials; refinement_trial++) 1531 { 1532 const uint8* pSelectors = m_best_solution.m_selectors; 1533 const int* pInten_table = g_etc1_inten_tables[m_best_solution.m_coords.m_inten_table]; 1534 1535 int delta_sum_r = 0, delta_sum_g = 0, delta_sum_b = 0; 1536 const color_quad_u8 base_color(m_best_solution.m_coords.get_scaled_color()); 1537 for (uint r = 0; r < n; r++) 1538 { 1539 const uint s = *pSelectors++; 1540 const int yd = pInten_table[s]; 1541 // Compute actual delta being applied to each pixel, taking into account clamping. 1542 delta_sum_r += rg_etc1::clamp<int>(base_color.r + yd, 0, 255) - base_color.r; 1543 delta_sum_g += rg_etc1::clamp<int>(base_color.g + yd, 0, 255) - base_color.g; 1544 delta_sum_b += rg_etc1::clamp<int>(base_color.b + yd, 0, 255) - base_color.b; 1545 } 1546 if ((!delta_sum_r) && (!delta_sum_g) && (!delta_sum_b)) 1547 break; 1548 const float avg_delta_r_f = static_cast<float>(delta_sum_r) / n; 1549 const float avg_delta_g_f = static_cast<float>(delta_sum_g) / n; 1550 const float avg_delta_b_f = static_cast<float>(delta_sum_b) / n; 1551 const int br1 = rg_etc1::clamp<int>(static_cast<uint>((m_avg_color[0] - avg_delta_r_f) * m_limit / 255.0f + .5f), 0, m_limit); 1552 const int bg1 = rg_etc1::clamp<int>(static_cast<uint>((m_avg_color[1] - avg_delta_g_f) * m_limit / 255.0f + .5f), 0, m_limit); 1553 const int bb1 = rg_etc1::clamp<int>(static_cast<uint>((m_avg_color[2] - avg_delta_b_f) * m_limit / 255.0f + .5f), 0, m_limit); 1554 1555 bool skip = false; 1556 1557 if ((mbr == br1) && (mbg == bg1) && (mbb == bb1)) 1558 skip = true; 1559 else if ((br1 == m_best_solution.m_coords.m_unscaled_color.r) && (bg1 == m_best_solution.m_coords.m_unscaled_color.g) && (bb1 == m_best_solution.m_coords.m_unscaled_color.b)) 1560 skip = true; 1561 else if ((m_br == br1) && (m_bg == bg1) && (m_bb == bb1)) 1562 skip = true; 1563 1564 if (skip) 1565 break; 1566 1567 etc1_solution_coordinates coords1(br1, bg1, bb1, 0, m_pParams->m_use_color4); 1568 if (m_pParams->m_quality == cHighQuality) 1569 { 1570 if (!evaluate_solution(coords1, m_trial_solution, &m_best_solution)) 1571 break; 1572 } 1573 else 1574 { 1575 if (!evaluate_solution_fast(coords1, m_trial_solution, &m_best_solution)) 1576 break; 1577 } 1578 1579 } // refinement_trial 1580 1581 } // xdi 1582 } // ydi 1583 } // zdi 1584 1585 if (!m_best_solution.m_valid) 1586 { 1587 m_pResult->m_error = cUINT32_MAX; 1588 return false; 1589 } 1590 1591 const uint8* pSelectors = m_best_solution.m_selectors; 1592 1593 #ifdef RG_ETC1_BUILD_DEBUG 1594 { 1595 color_quad_u8 block_colors[4]; 1596 m_best_solution.m_coords.get_block_colors(block_colors); 1597 1598 const color_quad_u8* pSrc_pixels = m_pParams->m_pSrc_pixels; 1599 uint64 actual_error = 0; 1600 for (uint i = 0; i < n; i++) 1601 actual_error += pSrc_pixels[i].squared_distance_rgb(block_colors[pSelectors[i]]); 1602 1603 RG_ETC1_ASSERT(actual_error == m_best_solution.m_error); 1604 } 1605 #endif 1606 1607 m_pResult->m_error = m_best_solution.m_error; 1608 1609 m_pResult->m_block_color_unscaled = m_best_solution.m_coords.m_unscaled_color; 1610 m_pResult->m_block_color4 = m_best_solution.m_coords.m_color4; 1611 1612 m_pResult->m_block_inten_table = m_best_solution.m_coords.m_inten_table; 1613 memcpy(m_pResult->m_pSelectors, pSelectors, n); 1614 m_pResult->m_n = n; 1615 1616 return true; 1617 } 1618 init(const params & p,results & r)1619 void etc1_optimizer::init(const params& p, results& r) 1620 { 1621 // This version is hardcoded for 8 pixel subblocks. 1622 RG_ETC1_ASSERT(p.m_num_src_pixels == 8); 1623 1624 m_pParams = &p; 1625 m_pResult = &r; 1626 1627 const uint n = 8; 1628 1629 m_limit = m_pParams->m_use_color4 ? 