https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65482
Bill Schmidt <wschmidt at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2017-01-26
CC| |meissner at gcc dot gnu.org,
| |uros at gcc dot gnu.org,
| |wschmidt at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #2 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Looking at the history, the reference to -mno-allow-movmisalign in
testsuite/lib/target-supports.exp appears in check_vect_support_and_set_flags {
}, which was added by Uros in November 2010. It appears as a default flag when
VSX is present but later ISA additions are not (POWER8 and later).
I am not sure why this is a default option; I would assume that Uros consulted
with somebody on POWER regarding this, probably one of the Haifa folks who were
working on the vectorizer originally. This predates me joining the project by
a couple of months. CCing Uros and Mike to see if they have any recollection.
I expect we probably had problems with the movmisalign<mode> patterns at the
time of that change, and this is no longer necessary, but I'm hesitant to make
a change this late in the release. I will run an experiment sometime to see
what the fallout is from removing it, though.
