src/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 "c99_math.h"
  43 #include <assert.h>
  44 #include <float.h>
  45 #include <stdarg.h>
  46
  47 #include "bitscan.h"
  48 #include "u_endian.h" /* for UTIL_ARCH_BIG_ENDIAN */
  49
  50 #ifdef __cplusplus
  51 extern "C" {
  52 #endif
  53
  54
  55 #ifndef M_SQRT2
  56 #define M_SQRT2 1.41421356237309504880
  57 #endif
  58
  59 #define POW2_TABLE_SIZE_LOG2 9
  60 #define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)
  61 #define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)
  62 #define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))
  63 extern float pow2_table[POW2_TABLE_SIZE];
  64
  65
  66 /**
  67  * Initialize math module.  This should be called before using any
  68  * other functions in this module.
  69  */
  70 extern void
  71 util_init_math(void);
  72
  73
  74 union fi {
  75    float f;
  76    int32_t i;
  77    uint32_t ui;
  78 };
  79
  80
  81 union di {
  82    double d;
  83    int64_t i;
  84    uint64_t ui;
  85 };
  86
  87
  88 /**
  89  * Extract the IEEE float32 exponent.
  90  */
  91 static inline signed
  92 util_get_float32_exponent(float x)
  93 {
  94    union fi f;
  95
  96    f.f = x;
  97
  98    return ((f.ui >> 23) & 0xff) - 127;
  99 }
 100
 101
 102 /**
 103  * Fast version of 2^x
 104  * Identity: exp2(a + b) = exp2(a) * exp2(b)
 105  * Let ipart = int(x)
 106  * Let fpart = x - ipart;
 107  * So, exp2(x) = exp2(ipart) * exp2(fpart)
 108  * Compute exp2(ipart) with i << ipart
 109  * Compute exp2(fpart) with lookup table.
 110  */
 111 static inline float
 112 util_fast_exp2(float x)
 113 {
 114    int32_t ipart;
 115    float fpart, mpart;
 116    union fi epart;
 117
 118    if(x > 129.00000f)
 119       return 3.402823466e+38f;
 120
 121    if (x < -126.99999f)
 122       return 0.0f;
 123
 124    ipart = (int32_t) x;
 125    fpart = x - (float) ipart;
 126
 127    /* same as
 128     *   epart.f = (float) (1 << ipart)
 129     * but faster and without integer overflow for ipart > 31
 130     */
 131    epart.i = (ipart + 127 ) << 23;
 132
 133    mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)];
 134
 135    return epart.f * mpart;
 136 }
 137
 138
 139 /**
 140  * Fast approximation to exp(x).
 141  */
 142 static inline float
 143 util_fast_exp(float x)
 144 {
 145    const float k = 1.44269f; /* = log2(e) */
 146    return util_fast_exp2(k * x);
 147 }
 148
 149
 150 #define LOG2_TABLE_SIZE_LOG2 16
 151 #define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)
 152 #define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)
 153 extern float log2_table[LOG2_TABLE_SIZE];
 154
 155
 156 /**
 157  * Fast approximation to log2(x).
 158  */
 159 static inline float
 160 util_fast_log2(float x)
 161 {
 162    union fi num;
 163    float epart, mpart;
 164    num.f = x;
 165    epart = (float)(((num.i & 0x7f800000) >> 23) - 127);
 166    /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */
 167    mpart = log2_table[((num.i & 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2))) >> (23 - LOG2_TABLE_SIZE_LOG2)];
 168    return epart + mpart;
 169 }
 170
 171
 172 /**
 173  * Fast approximation to x^y.
 174  */
 175 static inline float
 176 util_fast_pow(float x, float y)
 177 {
 178    return util_fast_exp2(util_fast_log2(x) * y);
 179 }
 180
 181
 182 /**
 183  * Floor(x), returned as int.
 184  */
 185 static inline int
 186 util_ifloor(float f)
 187 {
 188 #if defined(USE_X86_ASM) && defined(__GNUC__) && defined(__i386__)
 189    /*
 190     * IEEE floor for computers that round to nearest or even.
 191     * 'f' must be between -4194304 and 4194303.
