A) Poke around in the integration tests for abinitio/abrelax. They use 1l2y.pdb, which is a trp-cage, which is conveniently only 20 residues (and thus computes fast). You should be able to use that integration test as a sanity check. (You may need to up the cycle counts from its default state). This is in test/integration/tests/abinitio.
The documentation is probably wrong on exactly which gcc version produces exactly which errors under whichever conditions. I’m pretty sure that line is supposed to read 4.4.1 (and 4.4.1 only). Where did you see that so I can try to fix it? One of our not-so-secrets is that “officially supported” really means “whatever architectures that some developer happens to use” – there are likely to be lots of gaps in our coverage (like, say, most of Windows land).
C) I don’t know much about it, but there is a membrane_abinitio you may want to consider.