This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH][X86_64] Set bit_Fast_Unaligned_Load for Excavator family CPU's
- From: OndÅej BÃlka <neleai at seznam dot cz>
- To: "Pawar, Amit" <Amit dot Pawar at amd dot com>
- Cc: "libc-alpha at sourceware dot org" <libc-alpha at sourceware dot org>
- Date: Wed, 16 Dec 2015 13:08:10 +0100
- Subject: Re: [PATCH][X86_64] Set bit_Fast_Unaligned_Load for Excavator family CPU's
- Authentication-results: sourceware.org; auth=none
- References: <SN1PR12MB07330EBFA0ED52D659F4F46C97EF0 at SN1PR12MB0733 dot namprd12 dot prod dot outlook dot com>
On Wed, Dec 16, 2015 at 10:03:25AM +0000, Pawar, Amit wrote:
> For latest AMD Excavator family 0x15h CPU's, it is found that by running GLIBC benchtest SSE2_Unaligned based implementations are performing faster compare to SSE2 based implementations for routines mentioned below. Based on this observation bit_Fast_Unaligned_Load flag is set for Excavator family CPU's. This makes SSE2_Unaligned implementations as default for these routines.
>
> Routines: strcmp, strcat, strncat, stpcpy, stpncpy, strcpy, strncpy and strstr.
>
> Is attached patch OK for trunk ?
>
Should be ok, while benchtests are as bad as random noise for measuring
performance I know that now amd processors are not optimized at all, I
had some patches for to improve that situation that I could ping 11th
time.
Particulary unaligned path is also better in bulldozer and these
routines are good only for older cpus of phemonII era. I recalled there
are some tweaks that I had for bulldozer but I don't remember these.