Avoid many small allocations and just allocate all the structs in one array, which is 4 times faster. In a test with an armature modifier and custom normals, corner normal calculation went from 2.7 to 2.3 ms.
Avoid many small allocations and just allocate all the structs in one array, which is 4 times faster. In a test with an armature modifier and custom normals, corner normal calculation went from 2.7 to 2.3 ms.