[9fans] LOCK XADD for i386 incref/decref

[9fans] LOCK XADD for i386 incref/decref

[Venkatesh Srinivas]

Hi,

A few months ago, I added a patch to inferno-npe to use LOCK XADD
instead of the current lock/add/unlock sequence for incref and decref:
(http://code.google.com/p/inferno-npe/source/detail?r=b83540e1e77e62a19cbd21d2eb54d43d338716a5
and http://code.google.com/p/inferno-npe/source/detail?r=82f13e6755218ecb7dec0f1392b2eb8bfe0bb2c7).
On a 2.66 GHz Core 2 Duo, we noticed a full 14% reduction in runtime
(9.6s vs 11.2s) of the thread-ring test. Similar performance gains
were noticed on a Core i7 machine, but I no longer have the numbers
handy.

Perhaps this change is interesting for Plan 9...

-- vs

Re: [9fans] LOCK XADD for i386 incref/decref

[Federico G. Benavento]

sorry, but why in the compiler and not as a library function like in libthread?

http://plan9.bell-labs.com/sources/plan9/sys/src/libthread/xinc386.s

On Wed, Apr 28, 2010 at 4:16 PM, Venkatesh Srinivas
<me@endeavour.zapto.org> wrote:
> Hi,
>
> A few months ago, I added a patch to inferno-npe to use LOCK XADD
> instead of the current lock/add/unlock sequence for incref and decref:
> (http://code.google.com/p/inferno-npe/source/detail?r=b83540e1e77e62a19cbd21d2eb54d43d338716a5
> and http://code.google.com/p/inferno-npe/source/detail?r=82f13e6755218ecb7dec0f1392b2eb8bfe0bb2c7).
> On a 2.66 GHz Core 2 Duo, we noticed a full 14% reduction in runtime
> (9.6s vs 11.2s) of the thread-ring test. Similar performance gains
> were noticed on a Core i7 machine, but I no longer have the numbers
> handy.
>
> Perhaps this change is interesting for Plan 9...
>
> -- vs
>
>



--
Federico G. Benavento

Re: [9fans] LOCK XADD for i386 incref/decref

[Federico G. Benavento]

sorry for the noise, I should rest a bit after this

On Wed, Apr 28, 2010 at 11:39 PM, Federico G. Benavento
<benavento@gmail.com> wrote:
> sorry, but why in the compiler and not as a library function like in libthread?
>
> http://plan9.bell-labs.com/sources/plan9/sys/src/libthread/xinc386.s
>
> On Wed, Apr 28, 2010 at 4:16 PM, Venkatesh Srinivas
> <me@endeavour.zapto.org> wrote:
>> Hi,
>>
>> A few months ago, I added a patch to inferno-npe to use LOCK XADD
>> instead of the current lock/add/unlock sequence for incref and decref:
>> (http://code.google.com/p/inferno-npe/source/detail?r=b83540e1e77e62a19cbd21d2eb54d43d338716a5
>> and http://code.google.com/p/inferno-npe/source/detail?r=82f13e6755218ecb7dec0f1392b2eb8bfe0bb2c7).
>> On a 2.66 GHz Core 2 Duo, we noticed a full 14% reduction in runtime
>> (9.6s vs 11.2s) of the thread-ring test. Similar performance gains
>> were noticed on a Core i7 machine, but I no longer have the numbers
>> handy.
>>
>> Perhaps this change is interesting for Plan 9...
>>
>> -- vs
>>
>>
>
>
>
> --
> Federico G. Benavento
>



--
Federico G. Benavento

Re: [9fans] LOCK XADD for i386 incref/decref

[erik quanstrom]

On Wed Apr 28 15:18:01 EDT 2010, me@endeavour.zapto.org wrote:
> Hi,
>
> A few months ago, I added a patch to inferno-npe to use LOCK XADD
> instead of the current lock/add/unlock sequence for incref and decref:
> (http://code.google.com/p/inferno-npe/source/detail?r=b83540e1e77e62a19cbd21d2eb54d43d338716a5
> and http://code.google.com/p/inferno-npe/source/detail?r=82f13e6755218ecb7dec0f1392b2eb8bfe0bb2

sorry for the really long lag.  finally clearing out my
my inbox.

this patch is incorrect because incref() and decref()
assume that they are reusing the same lock that protects
the whole structure.  this incref is equivalent to
	lock(structure)
	structure->ref++;
	unlock(structure)
and not something like lock xadd.

- erik