 192     * This floor operation is done by "(iround(f + .5) + iround(f - .5)) >> 1",
 193     * but uses some IEEE specific tricks for better speed.
 194     * Contributed by Josh Vanderhoof
 195     */
 196    int ai, bi;
 197    double af, bf;
 198    af = (3 << 22) + 0.5 + (double)f;
 199    bf = (3 << 22) + 0.5 - (double)f;
 200    /* GCC generates an extra fstp/fld without this. */
 201    __asm__ ("fstps %0" : "=m" (ai) : "t" (af) : "st");
 202    __asm__ ("fstps %0" : "=m" (bi) : "t" (bf) : "st");
 203    return (ai - bi) >> 1;
 204 #else
 205    int ai, bi;
 206    double af, bf;
 207    union fi u;
 208    af = (3 << 22) + 0.5 + (double) f;
 209    bf = (3 << 22) + 0.5 - (double) f;
 210    u.f = (float) af;  ai = u.i;
 211    u.f = (float) bf;  bi = u.i;
 212    return (ai - bi) >> 1;
 213 #endif
 214 }
 215
 216
 217 /**
 218  * Round float to nearest int.
 219  */
 220 static inline int
 221 util_iround(float f)
 222 {
 223 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
 224    int r;
 225    __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
 226    return r;
 227 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
 228    int r;
 229    _asm {
 230       fld f
 231       fistp r
 232    }
 233    return r;
 234 #else
 235    if (f >= 0.0f)
 236       return (int) (f + 0.5f);
 237    else
 238       return (int) (f - 0.5f);
 239 #endif
 240 }
 241
 242
 243 /**
 244  * Approximate floating point comparison
 245  */
 246 static inline bool
 247 util_is_approx(float a, float b, float tol)
 248 {
 249    return fabsf(b - a) <= tol;
 250 }
 251
 252
 253 /**
 254  * util_is_X_inf_or_nan = test if x is NaN or +/- Inf
 255  * util_is_X_nan        = test if x is NaN
 256  * util_X_inf_sign      = return +1 for +Inf, -1 for -Inf, or 0 for not Inf
 257  *
 258  * NaN can be checked with x != x, however this fails with the fast math flag
 259  **/
 260
 261
 262 /**
 263  * Single-float
 264  */
 265 static inline bool
 266 util_is_inf_or_nan(float x)
 267 {
 268    union fi tmp;
 269    tmp.f = x;
 270    return (tmp.ui & 0x7f800000) == 0x7f800000;
 271 }
 272
 273
 274 static inline bool
 275 util_is_nan(float x)
 276 {
 277    union fi tmp;
 278    tmp.f = x;
 279    return (tmp.ui & 0x7fffffff) > 0x7f800000;
 280 }
 281
 282
 283 static inline int
 284 util_inf_sign(float x)
 285 {
 286    union fi tmp;
 287    tmp.f = x;
 288    if ((tmp.ui & 0x7fffffff) != 0x7f800000) {
 289       return 0;
 290    }
 291
 292    return (x < 0) ? -1 : 1;
 293 }
 294
 295
 296 /**
 297  * Double-float
 298  */
 299 static inline bool
 300 util_is_double_inf_or_nan(double x)
 301 {
 302    union di tmp;
 303    tmp.d = x;
 304    return (tmp.ui & 0x7ff0000000000000ULL) == 0x7ff0000000000000ULL;
 305 }
 306
 307
 308 static inline bool
 309 util_is_double_nan(double x)
 310 {
 311    union di tmp;
 312    tmp.d = x;
 313    return (tmp.ui & 0x7fffffffffffffffULL) > 0x7ff0000000000000ULL;
 314 }
 315
 316
 317 static inline int
 318 util_double_inf_sign(double x)
 319 {
 320    union di tmp;
 321    tmp.d = x;
 322    if ((tmp.ui & 0x7fffffffffffffffULL) != 0x7ff0000000000000ULL) {
 323       return 0;
 324    }
 325
 326    return (x < 0) ? -1 : 1;
 327 }
 328
 329
 330 /**
 331  * Half-float
 332  */
 333 static inline bool
 334 util_is_half_inf_or_nan(int16_t x)
 335 {
 336    return (x & 0x7c00) == 0x7c00;
 337 }
 338
 339
 340 static inline bool
 341 util_is_half_nan(int16_t x)
 342 {
 343    return (x & 0x7fff) > 0x7c00;
 344 }
 345
 346
 347 static inline int
 348 util_half_inf_sign(int16_t x)
 349 {
 350    if ((x & 0x7fff) != 0x7c00) {
 351       return 0;
 352    }
 353
 354    return (x < 0) ? -1 : 1;
 355 }
 356
 357
 358 /**
 359  * Return float bits.
