Raul Fernandez Hernandez farsthary
  • Joined on 2008-06-18
Raul Fernandez Hernandez commented on issue blender/blender#120593 2024-04-17 19:50:52 +02:00
Blender 4.2 Dyntopo sculpt fist stroke performance is very laggy

do I upload the file to have a look at it?

Yes please, share the .blend file to repro the same steps as the video

Raul Fernandez Hernandez pushed to main at farsthary/blender 2024-04-17 19:38:48 +02:00
757da9dbc1 Cleanup: Remove redundant compositor arguments
2cf8b5c4e1 Compositor: Improve interactivity for GPU compositing
ad48827244 Fix duplicate gpencil_sculpt_tool property in PAINT_OT_brush_select
a7d1bd9735 cppcheck: use source relative paths in the summary
e4529592d6 Cleanup: sort cmake file-lists
Compare 72 commits »
Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-16 12:06:10 +02:00
Sculpt: Improve Expand performance

Yup, will give @Sergey 's code a shot, if is fully thread safe then it will be very worth it to save this PR ;)

Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-16 11:38:58 +02:00
Sculpt: Improve Expand performance

@mod_moder You're right, if even now is getting slower than the single threaded version, adding more thread safety will increase the overhead. When I closed the PR on Saturday was mainly because…

Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-16 11:14:19 +02:00
Sculpt: Improve Expand performance

Thanks @Sergey I just pushed my changes before seeing your review and code. What I found in the my thread safe implementation is that is actually slower than the single threaded one :( adding…

Raul Fernandez Hernandez pushed to improve-expand-perf at farsthary/blender 2024-04-16 11:06:43 +02:00
63364a9fef thread safety
Raul Fernandez Hernandez pushed to main at farsthary/.profile 2024-04-15 16:56:57 +02:00
f6f0cbd13c remove macos trash
Raul Fernandez Hernandez pushed to main at farsthary/.profile 2024-04-15 16:56:18 +02:00
42bef32070 update
Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-15 10:31:45 +02:00
Sculpt: Improve Expand performance

I invoke the Git Gurus to help me get rid of this XD

Raul Fernandez Hernandez reopened pull request blender/blender#120125 2024-04-14 03:45:47 +02:00
Sculpt: Improve Expand performance
Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-14 03:45:45 +02:00
Sculpt: Improve Expand performance

First of all: Data race is the bug that no one wouldn't love to fix. And more important is how data race is going to be confirmed by bug triggers. Do you planned to be cause of some bug…

Raul Fernandez Hernandez pushed to main at farsthary/blender 2024-04-14 01:41:24 +02:00
dd672c38b1 Fix #117515: Geometry Nodes: Count of unique face neighbors
48baa80a05 Fix #120287: Annotations GPv2 data must not be converted.
fdf5a05bde Fix: EEVEE-Next: Broken shader compilation because of displacement
f371624b2e Fix: EEVEE-Next: Displacement: Wrong output space
b0e291952b EEVEE-Next: Light: Explicit loop unrolling for better codegen
Compare 33 commits »
Raul Fernandez Hernandez pushed to main at farsthary/blender 2024-04-14 01:36:40 +02:00
fac451e513 Cleanup: Convert guarded allocator to C++
af88ff0484 Fix #107676 detection of "natural" touchpad scrolling on Wayland
c30afb7bf6 Fix #118694: snap cursor sometimes doesn't update when dragging and dropping assets
883af6dd63 Anim: Add custom pre and post frame color to motion path
3fda0d5f8f Fix #119909: Unkeyable custom properties receive keyframes
Compare 23 commits »
Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-13 17:24:30 +02:00
Sculpt: Improve Expand performance

Thanks everyone for their reviews and opinions.

Raul Fernandez Hernandez closed pull request blender/blender#120125 2024-04-13 17:22:19 +02:00
Sculpt: Improve Expand performance
Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-13 17:22:16 +02:00
Sculpt: Improve Expand performance

This was intended to be a stop-gap change to gain some performance while a deeper refactor that eliminates the need to pre-calculate a falloff field for the whole mesh lands, but now I'm even more…

Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-13 17:02:42 +02:00
Sculpt: Improve Expand performance

Can you add table of benchmarks of different parts of code to description of PR? If simple, I don't like idea of using threads in case where you can use log(N) search. I talk about N^2 search…

Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-13 16:53:09 +02:00
Sculpt: Improve Expand performance

That's an assumption that there's a data race, in the actual code running even with heavy meshes if there's data race it will be seen as unmasked vertices surrounded by masked vertices. I noticed…

Raul Fernandez Hernandez commented on pull request blender/blender#120125 2024-04-13 16:35:37 +02:00
Sculpt: Improve Expand performance

Ok lets drop this PR and leave the old slow code then since is what people want.

Raul Fernandez Hernandez pushed to improve-expand-perf at farsthary/blender 2024-04-13 02:09:33 +02:00
c37451d47b feedback