This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Improve performance for Haswell family.
- From: Yuri Rumyantsev <ysrumyan at gmail dot com>
- To: Uros Bizjak <ubizjak at gmail dot com>
- Cc: gcc-patches <gcc-patches at gcc dot gnu dot org>, Igor Zamyatin <izamyatin at gmail dot com>
- Date: Fri, 14 Aug 2015 18:09:51 +0300
- Subject: Re: [PATCH] Improve performance for Haswell family.
- Authentication-results: sourceware.org; auth=none
- References: <CAEoMCqSa3z0B1b6UzkKThGzcaq2aX7pewyRPj8cgSNwL38ELyg at mail dot gmail dot com> <CAFULd4Z8+TVkAwQeGk349LTyfYTBwBDAGcGLMKryxRjXDKBT4w at mail dot gmail dot com>
Thanks Uros.
Will it be fine for you if I split core2.md to instruction reservation
definition for core2 (old core2.md got old cpu core2) and new
nehalem.md (my changes for core2.md but for cpu nehalem only)?
Thanks.
Yuri.
2015-08-14 13:20 GMT+03:00 Uros Bizjak <ubizjak@gmail.com>:
> On Fri, Aug 14, 2015 at 11:44 AM, Yuri Rumyantsev <ysrumyan@gmail.com> wrote:
>> Hi All,
>>
>> Here is patch which contains
>> (1) modifying of core2.md to conform Haswell pipeline and adding of
>> missed instruction reservation for instructions with vector operands.
>> (2) increase reassociation width for float-point operations for
>> Haswell family and 64-bit target, this allows to get +10% speedup on
>> spec2006/410.bwaves for Skylake processor.
>>
>> Bootstrap and regression testing did not show any new failures.
>> Is it OK for trunk?
>>
>> ChangeLog:
>> 2015-08-14 Yuri Rumyantsev <ysrumyan@gmail.com>
>>
>> * config/i386/core2.md: Confrom instruction reservation to Haswell
>> pipeline.
>> * config/i386/i386.c (ix86_reassociation_width): Increase reassociation
>> width for float-point operations.
>
> The i386.c change is OK.
>
> Other that this, your patch substantially changes core2.md, and I'm
> afraid this would hurt pipeline description of older processors. IMO,
> it is better to introduce new haswell.md file and put the whole
> haswell description here.
>
> Uros.