 360  */
 361 static inline unsigned
 362 fui( float f )
 363 {
 364    union fi fi;
 365    fi.f = f;
 366    return fi.ui;
 367 }
 368
 369 static inline float
 370 uif(uint32_t ui)
 371 {
 372    union fi fi;
 373    fi.ui = ui;
 374    return fi.f;
 375 }
 376
 377
 378 /**
 379  * Convert uint8_t to float in [0, 1].
 380  */
 381 static inline float
 382 ubyte_to_float(uint8_t ub)
 383 {
 384    return (float) ub * (1.0f / 255.0f);
 385 }
 386
 387
 388 /**
 389  * Convert float in [0,1] to uint8_t in [0,255] with clamping.
 390  */
 391 static inline uint8_t
 392 float_to_ubyte(float f)
 393 {
 394    /* return 0 for NaN too */
 395    if (!(f > 0.0f)) {
 396       return (uint8_t) 0;
 397    }
 398    else if (f >= 1.0f) {
 399       return (uint8_t) 255;
 400    }
 401    else {
 402       union fi tmp;
 403       tmp.f = f;
 404       tmp.f = tmp.f * (255.0f/256.0f) + 32768.0f;
 405       return (uint8_t) tmp.i;
 406    }
 407 }
 408
 409 /**
 410  * Convert uint16_t to float in [0, 1].
 411  */
 412 static inline float
 413 ushort_to_float(uint16_t us)
 414 {
 415    return (float) us * (1.0f / 65535.0f);
 416 }
 417
 418
 419 /**
 420  * Convert float in [0,1] to uint16_t in [0,65535] with clamping.
 421  */
 422 static inline uint16_t
 423 float_to_ushort(float f)
 424 {
 425    /* return 0 for NaN too */
 426    if (!(f > 0.0f)) {
 427       return (uint16_t) 0;
 428    }
 429    else if (f >= 1.0f) {
 430       return (uint16_t) 65535;
 431    }
 432    else {
 433       union fi tmp;
 434       tmp.f = f;
 435       tmp.f = tmp.f * (65535.0f/65536.0f) + 128.0f;
 436       return (uint16_t) tmp.i;
 437    }
 438 }
 439
 440 static inline float
 441 byte_to_float_tex(int8_t b)
 442 {
 443    return (b == -128) ? -1.0F : b * 1.0F / 127.0F;
 444 }
 445
 446 static inline int8_t
 447 float_to_byte_tex(float f)
 448 {
 449    return (int8_t) (127.0F * f);
 450 }
 451
 452 /**
 453  * Calc log base 2
 454  */
 455 static inline unsigned
 456 util_logbase2(unsigned n)
 457 {
 458 #if defined(HAVE___BUILTIN_CLZ)
 459    return ((sizeof(unsigned) * 8 - 1) - __builtin_clz(n | 1));
 460 #else
 461    unsigned pos = 0;
 462    if (n >= 1<<16) { n >>= 16; pos += 16; }
 463    if (n >= 1<< 8) { n >>=  8; pos +=  8; }
 464    if (n >= 1<< 4) { n >>=  4; pos +=  4; }
 465    if (n >= 1<< 2) { n >>=  2; pos +=  2; }
 466    if (n >= 1<< 1) {           pos +=  1; }
 467    return pos;
 468 #endif
 469 }
 470
 471 static inline uint64_t
 472 util_logbase2_64(uint64_t n)
 473 {
 474 #if defined(HAVE___BUILTIN_CLZLL)
 475    return ((sizeof(uint64_t) * 8 - 1) - __builtin_clzll(n | 1));
 476 #else
 477    uint64_t pos = 0ull;
 478    if (n >= 1ull<<32) { n >>= 32; pos += 32; }
 479    if (n >= 1ull<<16) { n >>= 16; pos += 16; }
 480    if (n >= 1ull<< 8) { n >>=  8; pos +=  8; }
 481    if (n >= 1ull<< 4) { n >>=  4; pos +=  4; }
 482    if (n >= 1ull<< 2) { n >>=  2; pos +=  2; }
 483    if (n >= 1ull<< 1) {           pos +=  1; }
 484    return pos;
 485 #endif
 486 }
 487
 488 /**
 489  * Returns the ceiling of log n base 2, and 0 when n == 0. Equivalently,
 490  * returns the smallest x such that n <= 2**x.
