01:08 imirkin: Lyude: give writing a new test a shot for ARB_post_depth_coverage as i described above. if you run into trouble or feel like you can't, let me know, and i can try to hack it up.
07:02 leberus: RSpliet: sorry, since I'm not subcribed to the list (I dont get the confirmation e-mail), I can't see your e-mail. What's about?
08:57 leberus: karolherbst: i'm writing another patchset with all changes you pointed out and adding the "coverletter" explaining everzthing
08:57 leberus: i expected to be ready tomorrow (im quite busy today)
09:02 karolherbst: leberus: no worries
10:55 dboyan: um, it seems the blob is doing something really clever when generating code for atomic operations.
10:55 dboyan: https://hastebin.com/abucowowes.pl
10:57 RSpliet: leberus: https://lists.freedesktop.org/archives/nouveau/2017-April/027715.html
10:57 dboyan: The blob uses a fast path with SHFL.BFLY when it discovers that all the threads are active
11:02 RSpliet: leberus: but tl;dr: git format-patch automagically turns git commits into patch files for you, git send-email lets you send them to the mailinglist without worrying about the formatting. --compose lets you add a cover letter (interactive, like git commit)
11:02 karolherbst: dboyan: I think imirkin had the theory, that the blob does it for expensive conditional code as well
11:04 karolherbst: huh
11:04 karolherbst: "set $p2 0x1 not 0x1" ...
11:05 dboyan: that means $p2 = false
11:05 karolherbst: yeah, I was just surpirsed why this wasn't opted away
11:06 dboyan: well dunno
11:08 karolherbst: dboyan: is this on kepler2?
11:08 dboyan: yeah, gk208
11:09 dboyan: it was actually a shader in bug 99532. I was impressed because the blob uses vote instruction while there was no "vote" or "ballot" in my shader. After I finished my ARB_shader_ballot series, I can finally understand
12:38 imirkin: dboyan: blob does a few smart things... (a) it will use vote to skip over predicated blocks if all invocations have same value for that predicate and (b) it will use the current lanes to coalesce atomic writes, so e.g. atomicAdd(foo, 1) becomes atomicAdd(foo, popCount(lanes)) in the first active lane.
12:41 dboyan: well, the blob is even more sophisticated than I thought
18:33 eau4x6: hello folks. is hardware accelerated video decoding (https://nouveau.freedesktop.org/wiki/VideoAcceleration/ ) supposed to work in Optimus (https://nouveau.freedesktop.org/wiki/Optimus/) setups also?
18:34 eau4x6: DRI_PRIME=1 vdpauinfo gives me "Failed to open VDPAU backend libvdpau_i965.so"
18:34 eau4x6: (it tries to load non-existing intel driver)
18:49 eau4x6: holy crap, it works
18:49 eau4x6: VDPAU_DRIVER=nouveau DRI_PRIME=1 vdpauinfo
19:01 leberus: RSpliet: thanks a lot of for the comments. The truth is that i didn't use it yet, but I'll look into it. Looks much easier than what I did xD
21:40 nyef`: Hrm. There's some... interesting bits in the tegra driver with respect to audio infoframes.