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  *
  24  * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
  25  *
  26  * Permission is hereby granted, free of charge, to any person obtaining a
  27  * copy of this software and associated documentation files (the "Software"),
  28  * to deal in the Software without restriction, including without limitation
  29  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  30  * and/or sell copies of the Software, and to permit persons to whom the
  31  * Software is furnished to do so, subject to the following conditions:
  32  *
  33  * The above copyright notice and this permission notice shall be included
  34  * in all copies or substantial portions of the Software.
  35  *
  36  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  37  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  38  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  39  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  40  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  41  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  42  * OTHER DEALINGS IN THE SOFTWARE.
  43  */
  44
  45 #include <stdarg.h>
  46 #include "c99_math.h"
  47 #include "imports.h"
  48 #include "context.h"
  49 #include "mtypes.h"
  50 #include "version.h"
  51
  52 #ifdef _GNU_SOURCE
  53 #include <locale.h>
  54 #ifdef __APPLE__
  55 #include <xlocale.h>
  56 #endif
  57 #endif
  58
  59
  60 #ifdef _WIN32
  61 #define vsnprintf _vsnprintf
  62 #elif defined(__IBMC__) || defined(__IBMCPP__)
  63 extern int vsnprintf(char *str, size_t count, const char *fmt, va_list arg);
  64 #endif
  65
  66 /**********************************************************************/
  67 /** \name Memory */
  68 /*@{*/
  69
  70 /**
  71  * Allocate aligned memory.
  72  *
  73  * \param bytes number of bytes to allocate.
  74  * \param alignment alignment (must be greater than zero).
  75  *
  76  * Allocates extra memory to accommodate rounding up the address for
  77  * alignment and to record the real malloc address.
  78  *
  79  * \sa _mesa_align_free().
  80  */
  81 void *
  82 _mesa_align_malloc(size_t bytes, unsigned long alignment)
  83 {
  84 #if defined(HAVE_POSIX_MEMALIGN)
  85    void *mem;
  86    int err = posix_memalign(& mem, alignment, bytes);
  87    if (err)
  88       return NULL;
  89    return mem;
  90 #elif defined(_WIN32) && defined(_MSC_VER)
  91    return _aligned_malloc(bytes, alignment);
  92 #else
  93    uintptr_t ptr, buf;
  94
  95    assert( alignment > 0 );
  96
  97    ptr = (uintptr_t)malloc(bytes + alignment + sizeof(void *));
  98    if (!ptr)
  99       return NULL;
 100
 101    buf = (ptr + alignment + sizeof(void *)) & ~(uintptr_t)(alignment - 1);
 102    *(uintptr_t *)(buf - sizeof(void *)) = ptr;
 103
 104 #ifdef DEBUG
 105    /* mark the non-aligned area */
 106    while ( ptr < buf - sizeof(void *) ) {
 107       *(unsigned long *)ptr = 0xcdcdcdcd;
 108       ptr += sizeof(unsigned long);
 109    }
 110 #endif
 111
 112    return (void *) buf;
 113 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 114 }
 115
 116 /**
 117  * Same as _mesa_align_malloc(), but using calloc(1, ) instead of
 118  * malloc()
 119  */
 120 void *
 121 _mesa_align_calloc(size_t bytes, unsigned long alignment)
 122 {
 123 #if defined(HAVE_POSIX_MEMALIGN)
 124    void *mem;
 125
 126    mem = _mesa_align_malloc(bytes, alignment);
 127    if (mem != NULL) {
 128       (void) memset(mem, 0, bytes);
 129    }
 130
 131    return mem;
 132 #elif defined(_WIN32) && defined(_MSC_VER)
 133    void *mem;
 134
 135    mem = _aligned_malloc(bytes, alignment);
 136    if (mem != NULL) {
 137       (void) memset(mem, 0, bytes);
 138    }
 139
 140    return mem;
 141 #else
 142    uintptr_t ptr, buf;
 143
 144    assert( alignment > 0 );
 145
 146    ptr = (uintptr_t)calloc(1, bytes + alignment + sizeof(void *));
 147    if (!ptr)
 148       return NULL;
 149
 150    buf = (ptr + alignment + sizeof(void *)) & ~(uintptr_t)(alignment - 1);
 151    *(uintptr_t *)(buf - sizeof(void *)) = ptr;
 152
 153 #ifdef DEBUG
 154    /* mark the non-aligned area */
 155    while ( ptr < buf - sizeof(void *) ) {
 156       *(unsigned long *)ptr = 0xcdcdcdcd;
 157       ptr += sizeof(unsigned long);
 158    }
 159 #endif
 160
 161    return (void *)buf;
 162 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 163 }
 164
 165 /**
 166  * Free memory which was allocated with either _mesa_align_malloc()
 167  * or _mesa_align_calloc().
