src/gallium/auxiliary/util/u_math.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2008 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28
  29 /**
  30  * Math utilities and approximations for common math functions.
  31  * Reduced precision is usually acceptable in shaders...
  32  *
  33  * "fast" is used in the names of functions which are low-precision,
  34  * or at least lower-precision than the normal C lib functions.
  35  */
  36
  37
  38 #ifndef U_MATH_H
  39 #define U_MATH_H
  40
  41
  42 #include "pipe/p_compiler.h"
  43
  44 #include "c99_math.h"
  45 #include <assert.h>
  46 #include <float.h>
  47 #include <stdarg.h>
  48
  49 #ifdef PIPE_OS_UNIX
  50 #include <strings.h> /* for ffs */
  51 #endif
  52
  53 #if defined(_MSC_VER)
  54 #include <intrin.h>
  55 #endif
  56
  57
  58 #ifdef __cplusplus
  59 extern "C" {
  60 #endif
  61
  62
  63 #ifndef M_SQRT2
  64 #define M_SQRT2 1.41421356237309504880
  65 #endif
  66
  67 #define POW2_TABLE_SIZE_LOG2 9
  68 #define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)
  69 #define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)
  70 #define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))
  71 extern float pow2_table[POW2_TABLE_SIZE];
  72
  73
  74 /**
  75  * Initialize math module.  This should be called before using any
  76  * other functions in this module.
  77  */
  78 extern void
  79 util_init_math(void);
  80
  81
  82 union fi {
  83    float f;
  84    int32_t i;
  85    uint32_t ui;
  86 };
  87
  88
  89 union di {
  90    double d;
  91    int64_t i;
  92    uint64_t ui;
  93 };
  94
  95
  96 /**
  97  * Extract the IEEE float32 exponent.
  98  */
  99 static inline signed
 100 util_get_float32_exponent(float x)
 101 {
 102    union fi f;
 103
 104    f.f = x;
 105
 106    return ((f.ui >> 23) & 0xff) - 127;
 107 }
 108
 109
 110 /**
 111  * Fast version of 2^x
 112  * Identity: exp2(a + b) = exp2(a) * exp2(b)
 113  * Let ipart = int(x)
 114  * Let fpart = x - ipart;
 115  * So, exp2(x) = exp2(ipart) * exp2(fpart)
 116  * Compute exp2(ipart) with i << ipart
 117  * Compute exp2(fpart) with lookup table.
 118  */
 119 static inline float
 120 util_fast_exp2(float x)
 121 {
 122    int32_t ipart;
 123    float fpart, mpart;
 124    union fi epart;
 125
 126    if(x > 129.00000f)
 127       return 3.402823466e+38f;
 128
 129    if (x < -126.99999f)
 130       return 0.0f;
 131
 132    ipart = (int32_t) x;
 133    fpart = x - (float) ipart;
 134
 135    /* same as
 136     *   epart.f = (float) (1 << ipart)
 137     * but faster and without integer overflow for ipart > 31
 138     */
 139    epart.i = (ipart + 127 ) << 23;
 140
 141    mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)];
 142
 143    return epart.f * mpart;
 144 }
 145
 146
 147 /**
 148  * Fast approximation to exp(x).
 149  */
 150 static inline float
 151 util_fast_exp(float x)
 152 {
 153    const float k = 1.44269f; /* = log2(e) */
 154    return util_fast_exp2(k * x);
 155 }
 156
 157
 158 #define LOG2_TABLE_SIZE_LOG2 16
 159 #define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)
 160 #define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)
 161 extern float log2_table[LOG2_TABLE_SIZE];
 162
 163
 164 /**
 165  * Fast approximation to log2(x).
 166  */
 167 static inline float
 168 util_fast_log2(float x)
 169 {
 170    union fi num;
 171    float epart, mpart;
 172    num.f = x;
 173    epart = (float)(((num.i & 0x7f800000) >> 23) - 127);
 174    /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */
 175    mpart = log2_table[((num.i & 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2))) >> (23 - LOG2_TABLE_SIZE_LOG2)];
 176    return epart + mpart;
 177 }
 178
 179
 180 /**
 181  * Fast approximation to x^y.
