Practical Info
This is a weekly video chat meeting for planning and discussion of Blender Eevee/viewport module development. Any contributor (developer, UI/UX designer, writer, …) working on Eevee/viewport in Blender is welcome to join.
For users and other interested parties, we ask to read the meeting notes instead so that the meeting can remain focused.
- Google Meet
- Next Meeting: February 27, 2023, 11:30 AM to 12:00 PM Amsterdam Time (Your local time: 2023-02-27T10:30:00Z → 2023-02-27T11:00:00Z)
NOTE: Meeting of February 20th has been cancelled.
Attendees:
- Clement
- Jeroen
- Omar
- Michael
- Miguel
Module
- Gitea migration and setting up workboards
- New issues/pull requests should be tagged with the EEVEE & viewport project.
- Documented GLSL code-style.
- Including what to avoid in GLSL for MSL compatibility.
Blender 4.0
These are initial ideas for planning, and can still change depending on resources and actual effort that is required.
- Prioritize *-Next projects for Blender 4.0 project. This is a lot of work and might require everyone.
- NPR project would then be a target for after 4.0 release. There is no agreement on the proposed design yet. Time can still be spent on the design and making prototypes. Workshop should be held.
- Vulkan is important for AMD and Intel and should still be part of the planning. At Blender HQ there is an idea to do the major part of the development this year. This also depends on the developement resources we can get.
- Prioritise on a not optimized implementation.
- Optimizations could than delayed to after Blender 4.0 targets are done. And would give the vendors more time to get into the project as well.
- Current development is used to understand the optimal requirements for the Vulkan back-end. After understanding a design and implementation can be made. An example for this would be to improve command synchronizations.
Eevee-next
- EEVEE-Next’s new Virtual Shadow Map is now in master (now called main). A devtalk post explain a bit what to expect from it.
- The BLI_math_float4x4.hh removal was also finally merge with all compilation issues fixes.
- Rotation API task is created that can be picked up by a developer (also community).
- Working on transparent tagging for shadow maps. Mostly done, needs some work, but first the pull request should be reviewed. More refining can be done afterwards.
- Bump node optimizations have been commited.
Overlay-Next
- Porting empty drawing to the Overlay-Next engine and found a ways to improve the code quality.
Workbench-next
- Added task for mat-cap improvements. Sculpt, Texture & Paint wants to update the mat-caps so this should be aligned.
Metal backend
- Python: Suppress BGL deprecation messages after 100 times.
- Shadergraph optimization patch updated with new ideas. With parent shaders can remove safely remove stuttering. There are more stuff that can be balanced compilation times versus stutteling.
- New features will be uploaded after the bcon3 change
- Pull requests are uploaded to solve the AMD issues.
Vulkan backend
- Vulkan: Use guardedalloc for driver allocations (#104434 - Vulkan: Use guardedalloc for driver allocations. - blender - Blender Projects)
- Equalize API of readback for IBO/VBO and SSBOs. (#104571 - GPU: Use same read back API as SSBOs - blender - Blender Projects)
- Vulkan Compute shaders (https://projects.blender.org/blender/blender/pulls/104518)
- Direct dispatching of compute tasks
- Initial SSBO, VBO buffers working.
- Current DEVICE_ONLY usage requires host visibility during test cases. My proposal would be to have a separate usage flag for host visibility as this should be treaded as an exception. This way back-end can still do further optimizations when DEVICE_ONLY is used. Metal back-end currently doesn’t support DEVICE_ONLY with host visibility.