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