This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH][AArch64] Replace insn to zero up DF register
- From: Andrew Pinski <pinskia at gmail dot com>
- To: Evandro Menezes <e dot menezes at samsung dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>, Marcus Shawcroft <Marcus dot Shawcroft at arm dot com>, Kyrill Tkachov <kyrylo dot tkachov at arm dot com>
- Date: Tue, 20 Oct 2015 22:27:15 +0800
- Subject: Re: [PATCH][AArch64] Replace insn to zero up DF register
- Authentication-results: sourceware.org; auth=none
- References: <56257F53 dot 2000905 at samsung dot com> <CA+=Sn1kLDZAPBozErJtTt5HYnf3rmSyGsJhL2D8oW_v1vpCf6g at mail dot gmail dot com> <CA+=Sn1nUfBCsnFdh1WDGZCp6WYx63ns7naQUFDjECovYTWVZdQ at mail dot gmail dot com>
On Tue, Oct 20, 2015 at 7:59 AM, Andrew Pinski <pinskia@gmail.com> wrote:
> On Tue, Oct 20, 2015 at 7:51 AM, Andrew Pinski <pinskia@gmail.com> wrote:
>> On Tue, Oct 20, 2015 at 7:40 AM, Evandro Menezes <e.menezes@samsung.com> wrote:
>>> In the existing targets, it seems that it's always faster to zero up a DF
>>> register with "movi %d0, #0" instead of "fmov %d0, xzr".
>>
>> I think for ThunderX 1, this change will not make a difference. So I
>> am neutral on this change.
>
> Actually depending on fmov is decoded in our pipeline, this change
> might actually be worse. Currently fmov with an immediate is 1 cycle
> while movi is two cycles. Let me double check how internally on how
> it is decoded and if it is 1 cycle or two.
Ok, my objections are removed as I talked with the architectures here
at Cavium and using movi is better in this case.
Thanks,
Andrew
>
> Thanks,
> Andrew
>
>>
>> Thanks,
>> Andrew
>>
>>>
>>> This patch modifies the respective pattern.
>>>
>>> Please, commit if it's alright.
>>>
>>> Thank you,
>>>
>>> --
>>> Evandro Menezes
>>>