This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: [PATCH 1/3] Add a fast path for C rd/wrlock v2

From: Jeff Law <law at redhat dot com>
To: Andi Kleen <andi at firstfloor dot org>
Cc: "Carlos O'Donell" <carlos at redhat dot com>, libc-alpha at sourceware dot org, Andi Kleen <ak at linux dot intel dot com>
Date: Mon, 09 Jun 2014 13:16:12 -0600
Subject: Re: [PATCH 1/3] Add a fast path for C rd/wrlock v2
Authentication-results: sourceware.org; auth=none
References: <1399412209-28245-1-git-send-email-andi at firstfloor dot org> <1399412209-28245-2-git-send-email-andi at firstfloor dot org> <53960741 dot 4060306 at redhat dot com>

On 06/09/14 13:13, Carlos O'Donell wrote:

On 05/06/2014 05:36 PM, Andi Kleen wrote:

From: Andi Kleen <ak@linux.intel.com>

One difference of the C versions to the assembler wr/rdlock
is that the C compiler saves some registers which are unnecessary
for the fast path in the prologue of the functions. Split the
uncontended fast path out into a separate function. Only when contention is
detected is the full featured function called. This makes
the fast path code (nearly) identical to the assembler version,
and gives uncontended performance within a few cycles.


Sounds like the best of both worlds. Thanks for the cleanup.

Andi -- any change you could file a gcc bug for that? GCC knows how todo a limited form of shrink-wrapping as well as function splitting tooptimize these kinds of situations. If it's not working as it should,obviously we'd like to know and try to get it fixed.


Thanks,
jeff

References:
- Re: [PATCH 1/3] Add a fast path for C rd/wrlock v2
  - From: Carlos O'Donell

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]