#
#asahi-gpu · 2021-11-13 23:28
alyssawhich together make up the majority of fails
#asahi-gpu · 2021-11-13 23:27
alyssaor fixing mipmapping I guess
#asahi-gpu · 2021-11-13 23:27
alyssalooks like I never got around to depth/stencil attachments, discard(), or cube maps
#asahi-gpu · 2021-11-13 23:21
alyssa*technically*
#asahi-gpu · 2021-11-13 23:21
alyssabut... yes technically passing 93% of deqp-gles2
#asahi-gpu · 2021-11-13 23:20
alyssabut i digress
#asahi-gpu · 2021-11-13 23:20
alyssaso I would expect the M1 to be able to do a run in 10 seconds flat
#asahi-gpu · 2021-11-13 23:20
alyssathis is still too slow, my mali board does them in about 100s
#asahi-gpu · 2021-11-13 23:19
alyssaa bit more of glmark2 works, and with an embarassing little hack deqp-gles2 runs are down to 6:06 from well over 10min before
#asahi-gpu · 2021-11-13 23:19
alyssaso stack of bug fixes later....
#asahi-gpu · 2021-11-13 23:19
alyssadougall nerdsniped me into doing some Mesa stuff today for AGX
#asahi-gpu · 2021-10-29 16:20
alyssaagoldmints[m]: yep, that's the one
#asahi-gpu · 2021-10-29 16:12
agoldmints[m]You have to opt into using it in your compiler I think, and it only lowers the indirect accesses that your backend can't handle directly.
#asahi-gpu · 2021-10-29 16:10
agoldmints[m]Yes, Mesa has a lowering pass to lower indirect accesses into if trees, nir_lower_indirect_derefs
#asahi-gpu · 2021-10-29 01:53
alyssaOr can do it depending on backend options or something
#asahi-gpu · 2021-10-29 01:53
alyssaI believe Mesa does this fwiw
#asahi-gpu · 2021-10-29 01:53
alyssanods
#asahi-gpu · 2021-10-28 23:39
phireafter talking with me, they actually expermented with an optimsation pass that converted unform indirection of small arrays into if/else trees (which I was expermenting with manually to speed up Nvidia GPUs
#asahi-gpu · 2021-10-28 23:37
phireI was only using uniform indexing into the register file. When I was talking to an Apple driver engineer back in like 2015, it didn't support it
#asahi-gpu · 2021-10-28 22:53
alyssaI don't believe AGX can do dynamic indexing of the register file.
#asahi-gpu · 2021-10-28 22:52
alyssaThe former is still fast (since it'll be hot in cache) but performing a memory load is inherently more expensive than a pushed uniform that can be accessed anywhere for free
#asahi-gpu · 2021-10-28 22:51
alyssaAny reasonable ISA can do indirection on the former, I don't know if AGX can do indirection on the latter
#asahi-gpu · 2021-10-28 22:51
alyssaOne is literally doing a load from constant/uniform memory, and the other is pushing the uniform to a dedicated uniform register / push constant / fast access uniform / register-mapped uniform
#asahi-gpu · 2021-10-28 22:50
alyssaphire: There are two issues being mixed up here (and in the Apple presentation I dug up)
#asahi-gpu · 2021-10-28 22:28
phirebecause some shader ISAs simply don't support it, and their drivers will instead put the array you were dynamic indexing out into main memory, and preformance tanked
#asahi-gpu · 2021-10-28 22:26
phirethen again, the upershaders were more bound on indexing into the dynamic register file.
#asahi-gpu · 2021-10-28 22:22
phirealyssa: oh, I didn't know GPU/compilers had problems doing indirection on the constant/uniform file. I rely a lot on that in dolphin's ubershader
#asahi-gpu · 2021-10-28 18:08
alyssathere will be magic bits for HSR, [[early_fragment_tests]] may help identify which
#asahi-gpu · 2021-10-28 18:06
alyssasomething like that
#asahi-gpu · 2021-10-28 18:06
alyssatypedef struct { data[8] } T
#asahi-gpu · 2021-10-28 18:06
alyssaconstant T &arr [[buffer(0)]]
#asahi-gpu · 2021-10-28 18:06
alyssarereading the Metal optimization guide suggests "maybe, if you hint the compiler just right"
#asahi-gpu · 2021-10-28 18:06
alyssaCurious if AGX supports indirection on the constant (uniform) file
#asahi-gpu · 2021-10-21 12:22
marcanthis is a dev channel; you want #asahi for general questions about the project
#asahi-gpu · 2021-10-21 12:21
marcanand yes, please change your nick
#asahi-gpu · 2021-10-21 12:21
marcanfor windows games FEX-Emu is what you're looking for
#asahi-gpu · 2021-10-20 21:34
opticronand in the case of the GPU, also the userland shader compilers
#asahi-gpu · 2021-10-20 21:32
opticronmacOS conflates a bunch of tech beyond the core OS and asahi is really the core OS and drivers
#asahi-gpu · 2021-10-20 21:31
aeadGues__________________________: Can you change your nick to something not completely stupid please
#asahi-gpu · 2021-10-20 21:31
opticronas far as accelerated execution of non-native binaries, that would be a different project that I think is already underway and applies to ARM64/AArch64 in general
#asahi-gpu · 2021-10-20 21:29
opticronI'm mostly a lurker
#asahi-gpu · 2021-10-20 21:29
opticroniirc, vulkan is a secondary objective, but the eventual work that could be done to enable vuklan on linux should be directly applicable to running vulkan on macOS. someone else is likely to have better information than me, though
#asahi-gpu · 2021-10-20 21:28
Gues__________________________If not, will Window games work under Linux with box64 + proton? It doesn't look like box64 is nearly as fast or complete as Rosetta 2.
#asahi-gpu · 2021-10-20 21:26
Gues__________________________Will this project enable native Vulkan in macOS (not Linux)? Would it be good enough for VKD3D to run Windows games? If so that would be amazing.
#asahi-gpu · 2021-10-18 20:31
kode54I have to log in with a VNC client to find that my "displays" are still emitting pictures to nothing, and I have to temporarily switch the rotation on one of them to wake the displays up again
#asahi-gpu · 2021-10-18 20:30
kode54where force quitting a locked up BasiliskII or Sheepshaver causes the DCP to shut off or something
#asahi-gpu · 2021-10-18 20:30
kode54my M1 Mac did the thing again
#asahi-gpu · 2021-10-18 20:30
kode54cool
#asahi-gpu · 2021-10-18 19:52
phireno idea how it would handle mapping memory
#asahi-gpu · 2021-10-18 19:52
phireSo my original idea was to just modify qemu to bash-though the AGX interface.

plus ›