ref: 7a81b63abc2b3da0d7f0950f69377d2b3f54b0fb
parent: 432bb50ff539e7672723dc9161e44046710a1334
author: Alexei Podtelezhnikov <[email protected]>
date: Thu Apr 25 18:06:33 EDT 2019
Optimize Bézier bisections. * src/raster/ftraster.c (Split_Conic, Split_Cubic): Use shifts and refactor. * src/smooth/ftgrays.c (gray_split_conic, gray_split_cubic): Ditto. * src/base/ftstroke.c (ft_conic_split, ft_cubic_split): Ditto. * src/base/ftbbox.c (cubic_peak): Use shifts.
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,13 @@
+2019-04-25 Alexei Podtelezhnikov <[email protected]>
+
+ Optimize Bézier bisections.
+
+ * src/raster/ftraster.c (Split_Conic, Split_Cubic): Use shifts and
+ refactor.
+ * src/smooth/ftgrays.c (gray_split_conic, gray_split_cubic): Ditto.
+ * src/base/ftstroke.c (ft_conic_split, ft_cubic_split): Ditto.
+ * src/base/ftbbox.c (cubic_peak): Use shifts.
+
2019-04-23 Werner Lemberg <[email protected]>
* src/sfnt/ttcmap.c (tt_cmap12_next): Remove dead code.
--- a/src/base/ftbbox.c
+++ b/src/base/ftbbox.c
@@ -319,9 +319,9 @@
q2 = q2 + q1;
q4 = q4 + q3;
q3 = q3 + q2;
- q4 = ( q4 + q3 ) / 8;
- q3 = q3 / 4;
- q2 = q2 / 2;
+ q4 = ( q4 + q3 ) >> 3;
+ q3 = q3 >> 2;
+ q2 = q2 >> 1;
}
else /* second half */
{@@ -330,9 +330,9 @@
q3 = q3 + q4;
q1 = q1 + q2;
q2 = q2 + q3;
- q1 = ( q1 + q2 ) / 8;
- q2 = q2 / 4;
- q3 = q3 / 2;
+ q1 = ( q1 + q2 ) >> 3;
+ q2 = q2 >> 2;
+ q3 = q3 >> 1;
}
/* check whether either end reached the maximum */
--- a/src/base/ftstroke.c
+++ b/src/base/ftstroke.c
@@ -86,16 +86,18 @@
base[4].x = base[2].x;
- b = base[1].x;
- a = base[3].x = ( base[2].x + b ) / 2;
- b = base[1].x = ( base[0].x + b ) / 2;
- base[2].x = ( a + b ) / 2;
+ a = base[0].x + base[1].x;
+ b = base[1].x + base[2].x;
+ base[3].x = b >> 1;
+ base[2].x = ( a + b ) >> 2;
+ base[1].x = a >> 1;
base[4].y = base[2].y;
- b = base[1].y;
- a = base[3].y = ( base[2].y + b ) / 2;
- b = base[1].y = ( base[0].y + b ) / 2;
- base[2].y = ( a + b ) / 2;
+ a = base[0].y + base[1].y;
+ b = base[1].y + base[2].y;
+ base[3].y = b >> 1;
+ base[2].y = ( a + b ) >> 2;
+ base[1].y = a >> 1;
}
@@ -153,28 +155,32 @@
static void
ft_cubic_split( FT_Vector* base )
{- FT_Pos a, b, c, d;
+ FT_Pos a, b, c;
base[6].x = base[3].x;
- c = base[1].x;
- d = base[2].x;
- base[1].x = a = ( base[0].x + c ) / 2;
- base[5].x = b = ( base[3].x + d ) / 2;
- c = ( c + d ) / 2;
- base[2].x = a = ( a + c ) / 2;
- base[4].x = b = ( b + c ) / 2;
- base[3].x = ( a + b ) / 2;
+ a = base[0].x + base[1].x;
+ b = base[1].x + base[2].x;
+ c = base[2].x + base[3].x;
+ base[5].x = c >> 1;
+ c += b;
+ base[4].x = c >> 2;
+ base[1].x = a >> 1;
+ a += b;
+ base[2].x = a >> 2;
+ base[3].x = ( a + c ) >> 3;
base[6].y = base[3].y;
- c = base[1].y;
- d = base[2].y;
- base[1].y = a = ( base[0].y + c ) / 2;
- base[5].y = b = ( base[3].y + d ) / 2;
- c = ( c + d ) / 2;
- base[2].y = a = ( a + c ) / 2;
- base[4].y = b = ( b + c ) / 2;
- base[3].y = ( a + b ) / 2;
+ a = base[0].y + base[1].y;
+ b = base[1].y + base[2].y;
+ c = base[2].y + base[3].y;
+ base[5].y = c >> 1;
+ c += b;
+ base[4].y = c >> 2;
+ base[1].y = a >> 1;
+ a += b;
+ base[2].y = a >> 2;
+ base[3].y = ( a + c ) >> 3;
}
--- a/src/raster/ftraster.c
+++ b/src/raster/ftraster.c
@@ -914,16 +914,18 @@
base[4].x = base[2].x;
- b = base[1].x;
- a = base[3].x = ( base[2].x + b ) / 2;
- b = base[1].x = ( base[0].x + b ) / 2;
- base[2].x = ( a + b ) / 2;
