This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH Atom][PR middle-end/44382] Tree reassociation improvement
- From: Richard Guenther <richard dot guenther at gmail dot com>
- To: Uros Bizjak <ubizjak at gmail dot com>
- Cc: Ilya Enkovich <enkovich dot gnu at gmail dot com>, gcc-patches at gcc dot gnu dot org
- Date: Fri, 2 Sep 2011 15:06:47 +0200
- Subject: Re: [PATCH Atom][PR middle-end/44382] Tree reassociation improvement
- References: <CAMbmDYYhfpA845e7TA50coQxBrn8-ZgZCVrx6DeX-D8x4MkK7Q@mail.gmail.com> <20110713232837.GA31809@hungry-tiger.westford.ibm.com> <CAFiYyc31XGVFH9NiZuhmeBSg48Uju=ZYVR44cTWnDEbB0TEe4g@mail.gmail.com> <CAFiYyc3gZw9ouZb-4CmVWd9dtYnSMQxY-xZGOoe3txoaev3eCQ@mail.gmail.com> <CAMbmDYa1hwR1WD2kaFW0=x5CbXLt10vs0WcHVgZKjEUeQjAang@mail.gmail.com> <CAFiYyc0fj7gSbCrpPTXs2NjCaG8LOKKReaBvxnv8ZKUEQqaHmg@mail.gmail.com> <CAMbmDYYFckMTNEtJ0F+Mpr9EA5Qf-v-hBYzHHKBD6QZ=NC9XTQ@mail.gmail.com> <CAMbmDYZW_SX08bzQ9z1OdK5fEADc1oYD9GB8GYB8cYjBe+Svpw@mail.gmail.com> <CAFiYyc0aewXE8kxzdiMTfSMPrX=udWMRmM4iVaVRK9kQ_FBGzA@mail.gmail.com> <CAMbmDYb9m6iV3ZS1GS-_HSHezum1tB8G78mg_VYuah3wkA-cAA@mail.gmail.com> <CAFiYyc0t-einW82V6sYLzuwTm=WG1Ej3ngi_dogC4ZS02TtdRg@mail.gmail.com> <CAMbmDYaTTs4HndicGdQkJ4WwvzGMfF21FRgqSj7Rc31OhXwO0A@mail.gmail.com> <CAMbmDYZ+JU2=TJhP5Hdpa78NPOApf47oUbj0yAXrw4cZwP9ivw@mail.gmail.com> <CAFiYyc0opH76_+RRBcycJ7kycx92xfAPPR92hEn+aDM18jY+MQ@mail.gmail.com> <CAMbmDYYc_eYGZ2w+E+rY_zvERUVrhfXkMpnAjxCzKT9_s7-HAA@mail.gmail.com> <CAFiYyc06f5m2njPh3i1kbXc=jUAviUfv3dRPZ3NFxta3feX=mA@mail.gmail.com> <CAMbmDYYuwh9d2c1yBjreKRouU3QLmN5qD+yMKeVr3XqqWGKjMw@mail.gmail.com> <CAFULd4YZtfWPC1qD=XmtXX+QHCBap7COPq-3TVZ4KrW2qMBQUQ@mail.gmail.com>
On Fri, Sep 2, 2011 at 2:52 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
> On Thu, Sep 1, 2011 at 12:27 PM, Ilya Enkovich <enkovich.gnu@gmail.com> wrote:
>>>
>>> this seems to not allow cycles_best to drop with lower width, but
>>> that it can't should be an implementation detail of get_required_cycles.
>>> To make it not so, can you add a comment before the loop, like
>>>
>>> ?/* get_required_cycles is monotonically increasing with lower width
>>> ? ? so we can perform a binary search for the minimal width that still
>>> ? ? results in the optimal cycle count. ?*/
>>>
>>
>> Fixed. Thanks!
>>
>>>
>>> With the above change the non-x86 specifc parts are ok. ?Please get
>>> approval for them from a x86 maintainer.
>>>
>>
>> Could please someone review x86 part?
>
> I assume that you need to split tune attribute to int and FP part to
> handle reassociation for other targets, since Atom handles both in the
> same way.
>
> Please also describe function return value in the comment (and perhaps
> in documentation, too).
>
> OK with this addition.
Btw, I would expect integer add and integer multiply to have different
settings for some targets which would mean splitting this up even
further ...
Richard.
> Thanks,
> Uros.
>