Fix recent libgomp failures

[Jeff Law <law at redhat dot com>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


As detailed in PR 37273 and 47464, my patch exposed a latent bug between
the interactions of reload & ifcvt.  I didn't see the failures when I
original tested the patch for 37237, but changes since I did the testing
and final submission obviously changed the code presented to reload &
ifcvt in such a way as to uncover the bug.

Eric B suggested using may_trap_or_fault_p rather than may_trap_p in
noce_operand_ok.  Instead I put the check in can_move_insns_across
mostly because I was already more familiar with that code and knew it'd
address the problem both for ifcvt and head merging.

I'll be following up with a separate patch to make the same change to
noce_operand_ok.

Bootstrapped and regression tested on x86_64-unknown-linux-gnu.
Verified it fixes the libgomp and gcc omp tests.  Installed.

Jeff
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJNQKQfAAoJEBRtltQi2kC7ePcH/ieSwr2mtnxRWniaAMw9sX9M
+AIztcrYbhqrk57JPNs0XatMe6ZaJo+lWbpM0sBjhbKsZEbsgRAgg5bDrIhqT1H7
zixMC5jazpdiIYos2Ws9nNVFP1CnfMN/x5sKP7WqCTLeZhqkgE3jxpxI66fP535o
ArgJPSaD42+/5gqC630ilVh6RZMooPihtVV7oC+WTryMbuBmfdnUwZ0C8iGcGOuT
K6r9Qm+vkyLn/+DDfM6/guvJcdxPOSW49VcgBZxUI6gW+5nHt+XE9DSnUKXbpZqb
2xJ7CwmBEuBBzLKEuskEibaJ2iomU8X7fJrEQD//2KxMTCs+zCVWsz0o/PzvXLg=
=gJ/Z
-----END PGP SIGNATURE-----

Attachment: P
Description: Text document