 182  */
 183 static inline float
 184 util_fast_pow(float x, float y)
 185 {
 186    return util_fast_exp2(util_fast_log2(x) * y);
 187 }
 188
 189 /* Note that this counts zero as a power of two.
 190  */
 191 static inline boolean
 192 util_is_power_of_two( unsigned v )
 193 {
 194    return (v & (v-1)) == 0;
 195 }
 196
 197
 198 /**
 199  * Floor(x), returned as int.
 200  */
 201 static inline int
 202 util_ifloor(float f)
 203 {
 204    int ai, bi;
 205    double af, bf;
 206    union fi u;
 207    af = (3 << 22) + 0.5 + (double) f;
 208    bf = (3 << 22) + 0.5 - (double) f;
 209    u.f = (float) af;  ai = u.i;
 210    u.f = (float) bf;  bi = u.i;
 211    return (ai - bi) >> 1;
 212 }
 213
 214
 215 /**
 216  * Round float to nearest int.
 217  */
 218 static inline int
 219 util_iround(float f)
 220 {
 221 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
 222    int r;
 223    __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
 224    return r;
 225 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
 226    int r;
 227    _asm {
 228       fld f
 229       fistp r
 230    }
 231    return r;
 232 #else
 233    if (f >= 0.0f)
 234       return (int) (f + 0.5f);
 235    else
 236       return (int) (f - 0.5f);
 237 #endif
 238 }
 239
 240
 241 /**
 242  * Approximate floating point comparison
 243  */
 244 static inline boolean
 245 util_is_approx(float a, float b, float tol)
 246 {
 247    return fabsf(b - a) <= tol;
 248 }
 249
 250
 251 /**
 252  * util_is_X_inf_or_nan = test if x is NaN or +/- Inf
 253  * util_is_X_nan        = test if x is NaN
 254  * util_X_inf_sign      = return +1 for +Inf, -1 for -Inf, or 0 for not Inf
 255  *
 256  * NaN can be checked with x != x, however this fails with the fast math flag
 257  **/
 258
 259
 260 /**
 261  * Single-float
 262  */
 263 static inline boolean
 264 util_is_inf_or_nan(float x)
 265 {
 266    union fi tmp;
 267    tmp.f = x;
 268    return (tmp.ui & 0x7f800000) == 0x7f800000;
 269 }
 270
 271
 272 static inline boolean
 273 util_is_nan(float x)
 274 {
 275    union fi tmp;
 276    tmp.f = x;
 277    return (tmp.ui & 0x7fffffff) > 0x7f800000;
 278 }
 279
 280
 281 static inline int
 282 util_inf_sign(float x)
 283 {
 284    union fi tmp;
 285    tmp.f = x;
 286    if ((tmp.ui & 0x7fffffff) != 0x7f800000) {
 287       return 0;
 288    }
 289
 290    return (x < 0) ? -1 : 1;
 291 }
 292
 293
 294 /**
 295  * Double-float
 296  */
 297 static inline boolean
 298 util_is_double_inf_or_nan(double x)
 299 {
 300    union di tmp;
 301    tmp.d = x;
 302    return (tmp.ui & 0x7ff0000000000000ULL) == 0x7ff0000000000000ULL;
 303 }
 304
 305
 306 static inline boolean
 307 util_is_double_nan(double x)
 308 {
 309    union di tmp;
 310    tmp.d = x;
 311    return (tmp.ui & 0x7fffffffffffffffULL) > 0x7ff0000000000000ULL;
 312 }
 313
 314
 315 static inline int
 316 util_double_inf_sign(double x)
 317 {
 318    union di tmp;
 319    tmp.d = x;
 320    if ((tmp.ui & 0x7fffffffffffffffULL) != 0x7ff0000000000000ULL) {
 321       return 0;
 322    }
 323
 324    return (x < 0) ? -1 : 1;
 325 }
 326
 327
 328 /**
 329  * Half-float
 330  */
 331 static inline boolean
 332 util_is_half_inf_or_nan(int16_t x)
 333 {
 334    return (x & 0x7c00) == 0x7c00;
 335 }
 336
 337
 338 static inline boolean
 339 util_is_half_nan(int16_t x)
 340 {
 341    return (x & 0x7fff) > 0x7c00;
 342 }
 343
 344
 345 static inline int
 346 util_half_inf_sign(int16_t x)
 347 {
 348    if ((x & 0x7fff) != 0x7c00) {
 349       return 0;
 350    }
 351
 352    return (x < 0) ? -1 : 1;
 353 }
 354
 355
 356 /**
 357  * Find first bit set in word.  Least significant bit is 1.
