20:47 mannerov: Anyone remembers how ready were my patches about using memfd for memory allocation ?
20:47 mannerov: My memory seems to tell me they only needed cleanup, but I find that suspicious
20:50 mannerov: I'm quite impressed btw by all the recent news: Zink perf, ACO perf, Apple M1, etc
21:02 Venemo: I don't know mannerov I don't remember. last we spoke about it, it was just an idea
21:02 Venemo: thx for complimenting ACO :)
21:02 mannerov: I have seen the patches on VSKIP
21:03 mannerov: If you are working on that level of small details, it really shows how far you have gone
21:03 Venemo: that's Daniel's
21:03 mannerov: Was ever VSKIP produced by the closed source driver ?
21:03 Venemo: no idea
21:03 mannerov: I only found it in PS4 presentations for hand tuned shaders
21:04 mannerov: I know it's not your patch, but ACO is a team work :-)
21:05 mannerov: Sad at this point of detail it's not possible to give perf figures alongside the patches
21:05 mannerov: Sure he shows the delta in generated asm... But that doesn't mean anything performance wise
21:05 Venemo: Daniel is working on reducing branching, he has some NIR patches along these lines too
21:06 mannerov: That's great
21:07 Venemo: sure, the stats don't always mean a direct improvement in FPS, but we still want things to be optimal
21:07 mannerov: Now we only miss a feature complete Clover to show ROCm how it's done
21:07 Venemo: heheh
21:07 Venemo: there is a MR open for that
21:10 Venemo: though, that uses llvm directly
21:10 Venemo: but that's something at least
21:10 mannerov: I await the time PS5 updates perf by using the Mesa driver
21:11 mannerov: Hum though they probably ship asm directly
21:11 mannerov: btw how's ACO perf for radeonsi ?
21:11 mannerov: for example on Unigine Heaven
21:43 Venemo: ACO is not hooked up to radeonsi yet