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
warpme_ has quit [Quit: Connection closed for inactivity]
stikonas has quit [Ping timeout: 252 seconds]
jolan has quit [Quit: leaving]
NeuroScr has quit [Quit: NeuroScr]
jolan has joined #panfrost
jolan has quit [Remote host closed the connection]
tgall_foo has quit [Read error: Connection reset by peer]
jolan has joined #panfrost
NeuroScr has joined #panfrost
<icecream95> alyssa: I've finished getting tiling working for ETC textures
megi has quit [Ping timeout: 260 seconds]
<icecream95> Running dEQP-GLES3.functional.texture.format.*, I get `Passed: 538/538 (100.0%)`, so it looks like formats like RGBA32I work after all...
vstehle has quit [Ping timeout: 240 seconds]
nerdboy has quit [Ping timeout: 260 seconds]
nerdboy has joined #panfrost
tgall_foo has joined #panfrost
nerdboy has quit [Ping timeout: 258 seconds]
vstehle has joined #panfrost
davidlt has joined #panfrost
davidlt has quit [Read error: Connection reset by peer]
<tomeu> icecream95: wow, awesome!
davidlt has joined #panfrost
<tomeu> narmstrong: any plans to add afbc support to the meson drm driver?
<tomeu> leinax: if you give others the ability to push to your leaks MR, I can help it make it to master
<narmstrong> tomeu: done !
<tomeu> hehe, nice
* tomeu will be trying to get supertuxkart running as fast as possible on a s912
<icecream95> Be aware that the FPS number printed from running supertuxkart with --profile-* is not accurate.
<narmstrong> Nice !
bbrezillon has joined #panfrost
nerdboy has joined #panfrost
<tomeu> icecream95: oh, I just want a nice game experience
<tomeu> narmstrong: hmm, just partitioned the whole of emmc, and now I cannot boot from the sd card
<tomeu> so looks like the bootloader in emmc was involved when booting from sd
<narmstrong> tomeu: yep but you can copy this bootloader to the sdcard and it will boot from the sdcard
<tomeu> narmstrong: ah, got it
<tomeu> narmstrong: what's the tool for flashing via usb?
<narmstrong> tomeu: you have a amlogic format update package ? ou you want to dump a binary to the emmc ?
<tomeu> narmstrong: the latter
<narmstrong> tomeu: it can't be used like that, the tool (http://openlinux2.amlogic.com:8000/download/A113/Tool/flash-tool-v4.7/flash-tool/tools/update) or (https://github.com/superna9999/pyamlboot) is used to load an u-boot from usb then with this u-boot you can flash whatever you want
<tomeu> ah, cool
<tomeu> narmstrong: guess none of these files will work no the nexbox a1 and I should get them from somewhere else?
<narmstrong> tomeu: did you dump the emmc ? if yes you can write it to the sdcard and it will boot
<tomeu> ok, will do just that
<tomeu> I'm unwrapping the other unit to do so now
<icecream95> I think the 0x14-0x17 formats are ASTC. No idea why there are four which are all 4x4 block size.
megi has joined #panfrost
<icecream95> Probably the unknown fields in the texture format struct encode the block size...
<icecream95> Does Mesa have some sort of licensing deal with Khronos? The ARM ASTC encoder requires me to have "licences to develop such products under the relevant Khronos agreements" to encode images.
<chewitt> tomeu: would be interesting to see if u-boot.bin.sd.bin from https://github.com/BayLibre/u-boot/releases/download/v2017.11-libretech-cc/khadas-vim2_fip_20180207.tar.gz will boot the nexbox
warpme_ has joined #panfrost
<narmstrong> for the VIM2, khadas has a neat tool to build full mainline images: https://github.com/khadas/fenix
<tomeu> chewitt: even without a DT for it?
<chewitt> the VIM2 device tree is fairly generic
<chewitt> so you might get lucky :)
<tomeu> chewitt: I don't have serial though, so it could be quite painful to get it to do something useful
<chewitt> ahh, that's probably essential for getting somewhere with it
<chewitt> prob. not too hard to add mainline u-boot support tho
<tomeu> yeah, was hoping not to have to solder anything :)
<tomeu> narmstrong: chewitt any ideas on how to partition the sd card without overwriting the bootloader?
