This needs some more research on our side. The shadow buffers are view depenedent. Moving the viewport needs to recalculate the shadows that is an issue since the original implementation. It could be that the shadows are refreshed too much (every TAA sample?) but would like to dive deeper into this.

• Added mode to Set Alpha Node.
• Old files will use replace mode.
• New files will use multiply mode.

We could add a setting to the set alpha node (Premultiply). By default it is enabled. When loading older files we disable Premultiplication for any set alpha node.

Even when the viewport is finished rendering it still keeps all data around for quick scene updates. This means that memory on the GPU is still allocated. Could it be that in your test your RAM or GPURAM is full so it needs to swap? Or doesn't have enough head room to optimize.

Fixed spelling error

It should be able to return the uint hash and the float hash.
We could encapsulate this in an object or always return the uint hash and add a conversion function. The challenge with the encapsulation is to detect if we are looking at an object or asset hash.

• Separated the hashing and float conversion according to the spec.

• Fixed Comment
• Migrated requested render slot to use tagging

Replaced areas with tiles to improve the performance for many core systems

The effect what artists will notice was performance. But due to other changes (only redraw current frame) I don't think this is an important change. I would suggest that when we tackle this part of the code we would like to go to CPP. The current code structure is awkward due to the lack of templates and classes.

• Switch from partial refresh to full refresh based on the requested area size

Merged 2.91 into master to remove unrelated changes

• Removed has_other_size function

• Fixed crash during texture painting.

• Fix spelling in comment

• Small code review comments

Alignment of tiles to GPU texture

Ignore my previous investigation. And something like dumb questions do not exist :-)

First impression is that the back edges originates when colors are converted twice from straight alpha to pre-multiplied alpha.

Missing clangformat

Draw stencil overlay to the overlay buffer

Yes gpu_texture_update_from_ibuf handles this for texture painting.

When cycles is rendering it creates a render result for specific tiles (offset + dimension) to the actual render result. It calls the rect on the RenderPass. This will sync a tile to blender. and call the RE_engine_end_result; what will sync the tile result back to the full image (`render_result_merge).

The behavior is documented in the manual (https://docs.blender.org/manual/en/latest/interface/controls/templates/color_picker.html#color-picker) But can understand that it isn't read or understood by users. Color management is typically a difficult subject.

This code was intended for research. If we actually want a test case we should do this around the ImBuf API.