| 1 | //************************************ bs::framework - Copyright 2018 Marko Pintera **************************************// |
| 2 | //*********** Licensed under the MIT license. See LICENSE.md for full terms. This notice is not to be removed. ***********// |
| 3 | #pragma once |
| 4 | |
| 5 | #include "Prerequisites/BsPrerequisitesUtil.h" |
| 6 | #include "Math/BsMath.h" |
| 7 | |
| 8 | #if BS_COMPILER == BS_COMPILER_MSVC |
| 9 | #include <intrin.h> |
| 10 | #endif |
| 11 | |
| 12 | namespace bs |
| 13 | { |
| 14 | /** @addtogroup General |
| 15 | * @{ |
| 16 | */ |
| 17 | |
| 18 | /** Floating point number broken down into components for easier access. */ |
| 19 | union Float754 |
| 20 | { |
| 21 | UINT32 raw; |
| 22 | float value; |
| 23 | struct { |
| 24 | #if BS_ENDIAN == BS_ENDIAN_BIG |
| 25 | UINT32 negative : 1; |
| 26 | UINT32 exponent : 8; |
| 27 | UINT32 mantissa : 23; |
| 28 | #else |
| 29 | UINT32 mantissa : 23; |
| 30 | UINT32 exponent : 8; |
| 31 | UINT32 negative : 1; |
| 32 | #endif |
| 33 | } field; |
| 34 | }; |
| 35 | |
| 36 | /** 10-bit floating point number broken down into components for easier access. */ |
| 37 | union Float10 |
| 38 | { |
| 39 | UINT32 raw; |
| 40 | struct { |
| 41 | #if BS_ENDIAN == BS_ENDIAN_BIG |
| 42 | UINT32 exponent : 5; |
| 43 | UINT32 mantissa : 5; |
| 44 | #else |
| 45 | UINT32 mantissa : 5; |
| 46 | UINT32 exponent : 5; |
| 47 | #endif |
| 48 | } field; |
| 49 | }; |
| 50 | |
| 51 | /** 11-bit floating point number broken down into components for easier access. */ |
| 52 | union Float11 |
| 53 | { |
| 54 | UINT32 raw; |
| 55 | struct { |
| 56 | #if BS_ENDIAN == BS_ENDIAN_BIG |
| 57 | UINT32 exponent : 5; |
| 58 | UINT32 mantissa : 6; |
| 59 | #else |
| 60 | UINT32 mantissa : 6; |
| 61 | UINT32 exponent : 5; |
| 62 | #endif |
| 63 | } field; |
| 64 | }; |
| 65 | |
| 66 | /** Class for manipulating bit patterns. */ |
| 67 | class Bitwise |
| 68 | { |
| 69 | public: |
| 70 | /** Returns the power-of-two number greater or equal to the provided value. */ |
| 71 | static UINT32 nextPow2(UINT32 n) |
| 72 | { |
| 73 | --n; |
| 74 | n |= n >> 16; |
| 75 | n |= n >> 8; |
| 76 | n |= n >> 4; |
| 77 | n |= n >> 2; |
| 78 | n |= n >> 1; |
| 79 | ++n; |
| 80 | return n; |
| 81 | } |
| 82 | |
| 83 | /** Returns the power-of-two number closest to the provided value. */ |
| 84 | static UINT32 closestPow2(UINT32 n) |
| 85 | { |
| 86 | UINT32 next = nextPow2(n); |
| 87 | |
| 88 | UINT32 prev = next >> 1; |
| 89 | if (n - prev < next - n) |
| 90 | return prev; |
| 91 | |
| 92 | return next; |
| 93 | } |
| 94 | |
| 95 | |
| 96 | /** Returns base-2 logarithm for common bit counts (8, 16, 32, 64), as a constant expression. */ |
| 97 | static constexpr UINT32 bitsLog2(UINT32 v) |
| 98 | { |
| 99 | switch(v) |
| 100 | { |
| 101 | default: |
| 102 | case 8: return 3; |
| 103 | case 16: return 4; |
| 104 | case 32: return 5; |
| 105 | case 64: return 6; |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | /** Returns modular exponentiation for integers. */ |
| 110 | static UINT32 modPow(UINT32 val1, UINT32 val2, UINT32 t) |
| 111 | { |
| 112 | int res = 1; |
| 113 | |
| 114 | while (val2 != 0) |
| 115 | { |
| 116 | if (val2 & 1) |
| 117 | res = (res * val1) % t; |
| 118 | |
| 119 | val2 >>= 1; |
| 120 | val1 = (val1 * val1) % t; |
| 121 | } |
| 122 | |
| 123 | return res; |
| 124 | } |
| 125 | #if BS_COMPILER == BS_COMPILER_MSVC |
| 126 | #pragma intrinsic(_BitScanReverse,_BitScanForward) |
| 127 | #endif |
| 128 | |
| 129 | /** Finds the most-significant non-zero bit in the provided value and returns the index of that bit. */ |
| 130 | static UINT32 mostSignificantBit(UINT32 val) |
| 131 | { |
| 132 | #if BS_COMPILER == BS_COMPILER_MSVC |
| 133 | unsigned long index; |
| 134 | _BitScanReverse(&index, val); |
| 135 | return index; |
| 136 | #elif BS_COMPILER == BS_COMPILER_GNUC || BS_COMPILER == BS_COMPILER_CLANG |
| 137 | return 31 - __builtin_clz(val); |
| 138 | #else |
| 139 | static_assert(false, "Not implemented" ); |
