#
✕ nick:alyssa
#asahi-gpu · 2022-10-14 00:44
alyssaHW bug, maybe
#asahi-gpu · 2022-10-14 00:37
alyssaand this workaround is only because of a missing optimization!
#asahi-gpu · 2022-10-14 00:37
alyssait's all hw so far
#asahi-gpu · 2022-10-14 00:37
alyssanope.
#asahi-gpu · 2022-10-14 00:33
alyssathis is.. really iffy.
#asahi-gpu · 2022-10-14 00:27
alyssaunclear if that's actually flushing anything (acting as a barrier), or it's just masking the timing-dependent bug
#asahi-gpu · 2022-10-14 00:27
alyssaif I insert an intervening nop, the problem isn't visible
#asahi-gpu · 2022-10-14 00:25
alyssaso IDK if they have a workaround
#asahi-gpu · 2022-10-14 00:20
alyssaor they can just statically guarantee it doesn't happen
#asahi-gpu · 2022-10-14 00:20
alyssaso IDK if they have a workaround
#asahi-gpu · 2022-10-14 00:20
alyssaI don't know if it's possible to emit that sequence with the blob
#asahi-gpu · 2022-10-13 23:44
alyssaon Mali that's a special instruction
#asahi-gpu · 2022-10-13 23:44
alyssai.e. a fast path for "texture2D(tex, v_texcoords.xy)"
#asahi-gpu · 2022-10-13 23:44
alyssaDebugging *did* lead me to a bit that controls forwarding of varyings to sample coordinates
#asahi-gpu · 2022-10-13 23:43
alyssaAGX seems to be inconsistent?
#asahi-gpu · 2022-10-13 23:43
alyssaEvery GPU uarch does something different here
#asahi-gpu · 2022-10-13 23:43
alyssaTBD
#asahi-gpu · 2022-10-13 23:40
alyssaSo this might be an iter specific quirk
#asahi-gpu · 2022-10-13 23:40
alyssaI have a code sample from the blob doing writes-after-writes with ldcf
#asahi-gpu · 2022-10-13 23:29
alyssano idea about the uarch
#asahi-gpu · 2022-10-13 23:29
alyssaphire: yes, interpolates varyings
#asahi-gpu · 2022-10-13 23:28
alyssacr1901: yeah, in exceptionally rare cases
#asahi-gpu · 2022-10-13 23:06
alyssaAt a uarch level I men
#asahi-gpu · 2022-10-13 23:06
alyssaWhat I don't know is why.
#asahi-gpu · 2022-10-13 23:06
alyssaMy guess is that there is a write-after-write dependency between the iter instructions
#asahi-gpu · 2022-10-13 23:06
alyssaExtraordinarily rare flake
#asahi-gpu · 2022-10-13 23:03
alyssathis program is flaking. it really should not be
#asahi-gpu · 2022-10-13 23:03
alyssahttps://rosenzweig.io/flake.txt
#asahi-gpu · 2022-10-13 23:02
alyssaughhh does AGX do magical dependency handling for iter instructions
#asahi-gpu · 2022-10-13 22:30
alyssait doesn't
#asahi-gpu · 2022-10-13 15:11
alyssafearless concurrency
#asahi-gpu · 2022-10-12 21:43
alyssathat sounds dcp
#asahi-gpu · 2022-10-12 21:36
alyssaquery sven
#asahi-gpu · 2022-10-12 21:36
alyssahello fruit farmer
#asahi-gpu · 2022-10-12 19:59
alyssalol
#asahi-gpu · 2022-10-12 19:28
alyssao:)
#asahi-gpu · 2022-10-12 19:28
alyssayou're the dcp maintainer now, it's your call :p
#asahi-gpu · 2022-10-12 19:28
alyssashrugs
#asahi-gpu · 2022-10-12 19:19
alyssai don't
#asahi-gpu · 2022-10-12 19:17
alyssaacronym
#asahi-gpu · 2022-10-12 19:16
alyssaRiiR is a very important algorithm for being friends with lina~ ;-p
#asahi-gpu · 2022-10-12 19:14
alyssaI'd rather it weren't but otoh maybe the mode parser should be.
#asahi-gpu · 2022-10-12 19:13
alyssais dcp getting RiiR'd or not I can't remember
#asahi-gpu · 2022-10-12 19:12
alyssaseems legit
#asahi-gpu · 2022-10-12 19:10
alyssaone of those needs to change, idk what colour the bikeshed should be
#asahi-gpu · 2022-10-12 19:10
alyssabut dcpep.c assumes 0 means there's a valid mode
#asahi-gpu · 2022-10-12 19:09
alyssaactual bug though: if no modes match, the mode parser returns 0 (success) with best mode = -1
#asahi-gpu · 2022-10-12 19:08
alyssathis one is bogus
#asahi-gpu · 2022-10-12 19:07
alyssa WARN_ON(endpoint != DCP_ENDPOINT);
#asahi-gpu · 2022-10-12 19:07
alyssathese are bogus assertions that should be dev_err instead

plus ›