425ef825d1c7371dbb68ab5b6520339438d45bf1
2 * Mesa 3-D graphics library
5 * Copyright (C) 1999-2006 Brian Paul All Rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
28 * Standard C library function wrappers.
30 * This file provides wrappers for all the standard C library functions
31 * like malloc(), free(), printf(), getenv(), etc.
39 /* XXX some of the stuff in glheader.h should be moved into this file.
42 #include <GL/internal/glcore.h>
50 /**********************************************************************/
51 /** \name General macros */
61 /**********************************************************************/
65 /** Allocate \p BYTES bytes */
66 #define MALLOC(BYTES) _mesa_malloc(BYTES)
67 /** Allocate and zero \p BYTES bytes */
68 #define CALLOC(BYTES) _mesa_calloc(BYTES)
69 /** Allocate a structure of type \p T */
70 #define MALLOC_STRUCT(T) (struct T *) _mesa_malloc(sizeof(struct T))
71 /** Allocate and zero a structure of type \p T */
72 #define CALLOC_STRUCT(T) (struct T *) _mesa_calloc(sizeof(struct T))
74 #define FREE(PTR) _mesa_free(PTR)
76 /** Allocate \p BYTES aligned at \p N bytes */
77 #define ALIGN_MALLOC(BYTES, N) _mesa_align_malloc(BYTES, N)
78 /** Allocate and zero \p BYTES bytes aligned at \p N bytes */
79 #define ALIGN_CALLOC(BYTES, N) _mesa_align_calloc(BYTES, N)
80 /** Allocate a structure of type \p T aligned at \p N bytes */
81 #define ALIGN_MALLOC_STRUCT(T, N) (struct T *) _mesa_align_malloc(sizeof(struct T), N)
82 /** Allocate and zero a structure of type \p T aligned at \p N bytes */
83 #define ALIGN_CALLOC_STRUCT(T, N) (struct T *) _mesa_align_calloc(sizeof(struct T), N)
84 /** Free aligned memory */
85 #define ALIGN_FREE(PTR) _mesa_align_free(PTR)
87 /** Copy \p BYTES bytes from \p SRC into \p DST */
88 #define MEMCPY( DST, SRC, BYTES) _mesa_memcpy(DST, SRC, BYTES)
89 /** Set \p N bytes in \p DST to \p VAL */
90 #define MEMSET( DST, VAL, N ) _mesa_memset(DST, VAL, N)
96 * For GL_ARB_vertex_buffer_object we need to treat vertex array pointers
97 * as offsets into buffer stores. Since the vertex array pointer and
98 * buffer store pointer are both pointers and we need to add them, we use
100 * Both pointers/offsets are expressed in bytes.
102 #define ADD_POINTERS(A, B) ( (GLubyte *) (A) + (uintptr_t) (B) )
106 * Sometimes we treat GLfloats as GLints. On x86 systems, moving a float
107 * as a int (thereby using integer registers instead of FP registers) is
108 * a performance win. Typically, this can be done with ordinary casts.
109 * But with gcc's -fstrict-aliasing flag (which defaults to on in gcc 3.0)
110 * these casts generate warnings.
111 * The following union typedef is used to solve that.
113 typedef union { GLfloat f
; GLint i
; } fi_type
;
117 /**********************************************************************
121 #define MAX_GLUSHORT 0xffff
122 #define MAX_GLUINT 0xffffffff
125 #define M_PI (3.1415926536)
129 #define M_E (2.7182818284590452354)
133 /* XXX this is a bit of a hack needed for compilation within XFree86 */
135 #define FLT_MIN (1.0e-37)
138 /* Degrees to radians conversion: */
139 #define DEG2RAD (M_PI/180.0)
143 *** USE_IEEE: Determine if we're using IEEE floating point
145 #if defined(__i386__) || defined(__386__) || defined(__sparc__) || \
146 defined(__s390x__) || defined(__powerpc__) || \
147 defined(__amd64__) || \
148 defined(ia64) || defined(__ia64__) || \
149 defined(__hppa__) || defined(hpux) || \
150 defined(__mips) || defined(_MIPS_ARCH) || \
151 defined(__arm__) || \
153 (defined(__alpha__) && (defined(__IEEE_FLOAT) || !defined(VMS)))
155 #define IEEE_ONE 0x3f800000
160 *** SQRTF: single-precision square root
162 #if 0 /* _mesa_sqrtf() not accurate enough - temporarily disabled */
163 # define SQRTF(X) _mesa_sqrtf(X)
164 #elif defined(XFree86LOADER) && defined(IN_MODULE) && !defined(NO_LIBCWRAPPER)
165 # define SQRTF(X) (float) xf86sqrt((float) (X))
167 # define SQRTF(X) (float) sqrt((float) (X))
172 *** INV_SQRTF: single-precision inverse square root
175 #define INV_SQRTF(X) _mesa_inv_sqrt(X)
177 #define INV_SQRTF(X) (1.0F / SQRTF(X)) /* this is faster on a P4 */
182 *** LOG2: Log base 2 of float
186 /* This is pretty fast, but not accurate enough (only 2 fractional bits).
