ChanServ changed the topic of #lima to: Development channel for open source lima driver for ARM Mali4** GPUs - Kernel has landed in mainline, userspace driver is part of mesa - Logs at https://people.freedesktop.org/~cbrill/dri-log/index.php?channel=lima and https://freenode.irclog.whitequark.org/lima - Contact ARM for binary driver support!
yuq825 has joined #lima
<yuq825> rellla: still need help?
kaspter has joined #lima
camus1 has joined #lima
kaspter has quit [Ping timeout: 258 seconds]
camus1 is now known as kaspter
dllud has quit [Remote host closed the connection]
dllud has joined #lima
camus1 has joined #lima
kaspter has quit [Ping timeout: 264 seconds]
camus1 is now known as kaspter
buzzmarshall has quit [Remote host closed the connection]
kaspter has quit [Quit: kaspter]
Barada has joined #lima
<rellla> yuq825: yes :( i just do not understand where and how to add creating and updating the buffer...
dddddd has quit [Ping timeout: 256 seconds]
<yuq825> you should already find in the _lima_resource_create_with_modifiers for PIPE_BIND_INDEX_BUFFER check, just create shadow mem there
<yuq825> update related change is at lima_transfer_map/unmap
chewitt_ is now known as chewitt
<anarsoul> yuq825: what about user indices?
<anarsoul> we won't be doing optimization in this case?
<yuq825> the shadow buffer is for hiding the cost of read from gpu mem, user indices don't need shadow bo
<yuq825> and we can't know user index update
<anarsoul> right
<anarsoul> that's unfortunate :(
<rellla> yuq825: and i use setup_miptree to get the size?
<yuq825> width or res->base->width0 has the size
<rellla> ok. it's the same value. why is it 1048576, which is 4096 * 16 * 16 for 763 indices?
<anarsoul> rellla: there's no miptree for indices
<anarsoul> rellla: likely it's 1m because u_uploader allocates it and you may have user indices
<rellla> ok. so now i need to change transfer_map/_unmap
<anarsoul> yuq825: how's the situation in China with coronavirus? Everything's back to normal?
<rellla> yuq825: is it really as good as press says?
<yuq825> not yet, come down from first class alarm to second, still some limitation
<anarsoul> :( looks like summer of 2020 is lost
<yuq825> any plan for that time?
<anarsoul> I had some plans, but looks like everything is cancelled
<anarsoul> Canada is on semi-voluntary lockdown, most of the businesses are closed
<anarsoul> national and provincial parks are closed
<yuq825> like February in China
<anarsoul> not as strict
<anarsoul> it's only a recommendation to stay home, I still can go outside if I want to
<anarsoul> but I've never seen streets so empty
<yuq825> wire feel for me too, but I got about two weeks' extra vacation
<rellla> i guess we have to skip our trip to italy in august :(
<yuq825> rellla: we should not allocate shadow bo for uploader bo created for user index case
<yuq825> that's just one time use drop bo and we can op on user buffer already
bhoj has quit [Remote host closed the connection]
<rellla> what do i have to do in lima_transfer_map and unmap exactly?
<rellla> ...seems i'm a bit lost :)
<yuq825> check for !PIPE_USAGE_STREAM && PIPE_BIND_INDEX_BUFFER for creating shadow bo
<yuq825> return the shadow bo address when lima_transfer_map
<yuq825> if map write, when unmap, write the shadow bo content to gpu bo
bhoj has joined #lima
<anarsoul> btw, has anyone had a chance to look into https://gitlab.freedesktop.org/mesa/mesa/-/issues/2701 ?
kaspter has joined #lima
<anarsoul> IIRC last time we had LIMA_GP_IRQ_PTR_ARRAY_OUT_OF_BOUNDS when we had some issue with scissor
yuq825 has quit [Remote host closed the connection]
yuq825 has joined #lima
<anarsoul> yeah, "/* 0x06192758 (0x00000458) */▸0x80158045 0x7072005b▸ /* SCISSORS: minx: 366.000000, maxx: 913.000000, miny: 69.000000, maxy: 44.000000 */"
<anarsoul> it doesn't look right
afaerber has quit [Ping timeout: 240 seconds]
dddddd has joined #lima
cwabbott_ has joined #lima
cwabbott has quit [Ping timeout: 246 seconds]
cwabbott_ is now known as cwabbott
afaerber has joined #lima
<rellla> ok, so it works for !has_user_indices now, let's see what is missing...
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
jernej has joined #lima
Barada has quit [Quit: Barada]
enunes has quit [Quit: leaving]
kaspter has quit [Quit: kaspter]
<rellla> i still need some help, why has_user_indices does not work anymore ...
yuq825 has quit [Ping timeout: 265 seconds]
buzzmarshall has joined #lima
niceplace has quit [Ping timeout: 264 seconds]
niceplaces has joined #lima
MoeIcenowy has quit [Ping timeout: 256 seconds]
MoeIcenowy has joined #lima
<anarsoul> rellla: don't do this optimization when has_user_indices is set
<anarsoul> you don't know when the buffer changed so you can't do it once and doing it every draw is expensive and likely will result in worse performance
<rellla> ok, but why does the current code not work for has_user_indices?
<rellla> i think i misunderstand user indices ;)
<anarsoul> rellla: I'll look into your code later today
<rellla> thanks, no hurry!
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #lima
<anarsoul> jernej: hey, are you using lima in libreelec?
<jernej> yes
<anarsoul> any issues with it?
<jernej> GUI seems slower with lima than binary driver
<jernej> but that's about it
<anarsoul> jernej: you may want to make sure that you're running GPU at max freq
<anarsoul> sunxi-mali sets it to max, lima runs with default
<jernej> currently there is still property to set clock to 384 MHz
<anarsoul> H3?
<jernej> yes
<anarsoul> heh
<anarsoul> IIRC sunxi-mali sets it to 576
<jernej> isn't A64 same?
<anarsoul> I have patch in my tree to set it to 432 on A64
<anarsoul> and it give 25% fps boost in q3a
<anarsoul> *gives
<jernej> anyway, I think I'll pick devfreq patches
<jernej> btw, we are using mesa 20.0.2 now
<anarsoul> yeah, basically each block is 1 instruction
<jernej> ok, good to know
<anarsoul> s/block/clock
<jernej> do I have to worry about temperature?
<anarsoul> if your shader consists of one instruction (can be the case for simple pass-through shaders) your fill rate is 50% higher
<anarsoul> jernej: I don't think that blob cares about temp
<jernej> I don't have a clue how shaders look like
<jernej> blob doesn't care, but BSP kernel driver did
<jernej> let me check old sources
<anarsoul> set it to 432mhz then? :)
<anarsoul> it works fine for me on 1080p pinebook
<anarsoul> it doesn't overheat
<jernej> hm... I can't find any temperature references
<jernej> I guess I remember wrong
<jernej> I guess that GUI is lightweight enough so I don't have to worry about temperature
<anarsoul> jernej: thanks for the pointer
<anarsoul> but yeah, I think it should be safe to run GUI with GPU@576MHz
<anarsoul> especially if you're using sunxi-mali kernel driver
<jernej> I forgot about that sunxi-mali set maximum frequency
dddddd has quit [Ping timeout: 256 seconds]
warpme_ has quit [Quit: Connection closed for inactivity]