1 /**************************************************************************
3 * Copyright 2008 VMware, Inc.
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 VMWARE 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"
55 #include <strings.h> /* for ffs */
60 #define M_SQRT2 1.41421356237309504880
66 #if _MSC_VER < 1400 && !defined(__cplusplus)
68 static INLINE
float cosf( float f
)
70 return (float) cos( (double) f
);
73 static INLINE
float sinf( float f
)
75 return (float) sin( (double) f
);
78 static INLINE
float ceilf( float f
)
80 return (float) ceil( (double) f
);
83 static INLINE
float floorf( float f
)
85 return (float) floor( (double) f
);
88 static INLINE
float powf( float f
, float g
)
90 return (float) pow( (double) f
, (double) g
);
93 static INLINE
float sqrtf( float f
)
95 return (float) sqrt( (double) f
);
98 static INLINE
float fabsf( float f
)
100 return (float) fabs( (double) f
);
103 static INLINE
float logf( float f
)
105 return (float) log( (double) f
);
109 /* Work-around an extra semi-colon in VS 2005 logf definition */
112 #define logf(x) ((float)log((double)(x)))
115 #define isfinite(x) _finite((double)(x))
116 #define isnan(x) _isnan((double)(x))
117 #endif /* _MSC_VER < 1400 && !defined(__cplusplus) */
119 static INLINE
double log2( double x
)
121 const double invln2
= 1.442695041;
122 return log( x
) * invln2
;
128 return x
>= 0.0 ? floor(x
+ 0.5) : ceil(x
- 0.5);
134 return x
>= 0.0f
? floorf(x
+ 0.5f
) : ceilf(x
- 0.5f
);
137 #define INFINITY (DBL_MAX + DBL_MAX)
138 #define NAN (INFINITY - INFINITY)
140 #endif /* _MSC_VER */
143 #ifdef PIPE_OS_ANDROID
146 double log2(double d
)
148 return log(d
) * (1.0 / M_LN2
);
151 /* workaround a conflict with main/imports.h */
159 return logf(f
) * (float) (1.0 / M_LN2
);
165 #if __STDC_VERSION__ < 199901L && (!defined(__cplusplus) || defined(_MSC_VER))
166 static INLINE
long int
169 long int rounded
= (long int)(d
+ 0.5);
171 if (d
- floor(d
) == 0.5) {
172 if (rounded
% 2 != 0)
173 rounded
+= (d
> 0) ? -1 : 1;
179 static INLINE
long int
182 long int rounded
= (long int)(f
+ 0.5f
);
184 if (f
- floorf(f
) == 0.5f
) {
185 if (rounded
% 2 != 0)
186 rounded
+= (f
> 0) ? -1 : 1;
192 static INLINE
long long int
195 long long int rounded
= (long long int)(d
+ 0.5);
197 if (d
- floor(d
) == 0.5) {
198 if (rounded
% 2 != 0)
199 rounded
+= (d
> 0) ? -1 : 1;
205 static INLINE
long long int
208 long long int rounded
= (long long int)(f
+ 0.5f
);
210 if (f
- floorf(f
) == 0.5f
) {
211 if (rounded
% 2 != 0)
212 rounded
+= (f
> 0) ? -1 : 1;
219 #define POW2_TABLE_SIZE_LOG2 9
220 #define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)
221 #define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)
222 #define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))
223 extern float pow2_table
[POW2_TABLE_SIZE
];
227 * Initialize math module. This should be called before using any
228 * other functions in this module.
231 util_init_math(void);
249 * Extract the IEEE float32 exponent.
252 util_get_float32_exponent(float x
) {
257 return ((f
.ui
>> 23) & 0xff) - 127;
262 * Fast version of 2^x
263 * Identity: exp2(a + b) = exp2(a) * exp2(b)
265 * Let fpart = x - ipart;
266 * So, exp2(x) = exp2(ipart) * exp2(fpart)
267 * Compute exp2(ipart) with i << ipart
268 * Compute exp2(fpart) with lookup table.
271 util_fast_exp2(float x
)
278 return 3.402823466e+38f
;
284 fpart
= x
- (float) ipart
;
287 * epart.f = (float) (1 << ipart)
288 * but faster and without integer overflow for ipart > 31
290 epart
.i
= (ipart
+ 127 ) << 23;
292 mpart
= pow2_table
[POW2_TABLE_OFFSET
+ (int)(fpart
* POW2_TABLE_SCALE
)];
294 return epart
.f
* mpart
;
299 * Fast approximation to exp(x).
302 util_fast_exp(float x
)
304 const float k
= 1.44269f
; /* = log2(e) */
305 return util_fast_exp2(k
* x
);
309 #define LOG2_TABLE_SIZE_LOG2 16
310 #define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)
311 #define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)
312 extern float log2_table
[LOG2_TABLE_SIZE
];
316 * Fast approximation to log2(x).