15 : 31; 1630 1631 vec3F avg_color(0.0f); 1632 1633 for (uint i = 0; i < n; i++) 1634 { 1635 const color_quad_u8& c = m_pParams->m_pSrc_pixels[i]; 1636 const vec3F fc(c.r, c.g, c.b); 1637 1638 avg_color += fc; 1639 1640 m_luma[i] = static_cast<uint16>(c.r + c.g + c.b); 1641 m_sorted_luma[0][i] = i; 1642 } 1643 avg_color *= (1.0f / static_cast<float>(n)); 1644 m_avg_color = avg_color; 1645 1646 m_br = rg_etc1::clamp<int>(static_cast<uint>(m_avg_color[0] * m_limit / 255.0f + .5f), 0, m_limit); 1647 m_bg = rg_etc1::clamp<int>(static_cast<uint>(m_avg_color[1] * m_limit / 255.0f + .5f), 0, m_limit); 1648 m_bb = rg_etc1::clamp<int>(static_cast<uint>(m_avg_color[2] * m_limit / 255.0f + .5f), 0, m_limit); 1649 1650 if (m_pParams->m_quality <= cMediumQuality) 1651 { 1652 m_pSorted_luma_indices = indirect_radix_sort(n, m_sorted_luma[0], m_sorted_luma[1], m_luma, 0, sizeof(m_luma[0]), false); 1653 m_pSorted_luma = m_sorted_luma[0]; 1654 if (m_pSorted_luma_indices == m_sorted_luma[0]) 1655 m_pSorted_luma = m_sorted_luma[1]; 1656 1657 for (uint i = 0; i < n; i++) 1658 m_pSorted_luma[i] = m_luma[m_pSorted_luma_indices[i]]; 1659 } 1660 1661 m_best_solution.m_coords.clear(); 1662 m_best_solution.m_valid = false; 1663 m_best_solution.m_error = cUINT64_MAX; 1664 } 1665 evaluate_solution(const etc1_solution_coordinates & coords,potential_solution & trial_solution,potential_solution * pBest_solution)1666 bool etc1_optimizer::evaluate_solution(const etc1_solution_coordinates& coords, potential_solution& trial_solution, potential_solution* pBest_solution) 1667 { 1668 trial_solution.m_valid = false; 1669 1670 if (m_pParams->m_constrain_against_base_color5) 1671 { 1672 const int dr = coords.m_unscaled_color.r - m_pParams->m_base_color5.r; 1673 const int dg = coords.m_unscaled_color.g - m_pParams->m_base_color5.g; 1674 const int db = coords.m_unscaled_color.b - m_pParams->m_base_color5.b; 1675 1676 if ((rg_etc1::minimum(dr, dg, db) < cETC1ColorDeltaMin) || (rg_etc1::maximum(dr, dg, db) > cETC1ColorDeltaMax)) 1677 return false; 1678 } 1679 1680 const color_quad_u8 base_color(coords.get_scaled_color()); 1681 1682 const uint n = 8; 1683 1684 trial_solution.m_error = cUINT64_MAX; 1685 1686 for (uint inten_table = 0; inten_table < cETC1IntenModifierValues; inten_table++) 1687 { 1688 const int* pInten_table = g_etc1_inten_tables[inten_table]; 1689 1690 color_quad_u8 block_colors[4]; 1691 for (uint s = 0; s < 4; s++) 1692 { 1693 const int yd = pInten_table[s]; 1694 block_colors[s].set(base_color.r + yd, base_color.g + yd, base_color.b + yd, 0); 1695 } 1696 1697 uint64 total_error = 0; 1698 1699 const color_quad_u8* pSrc_pixels = m_pParams->m_pSrc_pixels; 1700 for (uint c = 0; c < n; c++) 1701 { 1702 const color_quad_u8& src_pixel = *pSrc_pixels++; 1703 1704 uint best_selector_index = 0; 1705 uint best_error = rg_etc1::square(src_pixel.r - block_colors[0].r) + rg_etc1::square(src_pixel.g - block_colors[0].g) + rg_etc1::square(src_pixel.b - block_colors[0].b); 1706 1707 uint trial_error = rg_etc1::square(src_pixel.r - block_colors[1].r) + rg_etc1::square(src_pixel.g - block_colors[1].g) + rg_etc1::square(src_pixel.b - block_colors[1].b); 1708 if (trial_error < best_error) 1709 { 1710 best_error = trial_error; 1711 best_selector_index = 1; 1712 } 1713 1714 trial_error = rg_etc1::square(src_pixel.r - block_colors[2].r) + rg_etc1::square(src_pixel.g - block_colors[2].g) + rg_etc1::square(src_pixel.b - block_colors[2].b); 1715 if (trial_error < best_error) 1716 { 1717 best_error = trial_error; 1718 best_selector_index = 2; 1719 } 1720 1721 trial_error = rg_etc1::square(src_pixel.r - block_colors[3].r) + rg_etc1::square(src_pixel.g - block_colors[3].g) + rg_etc1::square(src_pixel.b - block_colors[3].b); 1722 if (trial_error < best_error) 1723 { 1724 best_error = trial_error; 1725 best_selector_index = 3; 1726 } 1727 1728 m_temp_selectors[c] = static_cast<uint8>(best_selector_index); 1729 1730 total_error += best_error; 1731 if (total_error >= trial_solution.m_error) 1732 break; 1733 } 1734 1735 if (total_error < trial_solution.m_error) 1736 { 1737 trial_solution.m_error = total_error; 1738 trial_solution.m_coords.m_inten_table = inten_table; 1739 memcpy(trial_solution.m_selectors, m_temp_selectors, 8); 1740 trial_solution.m_valid = true; 1741 } 1742 } 1743 trial_solution.m_coords.m_unscaled_color = coords.m_unscaled_color; 1744 trial_solution.m_coords.m_color4 = m_pParams->m_use_color4; 1745 1746 bool success = false; 1747 if (pBest_solution) 1748 { 1749 if (trial_solution.m_error < pBest_solution->m_error) 1750 { 1751 *pBest_solution = trial_solution; 