- Kaunas, Lithuania
- https://aras-p.info/
- Joined on
2022-01-20
Block a user
a88dc0930b
VSE: Multi-threaded video proxy downscaling
5b5576993e
Transform: skip hidden edges when checking an vertices edge is visible
ed2d131887
Transform: use deformed geometry when checking edge visibility
a13513ab25
Fix #125646: Resolve edge-slide performance regression when snapping
2ab800b685
Cleanup: pass const arguments to BMBVHTree related functions
Aras Pranckevicius
created branch vse_proxy_scale_threaded in aras_p/blender
2024-09-24 09:33:59 +02:00
VSE proxies build slower than it could be
VSE proxies build slower than it could be
Gonna reopen this, since I'm about to fix lack of threading in proxy image scaling :)
VSE proxies build slower than it could be
thought, that swscale is threaded, but I may be mistaken
swscale is not threaded by default (yeah, surprisingly). I had to jump through some hoops to make it threaded in 4.1/4.2, but I did…
VSE proxies build slower than it could be
This may be a bug, but possibly in ffmpeg or x264 library. We are using ffmpeg backend and we are instructing it to use all available CPU cores. But obviously it does not do that
My wild…
VSE proxies build slower than it could be
The CPU could be used more and GPU too. Several cores for several strips at once. And commit work done in the middle and don't freeze the whole blender during this process
Sure, a bunch of…
VSE proxies build slower than it could be
Yeah I'm not sure if this is a "bug":
- Many parts of VSE are less efficient than they could be, yes. Building proxies is one of these parts. This is known about, no one fixed/addressed it…
Cleanup: rename MAXSEQ -> SEQ_MAX_CHANNELS and make it C++ constexpr
Aras Pranckevicius
deleted branch vse_timeline_curve_lookup_opt from aras_p/blender
2024-09-23 15:43:02 +02:00
VSE: Faster strip fcurve lookup for drawing the timeline
Cleanup: rename MAXSEQ -> SEQ_MAX_CHANNELS and make it C++ constexpr