4f5a2d11fa97d1e3c78c4d19aae07366b702f012
3 * Standard C library function wrappers.
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,
10 * Some drivers will want to override/replace this file with something
11 * specialized, but that'll be rare.
13 * Eventually, I want to move roll the glheader.h file into this.
15 * \todo Functions still needed:
22 * Mesa 3-D graphics library
24 * Copyright (C) 1999-2007 Brian Paul All Rights Reserved.
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:
33 * The above copyright notice and this permission notice shall be included
34 * in all copies or substantial portions of the Software.
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.
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
);
66 /**********************************************************************/
71 * Allocate aligned memory.
73 * \param bytes number of bytes to allocate.
74 * \param alignment alignment (must be greater than zero).
76 * Allocates extra memory to accommodate rounding up the address for
77 * alignment and to record the real malloc address.
79 * \sa _mesa_align_free().
82 _mesa_align_malloc(size_t bytes
, unsigned long alignment
)
84 #if defined(HAVE_POSIX_MEMALIGN)
86 int err
= posix_memalign(& mem
, alignment
, bytes
);
90 #elif defined(_WIN32) && defined(_MSC_VER)
91 return _aligned_malloc(bytes
, alignment
);
95 ASSERT( alignment
> 0 );
97 ptr
= (uintptr_t)malloc(bytes
+ alignment
+ sizeof(void *));
101 buf
= (ptr
+ alignment
+ sizeof(void *)) & ~(uintptr_t)(alignment
- 1);
102 *(uintptr_t *)(buf
- sizeof(void *)) = ptr
;
105 /* mark the non-aligned area */
106 while ( ptr
< buf
- sizeof(void *) ) {
107 *(unsigned long *)ptr
= 0xcdcdcdcd;
108 ptr
+= sizeof(unsigned long);
113 #endif /* defined(HAVE_POSIX_MEMALIGN) */
117 * Same as _mesa_align_malloc(), but using calloc(1, ) instead of
121 _mesa_align_calloc(size_t bytes
, unsigned long alignment
)
123 #if defined(HAVE_POSIX_MEMALIGN)
126 mem
= _mesa_align_malloc(bytes
, alignment
);
128 (void) memset(mem
, 0, bytes
);
132 #elif defined(_WIN32) && defined(_MSC_VER)
135 mem
= _aligned_malloc(bytes
, alignment
);
137 (void) memset(mem
, 0, bytes
);
144 ASSERT( alignment
> 0 );
146 ptr
= (uintptr_t)calloc(1, bytes
+ alignment
+ sizeof(void *));
150 buf
= (ptr
+ alignment
+ sizeof(void *)) & ~(uintptr_t)(alignment
- 1);
151 *(uintptr_t *)(buf
- sizeof(void *)) = ptr
;
154 /* mark the non-aligned area */
155 while ( ptr
< buf
- sizeof(void *) ) {
156 *(unsigned long *)ptr
= 0xcdcdcdcd;
157 ptr
+= sizeof(unsigned long);
162 #endif /* defined(HAVE_POSIX_MEMALIGN) */
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.
175 _mesa_align_free(void *ptr
)
177 #if defined(HAVE_POSIX_MEMALIGN)
179 #elif defined(_WIN32) && defined(_MSC_VER)
183 void **cubbyHole
= (void **) ((char *) ptr
- sizeof(void *));
184 void *realAddr
= *cubbyHole
;
187 #endif /* defined(HAVE_POSIX_MEMALIGN) */
191 * Reallocate memory, with alignment.
194 _mesa_align_realloc(void *oldBuffer
, size_t oldSize
, size_t newSize
,
195 unsigned long alignment
)
197 #if defined(_WIN32) && defined(_MSC_VER)
199 return _aligned_realloc(oldBuffer
, newSize
, alignment
);
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
);
207 _mesa_align_free(oldBuffer
);
215 /**********************************************************************/
220 #ifndef HAVE___BUILTIN_FFS
222 * Find the first bit set in a word.