187 * Based on code from http://www.stereopsis.com/log2.html
189 static INLINE GLfloat
LOG2(GLfloat x
)
191 const GLfloat y
= x
* x
* x
* x
;
192 const GLuint ix
= *((GLuint
*) &y
);
193 const GLuint exp
= (ix
>> 23) & 0xFF;
194 const GLint log2
= ((GLint
) exp
) - 127;
195 return (GLfloat
) log2
* (1.0 / 4.0); /* 4, because of x^4 above */
198 /* Pretty fast, and accurate.
199 * Based on code from http://www.flipcode.com/totd/
201 static INLINE GLfloat
LOG2(GLfloat val
)
206 log_2
= ((num
.i
>> 23) & 255) - 128;
207 num
.i
&= ~(255 << 23);
209 num
.f
= ((-1.0f
/3) * num
.f
+ 2) * num
.f
- 2.0f
/3;
210 return num
.f
+ log_2
;
212 #elif defined(XFree86LOADER) && defined(IN_MODULE) && !defined(NO_LIBCWRAPPER)
213 #define LOG2(x) ((GLfloat) (xf86log(x) * 1.442695))
216 * NOTE: log_base_2(x) = log(x) / log(2)
217 * NOTE: 1.442695 = 1/log(2).
219 #define LOG2(x) ((GLfloat) (log(x) * 1.442695F))
224 *** IS_INF_OR_NAN: test if float is infinite or NaN
227 static INLINE
int IS_INF_OR_NAN( float x
)
231 return !(int)((unsigned int)((tmp
.i
& 0x7fffffff)-0x7f800000) >> 31);
233 #elif defined(isfinite)
234 #define IS_INF_OR_NAN(x) (!isfinite(x))
235 #elif defined(finite)
236 #define IS_INF_OR_NAN(x) (!finite(x))
238 #define IS_INF_OR_NAN(x) (!finite(x))
239 #elif defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
240 #define IS_INF_OR_NAN(x) (!isfinite(x))
242 #define IS_INF_OR_NAN(x) (!finite(x))
247 *** IS_NEGATIVE: test if float is negative
249 #if defined(USE_IEEE)
250 static INLINE
int GET_FLOAT_BITS( float x
)
256 #define IS_NEGATIVE(x) (GET_FLOAT_BITS(x) < 0)
258 #define IS_NEGATIVE(x) (x < 0.0F)
263 *** DIFFERENT_SIGNS: test if two floats have opposite signs
265 #if defined(USE_IEEE)
266 #define DIFFERENT_SIGNS(x,y) ((GET_FLOAT_BITS(x) ^ GET_FLOAT_BITS(y)) & (1<<31))
268 /* Could just use (x*y<0) except for the flatshading requirements.
269 * Maybe there's a better way?