 168  * \param ptr pointer to the memory to be freed.
 169  * The actual address to free is stored in the word immediately before the
 170  * address the client sees.
 171  * Note that it is legal to pass NULL pointer to this function and will be
 172  * handled accordingly.
 173  */
 174 void
 175 _mesa_align_free(void *ptr)
 176 {
 177 #if defined(HAVE_POSIX_MEMALIGN)
 178    free(ptr);
 179 #elif defined(_WIN32) && defined(_MSC_VER)
 180    _aligned_free(ptr);
 181 #else
 182    if (ptr) {
 183       void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
 184       void *realAddr = *cubbyHole;
 185       free(realAddr);
 186    }
 187 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 188 }
 189
 190 /**
 191  * Reallocate memory, with alignment.
 192  */
 193 void *
 194 _mesa_align_realloc(void *oldBuffer, size_t oldSize, size_t newSize,
 195                     unsigned long alignment)
 196 {
 197 #if defined(_WIN32) && defined(_MSC_VER)
 198    (void) oldSize;
 199    return _aligned_realloc(oldBuffer, newSize, alignment);
 200 #else
 201    const size_t copySize = (oldSize < newSize) ? oldSize : newSize;
 202    void *newBuf = _mesa_align_malloc(newSize, alignment);
 203    if (newBuf && oldBuffer && copySize > 0) {
 204       memcpy(newBuf, oldBuffer, copySize);
 205    }
 206
 207    _mesa_align_free(oldBuffer);
 208    return newBuf;
 209 #endif
 210 }
 211
 212 /*@}*/
 213
 214
 215 /**********************************************************************/
 216 /** \name Math */
 217 /*@{*/
 218
 219
 220 #ifndef HAVE___BUILTIN_FFS
 221 /**
 222  * Find the first bit set in a word.
 223  */
 224 int
 225 ffs(int i)
 226 {
 227    register int bit = 0;
 228    if (i != 0) {
 229       if ((i & 0xffff) == 0) {
 230          bit += 16;
 231          i >>= 16;
 232       }
 233       if ((i & 0xff) == 0) {
 234          bit += 8;
 235          i >>= 8;
 236       }
 237       if ((i & 0xf) == 0) {
 238          bit += 4;
 239          i >>= 4;
 240       }
 241       while ((i & 1) == 0) {
 242          bit++;
 243          i >>= 1;
 244       }
 245       bit++;
 246    }
 247    return bit;
 248 }
 249 #endif
 250
 251 #ifndef HAVE___BUILTIN_FFSLL
 252 /**
 253  * Find position of first bit set in given value.
 254  * XXX Warning: this function can only be used on 64-bit systems!
 255  * \return  position of least-significant bit set, starting at 1, return zero
 256  *          if no bits set.
 257  */
 258 int
 259 ffsll(long long int val)
 260 {
 261    int bit;
 262
 263    assert(sizeof(val) == 8);
 264
 265    bit = ffs((int) val);
 266    if (bit != 0)
 267       return bit;
 268
 269    bit = ffs((int) (val >> 32));
 270    if (bit != 0)
 271       return 32 + bit;
 272
 273    return 0;
 274 }
 275 #endif
 276
 277
 278 #ifndef HAVE___BUILTIN_POPCOUNT
 279 /**
 280  * Return number of bits set in given GLuint.
