[AArch64] Change aarch64 vector cost to match vectorizer

On 03/08/15 17:26, James Greenhalgh wrote:
> On Mon, Jul 27, 2015 at 02:22:41PM +0100, Pawel Kupidura wrote:
>> diff --git a/gcc/ChangeLog b/gcc/ChangeLog
>> index 10df325..ffafc3f 100644
>> --- a/gcc/ChangeLog
>> +++ b/gcc/ChangeLog
>> @@ -1,3 +1,7 @@
>> +2015-07-27  Pawel Kupidura<pawel.kupidura@arm.com>
>
> Two spaces between your name and your email address, like so:
>
> 2015-07-27  Pawel Kupidura<pawel.kupidura@arm.com>
>
>> +
>> +    * config/aarch64/aarch64.c: Changed inner loop statement cost
>> +    to be consistent with vectorizer code.
>> +
>
> s/Changed/Change
>
>>    2015-07-26  Uros Bizjak<ubizjak@gmail.com>
>>
>>        * config/alpha/alpha.c: Use SUBREG_P predicate.
>> diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
>> index 020f63c..3b6f8c5 100644
>> --- a/gcc/config/aarch64/aarch64.c
>> +++ b/gcc/config/aarch64/aarch64.c
>> @@ -7079,15 +7079,9 @@ aarch64_add_stmt_cost (void *data, int count,
>> enum vect_cost_for_stmt kind,
>>
>>          /* Statements in an inner loop relative to the loop being
>>         vectorized are weighted more heavily.  The value here is
>> -     a function (linear for now) of the loop nest level.  */
>> +     arbitrary and could potentially be improved with analysis.  */
>
> Your mail client has mangled the tabs in this diff, so the patch will
> not apply in this form. Could you try posting again having resolved the
> issues with your mail client?
>
>>          if (where == vect_body&&  stmt_info&&  stmt_in_inner_loop_p
>> (stmt_info))
>> -    {
>> -      loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_info);
>> -      struct loop *loop =  LOOP_VINFO_LOOP (loop_info);
>> -      unsigned nest_level = loop_depth (loop);
>> -
>> -      count *= nest_level;
>> -    }
>> +    count *= 50; /* FIXME */
>
> Likewise here.
>
> Thanks,
> James
>
Hi,

I'm sorry about the issues with formatting, it should be fixed now. 
Here's corrected version with diff to current trunk.

Thanks,
Pawel Kupidura

        retval = (unsigned) (count * stmt_cost);
        cost[where] += retval;

On Tue, Aug 04, 2015 at 11:06:11AM +0100, Pawel Kupidura wrote:
> Hi,
> 
> I'm sorry about the issues with formatting, it should be fixed now. 
> Here's corrected version with diff to current trunk.

Hi Pawel,

I'm still having trouble getting this patch to apply, I'm not sure whether
it is the format=flowed in your mail headers, or the quoted-printable
encoding, or something else. Certainly when I open your emails I see :

        if (where == vect_body && stmt_info && stmt_in_inner_loop_p 
(stmt_info))

The content of the patch is OK to commit, but it would be good to
have a copy on list that can be easily applied.

Thanks,
James

> diff --git a/gcc/ChangeLog b/gcc/ChangeLog
> index fdc4a7e..d1c6663 100644
> --- a/gcc/ChangeLog
> +++ b/gcc/ChangeLog
> @@ -1,3 +1,7 @@
> +2015-08-04  Pawel Kupidura  <pawel.kupidura@arm.com>
> +	* config/aarch64/aarch64.c: Change inner loop statement cost
> +	to be consistent with other targets.
> +
>   2015-08-03  Abe Skolnik  <a.skolnik@samsung.com>
> 
>   	* tree-if-conv.c: Fix various typos in comments.
> diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
> index 2b1ae36..173a385 100644
> --- a/gcc/config/aarch64/aarch64.c
> +++ b/gcc/config/aarch64/aarch64.c
> @@ -7086,15 +7086,9 @@ aarch64_add_stmt_cost (void *data, int count, 
> enum vect_cost_for_stmt kind,
> 
>         /* Statements in an inner loop relative to the loop being
>   	 vectorized are weighted more heavily.  The value here is
> -	 a function (linear for now) of the loop nest level.  */
> +	 arbitrary and could potentially be improved with analysis.  */
>         if (where == vect_body && stmt_info && stmt_in_inner_loop_p 
> (stmt_info))
> -	{
> -	  loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_info);
> -	  struct loop *loop =  LOOP_VINFO_LOOP (loop_info);
> -	  unsigned nest_level = loop_depth (loop);
> -
> -	  count *= nest_level;
> -	}
> +	count *= 50; /*  FIXME  */
> 
>         retval = (unsigned) (count * stmt_cost);
>         cost[where] += retval;