 358  * Return 0 if no bits set.
 359  */
 360 #ifndef FFS_DEFINED
 361 #define FFS_DEFINED 1
 362
 363 #if defined(_MSC_VER) && (_M_IX86 || _M_AMD64 || _M_IA64)
 364 static inline
 365 unsigned long ffs( unsigned long u )
 366 {
 367    unsigned long i;
 368    if (_BitScanForward(&i, u))
 369       return i + 1;
 370    else
 371       return 0;
 372 }
 373 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
 374 static inline
 375 unsigned ffs( unsigned u )
 376 {
 377    unsigned i;
 378
 379    if (u == 0) {
 380       return 0;
 381    }
 382
 383    __asm bsf eax, [u]
 384    __asm inc eax
 385    __asm mov [i], eax
 386
 387    return i;
 388 }
 389 #elif defined(__MINGW32__) || defined(PIPE_OS_ANDROID) || \
 390     defined(HAVE___BUILTIN_FFS)
 391 #define ffs __builtin_ffs
 392 #endif
 393
 394 #ifdef HAVE___BUILTIN_FFSLL
 395 #define ffsll __builtin_ffsll
 396 #else
 397 static inline int
 398 ffsll(long long int val)
 399 {
 400    int bit;
 401
 402    bit = ffs((unsigned) (val & 0xffffffff));
 403    if (bit != 0)
 404       return bit;
 405
 406    bit = ffs((unsigned) (val >> 32));
 407    if (bit != 0)
 408       return 32 + bit;
 409
 410    return 0;
 411 }
 412 #endif
 413
 414 #endif /* FFS_DEFINED */
 415
 416 /**
 417  * Find first bit set in long long.  Least significant bit is 1.
 418  * Return 0 if no bits set.
 419  */
 420 #ifndef FFSLL_DEFINED
 421 #define FFSLL_DEFINED 1
 422
 423 #if defined(__MINGW32__) || defined(PIPE_OS_ANDROID) || \
 424     defined(HAVE___BUILTIN_FFSLL)
 425 #define ffsll __builtin_ffsll
 426 #endif
 427
 428 #endif /* FFSLL_DEFINED */
 429
 430 /**
 431  * Find last bit set in a word.  The least significant bit is 1.
 432  * Return 0 if no bits are set.
 433  */
 434 static inline unsigned
 435 util_last_bit(unsigned u)
 436 {
 437 #if defined(HAVE___BUILTIN_CLZ)
 438    return u == 0 ? 0 : 32 - __builtin_clz(u);
 439 #else
 440    unsigned r = 0;
 441    while (u) {
 442        r++;
 443        u >>= 1;
 444    }
 445    return r;
 446 #endif
 447 }
 448
 449 /**
 450  * Find last bit set in a word.  The least significant bit is 1.
 451  * Return 0 if no bits are set.
 452  */
 453 static inline unsigned
 454 util_last_bit64(uint64_t u)
 455 {
 456 #if defined(HAVE___BUILTIN_CLZLL)
 457    return u == 0 ? 0 : 64 - __builtin_clzll(u);
 458 #else
 459    unsigned r = 0;
 460    while (u) {
 461        r++;
 462        u >>= 1;
 463    }
 464    return r;
 465 #endif
 466 }
 467
 468 /**
 469  * Find last bit in a word that does not match the sign bit. The least
 470  * significant bit is 1.
 471  * Return 0 if no bits are set.
