src/mesa/main/imports.c

   1 /**
   2  * \file imports.c
   3  * Standard C library function wrappers.
   4  *
   5  * Imports are services which the device driver or window system or
   6  * operating system provides to the core renderer.  The core renderer (Mesa)
   7  * will call these functions in order to do memory allocation, simple I/O,
   8  * etc.
   9  *
  10  * Some drivers will want to override/replace this file with something
  11  * specialized, but that'll be rare.
  12  *
  13  * Eventually, I want to move roll the glheader.h file into this.
  14  *
  15  * \todo Functions still needed:
  16  * - scanf
  17  * - qsort
  18  * - rand and RAND_MAX
  19  */
  20
  21 /*
  22  * Mesa 3-D graphics library
  23  * Version:  7.1
  24  *
  25  * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
  26  *
  27  * Permission is hereby granted, free of charge, to any person obtaining a
  28  * copy of this software and associated documentation files (the "Software"),
  29  * to deal in the Software without restriction, including without limitation
  30  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  31  * and/or sell copies of the Software, and to permit persons to whom the
  32  * Software is furnished to do so, subject to the following conditions:
  33  *
  34  * The above copyright notice and this permission notice shall be included
  35  * in all copies or substantial portions of the Software.
  36  *
  37  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  38  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  39  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  40  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  41  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  42  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  43  * OTHER DEALINGS IN THE SOFTWARE.
  44  */
  45
  46
  47
  48 #include "imports.h"
  49 #include "context.h"
  50 #include "mtypes.h"
  51 #include "version.h"
  52
  53 #ifdef _GNU_SOURCE
  54 #include <locale.h>
  55 #ifdef __APPLE__
  56 #include <xlocale.h>
  57 #endif
  58 #endif
  59
  60
  61 #ifdef _WIN32
  62 #define vsnprintf _vsnprintf
  63 #elif defined(__IBMC__) || defined(__IBMCPP__)
  64 extern int vsnprintf(char *str, size_t count, const char *fmt, va_list arg);
  65 #endif
  66
  67 /**********************************************************************/
  68 /** \name Memory */
  69 /*@{*/
  70
  71 /**
  72  * Allocate aligned memory.
  73  *
  74  * \param bytes number of bytes to allocate.
  75  * \param alignment alignment (must be greater than zero).
  76  *
  77  * Allocates extra memory to accommodate rounding up the address for
  78  * alignment and to record the real malloc address.
  79  *
  80  * \sa _mesa_align_free().
  81  */
  82 void *
  83 _mesa_align_malloc(size_t bytes, unsigned long alignment)
  84 {
  85 #if defined(HAVE_POSIX_MEMALIGN)
  86    void *mem;
  87    int err = posix_memalign(& mem, alignment, bytes);
  88    if (err)
  89       return NULL;
  90    return mem;
  91 #elif defined(_WIN32) && defined(_MSC_VER)
  92    return _aligned_malloc(bytes, alignment);
  93 #else
  94    uintptr_t ptr, buf;
  95
  96    ASSERT( alignment > 0 );
  97
  98    ptr = (uintptr_t)malloc(bytes + alignment + sizeof(void *));
  99    if (!ptr)
 100       return NULL;
 101
 102    buf = (ptr + alignment + sizeof(void *)) & ~(uintptr_t)(alignment - 1);
 103    *(uintptr_t *)(buf - sizeof(void *)) = ptr;
 104
 105 #ifdef DEBUG
 106    /* mark the non-aligned area */