271 #define DIFFERENT_SIGNS(x,y) ((x) * (y) <= 0.0F && (x) - (y) != 0.0F)
276 *** CEILF: ceiling of float
277 *** FLOORF: floor of float
278 *** FABSF: absolute value of float
279 *** LOGF: the natural logarithm (base e) of the value
280 *** EXPF: raise e to the value
281 *** LDEXPF: multiply value by an integral power of two
282 *** FREXPF: extract mantissa and exponent from value
284 #if defined(XFree86LOADER) && defined(IN_MODULE) && !defined(NO_LIBCWRAPPER)
285 #define CEILF(x) ((GLfloat) xf86ceil(x))
286 #define FLOORF(x) ((GLfloat) xf86floor(x))
287 #define FABSF(x) ((GLfloat) xf86fabs(x))
288 #define LOGF(x) ((GLfloat) xf86log(x))
289 #define EXPF(x) ((GLfloat) xf86exp(x))
290 #define LDEXPF(x,y) ((GLfloat) xf86ldexp(x,y))
291 #define FREXPF(x,y) ((GLfloat) xf86frexp(x,y))
292 #elif defined(__gnu_linux__)
294 #define CEILF(x) ceilf(x)
295 #define FLOORF(x) floorf(x)
296 #define FABSF(x) fabsf(x)
297 #define LOGF(x) logf(x)
298 #define EXPF(x) expf(x)
299 #define LDEXPF(x,y) ldexpf(x,y)
300 #define FREXPF(x,y) frexpf(x,y)
302 #define CEILF(x) ((GLfloat) ceil(x))
303 #define FLOORF(x) ((GLfloat) floor(x))
304 #define FABSF(x) ((GLfloat) fabs(x))
305 #define LOGF(x) ((GLfloat) log(x))
306 #define EXPF(x) ((GLfloat) exp(x))
307 #define LDEXPF(x,y) ((GLfloat) ldexp(x,y))
308 #define FREXPF(x,y) ((GLfloat) frexp(x,y))
313 *** IROUND: return (as an integer) float rounded to nearest integer
315 #if defined(USE_SPARC_ASM) && defined(__GNUC__) && defined(__sparc__)
316 static INLINE
int iround(float f
)
319 __asm__ ("fstoi %1, %0" : "=f" (r
) : "f" (f
));
322 #define IROUND(x) iround(x)
323 #elif defined(USE_X86_ASM) && defined(__GNUC__) && defined(__i386__) && \
324 (!defined(__BEOS__) || (__GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ >= 95)))
325 static INLINE
int iround(float f
)
328 __asm__ ("fistpl %0" : "=m" (r
) : "t" (f
) : "st");
331 #define IROUND(x) iround(x)
332 #elif defined(USE_X86_ASM) && defined(__MSC__) && defined(__WIN32__)
333 static INLINE
int iround(float f
)
342 #define IROUND(x) iround(x)
343 #elif defined(__WATCOMC__) && defined(__386__)
344 long iround(float f
);
345 #pragma aux iround = \
347 "fistp dword ptr [esp]" \
352 #define IROUND(x) iround(x)
354 #define IROUND(f) ((int) (((f) >= 0.0F) ? ((f) + 0.5F) : ((f) - 0.5F)))
359 *** IROUND_POS: return (as an integer) positive float rounded to nearest int
362 #define IROUND_POS(f) (assert((f) >= 0.0F), IROUND(f))
364 #define IROUND_POS(f) (IROUND(f))
369 *** IFLOOR: return (as an integer) floor of float
371 #if defined(USE_X86_ASM) && defined(__GNUC__) && defined(__i386__)
373 * IEEE floor for computers that round to nearest or even.
374 * 'f' must be between -4194304 and 4194303.
375 * This floor operation is done by "(iround(f + .5) + iround(f - .5)) >> 1",
376 * but uses some IEEE specific tricks for better speed.
377 * Contributed by Josh Vanderhoof
379 static INLINE
int ifloor(float f
)
383 af
= (3 << 22) + 0.5 + (double)f
;
384 bf
= (3 << 22) + 0.5 - (double)f
;
385 /* GCC generates an extra fstp/fld without this. */
386 __asm__ ("fstps %0" : "=m" (ai
) : "t" (af
) : "st");
387 __asm__ ("fstps %0" : "=m" (bi
) : "t" (bf
) : "st");
388 return (ai
- bi
) >> 1;
390 #define IFLOOR(x) ifloor(x)
391 #elif defined(USE_IEEE)
392 static INLINE
int ifloor(float f
)
398 af
= (3 << 22) + 0.5 + (double)f
;
399 bf
= (3 << 22) + 0.5 - (double)f
;
400 u
.f
= (float) af
; ai
= u
.i
;
401 u
.f
= (float) bf
; bi
= u
.i
;
402 return (ai
- bi
) >> 1;
404 #define IFLOOR(x) ifloor(x)
406 static INLINE
int ifloor(float f
)
409 return (i
> f
) ? i
- 1 : i
;
411 #define IFLOOR(x) ifloor(x)
416 *** ICEIL: return (as an integer) ceiling of float
418 #if defined(USE_X86_ASM) && defined(__GNUC__) && defined(__i386__)
420 * IEEE ceil for computers that round to nearest or even.
421 * 'f' must be between -4194304 and 4194303.
422 * This ceil operation is done by "(iround(f + .5) + iround(f - .5) + 1) >> 1",
423 * but uses some IEEE specific tricks for better speed.
