| 1 | // basisu.h |
| 2 | // Copyright (C) 2019-2021 Binomial LLC. All Rights Reserved. |
| 3 | // Important: If compiling with gcc, be sure strict aliasing is disabled: -fno-strict-aliasing |
| 4 | // |
| 5 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 6 | // you may not use this file except in compliance with the License. |
| 7 | // You may obtain a copy of the License at |
| 8 | // |
| 9 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | // |
| 11 | // Unless required by applicable law or agreed to in writing, software |
| 12 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | // See the License for the specific language governing permissions and |
| 15 | // limitations under the License. |
| 16 | #pragma once |
| 17 | |
| 18 | #ifdef _MSC_VER |
| 19 | |
| 20 | #pragma warning (disable : 4201) |
| 21 | #pragma warning (disable : 4127) // warning C4127: conditional expression is constant |
| 22 | #pragma warning (disable : 4530) // C++ exception handler used, but unwind semantics are not enabled. |
| 23 | |
| 24 | // Slamming this off always for v1.16 because we've gotten rid of most std containers. |
| 25 | #ifndef BASISU_NO_ITERATOR_DEBUG_LEVEL |
| 26 | #define BASISU_NO_ITERATOR_DEBUG_LEVEL (1) |
| 27 | #endif |
| 28 | |
| 29 | #ifndef BASISU_NO_ITERATOR_DEBUG_LEVEL |
| 30 | //#define _HAS_ITERATOR_DEBUGGING 0 |
| 31 | |
| 32 | #if defined(_DEBUG) || defined(DEBUG) |
| 33 | // This is madness, but we need to disable iterator debugging in debug builds or the encoder is unsable because MSVC's iterator debugging implementation is totally broken. |
| 34 | #ifndef _ITERATOR_DEBUG_LEVEL |
| 35 | #define _ITERATOR_DEBUG_LEVEL 1 |
| 36 | #endif |
| 37 | #ifndef _SECURE_SCL |
| 38 | #define _SECURE_SCL 1 |
| 39 | #endif |
| 40 | #else // defined(_DEBUG) || defined(DEBUG) |
| 41 | #ifndef _SECURE_SCL |
| 42 | #define _SECURE_SCL 0 |
| 43 | #endif |
| 44 | #ifndef _ITERATOR_DEBUG_LEVEL |
| 45 | #define _ITERATOR_DEBUG_LEVEL 0 |
| 46 | #endif |
| 47 | #endif // defined(_DEBUG) || defined(DEBUG) |
| 48 | |
| 49 | #endif // BASISU_NO_ITERATOR_DEBUG_LEVEL |
| 50 | |
| 51 | #endif // _MSC_VER |
| 52 | |
| 53 | #include <stdlib.h> |
| 54 | #include <stdio.h> |
| 55 | #include <math.h> |
| 56 | #include <stdarg.h> |
| 57 | #include <string.h> |
| 58 | #include <memory.h> |
| 59 | #include <limits.h> |
| 60 | #include <stdint.h> |
| 61 | |
| 62 | #include <algorithm> |
| 63 | #include <limits> |
| 64 | #include <functional> |
| 65 | #include <iterator> |
| 66 | #include <type_traits> |
| 67 | #include <assert.h> |
| 68 | #include <random> |
| 69 | |
| 70 | #include "basisu_containers.h" |
| 71 | |
| 72 | #ifdef max |
| 73 | #undef max |
| 74 | #endif |
| 75 | |
| 76 | #ifdef min |
| 77 | #undef min |
| 78 | #endif |
| 79 | |
| 80 | #ifdef _WIN32 |
| 81 | #define strcasecmp _stricmp |
| 82 | #endif |
| 83 | |
| 84 | // Set to one to enable debug printf()'s when any errors occur, for development/debugging. Especially useful for WebGL development. |
