ChanServ changed the topic of #zink to: official development channel for the mesa3d zink driver || https://docs.mesa3d.org/drivers/zink.html
hch12907 has joined #zink
<kurufu> I had to manually specify b_ndebug=false to get a debug build with asserts enabled.
hch12907 has quit [Ping timeout: 480 seconds]
hch12907 has joined #zink
hch12907 has quit [Ping timeout: 480 seconds]
hch12907 has joined #zink
hch12907 has quit [Remote host closed the connection]
hch12907 has joined #zink
<hch12907> zmike: it's basically done, just that I haven't tested it yet
<hch12907> my haswell machine is gone unfortunately
<hch12907> rebased and dropped the draft for !16265, still untested though
Akari has quit [Ping timeout: 480 seconds]
Akari has joined #zink
hch12907 has quit [Ping timeout: 480 seconds]
hch12907 has joined #zink
otokawa_ has joined #zink
hch12907 has quit [Ping timeout: 480 seconds]
<zmike> oh perfect, I'll get on that then
<zmike> hch12907[m]: fwiw you should be able to test this sort of thing on lavapipe too since it's just about enabling extensions
hch12907 has joined #zink
otokawa_ has quit []
<hch12907> grrr, building mesa w/ llvm=true always fails on my windows machine for some reason.
<hch12907> will investigate on it more tomorrow, hopefully I will get lavapipe working then
<xperia64> so it's not a version/feature set thing that's causing Dolphin to fail on AMD Windows, Yuzu can request a 4.6 compat context just fine
<xperia64> although Yuzu has a similar funny reflected triangle issue to minecraft: https://i.imgur.com/zfPlVd9.png
<xperia64> (also learned llvmpipe runs Citra faster than AMD's drivers on my machine, with hardware shader at least)
<hch12907> a software rasterizer runs faster than a GPU?
<hch12907> wow.
<hch12907> I know OpenGL on AMD Windows is horrible, but that is spectacularly bad.
<xperia64> AMD's drivers are definitely hitting some single-threaded CPU path for whatever Citra's "hardware shaders" are doing
<xperia64> I think hardwarer shader perf has severely degreaded on AMD at some point, I don't recall them being quite this awful. Qualcomm's drivers are definitely faster, and I almost want to say Mali might be too now.
<hch12907> tried running wglgears on AMD Windows with zink, ended up with a segfault in XXH32_finalize
<hch12907> works fine with softpipe (I still haven't figured out how to enable llvm)
* hch12907 can't ever get a proper build of C projects on windows
<xperia64> I have wglgears running with zink
<xperia64> MSVC mesa+zink, mingw static wglgears
<xperia64> at least I think it is, very hard to tell what wglgears is using
hch12907 has quit [Remote host closed the connection]
hch12907 has joined #zink
<hch12907> yeah, it's probably bad setup from my part (have to figure it out though...)
<xperia64> I def had trouble with it before, it's not just you
<hch12907> rebuilding everything from scratch right now (except LLVM) - the fact that mesa/demos switched from cmake to meson makes me smile.
<hch12907> "C:/msys64/mingw64/include/c++/11.3.0/cstdlib:75:15: fatal error: stdlib.h: No such file or directory"
<hch12907> that was from compiling gallivm, gah why won't llvm=enabled builds just work
<hch12907> hurray, got it building
<hch12907> somehow the ninja script wants to link "advapi32.lib" "ole32.lib" "psapi.lib" "pthread.lib" "shell32.lib" "uuid.lib", which the linker couldnt find
<hch12907> had to hand modify build.ninja and delete those
<hch12907> anyway, it doesn't segfault anymore, but I got this when running wglgears: "warning, this is cpu-based conditional rendering"
<hch12907> and no gears whatsoever
<hch12907> don't know where to start debugging it, either.
<hch12907> zmike: I think have_EXT_memory_budget etc. shouldn't be affected by the MR...?
<zmike> hch12907: dunno
<zmike> just posting what I saw
<zmike> I can retest
<hch12907> welp it's peculiar as hell that's for sure
<hch12907> but I have to sleep now so I will look at this further tomorrow
hch12907 has quit [Ping timeout: 480 seconds]
<xperia64> I mean wglgears works for me
<xperia64> verified it's using zink via renderdoc. cycled through zink, d3d12, and llvmpipe and all behaves as expected
<xperia64> oh, I made the wglgears window larger, and all the funny triangle issues on AMD windows zink are probably just provoking-vertex