This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [RFC, LRA] Incorrect subreg resolution?
- From: Richard Sandiford <rdsandiford at googlemail dot com>
- To: "H.J. Lu" <hjl dot tools at gmail dot com>
- Cc: Tejas Belagod <tbelagod at arm dot com>, "vmakarov\ at redhat dot com" <vmakarov at redhat dot com>, "gcc\ at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>
- Date: Sun, 12 Jan 2014 09:00:31 +0000
- Subject: Re: [RFC, LRA] Incorrect subreg resolution?
- Authentication-results: sourceware.org; auth=none
- References: <52D03D3C dot 60300 at arm dot com> <87k3e6q12u dot fsf at talisman dot default> <CAMe9rOocsgTB+6cUFVSPW9svH-PGYGixk8naey8CTewR3FAnZA at mail dot gmail dot com>
"H.J. Lu" <hjl.tools@gmail.com> writes:
> How about this patch
>
> http://gcc.gnu.org/git/?p=gcc.git;a=patch;h=23023006b946e06b6fd93786585f2f8cd4837956
>
> I tested it on Linux/x86-64 without any regressions.
I don't think we should key this off vectorness. It's a question of
whether the class contains a register for which the given subreg parameters
are valid. E.g. on some targets the FP registers are wider than the general
registers. And on some targets (e.g. NEON) you can use pairs of vector
registers, in which case there are two valid offsets for each "to" mode.
Target-independent code could figure this out from the information
it already has. I don't think new hooks should be needed.
Thanks,
Richard