buzzmarshall has quit [Remote host closed the connection]
Barada has joined #lima
cp- has quit [Quit: Disappeared in a puff of smoke]
cp- has joined #lima
_whitelogger has joined #lima
yann has quit [Ping timeout: 258 seconds]
yann has joined #lima
dddddd has quit [Ping timeout: 260 seconds]
rak-zero has joined #lima
gcl_ has joined #lima
gcl has quit [Ping timeout: 258 seconds]
UnivrslSuprBox has quit [Read error: Connection reset by peer]
warpme_ has quit [Read error: Connection reset by peer]
warpme_ has joined #lima
UnivrslSuprBox has joined #lima
ecloud is now known as ecloud_wfh
<rellla>
any reservations about marrying mali_syscall_tracker with shader disassemblers?
<enunes>
I would like to see disassembled shaders in the dumps
<enunes>
especially if it allows me to grep e.g. deqp dumps and see if the driver-compiled shader ever takes some specific decision
<enunes>
(e.g. in the past I tried to generate a shader that would use swizzle with the store instruction, I was unable to so now we assume it is not supported)
<enunes>
it probably isn't anyway in that case, but some times it would be nice to readily grep a large number of results to see if something ever happens
<rellla>
enunes: i will try to implement it
monstr has joined #lima
<rellla>
btw, does your ppir make process :)
<enunes>
what? :)
<rellla>
s/ppir/ppir work/
<enunes>
I'm still working on it, mostly making general improvements in the scheduler so we can combine more instructions
<enunes>
hopefully this week I have a working version so I can start making real comparisons with the current one (for example evaluate if that doesnt affect register pressure too much and has negative results in complicated cases...)
<enunes>
then based on that, I may have to tweak it more
<rellla>
sounds good :)
Barada has quit [Quit: Barada]
yann has quit [Ping timeout: 260 seconds]
monstr has quit [Remote host closed the connection]
<anarsoul>
enunes: btw we don't have complete disassembler since not all instructions were REd
<anarsoul>
at least for ppir
<anarsoul>
e.g. temp load/store are used for matrices in blob
<enunes>
well that sounds like future work and having the disassembler in syscall dumper will be helpful for that
<anarsoul>
yeah
<enunes>
rellla: I wonder if there is a way to do that without copying the disassembler... possibly linking lima_dri.so..? probably pretty bad... maybe calling lima_disasm externally if it exists in $PATH...
<anarsoul>
IIRC Erik said that there was no single unused bit in ISA :)
<anarsoul>
so yeah, there's a lot to RE
gcl has quit [Ping timeout: 260 seconds]
gcl has joined #lima
<rellla>
enunes: hm, i think this was the fastest way to implement it, because it's basically just copy & paste. i'm afraid of having mesa on my blob machine...
<rellla>
copying code means twice the effort for maintaining it of course, but imho that shouldn't be the problem to focus on atm
gcl has quit [Ping timeout: 256 seconds]
gcl has joined #lima
adjtm_ has joined #lima
adjtm has quit [Ping timeout: 256 seconds]
<enunes>
rellla: yeah I agree, I was just trying to think if some better way but I think the other options are not very worth the hassle
_whitelogger has joined #lima
anarsoul|c has joined #lima
anarsoul|c has quit [Ping timeout: 256 seconds]
_whitelogger has quit [Ping timeout: 256 seconds]
_whitelogger has joined #lima
gcl_ has joined #lima
gcl has quit [Ping timeout: 258 seconds]
gcl_ has quit [Ping timeout: 256 seconds]
gcl has joined #lima
buzzmarshall has joined #lima
megi has quit [Ping timeout: 260 seconds]
<anarsoul>
darn, out standalone compiler is not very useful :( we get different nir input comparing to when running on target