brl/mutter
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fixes for 64-bit platforms; use of SSE builtin when available (patches by Gwenole Beauchesne).

Browse Source
This commit is contained in:
Tomas Frydrych 2007-08-21 09:25:09 +00:00
parent 155c22b93a
commit b1c040ce19
2 changed files with 24 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 @@
2007-08-21 Tomas Frydrych <tf@openedhand.com>
* clutter/clutter-fixed.c:
(clutter_sqrti):
Fixes for 64-bit platforms; use of SSE builtin when available
(patches by Gwenole Beauchesne).
2007-08-20 Emmanuele Bassi <ebassi@openedhand.com> 2007-08-20 Emmanuele Bassi <ebassi@openedhand.com>
* clutter/Makefile.am: * clutter/Makefile.am:

20
clutter/clutter-fixed.c
View File

@ -539,7 +539,12 @@ clutter_sqrtx (ClutterFixed x)
/* make even (2n) */ /* make even (2n) */
bit &= 0xfffffffe; bit &= 0xfffffffe;
#else #else
/* TODO -- add i386 branch using bshr */ /* TODO -- add i386 branch using bshr
*
* NB: it's been said that the bshr instruction is poorly implemented
* and that it is possible to write a faster code in C using binary
* search -- at some point we should explore this
*/
int bit = 30; int bit = 30;
while (bit >= 0) while (bit >= 0)
{ {
@ -610,6 +615,12 @@ clutter_sqrtx (ClutterFixed x)
gint gint
clutter_sqrti (gint number) clutter_sqrti (gint number)
{ {
#if defined __SSE2__
/* The GCC built-in with SSE2 (sqrtsd) is up to twice as fast as
* the pure integer code below. It is also more accurate.
*/
return __builtin_sqrt (number);
#else
/* This is a fixed point implementation of the Quake III sqrt algorithm, /* This is a fixed point implementation of the Quake III sqrt algorithm,
* described, for example, at * described, for example, at
* http://www.codemaestro.com/reviews/review00000105.html * http://www.codemaestro.com/reviews/review00000105.html
@ -626,13 +637,13 @@ clutter_sqrti (gint number)
* elsewhere in clutter is not good enough, and 10.22 is used instead. * elsewhere in clutter is not good enough, and 10.22 is used instead.
*/ */
ClutterFixed x; ClutterFixed x;
unsigned long y_1; /* 10.22 fixed point */ guint32 y_1; /* 10.22 fixed point */
unsigned long f = 0x600000; /* '1.5' as 10.22 fixed */ guint32 f = 0x600000; /* '1.5' as 10.22 fixed */
union union
{ {
float f; float f;
unsigned long i; guint32 i;
} flt, flt2; } flt, flt2;
flt.f = number; flt.f = number;
@ -669,6 +680,7 @@ clutter_sqrti (gint number)
* better results than 0x200000. * better results than 0x200000.
*/ */
return (number * flt2.i + 0x1e3c68) >> 22; return (number * flt2.i + 0x1e3c68) >> 22;
#endif
} }
/** /**