319 util_fast_log2(float x
)
324 epart
= (float)(((num
.i
& 0x7f800000) >> 23) - 127);
325 /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */
326 mpart
= log2_table
[((num
.i
& 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2
))) >> (23 - LOG2_TABLE_SIZE_LOG2
)];
327 return epart
+ mpart
;
332 * Fast approximation to x^y.
335 util_fast_pow(float x
, float y
)
337 return util_fast_exp2(util_fast_log2(x
) * y
);
340 /* Note that this counts zero as a power of two.
342 static INLINE boolean
343 util_is_power_of_two( unsigned v
)
345 return (v
& (v
-1)) == 0;
350 * Floor(x), returned as int.
358 af
= (3 << 22) + 0.5 + (double) f
;
359 bf
= (3 << 22) + 0.5 - (double) f
;
360 u
.f
= (float) af
; ai
= u
.i
;
361 u
.f
= (float) bf
; bi
= u
.i
;
362 return (ai
- bi
) >> 1;
367 * Round float to nearest int.
372 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
374 __asm__ ("fistpl %0" : "=m" (r
) : "t" (f
) : "st");
376 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
385 return (int) (f
+ 0.5f
);
387 return (int) (f
- 0.5f
);
393 * Approximate floating point comparison
395 static INLINE boolean
396 util_is_approx(float a
, float b
, float tol
)
398 return fabs(b
- a
) <= tol
;
403 * util_is_X_inf_or_nan = test if x is NaN or +/- Inf
404 * util_is_X_nan = test if x is NaN
405 * util_X_inf_sign = return +1 for +Inf, -1 for -Inf, or 0 for not Inf
407 * NaN can be checked with x != x, however this fails with the fast math flag
414 static INLINE boolean
415 util_is_inf_or_nan(float x
)
419 return (tmp
.ui
& 0x7f800000) == 0x7f800000;
423 static INLINE boolean
428 return (tmp
.ui
& 0x7fffffff) > 0x7f800000;
433 util_inf_sign(float x
)
437 if ((tmp
.ui
& 0x7fffffff) != 0x7f800000) {
441 return (x
< 0) ? -1 : 1;
448 static INLINE boolean
449 util_is_double_inf_or_nan(double x
)
453 return (tmp
.ui
& 0x7ff0000000000000ULL
) == 0x7ff0000000000000ULL
;
457 static INLINE boolean
458 util_is_double_nan(double x
)
462 return (tmp
.ui
& 0x7fffffffffffffffULL
) > 0x7ff0000000000000ULL
;
467 util_double_inf_sign(double x
)
471 if ((tmp
.ui
& 0x7fffffffffffffffULL
) != 0x7ff0000000000000ULL
) {
475 return (x
< 0) ? -1 : 1;
482 static INLINE boolean
483 util_is_half_inf_or_nan(int16_t x
)
485 return (x
& 0x7c00) == 0x7c00;
489 static INLINE boolean
490 util_is_half_nan(int16_t x
)
492 return (x
& 0x7fff) > 0x7c00;
497 util_half_inf_sign(int16_t x
)
499 if ((x
& 0x7fff) != 0x7c00) {
503 return (x
< 0) ? -1 : 1;
508 * Find first bit set in word. Least significant bit is 1.
509 * Return 0 if no bits set.
512 #define FFS_DEFINED 1
514 #if defined(_MSC_VER) && _MSC_VER >= 1300 && (_M_IX86 || _M_AMD64 || _M_IA64)
515 unsigned char _BitScanForward(unsigned long* Index
, unsigned long Mask
);
516 #pragma intrinsic(_BitScanForward)
518 unsigned long ffs( unsigned long u
)
521 if (_BitScanForward(&i
, u
))
526 #elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
528 unsigned ffs( unsigned u
)
542 #elif defined(__MINGW32__) || defined(PIPE_OS_ANDROID)
543 #define ffs __builtin_ffs
546 #endif /* FFS_DEFINED */
549 * Find last bit set in a word. The least significant bit is 1.
550 * Return 0 if no bits are set.
552 static INLINE
unsigned util_last_bit(unsigned u
)
554 #if defined(__GNUC__) && ((__GNUC__ * 100 + __GNUC_MINOR__) >= 304)
555 return u
== 0 ? 0 : 32 - __builtin_clz(u
);
567 /* Destructively loop over all of the bits in a mask as in:
570 * int i = u_bit_scan(&mymask);
571 * ... process element i
575 static INLINE
int u_bit_scan(unsigned *mask
)
577 int i
= ffs(*mask
) - 1;
586 static INLINE
unsigned
596 * Convert ubyte to float in [0, 1].
597 * XXX a 256-entry lookup table would be slightly faster.
600 ubyte_to_float(ubyte ub
)
602 return (float) ub
* (1.0f
/ 255.0f
);
607 * Convert float in [0,1] to ubyte in [0,255] with clamping.