 107    while ( ptr < buf - sizeof(void *) ) {
 108       *(unsigned long *)ptr = 0xcdcdcdcd;
 109       ptr += sizeof(unsigned long);
 110    }
 111 #endif
 112
 113    return (void *) buf;
 114 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 115 }
 116
 117 /**
 118  * Same as _mesa_align_malloc(), but using calloc(1, ) instead of
 119  * malloc()
 120  */
 121 void *
 122 _mesa_align_calloc(size_t bytes, unsigned long alignment)
 123 {
 124 #if defined(HAVE_POSIX_MEMALIGN)
 125    void *mem;
 126
 127    mem = _mesa_align_malloc(bytes, alignment);
 128    if (mem != NULL) {
 129       (void) memset(mem, 0, bytes);
 130    }
 131
 132    return mem;
 133 #elif defined(_WIN32) && defined(_MSC_VER)
 134    void *mem;
 135
 136    mem = _aligned_malloc(bytes, alignment);
 137    if (mem != NULL) {
 138       (void) memset(mem, 0, bytes);
 139    }
 140
 141    return mem;
 142 #else
 143    uintptr_t ptr, buf;
 144
 145    ASSERT( alignment > 0 );
 146
 147    ptr = (uintptr_t)calloc(1, bytes + alignment + sizeof(void *));
 148    if (!ptr)
 149       return NULL;
 150
 151    buf = (ptr + alignment + sizeof(void *)) & ~(uintptr_t)(alignment - 1);
 152    *(uintptr_t *)(buf - sizeof(void *)) = ptr;
 153
 154 #ifdef DEBUG
 155    /* mark the non-aligned area */
 156    while ( ptr < buf - sizeof(void *) ) {
 157       *(unsigned long *)ptr = 0xcdcdcdcd;
 158       ptr += sizeof(unsigned long);
 159    }
 160 #endif
 161
 162    return (void *)buf;
 163 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 164 }
 165
 166 /**
 167  * Free memory which was allocated with either _mesa_align_malloc()
 168  * or _mesa_align_calloc().
 169  * \param ptr pointer to the memory to be freed.
 170  * The actual address to free is stored in the word immediately before the
 171  * address the client sees.
 172  */
 173 void
 174 _mesa_align_free(void *ptr)
 175 {
 176 #if defined(HAVE_POSIX_MEMALIGN)
 177    free(ptr);
 178 #elif defined(_WIN32) && defined(_MSC_VER)
 179    _aligned_free(ptr);
 180 #else
 181    void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
 182    void *realAddr = *cubbyHole;
 183    free(realAddr);
 184 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 185 }
 186
 187 /**
 188  * Reallocate memory, with alignment.
 189  */
 190 void *
 191 _mesa_align_realloc(void *oldBuffer, size_t oldSize, size_t newSize,
 192                     unsigned long alignment)
 193 {
 194 #if defined(_WIN32) && defined(_MSC_VER)
 195    (void) oldSize;
 196    return _aligned_realloc(oldBuffer, newSize, alignment);
 197 #else
 198    const size_t copySize = (oldSize < newSize) ? oldSize : newSize;
 199    void *newBuf = _mesa_align_malloc(newSize, alignment);
 200    if (newBuf && oldBuffer && copySize > 0) {
 201       memcpy(newBuf, oldBuffer, copySize);
 202    }
 203    if (oldBuffer)
 204       _mesa_align_free(oldBuffer);
 205    return newBuf;
 206 #endif
 207 }
 208
 209
 210
 211 /** Reallocate memory */
 212 void *
 213 _mesa_realloc(void *oldBuffer, size_t oldSize, size_t newSize)
 214 {
 215    const size_t copySize = (oldSize < newSize) ? oldSize : newSize;
 216    void *newBuffer = malloc(newSize);
 217    if (newBuffer && oldBuffer && copySize > 0)
 218       memcpy(newBuffer, oldBuffer, copySize);
 219    free(oldBuffer);
 220    return newBuffer;
 221 }
 222
 223 /*@}*/
 224
 225
 226 /**********************************************************************/
 227 /** \name Math */
 228 /*@{*/
 229
 230
 231 #ifndef __GNUC__
 232 /**
 233  * Find the first bit set in a word.
