This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [committed] PA long branch support
- From: Jeffrey A Law <law at redhat dot com>
- To: John David Anglin <dave at hiauly1 dot hia dot nrc dot ca>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Sun, 09 Apr 2006 13:06:35 -0600
- Subject: Re: [committed] PA long branch support
- References: <200604091740.k39He5tG018289@hiauly1.hia.nrc.ca>
- Reply-to: law at redhat dot com
On Sun, 2006-04-09 at 13:40 -0400, John David Anglin wrote:
> The enclosed patch adds long-branch support to all the PA conditional
> branch patterns. Conditional branch distances can now be larger than
> that available using the 17-bit pc-relative 'b' instruction.
>
> The patch has been tested on hppa-unknown-linux-gnu with no regressions
> and testcase for PR 26743. However, the patch is hard to test as a GCC
> build doesn't trigger the use of long branches. Thus, I've probably
> made a few mistakes ;)
>
> Thanks to Randolph Chung who provided some code for the output_movb
> case and a testcase. This showed that I'd somewhat botched the handling
> of nullification in my previous attempt at fixing output_cbranch.
>
> Once the technique of handling long branches was worked out, the rest
> of the changes were more or less mechanical.
>
> Committed to trunk.
One of the ways I used to test this this kind of stuff was to
drastically reduce the maximum displacement for the short branches.
ie, for a simple conditional I might claim the maximum displacement
was +- 8 bytes to force it to use a longer version.
You could probably do something similar to stress test this.
jeff