 491  */
 492 static inline unsigned
 493 util_logbase2_ceil(unsigned n)
 494 {
 495    if (n <= 1)
 496       return 0;
 497
 498    return 1 + util_logbase2(n - 1);
 499 }
 500
 501 static inline uint64_t
 502 util_logbase2_ceil64(uint64_t n)
 503 {
 504    if (n <= 1)
 505       return 0;
 506
 507    return 1ull + util_logbase2_64(n - 1);
 508 }
 509
 510 /**
 511  * Returns the smallest power of two >= x
 512  */
 513 static inline unsigned
 514 util_next_power_of_two(unsigned x)
 515 {
 516 #if defined(HAVE___BUILTIN_CLZ)
 517    if (x <= 1)
 518        return 1;
 519
 520    return (1 << ((sizeof(unsigned) * 8) - __builtin_clz(x - 1)));
 521 #else
 522    unsigned val = x;
 523
 524    if (x <= 1)
 525       return 1;
 526
 527    if (util_is_power_of_two_or_zero(x))
 528       return x;
 529
 530    val--;
 531    val = (val >> 1) | val;
 532    val = (val >> 2) | val;
 533    val = (val >> 4) | val;
 534    val = (val >> 8) | val;
 535    val = (val >> 16) | val;
 536    val++;
 537    return val;
 538 #endif
 539 }
 540
 541 static inline uint64_t
 542 util_next_power_of_two64(uint64_t x)
 543 {
 544 #if defined(HAVE___BUILTIN_CLZLL)
 545    if (x <= 1)
 546        return 1;
 547
 548    return (1ull << ((sizeof(uint64_t) * 8) - __builtin_clzll(x - 1)));
 549 #else
 550    uint64_t val = x;
 551
 552    if (x <= 1)
 553       return 1;
 554
 555    if (util_is_power_of_two_or_zero64(x))
 556       return x;
 557
 558    val--;
 559    val = (val >> 1)  | val;
 560    val = (val >> 2)  | val;
 561    val = (val >> 4)  | val;
 562    val = (val >> 8)  | val;
 563    val = (val >> 16) | val;
 564    val = (val >> 32) | val;
 565    val++;
 566    return val;
 567 #endif
 568 }
 569
 570 /**
 571  * Reverse bits in n
 572  * Algorithm taken from:
 573  * http://stackoverflow.com/questions/9144800/c-reverse-bits-in-unsigned-integer
 574  */
 575 static inline unsigned
 576 util_bitreverse(unsigned n)
 577 {
 578     n = ((n >> 1) & 0x55555555u) | ((n & 0x55555555u) << 1);
 579     n = ((n >> 2) & 0x33333333u) | ((n & 0x33333333u) << 2);
 580     n = ((n >> 4) & 0x0f0f0f0fu) | ((n & 0x0f0f0f0fu) << 4);
 581     n = ((n >> 8) & 0x00ff00ffu) | ((n & 0x00ff00ffu) << 8);
 582     n = ((n >> 16) & 0xffffu) | ((n & 0xffffu) << 16);
 583     return n;
 584 }
 585
 586 /**
 587  * Convert from little endian to CPU byte order.
