alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
<alyssa> So I think we're okay to implement nir_op_exp2 as half_exp2
raster has quit [Quit: Gettin' stinky!]
stikonas has quit [Quit: Konversation terminated!]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
vstehle has quit [Ping timeout: 265 seconds]
Lyude has joined #panfrost
empty_string is now known as zempty_string
zempty_string is now known as empty_string
icecream95 has quit [Quit: leaving]
icecream95 has joined #panfrost
Stary has quit [Ping timeout: 264 seconds]
Stary has joined #panfrost
chewitt has quit [Read error: Connection reset by peer]
chewitt has joined #panfrost
chewitt has quit [Read error: Connection reset by peer]
chewitt has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
Elpaulo has joined #panfrost
davidlt has joined #panfrost
ckeepax1 has joined #panfrost
chrisf_ has joined #panfrost
ckeepax has quit [*.net *.split]
chrisf has quit [*.net *.split]
abordado has quit [*.net *.split]
chrisf_ has quit [Client Quit]
chrisf has joined #panfrost
vstehle has joined #panfrost
<tomeu> anarsoul: I have a timeout of 1h20m in my mesa fork because of that
<anarsoul> how do I configure it?
<tomeu> anarsoul: somewhere in the pipeline settings in your fork
<tomeu> ci/cd settings, I think it's called
<anarsoul> thanks
kaspter has quit [Quit: kaspter]
kaspter has joined #panfrost
kaspter has quit [Ping timeout: 258 seconds]
kaspter has joined #panfrost
kaspter has quit [Ping timeout: 264 seconds]
kaspter has joined #panfrost
camus1 has joined #panfrost
NeuroScr has joined #panfrost
kaspter has quit [Ping timeout: 265 seconds]
camus1 is now known as kaspter
<daniels> CL has some pretty savage precision requirements
cwabbott has joined #panfrost
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
stikonas has joined #panfrost
thefloweringash has quit [Quit: killed]
clementp[m] has quit [Quit: killed]
nhp[m] has quit [Quit: killed]
paulk-leonov has quit [Ping timeout: 250 seconds]
paulk-leonov has joined #panfrost
paulk-leonov has quit [Ping timeout: 240 seconds]
paulk-leonov has joined #panfrost
thefloweringash has joined #panfrost
nhp[m] has joined #panfrost
icecream95 has quit [Quit: leaving]
<tomeu> alyssa: pushed decoding of textures and samplers and emission of hardcoded descs to https://gitlab.freedesktop.org/tomeu/mesa/-/commits/bifrost
<tomeu> the emission only compile-tested
bnieuwenhuizen has quit [Ping timeout: 250 seconds]
raster has joined #panfrost
kaichi has quit [Quit: Leaving]
rhyskidd has quit [Quit: rhyskidd]
rhyskidd has joined #panfrost
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
<tomeu> alyssa: wonder if we shouldn't store the sampler info in a gpu-independent way in panfrost_sampler_view, and then upload the right descriptors at draw time
<tomeu> it's not that they are that big, and as long as we follow the rules about writing to uncached memory, should be fine
buzzmarshall has joined #panfrost
tbueno has joined #panfrost
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
cwabbott has quit [Remote host closed the connection]
cwabbott has joined #panfrost
<alyssa> daniels: Alternatively, CL demands that the implementation is correct, and GL just does a big *shrug* and waves you through the CTS
<alyssa> ;P
<alyssa> tomeu: Hmm? What's the benefit of doing it that way?
<tomeu> alyssa: so the differences between midgard, bifrost, etc are more contained
<alyssa> I'm not sure how that would help, though?
<alyssa> In a perfect world, I'd think there would be a panfrost_create_sampler_midg and panfrost_create_sampler_bi pair of routines in root panfrost
<alyssa> and then the CSO has union { midg; bi; }
<alyssa> where would GPU independent sampler_view help? If you need to reference properties, that's already in the .base we keep :-)
<tomeu> I'm just fearing having to debug state leaks in those descriptors if they are kept around and we have more than one version of them
<daniels> alyssa: heh :) though you may note CL's native_* functions
<alyssa> tomeu: I don't think that would be a big problem... at least, no worse than what we have now
<alyssa> Those descriptors should be *immutable*
<alyssa> once emitted, they should not be modified whatsoever
<alyssa> in the rare case that you do something like switch on AFBC, the whole descriptor should be regenerated from ground up
<alyssa> so no state to leak
<raster> alyssa: any reason nir_native_to_shader() should need a special converter and not just use c_native ?