 472  */
 473 static inline unsigned
 474 util_last_bit_signed(int i)
 475 {
 476    if (i >= 0)
 477       return util_last_bit(i);
 478    else
 479       return util_last_bit(~(unsigned)i);
 480 }
 481
 482 /* Destructively loop over all of the bits in a mask as in:
 483  *
 484  * while (mymask) {
 485  *   int i = u_bit_scan(&mymask);
 486  *   ... process element i
 487  * }
 488  *
 489  */
 490 static inline int
 491 u_bit_scan(unsigned *mask)
 492 {
 493    int i = ffs(*mask) - 1;
 494    *mask &= ~(1u << i);
 495    return i;
 496 }
 497
 498 #ifndef _MSC_VER
 499 static inline int
 500 u_bit_scan64(uint64_t *mask)
 501 {
 502    int i = ffsll(*mask) - 1;
 503    *mask &= ~(1llu << i);
 504    return i;
 505 }
 506 #endif
 507
 508 /* For looping over a bitmask when you want to loop over consecutive bits
 509  * manually, for example:
 510  *
 511  * while (mask) {
 512  *    int start, count, i;
 513  *
 514  *    u_bit_scan_consecutive_range(&mask, &start, &count);
 515  *
 516  *    for (i = 0; i < count; i++)
 517  *       ... process element (start+i)
 518  * }
 519  */
 520 static inline void
 521 u_bit_scan_consecutive_range(unsigned *mask, int *start, int *count)
 522 {
 523    if (*mask == 0xffffffff) {
 524       *start = 0;
 525       *count = 32;
 526       *mask = 0;
 527       return;
 528    }
 529    *start = ffs(*mask) - 1;
 530    *count = ffs(~(*mask >> *start)) - 1;
 531    *mask &= ~(((1u << *count) - 1) << *start);
 532 }
 533
 534 static inline void
 535 u_bit_scan_consecutive_range64(uint64_t *mask, int *start, int *count)
 536 {
 537    if (*mask == ~0llu) {
 538       *start = 0;
 539       *count = 64;
 540       *mask = 0;
 541       return;
 542    }
 543    *start = ffsll(*mask) - 1;
 544    *count = ffsll(~(*mask >> *start)) - 1;
 545    *mask &= ~(((1llu << *count) - 1) << *start);
 546 }
 547
 548 /**
 549  * Return float bits.
 550  */
 551 static inline unsigned
 552 fui( float f )
 553 {
 554    union fi fi;
 555    fi.f = f;
 556    return fi.ui;
 557 }
 558
 559 static inline float
 560 uif(uint32_t ui)
 561 {
 562    union fi fi;
 563    fi.ui = ui;
 564    return fi.f;
 565 }
 566
 567
 568 /**
 569  * Convert ubyte to float in [0, 1].
 570  * XXX a 256-entry lookup table would be slightly faster.
 571  */
 572 static inline float
 573 ubyte_to_float(ubyte ub)
 574 {
 575    return (float) ub * (1.0f / 255.0f);
 576 }
 577
 578
 579 /**
 580  * Convert float in [0,1] to ubyte in [0,255] with clamping.
 581  */
 582 static inline ubyte
 583 float_to_ubyte(float f)
 584 {
 585    union fi tmp;
 586
 587    tmp.f = f;
 588    if (tmp.i < 0) {
 589       return (ubyte) 0;
 590    }
 591    else if (tmp.i >= 0x3f800000 /* 1.0f */) {
 592       return (ubyte) 255;
 593    }
 594    else {
 595       tmp.f = tmp.f * (255.0f/256.0f) + 32768.0f;
 596       return (ubyte) tmp.i;
 597    }
 598 }
 599
 600 static inline float
 601 byte_to_float_tex(int8_t b)
 602 {
 603    return (b == -128) ? -1.0F : b * 1.0F / 127.0F;
 604 }
 605
 606 static inline int8_t
 607 float_to_byte_tex(float f)
 608 {
 609    return (int8_t) (127.0F * f);
 610 }
 611
 612 /**
 613  * Calc log base 2
 614  */
 615 static inline unsigned
 616 util_logbase2(unsigned n)
 617 {
 618 #if defined(HAVE___BUILTIN_CLZ)
 619    return ((sizeof(unsigned) * 8 - 1) - __builtin_clz(n | 1));
 620 #else
 621    unsigned pos = 0;
 622    if (n >= 1<<16) { n >>= 16; pos += 16; }
 623    if (n >= 1<< 8) { n >>=  8; pos +=  8; }
 624    if (n >= 1<< 4) { n >>=  4; pos +=  4; }
 625    if (n >= 1<< 2) { n >>=  2; pos +=  2; }
 626    if (n >= 1<< 1) {           pos +=  1; }
 627    return pos;
 628 #endif
 629 }
 630
 631
 632 /**
 633  * Returns the smallest power of two >= x
 634  */
 635 static inline unsigned
 636 util_next_power_of_two(unsigned x)
 637 {
 638 #if defined(HAVE___BUILTIN_CLZ)
 639    if (x <= 1)
 640        return 1;
 641
 642    return (1 << ((sizeof(unsigned) * 8) - __builtin_clz(x - 1)));
 643 #else
 644    unsigned val = x;
 645
 646    if (x <= 1)
 647       return 1;
 648
 649    if (util_is_power_of_two(x))
 650       return x;
 651
 652    val--;
 653    val = (val >> 1) | val;
 654    val = (val >> 2) | val;
 655    val = (val >> 4) | val;
 656    val = (val >> 8) | val;
 657    val = (val >> 16) | val;
 658    val++;
 659    return val;
 660 #endif
 661 }
 662
 663
 664 /**
 665  * Return number of bits set in n.
