Awesome, thanks for the update. Sounds like things are now performing as originally intended.
Thanks so much for tracking down the inlining problem. I never would have spotted that. > On Jun 14, 2017, at 3:16 AM, Stefan Werner <[email protected]> wrote: > > > > On Fri, May 12, 2017 at 3:53 PM, Larry Gritz <[email protected] > <mailto:[email protected]>> wrote: > I'll be curious to hear what your new perf numbers are after the inline fix > (was that you? thanks!) and trying all the applicable things from my previous > email. > > Oh, things look much better now. sblur/tblur = 1/64 on diffuse rays made a > significant difference in the total # of texture reads. I haven't tried it on > glossy paths yet. > > I finally got some time to run some tests regarding performance. When > rendering frames from Agent 327*, I noticed barely any difference between > using TextureSys or loading all textures into memory - over a two hour > render, it was about a minute or two difference . This with all textures on a > local SSD - network access could slow things down. > > * https://www.youtube.com/watch?v=mN0zPOpADL4 > <https://www.youtube.com/watch?v=mN0zPOpADL4> > > Regarding the Windows inlining fix, I ran some benchmarks there too. Again on > one of the Agent 327 frames, it's 3h43m before the fix and 3h32m after the > fix. On a different scene that's lighter on geometry and render time is more > dominated by texture fetches, I'm seeing 1h50m before and 1h38 after - so > before, a good 10% of the render time were due to non-inlined methods. > > -Stefan > -- Larry Gritz [email protected]
_______________________________________________ Oiio-dev mailing list [email protected] http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