1752 success = true; 1753 } 1754 } 1755 1756 return success; 1757 } 1758 evaluate_solution_fast(const etc1_solution_coordinates & coords,potential_solution & trial_solution,potential_solution * pBest_solution)1759 bool etc1_optimizer::evaluate_solution_fast(const etc1_solution_coordinates& coords, potential_solution& trial_solution, potential_solution* pBest_solution) 1760 { 1761 if (m_pParams->m_constrain_against_base_color5) 1762 { 1763 const int dr = coords.m_unscaled_color.r - m_pParams->m_base_color5.r; 1764 const int dg = coords.m_unscaled_color.g - m_pParams->m_base_color5.g; 1765 const int db = coords.m_unscaled_color.b - m_pParams->m_base_color5.b; 1766 1767 if ((rg_etc1::minimum(dr, dg, db) < cETC1ColorDeltaMin) || (rg_etc1::maximum(dr, dg, db) > cETC1ColorDeltaMax)) 1768 { 1769 trial_solution.m_valid = false; 1770 return false; 1771 } 1772 } 1773 1774 const color_quad_u8 base_color(coords.get_scaled_color()); 1775 1776 const uint n = 8; 1777 1778 trial_solution.m_error = cUINT64_MAX; 1779 1780 for (int inten_table = cETC1IntenModifierValues - 1; inten_table >= 0; --inten_table) 1781 { 1782 const int* pInten_table = g_etc1_inten_tables[inten_table]; 1783 1784 uint block_inten[4]; 1785 color_quad_u8 block_colors[4]; 1786 for (uint s = 0; s < 4; s++) 1787 { 1788 const int yd = pInten_table[s]; 1789 color_quad_u8 block_color(base_color.r + yd, base_color.g + yd, base_color.b + yd, 0); 1790 block_colors[s] = block_color; 1791 block_inten[s] = block_color.r + block_color.g + block_color.b; 1792 } 1793 1794 // evaluate_solution_fast() enforces/assumesd a total ordering of the input colors along the intensity (1,1,1) axis to more quickly classify the inputs to selectors. 1795 // The inputs colors have been presorted along the projection onto this axis, and ETC1 block colors are always ordered along the intensity axis, so this classification is fast. 1796 // 0 1 2 3 1797 // 01 12 23 1798 const uint block_inten_midpoints[3] = { block_inten[0] + block_inten[1], block_inten[1] + block_inten[2], block_inten[2] + block_inten[3] }; 1799 1800 uint64 total_error = 0; 1801 const color_quad_u8* pSrc_pixels = m_pParams->m_pSrc_pixels; 1802 if ((m_pSorted_luma[n - 1] * 2) < block_inten_midpoints[0]) 1803 { 1804 if (block_inten[0] > m_pSorted_luma[n - 1]) 1805 { 1806 const uint min_error = labs(block_inten[0] - m_pSorted_luma[n - 1]); 1807 if (min_error >= trial_solution.m_error) 1808 continue; 1809 } 1810 1811 memset(&m_temp_selectors[0], 0, n); 1812 1813 for (uint c = 0; c < n; c++) 1814 total_error += block_colors[0].squared_distance_rgb(pSrc_pixels[c]); 1815 } 1816 else if ((m_pSorted_luma[0] * 2) >= block_inten_midpoints[2]) 1817 { 1818 if (m_pSorted_luma[0] > block_inten[3]) 1819 { 1820 const uint min_error = labs(m_pSorted_luma[0] - block_inten[3]); 1821 if (min_error >= trial_solution.m_error) 1822 continue; 1823 } 1824 1825 memset(&m_temp_selectors[0], 3, n); 1826 1827 for (uint c = 0; c < n; c++) 1828 total_error += block_colors[3].squared_distance_rgb(pSrc_pixels[c]); 1829 } 1830 else 1831 { 1832 uint cur_selector = 0, c; 1833 for (c = 0; c < n; c++) 1834 { 1835 const uint y = m_pSorted_luma[c]; 1836 while ((y * 2) >= block_inten_midpoints[cur_selector]) 1837 if (++cur_selector > 2) 1838 goto done; 1839 const uint sorted_pixel_index = m_pSorted_luma_indices[c]; 1840 m_temp_selectors[sorted_pixel_index] = static_cast<uint8>(cur_selector); 1841 total_error += block_colors[cur_selector].squared_distance_rgb(pSrc_pixels[sorted_pixel_index]); 1842 } 1843 done: 1844 while (c < n) 1845 { 1846 const uint sorted_pixel_index = m_pSorted_luma_indices[c]; 1847 m_temp_selectors[sorted_pixel_index] = 3; 1848 total_error += block_colors[3].squared_distance_rgb(pSrc_pixels[sorted_pixel_index]); 1849 ++c; 1850 } 1851 } 1852 1853 if (total_error < trial_solution.m_error) 1854 { 1855 trial_solution.m_error = total_error; 1856 trial_solution.m_coords.m_inten_table = inten_table; 1857 memcpy(trial_solution.m_selectors, m_temp_selectors, n); 1858 trial_solution.m_valid = true; 1859 if (!total_error) 1860 break; 1861 } 1862 } 1863 trial_solution.m_coords.m_unscaled_color = coords.m_unscaled_color; 1864 trial_solution.m_coords.m_color4 = m_pParams->m_use_color4; 1865 1866 bool success = false; 1867 if (pBest_solution) 1868 { 1869 if (trial_solution.m_error < pBest_solution->m_error) 1870 { 1871 *pBest_solution = trial_solution; 1872 success = true; 1873 } 1874 } 1875 1876 return success; 1877 } 1878 etc1_decode_value(uint diff,uint inten,uint selector,uint packed_c)1879 static uint etc1_decode_value(uint diff, uint inten, uint selector, uint packed_c) 1880 { 1881 const uint limit = diff ? 