610 float_to_ubyte(float f
)
618 else if (tmp
.i
>= 0x3f800000 /* 1.0f */) {
622 tmp
.f
= tmp
.f
* (255.0f
/256.0f
) + 32768.0f
;
623 return (ubyte
) tmp
.i
;
628 byte_to_float_tex(int8_t b
)
630 return (b
== -128) ? -1.0F
: b
* 1.0F
/ 127.0F
;
634 float_to_byte_tex(float f
)
636 return (int8_t) (127.0F
* f
);
642 static INLINE
unsigned
643 util_logbase2(unsigned n
)
645 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 304)
646 return ((sizeof(unsigned) * 8 - 1) - __builtin_clz(n
| 1));
649 if (n
>= 1<<16) { n
>>= 16; pos
+= 16; }
650 if (n
>= 1<< 8) { n
>>= 8; pos
+= 8; }
651 if (n
>= 1<< 4) { n
>>= 4; pos
+= 4; }
652 if (n
>= 1<< 2) { n
>>= 2; pos
+= 2; }
653 if (n
>= 1<< 1) { pos
+= 1; }
660 * Returns the smallest power of two >= x
662 static INLINE
unsigned
663 util_next_power_of_two(unsigned x
)
665 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 304)
669 return (1 << ((sizeof(unsigned) * 8) - __builtin_clz(x
- 1)));
676 if (util_is_power_of_two(x
))
680 val
= (val
>> 1) | val
;
681 val
= (val
>> 2) | val
;
682 val
= (val
>> 4) | val
;
683 val
= (val
>> 8) | val
;
684 val
= (val
>> 16) | val
;
692 * Return number of bits set in n.
694 static INLINE
unsigned
695 util_bitcount(unsigned n
)
697 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 304)
698 return __builtin_popcount(n
);
700 /* K&R classic bitcount.
702 * For each iteration, clear the LSB from the bitfield.
703 * Requires only one iteration per set bit, instead of
704 * one iteration per bit less than highest set bit.
707 for (bits
; n
; bits
++) {
716 * Convert from little endian to CPU byte order.
719 #ifdef PIPE_ARCH_BIG_ENDIAN
720 #define util_le32_to_cpu(x) util_bswap32(x)
721 #define util_le16_to_cpu(x) util_bswap16(x)
723 #define util_le32_to_cpu(x) (x)
724 #define util_le16_to_cpu(x) (x)
729 * Reverse byte order of a 32 bit word.
731 static INLINE
uint32_t
732 util_bswap32(uint32_t n
)
734 #if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 403)
735 return __builtin_bswap32(n
);
738 ((n
>> 8) & 0x0000ff00) |
739 ((n
<< 8) & 0x00ff0000) |
746 * Reverse byte order of a 16 bit word.
748 static INLINE
uint16_t
749 util_bswap16(uint16_t n
)
757 * Clamp X to [MIN, MAX].
758 * This is a macro to allow float, int, uint, etc. types.
760 #define CLAMP( X, MIN, MAX ) ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
762 #define MIN2( A, B ) ( (A)<(B) ? (A) : (B) )
763 #define MAX2( A, B ) ( (A)>(B) ? (A) : (B) )
765 #define MIN3( A, B, C ) ((A) < (B) ? MIN2(A, C) : MIN2(B, C))
766 #define MAX3( A, B, C ) ((A) > (B) ? MAX2(A, C) : MAX2(B, C))
768 #define MIN4( A, B, C, D ) ((A) < (B) ? MIN3(A, C, D) : MIN3(B, C, D))
769 #define MAX4( A, B, C, D ) ((A) > (B) ? MAX3(A, C, D) : MAX3(B, C, D))
773 * Align a value, only works pot alignemnts.
776 align(int value
, int alignment
)
778 return (value
+ alignment
- 1) & ~(alignment
- 1);
782 * Works like align but on npot alignments.
785 util_align_npot(size_t value
, size_t alignment
)
787 if (value
% alignment
)
788 return value
+ (alignment
- (value
% alignment
));
792 static INLINE
unsigned
793 u_minify(unsigned value
, unsigned levels
)
795 return MAX2(1, value
>> levels
);
799 #define COPY_4V( DST, SRC ) \
801 (DST)[0] = (SRC)[0]; \
802 (DST)[1] = (SRC)[1]; \
803 (DST)[2] = (SRC)[2]; \
804 (DST)[3] = (SRC)[3]; \
810 #define COPY_4FV( DST, SRC ) COPY_4V(DST, SRC)
815 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
825 static INLINE
uint32_t util_unsigned_fixed(float value
, unsigned frac_bits
)
827 return value
< 0 ? 0 : (uint32_t)(value
* (1<<frac_bits
));
830 static INLINE
int32_t util_signed_fixed(float value
, unsigned frac_bits
)
832 return (int32_t)(value
* (1<<frac_bits
));
836 util_fpstate_get(void);
838 util_fpstate_set_denorms_to_zero(unsigned current_fpstate
);
840 util_fpstate_set(unsigned fpstate
);
848 #endif /* U_MATH_H */