This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PARCH 1/2, x86, PR63534] Fix darwin bootstrap
- From: Evgeny Stupachenko <evstupac at gmail dot com>
- To: Mike Stump <mikestump at comcast dot net>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>, Uros Bizjak <ubizjak at gmail dot com>, Jakub Jelinek <jakub at redhat dot com>, Jeff Law <law at redhat dot com>, iains at gcc dot gnu dot org
- Date: Wed, 22 Oct 2014 21:24:51 +0400
- Subject: Re: [PARCH 1/2, x86, PR63534] Fix darwin bootstrap
- Authentication-results: sourceware.org; auth=none
- References: <CAOvf_xwDUU=gsQqHgybovAFHbn1+OVHYNBw+=xG0jau1wG2HDg at mail dot gmail dot com> <29736130-A4C0-41D1-8B90-670492F0E185 at comcast dot net>
>For example, is the pic register saved and restored?
Yes, if RA decided that it is better to save it.
>If restored, is the code to save it actually better than simply appearing it out of thin air as did previously?
"enabling ebx" gives performance mostly to loops. There still could be
some performance issues, however we got pretty good results on a set
of benchmarks:
https://gcc.gnu.org/ml/gcc-patches/2014-10/msg00892.html
To get performance diff now you can compare r216153 svn revision and
r216304 with the patch in 63534#c33
There was quite long discussion on enabling starting here:
https://gcc.gnu.org/ml/gcc-patches/2014-08/msg02186.html
On Fri, Oct 17, 2014 at 10:32 PM, Mike Stump <mikestump@comcast.net> wrote:
> On Oct 17, 2014, at 7:08 AM, Evgeny Stupachenko <evstupac@gmail.com> wrote:
>> The patch fixes 1st fail in darwin bootstarp.
>> When PIC register is pseudo we don't need to init it after setjmp or
>> non local goto.
>>
>> Is it ok?
>
> So, I donât see commentary in the PR that all fallout and all bugs introduced are fixed by the patch. :-(
>
> Given how central pic is to code-gen, I donât favor any patch, until all bugs and regressions are fixed. If they canât be, then I favor reversion of the patch that broke everything.
>
> Additionally, if given the types of changes to codegen, Iâd like to see the before and after changes to try and ensure that code-gen quality isnât regressed. For example, is the pic register saved and restored? If restored, is the code to save it actually better than simply appearing it out of thin air as did previously? If the patch that did all the pic work was in one patch, it would be easier for me to see the change in its entirety.