32 : 16; limit; 1882 RG_ETC1_ASSERT((diff < 2) && (inten < 8) && (selector < 4) && (packed_c < limit)); 1883 int c; 1884 if (diff) 1885 c = (packed_c >> 2) | (packed_c << 3); 1886 else 1887 c = packed_c | (packed_c << 4); 1888 c += g_etc1_inten_tables[inten][selector]; 1889 c = rg_etc1::clamp<int>(c, 0, 255); 1890 return c; 1891 } 1892 mul_8bit(int a,int b)1893 static inline int mul_8bit(int a, int b) { int t = a*b + 128; return (t + (t >> 8)) >> 8; } 1894 pack_etc1_block_init()1895 void pack_etc1_block_init() 1896 { 1897 for (uint diff = 0; diff < 2; diff++) 1898 { 1899 const uint limit = diff ? 32 : 16; 1900 1901 for (uint inten = 0; inten < 8; inten++) 1902 { 1903 for (uint selector = 0; selector < 4; selector++) 1904 { 1905 const uint inverse_table_index = diff + (inten << 1) + (selector << 4); 1906 for (uint color = 0; color < 256; color++) 1907 { 1908 uint best_error = cUINT32_MAX, best_packed_c = 0; 1909 for (uint packed_c = 0; packed_c < limit; packed_c++) 1910 { 1911 int v = etc1_decode_value(diff, inten, selector, packed_c); 1912 uint err = labs(v - static_cast<int>(color)); 1913 if (err < best_error) 1914 { 1915 best_error = err; 1916 best_packed_c = packed_c; 1917 if (!best_error) 1918 break; 1919 } 1920 } 1921 RG_ETC1_ASSERT(best_error <= 255); 1922 g_etc1_inverse_lookup[inverse_table_index][color] = static_cast<uint16>(best_packed_c | (best_error << 8)); 1923 } 1924 } 1925 } 1926 } 1927 1928 uint expand5[32]; 1929 for(int i = 0; i < 32; i++) 1930 expand5[i] = (i << 3) | (i >> 2); 1931 1932 for(int i = 0; i < 256 + 16; i++) 1933 { 1934 int v = clamp<int>(i - 8, 0, 255); 1935 g_quant5_tab[i] = static_cast<uint8>(expand5[mul_8bit(v,31)]); 1936 } 1937 } 1938 1939 // Packs solid color blocks efficiently using a set of small precomputed tables. 1940 // For random 888 inputs, MSE results are better than Erricson's ETC1 packer in "slow" mode ~9.5% of the time, is slightly worse only ~.01% of the time, and is equal the rest of the time. pack_etc1_block_solid_color(etc1_block & block,const uint8 * pColor,etc1_pack_params & pack_params)1941 static uint64 pack_etc1_block_solid_color(etc1_block& block, const uint8* pColor, etc1_pack_params& pack_params) 1942 { 1943 pack_params; 1944 RG_ETC1_ASSERT(g_etc1_inverse_lookup[0][255]); 1945 1946 static uint s_next_comp[4] = { 1, 2, 0, 1 }; 1947 1948 uint best_error = cUINT32_MAX, best_i = 0; 1949 int best_x = 0, best_packed_c1 = 0, best_packed_c2 = 0; 1950 1951 // For each possible 8-bit value, there is a precomputed list of diff/inten/selector configurations that allow that 8-bit value to be encoded with no error. 1952 for (uint i = 0; i < 3; i++) 1953 { 1954 const uint c1 = pColor[s_next_comp[i]], c2 = pColor[s_next_comp[i + 1]]; 1955 1956 const int delta_range = 1; 1957 for (int delta = -delta_range; delta <= delta_range; delta++) 1958 { 1959 const int c_plus_delta = rg_etc1::clamp<int>(pColor[i] + delta, 0, 255); 1960 1961 const uint16* pTable; 1962 if (!c_plus_delta) 1963 pTable = g_color8_to_etc_block_config_0_255[0]; 1964 else if (c_plus_delta == 255) 1965 pTable = g_color8_to_etc_block_config_0_255[1]; 1966 else 1967 pTable = g_color8_to_etc_block_config_1_to_254[c_plus_delta - 1]; 1968 1969 do 1970 { 1971 const uint x = *pTable++; 1972 1973 #ifdef RG_ETC1_BUILD_DEBUG 1974 const uint diff = x & 1; 1975 const uint inten = (x >> 1) & 7; 1976 const uint selector = (x >> 4) & 3; 1977 const uint p0 = (x >> 8) & 255; 1978 RG_ETC1_ASSERT(etc1_decode_value(diff, inten, selector, p0) == (uint)c_plus_delta); 1979 #endif 1980 1981 const uint16* pInverse_table = g_etc1_inverse_lookup[x & 0xFF]; 1982 uint16 p1 = pInverse_table[c1]; 1983 uint16 p2 = pInverse_table[c2]; 1984 const uint trial_error = rg_etc1::square(c_plus_delta - pColor[i]) + rg_etc1::square(p1 >> 8) + rg_etc1::square(p2 >> 8); 1985 if (trial_error < best_error) 1986 { 1987 best_error = trial_error; 1988 best_x = x; 1989 best_packed_c1 = p1 & 0xFF; 1990 best_packed_c2 = p2 & 0xFF; 1991 best_i = i; 1992 if (!best_error) 1993 goto found_perfect_match; 1994 } 1995 } while (*pTable != 0xFFFF); 1996 } 1997 } 1998 found_perfect_match: 1999 2000 const uint diff = best_x & 1; 2001 const uint inten = (best_x >> 1) & 7; 2002 2003 block.m_bytes[3] = static_cast<uint8>(((inten | (inten << 3)) << 2) | (diff << 1)); 2004 2005 const uint etc1_selector = g_selector_index_to_etc1[(best_x >> 4) & 3]; 2006 *reinterpret_cast<uint16*>(&block.m_bytes[4]) = (etc1_selector & 2) ? 0xFFFF : 0; 2007 *reinterpret_cast<uint16*>(&block.m_bytes[6]) = (etc1_selector & 1) ? 0xFFFF : 0; 2008 2009 const uint best_packed_c0 = (best_x >> 8) & 255; 2010 if (diff) 2011 { 2012 block.m_bytes[best_i] = static_cast<uint8>(best_packed_c0 << 3); 2013 block.m_bytes[s_next_comp[best_i]] = static_cast<uint8>(best_packed_c1 << 3); 2014 block.m_bytes[s_next_comp[best_i+1]] = static_cast<uint8>(best_packed_c2 << 3); 2015 } 2016 else 2017 { 2018 block.m_bytes[best_i] = static_cast<uint8>(best_packed_c0 | (best_packed_c0 << 4)); 2019 block.m_bytes[s_next_comp[best_i]] = static_cast<uint8>(best_packed_c1 | (best_packed_c1 << 4)); 2020 block.m_bytes[s_next_comp[best_i+1]] = static_cast<uint8>(best_packed_c2 | (best_packed_c2 << 4)); 2021 } 2022 2023 return best_error; 2024 } 2025 pack_etc1_block_solid_color_constrained(etc1_optimizer::results & results,uint num_colors,const uint8 * pColor,etc1_pack_params & pack_params,bool use_diff,const color_quad_u8 * pBase_color5_unscaled)2026 static uint pack_etc1_block_solid_color_constrained( 2027 etc1_optimizer::results& results, 2028 uint num_colors, const uint8* pColor, 2029 etc1_pack_params& pack_params, 2030 bool use_diff, 2031 const color_quad_u8* pBase_color5_unscaled) 2032 { 2033 RG_ETC1_ASSERT(g_etc1_inverse_lookup[0][255]); 2034 2035 pack_params; 2036 static uint s_next_comp[4] = { 1, 2, 0, 1 }; 2037 2038 uint best_error = cUINT32_MAX, best_i = 0; 2039 int best_x = 0, best_packed_c1 = 0, best_packed_c2 = 0; 2040 2041 // For each possible 8-bit value, there is a precomputed list of diff/inten/selector configurations that allow that 8-bit value to be encoded with no error. 2042 for (uint i = 0; i < 3; i++) 2043 { 2044 const uint c1 = pColor[s_next_comp[i]], c2 = pColor[s_next_comp[i + 1]]; 2045 2046 const int delta_range = 1; 2047 for (int delta = -delta_range; delta <= delta_range; delta++) 2048 { 2049 const int c_plus_delta = rg_etc1::clamp<int>(pColor[i] + delta, 0, 255); 2050 2051 const uint16* pTable; 2052 if (!c_plus_delta) 2053 pTable = g_color8_to_etc_block_config_0_255[0]; 2054 else if (c_plus_delta == 255) 2055 pTable = g_color8_to_etc_block_config_0_255[1]; 2056 else 2057 pTable = g_color8_to_etc_block_config_1_to_254[c_plus_delta - 1]; 2058 2059 do 2060 { 2061 const uint x = *pTable++; 2062 const uint diff = x & 1; 2063 if (static_cast<uint>(use_diff) != diff) 2064 { 2065 if (*pTable == 0xFFFF) 2066 break; 2067 continue; 2068 } 2069 2070 if ((diff) && (pBase_color5_unscaled)) 2071 { 2072 const int p0 = (x >> 8) & 255; 2073 int delta = p0 - static_cast<int>(pBase_color5_unscaled->c[i]); 2074 if ((delta < cETC1ColorDeltaMin) || (delta > cETC1ColorDeltaMax)) 2075 { 2076 if (*pTable == 0xFFFF) 2077 break; 2078 continue; 2079 } 2080 } 2081 2082 #ifdef RG_ETC1_BUILD_DEBUG 2083 { 2084 const uint inten = (x >> 1) & 7; 2085 const uint selector = (x >> 4) & 3; 2086 const uint p0 = (x >> 8) & 255; 2087 RG_ETC1_ASSERT(etc1_decode_value(diff, inten, selector, p0) == (uint)c_plus_delta); 2088 } 2089 #endif 2090 2091 const uint16* pInverse_table = g_etc1_inverse_lookup[x & 0xFF]; 2092 uint16 p1 = pInverse_table[c1]; 2093 uint16 p2 = pInverse_table[c2]; 2094 2095 if ((diff) && (pBase_color5_unscaled)) 2096 { 2097 int delta1 = (p1 & 0xFF) - static_cast<int>(pBase_color5_unscaled->c[s_next_comp[i]]); 2098 int delta2 = (p2 & 0xFF) - static_cast<int>(pBase_color5_unscaled->c[s_next_comp[i + 1]]); 2099 if ((delta1 < cETC1ColorDeltaMin) || (delta1 > cETC1ColorDeltaMax) || (delta2 < cETC1ColorDeltaMin) || (delta2 > cETC1ColorDeltaMax)) 2100 { 2101 if (*pTable == 0xFFFF) 2102 break; 2103 continue; 2104 } 2105 } 2106 2107 const uint trial_error = rg_etc1::square(c_plus_delta - pColor[i]) + rg_etc1::square(p1 >> 8) + rg_etc1::square(p2 >> 8); 2108 if (trial_error < best_error) 2109 { 2110 best_error = trial_error; 2111 best_x = x; 2112 best_packed_c1 = p1 & 0xFF; 2113 best_packed_c2 = p2 & 0xFF; 2114 best_i = i; 2115 if (!best_error) 2116 goto found_perfect_match; 2117 } 2118 } while (*pTable != 0xFFFF); 2119 } 2120 } 2121 found_perfect_match: 2122 2123 if (best_error == cUINT32_MAX) 2124 return best_error; 2125 2126 best_error *= num_colors; 2127 2128 results.m_n = num_colors; 2129 results.m_block_color4 = !(best_x & 1); 2130 results.m_block_inten_table = (best_x >> 1) & 7; 2131 memset(results.m_pSelectors, (best_x >> 4) & 3, num_colors); 2132 2133 const uint best_packed_c0 = (best_x >> 8) & 255; 2134 results.m_block_color_unscaled[best_i] = static_cast<uint8>(best_packed_c0); 2135 results.m_block_color_unscaled[s_next_comp[best_i]] = static_cast<uint8>(best_packed_c1); 2136 results.m_block_color_unscaled[s_next_comp[best_i + 1]] = static_cast<uint8>(best_packed_c2); 2137 results.m_error = best_error; 2138 2139 return best_error; 2140 } 2141 2142 // Function originally from RYG's public domain real-time DXT1 compressor, modified for 555. dither_block_555(color_quad_u8 * dest,const color_quad_u8 * block)2143 static void dither_block_555(color_quad_u8* dest, const color_quad_u8* block) 2144 { 2145 int err[8],*ep1 = err,*ep2 = err+4; 2146 uint8 *quant = g_quant5_tab+8; 2147 2148 memset(dest, 0xFF, sizeof(color_quad_u8)*16); 2149 2150 // process channels seperately 2151 for(int ch=0;ch<3;ch++) 2152 { 2153 uint8* bp = (uint8*)block; 2154 uint8* dp = (uint8*)dest; 2155 2156 bp += ch; dp += ch; 2157 2158 memset(err,0, sizeof(err)); 2159 for(int y = 0; y < 4; y++) 2160 { 2161 // pixel 0 2162 dp[ 0] = quant[bp[ 0] + ((3*ep2[1] + 5*ep2[0]) >> 4)]; 2163 ep1[0] = bp[ 0] - dp[ 0]; 2164 2165 // pixel 1 2166 dp[ 4] = quant[bp[ 4] + ((7*ep1[0] + 3*ep2[2] + 5*ep2[1] + ep2[0]) >> 4)]; 2167 ep1[1] = bp[ 4] - dp[ 4]; 2168 2169 // pixel 2 2170 dp[ 8] = quant[bp[ 8] + ((7*ep1[1] + 3*ep2[3] + 5*ep2[2] + ep2[1]) >> 4)]; 2171 ep1[2] = bp[ 8] - dp[ 8]; 2172 2173 // pixel 3 2174 dp[12] = quant[bp[12] + ((7*ep1[2] + 5*ep2[3] + ep2[2]) >> 4)]; 2175 ep1[3] = bp[12] - dp[12]; 2176 2177 // advance to next line 2178 int* tmp = ep1; ep1 = ep2; ep2 = tmp; 2179 bp += 16; 2180 dp += 16; 2181 } 2182 } 2183 } 2184 pack_etc1_block(void * pETC1_block,const unsigned int * pSrc_pixels_rgba,etc1_pack_params & pack_params)2185 unsigned int pack_etc1_block(void* pETC1_block, const unsigned int* pSrc_pixels_rgba, etc1_pack_params& pack_params) 2186 { 2187 const color_quad_u8* pSrc_pixels = reinterpret_cast<const color_quad_u8*>(pSrc_pixels_rgba); 2188 etc1_block& dst_block = *static_cast<etc1_block*>(pETC1_block); 2189 2190 #ifdef RG_ETC1_BUILD_DEBUG 2191 // Ensure all alpha values are 0xFF. 2192 for (uint i = 0; i < 16; i++) 2193 { 2194 RG_ETC1_ASSERT(pSrc_pixels[i].a == 255); 2195 } 2196 #endif 2197 2198 color_quad_u8 src_pixel0(pSrc_pixels[0]); 2199 2200 // Check for solid block. 2201 const uint32 first_pixel_u32 = pSrc_pixels->m_u32; 2202 int r; 2203 for (r = 15; r >= 1; --r) 2204 if (pSrc_pixels[r].m_u32 != first_pixel_u32) 2205 break; 2206 if (!r) 2207 return static_cast<unsigned int>(16 * pack_etc1_block_solid_color(dst_block, &pSrc_pixels[0].r, pack_params)); 2208 2209 color_quad_u8 dithered_pixels[16]; 2210 if (pack_params.m_dithering) 2211 { 2212 dither_block_555(dithered_pixels, pSrc_pixels); 2213 pSrc_pixels = dithered_pixels; 2214 } 2215 2216 etc1_optimizer optimizer; 2217 2218 uint64 best_error = cUINT64_MAX; 2219 uint best_flip = false, best_use_color4 = false; 2220 2221 uint8 best_selectors[2][8]; 2222 etc1_optimizer::results best_results[2]; 2223 for (uint i = 0; i < 2; i++) 2224 { 2225 best_results[i].m_n = 8; 2226 