This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Single Element Vectors
- From: Richard Henderson <rth at redhat dot com>
- To: Nick Clifton <nickc at redhat dot com>
- Cc: kumar dot gala at motorola dot com, gcc at gcc dot gnu dot org, aldyh at redhat dot com
- Date: Mon, 16 Sep 2002 21:33:30 -0700
- Subject: Re: Single Element Vectors
- References: <AC8CB01C-C6C8-11D6-8DC1-003065BFB10E@motorola.com> <m3elbyqroj.fsf@north-pole.nickc.cambridge.redhat.com>
On Fri, Sep 13, 2002 at 08:23:08AM +0100, Nick Clifton wrote:
> The second patch replaced the first patch and created a new vector
> type: V1D1 which then behaved as Motorola wanted. This second
> patch is much cleaner, and might even be suitable for submission to
> the FSF if the idea of single element vectors is accepted.
Aldy wandered by and re-explained the entire problem for me
from scratch. It seems that if we don't have _something_ to
distinguish the 64-bit vector result from regular DImode
values, we have really ugly register allocation problems even
discounting the function call abi issue.
The patch (forwarded to me off-list) is ok for mainline.
Nick, do post it to the list before you commit.
r~