[base] Small optimization of `FT_MulFix'.

* src/base/ftcalc.c (FT_MulFix): Loosen up the condition for direct
32-bit calculations.
This commit is contained in:
Alexei Podtelezhnikov 2014-08-20 00:57:22 -04:00
parent f09326a1a6
commit 8fd87d4e69
2 changed files with 18 additions and 2 deletions

View File

@ -1,3 +1,10 @@
2014-08-20 Alexei Podtelezhnikov <apodtele@gmail.com>
[base] Small optimization of `FT_MulFix'.
* src/base/ftcalc.c (FT_MulFix): Loosen up the condition for direct
32-bit calculations.
2014-08-19 Alexei Podtelezhnikov <apodtele@gmail.com>
[base] Use unsigned calculation in `FT_MulDiv'.

View File

@ -385,6 +385,15 @@
/* */
/* a + b <= 129895 - (c >> 17) */
/* */
/* FT_MulFix, on the other hand, is optimized for a small value of */
/* the first argument, when the second argument can be much larger. */
/* This can be achieved by scaling the second argument and the limit */
/* in the above inequalities. For example, */
/* */
/* a + (b >> 8) <= (131071 >> 4) */
/* */
/* should work well to avoid the overflow. */
/* */
/* documentation is in freetype.h */
@ -513,7 +522,7 @@
ua = (FT_ULong)a;
ub = (FT_ULong)b;
if ( ua <= 2048 && ub <= 1048576L )
if ( ua + ( ub >> 8 ) <= 8191UL )
ua = ( ua * ub + 0x8000U ) >> 16;
else
{
@ -544,7 +553,7 @@
ua = (FT_ULong)a;
ub = (FT_ULong)b;
if ( ua <= 2048 && ub <= 1048576L )
if ( ua + ( ub >> 8 ) <= 8191UL )
ua = ( ua * ub + 0x8000UL ) >> 16;
else
{