This is the mail archive of the
mailing list for the GCC project.
Re: Fw: [patch] [4.2 projects] vectorize strided accesses (4 patches)
- From: Roger Sayle <roger at eyesopen dot com>
- To: Victor Kaplansky <VICTORK at il dot ibm dot com>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Tue, 5 Sep 2006 09:13:27 -0600 (MDT)
- Subject: Re: Fw: [patch] [4.2 projects] vectorize strided accesses (4 patches)
On Tue, 5 Sep 2006, Victor Kaplansky wrote:
> I'm working on merging mainline changes into autovect branch just now.
> Which BOOT_CFLAGS did you use?
I didn't touch BOOT_CFLAGS at all, so the compiler itself wasn't
autovectorized. Just "../autovect/configure" and "make bootstrap".
I was interested in seeing whether dot-product reduction was
functional on the branch, what the anticipated performance benefits
are, and how many of vectorization related PRs on mainline have
already been resolved on the branch, but I was a bit surprised that a
branch that was targeted for 4.2, didn't bootstrap "out-of-the-box"
on x86. Both mainline and 4.1 build and test fine on the same machine.
As I mentioned previously, this failure may be unrelated to the
autovect changes; You may have been unlucky that java was broken
the last time you synchronized the tree, or not all of the tree was
synchronized. But its also not unlikely, that the breakage is fall-out
from vectorization related changes to common infrastructure if the
branch isn't regularly tested for building java, fortran, ada, etc...
Thanks again for investigating this.