This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
ia64 vector mode support
- From: "Jan Beulich" <JBeulich at novell dot com>
- To: <rth at redhat dot com>
- Cc: <gcc at gcc dot gnu dot org>
- Date: Tue, 04 Jan 2005 17:52:39 +0100
- Subject: ia64 vector mode support
Richard,
while I was about to get started with an attempt to implement these, I
got the notification that you just yesterday checked in some code going
into that same direction. Andrew comment in a bugzilla entry I had
opened a while ago for that pointed out that this is only for
autovectorization support, my intention was to get the intrinsics the
Intel compiler supports in a usable shape.
Now, the first and obvious question is: Are you intending to continue
your work into that direction? If so, I would probably want to stay away
from doing the same thing a second time.
What seems more problematic to me in this same area, however, is that
there appear to be more fundamental problems, at least with floating
point vector modes:
- Comparing with 3.4.3, the way they get passed as arguments or
returned has changed, mainly (as far as I can tell) because of changes
in common code. Is this intentional? If not, did anyone even notice?
- Regardless of whether I look at the old scheme or the new one, I am
fairly convinced that the parameter/return value passing is wrong. My
main problem here is to determine how to correct it: Matching the ABI
(according to which, as I'm reading it, these should be treated as
HFAs), matching the Intel compiler (which I can compare with), or
matching some other compilers (which I have no access to) that may use
yet different mechanisms.
Jan