424 * Contributed by Josh Vanderhoof
426 static INLINE
int iceil(float f
)
430 af
= (3 << 22) + 0.5 + (double)f
;
431 bf
= (3 << 22) + 0.5 - (double)f
;
432 /* GCC generates an extra fstp/fld without this. */
433 __asm__ ("fstps %0" : "=m" (ai
) : "t" (af
) : "st");
434 __asm__ ("fstps %0" : "=m" (bi
) : "t" (bf
) : "st");
435 return (ai
- bi
+ 1) >> 1;
437 #define ICEIL(x) iceil(x)
438 #elif defined(USE_IEEE)
439 static INLINE
int iceil(float f
)
444 af
= (3 << 22) + 0.5 + (double)f
;
445 bf
= (3 << 22) + 0.5 - (double)f
;
446 u
.f
= (float) af
; ai
= u
.i
;
447 u
.f
= (float) bf
; bi
= u
.i
;
448 return (ai
- bi
+ 1) >> 1;
450 #define ICEIL(x) iceil(x)
452 static INLINE
int iceil(float f
)
455 return (i
< f
) ? i
+ 1 : i
;
457 #define ICEIL(x) iceil(x)
462 *** UNCLAMPED_FLOAT_TO_UBYTE: clamp float to [0,1] and map to ubyte in [0,255]
463 *** CLAMPED_FLOAT_TO_UBYTE: map float known to be in [0,1] to ubyte in [0,255]
465 #if defined(USE_IEEE) && !defined(DEBUG)
466 #define IEEE_0996 0x3f7f0000 /* 0.996 or so */
467 /* This function/macro is sensitive to precision. Test very carefully
470 #define UNCLAMPED_FLOAT_TO_UBYTE(UB, F) \
476 else if (__tmp.i >= IEEE_0996) \
477 UB = (GLubyte) 255; \
479 __tmp.f = __tmp.f * (255.0F/256.0F) + 32768.0F; \
480 UB = (GLubyte) __tmp.i; \
483 #define CLAMPED_FLOAT_TO_UBYTE(UB, F) \
486 __tmp.f = (F) * (255.0F/256.0F) + 32768.0F; \
487 UB = (GLubyte) __tmp.i; \
490 #define UNCLAMPED_FLOAT_TO_UBYTE(ub, f) \
491 ub = ((GLubyte) IROUND(CLAMP((f), 0.0F, 1.0F) * 255.0F))
492 #define CLAMPED_FLOAT_TO_UBYTE(ub, f) \
493 ub = ((GLubyte) IROUND((f) * 255.0F))
498 *** START_FAST_MATH: Set x86 FPU to faster, 32-bit precision mode (and save
499 *** original mode to a temporary).
500 *** END_FAST_MATH: Restore x86 FPU to original mode.
502 #if defined(__GNUC__) && defined(__i386__)
504 * Set the x86 FPU control word to guarentee only 32 bits of precision
505 * are stored in registers. Allowing the FPU to store more introduces
506 * differences between situations where numbers are pulled out of memory
507 * vs. situations where the compiler is able to optimize register usage.
509 * In the worst case, we force the compiler to use a memory access to
510 * truncate the float, by specifying the 'volatile' keyword.
512 /* Hardware default: All exceptions masked, extended double precision,
513 * round to nearest (IEEE compliant):
515 #define DEFAULT_X86_FPU 0x037f
516 /* All exceptions masked, single precision, round to nearest:
518 #define FAST_X86_FPU 0x003f
519 /* The fldcw instruction will cause any pending FP exceptions to be
520 * raised prior to entering the block, and we clear any pending
521 * exceptions before exiting the block. Hence, asm code has free
522 * reign over the FPU while in the fast math block.
524 #if defined(NO_FAST_MATH)
525 #define START_FAST_MATH(x) \
527 static GLuint mask = DEFAULT_X86_FPU; \
528 __asm__ ( "fnstcw %0" : "=m" (*&(x)) ); \
529 __asm__ ( "fldcw %0" : : "m" (mask) ); \
532 #define START_FAST_MATH(x) \
534 static GLuint mask = FAST_X86_FPU; \
535 __asm__ ( "fnstcw %0" : "=m" (*&(x)) ); \
536 __asm__ ( "fldcw %0" : : "m" (mask) ); \
539 /* Restore original FPU mode, and clear any exceptions that may have
540 * occurred in the FAST_MATH block.
