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.
48 #include "util/rounding.h" /* for _mesa_roundeven */
63 #define vsnprintf _vsnprintf
64 #elif defined(__IBMC__) || defined(__IBMCPP__)
65 extern int vsnprintf(char *str
, size_t count
, const char *fmt
, va_list arg
);
68 /**********************************************************************/
73 * Allocate aligned memory.
75 * \param bytes number of bytes to allocate.
76 * \param alignment alignment (must be greater than zero).
78 * Allocates extra memory to accommodate rounding up the address for
79 * alignment and to record the real malloc address.
81 * \sa _mesa_align_free().
84 _mesa_align_malloc(size_t bytes
, unsigned long alignment
)
86 #if defined(HAVE_POSIX_MEMALIGN)
88 int err
= posix_memalign(& mem
, alignment
, bytes
);
92 #elif defined(_WIN32) && defined(_MSC_VER)
93 return _aligned_malloc(bytes
, alignment
);
97 assert( alignment
> 0 );
99 ptr
= (uintptr_t)malloc(bytes
+ alignment
+ sizeof(void *));
103 buf
= (ptr
+ alignment
+ sizeof(void *)) & ~(uintptr_t)(alignment
- 1);
104 *(uintptr_t *)(buf
- sizeof(void *)) = ptr
;
107 /* mark the non-aligned area */
108 while ( ptr
< buf
- sizeof(void *) ) {
109 *(unsigned long *)ptr
= 0xcdcdcdcd;
110 ptr
+= sizeof(unsigned long);
115 #endif /* defined(HAVE_POSIX_MEMALIGN) */
119 * Same as _mesa_align_malloc(), but using calloc(1, ) instead of
123 _mesa_align_calloc(size_t bytes
, unsigned long alignment
)
125 #if defined(HAVE_POSIX_MEMALIGN)
128 mem
= _mesa_align_malloc(bytes
, alignment
);
130 (void) memset(mem
, 0, bytes
);
134 #elif defined(_WIN32) && defined(_MSC_VER)
137 mem
= _aligned_malloc(bytes
, alignment
);
139 (void) memset(mem
, 0, bytes
);
146 assert( alignment
> 0 );
148 ptr
= (uintptr_t)calloc(1, bytes
+ alignment
+ sizeof(void *));
152 buf
= (ptr
+ alignment
+ sizeof(void *)) & ~(uintptr_t)(alignment
- 1);
153 *(uintptr_t *)(buf
- sizeof(void *)) = ptr
;
156 /* mark the non-aligned area */
157 while ( ptr
< buf
- sizeof(void *) ) {
158 *(unsigned long *)ptr
= 0xcdcdcdcd;
159 ptr
+= sizeof(unsigned long);
164 #endif /* defined(HAVE_POSIX_MEMALIGN) */
168 * Free memory which was allocated with either _mesa_align_malloc()
169 * or _mesa_align_calloc().
170 * \param ptr pointer to the memory to be freed.
171 * The actual address to free is stored in the word immediately before the
172 * address the client sees.
173 * Note that it is legal to pass NULL pointer to this function and will be
174 * handled accordingly.
177 _mesa_align_free(void *ptr
)
179 #if defined(HAVE_POSIX_MEMALIGN)
181 #elif defined(_WIN32) && defined(_MSC_VER)
185 void **cubbyHole
= (void **) ((char *) ptr
- sizeof(void *));
186 void *realAddr
= *cubbyHole
;
189 #endif /* defined(HAVE_POSIX_MEMALIGN) */
193 * Reallocate memory, with alignment.
196 _mesa_align_realloc(void *oldBuffer
, size_t oldSize
, size_t newSize
,
197 unsigned long alignment
)
199 #if defined(_WIN32) && defined(_MSC_VER)
201 return _aligned_realloc(oldBuffer
, newSize
, alignment
);
203 const size_t copySize
= (oldSize
< newSize
) ? oldSize
: newSize
;
204 void *newBuf
= _mesa_align_malloc(newSize
, alignment
);
205 if (newBuf
&& oldBuffer
&& copySize
> 0) {
206 memcpy(newBuf
, oldBuffer
, copySize
);
209 _mesa_align_free(oldBuffer
);
217 /**********************************************************************/
222 #ifndef HAVE___BUILTIN_FFS
224 * Find the first bit set in a word.
229 register int bit
= 0;
231 if ((i
& 0xffff) == 0) {
235 if ((i
& 0xff) == 0) {
239 if ((i
& 0xf) == 0) {
243 while ((i
& 1) == 0) {
253 #ifndef HAVE___BUILTIN_FFSLL
255 * Find position of first bit set in given value.
256 * XXX Warning: this function can only be used on 64-bit systems!
