551463f75cd1d7db3df73a2accb2b2515e7fb6ee
1 /**************************************************************************
3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Math utilities and approximations for common math functions.
31 * Reduced precision is usually acceptable in shaders...
33 * "fast" is used in the names of functions which are low-precision,
34 * or at least lower-precision than the normal C lib functions.
42 #include "pipe/p_compiler.h"
43 #include "util/u_debug.h"
51 #if defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
52 __inline
double ceil(double val
)
56 if ((val
- (long) val
) == 0) {
61 ceil_val
= (long) val
+ 1;
64 ceil_val
= (long) val
;
71 #ifndef PIPE_SUBSYSTEM_WINDOWS_CE_OGL
72 __inline
double floor(double val
)
76 if ((val
- (long) val
) == 0) {
81 floor_val
= (long) val
;
84 floor_val
= (long) val
- 1;
93 __inline
double __cdecl
pow(double val
, double exponent
)
100 #pragma function(log)
101 __inline
double __cdecl
log(double val
)
108 #pragma function(atan2)
109 __inline
double __cdecl
atan2(double val
)
122 #define M_SQRT2 1.41421356237309504880
126 #if defined(_MSC_VER)
128 #if _MSC_VER < 1400 && !defined(__cplusplus) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
130 static INLINE
float cosf( float f
)
132 return (float) cos( (double) f
);
135 static INLINE
float sinf( float f
)
137 return (float) sin( (double) f
);
140 static INLINE
float ceilf( float f
)
142 return (float) ceil( (double) f
);
145 static INLINE
float floorf( float f
)
147 return (float) floor( (double) f
);
150 static INLINE
float powf( float f
, float g
)
152 return (float) pow( (double) f
, (double) g
);
155 static INLINE
float sqrtf( float f
)
157 return (float) sqrt( (double) f
);
160 static INLINE
float fabsf( float f
)
162 return (float) fabs( (double) f
);
165 static INLINE
float logf( float f
)
167 return (float) log( (double) f
);
171 /* Work-around an extra semi-colon in VS 2005 logf definition */
174 #define logf(x) ((float)log((double)(x)))
177 #define isfinite(x) _finite((double)(x))
178 #define isnan(x) _isnan((double)(x))
179 #endif /* _MSC_VER < 1400 && !defined(__cplusplus) */
181 static INLINE
double log2( double x
)
183 const double invln2
= 1.442695041;
184 return log( x
) * invln2
;
190 return x
>= 0.0 ? floor(x
+ 0.5) : ceil(x
- 0.5);
196 return x
>= 0.0f
? floorf(x
+ 0.5f
) : ceilf(x
- 0.5f
);
199 #endif /* _MSC_VER */
202 #ifdef PIPE_OS_ANDROID
205 double log2(double d
)
207 return log(d
) * (1.0 / M_LN2
);
210 /* workaround a conflict with main/imports.h */
218 return logf(f
) * (float) (1.0 / M_LN2
);
226 #define POW2_TABLE_SIZE_LOG2 9
227 #define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)
228 #define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)
229 #define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))
230 extern float pow2_table
[POW2_TABLE_SIZE
];
234 * Initialize math module. This should be called before using any
235 * other functions in this module.
238 util_init_math(void);
249 * Fast version of 2^x
250 * Identity: exp2(a + b) = exp2(a) * exp2(b)
252 * Let fpart = x - ipart;
253 * So, exp2(x) = exp2(ipart) * exp2(fpart)
254 * Compute exp2(ipart) with i << ipart
255 * Compute exp2(fpart) with lookup table.
258 util_fast_exp2(float x
)
265 return 3.402823466e+38f
;
271 fpart
= x
- (float) ipart
;
274 * epart.f = (float) (1 << ipart)
275 * but faster and without integer overflow for ipart > 31
277 epart
.i
= (ipart
+ 127 ) << 23;
279 mpart
= pow2_table
[POW2_TABLE_OFFSET
+ (int)(fpart
* POW2_TABLE_SCALE
)];
281 return epart
.f
* mpart
;
286 * Fast approximation to exp(x).