| 140 | #endif |
| 141 | } |
| 142 | /** Finds the least-significant non-zero bit in the provided value and returns the index of that bit. */ |
| 143 | static UINT32 leastSignificantBit(UINT32 val) |
| 144 | { |
| 145 | #if BS_COMPILER == BS_COMPILER_MSVC |
| 146 | unsigned long index; |
| 147 | _BitScanForward(&index, val); |
| 148 | return index; |
| 149 | #elif BS_COMPILER == BS_COMPILER_GNUC || BS_COMPILER == BS_COMPILER_CLANG |
| 150 | return __builtin_ctz(val); |
| 151 | #else |
| 152 | static_assert(false, "Not implemented" ); |
| 153 | #endif |
| 154 | } |
| 155 | |
| 156 | /** Finds the most-significant non-zero bit in the provided value and returns the index of that bit. */ |
| 157 | static UINT32 mostSignificantBit(UINT64 val) |
| 158 | { |
| 159 | #if BS_COMPILER == BS_COMPILER_MSVC |
| 160 | #if BS_ARCH_TYPE == BS_ARCHITECTURE_x86_64 |
| 161 | unsigned long index; |
| 162 | _BitScanReverse64(&index, val); |
| 163 | return index; |
| 164 | #else // BS_ARCH_TYPE |
| 165 | if (static_cast<UINT32>(val >> 32) != 0) |
| 166 | { |
| 167 | _BitScanReverse(&index, static_cast<UINT32>(val >> 32)); |
| 168 | return index + 32; |
| 169 | } |
| 170 | else |
| 171 | { |
| 172 | _BitScanReverse(&index, static_cast<UINT32>(val)); |
| 173 | return index; |
| 174 | } |
| 175 | #endif // BS_ARCH_TYPE |
| 176 | #elif BS_COMPILER == BS_COMPILER_GNUC || BS_COMPILER == BS_COMPILER_CLANG |
| 177 | return 31 - __builtin_clzll(val); |
| 178 | #else // BS_COMPILER |
| 179 | static_assert(false, "Not implemented" ); |
| 180 | #endif // BS_COMPILER |
| 181 | } |
| 182 | /** Finds the least-significant non-zero bit in the provided value and returns the index of that bit. */ |
| 183 | static UINT32 leastSignificantBit(UINT64 val) |
| 184 | { |
| 185 | #if BS_COMPILER == BS_COMPILER_MSVC |
| 186 | #if BS_ARCH_TYPE == BS_ARCHITECTURE_x86_64 |
| 187 | unsigned long index; |
| 188 | _BitScanForward64(&index, val); |
| 189 | return index; |
| 190 | #else // BS_ARCH_TYPE |
| 191 | if (static_cast<UINT32>(val) != 0) |
| 192 | { |
| 193 | _BitScanForward(&index, static_cast<UINT32>(val)); |
| 194 | return index; |
| 195 | } |
| 196 | else |
| 197 | { |
| 198 | _BitScanForward(&index, static_cast<UINT32>(val >> 32)); |
| 199 | return index + 32; |
| 200 | } |
| 201 | #endif // BS_ARCH_TYPE |
| 202 | #elif BS_COMPILER == BS_COMPILER_GNUC || BS_COMPILER == BS_COMPILER_CLANG |
| 203 | return __builtin_ctzll(val); |
| 204 | #else // BS_COMPILER |
| 205 | static_assert(false, "Not implemented" ); |
| 206 | #endif // BS_COMPILER |
| 207 | } |
| 208 | |
| 209 | /** Determines whether the number is power-of-two or not. */ |
| 210 | template<typename T> |
| 211 | static bool isPow2(T n) |
| 212 | { |
| 213 | return (n & (n - 1)) == 0; |
| 214 | } |
| 215 | |
| 216 | /** Returns the number of bits a pattern must be shifted right by to remove right-hand zeros. */ |
| 217 | template<typename T> |
| 218 | static uint32_t getBitShift(T mask) |
| 219 | { |
| 220 | if (mask == 0) |
| 221 | return 0; |
| 222 | |
| 223 | uint32_t result = 0; |
| 224 | while ((mask & 1) == 0) { |
| 225 | ++result; |
| 226 | mask >>= 1; |
| 227 | } |
| 228 | return result; |
| 229 | } |
| 230 | |
| 231 | /** Count the number of set bits in a mask. */ |
| 232 | static uint32_t countSetBits(uint32_t val) |
| 233 | { |
| 234 | uint32_t count = 0; |
| 235 | for (count = 0; val; count++) |
| 236 | val &= val - 1; |
| 237 | |
| 238 | return count; |
| 239 | } |
| 240 | |
| 241 | /** Takes a value with a given src bit mask, and produces another value with a desired bit mask. */ |
| 242 | template<typename SrcT, typename DestT> |
| 243 | static DestT convertBitPattern(SrcT srcValue, SrcT srcBitMask, DestT destBitMask) |
| 244 | { |
| 245 | // Mask off irrelevant source value bits (if any) |
| 246 | srcValue = srcValue & srcBitMask; |
| 247 | |
| 248 | // Shift source down to bottom of DWORD |
| 249 | const uint32_t srcBitShift = getBitShift(srcBitMask); |
| 250 | srcValue >>= srcBitShift; |
