13:18 <Guest24> hello, anybody who knows AC4 decoder?

13:19 <Guest24> I could see AC4 decode could decode 9.1.6 but with wrong channel mapping only merge to low 8 channels.

13:20 <Guest24> I think I could help on testing and might correct the bug.

14:17 <Lynne> we don't have an ac-4 decoder for now

14:25 <JEEB> they probably were poking at some fork or applied the patches posted ages ago or so?

14:25 <JEEB> unless someone made a wrapper for the proprietary decoder lib

15:18 <cone-255> ffmpeg James Almer master:5b87869c09ce: avformat/mov: fix track handling when mixing IAMF and video tracks

15:18 <cone-255> ffmpeg James Almer master:eb3cc508d85e: fate/mov: add an IAMF+video muxing test

15:23 <cone-255> ffmpeg James Almer release/7.0:714635937a92: avformat/mov: fix track handling when mixing IAMF and video tracks

16:01 <sdc> just for clarification, I can use SWAP to rename registers essentially? For context I'm using m0, ..., m7 to store/compute 8 rows of pixels. For a slightly different version I need to transpose the 8 rows first (and then back) but it seems like TRANSPOSE8x8W doesn't keep the ordering for m0, ..., m7 so I want to use SWAP to rename the registers

16:03 <Lynne> SWAP doesn't issue any commands, it just swaps the register names

16:04 <Lynne> entirely in the preprocessor

16:06 <Lynne> IndecisiveTurtle: ping

16:12 <elenril> word on the street is, mov does not issue any moves, it just renames registers in microcode

16:13 <IndecisiveTurtle> Lynne: Heya. Logged in to give a status update but ended up chasing a small bug before that xD. Well to say, I worked out a ton of bugs this week but now it can encode a valid vc2 that decodes properly for most videos! Pretty happy with it so far, need to resolve some edge cases I saw in some tests

16:13 <Lynne> awesome!

16:13 <IndecisiveTurtle> Fixed haar with mutliple planes, fixed encode shader to work correctly and integrated slice calculation shader in the pipeline with a simple 2 pass prefix sum to find the correct location

16:14 <Lynne> pmozil also got his decoder working

16:14 <IndecisiveTurtle> Oh nice

16:14 <Lynne> wow, so slice calculation for the max slice size is working?

16:15 <IndecisiveTurtle> Yes for the most part. The initial video I had matches exactly with cpu. Now I'm testing another video where some slices are off by a bit and causes blocky artifacts

16:16 <IndecisiveTurtle> https://imgur.com/a/OG7Vr7d

16:16 <IndecisiveTurtle> Around the eyes and moving parts it has some garbage

16:18 <IndecisiveTurtle> Can check my github repo if you want, pushed the code yesterday

16:21 <IndecisiveTurtle> I noticed it might be due to quant indices. I compared shader with cpu and shader figures out the optimal size on first pass and remains in quant index 0 while cpu takes 5 iterations. Its probably a bug somewhere

16:23 <IndecisiveTurtle> The initial video I tested btw was this one https://imgur.com/a/Glj0zsZ

16:23 <IndecisiveTurtle> This is generated from the vulkan encoder and looks correct to me

16:24 <IndecisiveTurtle> It works on all wavelet levels 1-4. The fourth level tho while it has good picture some colors are inverted which I also need to look into

16:53 <IndecisiveTurtle> Lynne: My client showed something like a massive channel purge. Did I miss any messages?

17:40 <elenril> no

17:52 <Lynne> IndecisiveTurtle: if the second video you linked is the one on 4 levels, then that looks like a bitstream desync

17:53 <Lynne> that's amazing progress, though

17:57 <Lynne> keep in mind for 4 levels you need a large slice size

17:59 <Lynne> as you run out of subdivisions

18:00 <IndecisiveTurtle> Lynne: The second video was with 3 levels

18:00 <IndecisiveTurtle> With 4 levels it looks like this https://imgur.com/a/eehzoNd

18:02 <IndecisiveTurtle> From what I looked its higly likely to be slice size issue yeah. With 4 levels it reaches a point where it has to align 720p a bit to make it divide integer fashion

23:22 <cone-616> ffmpeg Michael Niedermayer release/5.1:8c689f4d0cfa: avcodec/snow: Fix off by 1 error in run_buffer

23:22 <cone-616> ffmpeg Michael Niedermayer release/5.1:999c1619f9dd: avcodec/utils: apply the same alignment to YUV410 as we do to YUV420 for snow

23:22 <cone-616> ffmpeg Michael Niedermayer release/5.1:6338d19eb349: Changelog: update

23:33 <cone-616> ffmpeg James Almer n5.1.6:HEAD: fate/mov: add an IAMF+video muxing test