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 <stdio.h>
  46 #include <stdarg.h>
  47 #include "c99_math.h"
  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  * Note that it is legal to pass NULL pointer to this function and will be
 173  * handled accordingly.
 174  */
 175 void
 176 _mesa_align_free(void *ptr)
 177 {
 178 #if defined(HAVE_POSIX_MEMALIGN)
 179    free(ptr);
 180 #elif defined(_WIN32) && defined(_MSC_VER)
 181    _aligned_free(ptr);
 182 #else
 183    if (ptr) {
 184       void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
 185       void *realAddr = *cubbyHole;
 186       free(realAddr);
 187    }
 188 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 189 }
 190
 191 /**
 192  * Reallocate memory, with alignment.
 193  */
 194 void *
 195 _mesa_align_realloc(void *oldBuffer, size_t oldSize, size_t newSize,
 196                     unsigned long alignment)
 197 {
 198 #if defined(_WIN32) && defined(_MSC_VER)
 199    (void) oldSize;
 200    return _aligned_realloc(oldBuffer, newSize, alignment);
 201 #else
 202    const size_t copySize = (oldSize < newSize) ? oldSize : newSize;
 203    void *newBuf = _mesa_align_malloc(newSize, alignment);
 204    if (newBuf && oldBuffer && copySize > 0) {
 205       memcpy(newBuf, oldBuffer, copySize);
 206    }
 207
 208    _mesa_align_free(oldBuffer);
 209    return newBuf;
 210 #endif
 211 }
 212
 213 /*@}*/
 214
 215
 216 /**********************************************************************/
 217 /** \name Math */
 218 /*@{*/
 219
 220
 221 #ifndef HAVE___BUILTIN_FFS
 222 /**
 223  * Find the first bit set in a word.
 224  */
 225 int
 226 ffs(int i)
 227 {
 228    register int bit = 0;
 229    if (i != 0) {
 230       if ((i & 0xffff) == 0) {
 231          bit += 16;
 232          i >>= 16;
 233       }
 234       if ((i & 0xff) == 0) {
 235          bit += 8;
 236          i >>= 8;
 237       }
 238       if ((i & 0xf) == 0) {
 239          bit += 4;
 240          i >>= 4;
 241       }
 242       while ((i & 1) == 0) {
 243          bit++;
 244          i >>= 1;
 245       }
 246       bit++;
 247    }
 248    return bit;
 249 }
 250 #endif
 251
 252 #ifndef HAVE___BUILTIN_FFSLL
 253 /**
 254  * Find position of first bit set in given value.
 255  * XXX Warning: this function can only be used on 64-bit systems!
 256  * \return  position of least-significant bit set, starting at 1, return zero
 257  *          if no bits set.
 258  */
 259 int
 260 ffsll(long long int val)
 261 {
 262    int bit;
 263
 264    assert(sizeof(val) == 8);
 265
 266    bit = ffs((int) val);
 267    if (bit != 0)
 268       return bit;
 269
 270    bit = ffs((int) (val >> 32));
 271    if (bit != 0)
 272       return 32 + bit;
 273
 274    return 0;
 275 }
 276 #endif
 277
 278
 279 #ifndef HAVE___BUILTIN_POPCOUNT
 280 /**
 281  * Return number of bits set in given GLuint.
 282  */
 283 unsigned int
 284 _mesa_bitcount(unsigned int n)
 285 {
 286    unsigned int bits;
 287    for (bits = 0; n > 0; n = n >> 1) {
 288       bits += (n & 1);
 289    }
 290    return bits;
 291 }
 292 #endif
 293
 294 #ifndef HAVE___BUILTIN_POPCOUNTLL
 295 /**
 296  * Return number of bits set in given 64-bit uint.
 297  */
 298 unsigned int
 299 _mesa_bitcount_64(uint64_t n)
 300 {
 301    unsigned int bits;
 302    for (bits = 0; n > 0; n = n >> 1) {
 303       bits += (n & 1);
 304    }
 305    return bits;
 306 }
 307 #endif
 308
 309
 310 /* Using C99 rounding functions for roundToEven() implementation is
 311  * difficult, because round(), rint, and nearbyint() are affected by
 312  * fesetenv(), which the application may have done for its own
 313  * purposes.  Mesa's IROUND macro is close to what we want, but it
 314  * rounds away from 0 on n + 0.5.
 315  */
 316 int
 317 _mesa_round_to_even(float val)
 318 {
 319    int rounded = IROUND(val);
 320
 321    if (val - floor(val) == 0.5) {
 322       if (rounded % 2 != 0)
 323          rounded += val > 0 ? -1 : 1;
 324    }
 325
 326    return rounded;
 327 }
 328
 329
 330 /**
 331  * Convert a 4-byte float to a 2-byte half float.
 332  *
 333  * Not all float32 values can be represented exactly as a float16 value. We
 334  * round such intermediate float32 values to the nearest float16. When the
 335  * float32 lies exactly between to float16 values, we round to the one with
 336  * an even mantissa.
 337  *
 338  * This rounding behavior has several benefits:
 339  *   - It has no sign bias.
 340  *
 341  *   - It reproduces the behavior of real hardware: opcode F32TO16 in Intel's
 342  *     GPU ISA.
 343  *
 344  *   - By reproducing the behavior of the GPU (at least on Intel hardware),
 345  *     compile-time evaluation of constant packHalf2x16 GLSL expressions will
 346  *     result in the same value as if the expression were executed on the GPU.