 234  */
 235 int
 236 ffs(int i)
 237 {
 238    register int bit = 0;
 239    if (i != 0) {
 240       if ((i & 0xffff) == 0) {
 241          bit += 16;
 242          i >>= 16;
 243       }
 244       if ((i & 0xff) == 0) {
 245          bit += 8;
 246          i >>= 8;
 247       }
 248       if ((i & 0xf) == 0) {
 249          bit += 4;
 250          i >>= 4;
 251       }
 252       while ((i & 1) == 0) {
 253          bit++;
 254          i >>= 1;
 255       }
 256       bit++;
 257    }
 258    return bit;
 259 }
 260
 261
 262 /**
 263  * Find position of first bit set in given value.
 264  * XXX Warning: this function can only be used on 64-bit systems!
 265  * \return  position of least-significant bit set, starting at 1, return zero
 266  *          if no bits set.
 267  */
 268 int
 269 ffsll(long long int val)
 270 {
 271    int bit;
 272
 273    assert(sizeof(val) == 8);
 274
 275    bit = ffs((int) val);
 276    if (bit != 0)
 277       return bit;
 278
 279    bit = ffs((int) (val >> 32));
 280    if (bit != 0)
 281       return 32 + bit;
 282
 283    return 0;
 284 }
 285 #endif /* __GNUC__ */
 286
 287
 288 #if !defined(__GNUC__) ||\
 289    ((__GNUC__ * 100 + __GNUC_MINOR__) < 304) /* Not gcc 3.4 or later */
 290 /**
 291  * Return number of bits set in given GLuint.
 292  */
 293 unsigned int
 294 _mesa_bitcount(unsigned int n)
 295 {
 296    unsigned int bits;
 297    for (bits = 0; n > 0; n = n >> 1) {
 298       bits += (n & 1);
 299    }
 300    return bits;
 301 }
 302
 303 /**
 304  * Return number of bits set in given 64-bit uint.
 305  */
 306 unsigned int
 307 _mesa_bitcount_64(uint64_t n)
 308 {
 309    unsigned int bits;
 310    for (bits = 0; n > 0; n = n >> 1) {
 311       bits += (n & 1);
 312    }
 313    return bits;
 314 }
 315 #endif
 316
 317
 318 /* Using C99 rounding functions for roundToEven() implementation is
 319  * difficult, because round(), rint, and nearbyint() are affected by
 320  * fesetenv(), which the application may have done for its own
 321  * purposes.  Mesa's IROUND macro is close to what we want, but it
 322  * rounds away from 0 on n + 0.5.
 323  */
 324 int
 325 _mesa_round_to_even(float val)
 326 {
 327    int rounded = IROUND(val);
 328
 329    if (val - floor(val) == 0.5) {
 330       if (rounded % 2 != 0)
 331          rounded += val > 0 ? -1 : 1;
 332    }
 333
 334    return rounded;
 335 }
 336
 337
 338 /**
 339  * Convert a 4-byte float to a 2-byte half float.
 340  *
 341  * Not all float32 values can be represented exactly as a float16 value. We
 342  * round such intermediate float32 values to the nearest float16. When the
 343  * float32 lies exactly between to float16 values, we round to the one with
 344  * an even mantissa.
 345  *
 346  * This rounding behavior has several benefits:
 347  *   - It has no sign bias.
 348  *
 349  *   - It reproduces the behavior of real hardware: opcode F32TO16 in Intel's
 350  *     GPU ISA.
 351  *
 352  *   - By reproducing the behavior of the GPU (at least on Intel hardware),
 353  *     compile-time evaluation of constant packHalf2x16 GLSL expressions will
 354  *     result in the same value as if the expression were executed on the GPU.