<chewitt> in LE we start partitions from 2048 sectors in, I think
<chewitt> Number Start End Size Type File system Flags
<chewitt> 1 8192s 1056767s 1048576s primary fat16 boot, lba
<chewitt> 2 1056768s 30310399s 29253632s primary ext4
<chewitt> 8192s, not 2048
<chewitt> the signed u-boot needs to be in a specific location for the SoC to see/load it
<chewitt> that's in the URL
<icecream95> 0x14 is definitely ASTC. I've already got an [illegally encoded] texture displaying...
<icecream95> 0x15-0x17 seem to be exactly the same.
<tomeu> chewitt: oh, I already have the bootloader in the SD card and works fine, but I also need to have the boot and root partitions there because I don't have anywhere else
<tomeu> looks like the partition table also needs to be in the sector 0
<tomeu> how can both the partition table and the bootloader coexist in the same sd card?
<icecream95> Use GPT and only the secondary table at the end of the drive?
<chewitt> MBR is sector 0, the SoC looks for something at a specific offset, which on GXL/GXM is 512 bytes or sectors (forget which) in
<chewitt> if you write u-boot.bin.sd.bin starting sector 0 the offset is in the right place automatically
<tomeu> oh, got it
<chewitt> then create the partition scheme
<chewitt> this overwrites sector 0, but there's enough space before you get to signed firmware
<chewitt> (unlike older GXBB which looks for signed firmware in sector 0 .. bit of a design fubar)
yann has quit [Ping timeout: 240 seconds]
NeuroScr has quit [Quit: NeuroScr]
<tomeu> hrm, looks like creating the partition table has overwritten the bootloader
yann has joined #panfrost
<icecream95> Does anyone have any ideas for where the field for ASTC block size could be?
icecream95 has quit [Ping timeout: 265 seconds]
raster has joined #panfrost
<chewitt> unzip and dd to an sd card
* tomeu gives it a try
<tomeu> chewitt: no luck
<tomeu> chewitt: guess the offset at which the bootloader needs to be isn't known?
<tomeu> guess I could try writing 1-2 sectors after sector 0
<narmstrong> tomeu: dd if="$INSTALL/usr/share/bootloader/u-boot.bin.sd.bin" of="$DISK" conv=fsync,notrunc bs=1 count=444 >"$SAVE_ERROR" 2>&1 || show_error
<narmstrong> dd if="$INSTALL/usr/share/bootloader/u-boot.bin.sd.bin" of="$DISK" conv=fsync,notrunc bs=512 skip=1 seek=1 >"$SAVE_ERROR" 2>&1 || show_error
<chewitt> ^ that's the same as used in the image I posted
<chewitt> which is using the vim2 fip sources (on a chance)
<chewitt> and u-boot 2020.01
<tomeu> isn't that overwriting the first sector?
<narmstrong> the last byte of the first sector must be preserved
<tomeu> I still don't get it: the first image is going to be written to bytes 0-444, thus overwiting the MBR
<narmstrong> nop, the MBR is later
<tomeu> ok, looks like writing the second block worked
megi has quit [Quit: WeeChat 2.7]
<urjaman> the MBR partition table is bytes 446 to 509
<tomeu> oh, that explains, thanks
<tomeu> the second write seems to corrupt the GPT table somehow
<urjaman> i thought you were only using the MBR
<urjaman> the GPT is later
<tomeu> a ok, that's it then
<urjaman> the primary GPT is sectors 1 to 33
<tomeu> ok, will try writing only after 33
<tomeu> hrm, the bootloader seems to be in those sectors
<tomeu> would a DOS partition table be better?
<urjaman> that's what i meant when i said MBR
<tomeu> ok, let's see
<tomeu> no luck
<tomeu> maybe I didn't copy out a big enough chunk of the working emmc
<tomeu> was expecting 1M to be enough
davidlt has quit [Ping timeout: 260 seconds]
bbrezillon has quit [Quit: WeeChat 2.7]
bbrezillon has joined #panfrost
<narmstrong> yeah GPT can't work with the amlogic boot binary, only MBR
megi has joined #panfrost
<alyssa> icecream95: Woohoo!