| 251 | |
| 252 | // Get max value possible in source from srcMask |
| 253 | const SrcT srcMax = srcBitMask >> srcBitShift; |
| 254 | |
| 255 | // Get max available in dest |
| 256 | const uint32_t destBitShift = getBitShift(destBitMask); |
| 257 | const DestT destMax = destBitMask >> destBitShift; |
| 258 | |
| 259 | // Scale source value into destination, and shift back |
| 260 | DestT destValue = (srcValue * destMax) / srcMax; |
| 261 | return (destValue << destBitShift); |
| 262 | } |
| 263 | |
| 264 | /** |
| 265 | * Convert N bit color channel value to P bits. It fills P bits with the bit pattern repeated. |
| 266 | * (this is /((1<<n)-1) in fixed point). |
| 267 | */ |
| 268 | static uint32_t fixedToFixed(UINT32 value, uint32_t n, uint32_t p) |
| 269 | { |
| 270 | if (n > p) |
| 271 | { |
| 272 | // Less bits required than available; this is easy |
| 273 | value >>= n - p; |
| 274 | } |
| 275 | else if (n < p) |
| 276 | { |
| 277 | // More bits required than are there, do the fill |
| 278 | // Use old fashioned division, probably better than a loop |
| 279 | if (value == 0) |
| 280 | value = 0; |
| 281 | else if (value == (static_cast<uint32_t>(1) << n) - 1) |
| 282 | value = (1 << p) - 1; |
| 283 | else value = value*(1 << p) / ((1 << n) - 1); |
| 284 | } |
| 285 | return value; |
| 286 | } |
| 287 | |
| 288 | /** |
| 289 | * Converts floating point value in range [0, 1] to an unsigned integer of a certain number of bits. Works for any |
| 290 | * value of bits between 0 and 31. |
| 291 | */ |
| 292 | static uint32_t unormToUint(float value, uint32_t bits) |
| 293 | { |
| 294 | if (value <= 0.0f) return 0; |
| 295 | if (value >= 1.0f) return (1 << bits) - 1; |
| 296 | return Math::roundToInt(value * (1 << bits)); |
| 297 | } |
| 298 | |
| 299 | /** |
| 300 | * Converts floating point value in range [-1, 1] to an unsigned integer of a certain number of bits. Works for any |
| 301 | * value of bits between 0 and 31. |
| 302 | */ |
| 303 | static uint32_t snormToUint(float value, uint32_t bits) |
| 304 | { |
| 305 | return unormToUint((value + 1.0f) * 0.5f, bits); |
| 306 | } |
| 307 | |
| 308 | /** Converts an unsigned integer to a floating point in range [0, 1]. */ |
| 309 | static float uintToUnorm(uint32_t value, uint32_t bits) |
| 310 | { |
| 311 | return (float)value / (float)((1 << bits) - 1); |
| 312 | } |
| 313 | |
| 314 | /** Converts an unsigned integer to a floating point in range [-1, 1]. */ |
| 315 | static float uintToSnorm(uint32_t value, uint32_t bits) |
| 316 | { |
| 317 | return uintToUnorm(value) * 2.0f - 1.0f; |
| 318 | } |
| 319 | |
| 320 | /** |
| 321 | * Converts floating point value in range [0, 1] to an unsigned integer of a certain number of bits. Works for any |
| 322 | * value of bits between 0 and 31. |
| 323 | */ |
| 324 | template<uint32_t bits = 8> |
| 325 | static uint32_t unormToUint(float value) |
| 326 | { |
| 327 | if (value <= 0.0f) return 0; |
| 328 | if (value >= 1.0f) return (1 << bits) - 1; |
| 329 | return Math::roundToInt(value * (1 << bits)); |
| 330 | } |
| 331 | |
| 332 | /** |
| 333 | * Converts floating point value in range [-1, 1] to an unsigned integer of a certain number of bits. Works for any |
| 334 | * value of bits between 0 and 31. |
| 335 | */ |
| 336 | template<uint32_t bits = 8> |
| 337 | static uint32_t snormToUint(float value) |
| 338 | { |
| 339 | return unormToUint<bits>((value + 1.0f) * 0.5f); |
| 340 | } |
| 341 | |
| 342 | /** Converts an unsigned integer to a floating point in range [0, 1]. */ |
| 343 | template<uint32_t bits = 8> |
| 344 | static float uintToUnorm(uint32_t value) |
| 345 | { |
| 346 | return (float)value / (float)((1 << bits) - 1); |
| 347 | } |
| 348 | |
| 349 | /** Converts an unsigned integer to a floating point in range [-1, 1]. */ |
| 350 | template<uint32_t bits = 8> |
| 351 | static float uintToSnorm(uint32_t value) |
| 352 | { |
| 353 | return uintToUnorm<bits>(value) * 2.0f - 1.0f; |
| 354 | } |
| 355 | |
| 356 | /** |