 588  */
 589
 590 #if UTIL_ARCH_BIG_ENDIAN
 591 #define util_le64_to_cpu(x) util_bswap64(x)
 592 #define util_le32_to_cpu(x) util_bswap32(x)
 593 #define util_le16_to_cpu(x) util_bswap16(x)
 594 #else
 595 #define util_le64_to_cpu(x) (x)
 596 #define util_le32_to_cpu(x) (x)
 597 #define util_le16_to_cpu(x) (x)
 598 #endif
 599
 600 #define util_cpu_to_le64(x) util_le64_to_cpu(x)
 601 #define util_cpu_to_le32(x) util_le32_to_cpu(x)
 602 #define util_cpu_to_le16(x) util_le16_to_cpu(x)
 603
 604 /**
 605  * Reverse byte order of a 32 bit word.
 606  */
 607 static inline uint32_t
 608 util_bswap32(uint32_t n)
 609 {
 610 #if defined(HAVE___BUILTIN_BSWAP32)
 611    return __builtin_bswap32(n);
 612 #else
 613    return (n >> 24) |
 614           ((n >> 8) & 0x0000ff00) |
 615           ((n << 8) & 0x00ff0000) |
 616           (n << 24);
 617 #endif
 618 }
 619
 620 /**
 621  * Reverse byte order of a 64bit word.
 622  */
 623 static inline uint64_t
 624 util_bswap64(uint64_t n)
 625 {
 626 #if defined(HAVE___BUILTIN_BSWAP64)
 627    return __builtin_bswap64(n);
 628 #else
 629    return ((uint64_t)util_bswap32((uint32_t)n) << 32) |
 630           util_bswap32((n >> 32));
 631 #endif
 632 }
 633
 634
 635 /**
 636  * Reverse byte order of a 16 bit word.
 637  */
 638 static inline uint16_t
 639 util_bswap16(uint16_t n)
 640 {
 641    return (n >> 8) |
 642           (n << 8);
 643 }
 644
 645 static inline void*
 646 util_memcpy_cpu_to_le32(void * restrict dest, const void * restrict src, size_t n)
 647 {
 648 #if UTIL_ARCH_BIG_ENDIAN
 649    size_t i, e;
 650    assert(n % 4 == 0);
 651
 652    for (i = 0, e = n / 4; i < e; i++) {
 653       uint32_t * restrict d = (uint32_t* restrict)dest;
 654       const uint32_t * restrict s = (const uint32_t* restrict)src;
 655       d[i] = util_bswap32(s[i]);
 656    }
 657    return dest;
 658 #else
 659    return memcpy(dest, src, n);
 660 #endif
 661 }
 662
 663 /**
 664  * Clamp X to [MIN, MAX].
 665  * This is a macro to allow float, int, uint, etc. types.
 666  * We arbitrarily turn NaN into MIN.
 667  */
 668 #define CLAMP( X, MIN, MAX )  ( (X)>(MIN) ? ((X)>(MAX) ? (MAX) : (X)) : (MIN) )
 669
 670 /* Syntax sugar occuring frequently in graphics code */
 671 #define SATURATE( X ) CLAMP(X, 0.0f, 1.0f)
 672
 673 #define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
 674 #define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
 675
 676 #define MIN3( A, B, C ) ((A) < (B) ? MIN2(A, C) : MIN2(B, C))
 677 #define MAX3( A, B, C ) ((A) > (B) ? MAX2(A, C) : MAX2(B, C))
 678
 679 #define MIN4( A, B, C, D ) ((A) < (B) ? MIN3(A, C, D) : MIN3(B, C, D))
 680 #define MAX4( A, B, C, D ) ((A) > (B) ? MAX3(A, C, D) : MAX3(B, C, D))
 681
 682
 683 /**
 684  * Align a value up to an alignment value
 685  *
 686  * If \c value is not already aligned to the requested alignment value, it
 687  * will be rounded up.
 688  *
 689  * \param value  Value to be rounded
 690  * \param alignment  Alignment value to be used.  This must be a power of two.
 691  *
 692  * \sa ROUND_DOWN_TO()
 693  */
 694 static inline uintptr_t
 695 ALIGN(uintptr_t value, int32_t alignment)
 696 {
 697    assert(util_is_power_of_two_nonzero(alignment));
 698    return (((value) + (alignment) - 1) & ~((alignment) - 1));
 699 }
 700
 701 /**
 702  * Like ALIGN(), but works with a non-power-of-two alignment.