 281  */
 282 unsigned int
 283 _mesa_bitcount(unsigned int n)
 284 {
 285    unsigned int bits;
 286    for (bits = 0; n > 0; n = n >> 1) {
 287       bits += (n & 1);
 288    }
 289    return bits;
 290 }
 291 #endif
 292
 293 #ifndef HAVE___BUILTIN_POPCOUNTLL
 294 /**
 295  * Return number of bits set in given 64-bit uint.
 296  */
 297 unsigned int
 298 _mesa_bitcount_64(uint64_t n)
 299 {
 300    unsigned int bits;
 301    for (bits = 0; n > 0; n = n >> 1) {
 302       bits += (n & 1);
 303    }
 304    return bits;
 305 }
 306 #endif
 307
 308
 309 /* Using C99 rounding functions for roundToEven() implementation is
 310  * difficult, because round(), rint, and nearbyint() are affected by
 311  * fesetenv(), which the application may have done for its own
 312  * purposes.  Mesa's IROUND macro is close to what we want, but it
 313  * rounds away from 0 on n + 0.5.
 314  */
 315 int
 316 _mesa_round_to_even(float val)
 317 {
 318    int rounded = IROUND(val);
 319
 320    if (val - floor(val) == 0.5) {
 321       if (rounded % 2 != 0)
 322          rounded += val > 0 ? -1 : 1;
 323    }
 324
 325    return rounded;
 326 }
 327
 328
 329 /**
 330  * Convert a 4-byte float to a 2-byte half float.
 331  *
 332  * Not all float32 values can be represented exactly as a float16 value. We
 333  * round such intermediate float32 values to the nearest float16. When the
 334  * float32 lies exactly between to float16 values, we round to the one with
 335  * an even mantissa.
 336  *
 337  * This rounding behavior has several benefits:
 338  *   - It has no sign bias.
 339  *
 340  *   - It reproduces the behavior of real hardware: opcode F32TO16 in Intel's
 341  *     GPU ISA.
 342  *
 343  *   - By reproducing the behavior of the GPU (at least on Intel hardware),
 344  *     compile-time evaluation of constant packHalf2x16 GLSL expressions will
 345  *     result in the same value as if the expression were executed on the GPU.
 346  */
 347 GLhalfARB
 348 _mesa_float_to_half(float val)
 349 {
 350    const fi_type fi = {val};
 351    const int flt_m = fi.i & 0x7fffff;
 352    const int flt_e = (fi.i >> 23) & 0xff;
 353    const int flt_s = (fi.i >> 31) & 0x1;
 354    int s, e, m = 0;
 355    GLhalfARB result;
 356
 357    /* sign bit */
 358    s = flt_s;
 359
 360    /* handle special cases */
 361    if ((flt_e == 0) && (flt_m == 0)) {
 362       /* zero */
 363       /* m = 0; - already set */
 364       e = 0;
 365    }
 366    else if ((flt_e == 0) && (flt_m != 0)) {
 367       /* denorm -- denorm float maps to 0 half */
 368       /* m = 0; - already set */
 369       e = 0;
 370    }
 371    else if ((flt_e == 0xff) && (flt_m == 0)) {
 372       /* infinity */
 373       /* m = 0; - already set */
 374       e = 31;
 375    }
 376    else if ((flt_e == 0xff) && (flt_m != 0)) {
 377       /* NaN */
 378       m = 1;
 379       e = 31;
 380    }
 381    else {
 382       /* regular number */
 383       const int new_exp = flt_e - 127;
 384       if (new_exp < -14) {
 385          /* The float32 lies in the range (0.0, min_normal16) and is rounded
 386           * to a nearby float16 value. The result will be either zero, subnormal,
 387           * or normal.
 388           */
 389          e = 0;
 390          m = _mesa_round_to_even((1 << 24) * fabsf(fi.f));
 391       }
 392       else if (new_exp > 15) {
 393          /* map this value to infinity */
 394          /* m = 0; - already set */
 395          e = 31;
 396       }
 397       else {
 398          /* The float32 lies in the range
 399           *   [min_normal16, max_normal16 + max_step16)
 400           * and is rounded to a nearby float16 value. The result will be
 401           * either normal or infinite.