| 357 | * Interpolates between two values using the @p t parameter. All parameters must be in [0, 255] range. When @p t |
| 358 | * is zero, @p from value will be returned, and when it is 255 @p to value will be returned, and interpolation |
| 359 | * between @p from and @p to will occurr for in-between values. |
| 360 | */ |
| 361 | static uint32_t lerpByte(uint32_t from, uint32_t to, uint32_t t) |
| 362 | { |
| 363 | assert((from & 0xFF) == from); |
| 364 | assert((to & 0xFF) == to); |
| 365 | assert((t & 0xFF) == t); |
| 366 | assert(from <= to); |
| 367 | |
| 368 | return (from + (((to - from) * t) >> 8)) & 0xFF; |
| 369 | } |
| 370 | |
| 371 | /** |
| 372 | * Interpolates between two values using the @p t parameter. All parameters must be in [0, 65536] range. When @p t |
| 373 | * is zero, @p from value will be returned, and when it is 65536 @p to value will be returned, and interpolation |
| 374 | * between @p from and @p to will occurr for in-between values. |
| 375 | */ |
| 376 | static uint32_t lerpWord(uint32_t from, uint32_t to, uint32_t t) |
| 377 | { |
| 378 | assert((from & 0xFFFF) == from); |
| 379 | assert((to & 0xFFFF) == to); |
| 380 | assert((t & 0xFFFF) == t); |
| 381 | assert(from <= to); |
| 382 | |
| 383 | return (from + (((to - from) * t) >> 16)) & 0xFFFF; |
| 384 | } |
| 385 | |
| 386 | /** |
| 387 | * Determines the position of the @p val parameter in the [from, to] range, returned in [0, 255] range where 0 is |
| 388 | * returned if @p val is less or equal than @p from, and 255 is returned if @p val is equal to greater to @p to, |
| 389 | * and in-between values returned accordingly. All values must be in [0, 255] range. |
| 390 | */ |
| 391 | static uint32_t invLerpByte(uint32_t from, uint32_t to, uint32_t val) |
| 392 | { |
| 393 | assert((from & 0xFF) == from); |
| 394 | assert((to & 0xFF) == to); |
| 395 | assert((val & 0xFF) == val); |
| 396 | assert(from <= to); |
| 397 | |
| 398 | return ((val - from) << 8) / std::max(to - from, 1u); |
| 399 | } |
| 400 | |
| 401 | /** |
| 402 | * Determines the position of the @p val parameter in the [from, to] range, returned in [0, 65536] range where 0 is |
| 403 | * returned if @p val is less or equal than @p from, and 65536 is returned if @p val is equal to greater to @p to, |
| 404 | * and in-between values returned accordingly. All values must be in [0, 65536] range. |
| 405 | */ |
| 406 | static uint32_t invLerpWord(uint32_t from, uint32_t to, uint32_t val) |
| 407 | { |
| 408 | assert((from & 0xFFFF) == from); |
| 409 | assert((to & 0xFFFF) == to); |
| 410 | assert((val & 0xFFFF) == val); |
| 411 | assert(from <= to); |
| 412 | |
| 413 | return ((val - from) << 16) / std::max(to - from, 1u); |
| 414 | } |
| 415 | |
| 416 | /** Write a n*8 bits integer value to memory in native endian. */ |
| 417 | static void intWrite(void *dest, const int32_t n, const uint32_t value) |
| 418 | { |
| 419 | switch(n) { |
| 420 | case 1: |
| 421 | ((UINT8*)dest)[0] = (UINT8)value; |
| 422 | break; |
| 423 | case 2: |
| 424 | ((UINT16*)dest)[0] = (UINT16)value; |
| 425 | break; |
| 426 | case 3: |
| 427 | #if BS_ENDIAN == BS_ENDIAN_BIG |
| 428 | ((UINT8*)dest)[0] = (UINT8)((value >> 16) & 0xFF); |
| 429 | ((UINT8*)dest)[1] = (UINT8)((value >> 8) & 0xFF); |
| 430 | ((UINT8*)dest)[2] = (UINT8)(value & 0xFF); |
| 431 | #else |
| 432 | ((UINT8*)dest)[2] = (UINT8)((value >> 16) & 0xFF); |
| 433 | ((UINT8*)dest)[1] = (UINT8)((value >> 8) & 0xFF); |
| 434 | ((UINT8*)dest)[0] = (UINT8)(value & 0xFF); |
| 435 | #endif |
| 436 | break; |
| 437 | case 4: |
| 438 | ((UINT32*)dest)[0] = (UINT32)value; |
| 439 | break; |
| 440 | } |
| 441 | } |
| 442 | |
| 443 | /** Read a n*8 bits integer value to memory in native endian. */ |
| 444 | static uint32_t intRead(const void *src, int32_t n) { |
| 445 | switch(n) { |
| 446 | case 1: |
| 447 | return ((UINT8*)src)[0]; |
| 448 | case 2: |
| 449 | return ((UINT16*)src)[0]; |
| 450 | case 3: |
| 451 | #if BS_ENDIAN == BS_ENDIAN_BIG |
| 452 | return ((UINT32)((UINT8*)src)[0]<<16)| |
