Assertion failure in tnutil.cxx

Has anyone seen the following error?

Assertion failure at line 686 of …/…/be/cg/tnutil.cxx:

Compiler Error in file test.cpp3.i during Code_Expansion phase:

don’t know how to make dedicated TN for class (null)

nvopencc INTERNAL ERROR: /usr/local/cuda/open64/lib//be returned non-zero status 1

I see the error when compiling a program on CUDA 3.1 for both the Mac and Linux64 (other platforms not tested). The same piece of code compiles and runs without error on CUDA 3.0. The error on 3.1 disappears when debugging is enabled (-g -G). I’ve been working to isolate the cause but the error message gives little indication to what the problem is. I’ve been trying to brute force and comment out large sections of the code to find the problem, but thus far to no avail. Does anyone have any insight as to what I might be looking for?

This kind of error message indicates a problem internal to the compiler. If you are a registered developer, please file a bug, attaching the failing code. If you are not a registered developer, I can take a look at this if you could provide a self-contained application that repros the issue on Linux with CUDA 3.1. If you don’t want to post the code publically, you can send me a personal message via the forum. Thanks.

This kind of error message indicates a problem internal to the compiler. If you are a registered developer, please file a bug, attaching the failing code. If you are not a registered developer, I can take a look at this if you could provide a self-contained application that repros the issue on Linux with CUDA 3.1. If you don’t want to post the code publically, you can send me a personal message via the forum. Thanks.

I applied to the registered developer program earlier this week but have not received a response yet.

I applied to the registered developer program earlier this week but have not received a response yet.

If you have a self-contained repro case (Linux64 / CUDA 3.1), I’d be happy to look at it. Feel free to send it in a personal message via the forums.

If you have a self-contained repro case (Linux64 / CUDA 3.1), I’d be happy to look at it. Feel free to send it in a personal message via the forums.

My application to the developer program was accepted. I plan to submit a bug report soon. It will take me some time to create a self-contained case to reproduce the issue.

My application to the developer program was accepted. I plan to submit a bug report soon. It will take me some time to create a self-contained case to reproduce the issue.

I noticed that 3.2RC2 was posted on Wednesday. I downloaded and installed it and the problem does not appear. Can I assume that the issue is fixed, or should I still try to submit a bug report for 3.1?

I noticed that 3.2RC2 was posted on Wednesday. I downloaded and installed it and the problem does not appear. Can I assume that the issue is fixed, or should I still try to submit a bug report for 3.1?

If the problem no longer manifests with the CUDA 3.2 release candidate, it has likely been fixed at this point. Impossible to say for sure of course. Since it sounds like it would create considerably work on your side to prepare a repro case for the issue you saw with 3.1, I would say just move forward with 3.2.

If the problem no longer manifests with the CUDA 3.2 release candidate, it has likely been fixed at this point. Impossible to say for sure of course. Since it sounds like it would create considerably work on your side to prepare a repro case for the issue you saw with 3.1, I would say just move forward with 3.2.

Thank you. I will forgo submitting a bug report for 3.1.

Thank you. I will forgo submitting a bug report for 3.1.