 703  */
 704 static inline uintptr_t
 705 ALIGN_NPOT(uintptr_t value, int32_t alignment)
 706 {
 707    assert(alignment > 0);
 708    return (value + alignment - 1) / alignment * alignment;
 709 }
 710
 711 /**
 712  * Align a value down to an alignment value
 713  *
 714  * If \c value is not already aligned to the requested alignment value, it
 715  * will be rounded down.
 716  *
 717  * \param value  Value to be rounded
 718  * \param alignment  Alignment value to be used.  This must be a power of two.
 719  *
 720  * \sa ALIGN()
 721  */
 722 static inline uintptr_t
 723 ROUND_DOWN_TO(uintptr_t value, int32_t alignment)
 724 {
 725    assert(util_is_power_of_two_nonzero(alignment));
 726    return ((value) & ~(alignment - 1));
 727 }
 728
 729 /**
 730  * Align a value, only works pot alignemnts.
 731  */
 732 static inline int
 733 align(int value, int alignment)
 734 {
 735    return (value + alignment - 1) & ~(alignment - 1);
 736 }
 737
 738 static inline uint64_t
 739 align64(uint64_t value, unsigned alignment)
 740 {
 741    return (value + alignment - 1) & ~((uint64_t)alignment - 1);
 742 }
 743
 744 /**
 745  * Works like align but on npot alignments.
 746  */
 747 static inline size_t
 748 util_align_npot(size_t value, size_t alignment)
 749 {
 750    if (value % alignment)
 751       return value + (alignment - (value % alignment));
 752    return value;
 753 }
 754
 755 static inline unsigned
 756 u_minify(unsigned value, unsigned levels)
 757 {
 758     return MAX2(1, value >> levels);
 759 }
 760
 761 #ifndef COPY_4V
 762 #define COPY_4V( DST, SRC )         \
 763 do {                                \
 764    (DST)[0] = (SRC)[0];             \
 765    (DST)[1] = (SRC)[1];             \
 766    (DST)[2] = (SRC)[2];             \
 767    (DST)[3] = (SRC)[3];             \
 768 } while (0)
 769 #endif
 770
 771
 772 #ifndef COPY_4FV
 773 #define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
 774 #endif
 775
 776
 777 #ifndef ASSIGN_4V
 778 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
 779 do {                                     \
 780    (DST)[0] = (V0);                      \
 781    (DST)[1] = (V1);                      \
 782    (DST)[2] = (V2);                      \
 783    (DST)[3] = (V3);                      \
 784 } while (0)
 785 #endif
 786
 787
 788 static inline uint32_t
 789 util_unsigned_fixed(float value, unsigned frac_bits)
 790 {
 791    return value < 0 ? 0 : (uint32_t)(value * (1<<frac_bits));
 792 }
 793
 794 static inline int32_t
 795 util_signed_fixed(float value, unsigned frac_bits)
 796 {
 797    return (int32_t)(value * (1<<frac_bits));
 798 }
 799
 800 unsigned
 801 util_fpstate_get(void);
 802 unsigned
 803 util_fpstate_set_denorms_to_zero(unsigned current_fpstate);
 804 void
 805 util_fpstate_set(unsigned fpstate);
 806
 807 /**
 808  * For indexed draw calls, return true if the vertex count to be drawn is
 809  * much lower than the vertex count that has to be uploaded, meaning
 810  * that the driver should flatten indices instead of trying to upload
 811  * a too big range.
 812  *
 813  * This is used by vertex upload code in u_vbuf and glthread.
 814  */
 815 static inline bool
 816 util_is_vbo_upload_ratio_too_large(unsigned draw_vertex_count,
 817                                    unsigned upload_vertex_count)
 818 {
 819    if (draw_vertex_count > 1024)
 820       return upload_vertex_count > draw_vertex_count * 4;
 821    else if (draw_vertex_count > 32)
 822       return upload_vertex_count > draw_vertex_count * 8;
 823    else
 824       return upload_vertex_count > draw_vertex_count * 16;
 825 }
 826
 827 #ifdef __cplusplus
 828 }
 829 #endif
 830
 831 #endif /* U_MATH_H */