| 453 | ((UINT32)((UINT8*)src)[1]<<8)| |
| 454 | ((UINT32)((UINT8*)src)[2]); |
| 455 | #else |
| 456 | return ((UINT32)((UINT8*)src)[0])| |
| 457 | ((UINT32)((UINT8*)src)[1]<<8)| |
| 458 | ((UINT32)((UINT8*)src)[2]<<16); |
| 459 | #endif |
| 460 | case 4: |
| 461 | return ((UINT32*)src)[0]; |
| 462 | } |
| 463 | return 0; // ? |
| 464 | } |
| 465 | |
| 466 | /** Convert a float32 to a float16 (NV_half_float). */ |
| 467 | static UINT16 floatToHalf(float i) |
| 468 | { |
| 469 | union { float f; UINT32 i; } v; |
| 470 | v.f = i; |
| 471 | return floatToHalfI(v.i); |
| 472 | } |
| 473 | |
| 474 | /** Converts float in UINT32 format to a a half in UINT16 format. */ |
| 475 | static UINT16 floatToHalfI(UINT32 i) |
| 476 | { |
| 477 | int32_t s = (i >> 16) & 0x00008000; |
| 478 | int32_t e = ((i >> 23) & 0x000000ff) - (127 - 15); |
| 479 | int32_t m = i & 0x007fffff; |
| 480 | |
| 481 | if (e <= 0) |
| 482 | { |
| 483 | if (e < -10) |
| 484 | { |
| 485 | return 0; |
| 486 | } |
| 487 | m = (m | 0x00800000) >> (1 - e); |
| 488 | |
| 489 | return static_cast<UINT16>(s | (m >> 13)); |
| 490 | } |
| 491 | else if (e == 0xff - (127 - 15)) |
| 492 | { |
| 493 | if (m == 0) // Inf |
| 494 | { |
| 495 | return static_cast<UINT16>(s | 0x7c00); |
| 496 | } |
| 497 | else // NAN |
| 498 | { |
| 499 | m >>= 13; |
| 500 | return static_cast<UINT16>(s | 0x7c00 | m | (m == 0)); |
| 501 | } |
| 502 | } |
| 503 | else |
| 504 | { |
| 505 | if (e > 30) // Overflow |
| 506 | { |
| 507 | return static_cast<UINT16>(s | 0x7c00); |
| 508 | } |
| 509 | |
| 510 | return static_cast<UINT16>(s | (e << 10) | (m >> 13)); |
| 511 | } |
| 512 | } |
| 513 | |
| 514 | /** Convert a float16 (NV_half_float) to a float32. */ |
| 515 | static float halfToFloat(UINT16 y) |
| 516 | { |
| 517 | union { float f; UINT32 i; } v; |
| 518 | v.i = halfToFloatI(y); |
| 519 | return v.f; |
| 520 | } |
| 521 | |
| 522 | /** Converts a half in UINT16 format to a float in UINT32 format. */ |
| 523 | static UINT32 halfToFloatI(UINT16 y) |
| 524 | { |
| 525 | int32_t s = (y >> 15) & 0x00000001; |
| 526 | int32_t e = (y >> 10) & 0x0000001f; |
| 527 | int32_t m = y & 0x000003ff; |
| 528 | |
| 529 | if (e == 0) |
| 530 | { |
| 531 | if (m == 0) // Plus or minus zero |
| 532 | { |
| 533 | return s << 31; |
| 534 | } |
| 535 | else // Denormalized number -- renormalize it |
| 536 | { |
| 537 | while (!(m & 0x00000400)) |
| 538 | { |
| 539 | m <<= 1; |
| 540 | e -= 1; |
| 541 | } |
| 542 | |
| 543 | e += 1; |
| 544 | m &= ~0x00000400; |
| 545 | } |
| 546 | } |
| 547 | else if (e == 31) |
| 548 | { |
| 549 | if (m == 0) // Inf |
| 550 | { |
| 551 | return (s << 31) | 0x7f800000; |
| 552 | } |
| 553 | else // NaN |
| 554 | { |
| 555 | return (s << 31) | 0x7f800000 | (m << 13); |
| 556 | } |
| 557 | } |
| 558 | |
| 559 | e = e + (127 - 15); |
| 560 | m = m << 13; |
| 561 | |
| 562 | return (s << 31) | (e << 23) | m; |
| 563 | } |
| 564 | |
| 565 | /** Converts a 32-bit float to a 10-bit float according to OpenGL packed_float extension. */ |
| 566 | static UINT32 floatToFloat10(float v) |
| 567 | { |
| 568 | Float754 f; |
| 569 | f.value = v; |
| 570 | |
| 571 | if (f.field.exponent == 0xFF) |
| 572 | { |
| 573 | // NAN or INF |
| 574 | if (f.field.mantissa > 0) |
| 575 | return 0x3E0 | (((f.raw >> 18) | (f.raw >> 13) | (f.raw >> 3) | f.raw) & 0x1F); |
| 576 | else if (f.field.negative) |
| 577 | return 0; // Negative infinity clamped to 0 |
| 578 | else |
| 579 | return 0x3E0; // Positive infinity |
| 580 | |
| 581 | } |
| 582 | else if (f.field.negative) |
| 583 | return 0; // Negative clamped to 0, no negatives allowed |
| 584 | else if (f.raw > 0x477C0000) |
| 585 | return 0x3DF; // Too large, clamp to max value |
| 586 | else |
| 587 | { |
| 588 | UINT32 val; |
| 589 | if (f.raw < 0x38800000U) |
| 590 | { |
| 591 | // Too small to be represented as a normalized float, convert to denormalized value |
| 592 | UINT32 shift = 113 - f.field.exponent; |
| 593 | val = (0x800000U | f.field.mantissa) >> shift; |