 347  */
 348 GLhalfARB
 349 _mesa_float_to_half(float val)
 350 {
 351    const fi_type fi = {val};
 352    const int flt_m = fi.i & 0x7fffff;
 353    const int flt_e = (fi.i >> 23) & 0xff;
 354    const int flt_s = (fi.i >> 31) & 0x1;
 355    int s, e, m = 0;
 356    GLhalfARB result;
 357
 358    /* sign bit */
 359    s = flt_s;
 360
 361    /* handle special cases */
 362    if ((flt_e == 0) && (flt_m == 0)) {
 363       /* zero */
 364       /* m = 0; - already set */
 365       e = 0;
 366    }
 367    else if ((flt_e == 0) && (flt_m != 0)) {
 368       /* denorm -- denorm float maps to 0 half */
 369       /* m = 0; - already set */
 370       e = 0;
 371    }
 372    else if ((flt_e == 0xff) && (flt_m == 0)) {
 373       /* infinity */
 374       /* m = 0; - already set */
 375       e = 31;
 376    }
 377    else if ((flt_e == 0xff) && (flt_m != 0)) {
 378       /* NaN */
 379       m = 1;
 380       e = 31;
 381    }
 382    else {
 383       /* regular number */
 384       const int new_exp = flt_e - 127;
 385       if (new_exp < -14) {
 386          /* The float32 lies in the range (0.0, min_normal16) and is rounded
 387           * to a nearby float16 value. The result will be either zero, subnormal,
 388           * or normal.
 389           */
 390          e = 0;
 391          m = _mesa_round_to_even((1 << 24) * fabsf(fi.f));
 392       }
 393       else if (new_exp > 15) {
 394          /* map this value to infinity */
 395          /* m = 0; - already set */
 396          e = 31;
 397       }
 398       else {
 399          /* The float32 lies in the range
 400           *   [min_normal16, max_normal16 + max_step16)
 401           * and is rounded to a nearby float16 value. The result will be
 402           * either normal or infinite.
 403           */
 404          e = new_exp + 15;
 405          m = _mesa_round_to_even(flt_m / (float) (1 << 13));
 406       }
 407    }
 408
 409    assert(0 <= m && m <= 1024);
 410    if (m == 1024) {
 411       /* The float32 was rounded upwards into the range of the next exponent,
 412        * so bump the exponent. This correctly handles the case where f32
 413        * should be rounded up to float16 infinity.
 414        */
 415       ++e;
 416       m = 0;
 417    }
 418
 419    result = (s << 15) | (e << 10) | m;
 420    return result;
 421 }
 422
 423
 424 /**
 425  * Convert a 2-byte half float to a 4-byte float.
 426  * Based on code from:
 427  * http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/008786.html
 428  */
 429 float
 430 _mesa_half_to_float(GLhalfARB val)
 431 {
 432    /* XXX could also use a 64K-entry lookup table */
 433    const int m = val & 0x3ff;
 434    const int e = (val >> 10) & 0x1f;
 435    const int s = (val >> 15) & 0x1;
 436    int flt_m, flt_e, flt_s;
 437    fi_type fi;
 438    float result;
 439
 440    /* sign bit */
 441    flt_s = s;
 442
 443    /* handle special cases */
 444    if ((e == 0) && (m == 0)) {
 445       /* zero */
 446       flt_m = 0;
 447       flt_e = 0;
 448    }
 449    else if ((e == 0) && (m != 0)) {
 450       /* denorm -- denorm half will fit in non-denorm single */
 451       const float half_denorm = 1.0f / 16384.0f; /* 2^-14 */
 452       float mantissa = ((float) (m)) / 1024.0f;
 453       float sign = s ? -1.0f : 1.0f;
 454       return sign * mantissa * half_denorm;
 455    }
 456    else if ((e == 31) && (m == 0)) {
 457       /* infinity */
 458       flt_e = 0xff;
 459       flt_m = 0;
 460    }
 461    else if ((e == 31) && (m != 0)) {
 462       /* NaN */
 463       flt_e = 0xff;
 464       flt_m = 1;
 465    }
 466    else {
 467       /* regular */
 468       flt_e = e + 112;
 469       flt_m = m << 13;
 470    }
 471
 472    fi.i = (flt_s << 31) | (flt_e << 23) | flt_m;
 473    result = fi.f;
 474    return result;
 475 }
 476
 477 /*@}*/
 478
 479
 480 /**********************************************************************/
 481 /** \name String */
 482 /*@{*/
 483
 484 /**
 485  * Implemented using malloc() and strcpy.
 486  * Note that NULL is handled accordingly.
 487  */
 488 char *
 489 _mesa_strdup( const char *s )
 490 {
 491    if (s) {
 492       size_t l = strlen(s);
 493       char *s2 = malloc(l + 1);
 494       if (s2)
 495          strcpy(s2, s);
 496       return s2;
 497    }
 498    else {
 499       return NULL;
 500    }
 501 }
 502
 503 /** Compute simple checksum/hash for a string */
 504 unsigned int
 505 _mesa_str_checksum(const char *str)
 506 {
 507    /* This could probably be much better */
 508    unsigned int sum, i;
 509    const char *c;
 510    sum = i = 1;
 511    for (c = str; *c; c++, i++)
 512       sum += *c * (i % 100);
 513    return sum + i;
 514 }
 515
 516
 517 /*@}*/
 518
 519
 520 /** Needed due to #ifdef's, above. */
 521 int
 522 _mesa_vsnprintf(char *str, size_t size, const char *fmt, va_list args)
 523 {
 524    return vsnprintf( str, size, fmt, args);
 525 }
 526
 527 /** Wrapper around vsnprintf() */
 528 int
 529 _mesa_snprintf( char *str, size_t size, const char *fmt, ... )
 530 {
 531    int r;
 532    va_list args;
 533    va_start( args, fmt );
 534    r = vsnprintf( str, size, fmt, args );
 535    va_end( args );
 536    return r;
 537 }
 538
 539