<alyssa> icecream95: Woohoo!
<alyssa> Uh, double ok
<alyssa> icecream95: I would expect ASTC parameters, if they're implicit, would be in an unknown from mali_texture_format or mali_texture_descriptor
<alyssa> Of which there are many
<alyssa> Anisotropic filtering parameters I would expect to be in mali_sampler_descriptor, perhaps in the upper unknown bits of filter_mode currently seen just as 0
<alyssa> But that's Bifrost-only, I think
alpernebbi has joined #panfrost
davidlt has joined #panfrost
<tomeu> narmstrong: btw, booting after writing the emmc contents to a sd card doesn't seem to work
BenG83 has joined #panfrost
<leinax> tomeu: thanks, just enabled that setting
guillaume_g has quit [Quit: Konversation terminated!]
<BenG83> so I found out how to switch glamor to use either GL or GLES2, but both variants break panfrost in xwayland, but was worth a shot...
<BenG83> running out of easy things to poke at, I wonder what is the difference in xwayland/glamor using panfrost and native wayland applications
raster has quit [Quit: Gettin' stinky!]
stikonas has joined #panfrost
karolherbst has quit [Quit: duh 🐧]
jernej has quit [Ping timeout: 260 seconds]
karolherbst has joined #panfrost
warpme_ has quit [Quit: Connection closed for inactivity]
davidlt has quit [Ping timeout: 258 seconds]
jernej has joined #panfrost
alpernebbi has quit [Quit: alpernebbi]
<HdkR> :D
<HdkR> $280 Helio P60T Chromebook
<alyssa> HdkR: "Lenovo plans to start sales of the IdeaPad Duet Chromebook in April."
<alyssa> Don't tell me you're thinking of a launch-day Bifrost driver? =P
<HdkR> hah
<alyssa> Be the change..
icecream95 has joined #panfrost
icecream95 has quit [Ping timeout: 260 seconds]
icecream95 has joined #panfrost
chewitt has quit [Quit: Zzz..]
<icecream95> unknown3 in mali_texture_descriptor is two 8-bit fields, not 1 16-bit one.
<alyssa> icecream95: Ah?
<icecream95> Scratch that. It's just the two high bits of unknown3 seem to be flags that can be set to one, the rest of the bits have to be zero. No idea what the flags do yet...
<alyssa> Weeeeeee.
<alyssa> icecream95: isn't witchcraft fun? :O
<anarsoul> sure it is
<icecream95> Especially when you are trying to go through all the values in a 32-bit field...
BenG83 has quit [Ping timeout: 268 seconds]
<icecream95> If anyone wants to join in with the hunt for the secret block size field: https://gitlab.freedesktop.org/snippets/794
<alyssa> :eyes:
<daniels> stk shadows! \o/
NeuroScr has joined #panfrost
<alyssa> daniels: Hush this was supposed to be a surprise for when you got back :V
<alyssa> =P
* daniels goes back to lurking in the shadows
<alyssa> =P
<HdkR> The shadows must be known
<alyssa> You know, "Tomeu, Boris and I go and write an ES3.0 implementation and codedrop when you come back from holiday"
<alyssa> It would've worked, too, if I wasn't so chatty ;)
<icecream95> Does that mean OpenMW will stop causing kernel panics now?
<alyssa> Hmm?
<icecream95> After the first shadows commits to Mesa it started causing hangs...
* mmind00 feels a Scooby Doo vibe from his childhood ... "And I would have gotten away with it too, if it weren't for you meddling kids"
<daniels> alyssa: :D
<alyssa> 13 files changed, 169 insertions(+), 114 deletions(-)
<alyssa> diff from master for the stk branch isn't awful
<HdkR> alyssa: You can do indirect UBO binding indexing, but I'm not sure which version of GL supports it
<alyssa> ES3.2 appears to
NeuroScr has quit [Quit: NeuroScr]
leinax has quit [Remote host closed the connection]