This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: powerpc & unaligned block moves with fp registers
- To: dje at watson dot ibm dot com
- Subject: Re: powerpc & unaligned block moves with fp registers
- From: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- Date: Thu, 8 Nov 01 16:28:39 EST
- Cc: gcc at gcc dot gnu dot org
I guess I did not explain this clearly enough. The DImode objects
do not need to be naturally aligned. Following the definition of
STRICT_ALIGNMENT to 0, these instructions will work, just work more
slowly.
You can't have it both ways! Either they are too slow, in which case
STRICT_ALIGNMENT should be set or they are *not* too slow, in which
case there is no problem.
We discussed STRICT_ALIGNMENT a few years ago and agreed that
STRICT_ALIGNMENT was not the solution. This is why we expanded
SLOW_UNALIGNED_ACCESS in the first place.
Right, but the question is *how slow*? If they are just a few times
slower than aligned, that's fine, but if they are many times slower,
then I think they simply shouldn't be used at all and STRICT_ALIGNMENT
should be true.