This is the mail archive of the
mailing list for the GCC project.
Re: [RFC] PR/27390, latent complex-mode bug in regstack
[snip long explanation of possibilities to try]
Does all this make sense? (and rationalizes the strange behaviours
you've been seeing). Let me know if your happy to investigate the above
refinements or you'd prefer to commit something similar to what you have
(to resolve the regressions) and leave it to someone else to tidy up
It does, but given that we do not have a testcase for this above -O0,
and that I'd like the fix to be as little invasive as possible, I'd be
inclined to go with the patch that I posted if Richi's testing (which
now I humbly request) shows that it does not introduce new regressions.
The patch is accessible at
http://gcc.gnu.org/bugzilla/attachment.cgi?id=11597 on Bugzilla, to
avoid any confusion.
If somebody finds a testcase for optimizing compilation, I'll be happy
to look at your proposed improvements for 4.3.
Many thanks for investigating this. It's now clear that this was a
pre-existing latent bug that your reload preference changes had
No problem. Unfortunately, even if I knew that it was quite easy a bug,
I would not have had time to look at it before. :-(