This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] Add support for pragma Loop_Optimize ([No_]Vector)
- From: Richard Biener <richard dot guenther at gmail dot com>
- To: Eric Botcazou <ebotcazou at adacore dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>, Jakub Jelinek <jakub at redhat dot com>
- Date: Tue, 15 Apr 2014 10:11:30 +0200
- Subject: Re: [patch] Add support for pragma Loop_Optimize ([No_]Vector)
- Authentication-results: sourceware.org; auth=none
- References: <1411548 dot 4VpsVizfpc at polaris> <CAFiYyc1xMKnkVK+g_YJ-qc1KHiWqcV+5VDEwcocRR3e2nE4KJw at mail dot gmail dot com> <5123786 dot 1DhOWysV6X at polaris>
On Tue, Apr 15, 2014 at 10:01 AM, Eric Botcazou <ebotcazou@adacore.com> wrote:
>> The loop flags copying should go into copy_loop_info instead of only to
>> copy_loops. Jakub - I see you remap simduid on copy - you have to do
>> sth in copy_loop_info instead I suppose. See the other callers.
>
> That also occurred to me, but IMO it's not crystal clear; for example, ivdep
> (aka safelen) is not in copy_loop_info either. So I think this needs to be
> further discussed.
Well, there are passes that can end up duplicating loops and thus you lose
no_vectorize on the copy for example. Clearly that's undesired, no? For
safelen and simduid not copying them is erroring on the safe side at least,
likewise for force_vectorize. But we do have the copy_loop_info abstraction
for a reason. Otherwise we should simply discard it.
Richard.
>> Otherwise I'm fine with this patch.
>
> Thanks, I have applied it as-is for now.
>
> --
> Eric Botcazou