best_results[i].m_pSelectors = best_selectors[i]; 2227 } 2228 2229 uint8 selectors[3][8]; 2230 etc1_optimizer::results results[3]; 2231 2232 for (uint i = 0; i < 3; i++) 2233 { 2234 results[i].m_n = 8; 2235 results[i].m_pSelectors = selectors[i]; 2236 } 2237 2238 color_quad_u8 subblock_pixels[8]; 2239 2240 etc1_optimizer::params params(pack_params); 2241 params.m_num_src_pixels = 8; 2242 params.m_pSrc_pixels = subblock_pixels; 2243 2244 for (uint flip = 0; flip < 2; flip++) 2245 { 2246 for (uint use_color4 = 0; use_color4 < 2; use_color4++) 2247 { 2248 uint64 trial_error = 0; 2249 2250 uint subblock; 2251 for (subblock = 0; subblock < 2; subblock++) 2252 { 2253 if (flip) 2254 memcpy(subblock_pixels, pSrc_pixels + subblock * 8, sizeof(color_quad_u8) * 8); 2255 else 2256 { 2257 const color_quad_u8* pSrc_col = pSrc_pixels + subblock * 2; 2258 subblock_pixels[0] = pSrc_col[0]; subblock_pixels[1] = pSrc_col[4]; subblock_pixels[2] = pSrc_col[8]; subblock_pixels[3] = pSrc_col[12]; 2259 subblock_pixels[4] = pSrc_col[1]; subblock_pixels[5] = pSrc_col[5]; subblock_pixels[6] = pSrc_col[9]; subblock_pixels[7] = pSrc_col[13]; 2260 } 2261 2262 results[2].m_error = cUINT64_MAX; 2263 if ((params.m_quality >= cMediumQuality) && ((subblock) || (use_color4))) 2264 { 2265 const uint32 subblock_pixel0_u32 = subblock_pixels[0].m_u32; 2266 for (r = 7; r >= 1; --r) 2267 if (subblock_pixels[r].m_u32 != subblock_pixel0_u32) 2268 break; 2269 if (!r) 2270 { 2271 pack_etc1_block_solid_color_constrained(results[2], 8, &subblock_pixels[0].r, pack_params, !use_color4, (subblock && !use_color4) ? &results[0].m_block_color_unscaled : NULL); 2272 } 2273 } 2274 2275 params.m_use_color4 = (use_color4 != 0); 2276 params.m_constrain_against_base_color5 = false; 2277 2278 if ((!use_color4) && (subblock)) 2279 { 2280 params.m_constrain_against_base_color5 = true; 2281 params.m_base_color5 = results[0].m_block_color_unscaled; 2282 } 2283 2284 if (params.m_quality == cHighQuality) 2285 { 2286 static const int s_scan_delta_0_to_4[] = { -4, -3, -2, -1, 0, 1, 2, 3, 4 }; 2287 params.m_scan_delta_size = RG_ETC1_ARRAY_SIZE(s_scan_delta_0_to_4); 2288 params.m_pScan_deltas = s_scan_delta_0_to_4; 2289 } 2290 else if (params.m_quality == cMediumQuality) 2291 { 2292 static const int s_scan_delta_0_to_1[] = { -1, 0, 1 }; 2293 params.m_scan_delta_size = RG_ETC1_ARRAY_SIZE(s_scan_delta_0_to_1); 2294 params.m_pScan_deltas = s_scan_delta_0_to_1; 2295 } 2296 else 2297 { 2298 static const int s_scan_delta_0[] = { 0 }; 2299 params.m_scan_delta_size = RG_ETC1_ARRAY_SIZE(s_scan_delta_0); 2300 params.m_pScan_deltas = s_scan_delta_0; 2301 } 2302 2303 optimizer.init(params, results[subblock]); 2304 if (!optimizer.compute()) 2305 break; 2306 2307 if (params.m_quality >= cMediumQuality) 2308 { 2309 // TODO: Fix fairly arbitrary/unrefined thresholds that control how far away to scan for potentially better solutions. 2310 const uint refinement_error_thresh0 = 3000; 2311 const uint refinement_error_thresh1 = 6000; 2312 if (results[subblock].m_error > refinement_error_thresh0) 2313 { 2314 if (params.m_quality == cMediumQuality) 2315 { 2316 static const int s_scan_delta_2_to_3[] = { -3, -2, 2, 3 }; 2317 params.m_scan_delta_size = RG_ETC1_ARRAY_SIZE(s_scan_delta_2_to_3); 2318 params.m_pScan_deltas = s_scan_delta_2_to_3; 2319 } 2320 else 2321 { 2322 static const int s_scan_delta_5_to_5[] = { -5, 5 }; 2323 static const int s_scan_delta_5_to_8[] = { -8, -7, -6, -5, 5, 6, 7, 8 }; 2324 if (results[subblock].m_error > refinement_error_thresh1) 2325 { 2326 params.m_scan_delta_size = RG_ETC1_ARRAY_SIZE(s_scan_delta_5_to_8); 2327 params.m_pScan_deltas = s_scan_delta_5_to_8; 2328 } 2329 else 2330 { 2331 params.m_scan_delta_size = RG_ETC1_ARRAY_SIZE(s_scan_delta_5_to_5); 2332 params.m_pScan_deltas = s_scan_delta_5_to_5; 2333 } 2334 } 2335 2336 if (!optimizer.compute()) 2337 break; 2338 } 2339 2340 if (results[2].m_error < results[subblock].m_error) 2341 results[subblock] = results[2]; 2342 } 2343 2344 trial_error += results[subblock].m_error; 2345 if (trial_error >= best_error) 2346 break; 2347 } 2348 2349 if (subblock < 2) 2350 continue; 2351 