freenode :: #aomedia

13 May 2017
00:00rillianwwcv: oh, you mean to get the appropriate list of source files.
00:00rillianI guess that would work. I was just trying to find what would be dead code.
00:00wwcvyes
00:00wwcvgit grep HAVE_NEON_ASM
00:04rillianwwcv: hmm, I guess I should remove the armv6 configure targets from aom too?
00:04wwcvwe removed them from libvpx a long time ago. there is no armv6 code that affects vp9, and therefore av1
00:04wwcv(long time ago ~= last summer)
00:06rillianok.
00:06rillianthanks for the libvpx import script, btw. Was very helpful for libaom.
00:07wwcvi think frank wrote the initial version for chromium, but yeah it's been handy for all the places libvpx builds
00:20rillianwwcv: who do we need to wait on to remove the armv6 jenkins jobs?
00:20wwcvjzern@, but if you just fix the rtcd that patch should pass
00:20wwcvi thought you removed armv6 from the configure file
00:21rillianwwcv: I thought I did. Do you mean put back the armv6 support in rtcd.pl?
00:21wwcvaom_dsp/aom_dsp_rtcd_defs.pl:specialize qw/aom_sad16x16 media neon msa sse2/;
00:22wwcvyou need to remove 'media'
00:23rillianah, thanks.
00:27wwcvgit grep -w media shows some other things you can probably remove too
00:27rillianyeah, found those
00:28wwcvas long as you leave "all_platforms="${all_platforms} armv6-linux-gcc"" in configure it should pass jenkins
00:29rillianwhat does the `string(STRIP AOM_RTCD_FLAGS)` line do in armv7-ios.cmake?
00:29rilliandoes that just remove the leading whitespace if AOM_RTCD_FLAGS was previously empty?
00:30wwcv\_()_/
00:30rillianlol
00:30wwcvI haven't been paying close attention to libaom
00:30rillianwell, this is what ci tests are for
00:50* rillian is confused how removing armv6 support resulted in a missing uint16 typedef
00:50rillianwwcv: I guess I should fix that in a third commit?
00:50wwcvit's failing for other changes as well, not sure how it got the broken state, maybe someone forced it in?
00:51rillianhmm uint16 is all over the palce
00:52rillianfrom libyuv/row.h
00:52wwcvfile probalby needs the aom equiv of vpx/vpx_integer.h
00:52wwcvoh libyuv? weird, but i didn't look closely
00:53rillianI guess the question is whether this particular instance should be uint16 or uint16_t
00:53rilliancolor_cache is uint16_t*, so I think the latet.
00:53rillian*later.
00:54rillianwwcv: what else is broken?
00:54wwcvrillian: i dunno? just saw something about types and assumed it needed the header file
00:54rillianah ok.
00:55wwcvthe bigger question is why did it start failing now
00:55rillianit's also #ifdef HIGH_BITDEPTH
00:55wwcvoh, maybe since that became default
00:59rillianwell, I pushed a type change. We'll see how that works.
00:59rillianGotta go, coffee shop is closing.
00:59rillianwwcv: thanks for all the help!
00:59wwcvnp
01:02derfwwcv: Unfortunatey, the test infrastructure is kind of useless right now, since more than half the time the Jenkins runs time out before completing.
01:02derfI assume people are getting fed up and clicking Verify themselves.
01:03wwcvwell that just makes it worse, so maybe the shouldn't?
01:10rillianinadequite provisioning and bypassing tests are both symptoms of the same issue.
01:21rillianwwcv: I don't see my fixup. Does it need to be a new issue?
01:22wwcvwhere did you expect to see it?
01:23rillianin gerrit, attached to the original change
01:24rilliandoes gerrit not support more than one commit per issue?
01:24wwcvonly one per change-id
01:24wwcvbut if they are in sequence in your git history, they should show up that way in gerrit
01:24wwcvhttps://aomedia-review.googlesource.com/c/11974/
01:25wwcvit's on top of the armv6 changes
01:25wwcvsee the relation change on the right side
01:25wwcvyou can click 'rebase' near the top right and put it on master
01:25rillianaha! so there's an issue for each commit.
01:25wwcvyes
01:25wwcvcommit --amend is how gerrit expects you to work
01:25wwcvor rebase/squash
01:26rillianok, sorry.
01:26rillianI know it says that in the contributor guide
01:26rillianbut these seemed like separate issues...so.
01:27wwcvif you do them on the same branch then they depend on each other
01:27wwcvso the armv6 changes will still fail, because the type change that fixes the build comes after them
01:28rillianok. sadly it looks like there are in fact more of of the uint16 issues
01:28wwcvopen an aomedia bug, assign it to whomever touch them last?
01:29rillianok. I'm almost out of time.
01:29rillianunless eurovision is boring tomorrow :)
01:31rillianit would help if I could reproduce the failure...
01:31rillianoh well, enough for now
14 May 2017
No messages
   
Last message: 45 days and 18 hours ago