| 85 | #ifndef BASISU_FORCE_DEVEL_MESSAGES |
| 86 | #define BASISU_FORCE_DEVEL_MESSAGES 0 |
| 87 | #endif |
| 88 | |
| 89 | #define BASISU_NOTE_UNUSED(x) (void)(x) |
| 90 | #define BASISU_ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0])) |
| 91 | #define BASISU_NO_EQUALS_OR_COPY_CONSTRUCT(x) x(const x &) = delete; x& operator= (const x &) = delete; |
| 92 | #define BASISU_ASSUME(x) static_assert(x, #x); |
| 93 | #define BASISU_OFFSETOF(s, m) offsetof(s, m) |
| 94 | #define BASISU_STRINGIZE(x) #x |
| 95 | #define BASISU_STRINGIZE2(x) BASISU_STRINGIZE(x) |
| 96 | |
| 97 | #if BASISU_FORCE_DEVEL_MESSAGES |
| 98 | #define BASISU_DEVEL_ERROR(...) do { basisu::debug_printf(__VA_ARGS__); } while(0) |
| 99 | #else |
| 100 | #define BASISU_DEVEL_ERROR(...) |
| 101 | #endif |
| 102 | |
| 103 | namespace basisu |
| 104 | { |
| 105 | // Types/utilities |
| 106 | |
| 107 | #ifdef _WIN32 |
| 108 | const char BASISU_PATH_SEPERATOR_CHAR = '\\'; |
| 109 | #else |
| 110 | const char BASISU_PATH_SEPERATOR_CHAR = '/'; |
| 111 | #endif |
| 112 | |
| 113 | typedef basisu::vector<uint8_t> uint8_vec; |
| 114 | typedef basisu::vector<int16_t> int16_vec; |
| 115 | typedef basisu::vector<uint16_t> uint16_vec; |
| 116 | typedef basisu::vector<uint32_t> uint_vec; |
| 117 | typedef basisu::vector<uint64_t> uint64_vec; |
| 118 | typedef basisu::vector<int> int_vec; |
| 119 | typedef basisu::vector<bool> bool_vec; |
| 120 | |
| 121 | void enable_debug_printf(bool enabled); |
| 122 | void debug_printf(const char *pFmt, ...); |
| 123 | |
| 124 | |
| 125 | template <typename T> inline void clear_obj(T& obj) { memset(&obj, 0, sizeof(obj)); } |
| 126 | |
| 127 | template <typename T0, typename T1> inline T0 lerp(T0 a, T0 b, T1 c) { return a + (b - a) * c; } |
| 128 | |
| 129 | template <typename S> inline S maximum(S a, S b) { return (a > b) ? a : b; } |
| 130 | template <typename S> inline S maximum(S a, S b, S c) { return maximum(maximum(a, b), c); } |
| 131 | template <typename S> inline S maximum(S a, S b, S c, S d) { return maximum(maximum(maximum(a, b), c), d); } |
| 132 | |
| 133 | template <typename S> inline S minimum(S a, S b) { return (a < b) ? a : b; } |
| 134 | template <typename S> inline S minimum(S a, S b, S c) { return minimum(minimum(a, b), c); } |
| 135 | template <typename S> inline S minimum(S a, S b, S c, S d) { return minimum(minimum(minimum(a, b), c), d); } |
| 136 | |
| 137 | inline float clampf(float value, float low, float high) { if (value < low) value = low; else if (value > high) value = high; return value; } |
| 138 | inline float saturate(float value) { return clampf(value, 0, 1.0f); } |
| 139 | inline uint8_t minimumub(uint8_t a, uint8_t b) { return (a < b) ? a : b; } |
| 140 | inline uint32_t minimumu(uint32_t a, uint32_t b) { return (a < b) ? a : b; } |
| 141 | inline int32_t minimumi(int32_t a, int32_t b) { return (a < b) ? a : b; } |
| 142 | inline float minimumf(float a, float b) { return (a < b) ? a : b; } |
| 143 | inline uint8_t maximumub(uint8_t a, uint8_t b) { return (a > b) ? a : b; } |
