[fedora-arm] iwmmxt optimizations

[Daniel Drake]

On Tue, Aug 21, 2012 at 10:29 AM, Peter Robinson <pbrobinson at gmail.com> wrote:
> It would also be of use for the Marvell Armada server SoC bits. It was
> my understanding from Matt that it had already been pulled back into
> the gcc 4.7 branch, is that not the case? Ultimately if Matt's
> information about it being committed to the 4.7 branch was incorrect
> and Jakub isn't happy to backport it there's not a lot we can do. If
> it is in fact backported to 4.7 we can look at the release schedule
> for 4.7.2 and it should just arrive if 4.7.2 lands in F-18. I don't
> see anything about it in the PR Target Milestone report [1] but I also
> don't remember if Matt filed a PR (bug) report about it as I think it
> might have just been dealt with on the mailing list.

gcc-4.7 has been closed to new features since November 8th 2011.

Then on March 2nd 2012, gcc-4.7 was moved to its own branch, and the
trunk was opened for new developments into what will become gcc-4.8.

Then on June 13th 2012, iwmmxt support went into gcc trunk (but not
the gcc-4.7 branch, since that was already closed to new features 7
months beforehand).

So, this functionality was accepted waaay too late for gcc-4.7 and
will never be included in any gcc-4.7 release (right now they only
accept regression fixes and docs).
The first official release where it will appear is 4.8.0.

In terms of persuading Jakub, I wonder if having support from more
core members of the Fedora-ARM efforts would help (if it interests
them), and we could also ask Nick Clifton for his opinion (fedora
contributor who committed this upstream). The other option is for OLPC
to run a locally-compiled pixman package until gcc-4.8 hits Fedora,
which wouldn't be the end of the world.

Daniel