austriancoder changed the topic of #etnaviv to: #etnaviv - the home of the reverse-engineered Vivante GPU driver - Logs https://freenode.irclog.whitequark.org/etnaviv
cphealy has quit [Quit: Leaving]
pcercuei has quit [Quit: dodo]
rhyskidd has quit [Read error: Connection reset by peer]
rhyskidd has joined #etnaviv
juanrubio_ has quit [Ping timeout: 240 seconds]
_whitelogger has joined #etnaviv
JohnnyonF has quit [Read error: Connection reset by peer]
_whitelogger has joined #etnaviv
_whitelogger has joined #etnaviv
RzR has quit [Ping timeout: 258 seconds]
rhyskidd has quit [Remote host closed the connection]
Jookia has quit [Remote host closed the connection]
Jookia has joined #etnaviv
pcercuei has joined #etnaviv
juanrubio_ has joined #etnaviv
hanzelpeter has joined #etnaviv
hanzelpeter has quit [Quit: leaving]
JohnnyonFlame has joined #etnaviv
embden has joined #etnaviv
cphealy has joined #etnaviv
<JohnnyonFlame> I'm not sure how to report this one, it seems like the shader compiler is failing internally? (using https://github.com/glmark2/glmark2/commits/master) - https://gist.github.com/JohnnyonFlame/c419cf7b8b271c0d701d2f1984733098#gistcomment-3349703
<JohnnyonFlame> (ignore gistfile 1 to 3, that gist is where I'm keeping historic performance data etc)
<JohnnyonFlame> it seems like the application has no idea the compilation failed either
<austriancoder> JohnnyonFlame: is a known issue
<JohnnyonFlame> alrite, thank you
<austriancoder> too many shader instructions
<austriancoder> btw. there is an alternative nir based compiler: ETNA_MESA_DEBUG=nir
<JohnnyonFlame> I tried enabling it, but ETNA_MESA_DEBUG seems to not be doing anything here- I'm not sure how to check it
<JohnnyonFlame> IIRC that flag is by default enabled here due to a system flag, let me reboot to check
<JohnnyonFlame> gcw0:~ $ echo $ETNA_MESA_DEBUG
<JohnnyonFlame> nir
<JohnnyonFlame> running the application with $ LIBGL_DRIVERS_PATH=$(pwd) ./glmark2 --data-path=. -bterrain still results in the same error messages
<JohnnyonFlame> running "OpenGL ES 2.0 Mesa 20.1.1 (git-1602ffbdb0)"
<pcercuei> run with ETNA_MESA_DEBUG=''
<pcercuei> for TGSI
<JohnnyonFlame> pcercuei: I set compiler_msgs,linker_msgs on that run
<JohnnyonFlame> so that should've disabled nir
<pcercuei> yes
<JohnnyonFlame> iirc I did export ETNA_MESA_DEBUG='' at one point due to frustration
<JohnnyonFlame> > Creating 4.0 filesystem on /content/buildroot-gcw0/output/gcw0/images/rootfs.squashfs, block size 131072.
<JohnnyonFlame> sorry wrong channel
tlwoerner has quit [Quit: Leaving]
<flto> terrain scene isn't going to work on pre-GC3000, there's just not enough room for the shader (and NIR backend isn't good enough to make a small enough shader.. maybe if it was perfect it could, IIRC it was close)
<pcercuei> What's the max number of instructions?
<cphealy> 512
<cphealy> With GC3000 and newer it's "unlimited"
<pcercuei> How can it be unlimited?
<pcercuei> it has a MMU? :D
<cphealy> Well, not really unlimited but a lot.
<cphealy> And yes, the GPU has an MMU and starting with the GC3000 it's a improved MMU design which supports PPA (per process address space.)
<cphealy> Actually, according to NXP, the GC3000 supports "1 Million shader instructions (with I-Cache) per shader type"
<pcercuei> That's a lot
<pcercuei> So the 512 instructions on older GCs is a hard limit?
<cphealy> Here's the NXP document that references this limit delta between GC2000 and GC3000: https://www.nxp.com/docs/en/engineering-bulletin/EB810.pdf
<mntmn> hmm, i have a fun thing here where an application crashes because it reallocates/resizes a vertex buffer
<mntmn> glCopyBufferSubData
<mntmn> that ends up in util_blitter_blit_generic
<cphealy> Is it a common app that would be easy to reproduce?
<mntmn> ah they already have a vendor workaround to disable that in the code, for Intel :)
<mntmn> cphealy: nope.
<mntmn> cphealy: would be easy to construct a test case for it i think
<mntmn> but i'll try the workaround first...
<cphealy> ack
<cphealy> It would be good if there were a test case though so it could be fixed/validated.
<mntmn> yep, i'll note it