2352 best_error = trial_error; 2353 best_results[0] = results[0]; 2354 best_results[1] = results[1]; 2355 best_flip = flip; 2356 best_use_color4 = use_color4; 2357 2358 } // use_color4 2359 2360 } // flip 2361 2362 int dr = best_results[1].m_block_color_unscaled.r - best_results[0].m_block_color_unscaled.r; 2363 int dg = best_results[1].m_block_color_unscaled.g - best_results[0].m_block_color_unscaled.g; 2364 int db = best_results[1].m_block_color_unscaled.b - best_results[0].m_block_color_unscaled.b; 2365 RG_ETC1_ASSERT(best_use_color4 || (rg_etc1::minimum(dr, dg, db) >= cETC1ColorDeltaMin) && (rg_etc1::maximum(dr, dg, db) <= cETC1ColorDeltaMax)); 2366 2367 if (best_use_color4) 2368 { 2369 dst_block.m_bytes[0] = static_cast<uint8>(best_results[1].m_block_color_unscaled.r | (best_results[0].m_block_color_unscaled.r << 4)); 2370 dst_block.m_bytes[1] = static_cast<uint8>(best_results[1].m_block_color_unscaled.g | (best_results[0].m_block_color_unscaled.g << 4)); 2371 dst_block.m_bytes[2] = static_cast<uint8>(best_results[1].m_block_color_unscaled.b | (best_results[0].m_block_color_unscaled.b << 4)); 2372 } 2373 else 2374 { 2375 if (dr < 0) dr += 8; dst_block.m_bytes[0] = static_cast<uint8>((best_results[0].m_block_color_unscaled.r << 3) | dr); 2376 if (dg < 0) dg += 8; dst_block.m_bytes[1] = static_cast<uint8>((best_results[0].m_block_color_unscaled.g << 3) | dg); 2377 if (db < 0) db += 8; dst_block.m_bytes[2] = static_cast<uint8>((best_results[0].m_block_color_unscaled.b << 3) | db); 2378 } 2379 2380 dst_block.m_bytes[3] = static_cast<uint8>( (best_results[1].m_block_inten_table << 2) | (best_results[0].m_block_inten_table << 5) | ((~best_use_color4 & 1) << 1) | best_flip ); 2381 2382 uint selector0 = 0, selector1 = 0; 2383 if (best_flip) 2384 { 2385 // flipped: 2386 // { 0, 0 }, { 1, 0 }, { 2, 0 }, { 3, 0 }, 2387 // { 0, 1 }, { 1, 1 }, { 2, 1 }, { 3, 1 } 2388 // 2389 // { 0, 2 }, { 1, 2 }, { 2, 2 }, { 3, 2 }, 2390 // { 0, 3 }, { 1, 3 }, { 2, 3 }, { 3, 3 } 2391 const uint8* pSelectors0 = best_results[0].m_pSelectors; 2392 const uint8* pSelectors1 = best_results[1].m_pSelectors; 2393 for (int x = 3; x >= 0; --x) 2394 { 2395 uint b; 2396 b = g_selector_index_to_etc1[pSelectors1[4 + x]]; 2397 selector0 = (selector0 << 1) | (b & 1); selector1 = (selector1 << 1) | (b >> 1); 2398 2399 b = g_selector_index_to_etc1[pSelectors1[x]]; 2400 selector0 = (selector0 << 1) | (b & 1); selector1 = (selector1 << 1) | (b >> 1); 2401 2402 b = g_selector_index_to_etc1[pSelectors0[4 + x]]; 2403 selector0 = (selector0 << 1) | (b & 1); selector1 = (selector1 << 1) | (b >> 1); 2404 2405 b = g_selector_index_to_etc1[pSelectors0[x]]; 2406 selector0 = (selector0 << 1) | (b & 1); selector1 = (selector1 << 1) | (b >> 1); 2407 } 2408 } 2409 else 2410 { 2411 // non-flipped: 2412 // { 0, 0 }, { 0, 1 }, { 0, 2 }, { 0, 3 }, 2413 // { 1, 0 }, { 1, 1 }, { 1, 2 }, { 1, 3 } 2414 // 2415 // { 2, 0 }, { 2, 1 }, { 2, 2 }, { 2, 3 }, 2416 // { 3, 0 }, { 3, 1 }, { 3, 2 }, { 3, 3 } 2417 for (int subblock = 1; subblock >= 0; --subblock) 2418 { 2419 const uint8* pSelectors = best_results[subblock].m_pSelectors + 4; 2420 for (uint i = 0; i < 2; i++) 2421 { 2422 uint b; 2423 b = g_selector_index_to_etc1[pSelectors[3]]; 2424 selector0 = (selector0 << 1) | (b & 1); selector1 = (selector1 << 1) | (b >> 1); 2425 2426 b = g_selector_index_to_etc1[pSelectors[2]]; 2427 selector0 = (selector0 << 1) | (b & 1); selector1 = (selector1 << 1) | (b >> 1); 2428 2429 b = g_selector_index_to_etc1[pSelectors[1]]; 2430 selector0 = (selector0 << 1) | (b & 1); selector1 = (selector1 << 1) | (b >> 1); 2431 2432 b = g_selector_index_to_etc1[pSelectors[0]]; 2433 selector0 = (selector0 << 1) | (b & 1);selector1 = (selector1 << 1) | (b >> 1); 2434 2435 pSelectors -= 4; 2436 } 2437 } 2438 } 2439 2440 dst_block.m_bytes[4] = static_cast<uint8>(selector1 >> 8); dst_block.m_bytes[5] = static_cast<uint8>(selector1 & 0xFF); 2441 dst_block.m_bytes[6] = static_cast<uint8>(selector0 >> 8); dst_block.m_bytes[7] = static_cast<uint8>(selector0 & 0xFF); 2442 2443 return static_cast<unsigned int>(best_error); 2444 } 2445 2446 } // namespace rg_etc1 2447