542 #define END_FAST_MATH(x) \
544 __asm__ ( "fnclex ; fldcw %0" : : "m" (*&(x)) ); \
547 #elif defined(__WATCOMC__) && defined(__386__)
548 #define DEFAULT_X86_FPU 0x037f /* See GCC comments above */
549 #define FAST_X86_FPU 0x003f /* See GCC comments above */
550 void _watcom_start_fast_math(unsigned short *x
,unsigned short *mask
);
551 #pragma aux _watcom_start_fast_math = \
552 "fnstcw word ptr [eax]" \
553 "fldcw word ptr [ecx]" \
556 void _watcom_end_fast_math(unsigned short *x
);
557 #pragma aux _watcom_end_fast_math = \
559 "fldcw word ptr [eax]" \
562 #if defined(NO_FAST_MATH)
563 #define START_FAST_MATH(x) \
565 static GLushort mask = DEFAULT_X86_FPU; \
566 _watcom_start_fast_math(&x,&mask); \
569 #define START_FAST_MATH(x) \
571 static GLushort mask = FAST_X86_FPU; \
572 _watcom_start_fast_math(&x,&mask); \
575 #define END_FAST_MATH(x) _watcom_end_fast_math(&x)
577 #elif defined(_MSC_VER) && defined(_M_IX86)
578 #define DEFAULT_X86_FPU 0x037f /* See GCC comments above */
579 #define FAST_X86_FPU 0x003f /* See GCC comments above */
580 #if defined(NO_FAST_MATH)
581 #define START_FAST_MATH(x) do {\
582 static GLuint mask = DEFAULT_X86_FPU;\
583 __asm fnstcw word ptr [x]\
584 __asm fldcw word ptr [mask]\
587 #define START_FAST_MATH(x) do {\
588 static GLuint mask = FAST_X86_FPU;\
589 __asm fnstcw word ptr [x]\
590 __asm fldcw word ptr [mask]\
593 #define END_FAST_MATH(x) do {\
595 __asm fldcw word ptr [x]\
599 #define START_FAST_MATH(x) x = 0
600 #define END_FAST_MATH(x) (void)(x)
605 /**********************************************************************
610 _mesa_malloc( size_t bytes
);
613 _mesa_calloc( size_t bytes
);
616 _mesa_free( void *ptr
);
619 _mesa_align_malloc( size_t bytes
, unsigned long alignment
);
622 _mesa_align_calloc( size_t bytes
, unsigned long alignment
);
625 _mesa_align_free( void *ptr
);
628 _mesa_exec_malloc( GLuint size
);
631 _mesa_exec_free( void *addr
);
634 _mesa_realloc( void *oldBuffer
, size_t oldSize
, size_t newSize
);
637 _mesa_memcpy( void *dest
, const void *src
, size_t n
);
640 _mesa_memset( void *dst
, int val
, size_t n
);
643 _mesa_memset16( unsigned short *dst
, unsigned short val
, size_t n
);
646 _mesa_bzero( void *dst
, size_t n
);
649 _mesa_memcmp( const void *s1
, const void *s2
, size_t n
);
661 _mesa_asinf(float x
);
664 _mesa_atanf(float x
);
667 _mesa_sqrtd(double x
);
670 _mesa_sqrtf(float x
);
673 _mesa_inv_sqrtf(float x
);
676 _mesa_pow(double x
, double y
);
682 _mesa_bitcount(unsigned int n
);
685 _mesa_float_to_half(float f
);
688 _mesa_half_to_float(GLhalfARB h
);
692 _mesa_bsearch( const void *key
, const void *base
, size_t nmemb
, size_t size
,
693 int (*compar
)(const void *, const void *) );
696 _mesa_getenv( const char *var
);
699 _mesa_strstr( const char *haystack
, const char *needle
);
702 _mesa_strncat( char *dest
, const char *src
, size_t n
);
705 _mesa_strcpy( char *dest
, const char *src
);
708 _mesa_strncpy( char *dest
, const char *src
, size_t n
);
711 _mesa_strlen( const char *s
);
714 _mesa_strcmp( const char *s1
, const char *s2
);
717 _mesa_strncmp( const char *s1
, const char *s2
, size_t n
);
720 _mesa_strdup( const char *s
);
723 _mesa_atoi( const char *s
);
726 _mesa_strtod( const char *s
, char **end
);
729 _mesa_sprintf( char *str
, const char *fmt
, ... );
732 _mesa_printf( const char *fmtString
, ... );
735 _mesa_vsprintf( char *str
, const char *fmt
, va_list args
);
739 _mesa_warning( __GLcontext
*gc
, const char *fmtString
, ... );
742 _mesa_problem( const __GLcontext
*ctx
, const char *fmtString
, ... );
745 _mesa_error( __GLcontext
*ctx
, GLenum error
, const char *fmtString
, ... );
748 _mesa_debug( const __GLcontext
*ctx
, const char *fmtString
, ... );
751 _mesa_exit( int status
);
755 _mesa_init_default_imports( __GLimports
*imports
, void *driverCtx
);
763 #endif /* IMPORTS_H */