Internal Compiler Errors gcc 4.7.2 about albany HOT 8 CLOSED

Hi Brian,

My initial reaction is that we should pursue option (2) above. The features in these models that are causing issues were added in our effort to move towards kokkos compatibility. It seems like we could create a flag based on known compiler versions.

I don't like the fact that you have to comment out code to get things to compile.

Any other thoughts @lxmota ?

from albany.

Yes, I agree that the preferred option is (2).

from albany.

Brian, as per our conversation today, go ahead and take a stab at providing an internal flag and appropriate code guards. Please let me know if you encounter any issues.

from albany.

Hi Jake. This commit
https://github.com/bgranzow/Albany/commit/a852c9d3398de7cc6a10bf9006e60ef9601aa78c
on my fork of Albany seems to have resolved the issues for me, while not adversely affecting normal Albany builds.

Specifying -D DISABLE_LAMBDAS=ON in the Albany config script optionally disables the three models which were problematic with the gcc 4.7.2. compiler. The variable defaults to OFF and should maintain the previous behavior by default.

This is a pretty easy fix to apply, but it might make the upcoming merges we discussed more difficult than they need to be. My plan is to hold off applying these changes until those merges are complete. Sound reasonable?

from albany.

Hi Brian, looks good to me, and what you suggest sounds reasonable. When the upcoming kokkos related merges occur in the LCM source, we can test and hopefully close this issue.

from albany.

@bgranzow says w.r.t. 3a53f55:

Unfortunately, these changes did not resolve the error associated with #9 when I compiled at this ocmmit 64558eb.

My guess is this has more to do with how ParallelConstitutiveModel<...>::computeStateParallel(...) is implemented, not really how it was instantiated.

from albany.

@bgranzow here is another attempted fix for you to try: efa6968. Its in the dont_capture_this branch. Since the compiler bug has to do with a lambda capturing the this pointer, I tried to avoid doing that.

from albany.

The 'dont_capture_this' branch compiles without any warnings, and is more correct than the workarounds I suggested. I'm closing this issue.

Thanks @ibaned!

from albany.