QuadMesh normal issue on a single workstation

QuadMesh normal issue on a single workstation

Hi,

We are currently using Embree 2.9.0. - a single static scene with one or more TriangleMesh _or_ QuadMesh.

Everything is running and working fine _almost_ everywhere. We tested the software on Intel and non-intel CPUs, Opteron racks, VMs, 6 years old laptopts, whatever, and _everything_ is working great.

Except, that is, for a single PC with a core i7 2700k, There (and only there) QuadMeshes have an inverted normal on one of its 2 tris. And we simply cannot figure out why. Its a Windows 10 machine 64bit with 8 GB RAM and a 2700k. nothing special. still, QuadMeshes simply refuse to have the same normal direction on both triangles. I just would like to stress that this behaviour can only be seen on that particular machine only. Wherever else, the software handles QuadMeshes pretty good.

Now, I understand that without having a piece of source code it may be quite difficult to have an idea, I was just wondering if you ever experienced anything like that, or if it was a rare but known issue.

We tried setting RTC_SCENE_ROBUST, but aside from that, there are not much more flags we can play with. Anyway we simply dropped QuadMeshes and use TriangleMeshes everywhere (which I believe Embree internally does anyway), and everthing is fine.

Just FYI

thanks

PS: should you need pieces of source just contact me

4 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hi,

the inverted normal sounds strange and could be a buggy code path. Could you give us some more information:

- Does the inverted normal happen for the Embree tutorials too or only inside your app?

- Is this for the single ray interface only or do you trace packets/streams of rays?

- Could you pass 'verbose=2' to the embree configuration string to enable a more verbose output and send us the output?

If this does not provide further inside we might need a small reproducer (either binary or source code).

Thanks,

Carsten

I will make some test and post the results

thanks for your quick answer

Hi Germano, we located the issue and will fix it for the next release. Thus there is no need for you to provide additional information.

Leave a Comment

Please sign in to add a comment. Not a member? Join today