<raster> ooh i missed a spot
<raster> for PIPE_FORMAT_B5G6R5_UNORM i mean above
<alyssa> raster: On chip, B5G6R5 is stored as a 16-bit field
<alyssa> In the shader, when you read/write from it, you get 4x 32-bit floats (or fp16)
<alyssa> So you need a routine to convert from packed 16-bit to unpacked floats
<raster> interesting as nir_native_to_shader() and nir_shader_to_native() are not very symetrical in format handling
<alyssa> yes, the file is very incomplete / minimal optimizations / probably broken
<alyssa> Kinda hard to get right w/o ISA docs.
<raster> aaah thats why i'm a bit confused
<raster> so like PIPE_FORMAT_B5G5R5A1_UNORM is handled in nir_shader_to_native but seemingly not in nir_native_to_shader() as it wouldn't have homogeneous bits... and not util_format_is_unorm8()
<alyssa> Yeah, incomplete
<raster> got it
<alyssa> Enough to write out rgb555a1 but not enough to read it
<alyssa> (This is all for blend shaders, BTW.)
<raster> yeah
<raster> now it's making sense - i just kind of went "ok c_native is a special read/write that knows a set of common texture formats inside"
<raster> i didnt hunt more
<raster> like a native texture format :)
<raster> got it
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
<raster> oooh that be some rendering bugs
<raster> thar be... i mean. pirate-like
nerdboy has joined #panfrost
cwabbott has quit [Ping timeout: 246 seconds]
cwabbott has joined #panfrost
nerdboy has quit [Ping timeout: 260 seconds]
<alyssa> yarr harr, fiddle-dee-dee. do whatcha ya want..
cwabbott has quit [Quit: cwabbott]
nerdboy has joined #panfrost
nerdboy has quit [Ping timeout: 256 seconds]
bnieuwenhuizen has joined #panfrost
cwabbott has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
davidlt has quit [Ping timeout: 265 seconds]
tomboy64 has quit [Remote host closed the connection]
tomboy64 has joined #panfrost
nerdboy has joined #panfrost
icecream95 has joined #panfrost
buzzmarshall has joined #panfrost
<mixfix41> i was really having trouble with a arm bug in firefox browser did you guys experience it on armv7?
<mixfix41> since panfrost enhances firefox so well and all that
<mixfix41> but 72.0b2 gets it done though i think arch linux has not been experience this i would guess
<alyssa> ?
<mixfix41> something where firefox ran only as root
<mixfix41> and would seg fault as non root
<mixfix41> so i was really absorbed since i launched firefox as root not knowing and was wondering how it ran from when i encountered the update
<HdkR> Is it an ARM bug or just a permissions problem from having run it as root?
<HdkR> strace it as the user to see if it tries accessing something and fails?
<mixfix41> i might have those recorded.. i was doing that but to new to strace and the debug symbols are hard to complete a build
<mixfix41> yea
<mixfix41> if i changed Xauthority to root
<mixfix41> it would launch with sudo
<mixfix41> something hard coded
<mixfix41> and simple(sp?)_ it felt
<mixfix41> i came upon a bunch of threads yes its just armv7 32bit i think but i was wondering since the official precompiled package did this from slackware it would be more noted i guess
<mixfix41> if i could figure it out seems like a cool internal thing to know about ff but those functions stuff is realyl over my head
<HdkR> That's just someone hecking up permissions of the /dev/shm region
<mixfix41> no but firefox wont launch on 72 so i just need it to run
<HdkR> If you also have /dev/shm permission problems then it would explain the problem
<mixfix41> yea its more friendly then the ff-6x.x.x seg faults
nerdboy has quit [Ping timeout: 264 seconds]
nerdboy has joined #panfrost
kaichi has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
<alyssa> Okay, flog2/fexp2 work
<alyssa> which theoretically should mean I'm clear for any exponential/logarithm/pow for GL
<alyssa> hopefully
<alyssa> not out of the clear for special functions, sin/cos will still be monsters.