Josh Vanderhoof wrote:
>
> Keith Whitwell <[EMAIL PROTECTED]> writes:
>
> > This hasn't really been the case since the fxfastpath code was
> introduced. A
> > sample profile (from a *long* time ago, but more or less still correct)
> is
> > attached. Clipping is actually done in routines called called
> > render_vb_triangles_TMU0_TMU1 and also in fx_tri_clip_RGBA_TMU0 (the
> equivalent
> > routine is inlined in the TMU0_TMU1 case, which is why it doesn't show
> up).
> > Clipping is still a significant percentage, but it is not 'number 1'...
> >
> > % cumulative self self total
> > time seconds seconds calls us/call us/call name
> >
> > 10.03 1.75 1.75 163106 10.73 10.73
> > gl_3dnow_transform_points3_general_raw_fx2
> > 8.43 3.22 1.47 48173 30.52 30.52
> > render_vb_triangles_TMU0_TMU1
> > 7.51 4.53 1.31 48173 27.19 27.19
>
> That's a lot better than I would have expected (nice job, Keith!).
> SSE transformation functions may be a larger improvement than I
> thought. I still don't think it will be 10% though.
Squeezing 10% out of Quake might be hard but other apps could
realize a more significant improvement. Visualization programs,
for example, often have long primitives with lots of vertices
and smaller triangles.
That's one thing with OpenGL programs: different apps can stress
OpenGL in very different ways.
-Brian
----------------------------------------------------------------------
Brian Paul Avid Technology / Softimage [EMAIL PROTECTED]
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
