[Smtk-developers] Tessellation experiments

[David Thompson]

Hi all,

I've explored the option of asking for all the tessellation info in a single call instead of one call per edge and face... and discovered that CGM's OpenCascade engine does not implement the method.

Furthermore, when loading the Audi_7.stp file with no tessellations computed, it takes 115 seconds on my machine (a debug build using a modified version of smtk/bridge/cgm/testing/python/cgmLoadFile.py). Of this, 93% of the time is spent inside 2 functions in OpenCascade's STEP importer. The remaining 7% is overhead added by CGM and SMTK. So even without tessellation, the bulk of the time importing STEP files appears to be in OpenCascade. I have not tried loading the Audi model with tessellation enabled (yet).

The Audi has 2539 "volumes" (probably sheet bodies) according to some OCC log messages and SMTK reports 36138 entities total (includes verts, edges, faces, volumes, and models).

I can import the Audi model into Cubit 14.1 in roughly 2 minutes, of which I *suspect* ACIS spends about 1 minute tessellating faces and edges (based on the fact that the progress bar suddenly starts moving about 1 minute into the load).

It appears that the OCC STEP translator is about half the speed of the Cubit one (although my OCC build is in debug mode) but that tessellation is really slow since it took Bob 40 minutes to get the model displayed in ModelBuilder. I suspect a good bit of that is OCC but some of it is certainly SMTK since it copies the triangulations into its own format and then again into a VTK dataset.

	David