05:53dhewg: mannerov: alright, let's release then. can you please summarize the recent few commits for a changelog?
05:54dhewg: not every change, relevant fixes from a user perspective
07:27dhewg: mannerov: how about https://pastebin.com/BGZF25Pd ?
07:29dhewg: is that too vague? did I miss anything?
09:24dhewg: I'm going ahead and use something silimar to that
09:29dhewg: nice, github actions worked
09:37dhewg: I didn't do anything for that ^, just pushed a signed tag :)
12:10linkmauve: Hi, someone was asking in another room, did anyone ever test Nine on Lima?
12:11linkmauve: The hardware supports pretty much only GLES 2.0, and the driver attempts to provide incomplete GL 2.1 support so that more programs can run at all.
12:44dhewg: mannerov: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4647
12:46dhewg: linkmauve: there're issues about nine on panfrost and v3d on mesa's tracker, but I've not heard about attempts on lima
19:04dungeon007: orbea: That Trails in the Sky game i think suffered from slowdowns like BloodRayne, so i guess that helped
19:07dungeon007: Well, you probably didnt noticed that as on Big GPU, as many perf issues cant be noticed easely on biggies 🤣
19:20mannerov: dhewg: congrats for the release
19:22mannerov: linkmauve: haven't heard of anything either. Did you test ?
19:23linkmauve: Nope, even native games (like my own) don’t really run well yet.
19:23linkmauve: So I’m not adding an ABI change plus an emulator on top of it. :D
19:24linkmauve: Or is there any d3d9 game compiled for AArch64 already?
19:24mannerov: not that I'm aware of
19:31orbea: dungeon007: no slowdown with nine for me in that game
19:35dungeon007: Sure, one have to run that on potato to see that there was a perf issue.
19:36dungeon007: Observed only on low power Athlon 5350, as that has a GPU just 5 times faster than V3D on Raspeberry Pi 4 and CPU is similar performant 🤣
19:36orbea: dungeon007: i played a lot of falcom games with wine, nine always makes thing work better
19:46dungeon007: Or that Athlon if OCed is just like RK3399 on panfrost, just that GPU is probably still 2 times faster. 🤣 If these could make it working on nine, it should perform fine i guess.
19:50orbea: i dont know if it still happens, but wine-vanilla used to make trails in the sky throttle a single CPU at 100%, nine stopped that
19:52mannerov: dungeon007: did you bisect your perf loss
19:57dungeon007: Well, on one hand bisected it on SDMA removal.
20:00mannerov: well it only makes me curious why SDMA was giving that much better perf
20:00mannerov: CPU limitation or GPU limitation ?
20:01dungeon007: Severe... On a second hand there is some mess with these SSE flags, when Marek was atempting to add some F16C optimizations, seems he scrawed something there 🤣
20:04dungeon007: On the third hand there are some perf difference on D3D8to9, when we swtiched to MANAGED. Somewhere better, somewhere same, somewhere bugs...
20:04dungeon007: I mean several reasons of perf ups and downs, depending what i test 🤣
20:11dungeon007: Well, this HW is both CPU and GPU limited.... on SDMA removal got 20% slower on Rove bench.
20:11dungeon007: Phoronix was claiming there wont be perf regressions, but who on his right mind reads that anyway 🤣
20:20mannerov: dungeon007: for the csmt_force=0 fixing sse bugs, might I ask you to test stuffs to fix csmt_force=1
20:23mannerov: basically in nine_state.c: nine_csmt_worker
20:23mannerov: at the beginning of the function, call device9.c's nine_setup_fpu()
20:24mannerov: (you'll have to copy paste the function in nine_state.c)
20:24mannerov: I hope it will fix it
20:26dungeon007: You mean about that NFS:Carbon case... not sure what you are thinking, drops it on paste
20:28mannerov: in nine_state.c
21:14dungeon007: Nope, that does not make it working.
21:21dungeon007: It is working with native override msvcr71. Seems like WINE implementation of that only works one way, or maybe WINE have to be compiled with mfpmath forced too... sure unsure in all that 🤣
21:31mannerov: I don't understand the difference with csmt_force=0 or 1
21:31mannerov: so weird
23:11dungeon007: This is weird to me https://gitlab.freedesktop.org/mesa/mesa/-/commits/20.3/src/gallium/frontends/nine
23:12dungeon007: Adding FC16C, adding, removing, adding removing... and on b42c6ff6 i got slowdowns 🤣
23:14dungeon007: On mfpmath sse2 build, mine CPU also supports f16c... not sure what happend there 🤣
23:25dungeon007: And they break it on last stable point, meanwhile promoting sse enablement, ha, ha 🤣
23:28dungeon007: Yeah, enable it and got nothing or worse. And call it IMPROVEMENT, as Phoronix have a time to write an article for that 🤣