| 144 | inline uint32_t maximumu(uint32_t a, uint32_t b) { return (a > b) ? a : b; } |
| 145 | inline int32_t maximumi(int32_t a, int32_t b) { return (a > b) ? a : b; } |
| 146 | inline float maximumf(float a, float b) { return (a > b) ? a : b; } |
| 147 | inline int squarei(int i) { return i * i; } |
| 148 | inline float squaref(float i) { return i * i; } |
| 149 | template<typename T> inline T square(T a) { return a * a; } |
| 150 | |
| 151 | template <typename S> inline S clamp(S value, S low, S high) { return (value < low) ? low : ((value > high) ? high : value); } |
| 152 | |
| 153 | inline uint32_t iabs(int32_t i) { return (i < 0) ? static_cast<uint32_t>(-i) : static_cast<uint32_t>(i); } |
| 154 | inline uint64_t iabs64(int64_t i) { return (i < 0) ? static_cast<uint64_t>(-i) : static_cast<uint64_t>(i); } |
| 155 | |
| 156 | template<typename T> inline void clear_vector(T &vec) { vec.erase(vec.begin(), vec.end()); } |
| 157 | template<typename T> inline typename T::value_type *enlarge_vector(T &vec, size_t n) { size_t cs = vec.size(); vec.resize(cs + n); return &vec[cs]; } |
| 158 | |
| 159 | inline bool is_pow2(uint32_t x) { return x && ((x & (x - 1U)) == 0U); } |
| 160 | inline bool is_pow2(uint64_t x) { return x && ((x & (x - 1U)) == 0U); } |
| 161 | |
| 162 | template<typename T> inline T open_range_check(T v, T minv, T maxv) { assert(v >= minv && v < maxv); BASISU_NOTE_UNUSED(minv); BASISU_NOTE_UNUSED(maxv); return v; } |
| 163 | template<typename T> inline T open_range_check(T v, T maxv) { assert(v < maxv); BASISU_NOTE_UNUSED(maxv); return v; } |
| 164 | |
| 165 | inline uint32_t total_bits(uint32_t v) { uint32_t l = 0; for ( ; v > 0U; ++l) v >>= 1; return l; } |
| 166 | |
| 167 | template<typename T> inline T saturate(T val) { return clamp(val, 0.0f, 1.0f); } |
| 168 | |
| 169 | template<typename T, typename R> inline void append_vector(T &vec, const R *pObjs, size_t n) |
| 170 | { |
| 171 | if (n) |
| 172 | { |
| 173 | if (vec.size()) |
| 174 | { |
| 175 | assert((pObjs + n) <= vec.begin() || (pObjs >= vec.end())); |
| 176 | } |
| 177 | const size_t cur_s = vec.size(); |
| 178 | vec.resize(cur_s + n); |
| 179 | memcpy(&vec[cur_s], pObjs, sizeof(R) * n); |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | template<typename T> inline void append_vector(T &vec, const T &other_vec) |
| 184 | { |
| 185 | assert(&vec != &other_vec); |
| 186 | if (other_vec.size()) |
| 187 | append_vector(vec, &other_vec[0], other_vec.size()); |
| 188 | } |
| 189 | |
| 190 | template<typename T> inline void vector_ensure_element_is_valid(T &vec, size_t idx) |
| 191 | { |
| 192 | if (idx >= vec.size()) |
| 193 | vec.resize(idx + 1); |
| 194 | } |
| 195 | |
| 196 | template<typename T> inline void vector_sort(T &vec) |
| 197 | { |
| 198 | if (vec.size()) |
| 199 | std::sort(vec.begin(), vec.end()); |
| 200 | } |
| 201 | |
| 202 | template<typename T, typename U> inline bool unordered_set_contains(T& set, const U&obj) |
| 203 | { |
| 204 | return set.find(obj) != set.end(); |
