00:16 <mntmn> how do i know which version of GCABI is in use on an imx8mq image?

00:23 <flto> its part of the driver version string

00:45 <mntmn> flto: what binary do i feed to build_json.sh to make a new .json?

00:47 <mntmn> this image has > Galcore version 6.2.4.150331

00:47 <mntmn> according to boundary it is > 6.2.4p1.8

00:48 <mntmn> i made a headers package and built libvivhook and made an fdr trace

00:48 <mntmn> but i cannot decode the trace because i don't have a fitting json. how do i generate it?

00:53 <mntmn> this maybe? https://gist.github.com/laanwj/b5ddea8d9e509c9e06fbd4e5c65bec8f

00:56 <mntmn> oh yeah this works... this should be included in etna_viv repo and documented

00:58 <mntmn> http://dump.mntmn.com/zdepth-blob-trace-20200609.txt

00:58 <mntmn> > RA.EARLY_DEPTH := 0x50000031

00:58 <mntmn> weird

00:59 <flto> mntmn: I did this process only once, I don't remember what binary I passed to it.. but I don't remember having to build something like that gcabi.c, anyway if it works thats good

01:00 <mntmn> yeah i finally have a blob tracing setup now

01:31 <mntmn> ah, seeing some new things now, like RA.EARLY_DEPTH := 0x11000031

01:43 <flto> mntmn: I should mention: I am assuming you are always trying with ETNA_MESA_DEBUG=no_early_z etnaviv does not try to disable early z with discard (for disabling based on shader, it only checks if fragdepth is written) - possibly that was fine on pre-GC7000 because it does not early-write depth

01:43 <mntmn> flto: yes, i have that, but also it doesn't seem to make a difference (yet)

01:44 <mntmn> it could be that that bit is somewhere else now?

01:47 <mntmn> woah

01:47 <mntmn> i think i have it

01:49 <cphealy> mntmn: if you figure it out, do you pay yourself? ;-)

01:50 <mntmn> yes

01:50 <mntmn> i think i really figured it out

01:50 <cphealy> Wow, that's awesome!

01:50 <pcercuei> it works?

01:50 <mntmn> yes it works

01:51 <pcercuei> kudos

01:53 <mntmn> i think RA_EARLY_DEPTH bit 0x10000000 in combination with DISABLE_ZS=0 (which has another meaning now and is normally always forced to =1 when halti>5) does the trick. i also set SUPER_TILED=1 because i thought it might have a different meaning, need to double check that

01:54 <mntmn> i'll do some quick cross checks...

01:58 <flto> mntmn: DISABLE_ZS=0 makes sense, I actually had that disabled in that gles3 branch I mentioned earlier.. SUPER_TILED would be always 1 for depth anyway so you didn't change anything there

01:58 <mntmn> ok, bit 0x01000000 was also set here, but that doesn't seem to have an effect

01:58 <mntmn> sorry, in RA_EARLY_DEPTH

01:59 <mntmn> oh, actually has an effect

02:00 <mntmn> yeah 0x01000000 also needs to be set in RA_EARLY_DEPTH

02:04 <mntmn> ok 0x10000000 needs to be on if DISABLE_ZS=0, or depth freezes. the comment in the code about that bits says depth will freeze the GPU, the missing info is that it has to be used in conjunction with that bit in RA_EARLY_DEPTH

02:10 <mntmn> ok so to boil it down, the solution to switch of early z is: RA_EARLY_DEPTH=0x11000031 and PE_DEPTH_CONFIG &= ~VIVS_PE_DEPTH_CONFIG_DISABLE_ZS

02:52 <cphealy> Way to go mntmn!

04:21 <austriancoder> this could correlate to the feature bit RA_WRITE_DEPTH which is only set on gc7000l

10:09 <Marex> mntmn: nice, now for the last step, patch

10:18 <mntmn> alright