New releases on the website?

[wpoe...@gmail.com]

Hi all,

I see that in the git tree of CLooG, version 0.18.3 is already tagged but there is no mention of it on the website? No tarballs to download etc?

Does this have a specific reason or have the maintainers simply not yet found the time to update the website?

I'm asking this because it seems that Clang 3.6rc2 with polly needs CLooG 0.18.3 to compile?

Best regards,

Ward

[Sven Verdoolaege]

On Wed, Feb 04, 2015 at 04:37:15AM -0800, wpoe...@gmail.com wrote:
> I'm asking this because it seems that Clang 3.6rc2 with polly needs CLooG
> 0.18.3 to compile?

What made you get the impression that you need cloog to compile clang 3.6?

skimo

[Ward Poelmans]

When I tried to compile the 3.6rc2, I got:
/home/opt/easybuild/build/Clang/3.6.0/GCC-4.9.2/llvm-3.6.0rc2.src/tools/polly/lib/CodeGen/BlockGenerators.cpp:
In member function ‘llvm::Value*
polly::BlockGenerator::getNewAccessOperand(const
polly::MemoryAccess&)’:
/home/opt/easybuild/build/Clang/3.6.0/GCC-4.9.2/llvm-3.6.0rc2.src/tools/polly/lib/CodeGen/BlockGenerators.cpp:156:38:
error: ‘isl_ast_expr_address_of’ was not declared in this scope

It could be that I messed up versions somewhere. I'm currently
retrying with CLoog 0.18.3 (+ ISL 0.14). The `isl_ast_expr_address_of`
is defined in ISL 0.14.

Ward

[Tobias Grosser]

This is not the need for CLooG, but rather the need for a newer version
of isl.

Can you try the attached patch (which includes isl into Polly). It may
solve your issue.

Cheers,
Tobias

[Ward Poelmans]

On Wed, Feb 4, 2015 at 2:00 PM, Tobias Grosser <tob...@grosser.es> wrote:

> This is not the need for CLooG, but rather the need for a newer version of
> isl.

Yes, but as I understood it, those two are coupled? Anyway, I tried
CLooG-0.18.3 + ISL 0.14 and found a new problem:
git commit 05677d3b35f35 adds `isl_ast_op_zdiv_r` to ast_types but for
some reason, this commit is not part of the isl-0.14 tag. Is that
intentional? Polly doesn't like it ;)

> Can you try the attached patch (which includes isl into Polly). It may solve
> your issue.

I will give it a try.. Can we still use GMP with this patch or is
imath now mandatory?

Ward

[Tobias Grosser]

On 04.02.2015 14:52, Ward Poelmans wrote:
> On Wed, Feb 4, 2015 at 2:00 PM, Tobias Grosser <tob...@grosser.es> wrote:
>
>> This is not the need for CLooG, but rather the need for a newer version of
>> isl.
>
> Yes, but as I understood it, those two are coupled? Anyway, I tried
> CLooG-0.18.3 + ISL 0.14 and found a new problem:
> git commit 05677d3b35f35 adds `isl_ast_op_zdiv_r` to ast_types but for
> some reason, this commit is not part of the isl-0.14 tag. Is that
> intentional? Polly doesn't like it ;)

Polly does not use CLooG any more. It only requires isl, so just leaving
CLooG out of the picture is probably best.

Also, the latest version of Polly currently requires a snapshot of isl
that is newer than isl 0.14. The version of isl that should work is:

b3e0fa7a05d32f1e0e36e0a42b0b83fa2ba1f609

>> Can you try the attached patch (which includes isl into Polly). It may solve
>> your issue.
>
> I will give it a try.. Can we still use GMP with this patch or is
> imath now mandatory?

In this case imath is mandatory. Adding GMP back in is possible, but was
not done as I did not know that there was a need for it. Do you need GMP
for something specific?

Cheers,
Tobias

[Ward Poelmans]

On Wed, Feb 4, 2015 at 3:05 PM, Tobias Grosser <tob...@grosser.es> wrote:
>
> Polly does not use CLooG any more. It only requires isl, so just leaving
> CLooG out of the picture is probably best.

OK, Thanks for clarifying that!

>> I will give it a try.. Can we still use GMP with this patch or is
>> imath now mandatory?
>
> In this case imath is mandatory. Adding GMP back in is possible, but was not
> done as I did not know that there was a need for it. Do you need GMP for
> something specific?

Not really but why not keep the choice? Use imath by default and give
the user the optional choice to replace it with GMP? If GMP still is
faster, I don't see why not?

I don't mind dependencies, only tightly version coupled dependencies are a pain.

Ward

[Tobias Grosser]

I don't mind the choice, but the more options we have the more time is
spent on keeping them working. For GMP, it just adds complexity to the
build scripts.

I would rather spend my time tuning imath (which we need for license
reasons anyhow), than for keeping GMP working.

> I don't mind dependencies, only tightly version coupled dependencies are a pain.

That's the problem. isl is moving fast and Polly tends to use the latest
features, such that we regularly run into situations where we need a new
isl release just to be able to make a proper Polly release.

Could you possibly try the patch, such that I can get some feedback if
it is working for you.

Also, do you currently see compile-time performance issues with Polly
that would cause troubles and where (temporary) additional slowdown
would be an issue?

Cheers,
Tobias

[Ward Poelmans]

On Wed, Feb 4, 2015 at 3:23 PM, Tobias Grosser <tob...@grosser.es> wrote:
> On 04.02.2015 15:15, Ward Poelmans wrote:
>>
>> Not really but why not keep the choice? Use imath by default and give
>> the user the optional choice to replace it with GMP? If GMP still is
>> faster, I don't see why not?
>
> I don't mind the choice, but the more options we have the more time is spent
> on keeping them working. For GMP, it just adds complexity to the build
> scripts.

