Struct of Arrays Refactor for Mesh Edges #95966

New Issue

Hans Goudey · 2022-02-22T20:27:22+01:00

Hans Goudey commented

2022-02-22 20:27:22 +01:00

Currently, MEdge stores quite a bit of data:

typedef struct MEdge {
  unsigned int v1, v2;
  char crease, bweight;
  short flag;
} MEdge;

For the reasons described in #95965, this is an inefficient way to organize data.
This task describes how MEdge should be changed to only store the two vertex indices.
In the end, edges should just store two vertex index integers. Everything else should be optionally stored elsewhere.

Span<int2> edges;

Bevel Weight

In the longer term, this should be stored as an optional named generic float attribute (see #89054).
In the shorter term, this can be stored as CD_BWEIGHT. This means slightly increased memory usage when this data is used,
but it also gives increased precision, and it doesn't have to be allocated when it isn't used.

291c313f80

Flag

In the flag is:

  /*  SELECT = (1 << 0), */
  ME_EDGEDRAW = (1 << 1),
  ME_SEAM = (1 << 2),
  /*  ME_HIDE = (1 << 4), */
  ME_EDGERENDER = (1 << 5),
  ME_LOOSEEDGE = (1 << 7),
  ME_EDGE_TMP_TAG = (1 << 8),
  ME_SHARP = (1 << 9),

Selection 12becbf0df
It's not clear to me how selection interacts with mselect (MSelect), but perhaps it can be moved there.
It can also be moved to a separate boolean custom data layer with a name that keeps it from being displayed in the UI.

ME_SEAM cccf91ff83
Ideally this would move to a completely generic boolean attribute, since only having one seam layer is limiting.

ME_HIDE 2480b55f21
This can be moved to a separate boolean custom data layer, likely with a reserved name.

ME_EDGEDRAW 7ca651d182
ME_EDGERENDER 10131a6f62
As far as I can tell, these are always used together, like ME_EDGEDRAW | ME_EDGERENDER. It looks like they are only runtime data.
They should be stored in Mesh_Runtime as a bitmask or an array of booleans. This way they can only be allocated when they are actually needed,
like drawing a mesh after a subdivision surface modifier.

ME_LOOSEEDGE 1ea169d90e
This is also runtime data, so it should be stored in Mesh_Runtime, with some special API like mesh normals have now.

ME_EDGE_TMP_TAG 17567c235a
Can be easily removed, it's only used in one place.

ME_SHARP dd9e1eded0
Should be moved to a separate boolean layer and exposed as a builtin attribute.

Currently, `MEdge` stores quite a bit of data: ``` typedef struct MEdge { unsigned int v1, v2; char crease, bweight; short flag; } MEdge; ``` For the reasons described in #95965, this is an inefficient way to organize data. This task describes how `MEdge` should be changed to only store the two vertex indices. In the end, edges should just store two vertex index integers. Everything else should be optionally stored elsewhere. ``` Span<int2> edges; ``` ### Bevel Weight In the longer term, this should be stored as an optional named generic float attribute (see #89054). In the shorter term, this can be stored as `CD_BWEIGHT`. This means slightly increased memory usage when this data is used, but it also gives increased precision, and it doesn't have to be allocated when it isn't used. 291c313f80 ### Flag In the flag is: ``` /* SELECT = (1 << 0), */ ME_EDGEDRAW = (1 << 1), ME_SEAM = (1 << 2), /* ME_HIDE = (1 << 4), */ ME_EDGERENDER = (1 << 5), ME_LOOSEEDGE = (1 << 7), ME_EDGE_TMP_TAG = (1 << 8), ME_SHARP = (1 << 9), ``` **Selection** 12becbf0df It's not clear to me how selection interacts with `mselect` (`MSelect`), but perhaps it can be moved there. It can also be moved to a separate boolean custom data layer with a name that keeps it from being displayed in the UI. **ME_SEAM** cccf91ff832d119dbf048b0518a696b9aa83bce4 Ideally this would move to a completely generic boolean attribute, since only having one seam layer is limiting. **ME_HIDE** 2480b55f21 This can be moved to a separate boolean custom data layer, likely with a reserved name. **ME_EDGEDRAW** 7ca651d18274acacfe8e0ecb7bc6fdd3399ea01b **ME_EDGERENDER** 10131a6f62 As far as I can tell, these are always used together, like `ME_EDGEDRAW | ME_EDGERENDER`. It looks like they are only runtime data. They should be stored in `Mesh_Runtime` as a bitmask or an array of booleans. This way they can only be allocated when they are actually needed, like drawing a mesh after a subdivision surface modifier. **ME_LOOSEEDGE** 1ea169d90e This is also runtime data, so it should be stored in `Mesh_Runtime`, with some special API like mesh normals have now. **ME_EDGE_TMP_TAG** 17567c235a Can be easily removed, it's only used in one place. **ME_SHARP** dd9e1eded0 Should be moved to a separate boolean layer and exposed as a builtin attribute.

Hans Goudey commented

2022-02-22 20:27:22 +01:00

Added subscriber: @HooglyBoogly

Hans Goudey commented

2022-02-22 20:43:40 +01:00

Changed status from 'Needs Triage' to: 'Confirmed'

Campbell Barton commented

2022-06-02 07:57:52 +02:00

Added subscriber: @ideasman42

Campbell Barton commented

2022-06-02 07:57:52 +02:00

edit replied to #95965.

