From 59c449f6c4adf1296f0aa55196a7a9440eaec721 Mon Sep 17 00:00:00 2001
From: kramm <kramm>
Date: Sat, 21 May 2005 17:01:58 +0000
Subject: [PATCH] optimized spline approximation
---
lib/gfxtools.c | 17 +++++++++++------
1 file changed, 11 insertions(+), 6 deletions(-)
diff --git a/lib/gfxtools.c b/lib/gfxtools.c
index 938f6e4..ec98609 100644
--- a/lib/gfxtools.c
+++ b/lib/gfxtools.c
@@ -411,7 +411,7 @@ static int approximate3(const cspline_t*s, qspline_t*q, int size, double quality
test.control.y += test.end.y;
}
-#define PROBES
+//#define PROBES
#ifdef PROBES
/* measure the spline's accurancy, by taking a number of probes */
for(t=0;t<probes;t++) {
@@ -444,16 +444,21 @@ static int approximate3(const cspline_t*s, qspline_t*q, int size, double quality
/* convert control point representation to
d*x^3 + c*x^2 + b*x + a */
-
- /* FIXME: we need to do this for the subspline between [start,end],
- not [0,1] */
dx= s->end.x - s->control2.x*3 + s->control1.x*3 - s->start.x;
dy= s->end.y - s->control2.y*3 + s->control1.y*3 - s->start.y;
+ /* we need to do this for the subspline between [start,end], not [0,1]
+ as a transformation of t->a*t+b does nothing to highest coefficient
+ of the spline except multiply it with a^3, we just need to modify
+ d here. */
+ {double m = end-start;
+ dx*=m*m*m;
+ dy*=m*m*m;
+ }
+
/* use the integral over (f(x)-g(x))^2 between 0 and 1
to measure the approximation quality.
- (it boils down to const*d^2)
- */
+ (it boils down to const*d^2) */
recurse = (dx*dx + dy*dy > quality2);
#endif
--
1.7.10.4