What's the current status of type specialization?

Yes, it should work. Development has slowed down quite a bit since I am writing my thesis, and that's part of the reason specialization is still a -Y option (in other words, use on your own risk). It is for the most part feature-complete, so please file tickets with the bugs you find, that would help us a lot.
thanks,iulian

On Sat, Feb 20, 2010 at 4:40 PM, Heiko Seeberger <heiko.seeberger@googlemail.com> wrote:

Hi,
Today I tried to implement a demo for @specialized. The compiler did its job, but at runtime I got an AbstractMethodError. What's the current status of type specialization? Should it work?

Heiko

Work: weiglewilczek.com
Blog: heikoseeberger.name
Follow me: twitter.com/hseeberger
OSGi on Scala: scalamodules.org
Lift, the simply functional web framework: liftweb.net

--
« Je déteste la montagne, ça cache le paysage »
Alphonse Allais

Iulian,
Thanks for your quick response!

On 20 February 2010 17:20, Iulian Dragos <iulian.dragos@epfl.ch> wrote:

Yes, it should work.

Sounds good ;-) 

Development has slowed down quite a bit since I am writing my thesis, and that's part of the reason specialization is still a -Y option (in other words, use on your own risk). It is for the most part feature-complete, so please file tickets with the bugs you find, that would help us a lot.

Did it: https://lampsvn.epfl.ch/trac/scala/ticket/3085
Thanks,
Heiko
Work: weiglewilczek.com
Blog: heikoseeberger.name
Follow me: twitter.com/hseeberger
OSGi on Scala: scalamodules.org
Lift, the simply functional web framework: liftweb.net

Iulian,
You closed the ticket commenting "(In [20949]) Specialized types are not substituted inside type arguments.". I am not sure if I fully understand => Could you please explain?
Thank you,
Heiko
On 21 February 2010 10:05, Heiko Seeberger <heiko.seeberger@googlemail.com> wrote:

Iulian,
Thanks for your quick response!

On 20 February 2010 17:20, Iulian Dragos <iulian.dragos@epfl.ch> wrote:

Yes, it should work.

Sounds good ;-) 

Development has slowed down quite a bit since I am writing my thesis, and that's part of the reason specialization is still a -Y option (in other words, use on your own risk). It is for the most part feature-complete, so please file tickets with the bugs you find, that would help us a lot.

Did it: https://lampsvn.epfl.ch/trac/scala/ticket/3085
Thanks,
Heiko
Work: weiglewilczek.com
Blog: heikoseeberger.name
Follow me: twitter.com/hseeberger
OSGi on Scala: scalamodules.org
Lift, the simply functional web framework: liftweb.net

Hi Heiko,

On Mon, Feb 22, 2010 at 1:33 PM, Heiko Seeberger
wrote:
> Iulian,
> You closed the ticket commenting "(In [20949]) Specialized types are not
> substituted inside type arguments.".
> I am not sure if I fully understand => Could you please explain?

It means that Iulian fixed the bug in that changeset:

https://lampsvn.epfl.ch/trac/scala/changeset/20949

Best,
Ismael

Ah, great!
Thanks,
Heiko

On 22 February 2010 14:38, Ismael Juma <mlists@juma.me.uk> wrote:

Hi Heiko,

On Mon, Feb 22, 2010 at 1:33 PM, Heiko Seeberger
<heiko.seeberger@googlemail.com> wrote:
> Iulian,
> You closed the ticket commenting "(In [20949]) Specialized types are not
> substituted inside type arguments.".
> I am not sure if I fully understand => Could you please explain?

It means that Iulian fixed the bug in that changeset:

https://lampsvn.epfl.ch/trac/scala/changeset/20949

Best,
Ismael

--
Heiko Seeberger

Work: weiglewilczek.com
Blog: heikoseeberger.name
Follow me: twitter.com/hseeberger
OSGi on Scala: scalamodules.org
Lift, the simply functional web framework: liftweb.net

On 22 February 2010 14:38, Ismael Juma <mlists@juma.me.uk> wrote:

Hi Heiko,

On Mon, Feb 22, 2010 at 1:33 PM, Heiko Seeberger
<heiko.seeberger@googlemail.com> wrote:
> Iulian,
> You closed the ticket commenting "(In [20949]) Specialized types are not
> substituted inside type arguments.".
> I am not sure if I fully understand => Could you please explain?

It means that Iulian fixed the bug in that changeset:

In short, yes. The problem was that specialized types were propagated too eagerly, so in your example the function literal was typed "Int => SpecializedPair$mcI$sp" instead of "Int => SpecializedPair[Int]". The AbstractMethodError appeared because the apply method in the Function1 trait was implemented at SpecializedPair$mcI$sp.
I ran you example and the speedup was not very impressive. I think that's because the running time was dominated by memory operations (on my system I got around 10.000 L2 cache misses). A better benchmark would be the matrix multiplication routine I checked in in the same commit, under test/files/run/spec-matrix.scala.
iulian

https://lampsvn.epfl.ch/trac/scala/changeset/20949

Best,
Ismael

--
Heiko Seeberger

Work: weiglewilczek.com
Blog: heikoseeberger.name
Follow me: twitter.com/hseeberger
OSGi on Scala: scalamodules.org
Lift, the simply functional web framework: liftweb.net

--
« Je déteste la montagne, ça cache le paysage »
Alphonse Allais

Hi Iulian,

On Mon, Feb 22, 2010 at 2:36 PM, Iulian Dragos wrote:
> I think that's
> because the running time was dominated by memory operations (on my system I
> got around 10.000 L2 cache misses).

Out of curiosity, what did you use to measure the L2 cache misses?

Thanks,
Ismael

On Mon, Feb 22, 2010 at 3:45 PM, Ismael Juma <mlists@juma.me.uk> wrote:

Hi Iulian,

On Mon, Feb 22, 2010 at 2:36 PM, Iulian Dragos <iulian.dragos@epfl.ch> wrote:
> I think that's
> because the running time was dominated by memory operations (on my system I
> got around 10.000 L2 cache misses).

Out of curiosity, what did you use to measure the L2 cache misses?

Shark, a Mac OS X profiler can read hardware performance counters.
iulian 

Thanks,
Ismael

--
« Je déteste la montagne, ça cache le paysage »
Alphonse Allais

On Mon, Feb 22, 2010 at 4:14 PM, Iulian Dragos wrote:
> Shark, a Mac OS X profiler can read hardware performance counters.
> iulian

Cool, thanks.

Ismael