Subsong performance issues in v2.x

2023-09-18 22:10:54

Subsong performance issues in v2.x

Hi, I've noticed an issue with subsongs in v2.x (up to latest preview) vs v1.x, which affects performance in foo_input_vgmstream for files with subsongs (esp. some bank formats, which may contain several thousands).

When adding a new file with subsongs, these methods get called:
- get_subsong_count(): returns number of subsongs
- per subsong:
  - get_subsong(..): matches current index vs internal subsong number
  - get_info(..): queries info for current subsong
Then all subsongs get added to playlist with queried info. This works the same in 1.x and 2.x.

When playing a single subsong, this happens in v1.x:
- get_subsong_count()
- per subsong:
  - get_subsong(..): a bit odd it's re-called again per subsong rather than current index, but no matter
- decode_initialize(..): for current index (ex. subsong 3)
- get_info(..): for current index, to update current info I assume

However this happens in v2.x:
- get_subsong_count()
- per subsong:
  - get_subsong(..)
  - get_info() !!!
- decode_initialize(..)
- get_info(..)
So if any subsong is played, foobar re-queries all info for *all subsongs*, every time a subsong is played.

vgmstream only has info for current subsong (assumes players will only ask for 1 at a time) so it's slow to re-query all subsongs on every play. But even if we manually cache'd that, seems like a bug as I don't think there is need to re-ask about other subsongs when only trying to play one.

Could be that vgmstream isn't doing something that foobar expects too, code here is somebody has suggestions:
https://github.com/vgmstream/vgmstream/blob/master/fb2k/foo_vgmstream.cpp

Re: Subsong performance issues in v2.x

Reply #1 – 2023-09-19 11:12:33

Indeed, asks decoder for all infos even if they're not actually needed. This was NOT intended.
Fixed for the next update, thanks for reporting.