| 205 | } |
| 206 | |
| 207 | template<typename T> int vector_find(const T &vec, const typename T::value_type &obj) |
| 208 | { |
| 209 | assert(vec.size() <= INT_MAX); |
| 210 | for (size_t i = 0; i < vec.size(); i++) |
| 211 | if (vec[i] == obj) |
| 212 | return static_cast<int>(i); |
| 213 | return -1; |
| 214 | } |
| 215 | |
| 216 | template<typename T> void vector_set_all(T &vec, const typename T::value_type &obj) |
| 217 | { |
| 218 | for (size_t i = 0; i < vec.size(); i++) |
| 219 | vec[i] = obj; |
| 220 | } |
| 221 | |
| 222 | inline uint64_t read_be64(const void *p) |
| 223 | { |
| 224 | uint64_t val = 0; |
| 225 | for (uint32_t i = 0; i < 8; i++) |
| 226 | val |= (static_cast<uint64_t>(static_cast<const uint8_t *>(p)[7 - i]) << (i * 8)); |
| 227 | return val; |
| 228 | } |
| 229 | |
| 230 | inline void write_be64(void *p, uint64_t x) |
| 231 | { |
| 232 | for (uint32_t i = 0; i < 8; i++) |
| 233 | static_cast<uint8_t *>(p)[7 - i] = static_cast<uint8_t>(x >> (i * 8)); |
| 234 | } |
| 235 | |
| 236 | static inline uint16_t byteswap16(uint16_t x) { return static_cast<uint16_t>((x << 8) | (x >> 8)); } |
| 237 | static inline uint32_t byteswap32(uint32_t x) { return ((x << 24) | ((x << 8) & 0x00FF0000) | ((x >> 8) & 0x0000FF00) | (x >> 24)); } |
| 238 | |
| 239 | inline uint32_t floor_log2i(uint32_t v) |
| 240 | { |
| 241 | uint32_t b = 0; |
| 242 | for (; v > 1U; ++b) |
| 243 | v >>= 1; |
| 244 | return b; |
| 245 | } |
| 246 | |
| 247 | inline uint32_t ceil_log2i(uint32_t v) |
| 248 | { |
| 249 | uint32_t b = floor_log2i(v); |
| 250 | if ((b != 32) && (v > (1U << b))) |
| 251 | ++b; |
| 252 | return b; |
| 253 | } |
| 254 | |
| 255 | inline int posmod(int x, int y) |
| 256 | { |
| 257 | if (x >= 0) |
| 258 | return (x < y) ? x : (x % y); |
| 259 | int m = (-x) % y; |
| 260 | return (m != 0) ? (y - m) : m; |
| 261 | } |
| 262 | |
| 263 | inline bool do_excl_ranges_overlap(int la, int ha, int lb, int hb) |
| 264 | { |
| 265 | assert(la < ha && lb < hb); |
| 266 | if ((ha <= lb) || (la >= hb)) return false; |
| 267 | return true; |
| 268 | } |
| 269 | |
| 270 | static inline uint32_t read_le_dword(const uint8_t *pBytes) |
| 271 | { |
| 272 | return (pBytes[3] << 24U) | (pBytes[2] << 16U) | (pBytes[1] << 8U) | (pBytes[0]); |
| 273 | } |
| 274 | |
| 275 | static inline void write_le_dword(uint8_t* pBytes, uint32_t val) |
| 276 | { |
| 277 | pBytes[0] = (uint8_t)val; |
| 278 | pBytes[1] = (uint8_t)(val >> 8U); |
| 279 | pBytes[2] = (uint8_t)(val >> 16U); |
| 280 | pBytes[3] = (uint8_t)(val >> 24U); |
| 281 | } |
| 282 | |
| 283 | // Always little endian 1-8 byte unsigned int |
| 284 | template<uint32_t NumBytes> |
| 285 | struct packed_uint |
| 286 | { |
| 287 | uint8_t m_bytes[NumBytes]; |
| 288 | |
| 289 | inline packed_uint() { static_assert(NumBytes <= sizeof(uint64_t), "Invalid NumBytes" ); } |
| 290 | inline packed_uint(uint64_t v) { *this = v; } |
| 291 | inline packed_uint(const packed_uint& other) { *this = other; } |
| 292 | |
