I was thinking of a different problem than prefetch and cache disk solve. What I think could be faster is tweaking strip parameters and effect parameters within a single frame. Right now it seems to load raw images and re-render scene strips for that all the time. Using a variation of rna_Sequence_update for certain properties that don't need to clear raw images in the cache would be relatively simple to add I expect. And as I understand it the raw images for the current frame are always in the cache regardless of the cache settings.
Mon, Mar 25
Works well in my tests, still some comments which are hopefully easy to address
Sun, Mar 24
- change type of BKE_sequencer_cache_put_if_possible to bool
- implement ghash iterator utilizing callback
Thanks, I was busy this week haven't got even to respond to you.
Fri, Mar 22
Mon, Mar 18
Yes, there's some instructions here:
@Richard Antalik (ISS), you actually have commit rights to the documentation repository (username and password and are the same as for logging into this site).
@Olly Funkster (Funkster) this is problem with 3D scenes..
I've been using sequence strips in 2.8, and haven't noticed a difference in framerate between the sequence strip and the scene it came from. I do have a chunky processor though, and some patches applied.
No need to check for zero, this value is guaranteed to be higher.
Sat, Mar 16
I'm also experiencing this, but I'm sure I had the same issue back when I first tried 2.8 (December 2018)
Task is more or less invalid, but docs should be fixed so I guess..
I can not make those changes (I think)
I am dividing by scene->r.*sch - should I check for zero?
Fri, Mar 15
@Richard Antalik (ISS) I guess I can mark this as confirmed then?
Thanks for pointing this out. Adding scene to itself is not supported for quite a while.
Thu, Mar 14
@hudson barkley (snuq) Brecht probably meant that VSE code does not have setup of new viewport in place to allow fast rendering.
...I expect a big part is the new viewport and Eevee in particular doing a lot more setup work than the old viewport.
The assert does not affect release builds end users, only debug builds, so would not consider that high priority.
Wed, Mar 13
I think we have reserved high priority tasks to severe crashes and widely reported issues ATM.
Can confirm this (also can not open file).
Thanks, I wasn't looking hard enough :)
Getting an assert when opening that file (might not be related but I'll report it just in case):
BLI_assert failed: /home/zed/programmering/blender_master/blender/source/blender/windowmanager/intern/wm_window.c:2332, WM_opengl_context_activate(), at 'GPU_framebuffer_active_get() == ((void *)0)'
Can this just use V2D_SCROLL_WIDTH_TEXT? Computing scrollers for every strip drawn seems a little expensive and complicated.
Tue, Mar 12
Added default values for RNA properties
SeqCache is "self contained" - no reference to IMB_moviecache
Fri, Mar 8
I was thinking about this IMB_moviecache issue we have here - in D3934: VSE cache with frame prefetching (WIP) I pointed to a direction I want to go with cache
Thu, Mar 7
renamed free_after_render -> is_temp_cache
removed some prefetch specific code
Added this text to seqcache.c: