This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Simple optimization for MASK_STORE.
- From: Richard Biener <richard dot guenther at gmail dot com>
- To: Jeff Law <law at redhat dot com>,Yuri Rumyantsev <ysrumyan at gmail dot com>
- Cc: gcc-patches <gcc-patches at gcc dot gnu dot org>,Igor Zamyatin <izamyatin at gmail dot com>
- Date: Fri, 08 May 2015 21:16:48 +0200
- Subject: Re: [PATCH] Simple optimization for MASK_STORE.
- Authentication-results: sourceware.org; auth=none
- References: <CAEoMCqRmV48Ytdew0azyTQWZcmfFmjX-JaLtYUz8Q3wejL2RnQ at mail dot gmail dot com> <CAFiYyc38QMSXL058QuV0TZMAku=Ur0FXhF9TEm2Lp7C_HHmWLg at mail dot gmail dot com> <554D03C3 dot 4080700 at redhat dot com>
On May 8, 2015 8:43:15 PM GMT+02:00, Jeff Law <email@example.com> wrote:
>On 05/08/2015 03:27 AM, Richard Biener wrote:
>> On Wed, May 6, 2015 at 4:04 PM, Yuri Rumyantsev <firstname.lastname@example.org>
>>> Hi All,
>>> Here is a patch which gives us significant speed-up on HASWELL for
>>> test containing masked stores. The main goal of that patch is
>>> to avoid HW hazard for maskmove instructions through inserting
>>> additional check on zero mask and putting all masked store
>>> into separate block on false edge.All MASK_STORE statements having
>>> same mask put into one block. Any comments will be appreciate.
>> Hmm. I'm not very happy with this "optimization" happening at the
>> GIMPLE level - it feels more like a mdreorg thing...
>I haven't looked at the patch, but it sounds like just specialization
>using a runtime check to select between the implementations which is
>fairly natural to do at the gimple level.
Well, sure. Which is why I suggested a un-ifcvt pass instead.
Anyway, the implementation has multiple issues.