| 293 | inline packed_uint& operator= (uint64_t v) |
| 294 | { |
| 295 | for (uint32_t i = 0; i < NumBytes; i++) |
| 296 | m_bytes[i] = static_cast<uint8_t>(v >> (i * 8)); |
| 297 | return *this; |
| 298 | } |
| 299 | |
| 300 | inline packed_uint& operator= (const packed_uint& rhs) |
| 301 | { |
| 302 | memcpy(m_bytes, rhs.m_bytes, sizeof(m_bytes)); |
| 303 | return *this; |
| 304 | } |
| 305 | |
| 306 | inline operator uint32_t() const |
| 307 | { |
| 308 | switch (NumBytes) |
| 309 | { |
| 310 | case 1: |
| 311 | { |
| 312 | return m_bytes[0]; |
| 313 | } |
| 314 | case 2: |
| 315 | { |
| 316 | return (m_bytes[1] << 8U) | m_bytes[0]; |
| 317 | } |
| 318 | case 3: |
| 319 | { |
| 320 | return (m_bytes[2] << 16U) | (m_bytes[1] << 8U) | m_bytes[0]; |
| 321 | } |
| 322 | case 4: |
| 323 | { |
| 324 | return read_le_dword(m_bytes); |
| 325 | } |
| 326 | case 5: |
| 327 | { |
| 328 | uint32_t l = read_le_dword(m_bytes); |
| 329 | uint32_t h = m_bytes[4]; |
| 330 | return static_cast<uint64_t>(l) | (static_cast<uint64_t>(h) << 32U); |
| 331 | } |
| 332 | case 6: |
| 333 | { |
| 334 | uint32_t l = read_le_dword(m_bytes); |
| 335 | uint32_t h = (m_bytes[5] << 8U) | m_bytes[4]; |
| 336 | return static_cast<uint64_t>(l) | (static_cast<uint64_t>(h) << 32U); |
| 337 | } |
| 338 | case 7: |
| 339 | { |
| 340 | uint32_t l = read_le_dword(m_bytes); |
| 341 | uint32_t h = (m_bytes[6] << 16U) | (m_bytes[5] << 8U) | m_bytes[4]; |
| 342 | return static_cast<uint64_t>(l) | (static_cast<uint64_t>(h) << 32U); |
| 343 | } |
| 344 | case 8: |
| 345 | { |
| 346 | uint32_t l = read_le_dword(m_bytes); |
| 347 | uint32_t h = read_le_dword(m_bytes + 4); |
| 348 | return static_cast<uint64_t>(l) | (static_cast<uint64_t>(h) << 32U); |
| 349 | } |
| 350 | default: |
| 351 | { |
| 352 | assert(0); |
| 353 | return 0; |
| 354 | } |
| 355 | } |
| 356 | } |
| 357 | }; |
| 358 | |
| 359 | enum eZero { cZero }; |
| 360 | enum eNoClamp { cNoClamp }; |
| 361 | |
| 362 | // Rice/Huffman entropy coding |
| 363 | |
| 364 | // This is basically Deflate-style canonical Huffman, except we allow for a lot more symbols. |
| 365 | enum |
| 366 | { |
| 367 | cHuffmanMaxSupportedCodeSize = 16, cHuffmanMaxSupportedInternalCodeSize = 31, |
| 368 | cHuffmanFastLookupBits = 10, |
| 369 | cHuffmanMaxSymsLog2 = 14, cHuffmanMaxSyms = 1 << cHuffmanMaxSymsLog2, |
| 370 | |
| 371 | // Small zero runs |
| 372 | cHuffmanSmallZeroRunSizeMin = 3, cHuffmanSmallZeroRunSizeMax = 10, = 3, |
| 373 | |
| 374 | // Big zero run |
| 375 | cHuffmanBigZeroRunSizeMin = 11, cHuffmanBigZeroRunSizeMax = 138, = 7, |
| 376 | |
| 377 | // Small non-zero run |
| 378 | cHuffmanSmallRepeatSizeMin = 3, cHuffmanSmallRepeatSizeMax = 6, = 2, |
| 379 | |
| 380 | // Big non-zero run |
| 381 | cHuffmanBigRepeatSizeMin = 7, cHuffmanBigRepeatSizeMax = 134, = 7, |
| 382 | |
| 383 | cHuffmanTotalCodelengthCodes = 21, cHuffmanSmallZeroRunCode = 17, cHuffmanBigZeroRunCode = 18, cHuffmanSmallRepeatCode = 19, cHuffmanBigRepeatCode = 20 |
| 384 | }; |
| 385 | |