 355  */
 356 GLhalfARB
 357 _mesa_float_to_half(float val)
 358 {
 359    const fi_type fi = {val};
 360    const int flt_m = fi.i & 0x7fffff;
 361    const int flt_e = (fi.i >> 23) & 0xff;
 362    const int flt_s = (fi.i >> 31) & 0x1;
 363    int s, e, m = 0;
 364    GLhalfARB result;
 365
 366    /* sign bit */
 367    s = flt_s;
 368
 369    /* handle special cases */
 370    if ((flt_e == 0) && (flt_m == 0)) {
 371       /* zero */
 372       /* m = 0; - already set */
 373       e = 0;
 374    }
 375    else if ((flt_e == 0) && (flt_m != 0)) {
 376       /* denorm -- denorm float maps to 0 half */
 377       /* m = 0; - already set */
 378       e = 0;
 379    }
 380    else if ((flt_e == 0xff) && (flt_m == 0)) {
 381       /* infinity */
 382       /* m = 0; - already set */
 383       e = 31;
 384    }
 385    else if ((flt_e == 0xff) && (flt_m != 0)) {
 386       /* NaN */
 387       m = 1;
 388       e = 31;
 389    }
 390    else {
 391       /* regular number */
 392       const int new_exp = flt_e - 127;
 393       if (new_exp < -14) {
 394          /* The float32 lies in the range (0.0, min_normal16) and is rounded
 395           * to a nearby float16 value. The result will be either zero, subnormal,
 396           * or normal.
 397           */
 398          e = 0;
 399          m = _mesa_round_to_even((1 << 24) * fabsf(fi.f));
 400       }
 401       else if (new_exp > 15) {
 402          /* map this value to infinity */
 403          /* m = 0; - already set */
 404          e = 31;
 405       }
 406       else {
 407          /* The float32 lies in the range
 408           *   [min_normal16, max_normal16 + max_step16)
 409           * and is rounded to a nearby float16 value. The result will be
 410           * either normal or infinite.
 411           */
 412          e = new_exp + 15;
 413          m = _mesa_round_to_even(flt_m / (float) (1 << 13));
 414       }
 415    }
 416
 417    assert(0 <= m && m <= 1024);
 418    if (m == 1024) {
 419       /* The float32 was rounded upwards into the range of the next exponent,
 420        * so bump the exponent. This correctly handles the case where f32
 421        * should be rounded up to float16 infinity.
 422        */
 423       ++e;
 424       m = 0;
 425    }
 426
 427    result = (s << 15) | (e << 10) | m;
 428    return result;
 429 }
 430
 431
 432 /**
 433  * Convert a 2-byte half float to a 4-byte float.
 434  * Based on code from:
 435  * http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/008786.html
 436  */
 437 float
 438 _mesa_half_to_float(GLhalfARB val)
 439 {
 440    /* XXX could also use a 64K-entry lookup table */
 441    const int m = val & 0x3ff;
 442    const int e = (val >> 10) & 0x1f;
 443    const int s = (val >> 15) & 0x1;
 444    int flt_m, flt_e, flt_s;
 445    fi_type fi;
 446    float result;
 447
 448    /* sign bit */
 449    flt_s = s;
 450
 451    /* handle special cases */
 452    if ((e == 0) && (m == 0)) {
 453       /* zero */
 454       flt_m = 0;
 455       flt_e = 0;
 456    }
 457    else if ((e == 0) && (m != 0)) {
 458       /* denorm -- denorm half will fit in non-denorm single */
 459       const float half_denorm = 1.0f / 16384.0f; /* 2^-14 */
 460       float mantissa = ((float) (m)) / 1024.0f;
 461       float sign = s ? -1.0f : 1.0f;
 462       return sign * mantissa * half_denorm;
 463    }
 464    else if ((e == 31) && (m == 0)) {
 465       /* infinity */
 466       flt_e = 0xff;
 467       flt_m = 0;
 468    }
 469    else if ((e == 31) && (m != 0)) {
 470       /* NaN */
 471       flt_e = 0xff;
 472       flt_m = 1;
 473    }
 474    else {
 475       /* regular */
 476       flt_e = e + 112;
 477       flt_m = m << 13;
 478    }
 479
 480    fi.i = (flt_s << 31) | (flt_e << 23) | flt_m;
 481    result = fi.f;
 482    return result;
 483 }
 484
 485 /*@}*/
 486
 487
 488 /**********************************************************************/
 489 /** \name Sort & Search */
 490 /*@{*/
 491
 492 /**
 493  * Wrapper for bsearch().