| 594 | } |
| 595 | else |
| 596 | { |
| 597 | // Rebias exponent |
| 598 | val = f.raw + 0xC8000000; |
| 599 | } |
| 600 | |
| 601 | return ((val + 0x1FFFFU + ((val >> 18) & 1)) >> 18) & 0x3FF; |
| 602 | } |
| 603 | } |
| 604 | |
| 605 | /** Converts a 32-bit float to a 11-bit float according to OpenGL packed_float extension. */ |
| 606 | static UINT32 floatToFloat11(float v) |
| 607 | { |
| 608 | Float754 f; |
| 609 | f.value = v; |
| 610 | |
| 611 | if (f.field.exponent == 0xFF) |
| 612 | { |
| 613 | // NAN or INF |
| 614 | if (f.field.mantissa > 0) |
| 615 | return 0x7C0 | (((f.raw >> 17) | (f.raw >> 11) | (f.raw >> 6) | f.raw) & 0x3F); |
| 616 | else if (f.field.negative) |
| 617 | return 0; // Negative infinity clamped to 0 |
| 618 | else |
| 619 | return 0x7C0; // Positive infinity |
| 620 | |
| 621 | } |
| 622 | else if (f.field.negative) |
| 623 | return 0; // Negative clamped to 0, no negatives allowed |
| 624 | else if (f.raw > 0x477E0000) |
| 625 | return 0x7BF; // Too large, clamp to max value |
| 626 | else |
| 627 | { |
| 628 | UINT32 val; |
| 629 | if(f.raw < 0x38800000U) |
| 630 | { |
| 631 | // Too small to be represented as a normalized float, convert to denormalized value |
| 632 | UINT32 shift = 113 - f.field.exponent; |
| 633 | val = (0x800000U | f.field.mantissa) >> shift; |
| 634 | } |
| 635 | else |
| 636 | { |
| 637 | // Rebias exponent |
| 638 | val = f.raw + 0xC8000000; |
| 639 | } |
| 640 | |
| 641 | return ((val + 0xFFFFU + ((val >> 17) & 1)) >> 17) & 0x7FF; |
| 642 | } |
| 643 | } |
| 644 | |
| 645 | /** Converts a 10-bit float to a 32-bit float according to OpenGL packed_float extension. */ |
| 646 | static float float10ToFloat(UINT32 v) |
| 647 | { |
| 648 | Float10 f; |
| 649 | f.raw = v; |
| 650 | |
| 651 | UINT32 output; |
| 652 | if (f.field.exponent == 0x1F) // INF or NAN |
| 653 | { |
| 654 | output = 0x7f800000 | (f.field.mantissa << 17); |
| 655 | } |
| 656 | else |
| 657 | { |
| 658 | UINT32 exponent; |
| 659 | UINT32 mantissa = f.field.mantissa; |
| 660 | |
| 661 | if (f.field.exponent != 0) // The value is normalized |
| 662 | exponent = f.field.exponent; |
| 663 | else if (mantissa != 0) // The value is denormalized |
| 664 | { |
| 665 | // Normalize the value in the resulting float |
| 666 | exponent = 1; |
| 667 | |
| 668 | do |
| 669 | { |
| 670 | exponent--; |
| 671 | mantissa <<= 1; |
| 672 | } while ((mantissa & 0x20) == 0); |
| 673 | |
| 674 | mantissa &= 0x1F; |
| 675 | } |
| 676 | else // The value is zero |
| 677 | exponent = (UINT32)-112; |
| 678 | |
| 679 | output = ((exponent + 112) << 23) | (mantissa << 18); |
| 680 | } |
| 681 | |
| 682 | return *(float*)&output; |
| 683 | } |
| 684 | |
| 685 | /** Converts a 11-bit float to a 32-bit float according to OpenGL packed_float extension. */ |
| 686 | static float float11ToFloat(UINT32 v) |
| 687 | { |
| 688 | Float11 f; |
| 689 | f.raw = v; |
| 690 | |
| 691 | UINT32 output; |
| 692 | if (f.field.exponent == 0x1F) // INF or NAN |
| 693 | { |
| 694 | output = 0x7f800000 | (f.field.mantissa << 17); |
| 695 | } |
| 696 | else |
| 697 | { |
| 698 | UINT32 exponent; |
| 699 | UINT32 mantissa = f.field.mantissa; |
| 700 | |
| 701 | if (f.field.exponent != 0) // The value is normalized |
| 702 | exponent = f.field.exponent; |
| 703 | else if (mantissa != 0) // The value is denormalized |
| 704 | { |
| 705 | // Normalize the value in the resulting float |
| 706 | exponent = 1; |
| 707 | |
| 708 | do |
| 709 | { |
| 710 | exponent--; |
| 711 | mantissa <<= 1; |
| 712 | } while ((mantissa & 0x40) == 0); |
| 713 | |
| 714 | mantissa &= 0x3F; |
| 715 | } |
| 716 | else // The value is zero |
| 717 | exponent = (UINT32)-112; |
| 718 | |
| 719 | output = ((exponent + 112) << 23) | (mantissa << 17); |
| 720 | } |
| 721 | |
| 722 | return *(float*)&output; |
| 723 | } |
| 724 | |
| 725 | /** |
| 726 | * Encodes a 32-bit integer value as a base-128 varint. Varints are a method of serializing integers using one or |
| 727 | * more bytes, where smaller values use less bytes. |
