This is the mail archive of the
mailing list for the GCC project.
Re: Possible patches for bug 27565
Ian Lance Taylor <email@example.com> writes:
> "Joseph S. Myers" <firstname.lastname@example.org> writes:
> > but LOCAL_ALIGNMENT is defined as
> > #define LOCAL_ALIGNMENT(TYPE, ALIGN) \
> > ((TARGET_ALTIVEC && TREE_CODE (TYPE) == VECTOR_TYPE) ? 128 : \
> > (TARGET_E500_DOUBLE && TYPE_MODE (TYPE) == DFmode) ? 64 : \
> > (TARGET_SPE && TREE_CODE (TYPE) == VECTOR_TYPE) ? 64 : ALIGN)
> > which reduces the alignment from BIGGEST_ALIGNMENT (128) to 64 so causing
> > an assertion failure.
> I don't understand why we check this condition.
> In general we assume that the alignment of BLKmode is BITS_PER_UNIT.
> There is no other assumption that we can make. In particular, we
> certainly can't assume that a BLKmode value is aligned to
> BIGGEST_ALIGNMENT. So why do we insist on that here?
Of course, this also leads to the question of why LOCAL_ALIGNMENT
should return anything other than ALIGN when TYPE_MODE (TYPE) ==
BLKmode. Or: why is the mode BLKmode here?