 666  */
 667 static inline unsigned
 668 util_bitcount(unsigned n)
 669 {
 670 #if defined(HAVE___BUILTIN_POPCOUNT)
 671    return __builtin_popcount(n);
 672 #else
 673    /* K&R classic bitcount.
 674     *
 675     * For each iteration, clear the LSB from the bitfield.
 676     * Requires only one iteration per set bit, instead of
 677     * one iteration per bit less than highest set bit.
 678     */
 679    unsigned bits;
 680    for (bits = 0; n; bits++) {
 681       n &= n - 1;
 682    }
 683    return bits;
 684 #endif
 685 }
 686
 687
 688 static inline unsigned
 689 util_bitcount64(uint64_t n)
 690 {
 691 #ifdef HAVE___BUILTIN_POPCOUNTLL
 692    return __builtin_popcountll(n);
 693 #else
 694    return util_bitcount(n) + util_bitcount(n >> 32);
 695 #endif
 696 }
 697
 698
 699 /**
 700  * Reverse bits in n
 701  * Algorithm taken from:
 702  * http://stackoverflow.com/questions/9144800/c-reverse-bits-in-unsigned-integer
 703  */
 704 static inline unsigned
 705 util_bitreverse(unsigned n)
 706 {
 707     n = ((n >> 1) & 0x55555555u) | ((n & 0x55555555u) << 1);
 708     n = ((n >> 2) & 0x33333333u) | ((n & 0x33333333u) << 2);
 709     n = ((n >> 4) & 0x0f0f0f0fu) | ((n & 0x0f0f0f0fu) << 4);
 710     n = ((n >> 8) & 0x00ff00ffu) | ((n & 0x00ff00ffu) << 8);
 711     n = ((n >> 16) & 0xffffu) | ((n & 0xffffu) << 16);
 712     return n;
 713 }
 714
 715 /**
 716  * Convert from little endian to CPU byte order.
 717  */
 718
 719 #ifdef PIPE_ARCH_BIG_ENDIAN
 720 #define util_le64_to_cpu(x) util_bswap64(x)
 721 #define util_le32_to_cpu(x) util_bswap32(x)
 722 #define util_le16_to_cpu(x) util_bswap16(x)
 723 #else
 724 #define util_le64_to_cpu(x) (x)
 725 #define util_le32_to_cpu(x) (x)
 726 #define util_le16_to_cpu(x) (x)
 727 #endif
 728
 729 #define util_cpu_to_le64(x) util_le64_to_cpu(x)
 730 #define util_cpu_to_le32(x) util_le32_to_cpu(x)
 731 #define util_cpu_to_le16(x) util_le16_to_cpu(x)
 732
 733 /**
 734  * Reverse byte order of a 32 bit word.
 735  */
 736 static inline uint32_t
 737 util_bswap32(uint32_t n)
 738 {
 739 #if defined(HAVE___BUILTIN_BSWAP32)
 740    return __builtin_bswap32(n);
 741 #else
 742    return (n >> 24) |
 743           ((n >> 8) & 0x0000ff00) |
 744           ((n << 8) & 0x00ff0000) |
 745           (n << 24);
 746 #endif
 747 }
 748
 749 /**
 750  * Reverse byte order of a 64bit word.
