This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Lock elision test results
- From: Andi Kleen <andi at firstfloor dot org>
- To: Torvald Riegel <triegel at redhat dot com>
- Cc: vogt at linux dot vnet dot ibm dot com, libc-alpha at sourceware dot org, andi <andi at firstfloor dot org>
- Date: Wed, 3 Jul 2013 16:15:03 +0200
- Subject: Re: Lock elision test results
- References: <20130614102653 dot GA21917 at linux dot vnet dot ibm dot com> <1372767484 dot 22198 dot 4505 dot camel at triegel dot csb> <20130703065355 dot GA4522 at linux dot vnet dot ibm dot com> <1372849432 dot 14172 dot 20 dot camel at triegel dot csb>
First I must say I'm pretty sceptic on the idea of assembling a
micro benchmark suite for lock elision. It is unlikely to give
you a true picture of the performance, as small programs tend to not
model real applications well for something as complex as transactions.
On Wed, Jul 03, 2013 at 01:03:52PM +0200, Torvald Riegel wrote:
> Andi, do you have measurements of how costly aborts are on Haswell, or
> can you give a rough estimation (i.e., not counting the work one wasted,
Normally the mis-speculated work should dominate, but it's not always
wasted (e.g. it may serve as a prefetch)
> just the looking at the rollback and jump back to the xbegin)? I'm
It depends on what the transaction was doing, in particular how much
synchronization the abort needs. But it's typically within one
order of magnitude of a commit.
-Andi
--
ak@linux.intel.com -- Speaking for myself only.