04:57infinity0: what's the best gfx card these days where nouveau supports reclocking?
04:57infinity0: some kepler or maxwell card i guess?
04:57infinity0: (i need to get a new card)
10:04pmoreau: Submission done, I should have some time again for Nouveau! :-)
11:51imirkin: infinity0: best gfx card is an amd one
11:51imirkin: infinity0: if you insist on nvidia + nouveau, you'll get the most perf with GTX 780 Ti
12:41chithead: not with the kepler based titan? ;)
12:55imirkin_: not sure there's any difference
12:57imirkin_: i don't really see the tital/tesla series as made for home use... the K80 could be interesting since it's theoretically a different chip
12:57imirkin_: but it's a bit on the pricey side, and would probably require changes to nouveau to support
14:42infinity0: imirkin_: any thoughts on 780 ti (kepler) vs 750 ti (maxwell)
14:42infinity0: i will look into amd as well though, is that just supported "out of the box" on linux
14:44karolherbst: infinity0: the 780 ti is much faster
14:44karolherbst: factor 4
14:45infinity0: wow ok good to know, thanks
14:45karolherbst: and maxwell is worse supported than kepler generally
15:05imirkin_: infinity0: yeah, maxwell has rendering issues
15:05imirkin_: and just generically, 780 ti is a lot more powerful than 750 ti, irrespective of driver
15:24infinity0: seems like they're pretty hard to get these days due to being old i guess :(
15:24infinity0: http://gpu.userbenchmark.com/Compare/Nvidia-GTX-780-Ti-vs-AMD-RX-580/2165vs3923 think i'll try this one instead
15:25karolherbst: infinity0: hard meaning they are expensive?
15:26infinity0: expensive but also not in stock in the shops physically near me
15:26pmoreau: karolherbst: I’ll reformat the translator code over the weekend, to make it easier to send patches afterwards.
15:26karolherbst: okay
15:26karolherbst: pmoreau: I've done more cmake/travis hacking: https://travis-ci.org/karolherbst/SPIRV-LLVM-Translator
15:27karolherbst: ...
15:27imirkin_: infinity0: well, any kepler should be fine
15:27imirkin_: 770's and 760's should be fairly powerful too
15:27imirkin_: the nice thing about the 780 is that it has more ISA registers
15:27imirkin_: which puts less pressure on nouveau's register allocator
15:28pmoreau: karolherbst: I saw your comments on the PR. You haven’t pushed the reworked version to your branch yet, have you?
15:28karolherbst: pmoreau: to travis-test
15:28imirkin_: infinity0: but if you want to do actual gaming, definitely go with AMD
15:28pmoreau: Ah, okay :-)
15:28karolherbst: I hope this is the final build and I can push it to test
15:28pmoreau: Quite a few tests running :-D
15:28imirkin_: their driver is well-supported, they have vulkan, and the hardware is made by a company that supports open-source graphics.
15:32karolherbst: pmoreau: yeah well, only 5 concurrent jobs though :(
15:32pmoreau: Thankfully it shouldn’t take too long to compile.
15:32infinity0: imirkin_: thanks for all the info! yeah i am considering that
15:33karolherbst: pmoreau: ~ 15 mins for the builds inside a llvm tree
15:33pmoreau: I’m surprised at how long SPIRV-Tools takes to compile, mainly due to all the tests.
15:34karolherbst: pmoreau: I think the allow_failures: - compiler: clang doesn't work :(
15:34karolherbst: so it seems like I really have to remove it, because it seems to be a bug inside clang itself
15:34pmoreau: That’s unfortunate
15:35karolherbst: and for whatever reason they are super slow upgrading the env
15:35karolherbst: I mean, trusty is still the "newest" and 0 plans to change that
15:35karolherbst: but caring enough to add like every single xcode version...
15:37pmoreau: :-/
15:39karolherbst: pmoreau: ohh if you want, you can also fix those compile warnings in the clang builds :D
15:39karolherbst: I think there are in fact 2
15:39karolherbst: 90% caused by std::move() used in a silly place
15:40karolherbst: and the second issue is missing override
15:40pmoreau: I’ll just run clang-tidy on it, and that should fix the moves and overrides.
15:40karolherbst: nice
15:41pmoreau: That’s planned to be part of the reformat series; I already did that for the spirv-6.0 branch on the old repo.
15:43karolherbst: ahh, nice
15:44karolherbst: pmoreau: I am currently thinking about enabling ccache... but mhh
15:44pmoreau: I have no experience with ccache, so I won’t be able to help.
15:45karolherbst: I mean you just use it as a gcc/g++ replacement
15:45karolherbst: but
15:45karolherbst: it can generate false positives
15:45karolherbst: and use garbage binaries
15:45pmoreau: Sounds like a source of issues that will be painful to debug.
15:46karolherbst: nah
15:46karolherbst: the compile usually fails
15:46karolherbst: and you see the issu
15:46karolherbst: e
15:46karolherbst: but still
15:46karolherbst: then you have to clear the cache manually and so on
15:46karolherbst: it is just annoying
15:46karolherbst: and just saves a bit of time
15:49pmoreau:should have cloned LLVM back home: only getting 300KiB/s at the hotel, compared to 40-50MiB/s at home.
15:49karolherbst: :D
15:51karolherbst: pmoreau: uhhh...
15:51karolherbst: pmoreau: the test passes with clang if build in tree...
15:51pmoreau: Weird! :o
15:52pmoreau: karolherbst: FYI, https://github.com/pierremoreau/SPIRV-LLVM-Translator/tree/reformat
15:53karolherbst: nice
15:54karolherbst: pmoreau: duh, on mac os x: "/Users/travis/build/karolherbst/SPIRV-LLVM-Translator/llvm/tools/llvm-spirv/lib/SPIRV/OCL20ToSPIRV.cpp:1143:17: error: no viable conversion from 'const char [16]' to 'llvm::StringLiteral'"
15:54pmoreau: (There is only the switch from DOS line endings to UNIX ones.
15:55pmoreau: Might be using an old'n'buggy compiler?
15:56karolherbst: ohh, maybe
15:56karolherbst: this was with xcode7.3
15:57karolherbst: well, I will finish the new build now and head home :)
15:59pmoreau: :-)
16:44pmoreau: Phew, clang-format: done. The annoying part was adding some additional fixes and checking what it did.
16:44pmoreau: Now onto clang-tidy, but I need to compile the thing first.
19:14karolherbst: pmoreau: https://travis-ci.org/karolherbst/SPIRV-LLVM-Translator :)
19:15karolherbst: pmoreau: sadly it didn't build with xcode-8 either, so only enabling xcode-9.3 for now
21:15Lyude: sweet, found a bonus set of slcg registers that I missed on kepler2 somehow
23:11Lyude: karolherbst: poke; any idea why a gm107 might not have support for showing the wattage sensors?
23:12Lyude: got maxwell1 cg ready to go, but need to actually see the wattage so I can figure out if it's working as intended or not
23:14Lyude: oh, the problem seems to probably be that it's not written
23:14Lyude: i'll go to making that work, someone stop me if for some dumb reason maxwell1 somehow does not have power sensors
23:20Lyude: mupuf: poke; you around?