*edit* replied to #95965.

blender-admin commented

2022-07-31 18:51:38 +02:00

This issue was referenced by 17567c235a

This issue was referenced by 17567c235ac62f6d0f7e3bf0cea50593633099da

blender-admin commented

2022-08-11 18:54:24 +02:00

This issue was referenced by 2480b55f21

This issue was referenced by 2480b55f216c31373a84bc5c5d2b0cc158497c44

blender-admin commented

2022-09-09 15:29:07 +02:00

This issue was referenced by 291c313f80

This issue was referenced by 291c313f80b4cccc8fcce3035584caeaa654844f

blender-admin commented

2022-09-23 16:02:05 +02:00

This issue was referenced by a8a454287a

This issue was referenced by a8a454287a27d408668f8adc6fe1b3aa988de1ac

blender-admin commented

2022-09-23 16:38:37 +02:00

This issue was referenced by 12becbf0df

This issue was referenced by 12becbf0dffe06b6f28c4cc444fe0312cf9249b9

Hans Goudey commented

2022-11-16 23:39:22 +01:00

ME_EDGEDRAW Flag

Currently the ME_EDGEDRAW flag is only ever set to false for the "optimal display" feature of the subdivision surface modifier.
When applying the modifier back to the original mesh this is turned off though.
Blender only produces original (in Main) meshes with every ME_EDGEDRAW flag set.

So, moving forward it makes sense to remove the BM_ELEM_DRAW flag from edges completely.
If that functionality is needed, disabling drawing of certain edges should be a viewport/overlay feature rather than being stored in the original mesh.
ME_EDGEDRAW should become runtime data-- a bitmap just like MeshRuntime::subsurf_face_dot_tags.

**ME_EDGEDRAW Flag** Currently the `ME_EDGEDRAW` flag is only ever set to false for the "optimal display" feature of the subdivision surface modifier. When applying the modifier back to the original mesh this is turned off though. Blender only produces original (in Main) meshes with every `ME_EDGEDRAW` flag set. So, moving forward it makes sense to remove the `BM_ELEM_DRAW` flag from edges completely. If that functionality is needed, disabling drawing of certain edges should be a viewport/overlay feature rather than being stored in the original mesh. `ME_EDGEDRAW` should become runtime data-- a bitmap just like `MeshRuntime::subsurf_face_dot_tags`.

blender-admin commented

2022-11-18 23:05:06 +01:00

This issue was referenced by 1ea169d90e

This issue was referenced by 1ea169d90e39647eac721cc1fee9927b0432bf97

Jamell Moore commented

2022-12-13 14:43:44 +01:00

Added subscriber: @JamellMoore

Jamell Moore commented

2023-01-05 19:13:35 +01:00

@HooglyBoogly I' currently working with a patched build where I've exposed CD_BWEIGHT as a built-in named CD_PROP_FLOAT attribute (bevel_weight) as it allows controlling bevels using a bevel modifier. Would you welcome this in the short term?

The bevel node seems to be a community project that may take many releases to complete. This may be useful to other users until the node is complete or/and the bevel modifier allows generic attributes.

@HooglyBoogly I' currently working with a patched build where I've exposed CD_BWEIGHT as a built-in named CD_PROP_FLOAT attribute (bevel_weight) as it allows controlling bevels using a bevel modifier. Would you welcome this in the short term? The bevel node seems to be a community project that may take many releases to complete. This may be useful to other users until the node is complete or/and the bevel modifier allows generic attributes.

Hans Goudey commented

2023-01-05 19:22:30 +01:00

That is tempting, personally I don't think it's a great approach though. It's hard to add that sort of temporary feature, because people start to depend on it, and then it becomes somewhat permanent because we try hard to keep backwards compatibility for new releases. I would like to avoid that sort of temporary compromise becoming part of the permanent design.

It's not procedural, but would adding a "Bevel Weight" option to the convert attribute operator be helpful, so at least you could apply the modifier and convert an edge float attribute to bevel weights?

That is tempting, personally I don't think it's a great approach though. It's hard to add that sort of temporary feature, because people start to depend on it, and then it becomes somewhat permanent because we try hard to keep backwards compatibility for new releases. I would like to avoid that sort of temporary compromise becoming part of the permanent design. It's not procedural, but would adding a "Bevel Weight" option to the convert attribute operator be helpful, so at least you could apply the modifier and convert an edge float attribute to bevel weights?

Jamell Moore commented

2023-01-05 21:08:03 +01:00

I can't foresee any issues myself with regards to backwards compatibility other than ones you would have to deal with when transitioning to generic attributes, but that's fair enough though. I'm sure I've overseen something and it's better to be safer than cause issues down line.

Adding a "Bevel Weight" option to the Convert Attribute Operator wouldn't help for my use because I'm making procedural assets. It could be useful for other contexts though where the final output is non-procedural.

I can't foresee any issues myself with regards to backwards compatibility other than ones you would have to deal with when transitioning to generic attributes, but that's fair enough though. I'm sure I've overseen something and it's better to be safer than cause issues down line. Adding a "Bevel Weight" option to the Convert Attribute Operator wouldn't help for my use because I'm making procedural assets. It could be useful for other contexts though where the final output is non-procedural.

blender-admin commented

2023-01-10 22:12:14 +01:00

This issue was referenced by dd9e1eded0

This issue was referenced by dd9e1eded0d48e19a500233b20e8c5f18ed12a09

Hans Goudey referenced this issue

2023-02-07 18:45:26 +01:00

Mesh: Remove unnecessary edge draw flag #104417

Hans Goudey referenced this issue from a commit

2023-02-09 15:56:07 +01:00

Mesh: Remove unnecessary edge draw flag

Joseph Eagar referenced this issue from a commit

2023-02-11 21:53:09 +01:00