<hexdump0815>
i can add debug options or code to both the kernel or mesa if you can tell me what to add or do any additional tests which might help to get this problem debugged - any idea to bring this forward is welcome
<hexdump0815>
please also respond if i'm offline here on irc - i'll read it in the logs
monstr has joined #lima
<hexdump0815>
the problem mentioned in the ticket does not happen in panfrost, it does not happen with the legacy mali blob and also the rpi mesa is handling this case well
Viciouss has quit [Quit: Ping timeout (120 seconds)]
Viciouss has joined #lima
camus has joined #lima
kaspter has quit [Ping timeout: 265 seconds]
camus is now known as kaspter
<linkmauve>
anarsoul|2, 7.89s on 20.3.4 built without LTO and with debug symbols, exactly like I built master.
<linkmauve>
So it’s a ~9s regression, more than doubling.
monstr has quit [Remote host closed the connection]
<anarsoul|2>
hexdump0815: Erico was able to reproduce your problem with the trace provided, so you just need to wait till someone gets some time to work on it
<anarsoul|2>
linkmauve: yeah, if you can bisect that would definitely help
<linkmauve>
In the middle of it, for now I got:
<linkmauve>
00944f80e26 7.56s
<linkmauve>
efc75e13ea1 14.02s
<linkmauve>
ac2b8a56cdb 7.62s
<linkmauve>
I really need a more powerful ARM machine than this Nintendo Switch. :x
<linkmauve>
At least now I gave it a SSD over USB 3.0.
<linkmauve>
14s still isn’t 17s, but it’s already far from ~7.5s.
warpme_ has quit [Quit: Connection closed for inactivity]
<linkmauve>
anarsoul|2, 1d5ce66b02016a8c985e3e3cc44df65ccfacca48 is the first bad commit
<linkmauve>
“mesa/st: Always precompile the first shader variant.”
<anarsoul|2>
I guess the question is why the first variant doesn't match following variant(s) and it's not picked up from cache
<linkmauve>
So atm given it takes about twice as much time, we probably never get the first variant?
<linkmauve>
I mean, GTK never uses the first variant it compiles?