00:39 anarsoul: is there any way to propagate varying precision to nir_variable without enabling PIPE_SHADER_CAP_FP16?
01:01 anarsoul: so basically what I want is to be able to specify varyings precision to the hardware, but I don't want to deal with halfs in NIR
01:03 anarsoul: mali4x0 supports only float32 in GP (vertex shader) and float16 in PP (fragment shader), but it's possible to specify precision for GP output and PP input to save some memory bandwidth
01:26 mareko: anarsoul: it should already be propagated
01:27 mareko: anarsoul: except that linked shaders might have mismatching precision qualifiers
02:14 anarsoul: hm
02:15 anarsoul: I think glsl_get_base_type() returns only FLOAT for lima
10:31 MrCooper: panfrost-t760-gles2 jobs are timing out: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/4821935
10:37 daniels: heh, that job is processing as we speak
10:37 daniels: those two machines were down for a bit earlier this morning, and they're just rather backlogged atm
10:45 MrCooper: ah good then
10:51 daniels: looks like it's now caught up
10:51 karolherbst: Plagman: one thing which always causes issues.. some games were just developed against xboxdrv and if you use xpad the mappings are quite wrong
10:51 karolherbst: are there any good ideas on how to solve this mess?
10:51 daniels: just in time to disable it all ...
13:16 orbea: karolherbst: re: xboxdrv/xpad, I gave up and went with a ds3 gamepad which actually works most of the time and doesn't break me with force pushes...
13:17 orbea: occasionally xboddrv is still useful to trick the game into giving me the right layout.
16:58 AndrewR: strange, now OpenCL piglit tests mostly failing again with segfault in libMesaOpenCL.so.1.0. (32-bit build against llvm-10).
17:05 EdB: do you have a bt ?
17:05 EdB: stacktrace
17:05 EdB: what is the failling test ?
17:14 AndrewR: EdB, it appears in dmesg, need to re-learn how to run one piglit test with output to terminal
17:14 EdB: you can run de command detailed in the test
17:15 EdB: if you've created a html summary
17:17 EdB: if you give a test name I can help you
17:18 dcbaker[m]: Piglit print-cmd
17:19 AndrewR: EdB, https://pastebin.com/8rVq92XZ
17:20 AndrewR: EdB, for tests/cl/program/execute/builtin/atomic/atomic_add-global.cl
17:20 AndrewR: EdB, not very useful, need to reinstall mesa with symbols
17:22 EdB: here it run with /home/edb/projects/mesa-opencl/piglit-git/bin/cl-program-tester /home/edb/projects/mesa-opencl/piglit-git/tests/cl/program/execute/builtin/atomic/atomic_add-global.cl for exemple
17:22 EdB: but it run fine one my x64 build
17:28 AndrewR: EdB, yeah, most likely bothed build on my side, just interesting to learn why I have it failing now while it was working better some days ago ...
17:51 AndrewR: EdB, and now it runs better, this test doesn't crash anymore, and whole run seems to have much less crashes ...
18:02 EdB: :)
18:09 EdB: Why my latest dev, I managed to have 4033/4039
18:12 AndrewR: EdB, 4974 ?
18:15 EdB: *with
18:15 EdB: only for CL
18:16 AndrewR: EdB, I was running piglit like this: LP_CL=1 ./piglit run cl cl.results-1.1-3/
18:16 EdB: huhu
18:16 EdB: strange
18:17 AndrewR: EdB, I mean by latest dev you mean https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4974 merged with master? I run into 32-bit compilation problem, I see you updating it ....
18:21 EdB: yes
18:22 EdB: I think I've fix the 32b build as spotted by karolherbst
18:22 AndrewR: EdB, also, I think 6040 incompatible with it, or was last time when I tried (download plain diffs, git apply them one after another)
18:22 AndrewR: EdB, moment, will try reinstall my --strip build first, then patch and recompile
18:24 EdB: indeed 6040 is // to 4974
18:24 EdB: I will make them work depending of wich one is the first to land
18:31 AndrewR: EdB, [710/710] skip: 73, pass: 495, fail: 23, timeout: 1, crash: 118
18:34 AndrewR: EdB, for exampl builtin-float-lgamma-1.0.generated runs into Unsupported intrinsic: vec1 64 ssa_302 = intrinsic load_constant_base_ptr () ()cl-program-tester: ../src/gallium/auxiliary/gallivm/lp_bld_nir.c:1654: void visit_intrinsic(struct lp_build_nir_context *, nir_intrinsic_instr *): Assertion `0' failed.
18:36 AndrewR: EdB, also I only have 2825/3010 tests - do you run with CL version override?
19:27 EdB: what card to you have ?
19:28 EdB: AndrewR: got to go to sleep. But I will be happy to continue latter
19:49 AndrewR: EdB_, I have two nv50-era cards (NV92 + NVA3). Was semi-sleepin / resting for last hour, too ... will re-run piglit test with those overrides. (it takes just 7 min)
19:54 AndrewR: EdB_, [710/710] skip: 57, pass: 499, fail: 32, timeout: 1, crash: 121
19:56 airlied: AndrewR: there iscan mr to fix load constant prr
19:57 airlied: 6569
19:57 AndrewR: airlied, yeah, will try to apply this too
19:58 AndrewR: https://yadi.sk/d/zKrzKiJBkiMpkw?w=1 - results from 1.1 and 1.2 piglit runs ( results.json.bz2 )
20:01 AndrewR: airlied, this one also having 32-bit issues ?
20:02 airlied: no idea, havent tried a 32 bit run
20:02 AndrewR: airlied, https://pastebin.com/pnv9S9T1