Re: [isabelle-dev] Poly/ML 5.7

[Makarius]

On 15/05/17 14:33, Makarius wrote:
> On 15/05/17 12:14, Makarius wrote:
>> Some results can be seen here:
>>
>> http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux/index.html
>> http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux_2_threads/index.html
>>
>> The test hardware is similar or actually the same as "Linux A", but this
>> needs to be investigated further. It is also important to compare
>> timings for the *same* Isabelle versions, but the Poly_ML_5.7_Linux task
>> is still busy digging into older history.
> 
> Maybe I also manage to get ml_statistics into the charts soon. Then we
> can look at heap, task, thread details etc.

I have now done that in various changes leading to Isabelle/65e132abab1e.

Moreover, I have made a fresh start for "Poly/ML 5.7" measurements, with
precisely the same parameters as "Linux A" (Isabelle/d547173212d2). This
explains why there are only a few data points for Poly/ML 5.7, but the
situation will improve every day.


This means we can now compare the measurements and analyze the (slight)
performance losses of Poly/ML 5.7.


Makarius
___
isabelle-dev mailing list
isabelle-...@in.tum.de
https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev

Re: [isabelle-dev] Poly/ML 5.7

[Makarius]

On 15/05/17 14:24, Makarius wrote:
> On 15/05/17 12:29, Lawrence Paulson wrote:
>> Version 5.7 doesn’t even build on my main workstation, though it works
>> on my MacBook Pro running broadly similar software. No idea what is
>> going on here, but I’m not happy about it.
> 
> There are already some mail threads on the Poly/ML list, where various
> people have helped out to consolidate the situation.
> 
> Anyone who is proficient with C/C++ and various platforms (Linux,
> Windows, Mac OS X varieties) is invited to join there -- not just now,
> but also in the longer term.

Some of this activity is actually on the HOL and ProofPower mailing
list, apart from the Isabelle side here.

We should make sure that everything relevant to Poly/ML also goes over
its official mailing list:
http://lists.inf.ed.ac.uk/mailman/listinfo/polyml


Makarius

___
isabelle-dev mailing list
isabelle-...@in.tum.de
https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev

Re: [isabelle-dev] Poly/ML 5.7

[Makarius]

On 15/05/17 12:14, Makarius wrote:
> Some results can be seen here:
> 
> http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux/index.html
> http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux_2_threads/index.html
> 
> The test hardware is similar or actually the same as "Linux A", but this
> needs to be investigated further. It is also important to compare
> timings for the *same* Isabelle versions, but the Poly_ML_5.7_Linux task
> is still busy digging into older history.

I have now changed that here:

changeset:   65840:8d7b2ac9a245
tag: tip
user:wenzelm
date:Mon May 15 14:27:14 2017 +0200
files:   src/Pure/Admin/isabelle_cronjob.scala
description:
history parameters like "Linux A", for more comparable results;


Maybe I also manage to get ml_statistics into the charts soon. Then we
can look at heap, task, thread details etc.


Makarius

___
isabelle-dev mailing list
isabelle-...@in.tum.de
https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev

Re: [isabelle-dev] Poly/ML 5.7

[Makarius]

On 15/05/17 12:29, Lawrence Paulson wrote:
> Version 5.7 doesn’t even build on my main workstation, though it works
> on my MacBook Pro running broadly similar software. No idea what is
> going on here, but I’m not happy about it.

There are already some mail threads on the Poly/ML list, where various
people have helped out to consolidate the situation.

Anyone who is proficient with C/C++ and various platforms (Linux,
Windows, Mac OS X varieties) is invited to join there -- not just now,
but also in the longer term.


Note that for Isabelle, the build process works via "isabelle
build_polyml" from recent repository versions, see also
http://isabelle.in.tum.de/repos/isabelle/annotate/8007f10195af/src/Pure/Admin/build_polyml.scala

Going through Isabelle/Scala has the slight disadvantage of obscuring
the description of the build process, but it is more robust and
repeatable on all these platform variations.


Makarius

___
isabelle-dev mailing list
isabelle-...@in.tum.de
https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev

Re: [isabelle-dev] Poly/ML 5.7

[Lawrence Paulson]

Version 5.7 doesn’t even build on my main workstation, though it works on my 
MacBook Pro running broadly similar software. No idea what is going on here, 
but I’m not happy about it.
Larry

> On 15 May 2017, at 11:14, Makarius  wrote:
> 
> David Matthews has made the Poly/ML 5.7 release snapshot some weeks ago
> (see https://github.com/polyml/polyml/releases/tag/v5.7), but announced
> it only last week.
> 
> I have wrapped that up as a component in Isabelle/d3c5898f1a5e, but that
> is only for testing, not for production use. Some results can be seen here:
> 
> http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux/index.html
> http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux_2_threads/index.html
> 
> The test hardware is similar or actually the same as "Linux A", but this
> needs to be investigated further. It is also important to compare
> timings for the *same* Isabelle versions, but the Poly_ML_5.7_Linux task
> is still busy digging into older history.
> 
> 
> After the substantial changes of the code generator and representation
> of ML values in memory, there are various open questions concerning
> performance and robustness of Isabelle applications.
> 
> This means, we need to skip the Poly/ML 5.7 release and stay on the last
> Poly/ML 5.6 until the situation becomes more clear.
> 
> 
> This is how to use the polyml-5.7 component locally:
> 
>  $ edit $ISABELLE_HOME_USER/etc/settings
> 
>  init_component "$HOME/.isabelle/contrib/polyml-5.7"
> 
>  $ isabelle components -a
> 
> 
>   Makarius
> ___
> isabelle-dev mailing list
> isabelle-...@in.tum.de
> https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev

___
isabelle-dev mailing list
isabelle-...@in.tum.de
https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev

[isabelle-dev] Poly/ML 5.7

[Makarius]

David Matthews has made the Poly/ML 5.7 release snapshot some weeks ago
(see https://github.com/polyml/polyml/releases/tag/v5.7), but announced
it only last week.

I have wrapped that up as a component in Isabelle/d3c5898f1a5e, but that
is only for testing, not for production use. Some results can be seen here:

http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux/index.html
http://isabelle.in.tum.de/devel/build_status/Poly_ML_5.7_Linux_2_threads/index.html

The test hardware is similar or actually the same as "Linux A", but this
needs to be investigated further. It is also important to compare
timings for the *same* Isabelle versions, but the Poly_ML_5.7_Linux task
is still busy digging into older history.


After the substantial changes of the code generator and representation
of ML values in memory, there are various open questions concerning
performance and robustness of Isabelle applications.

This means, we need to skip the Poly/ML 5.7 release and stay on the last
Poly/ML 5.6 until the situation becomes more clear.


This is how to use the polyml-5.7 component locally:

  $ edit $ISABELLE_HOME_USER/etc/settings

  init_component "$HOME/.isabelle/contrib/polyml-5.7"

  $ isabelle components -a


Makarius
___
isabelle-dev mailing list
isabelle-...@in.tum.de
https://mailmanbroy.informatik.tu-muenchen.de/mailman/listinfo/isabelle-dev