Sorry for answering out of thread, but the message is long gone.
I just remembered very old x86 code of mine.
More than fifteen years ago it was still faster to go for 32-bit
wide testing when more than 20 bytes had to be compared, and to
redo a short byte loop to work around the fact that the car
Hi,
Andrew Doran writes:
> Hi,
>
> Change backed out. Sorry about the disruption.
Thank you.
The back-out works fine for me.
> Andrew
>
> On Thu, Jan 16, 2020 at 05:30:20PM +0900, Ryo ONODERA wrote:
>> Hi,
>>
>> pkgsrc/www/firefox and mail/notmuch are also
>> broken after this commit.
>>
>>
Hi,
Change backed out. Sorry about the disruption.
Andrew
On Thu, Jan 16, 2020 at 05:30:20PM +0900, Ryo ONODERA wrote:
> Hi,
>
> pkgsrc/www/firefox and mail/notmuch are also
> broken after this commit.
>
>
> On January 16, 2020 5:23:47 PM GMT+09:00, Kamil Rytarowski
> wrote:
> >On 15.01.20
Hi,
pkgsrc/www/firefox and mail/notmuch are also
broken after this commit.
On January 16, 2020 5:23:47 PM GMT+09:00, Kamil Rytarowski wrote:
>On 15.01.2020 11:56, Andrew Doran wrote:
>> Module Name: src
>> Committed By:ad
>> Date:Wed Jan 15 10:56:49 UTC 2020
>>
>> Modif
On 15.01.2020 11:56, Andrew Doran wrote:
> Module Name: src
> Committed By: ad
> Date: Wed Jan 15 10:56:49 UTC 2020
>
> Modified Files:
> src/common/lib/libc/arch/x86_64/string: bcmp.S memcmp.S
>
> Log Message:
> Rewrite bcmp() & memcmp() to not use REP CMPS. Seems about 5-10x fas