| 728 | * |
| 729 | * @param[in] value Value to encode. |
| 730 | * @param[out] output Buffer to store the encoded bytes in. Must be at least 5 bytes in length. |
| 731 | * @return Number of bytes required to store the value, in range [1, 5] |
| 732 | */ |
| 733 | static UINT32 encodeVarInt(UINT32 value, UINT8* output) |
| 734 | { |
| 735 | UINT32 idx = 0; |
| 736 | if (value & 0xFFFFFF80U) |
| 737 | { |
| 738 | output[idx++] = (UINT8)(value | 0x80); |
| 739 | value >>= 7; |
| 740 | |
| 741 | if (value & 0xFFFFFF80U) |
| 742 | { |
| 743 | output[idx++] = (UINT8)(value | 0x80); |
| 744 | value >>= 7; |
| 745 | |
| 746 | if (value & 0xFFFFFF80U) |
| 747 | { |
| 748 | output[idx++] = (UINT8)(value | 0x80); |
| 749 | value >>= 7; |
| 750 | |
| 751 | if (value & 0xFFFFFF80U) |
| 752 | { |
| 753 | output[idx++] = (UINT8)(value | 0x80); |
| 754 | value >>= 7; |
| 755 | } |
| 756 | } |
| 757 | } |
| 758 | } |
| 759 | |
| 760 | output[idx++] = (UINT8)value; |
| 761 | return idx; |
| 762 | } |
| 763 | |
| 764 | /** |
| 765 | * Decodes a value encoded using encodeVarInt(UINT32, UINT8*). |
| 766 | * |
| 767 | * @param[out] value Variable to receive the decoded value. |
| 768 | * @param[in] input Input buffer to decode the data from. |
| 769 | * @param[in] size Size of the input buffer. |
| 770 | * @return Number of bytes read. |
| 771 | */ |
| 772 | static UINT32 decodeVarInt(UINT32& value, const UINT8* input, UINT32 size) |
| 773 | { |
| 774 | if(size == 0) |
| 775 | return 0; |
| 776 | |
| 777 | UINT32 idx = 0; |
| 778 | value = (UINT32)(input[idx] & 0x7F); |
| 779 | if (input[idx++] & 0x80 && --size) |
| 780 | { |
| 781 | value |= (UINT32)(input[idx] & 0x7F) << 7; |
| 782 | |
| 783 | if (input[idx++] & 0x80 && --size) |
| 784 | { |
| 785 | value |= (UINT32)(input[idx] & 0x7F) << 14; |
| 786 | |
| 787 | if (input[idx++] & 0x80 && --size) |
| 788 | { |
| 789 | value |= (UINT32)(input[idx] & 0x7F) << 21; |
| 790 | |
| 791 | if (input[idx++] & 0x80 && --size) |
| 792 | value |= (UINT32)(input[idx++]) << 28; |
| 793 | } |
| 794 | } |
| 795 | } |
| 796 | |
| 797 | return !size || input[idx - 1] & 0x80 ? 0 : idx; |
| 798 | } |
| 799 | |
| 800 | /** @copydoc encodeVarInt(UINT32, UINT8*) */ |
| 801 | static UINT32 encodeVarInt(INT32 value, UINT8* output) |
| 802 | { |
| 803 | // Encode using zig-zag pattern so that negative values don't take up max byte count |
| 804 | UINT32 temp = (value << 1) ^ (value >> 31); |
| 805 | return encodeVarInt(temp, output); |
| 806 | } |
| 807 | |
| 808 | /** @copydoc decodeVarInt(UINT32, UINT8*) */ |
| 809 | static UINT32 decodeVarInt(INT32& value, const UINT8* input, UINT32 size) |
| 810 | { |
| 811 | UINT32 temp; |
| 812 | |
| 813 | UINT32 readBytes = decodeVarInt(temp, input, size); |
| 814 | value = (INT32)((temp >> 1) ^ -((INT32)temp & 1)); |
| 815 | |
| 816 | return readBytes; |
| 817 | } |
| 818 | |
| 819 | /** |
| 820 | * Encodes a 64-bit integer value as a base-128 varint. Varints are a method of serializing integers using one or |
| 821 | * more bytes, where smaller values use less bytes. |
| 822 | * |
| 823 | * @param[in] value Value to encode. |
| 824 | * @param[out] output Buffer to store the encoded bytes in. Must be at least 10 bytes in length. |
| 825 | * @return Number of bytes required to store the value, in range [1, 10] |
| 826 | */ |
| 827 | static UINT32 encodeVarInt(UINT64 value, UINT8* output) |
| 828 | { |
| 829 | UINT32 idx = 0; |
| 830 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 831 | { |
| 832 | output[idx++] = (UINT8)(value | 0x80); |
| 833 | value >>= 7; |
| 834 | |
| 835 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 836 | { |
| 837 | output[idx++] = (UINT8)(value | 0x80); |
| 838 | value >>= 7; |
| 839 | |
| 840 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 841 | { |
| 842 | output[idx++] = (UINT8)(value | 0x80); |
| 843 | value >>= 7; |
| 844 | |
| 845 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 846 | { |
