Patchwork [PR50572] Tune loop alignment for Atom

login
register
mail settings
Submitter Sergey Ostanevich
Date Oct. 19, 2011, 1:09 p.m.
Message ID <CAGYS_T+F61cZL2snCj5ZJXsVe6+MQodPO+pZVCP-OcHmx3q9iQ@mail.gmail.com>
Download mbox | patch
Permalink /patch/120625/
State New
Headers show

Comments

Sergey Ostanevich - Oct. 19, 2011, 1:09 p.m.
On Wed, Oct 19, 2011 at 4:46 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
> On Wed, Oct 19, 2011 at 2:26 PM, Sergey Ostanevich <sergos.gnu@gmail.com> wrote:
>
>>> You can just change the default in processor_target_table.
>>>
>>> Uros.
>>>
>>
>> Will it be applicable during optimizations for size? It will hurt,
>> although not much (see PR).
>
> Looking at the code, I'd say that we don't handle -Os in different way.
>
>> New patch is below. Ok for trunk as obvious?
>>
>> Sergos
>>
>> 2011-10-19  Sergey Ostanevich  <sergos.gnu@gmail.com>
>>
>>        * gcc/config/i386/i386.c (ix86_option_override_internal): use loop
>>        align by 16 bytes for Atom platform
>
> Please update ChangeLog, like:
>
> * gcc/config/i386/i386.c (processor_target_table): Change Atom
> align_loop_max_skip to 15.
>
>> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
>> index 2c53423..8c60086 100644
>> --- a/gcc/config/i386/i386.c
>> +++ b/gcc/config/i386/i386.c
>> @@ -2596,7 +2596,7 @@ static const struct ptt
>> processor_target_table[PROCESSOR_max] =
>>   {&bdver1_cost, 32, 24, 32, 7, 32},
>>   {&bdver2_cost, 32, 24, 32, 7, 32},
>>   {&btver1_cost, 32, 24, 32, 7, 32},
>> -  {&atom_cost, 16, 7, 16, 7, 16}
>> +  {&atom_cost, 16, 15, 16, 7, 16}
>>  };
>
>
> OK.
>
> Thanks,
> Uros.
>

Thanks for comments!
I double checked: for -Os there's no .p2align appeared.
For -O2 I see ".p2align 4,,15" instead of ".p2align 4,,7", as expected.

Can someone commit it please?

Regards,
Sergos


2011-10-18  Sergey Ostanevich <sergos.gnu@gmail.com>

	* gcc/config/i386/i386.c (processor_target_table): Change Atom
	align_loops_max_skip to 15.

 static const char *const cpu_names[TARGET_CPU_DEFAULT_max] =
H.J. Lu - Oct. 19, 2011, 2:59 p.m.
On Wed, Oct 19, 2011 at 6:09 AM, Sergey Ostanevich <sergos.gnu@gmail.com> wrote:
> On Wed, Oct 19, 2011 at 4:46 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
>> On Wed, Oct 19, 2011 at 2:26 PM, Sergey Ostanevich <sergos.gnu@gmail.com> wrote:
>>
>>>> You can just change the default in processor_target_table.
>>>>
>>>> Uros.
>>>>
>>>
>>> Will it be applicable during optimizations for size? It will hurt,
>>> although not much (see PR).
>>
>> Looking at the code, I'd say that we don't handle -Os in different way.
>>
>>> New patch is below. Ok for trunk as obvious?
>>>
>>> Sergos
>>>
>>> 2011-10-19  Sergey Ostanevich  <sergos.gnu@gmail.com>
>>>
>>>        * gcc/config/i386/i386.c (ix86_option_override_internal): use loop
>>>        align by 16 bytes for Atom platform
>>
>> Please update ChangeLog, like:
>>
>> * gcc/config/i386/i386.c (processor_target_table): Change Atom
>> align_loop_max_skip to 15.
>>
>>> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
>>> index 2c53423..8c60086 100644
>>> --- a/gcc/config/i386/i386.c
>>> +++ b/gcc/config/i386/i386.c
>>> @@ -2596,7 +2596,7 @@ static const struct ptt
>>> processor_target_table[PROCESSOR_max] =
>>>   {&bdver1_cost, 32, 24, 32, 7, 32},
>>>   {&bdver2_cost, 32, 24, 32, 7, 32},
>>>   {&btver1_cost, 32, 24, 32, 7, 32},
>>> -  {&atom_cost, 16, 7, 16, 7, 16}
>>> +  {&atom_cost, 16, 15, 16, 7, 16}
>>>  };
>>
>>
>> OK.
>>
>> Thanks,
>> Uros.
>>
>
> Thanks for comments!
> I double checked: for -Os there's no .p2align appeared.
> For -O2 I see ".p2align 4,,15" instead of ".p2align 4,,7", as expected.
>
> Can someone commit it please?
>
> Regards,
> Sergos
>
>
> 2011-10-18  Sergey Ostanevich <sergos.gnu@gmail.com>
>
>        * gcc/config/i386/i386.c (processor_target_table): Change Atom
^^^^^^^^^^^^^^^^^^^^^^^^ Please remove gcc/.

>        align_loops_max_skip to 15.
>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index 2c53423..8c60086 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -2596,7 +2596,7 @@ static const struct ptt
> processor_target_table[PROCESSOR_max] =
>   {&bdver1_cost, 32, 24, 32, 7, 32},
>   {&bdver2_cost, 32, 24, 32, 7, 32},
>   {&btver1_cost, 32, 24, 32, 7, 32},
> -  {&atom_cost, 16, 7, 16, 7, 16}
> +  {&atom_cost, 16, 15, 16, 7, 16}
>  };
>
>  static const char *const cpu_names[TARGET_CPU_DEFAULT_max] =
>
H.J. Lu - Oct. 19, 2011, 4:49 p.m.
On Wed, Oct 19, 2011 at 9:30 AM, Sergey Ostanevich <sergos.gnu@gmail.com> wrote:

> 2011-10-18  Sergey Ostanevich <sergos.gnu@gmail.com>
>
>    * config/i386/i386.c (processor_target_table): Change Atom
>    align_loops_max_skip to 15.

Please add PR # in ChangeLog.

>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index 2c53423..8c60086 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -2596,7 +2596,7 @@ static const struct ptt
> processor_target_table[PROCESSOR_max] =
>   {&bdver1_cost, 32, 24, 32, 7, 32},
>   {&bdver2_cost, 32, 24, 32, 7, 32},
>   {&btver1_cost, 32, 24, 32, 7, 32},
> -  {&atom_cost, 16, 7, 16, 7, 16}
> +  {&atom_cost, 16, 15, 16, 7, 16}
> };
>
> static const char *const cpu_names[TARGET_CPU_DEFAULT_max] =

Please provide a patch which can be applied.  Cut/paste doesn't create
a working patch.  Please attach it.

Patch

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 2c53423..8c60086 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -2596,7 +2596,7 @@  static const struct ptt
processor_target_table[PROCESSOR_max] =
   {&bdver1_cost, 32, 24, 32, 7, 32},
   {&bdver2_cost, 32, 24, 32, 7, 32},
   {&btver1_cost, 32, 24, 32, 7, 32},
-  {&atom_cost, 16, 7, 16, 7, 16}
+  {&atom_cost, 16, 15, 16, 7, 16}
 };