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!
tlwoerner has joined #lima
<anarsoul> here PLBU_CMD_ARRAYS_SEMAPHORE_END() usually goes *after* PLBU_CMD_DRAW_ELEMENTS()
<anarsoul> nevermind, it's actually PLBU_CMD_DRAW_ARRAYS()
<anarsoul> so I have a hypothesis
<anarsoul> semaphore synchronizes PLBU and VS
<anarsoul> VS_CMD_ARRAYS_SEMAPHORE_BEGIN_1() is 0x00028000, 0x50000000, immediately follows by VS_CMD_ARRAYS_SEMAPHORE_BEGIN_2() which is 0x00000001, 0x50000000
<anarsoul> likely means: wait for semaphore to become 0x8000 (likely -1), then increase it by 0x0002, then wait for semaphore to become 0x0001, increase it by 0x0000
<anarsoul> then:
<anarsoul> PLBU_CMD_ARRAYS_SEMAPHORE_BEGIN() is 0x00010002, 0x60000000 and PLBU_CMD_ARRAYS_SEMAPHORE_END() is 0x00010001, 0x60000000
<anarsoul> 0x00010002, 0x60000000 is wait semaphore to become 0x0002, then decrease it by 0x0001
<anarsoul> and 0x00010001, 0x60000000 means wait for semaphore to become 0x0001, then decrease it by 0x0001
<anarsoul> so PLBU at the end of cmd stream should have semaphore at 0x0
<anarsoul> while for VS it depends on whether draw is indexed or not
<anarsoul> hm, however for VS it's also should be 0x0
<anarsoul> but for non-indexed draw semaphore should have 0x8000 at start (-1)? wtf?
<anarsoul> rellla: ^^
<anarsoul> likely we have issues with indexed and non-indexed draws mixed in the same frame
<anarsoul> it all adds up nicely except for non-indexed draws
<anarsoul> doesn't add up at all
<anarsoul> I guess I need to think a bit about it :)
dddddd has quit [Remote host closed the connection]
kaspter has quit [Ping timeout: 260 seconds]
kaspter has joined #lima
_whitelogger has joined #lima
Barada has joined #lima
_whitelogger has joined #lima
<anarsoul> another guess: "0x00028000 0x50000000" in vs cmd could mean "wait for any (where 0x8000 is any), increment by 2
<anarsoul> then following "0x00000001 0x50000000" waits till PLBU decreses semaphore value via "0x00010002 0x60000000" plbu cmd which is "wait for 0x2, decrease by 0x1"
<anarsoul> vs proceeds further, gets to "draw", cmd, then to "0x00000000 0x50000000" which is wait for 0x0, don't increment
<anarsoul> plbu at the same time does "DRAW_ARRAYS", then gets to "0x00010001 0x60000000"
<anarsoul> which is wait for 0x1, decrement by 0x1 and it unblocks VS
<anarsoul> semaphore is at 0 now
<anarsoul> vs proceeds, does another draw, then gets to "0x00018000 0x50000000" which is "wait for any, increment by 1", semaphore becomes 1
<anarsoul> I suspect that semaphore usage in lima is likely implemented incorrectly
<rellla> anarsoul: i will also think about it, but likely only after the christmas days :)
<anarsoul> :)
<rellla> sounds complicated, and i'm still on finishing syscall-tracker, only texture left ...
<rellla> in case you want to use it, take https://gitlab.freedesktop.org/rellla/mali-syscall-tracker/commits/parser, it fixes 2-3 issues with decoding.
<anarsoul> rellla: if you have it up and running can you make a dump of 'glmark2 -b ideas' of blob?
<rellla> i think i can do that later, but there should already be an older one online somewhere
<rellla> an older one, no clue how pretty it is parsed, but maybe it helps for now
<rellla> s/clue/idea/
_whitelogger has joined #lima
Elpaulo has joined #lima
kaspter has quit [Remote host closed the connection]
kaspter has joined #lima
<rellla> anarsoul, enunes: probably useful as a reference http://imkreisrum.de/deqp/mali400-r6p2_on-allwinner-a20_all-tests/
<rellla> 107 fails
_whitelogger has joined #lima
adjtm has quit [Ping timeout: 258 seconds]
Barada has quit [Quit: Barada]
kaspter has quit [Ping timeout: 265 seconds]
kaspter has joined #lima
dddddd has joined #lima
embed-3d has quit [Quit: http://quassel-irc.org - Chat comfortably. Anywhere.]
embed-3d has joined #lima
embed-3d has quit [Client Quit]
embed-3d has joined #lima
<anarsoul> rellla: thanks for ideas dump
adjtm has joined #lima
<anarsoul> yeah, it's complicated :)
yann has quit [Ping timeout: 260 seconds]