isa/mutter
1
0
Fork 0
mirror of https://github.com/brl/mutter.git synced 2024-12-23 11:32:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

2007-12-06 Tomas Frydrych <tf@openedhand.com>

Browse Source 
* clutter/clutter-fixed.c:
	* clutter/clutter-fixed.h:
	(clutter_sqrti):
	Added extra iteration to the Newton-Rapson algorithm for argumens
	lesser than 342 to improve precission.
This commit is contained in:
Tomas Frydrych 2007-12-06 16:21:48 +00:00
parent dad823f123
commit 58142d7ee7
3 changed files with 25 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
ChangeLog
View File

@ -1,3 +1,11 @@
2007-12-06 Tomas Frydrych <tf@openedhand.com>
* clutter/clutter-fixed.c:
* clutter/clutter-fixed.h:
(clutter_sqrti):
Added extra iteration to the Newton-Rapson algorithm for arguments
lesser than 342 to improve precission.
2007-12-06 Emmanuele Bassi <ebassi@openedhand.com>
* clutter/clutter-main.c:

19
clutter/clutter-fixed.c
View File

@ -606,8 +606,8 @@ clutter_sqrtx (ClutterFixed x)
*
* Very fast fixed point implementation of square root for integers.
*
* This function is about 10x faster than clib sqrt() on x86, and (this is
* not a typo!) more than 800x faster on ARM without FPU. It's error is < 5%
* This function is at least 6x faster than clib sqrt() on x86, and (this is
* not a typo!) about 500x faster on ARM without FPU. It's error is < 5%
* for arguments < #CLUTTER_SQRTI_ARG_5_PERCENT and < 10% for arguments <
* #CLUTTER_SQRTI_ARG_10_PERCENT. The maximum argument that can be passed to
* this function is CLUTTER_SQRTI_ARG_MAX.
@ -673,13 +673,26 @@ clutter_sqrti (gint number)
flt2.f = flt.f + 2.0;
flt2.i &= 0x7FFFFF;
/* Now we correct the estimate, only single iterration is needed */
/* Now we correct the estimate */
y_1 = (flt2.i >> 11) * (flt2.i >> 11);
y_1 = (y_1 >> 8) * (x >> 8);
y_1 = f - y_1;
flt2.i = (flt2.i >> 11) * (y_1 >> 11);
/* If the original argument is less than 342, we do another
* iteration to improve precission (for arguments >= 342, the single
* iteration produces generally better results).
*/
if (x < 171)
{
y_1 = (flt2.i >> 11) * (flt2.i >> 11);
y_1 = (y_1 >> 8) * (x >> 8);
y_1 = f - y_1;
flt2.i = (flt2.i >> 11) * (y_1 >> 11);
}
/* Invert, round and convert from 10.22 to an integer
* 0x1e3c68 is a magical rounding constant that produces slightly
* better results than 0x200000.

2
clutter/clutter-fixed.h
View File

@ -338,7 +338,7 @@ ClutterFixed clutter_tani (ClutterAngle angle);
* Since: 0.6
*/
#ifndef __SSE2__
#define CLUTTER_SQRTI_ARG_5_PERCENT 131
#define CLUTTER_SQRTI_ARG_5_PERCENT 210
#else
#define CLUTTER_SQRTI_ARG_5_PERCENT INT_MAX
#endif