 751  */
 752 static inline uint64_t
 753 util_bswap64(uint64_t n)
 754 {
 755 #if defined(HAVE___BUILTIN_BSWAP64)
 756    return __builtin_bswap64(n);
 757 #else
 758    return ((uint64_t)util_bswap32((uint32_t)n) << 32) |
 759           util_bswap32((n >> 32));
 760 #endif
 761 }
 762
 763
 764 /**
 765  * Reverse byte order of a 16 bit word.
 766  */
 767 static inline uint16_t
 768 util_bswap16(uint16_t n)
 769 {
 770    return (n >> 8) |
 771           (n << 8);
 772 }
 773
 774 static inline void*
 775 util_memcpy_cpu_to_le32(void * restrict dest, const void * restrict src, size_t n)
 776 {
 777 #ifdef PIPE_ARCH_BIG_ENDIAN
 778    size_t i, e;
 779    assert(n % 4 == 0);
 780
 781    for (i = 0, e = n / 4; i < e; i++) {
 782       uint32_t * restrict d = (uint32_t* restrict)dest;
 783       const uint32_t * restrict s = (const uint32_t* restrict)src;
 784       d[i] = util_bswap32(s[i]);
 785    }
 786    return dest;
 787 #else
 788    return memcpy(dest, src, n);
 789 #endif
 790 }
 791
 792 /**
 793  * Clamp X to [MIN, MAX].
 794  * This is a macro to allow float, int, uint, etc. types.
 795  */
 796 #define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
 797
 798 #define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
 799 #define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
 800
 801 #define MIN3( A, B, C ) ((A) < (B) ? MIN2(A, C) : MIN2(B, C))
 802 #define MAX3( A, B, C ) ((A) > (B) ? MAX2(A, C) : MAX2(B, C))
 803
 804 #define MIN4( A, B, C, D ) ((A) < (B) ? MIN3(A, C, D) : MIN3(B, C, D))
 805 #define MAX4( A, B, C, D ) ((A) > (B) ? MAX3(A, C, D) : MAX3(B, C, D))
 806
 807
 808 /**
 809  * Align a value, only works pot alignemnts.
 810  */
 811 static inline int
 812 align(int value, int alignment)
 813 {
 814    return (value + alignment - 1) & ~(alignment - 1);
 815 }
 816
 817 static inline uint64_t
 818 align64(uint64_t value, unsigned alignment)
 819 {
 820    return (value + alignment - 1) & ~(alignment - 1);
 821 }
 822
 823 /**
 824  * Works like align but on npot alignments.
 825  */
 826 static inline size_t
 827 util_align_npot(size_t value, size_t alignment)
 828 {
 829    if (value % alignment)
 830       return value + (alignment - (value % alignment));
 831    return value;
 832 }
 833
 834 static inline unsigned
 835 u_minify(unsigned value, unsigned levels)
 836 {
 837     return MAX2(1, value >> levels);
 838 }
 839
 840 #ifndef COPY_4V
 841 #define COPY_4V( DST, SRC )         \
 842 do {                                \
 843    (DST)[0] = (SRC)[0];             \
 844    (DST)[1] = (SRC)[1];             \
 845    (DST)[2] = (SRC)[2];             \
 846    (DST)[3] = (SRC)[3];             \
 847 } while (0)
 848 #endif
 849
 850
 851 #ifndef COPY_4FV
 852 #define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
 853 #endif
 854
 855
 856 #ifndef ASSIGN_4V
 857 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
 858 do {                                     \
 859    (DST)[0] = (V0);                      \
 860    (DST)[1] = (V1);                      \
 861    (DST)[2] = (V2);                      \
 862    (DST)[3] = (V3);                      \
 863 } while (0)
 864 #endif
 865
 866
 867 static inline uint32_t
 868 util_unsigned_fixed(float value, unsigned frac_bits)
 869 {
 870    return value < 0 ? 0 : (uint32_t)(value * (1<<frac_bits));
 871 }
 872
 873 static inline int32_t
 874 util_signed_fixed(float value, unsigned frac_bits)
 875 {
 876    return (int32_t)(value * (1<<frac_bits));
 877 }
 878
 879 unsigned
 880 util_fpstate_get(void);
 881 unsigned
 882 util_fpstate_set_denorms_to_zero(unsigned current_fpstate);
 883 void
 884 util_fpstate_set(unsigned fpstate);
 885
 886
 887
 888 #ifdef __cplusplus
 889 }
 890 #endif
 891
 892 #endif /* U_MATH_H */