289 util_fast_exp(float x
)
291 const float k
= 1.44269f
; /* = log2(e) */
292 return util_fast_exp2(k
* x
);
296 #define LOG2_TABLE_SIZE_LOG2 16
297 #define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)
298 #define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)
299 extern float log2_table
[LOG2_TABLE_SIZE
];
303 * Fast approximation to log2(x).
306 util_fast_log2(float x
)
311 epart
= (float)(((num
.i
& 0x7f800000) >> 23) - 127);
312 /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */
313 mpart
= log2_table
[((num
.i
& 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2
))) >> (23 - LOG2_TABLE_SIZE_LOG2
)];
314 return epart
+ mpart
;
319 * Fast approximation to x^y.
322 util_fast_pow(float x
, float y
)
324 return util_fast_exp2(util_fast_log2(x
) * y
);
327 /* Note that this counts zero as a power of two.
329 static INLINE boolean
330 util_is_power_of_two( unsigned v
)
332 return (v
& (v
-1)) == 0;
337 * Floor(x), returned as int.
345 af
= (3 << 22) + 0.5 + (double) f
;
346 bf
= (3 << 22) + 0.5 - (double) f
;
347 u
.f
= (float) af
; ai
= u
.i
;
348 u
.f
= (float) bf
; bi
= u
.i
;
349 return (ai
- bi
) >> 1;
354 * Round float to nearest int.
359 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
361 __asm__ ("fistpl %0" : "=m" (r
) : "t" (f
) : "st");
363 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
372 return (int) (f
+ 0.5f
);
374 return (int) (f
- 0.5f
);
380 * Approximate floating point comparison
382 static INLINE boolean
383 util_is_approx(float a
, float b
, float tol
)
385 return fabs(b
- a
) <= tol
;
390 * Test if x is NaN or +/- infinity.
392 static INLINE boolean
393 util_is_inf_or_nan(float x
)
397 return !(int)((unsigned int)((tmp
.i
& 0x7fffffff)-0x7f800000) >> 31);
402 * Find first bit set in word. Least significant bit is 1.
403 * Return 0 if no bits set.
405 #if defined(_MSC_VER) && _MSC_VER >= 1300 && (_M_IX86 || _M_AMD64 || _M_IA64)
406 unsigned char _BitScanForward(unsigned long* Index
, unsigned long Mask
);
407 #pragma intrinsic(_BitScanForward)
409 unsigned long ffs( unsigned long u
)
412 if (_BitScanForward(&i
, u
))
417 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
419 unsigned ffs( unsigned u
)
433 #elif defined(__MINGW32__) || defined(PIPE_OS_ANDROID)
434 #define ffs __builtin_ffs
438 /* Destructively loop over all of the bits in a mask as in:
441 * int i = u_bit_scan(&mymask);
442 * ... process element i
446 static INLINE
int u_bit_scan(unsigned *mask
)
448 int i
= ffs(*mask
) - 1;
457 static INLINE
unsigned
467 * Convert ubyte to float in [0, 1].
468 * XXX a 256-entry lookup table would be slightly faster.
471 ubyte_to_float(ubyte ub
)
473 return (float) ub
* (1.0f
/ 255.0f
);
478 * Convert float in [0,1] to ubyte in [0,255] with clamping.
481 float_to_ubyte(float f
)
483 const int ieee_0996
= 0x3f7f0000; /* 0.996 or so */
490 else if (tmp
.i
>= ieee_0996
) {
494 tmp
.f
= tmp
.f
* (255.0f
/256.0f
) + 32768.0f
;
495 return (ubyte
) tmp
.i
;
500 byte_to_float_tex(int8_t b
)
502 return (b
== -128) ? -1.0F
: b
* 1.0F
/ 127.0F
;
506 float_to_byte_tex(float f
)
508 return (int8_t) (127.0F
* f
);
514 static INLINE
unsigned
515 util_logbase2(unsigned n
)
517 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 304)
518 return ((sizeof(unsigned) * 8 - 1) - __builtin_clz(n
| 1));
521 if (n
>= 1<<16) { n
>>= 16; pos
+= 16; }
522 if (n
>= 1<< 8) { n
>>= 8; pos
+= 8; }
523 if (n
>= 1<< 4) { n
>>= 4; pos
+= 4; }
524 if (n
>= 1<< 2) { n
>>= 2; pos
+= 2; }
525 if (n
>= 1<< 1) { pos
+= 1; }
532 * Returns the smallest power of two >= x
534 static INLINE
unsigned
535 util_next_power_of_two(unsigned x
)
537 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 304)
541 return (1 << ((sizeof(unsigned) * 8) - __builtin_clz(x
- 1)));
548 if (util_is_power_of_two(x
))
552 val
= (val
>> 1) | val
;
553 val
= (val
>> 2) | val
;
554 val
= (val
>> 4) | val
;
555 val
= (val
>> 8) | val
;
556 val
= (val
>> 16) | val
;
564 * Return number of bits set in n.