| 386 | static const uint8_t g_huffman_sorted_codelength_codes[] = { cHuffmanSmallZeroRunCode, cHuffmanBigZeroRunCode, cHuffmanSmallRepeatCode, cHuffmanBigRepeatCode, 0, 8, 7, 9, 6, 0xA, 5, 0xB, 4, 0xC, 3, 0xD, 2, 0xE, 1, 0xF, 0x10 }; |
| 387 | const uint32_t cHuffmanTotalSortedCodelengthCodes = sizeof(g_huffman_sorted_codelength_codes) / sizeof(g_huffman_sorted_codelength_codes[0]); |
| 388 | |
| 389 | // GPU texture formats |
| 390 | |
| 391 | enum class texture_format |
| 392 | { |
| 393 | cInvalidTextureFormat = -1, |
| 394 | |
| 395 | // Block-based formats |
| 396 | cETC1, // ETC1 |
| 397 | cETC1S, // ETC1 (subset: diff colors only, no subblocks) |
| 398 | cETC2_RGB, // ETC2 color block (basisu doesn't support ETC2 planar/T/H modes - just basic ETC1) |
| 399 | cETC2_RGBA, // ETC2 EAC alpha block followed by ETC2 color block |
| 400 | cETC2_ALPHA, // ETC2 EAC alpha block |
| 401 | cBC1, // DXT1 |
| 402 | cBC3, // DXT5 (BC4/DXT5A block followed by a BC1/DXT1 block) |
| 403 | cBC4, // DXT5A |
| 404 | cBC5, // 3DC/DXN (two BC4/DXT5A blocks) |
| 405 | cBC7, |
| 406 | cASTC4x4, // LDR only |
| 407 | cPVRTC1_4_RGB, |
| 408 | cPVRTC1_4_RGBA, |
| 409 | cATC_RGB, |
| 410 | cATC_RGBA_INTERPOLATED_ALPHA, |
| 411 | cFXT1_RGB, |
| 412 | cPVRTC2_4_RGBA, |
| 413 | cETC2_R11_EAC, |
| 414 | cETC2_RG11_EAC, |
| 415 | cUASTC4x4, |
| 416 | cBC1_NV, |
| 417 | cBC1_AMD, |
| 418 | |
| 419 | // Uncompressed/raw pixels |
| 420 | cRGBA32, |
| 421 | cRGB565, |
| 422 | cBGR565, |
| 423 | cRGBA4444, |
| 424 | cABGR4444 |
| 425 | }; |
| 426 | |
| 427 | inline uint32_t get_bytes_per_block(texture_format fmt) |
| 428 | { |
| 429 | switch (fmt) |
| 430 | { |
| 431 | case texture_format::cETC1: |
| 432 | case texture_format::cETC1S: |
| 433 | case texture_format::cETC2_RGB: |
| 434 | case texture_format::cETC2_ALPHA: |
| 435 | case texture_format::cBC1: |
| 436 | case texture_format::cBC1_NV: |
| 437 | case texture_format::cBC1_AMD: |
| 438 | case texture_format::cBC4: |
| 439 | case texture_format::cPVRTC1_4_RGB: |
| 440 | case texture_format::cPVRTC1_4_RGBA: |
| 441 | case texture_format::cATC_RGB: |
| 442 | case texture_format::cPVRTC2_4_RGBA: |
| 443 | case texture_format::cETC2_R11_EAC: |
| 444 | return 8; |
| 445 | case texture_format::cRGBA32: |
| 446 | return sizeof(uint32_t) * 16; |
| 447 | default: |
| 448 | break; |
| 449 | } |
| 450 | return 16; |
| 451 | } |
| 452 | |
| 453 | inline uint32_t get_qwords_per_block(texture_format fmt) |
| 454 | { |
| 455 | return get_bytes_per_block(fmt) >> 3; |
| 456 | } |
| 457 | |
| 458 | inline uint32_t get_block_width(texture_format fmt) |
| 459 | { |
| 460 | BASISU_NOTE_UNUSED(fmt); |
| 461 | switch (fmt) |
| 462 | { |
| 463 | case texture_format::cFXT1_RGB: |
| 464 | return 8; |
| 465 | default: |
| 466 | break; |
| 467 | } |
| 468 | return 4; |
| 469 | } |
| 470 | |
| 471 | inline uint32_t get_block_height(texture_format fmt) |
| 472 | { |
| 473 | BASISU_NOTE_UNUSED(fmt); |
| 474 | return 4; |
| 475 | } |
| 476 | |
| 477 | } // namespace basisu |
| 478 | |
| 479 | |