257 * \return position of least-significant bit set, starting at 1, return zero
261 ffsll(long long int val
)
265 assert(sizeof(val
) == 8);
267 bit
= ffs((int) val
);
271 bit
= ffs((int) (val
>> 32));
280 #ifndef HAVE___BUILTIN_POPCOUNT
282 * Return number of bits set in given GLuint.
285 _mesa_bitcount(unsigned int n
)
288 for (bits
= 0; n
> 0; n
= n
>> 1) {
295 #ifndef HAVE___BUILTIN_POPCOUNTLL
297 * Return number of bits set in given 64-bit uint.
300 _mesa_bitcount_64(uint64_t n
)
303 for (bits
= 0; n
> 0; n
= n
>> 1) {
312 * Convert a 4-byte float to a 2-byte half float.
314 * Not all float32 values can be represented exactly as a float16 value. We
315 * round such intermediate float32 values to the nearest float16. When the
316 * float32 lies exactly between to float16 values, we round to the one with
319 * This rounding behavior has several benefits:
320 * - It has no sign bias.
322 * - It reproduces the behavior of real hardware: opcode F32TO16 in Intel's
325 * - By reproducing the behavior of the GPU (at least on Intel hardware),
326 * compile-time evaluation of constant packHalf2x16 GLSL expressions will
327 * result in the same value as if the expression were executed on the GPU.
330 _mesa_float_to_half(float val
)
332 const fi_type fi
= {val
};
333 const int flt_m
= fi
.i
& 0x7fffff;
334 const int flt_e
= (fi
.i
>> 23) & 0xff;
335 const int flt_s
= (fi
.i
>> 31) & 0x1;
342 /* handle special cases */
343 if ((flt_e
== 0) && (flt_m
== 0)) {
345 /* m = 0; - already set */
348 else if ((flt_e
== 0) && (flt_m
!= 0)) {
349 /* denorm -- denorm float maps to 0 half */
350 /* m = 0; - already set */
353 else if ((flt_e
== 0xff) && (flt_m
== 0)) {
355 /* m = 0; - already set */
358 else if ((flt_e
== 0xff) && (flt_m
!= 0)) {
365 const int new_exp
= flt_e
- 127;
367 /* The float32 lies in the range (0.0, min_normal16) and is rounded
368 * to a nearby float16 value. The result will be either zero, subnormal,
372 m
= (int) _mesa_roundevenf((1 << 24) * fabsf(fi
.f
));
374 else if (new_exp
> 15) {
375 /* map this value to infinity */
376 /* m = 0; - already set */
380 /* The float32 lies in the range
381 * [min_normal16, max_normal16 + max_step16)
382 * and is rounded to a nearby float16 value. The result will be
383 * either normal or infinite.
386 m
= (int) _mesa_roundevenf(flt_m
/ (float) (1 << 13));
390 assert(0 <= m
&& m
<= 1024);
392 /* The float32 was rounded upwards into the range of the next exponent,
393 * so bump the exponent. This correctly handles the case where f32
394 * should be rounded up to float16 infinity.
400 result
= (s
<< 15) | (e
<< 10) | m
;
406 * Convert a 2-byte half float to a 4-byte float.
407 * Based on code from:
408 * http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/008786.html
411 _mesa_half_to_float(GLhalfARB val
)
413 /* XXX could also use a 64K-entry lookup table */
414 const int m
= val
& 0x3ff;
415 const int e
= (val
>> 10) & 0x1f;
416 const int s
= (val
>> 15) & 0x1;
417 int flt_m
, flt_e
, flt_s
;
424 /* handle special cases */
425 if ((e
== 0) && (m
== 0)) {
430 else if ((e
== 0) && (m
!= 0)) {
431 /* denorm -- denorm half will fit in non-denorm single */
432 const float half_denorm
= 1.0f
/ 16384.0f
; /* 2^-14 */
433 float mantissa
= ((float) (m
)) / 1024.0f
;
434 float sign
= s
? -1.0f
: 1.0f
;
435 return sign
* mantissa
* half_denorm
;
437 else if ((e
== 31) && (m
== 0)) {
442 else if ((e
== 31) && (m
!= 0)) {
453 fi
.i
= (flt_s
<< 31) | (flt_e
<< 23) | flt_m
;
461 /**********************************************************************/
466 /** Compute simple checksum/hash for a string */
468 _mesa_str_checksum(const char *str
)
470 /* This could probably be much better */
474 for (c
= str
; *c
; c
++, i
++)
475 sum
+= *c
* (i
% 100);
483 /** Needed due to #ifdef's, above. */
485 _mesa_vsnprintf(char *str
, size_t size
, const char *fmt
, va_list args
)
487 return vsnprintf( str
, size
, fmt
, args
);
490 /** Wrapper around vsnprintf() */
492 _mesa_snprintf( char *str
, size_t size
, const char *fmt
, ... )
496 va_start( args
, fmt
);
497 r
= vsnprintf( str
, size
, fmt
, args
);