This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] vectorizer related issues
- From: Diego Novillo <dnovillo at redhat dot com>
- To: Devang Patel <dpatel at apple dot com>
- Cc: Dorit Naishlos <DORIT at il dot ibm dot com>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
- Date: Thu, 08 Jan 2004 22:37:01 -0500
- Subject: Re: [tree-ssa] vectorizer related issues
- Organization: Red Hat Canada
- References: <OF84E7BD0B.816A0013-ONC2256DFE.002C8227-C2256DFE.0047E4E4@il.ibm.com> <1073580125.3165.93.camel@frodo.toronto.redhat.com> <5553DB3C-4235-11D8-9D2C-000393A91CAA@apple.com>
On Thu, 2004-01-08 at 18:49, Devang Patel wrote:
> Recently I spent some time understanding
> "Array SSA form and its use in Parallelization
> - Kathleen Knobe, Vivek Sarkar
> (POPL 98)"
>
> Any thoughts about their approach?
>
Yes, that's the one you'll see most often referenced. From what I
remember, we could use the VDEFs as their define-phi nodes. I wouldn't
want to use actual PHI_NODEs because those have specific control-flow
semantics that define-phis don't.
The "time-stamp" @ operators could be added to the virtual operands of
statements. Though we may also want to add them as real statements. We
will need to quickly collect iteration vectors with all the
corresponding index variables in nested loops for these time stamps. I
haven't thought much about that and it's been a while since I last read
the paper.
> What other papers you've in mind?
>
INPROCEEDINGS{bib:collard-99,
AUTHOR = {Jean-Francois Collard},
TITLE = {Array {SSA} for Explicitly Parallel Programs},
BOOKTITLE = {European Conference on Parallel Processing},
PAGES = {383-390},
YEAR = {1999},
}
Diego.