#asahi-gpu · 2024-08-10 21:34
karolherbstfixed the CL_HALF_FLOAT stuff :3#asahi-gpu · 2024-08-09 13:43
alyssayeehaw#asahi-gpu · 2024-08-09 07:44
karolherbst*upstream#asahi-gpu · 2024-08-09 07:44
karolherbstdoing another CTS run, because I've also rebased my CTS branch and I've already had to fix a few regressions upstreams#asahi-gpu · 2024-08-09 07:44
karolherbstseems to work, thanks :)#asahi-gpu · 2024-08-09 02:19
alyssacopies are... trickier#asahi-gpu · 2024-08-09 02:19
alyssakarolherbst: pushed a patch to agx/blockblit fixing idiv#asahi-gpu · 2024-08-08 16:07
karolherbsthave a good rest#asahi-gpu · 2024-08-08 16:06
alyssayeah i need a break too many things happening right now#asahi-gpu · 2024-08-08 16:02
karolherbstfair#asahi-gpu · 2024-08-08 16:02
karolherbst:D#asahi-gpu · 2024-08-08 16:02
alyssai was busy reading french wikipedia's page on newton's method :P#asahi-gpu · 2024-08-08 16:02
alyssaor I can just give a patch in a bit#asahi-gpu · 2024-08-08 16:01
karolherbst(the first is a full copy)#asahi-gpu · 2024-08-08 16:01
karolherbstthe second hits the bug#asahi-gpu · 2024-08-08 16:01
karolherbstyou need to skip the first blit#asahi-gpu · 2024-08-08 16:01
karolherbstimages/clCopyImage/test_cl_copy_images 1D CL_HALF_FLOAT#asahi-gpu · 2024-08-08 16:01
alyssaI think I know what's happening but I'd like to confirm#asahi-gpu · 2024-08-08 16:01
alyssawhich test case is that?#asahi-gpu · 2024-08-08 16:00
alyssaok#asahi-gpu · 2024-08-08 15:59
karolherbstalyssa: mhhh.. okay, maybe it's the blit.. but what seems to happen is, that the image gets initialized with 0xffff over it's entire size, then an API copy happens at offset srcOffset = 79, destOffset = 121 and then the destination becomes 0x7e00 starting at offset 96.... but 121+ is as expected copied from the source#asahi-gpu · 2024-08-08 15:49
alyssayeehaw#asahi-gpu · 2024-08-08 15:48
karolherbstI'll just be busy figuring out what's wrong with the blit, because it's not the actual copy being all wrong, so it's a bit confusing...#asahi-gpu · 2024-08-08 15:47
alyssai'll get to it soon#asahi-gpu · 2024-08-08 15:47
karolherbstyeah, fair#asahi-gpu · 2024-08-08 15:47
alyssathough not right now I have a big todo list this week#asahi-gpu · 2024-08-08 15:47
alyssai can work with this :+1:#asahi-gpu · 2024-08-08 15:47
alyssawas slightly more complicated than that but yeah, added the test and indeed we're failing for large inputs#asahi-gpu · 2024-08-08 15:46
karolherbstmaybe enable it and see if it works?#asahi-gpu · 2024-08-08 15:45
karolherbstonly disabled test besides tgamma#asahi-gpu · 2024-08-08 15:44
alyssareal#asahi-gpu · 2024-08-08 15:44
karolherbst🙃#asahi-gpu · 2024-08-08 15:44
karolherbst //ENTRY(reciprocal, 1.0f, 1.0f, FTZ_OFF, unaryF),#asahi-gpu · 2024-08-08 15:44
karolherbstreciprocal#asahi-gpu · 2024-08-08 15:44
karolherbstohhh#asahi-gpu · 2024-08-08 15:43
karolherbstis that even used#asahi-gpu · 2024-08-08 15:43
karolherbstwait..#asahi-gpu · 2024-08-08 15:43
karolherbstreference_recip#asahi-gpu · 2024-08-08 15:43
alyssaI would like math_bruteforce to test the full 2^32 input space because otherwise I have no confidence in writing the lowering correctly#asahi-gpu · 2024-08-08 15:42
alyssa:=/#asahi-gpu · 2024-08-08 15:42
karolherbstyeah.. I don't think it's tested besides through divide#asahi-gpu · 2024-08-08 15:42
alyssareciprocal is suspiciously absent from math_bruteforce#asahi-gpu · 2024-08-08 15:41
alyssaah. but does the CL CTS test it.#asahi-gpu · 2024-08-08 15:41
karolherbstthe spirv env just defines precision for "1.0 / x" explicitly #asahi-gpu · 2024-08-08 15:40
karolherbstehh clc#asahi-gpu · 2024-08-08 15:40
karolherbstlibclc doesn't have a frcp opcode #asahi-gpu · 2024-08-08 15:40
alyssaI guess libclc doesn't do divide?#asahi-gpu · 2024-08-08 15:38
alyssaalthough i dont actually see where we call libclc frcp#asahi-gpu · 2024-08-08 15:37
alyssaso that seems.. bad#asahi-gpu · 2024-08-08 15:36
alyssabut you need exact frcp! in the backend *anyway* for correct idiv