 494  */
 495 void *
 496 _mesa_bsearch( const void *key, const void *base, size_t nmemb, size_t size,
 497                int (*compar)(const void *, const void *) )
 498 {
 499 #if defined(_WIN32_WCE)
 500    void *mid;
 501    int cmp;
 502    while (nmemb) {
 503       nmemb >>= 1;
 504       mid = (char *)base + nmemb * size;
 505       cmp = (*compar)(key, mid);
 506       if (cmp == 0)
 507          return mid;
 508       if (cmp > 0) {
 509          base = (char *)mid + size;
 510          --nmemb;
 511       }
 512    }
 513    return NULL;
 514 #else
 515    return bsearch(key, base, nmemb, size, compar);
 516 #endif
 517 }
 518
 519 /*@}*/
 520
 521
 522 /**********************************************************************/
 523 /** \name Environment vars */
 524 /*@{*/
 525
 526 /**
 527  * Wrapper for getenv().
 528  */
 529 char *
 530 _mesa_getenv( const char *var )
 531 {
 532 #if defined(_XBOX) || defined(_WIN32_WCE)
 533    return NULL;
 534 #else
 535    return getenv(var);
 536 #endif
 537 }
 538
 539 /*@}*/
 540
 541
 542 /**********************************************************************/
 543 /** \name String */
 544 /*@{*/
 545
 546 /**
 547  * Implemented using malloc() and strcpy.
 548  * Note that NULL is handled accordingly.
 549  */
 550 char *
 551 _mesa_strdup( const char *s )
 552 {
 553    if (s) {
 554       size_t l = strlen(s);
 555       char *s2 = malloc(l + 1);
 556       if (s2)
 557          strcpy(s2, s);
 558       return s2;
 559    }
 560    else {
 561       return NULL;
 562    }
 563 }
 564
 565 /** Wrapper around strtof() */
 566 float
 567 _mesa_strtof( const char *s, char **end )
 568 {
 569 #if defined(_GNU_SOURCE) && !defined(__CYGWIN__) && !defined(__FreeBSD__) && \
 570    !defined(ANDROID) && !defined(__HAIKU__) && !defined(__UCLIBC__)
 571    static locale_t loc = NULL;
 572    if (!loc) {
 573       loc = newlocale(LC_CTYPE_MASK, "C", NULL);
 574    }
 575    return strtof_l(s, end, loc);
 576 #elif defined(_ISOC99_SOURCE) || (defined(_XOPEN_SOURCE) && _XOPEN_SOURCE >= 600)
 577    return strtof(s, end);
 578 #else
 579    return (float)strtod(s, end);
 580 #endif
 581 }
 582
 583 /** Compute simple checksum/hash for a string */
 584 unsigned int
 585 _mesa_str_checksum(const char *str)
 586 {
 587    /* This could probably be much better */
 588    unsigned int sum, i;
 589    const char *c;
 590    sum = i = 1;
 591    for (c = str; *c; c++, i++)
 592       sum += *c * (i % 100);
 593    return sum + i;
 594 }
 595
 596
 597 /*@}*/
 598
 599
 600 /** Needed due to #ifdef's, above. */
 601 int
 602 _mesa_vsnprintf(char *str, size_t size, const char *fmt, va_list args)
 603 {
 604    return vsnprintf( str, size, fmt, args);
 605 }
 606
 607 /** Wrapper around vsnprintf() */
 608 int
 609 _mesa_snprintf( char *str, size_t size, const char *fmt, ... )
 610 {
 611    int r;
 612    va_list args;
 613    va_start( args, fmt );
 614    r = vsnprintf( str, size, fmt, args );
 615    va_end( args );
 616    return r;
 617 }
 618
 619