freenode :: #aomedia

17 Mar 2017
16:42peloverdembx: I must say the analyzer has been quite useful to me lately
16:42mbxpeloverde glad to hear it
16:42mbxhow do you use it?
16:43peloverdeI've been using it to debug rate issues while converting more symbols to cdf
16:43mbxpeloverde i'm happy to add features, but I don't know exactly what people need
16:43peloverdeWhen I wind up down I can quickly identify where the extra bits are going
16:44mbxare you using the symbols table?
16:45peloverdeand using partition and bits overlays
16:46mbxpeloverde I see
16:46mbxpeloverde is there anything I can change to make it more useful
16:47peloverderight now it's doing everything i need from it so I don't think so
16:48mbxit may be useful to add better accounting strings
16:49mbxto get more accurate breakdowns of where bit are used
16:49mbxpeloverde are you toggling back and forth between one or more videos while looking at the symbols table?
16:49peloverdeframe symbols wasn't very discovrable
16:50peloverdeit's usually past the bottom of my browser window and it wasn't clear that it was a scrollable pane
16:50mbxyah, we fixed that
16:51mbxbut we haven't pushed it yet
16:51mbxthere are now several tabs you can select, instead of having to scroll down
16:53mbxpeloverde i'm glad you brought it up, i just looked at the fix now on my smaller screen and is even more broken :)
17:00peloverdembx: Is it possible to lock the cursor on a block?
17:06mbxnot yet, i could change it so that you'd have to click to set the cursor
17:09mbxpeloverde the selected block would change when you step through frames
18:18mbxpeloverde the mouse will lock now, when you click.
21:16TD-Linuxtell me what you think about this
21:17TD-Linuxa user selected --best without knowing the consequences
21:44TD-Linuxactually I kind of like remapping this to cpu-used -1
21:47kagami_IMO better to remove, otherwise someone will still advice these sort of "extra quality" options
21:48kagami_or rename to something like --bench which is obviously not for real use
22:16TD-Linuxyeah I think I'll change it to straight up delete
22:26smarteror have a really obnoxious option
22:27smarterI remember glxgears --iacknowledgethatthisisnotabenchmark
22:40peloverdeIs there an awcy option for a partial report?
23:10unlordpeloverde: what are you trying to do?
23:12peloverdeunlord: I've debugging a cdf conversion, I started a run this morning with a version that fixed a bunch of problem samples. Right now the run is at 14X/150. Partial results could help my identify problems earlier. This is my third pass at it and it takes >8 hours per run so it's tough to iterate quickly.
23:14unlordpeloverde: hmm, you could always run the samples locally on just some of the smaller files
23:14unlordpeloverde: the rd_collect scripts in the daala repo, the precursor to AWCY, work with AV1
23:15unlorddepending on your machine, it might be faster than waiting 8h per run
23:19TD-Linuxpeloverde, you can specify only one quantizer to run, there's no partial report feature yet but it's on my todo
23:19TD-Linuxe.g. showing results before it completes
23:19smarterpeloverde: you can also use --limit=5 or something like that
23:19unlordTD-Linux: would one quantizer be enough information to get any RD data?
23:20unlordsmarter: yah, that is probably better actually
23:20unlordsmarter: this is how codeview was completing PVQ runs
23:20smarterunlord: I had that experience too :)
23:29mbxunlord i don't think you get any meaningful RD data for 1 quantizer
23:29derfYou need at least 4.
23:31unlordderf: that was what I thought :)
23:46peloverdeEven only showing results for clips with all the points done would be super useful
23:46peloverderight now on my run most clips have all points done
23:49TD-Linuxokay yeah I can test this on beta
18 Mar 2017
No messages
Last message: 11 days and 4 hours ago