OK, I can understand that. But should I wish to use GMP, it's only a
matter of making some changes in the build system? The API is the
same?

> Could you possibly try the patch, such that I can get some feedback if it is
> working for you.

Almost there. I'm down to:
/home/opt/easybuild/build/Clang/3.6.0/GCC-4.9.2/llvm-3.6.0rc2.src/tools/polly/lib/External/isl/isl_version.c:2:24:
fatal error: gitversion.h: No such file or directory

This used to be generate by automake. As you checkout a specify
version of isl, it should be manually added?

> Also, do you currently see compile-time performance issues with Polly that
> would cause troubles and where (temporary) additional slowdown would be an
> issue?

None so far, but then again, I haven't been able to compile it successfully ;)

Do you prefer I the feedback directly to you or through this mailinglist?

Ward

[Tobias Grosser]

On 04.02.2015 16:11, Ward Poelmans wrote:
> On Wed, Feb 4, 2015 at 3:23 PM, Tobias Grosser <tob...@grosser.es> wrote:
>> On 04.02.2015 15:15, Ward Poelmans wrote:
>>>
>>> Not really but why not keep the choice? Use imath by default and give
>>> the user the optional choice to replace it with GMP? If GMP still is
>>> faster, I don't see why not?
>>
>> I don't mind the choice, but the more options we have the more time is spent
>> on keeping them working. For GMP, it just adds complexity to the build
>> scripts.
>
> OK, I can understand that. But should I wish to use GMP, it's only a
> matter of making some changes in the build system? The API is the
> same?

Right.

>> Could you possibly try the patch, such that I can get some feedback if it is
>> working for you.
>
> Almost there. I'm down to:
> /home/opt/easybuild/build/Clang/3.6.0/GCC-4.9.2/llvm-3.6.0rc2.src/tools/polly/lib/External/isl/isl_version.c:2:24:
> fatal error: gitversion.h: No such file or directory
>
> This used to be generate by automake. As you checkout a specify
> version of isl, it should be manually added?

Try the updated patch.

>> Also, do you currently see compile-time performance issues with Polly that
>> would cause troubles and where (temporary) additional slowdown would be an
>> issue?
>
> None so far, but then again, I haven't been able to compile it successfully ;)

;-)

> Do you prefer I the feedback directly to you or through this mailinglist?

Yes.

Tobias

[Cédric Bastoul]

Hi Ward,

even if Polly is not using it anymore, and thus you probably don't care, here it is now ;-). I delayed it for too much time :

http://cloog.org

http://www.bastoul.net/cloog/pages/download/cloog-0.18.3.tar.gz

Best and sorry for the delay,

Cedric

--
--
You got this message because you subscribed to the CLooG mailing list.
To send messages to this list, use cl...@googlegroups.com
To stop subscribing, send a mail to cloog+un...@googlegroups.com
For more options and to visit the group, http://groups.google.fr/group/cloog?hl=en

---
You received this message because you are subscribed to the Google Groups "CLooG" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cloog+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ward Poelmans]

We have progress! The build ended successfully but when running the
test suite, it hangs on these 2 tests:
polly/test/Isl/CodeGen/debug-intrinsics.ll
polly/test/Isl/CodeGen/LoopParallelMD/do_not_mutate_debug_info.ll

It's just burning cpu cycles without ending (running for 16h).

Ward

[Sven Verdoolaege]

You should probably drop the cloog mailing list from CC
on this thread. It is getting seriously off-topic.

skimo

[Johannes Doerfert]

How did u build this? My cmake + make debug build passed all tests.

--

Johannes Doerfert
Researcher / PhD Student

Compiler Design Lab (Prof. Hack)
Saarland University, Computer Science
Building E1.3, Room 4.31

Tel. +49 (0)681 302-57521 : doer...@cs.uni-saarland.de
Fax. +49 (0)681 302-3065 : http://www.cdl.uni-saarland.de/people/doerfert

[Tobias Grosser]

On 05.02.2015 13:14, Johannes Doerfert wrote:
> How did u build this? My cmake + make debug build passed all tests.
>
> On 02/05, Ward Poelmans wrote:
>> On Wed, Feb 4, 2015 at 4:24 PM, Tobias Grosser <tob...@grosser.es> wrote:
>>> On 04.02.2015 16:11, Ward Poelmans wrote:
>>>> On Wed, Feb 4, 2015 at 3:23 PM, Tobias Grosser <tob...@grosser.es> wrote:
>>>>> On 04.02.2015 15:15, Ward Poelmans wrote:
>>>>> Also, do you currently see compile-time performance issues with Polly
>>>>> that
>>>>> would cause troubles and where (temporary) additional slowdown would be
>>>>> an
>>>>> issue?
>>>>
>>>>
>>>> None so far, but then again, I haven't been able to compile it
>>>> successfully ;)
>>
>> We have progress! The build ended successfully but when running the
>> test suite, it hangs on these 2 tests:
>> polly/test/Isl/CodeGen/debug-intrinsics.ll
>> polly/test/Isl/CodeGen/LoopParallelMD/do_not_mutate_debug_info.ll
>>
>> It's just burning cpu cycles without ending (running for 16h).

I think this is the 3.6 release branch on which we possibly miss some of
the test case changes that updated the debug meta data.

Cheers,
Tobias