566 static INLINE
unsigned
567 util_bitcount(unsigned n
)
569 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 304)
570 return __builtin_popcount(n
);
572 /* K&R classic bitcount.
574 * For each iteration, clear the LSB from the bitfield.
575 * Requires only one iteration per set bit, instead of
576 * one iteration per bit less than highest set bit.
579 for (bits
; n
; bits
++) {
588 * Convert from little endian to CPU byte order.
591 #ifdef PIPE_ARCH_BIG_ENDIAN
592 #define util_le32_to_cpu(x) util_bswap32(x)
593 #define util_le16_to_cpu(x) util_bswap16(x)
595 #define util_le32_to_cpu(x) (x)
596 #define util_le16_to_cpu(x) (x)
601 * Reverse byte order of a 32 bit word.
603 static INLINE
uint32_t
604 util_bswap32(uint32_t n
)
606 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 403)
607 return __builtin_bswap32(n
);
610 ((n
>> 8) & 0x0000ff00) |
611 ((n
<< 8) & 0x00ff0000) |
618 * Reverse byte order of a 16 bit word.
620 static INLINE
uint16_t
621 util_bswap16(uint16_t n
)
629 * Clamp X to [MIN, MAX].
630 * This is a macro to allow float, int, uint, etc. types.
632 #define CLAMP( X, MIN, MAX ) ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
634 #define MIN2( A, B ) ( (A)<(B) ? (A) : (B) )
635 #define MAX2( A, B ) ( (A)>(B) ? (A) : (B) )
637 #define MIN3( A, B, C ) ((A) < (B) ? MIN2(A, C) : MIN2(B, C))
638 #define MAX3( A, B, C ) ((A) > (B) ? MAX2(A, C) : MAX2(B, C))
640 #define MIN4( A, B, C, D ) ((A) < (B) ? MIN3(A, C, D) : MIN3(B, C, D))
641 #define MAX4( A, B, C, D ) ((A) > (B) ? MAX3(A, C, D) : MAX3(B, C, D))
645 * Align a value, only works pot alignemnts.
648 align(int value
, int alignment
)
650 return (value
+ alignment
- 1) & ~(alignment
- 1);
654 * Works like align but on npot alignments.
657 util_align_npot(size_t value
, size_t alignment
)
659 if (value
% alignment
)
660 return value
+ (alignment
- (value
% alignment
));
664 static INLINE
unsigned
665 u_minify(unsigned value
, unsigned levels
)
667 return MAX2(1, value
>> levels
);
671 #define COPY_4V( DST, SRC ) \
673 (DST)[0] = (SRC)[0]; \
674 (DST)[1] = (SRC)[1]; \
675 (DST)[2] = (SRC)[2]; \
676 (DST)[3] = (SRC)[3]; \
682 #define COPY_4FV( DST, SRC ) COPY_4V(DST, SRC)
687 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
697 static INLINE
uint32_t util_unsigned_fixed(float value
, unsigned frac_bits
)
699 return value
< 0 ? 0 : (uint32_t)(value
* (1<<frac_bits
));
702 static INLINE
int32_t util_signed_fixed(float value
, unsigned frac_bits
)
704 return (int32_t)(value
* (1<<frac_bits
));
713 #endif /* U_MATH_H */