Josh Vanderhoof wrote:
>
> > In particular the vertex-snapping code? It would be good to restore
> > this in the current version - two unnecessary divides per vertex is
> > pretty wasteful.
>
> It's a float->int conversion, not a divide. Still pretty wasteful
> though.
True, the divide is just 'conceptual'... OK, in the interim I've
committed the obvious bit of C code:
# if defined(__linux__) && defined(__i386__)
# define DO_SETUP_XY { \
GLfloat t1 = win[i][0] + snapper; \
GLfloat t2 = win[i][1] + snapper; \
(v)->x = t1 - snapper; \
(v)->y = t2 - snapper; \
}
# else
...
# endif
I don't know how the old asm got away with using just one fadd, as it
appeared to. When enabled this code gives me about another 10 percent
in the best mode of isosurf.
Keith
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