227 register int bit
= 0;
229 if ((i
& 0xffff) == 0) {
233 if ((i
& 0xff) == 0) {
237 if ((i
& 0xf) == 0) {
241 while ((i
& 1) == 0) {
251 #ifndef HAVE___BUILTIN_FFSLL
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
259 ffsll(long long int val
)
263 assert(sizeof(val
) == 8);
265 bit
= ffs((int) val
);
269 bit
= ffs((int) (val
>> 32));
278 #ifndef HAVE___BUILTIN_POPCOUNT
280 * Return number of bits set in given GLuint.
283 _mesa_bitcount(unsigned int n
)
286 for (bits
= 0; n
> 0; n
= n
>> 1) {
293 #ifndef HAVE___BUILTIN_POPCOUNTLL
295 * Return number of bits set in given 64-bit uint.
298 _mesa_bitcount_64(uint64_t n
)
301 for (bits
= 0; n
> 0; n
= n
>> 1) {
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.
316 _mesa_round_to_even(float val
)
318 int rounded
= IROUND(val
);
320 if (val
- floor(val
) == 0.5) {
321 if (rounded
% 2 != 0)
322 rounded
+= val
> 0 ? -1 : 1;
330 * Convert a 4-byte float to a 2-byte half float.
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
337 * This rounding behavior has several benefits:
338 * - It has no sign bias.
340 * - It reproduces the behavior of real hardware: opcode F32TO16 in Intel's
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.
348 _mesa_float_to_half(float val
)
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;
360 /* handle special cases */
361 if ((flt_e
== 0) && (flt_m
== 0)) {
363 /* m = 0; - already set */
366 else if ((flt_e
== 0) && (flt_m
!= 0)) {
367 /* denorm -- denorm float maps to 0 half */
368 /* m = 0; - already set */
371 else if ((flt_e
== 0xff) && (flt_m
== 0)) {
373 /* m = 0; - already set */
376 else if ((flt_e
== 0xff) && (flt_m
!= 0)) {
383 const int new_exp
= flt_e
- 127;
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,
390 m
= _mesa_round_to_even((1 << 24) * fabsf(fi
.f
));
392 else if (new_exp
> 15) {
393 /* map this value to infinity */
394 /* m = 0; - already set */
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.
404 m
= _mesa_round_to_even(flt_m
/ (float) (1 << 13));
408 assert(0 <= m
&& 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.
418 result
= (s
<< 15) | (e
<< 10) | m
;
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
429 _mesa_half_to_float(GLhalfARB val
)
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
;
442 /* handle special cases */
443 if ((e
== 0) && (m
== 0)) {
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
;
455 else if ((e
== 31) && (m
== 0)) {
460 else if ((e
== 31) && (m
!= 0)) {
471 fi
.i
= (flt_s
<< 31) | (flt_e
<< 23) | flt_m
;
479 /**********************************************************************/
484 * Implemented using malloc() and strcpy.
485 * Note that NULL is handled accordingly.
488 _mesa_strdup( const char *s
)
491 size_t l
= strlen(s
);
492 char *s2
= malloc(l
+ 1);
502 /** Compute simple checksum/hash for a string */
504 _mesa_str_checksum(const char *str
)
506 /* This could probably be much better */
510 for (c
= str
; *c
; c
++, i
++)
511 sum
+= *c
* (i
% 100);
519 /** Needed due to #ifdef's, above. */
521 _mesa_vsnprintf(char *str
, size_t size
, const char *fmt
, va_list args
)
523 return vsnprintf( str
, size
, fmt
, args
);
526 /** Wrapper around vsnprintf() */
528 _mesa_snprintf( char *str
, size_t size
, const char *fmt
, ... )
532 va_start( args
, fmt
);
533 r
= vsnprintf( str
, size
, fmt
, args
);