This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH/AArch64 2/3] * config/tc-aarch64.h (MAX_MEM_FOR_RS_ALIGN_CODE): Increase to 63.
- From: Andrew Pinski <pinskia at gmail dot com>
- To: Philipp Tomsich <philipp dot tomsich at theobroma-systems dot com>
- Cc: binutils <binutils at sourceware dot org>
- Date: Thu, 5 Dec 2013 17:53:17 -0800
- Subject: Re: [PATCH/AArch64 2/3] * config/tc-aarch64.h (MAX_MEM_FOR_RS_ALIGN_CODE): Increase to 63.
- Authentication-results: sourceware.org; auth=none
- References: <1386181767-10323-1-git-send-email-philipp dot tomsich at theobroma-systems dot com> <1386181767-10323-2-git-send-email-philipp dot tomsich at theobroma-systems dot com> <CA+=Sn1nxGT+x9JgJo=B6iOBKcOz0xWiF5nf2t=6vgVz+eMSpDQ at mail dot gmail dot com>
On Thu, Dec 5, 2013 at 5:52 PM, Andrew Pinski <pinskia@gmail.com> wrote:
> On Wed, Dec 4, 2013 at 10:29 AM, Philipp Tomsich
> <philipp.tomsich@theobroma-systems.com> wrote:
>> While working on various benchmarks, we've seen improvements from aligning
>> function entries and hot loops to a cache-line boundary. For this reason we
>> need to increase the permissible maximum code alignment to 64 bytes.
>
> I think this should be increased to even 128 bytes as Cavium's thunder
> has 128byte cache line sizes.
> Also I think you should at least update this code to be closer to what
> the arm code does as in
> https://sourceware.org/ml/binutils/2009-06/msg00289.html .
Also it might be better to remove this limit if possible. See the
thread there on why there was a limit for ARM.
Thanks,
Andrew Pinski
>
> Thanks,
> Andrew
>
>>
>> Thanks,
>> Philipp Tomsich.
>>
>> * config/tc-aarch64.h (MAX_MEM_FOR_RS_ALIGN_CODE): Increase to 63.
>>
>> ---
>> gas/config/tc-aarch64.h | 4 ++--
>> 1 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/gas/config/tc-aarch64.h b/gas/config/tc-aarch64.h
>> index 74a81d6..ee60525 100644
>> --- a/gas/config/tc-aarch64.h
>> +++ b/gas/config/tc-aarch64.h
>> @@ -116,8 +116,8 @@ void aarch64_copy_symbol_attributes (symbolS *, symbolS *);
>>
>> #define TC_CONS_FIX_NEW cons_fix_new_aarch64
>>
>> -/* Max code alignment is 32 bytes */
>> -#define MAX_MEM_FOR_RS_ALIGN_CODE 31
>> +/* Max code alignment is 64 bytes */
>> +#define MAX_MEM_FOR_RS_ALIGN_CODE 63
>>
>> /* For frags in code sections we need to record whether they contain
>> code or data. */
>> --
>> 1.7.2.5
>>