+ a = base[0].x + base[1].x;
+ b = base[1].x + base[2].x;
+ base[3].x = b >> 1;
+ base[2].x = ( a + b ) >> 2;
+ base[1].x = a >> 1;
base[4].y = base[2].y;
- b = base[1].y;
- a = base[3].y = ( base[2].y + b ) / 2;
- b = base[1].y = ( base[0].y + b ) / 2;
- base[2].y = ( a + b ) / 2;
+ a = base[0].y + base[1].y;
+ b = base[1].y + base[2].y;
+ base[3].y = b >> 1;
+ base[2].y = ( a + b ) >> 2;
+ base[1].y = a >> 1;
/* hand optimized. gcc doesn't seem to be too good at common */
/* expression substitution and instruction scheduling ;-) */
@@ -947,28 +949,32 @@
static void
Split_Cubic( TPoint* base )
{- Long a, b, c, d;
+ Long a, b, c;
base[6].x = base[3].x;
- c = base[1].x;
- d = base[2].x;
- base[1].x = a = ( base[0].x + c + 1 ) >> 1;
- base[5].x = b = ( base[3].x + d + 1 ) >> 1;
- c = ( c + d + 1 ) >> 1;
- base[2].x = a = ( a + c + 1 ) >> 1;
- base[4].x = b = ( b + c + 1 ) >> 1;
- base[3].x = ( a + b + 1 ) >> 1;
+ a = base[0].x + base[1].x;
+ b = base[1].x + base[2].x;
+ c = base[2].x + base[3].x;
+ base[5].x = c >> 1;
+ c += b;
+ base[4].x = c >> 2;
+ base[1].x = a >> 1;
+ a += b;
+ base[2].x = a >> 2;
+ base[3].x = ( a + c ) >> 3;
base[6].y = base[3].y;
- c = base[1].y;
- d = base[2].y;
- base[1].y = a = ( base[0].y + c + 1 ) >> 1;
- base[5].y = b = ( base[3].y + d + 1 ) >> 1;
- c = ( c + d + 1 ) >> 1;
- base[2].y = a = ( a + c + 1 ) >> 1;
- base[4].y = b = ( b + c + 1 ) >> 1;
- base[3].y = ( a + b + 1 ) >> 1;
+ a = base[0].y + base[1].y;
+ b = base[1].y + base[2].y;
+ c = base[2].y + base[3].y;
+ base[5].y = c >> 1;
+ c += b;
+ base[4].y = c >> 2;
+ base[1].y = a >> 1;
+ a += b;
+ base[2].y = a >> 2;
+ base[3].y = ( a + c ) >> 3;
}
--- a/src/smooth/ftgrays.c
+++ b/src/smooth/ftgrays.c
@@ -979,16 +979,18 @@
base[4].x = base[2].x;
- b = base[1].x;
- a = base[3].x = ( base[2].x + b ) / 2;
- b = base[1].x = ( base[0].x + b ) / 2;
- base[2].x = ( a + b ) / 2;
+ a = base[0].x + base[1].x;
+ b = base[1].x + base[2].x;
+ base[3].x = b >> 1;
+ base[2].x = ( a + b ) >> 2;
+ base[1].x = a >> 1;
base[4].y = base[2].y;
- b = base[1].y;
- a = base[3].y = ( base[2].y + b ) / 2;
- b = base[1].y = ( base[0].y + b ) / 2;
- base[2].y = ( a + b ) / 2;
+ a = base[0].y + base[1].y;
+ b = base[1].y + base[2].y;
+ base[3].y = b >> 1;
+ base[2].y = ( a + b ) >> 2;
+ base[1].y = a >> 1;
}
@@ -1060,28 +1062,32 @@
static void
gray_split_cubic( FT_Vector* base )
{- TPos a, b, c, d;
+ TPos a, b, c;
base[6].x = base[3].x;
- c = base[1].x;
- d = base[2].x;
- base[1].x = a = ( base[0].x + c ) / 2;
- base[5].x = b = ( base[3].x + d ) / 2;
- c = ( c + d ) / 2;
- base[2].x = a = ( a + c ) / 2;
- base[4].x = b = ( b + c ) / 2;
- base[3].x = ( a + b ) / 2;
+ a = base[0].x + base[1].x;
+ b = base[1].x + base[2].x;
+ c = base[2].x + base[3].x;
+ base[5].x = c >> 1;
+ c += b;
+ base[4].x = c >> 2;
+ base[1].x = a >> 1;
+ a += b;
+ base[2].x = a >> 2;
+ base[3].x = ( a + c ) >> 3;
base[6].y = base[3].y;
- c = base[1].y;
- d = base[2].y;
- base[1].y = a = ( base[0].y + c ) / 2;
- base[5].y = b = ( base[3].y + d ) / 2;
- c = ( c + d ) / 2;
- base[2].y = a = ( a + c ) / 2;
- base[4].y = b = ( b + c ) / 2;
- base[3].y = ( a + b ) / 2;
+ a = base[0].y + base[1].y;
+ b = base[1].y + base[2].y;
+ c = base[2].y + base[3].y;
+ base[5].y = c >> 1;
+ c += b;
+ base[4].y = c >> 2;
+ base[1].y = a >> 1;
+ a += b;
+ base[2].y = a >> 2;
+ base[3].y = ( a + c ) >> 3;
}