Fix convergence errors on AWS EC2 Graviton instances about athena

[felker]

@jxuamazon

[tomidakn]

If I understand correctly, he said -O3 converges but the other configurations fail. I suspect that -O2 fails to unroll some functions and interfere vectorization with #pragma omp simd. (-O0 should work.) If this is the case, it is expected, but maybe it should be better documented.

[jxuamazon]

@tomidakn sorry, I remembered incorrectly when I talked to you about this. -O3 does not converge , but -O2 does.
@felker - I used Orszag-tan vortex test with 512x512 and 256x256 resolution. with -O3, the time step kept getting smaller and the program will run for a long time. I plan to run some more test on graviton 3 in the coming weeks.

[jxuamazon]

Tested this on C7g, the issue no longer exists. We can close this issue.

[jxuamazon]

tested on graviton3 c7g instance. the issue no longer exists.