#asahi-gpu · 2023-07-05 18:52
alyssaI could easily believe that#asahi-gpu · 2023-07-05 18:50
jannaulet me first test if I need all bits#asahi-gpu · 2023-07-05 18:50
jannauif bit 27 would not break it I'd say bit 20 to 27 are for each cluster#asahi-gpu · 2023-07-05 18:50
alyssayou going to write the patch then?#asahi-gpu · 2023-07-05 18:48
alyssa\o/#asahi-gpu · 2023-07-05 18:47
jannauall good with bits 4, 21-26 in addition#asahi-gpu · 2023-07-05 18:46
alyssaor maybe need even more bits set. IDK#asahi-gpu · 2023-07-05 18:45
alyssa[encoder memoryBarrier: MTLBarrierScopeBuffers afterStages: MTLRenderStageVertex beforeStages: MTLRenderStageVertex]#asahi-gpu · 2023-07-05 18:43
alyssamay be necessary to see what metal does after all :|#asahi-gpu · 2023-07-05 18:43
alyssainteresting#asahi-gpu · 2023-07-05 18:43
jannauyes, those are in addition#asahi-gpu · 2023-07-05 18:42
alyssathat's still setting all the other bits too?#asahi-gpu · 2023-07-05 18:42
alyssauhoh#asahi-gpu · 2023-07-05 18:41
jannaualyssa: becomes a flake with bit (4 and) 24 set#asahi-gpu · 2023-07-05 18:37
alyssaella-0: thx#asahi-gpu · 2023-07-05 18:36
ella-0alyssa: tests i'm using are VK.api.copy_and_blit.core.image_to_image.simple_tests.*, branch is https://gitlab.freedesktop.org/Ella-0/mesa/-/tree/wip/agxv/meta#asahi-gpu · 2023-07-05 18:21
alyssayes#asahi-gpu · 2023-07-05 18:20
jannauadding more bits at the end or in the gaps? I suppose the answer is yes#asahi-gpu · 2023-07-05 18:19
alyssaso sorry about that X_X#asahi-gpu · 2023-07-05 18:19
jannauyes#asahi-gpu · 2023-07-05 18:19
alyssahopefully that makes more sense X_X#asahi-gpu · 2023-07-05 18:18
alyssahttps://gitlab.freedesktop.org/alyssa/mesa/-/blob/agx/es31-v2/src/gallium/drivers/asahi/agx_state.c#L3228-3241#asahi-gpu · 2023-07-05 18:18
alyssaok i was definitely way off#asahi-gpu · 2023-07-05 18:17
alyssaam i wrong the wrong branch#asahi-gpu · 2023-07-05 18:17
alyssauh, no#asahi-gpu · 2023-07-05 18:15
jannaualyssa: no. I think we're looking at different agx_state.c files. is line 2892 'cfg.unknown_30 = frag_tex_count >= 4;'?#asahi-gpu · 2023-07-05 18:10
alyssajannau: any luck?#asahi-gpu · 2023-07-05 17:48
alyssabut in this case since we know what we're looking for bruteforcing might be faster anyway lol#asahi-gpu · 2023-07-05 17:47
alyssaIf someone has the ability to run wrap.dylib + agxdecode against t600x this corresponds to a metal memory barrier, that's where I got those magic bits in the first place#asahi-gpu · 2023-07-05 17:47
alyssatry setting more bits (this will require adding extra fields in asahi/lib/cmdbuf.xml)#asahi-gpu · 2023-07-05 17:46
alyssaSee line 2892 of agx_state.c#asahi-gpu · 2023-07-05 17:46
alyssathe barrier we're using is strong enough to flush a cluster but not the whole system#asahi-gpu · 2023-07-05 17:45
alyssaI'm wondering if this is morally the same issue as the atomics#asahi-gpu · 2023-07-05 17:45
alyssasince we dispatch xfb with the VDM (i.e. as vertex shaders with no fragment output), that's a forward VDM->VDM dependency, that does not require a full flush of the batch, but it does require a memory barrier#asahi-gpu · 2023-07-05 17:44
alyssajannau: The potentially "spicy" part of those tests is that they feed transform feedback results into the inidrect draw#asahi-gpu · 2023-07-05 17:44
jannauerror is in all cases: "(x,y)= (0,0). Color RGBA(0,0,0,1) is different than expected RGBA(0.1,0.2,0.3,1)"#asahi-gpu · 2023-07-05 17:43
alyssadefinitely passes on M1#asahi-gpu · 2023-07-05 17:43
alyssaInteresting#asahi-gpu · 2023-07-05 17:42
jannaufor all configs#asahi-gpu · 2023-07-05 17:41
jannaufor various configs#asahi-gpu · 2023-07-05 17:41
jannauKHR-GLES31.core.draw_indirect.advanced-twoPass-transformFeedback-elements#asahi-gpu · 2023-07-05 17:41
alyssaoof#asahi-gpu · 2023-07-05 17:41
alyssawhat this time?#asahi-gpu · 2023-07-05 17:41
jannauKHR-GLES31.core.draw_indirect.advanced-twoPass-transformFeedback-arrays#asahi-gpu · 2023-07-05 17:41
alyssa:'(#asahi-gpu · 2023-07-05 17:40
jannaufinished here as well after 52 minutes, still failed#asahi-gpu · 2023-07-05 17:40
alyssaM1 seemed to finish#asahi-gpu · 2023-07-05 16:49
alyssadeqp-runner does a bunch of debug stuff for failing tests, idk what the real CTS runner is doing#asahi-gpu · 2023-07-05 16:49
alyssasure#asahi-gpu · 2023-07-05 16:48
jannauiirc 41 minutes vs. 28 min