Hello,
Not sure why this is in a collada file, but seems this code is used when exporting glTF2 files

Can you explain what this change is supposed to do ? Is that simply like the for loop is distributed to run on multiple processors so that multiple curves can be calculated at the same time ? Then i believe this is a good thing :)

@Julien DUROURE (julien) This file is in the collada module because it originally was created for the collada importer/exporter. And i am happy to see that this stuff is reused elsewhere now :)

I had a choice of per object, per bone and per curve keyframes. So I moved it to the closest function to the work.

This blender 3d curve baking optimization is done sequentially curve by curve so each curve is divided among all the keyframes to all the cores.

@Ernest Lee (iFire) I suspect that when each core handles one entire curve at a time that might possibly be more efficient? I am not sure though, did you try ? Or maybe i misunderstand how it works..

I’m not sure how to do a build for you to test.

@Ernest Lee (iFire) Can you provide your test case and a brief instruction how to do the test ? I would like to debug this as it sounds a bit odd that you can improve the runtime that way by a factor of 3. If that is true then maybe the code itself is not efficient.

@Sybren A. Stüvel (sybren) I am not sure if we can use OMP wherever we want. Are there any rules when its allowed to use it or is there another method to use in Blender?

Here's the test I'm imagining.

1. Build Blender with the CICD automation with this patch in two locations
2. location 1 is the current patch with each curve is paralleled
3. location 2 is working on each object parallelized
4. I will provide a blender test scene in the near future.

For location two it'll be:

	  /* Add curves on Object->material actions*/
	  object_type = BC_ANIMATION_TYPE_MATERIAL;
#ifdef WITH_OMP
#pragma omp parallel for
#endf
	  for (int a = 0; a < ob->totcol; a++) {