#
#asahi-gpu · 2023-08-20 13:14
alyssacan you try building asahi/mesa:main and seeing if it still reproduces?
#asahi-gpu · 2023-08-20 13:14
alyssajannau: oh, hmm
#asahi-gpu · 2023-08-20 13:14
alyssasee st_getTexSubImage
#asahi-gpu · 2023-08-20 13:12
alyssaoff hand i'm not 100% sure why that isn't happening / how to make it happen
#asahi-gpu · 2023-08-20 13:11
alyssaideally, the format conversion would happen as part of the blit on the GPU
#asahi-gpu · 2023-08-20 13:11
alyssajannau: that would indeed be CPU format conversion :|
#asahi-gpu · 2023-08-20 10:48
jannauutil_format_b10g10r10x2_unorm_unpack_rgba_8unorm() called from _mesa_GetTexImage
#asahi-gpu · 2023-08-20 10:47
jannauhmm, it's not kpipewire but kwin causing the high cpu load
#asahi-gpu · 2023-08-20 00:20
alyssamight be worth fiddling with PIPE_CAP_TEXTURE_TRANSFER_MODES
#asahi-gpu · 2023-08-20 00:20
alyssaworth checking a profiler (I like sysprof) to see what's going wrong
#asahi-gpu · 2023-08-20 00:19
alyssaI thought we did that properly, but it's possible we're going down a slow path somewhere deep in mesa .. CPU format conversion directly from write-combine memory, for ex, would cause unusable slow..
#asahi-gpu · 2023-08-20 00:19
alyssajannau: that 'should' be implemented by a blit on the GPU to a writeback buffer and then a CPU read from the writeback buffer, which is b/w heavy but should not cause unusable slowness on m1-class hardware
#asahi-gpu · 2023-08-19 18:43
jannauresulting in an almost unusable slow system after starting screen sharing with krfb
#asahi-gpu · 2023-08-19 18:42
jannauasking for this gem: https://invent.kde.org/plasma/kpipewire/-/blob/master/src/dmabufhandler.cpp#L185
#asahi-gpu · 2023-08-19 18:41
jannauis there anything to be done in the driver to improve glReadPixels(..., GL_BGRA, ...) on a fbo backed by b10g10r10x2 texture?
#asahi-gpu · 2023-08-17 23:58
alyssaIDK why they're checking only the renderer
#asahi-gpu · 2023-08-17 23:58
alyssathis has the commit hash
#asahi-gpu · 2023-08-17 23:58
alyssaOpenGL core profile version string: 3.1 Mesa 23.3.0-devel (git-2866d6991e)
#asahi-gpu · 2023-08-17 23:56
alyssathanks :)
#asahi-gpu · 2023-08-17 23:56
alyssahere I was worried this was a regression I introduced ;~P
#asahi-gpu · 2023-08-17 23:56
alyssajannau: oh, good
#asahi-gpu · 2023-08-17 23:48
jannausee https://github.com/AsahiLinux/linux/issues/72 there is an upstream bug but I think we're partly to blame as well since the mesa version doesn't change
#asahi-gpu · 2023-08-17 23:46
jannau ~/.config/chromium/Default/GPUCache
#asahi-gpu · 2023-08-17 23:46
jannaualyssa: Chromium has its own shader cache
#asahi-gpu · 2023-08-17 23:46
alyssaThe on-screen symptom is that Chromium rendering is totally broken
#asahi-gpu · 2023-08-17 23:45
alyssaMESA_SHADER_CACHE_DISABLE=1 'seems' to workaround which points at a Mesa bug but it's hard to say
#asahi-gpu · 2023-08-17 23:45
alyssaI can't tell if this a Mesa bug or a Chromium one
#asahi-gpu · 2023-08-17 23:45
alyssa[4983:4983:0817/194416.716259:ERROR:shared_context_state.cc(81)] Skia shader compilation error
#asahi-gpu · 2023-08-17 23:45
alyssalink failed but did not provide an info log
#asahi-gpu · 2023-08-17 23:45
alyssaErrors:
#asahi-gpu · 2023-08-17 23:45
alyssasometimes it fails to start properly and spams stderr with messages like
#asahi-gpu · 2023-08-17 23:45
alyssasince rebasing I'm getting weird Chromium issues
#asahi-gpu · 2023-08-17 23:32
jannauother than that it seems to work ok. comes up reliably at boot and after display standby. only lightly tested
#asahi-gpu · 2023-08-17 23:29
jannauThis is a problem on fedora since clang and bindgen use libllvm.so from llvm-15/16 which breaks build with asahi
#asahi-gpu · 2023-08-17 23:28
jannaurequires 'pd_ignore_unused' (haven't checked why yet), there seems to be some kind of memory hazard in afk.c or dptx.c. works for me only if compiled with clang
#asahi-gpu · 2023-08-17 23:25
jannauonly atc1 works, should be on all machines the port next to the usb-c port with serial
#asahi-gpu · 2023-08-17 23:24
jannaukernel branch: https://github.com/jannau/linux/tree/asahi-6.4-7pre-dcpext_dp_t602x - only works on t602x with 13.5 fw, tested on j474s
#asahi-gpu · 2023-08-17 23:22
jannaulina, marcan: https://github.com/AsahiLinux/m1n1/pull/320 required m1n1 changes, can be merged immediately
#asahi-gpu · 2023-08-17 21:29
alyssahttps://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22334
#asahi-gpu · 2023-08-17 21:29
alyssanot compute
#asahi-gpu · 2023-08-17 21:29
alyssathat's an optimization for mesh shaders
#asahi-gpu · 2023-08-17 21:29
i509vcbok so agxv can't use that
#asahi-gpu · 2023-08-17 21:28
alyssawhich again you don't know for indirect dispatch
#asahi-gpu · 2023-08-17 21:27
alyssait just relies on the driver passing in a constant count
#asahi-gpu · 2023-08-17 21:27
alyssai509vcb: becuase that doesn't actually handle it
#asahi-gpu · 2023-08-17 21:17
i509vcb(v3dv and a few other drivers use lower_compute_system_values)
#asahi-gpu · 2023-08-17 21:17
i509vcbis there a reason the gallium driver has it's own bespoke thing there with AGX_SYSVAL_TABLE_GRID?
#asahi-gpu · 2023-08-17 21:16
i509vcbI learned of nir_lower_compute_system_values which seems to handle nir_intrinsic_load_num_workgroups
#asahi-gpu · 2023-08-17 20:54
alyssai509vcb: https://dougallj.github.io/applegpu/docs.html
#asahi-gpu · 2023-08-17 20:54
jannauanyway, I'll prepare a branch for testing on the mac pro (and the m2 max macbook pro)

plus ›