 402           */
 403          e = new_exp + 15;
 404          m = _mesa_round_to_even(flt_m / (float) (1 << 13));
 405       }
 406    }
 407
 408    assert(0 <= m && m <= 1024);
 409    if (m == 1024) {
 410       /* The float32 was rounded upwards into the range of the next exponent,
 411        * so bump the exponent. This correctly handles the case where f32
 412        * should be rounded up to float16 infinity.
 413        */
 414       ++e;
 415       m = 0;
 416    }
 417
 418    result = (s << 15) | (e << 10) | m;
 419    return result;
 420 }
 421
 422
 423 /**
 424  * Convert a 2-byte half float to a 4-byte float.
 425  * Based on code from:
 426  * http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/008786.html
 427  */
 428 float
 429 _mesa_half_to_float(GLhalfARB val)
 430 {
 431    /* XXX could also use a 64K-entry lookup table */
 432    const int m = val & 0x3ff;
 433    const int e = (val >> 10) & 0x1f;
 434    const int s = (val >> 15) & 0x1;
 435    int flt_m, flt_e, flt_s;
 436    fi_type fi;
 437    float result;
 438
 439    /* sign bit */
 440    flt_s = s;
 441
 442    /* handle special cases */
 443    if ((e == 0) && (m == 0)) {
 444       /* zero */
 445       flt_m = 0;
 446       flt_e = 0;
 447    }
 448    else if ((e == 0) && (m != 0)) {
 449       /* denorm -- denorm half will fit in non-denorm single */
 450       const float half_denorm = 1.0f / 16384.0f; /* 2^-14 */
 451       float mantissa = ((float) (m)) / 1024.0f;
 452       float sign = s ? -1.0f : 1.0f;
 453       return sign * mantissa * half_denorm;
 454    }
 455    else if ((e == 31) && (m == 0)) {
 456       /* infinity */
 457       flt_e = 0xff;
 458       flt_m = 0;
 459    }
 460    else if ((e == 31) && (m != 0)) {
 461       /* NaN */
 462       flt_e = 0xff;
 463       flt_m = 1;
 464    }
 465    else {
 466       /* regular */
 467       flt_e = e + 112;
 468       flt_m = m << 13;
 469    }
 470
 471    fi.i = (flt_s << 31) | (flt_e << 23) | flt_m;
 472    result = fi.f;
 473    return result;
 474 }
 475
 476 /*@}*/
 477
 478
 479 /**********************************************************************/
 480 /** \name String */
 481 /*@{*/
 482
 483 /**
 484  * Implemented using malloc() and strcpy.
 485  * Note that NULL is handled accordingly.
 486  */
 487 char *
 488 _mesa_strdup( const char *s )
 489 {
 490    if (s) {
 491       size_t l = strlen(s);
 492       char *s2 = malloc(l + 1);
 493       if (s2)
 494          strcpy(s2, s);
 495       return s2;
 496    }
 497    else {
 498       return NULL;
 499    }
 500 }
 501
 502 /** Compute simple checksum/hash for a string */
 503 unsigned int
 504 _mesa_str_checksum(const char *str)
 505 {
 506    /* This could probably be much better */
 507    unsigned int sum, i;
 508    const char *c;
 509    sum = i = 1;
 510    for (c = str; *c; c++, i++)
 511       sum += *c * (i % 100);
 512    return sum + i;
 513 }
 514
 515
 516 /*@}*/
 517
 518
 519 /** Needed due to #ifdef's, above. */
 520 int
 521 _mesa_vsnprintf(char *str, size_t size, const char *fmt, va_list args)
 522 {
 523    return vsnprintf( str, size, fmt, args);
 524 }
 525
 526 /** Wrapper around vsnprintf() */
 527 int
 528 _mesa_snprintf( char *str, size_t size, const char *fmt, ... )
 529 {
 530    int r;
 531    va_list args;
 532    va_start( args, fmt );
 533    r = vsnprintf( str, size, fmt, args );
 534    va_end( args );
 535    return r;
 536 }
 537
 538