Re: [PATCH] Mantaining turnstile aligned to 128 bytes in i386 CPUs

From: Bruce Evans <bde_at_zeta.org.au>
Date: Wed, 17 Jan 2007 15:50:41 +1100 (EST)
On Wed, 17 Jan 2007, Ivan Voras wrote:

> Kip Macy wrote:
>>> Maybe even someone finds a way to get optimized versions of memcpy in
>>> the kernel :)
>
>> It makes a huge difference in a proprietary file serving appliance
>> that I know of.
>
> Beneficial difference?

Heheh.

>> However, past measurements on FreeBSD have supposedly
>> indicated that it isn't that big win as a result of increased context
>> switch time.

No, they indicated that the win is not very large (sometimes negative),
and is very machine dependent.  E.g., it is a small pessimization all 64
bit i386's running 64-bit mode -- that's just all i386's you would want
to buy now.  On other CPU classes:

P2 (my old Celeron): +- epsilon difference
P3 (freefall): +- epsilon difference
P4 (nosedive's Xeon): movdqa 17% faster than movsl, but all other cached
    moves slower using MMX or SSE[1-2]; movnt with block prefetch 60% faster
    than movsl with no prefetch, but < 5% faster with no prefetch for both.
AXP: (my 5 year old system with a newer CPU): movq through MMX is 60%
    faster than movsl for cached moves, but movdqa through XMM is only 4%
    faster.  movnt with block prefetch is 155% faster than movsl with no
    prefetch, and 73% faster with no prefetch for both.
A64 in 32-bit mode: in between P4 and AXP (closer to AXP).  movsl doesn't
    lose by so much, and prefetchnta actually works so block prefetch is
    not needed and there is a better chance of prefetching helping more
    than benchmarks.

Bruce
Received on Wed Jan 17 2007 - 03:50:45 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:04 UTC