| 847 | output[idx++] = (UINT8)(value | 0x80); |
| 848 | value >>= 7; |
| 849 | |
| 850 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 851 | { |
| 852 | output[idx++] = (UINT8)(value | 0x80); |
| 853 | value >>= 7; |
| 854 | |
| 855 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 856 | { |
| 857 | output[idx++] = (UINT8)(value | 0x80); |
| 858 | value >>= 7; |
| 859 | |
| 860 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 861 | { |
| 862 | output[idx++] = (UINT8)(value | 0x80); |
| 863 | value >>= 7; |
| 864 | |
| 865 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 866 | { |
| 867 | output[idx++] = (UINT8)(value | 0x80); |
| 868 | value >>= 7; |
| 869 | |
| 870 | if (value & 0xFFFFFFFFFFFFFF80ULL) |
| 871 | { |
| 872 | output[idx++] = (UINT8)(value | 0x80); |
| 873 | value >>= 7; |
| 874 | } |
| 875 | } |
| 876 | } |
| 877 | } |
| 878 | } |
| 879 | } |
| 880 | } |
| 881 | } |
| 882 | } |
| 883 | |
| 884 | output[idx++] = (UINT8)value; |
| 885 | return idx; |
| 886 | } |
| 887 | |
| 888 | /** |
| 889 | * Decodes a value encoded using encodeVarInt(UINT64, UINT8*). |
| 890 | * |
| 891 | * @param[out] value Variable to receive the decoded value. |
| 892 | * @param[in] input Input buffer to decode the data from. |
| 893 | * @param[in] size Size of the input buffer. |
| 894 | * @return Number of bytes read. |
| 895 | */ |
| 896 | static UINT32 decodeVarInt(UINT64& value, const UINT8* input, UINT32 size) |
| 897 | { |
| 898 | if(size == 0) |
| 899 | return 0; |
| 900 | |
| 901 | UINT32 idx = 0; |
| 902 | value = (UINT64)(input[idx] & 0x7F); |
| 903 | if (input[idx++] & 0x80 && --size) |
| 904 | { |
| 905 | value |= (UINT64)(input[idx] & 0x7F) << 7; |
| 906 | |
| 907 | if (input[idx++] & 0x80 && --size) |
| 908 | { |
| 909 | value |= (UINT64)(input[idx] & 0x7F) << 14; |
| 910 | |
| 911 | if (input[idx++] & 0x80 && --size) |
| 912 | { |
| 913 | value |= (UINT64)(input[idx] & 0x7F) << 21; |
| 914 | |
| 915 | if (input[idx++] & 0x80 && --size) |
| 916 | { |
| 917 | value |= (UINT64)(input[idx] & 0x7F) << 28; |
| 918 | |
| 919 | if (input[idx++] & 0x80 && --size) |
| 920 | { |
| 921 | value |= (UINT64)(input[idx] & 0x7F) << 35; |
| 922 | |
| 923 | if (input[idx++] & 0x80 && --size) |
| 924 | { |
| 925 | value |= (UINT64)(input[idx] & 0x7F) << 42; |
| 926 | |
| 927 | if (input[idx++] & 0x80 && --size) |
| 928 | { |
| 929 | value |= (UINT64)(input[idx] & 0x7F) << 49; |
| 930 | |
| 931 | if (input[idx++] & 0x80 && --size) |
| 932 | { |
| 933 | value |= (UINT64)(input[idx] & 0x7F) << 56; |
| 934 | |
| 935 | if (input[idx++] & 0x80 && --size) |
| 936 | value |= (UINT64)(input[idx++]) << 63; |
| 937 | } |
| 938 | } |
| 939 | } |
| 940 | } |
| 941 | } |
| 942 | } |
| 943 | } |
| 944 | } |
| 945 | |
| 946 | return !size || input[idx - 1] & 0x80 ? 0 : idx; |
| 947 | } |
| 948 | |
| 949 | /** @copydoc encodeVarInt(UINT64, UINT8*) */ |
| 950 | static UINT32 encodeVarInt(INT64 value, UINT8* output) |
| 951 | { |
| 952 | // Encode using zig-zag pattern so that negative values don't take up max byte count |
| 953 | UINT64 temp = (value << 1) ^ (value >> 63); |
| 954 | return encodeVarInt(temp, output); |
| 955 | } |
| 956 | |
| 957 | /** @copydoc decodeVarInt(UINT64, UINT8*) */ |
| 958 | static UINT32 decodeVarInt(INT64& value, const UINT8* input, UINT32 size) |
| 959 | { |
| 960 | UINT64 temp; |
| 961 | |
| 962 | UINT32 readBytes = decodeVarInt(temp, input, size); |
| 963 | value = (INT64)((temp >> 1) ^ -((INT64)temp & 1)); |
| 964 | |
| 965 | return readBytes; |
| 966 | } |
| 967 | |
| 968 | /** Converts a float in range [-1,1] into an unsigned 8-bit integer. */ |
| 969 | static UINT8 quantize8BitSigned(float v) |
| 970 | { |
| 971 | return quantize8BitUnsigned(v * 0.5f + 0.5f); |
| 972 | } |
| 973 | |
| 974 | /** Converts a float in range [0,1] into an unsigned 8-bit integer. */ |
| 975 | static UINT8 quantize8BitUnsigned(float v) |
| 976 | { |
| 977 | return (UINT8)(v * 255.999f); |
| 978 | } |
| 979 | }; |
| 980 | |
| 981 | /** @} */ |
| 982 | } |
| 983 | |