On Thu, Jun 3, 2010 at 10:20 AM, Bill Hart <[email protected]> wrote:
> For any Sage people reading this thread:
>
> MPIR now uses the same XGCD normalisation as GMP 5.0.1. but *NOT*
> previous versions (e.g. NOT GMP 5.0.0) and GMP 4.3.2. This is the
> normalisation Pari requires to operate correctly. It is expected that
> the XGCD normalisation is now fixed forever and will not change again
> in MPIR (and if we understand correctly from the GMP list, not in GMP
> either).
>
> If you find a counterexample to that, please report the inputs you
> provide to the xgcd function and what GMP and MPIR return and we'll
> fix it. (We aren't expecting any counterexamples, as we have test code
> for this and have run hundreds of millions of XGCD's to check.)
>

Excellent.  We'll change our doctests to reflect this.  Thanks.

 -- William

> Bill.
>
> On 3 June 2010 18:11, Bill Hart <[email protected]> wrote:
>> A little bit more detail on the MPIR in Sage thing:
>>
>> It looks like MPIR 2.1.0 is slated to be released in Sage 5.0. It's
>> actually one of the main goals for that release. Woot!!
>>
>> http://trac.sagemath.org/sage_trac/milestone/sage-5.0
>>
>> That's only two weeks away, and it takes nearly a week to do the
>> release *after* all the patches are written. So we need to release a
>> solid MPIR ASAP so they have time to fix the remaining issues in Sage.
>>
>> It looks like someone patched ECM in Sage already. But there are now
>> Sage doctest failures. In all likelihood these are due to GCD
>> normalisation, and will require the attention of the people involved
>> with writing the Sage doctests. I think this will get dealt with when
>> the Sage 5.0 release process starts. It's a one off job that the Sage
>> guys know needs to be done now that the GCD normalisation has been
>> changed back to the original. See the following for more details:
>>
>> http://trac.sagemath.org/sage_trac/ticket/8664
>>
>> Note that MPIR 2.0.1. on that ticket is now meant to read 2.1.0. There
>> will be no MPIR 2.0.1.
>>
>> Bill.
>>
>> On 3 June 2010 17:43, Bill Hart <[email protected]> wrote:
>>> On 3 June 2010 15:00, Jason Moxham <[email protected]> wrote:
>>>> On Thursday 03 June 2010 04:58:30 Jason Moxham wrote:
>>>>> On Wednesday 02 June 2010 17:45:26 Jason Moxham wrote:
>>>>> > Looks good , there are only 2 or 3 real errors
>>>>> >
>>>>> > your cleo and iras errors are because you need to get icc in your path  
>>>>> > ,
>>>>> > you will hit 1 real error then , but we can easily fix that.
>>>>>
>>>>> I fixed it by just deleting the assert , we very rarely have them in asm
>>>>> anyway , and that fixed the problem until we hit the next little error
>>>>>
>>>>>
>>>>>  ../.libs/libmpir.so  -Wl,--rpath -Wl,/usr/local/lib
>>>>> icc: command line remark #10010: option '-c99' is deprecated and will be
>>>>> removed in a future release. See '-help deprecated'
>>>>> icc: command line remark #10010: option '-c99' is deprecated and will be
>>>>> removed in a future release. See '-help deprecated'
>>>>> /home/jasonmoxham/sourcecode/mpir-2.1.0/cleo/.libs/libmpir.so: undefined
>>>>> reference to `__gmpn_invert_limb'
>>>>> icc -c99 -o .libs/t-bswap t-bswap.o  ./.libs/libtests.a
>>>>> /home/jasonmoxham/sourcecode/mpir-2.1.0/cleo/.libs/libmpir.so
>>>>> ../.libs/libmpir.so  -Wl,--rpath -Wl,/usr/local/lib
>>>>> make[4]: *** [t-count_zeros] Error 1
>>>>> make[4]: *** Waiting for unfinished jobs....
>>>>> icc: command line remark #10010: option '-c99' is deprecated and will be
>>>>> removed in a future release. See '-help deprecated'
>>>>> t-constants.o: In function `main':
>>>>> /home/jasonmoxham/sourcecode/mpir-2.1.0/./tests/t-constants.c:(.text+0x4b2)
>>>>>: undefined reference to `__gmpn_invert_limb'
>>>>> /home/jasonmoxham/sourcecode/mpir-2.1.0/cleo/.libs/libmpir.so: undefined
>>>>> reference to `__gmpn_invert_limb'
>>>>> make[4]: *** [t-constants] Error 1
>>>>> make[4]: *** [t-gmpmax] Error 1
>>>>> /home/jasonmoxham/sourcecode/mpir-2.1.0/cleo/.libs/libmpir.so: undefined
>>>>> reference to `__gmpn_invert_limb'
>>>>> make[4]: *** [t-bswap] Error 1
>>>>> make[4]: Leaving directory
>>>>> `/home/jasonmoxham/sourcecode/mpir-2.1.0/cleo/tests'
>>>>> make[3]: *** [check-am] Error 2
>>>>> make[3]: Leaving directory
>>>>> `/home/jasonmoxham/sourcecode/mpir-2.1.0/cleo/tests'
>>>>> make[2]: *** [check-recursive] Error 1
>>>>> make[2]: Leaving directory
>>>>> `/home/jasonmoxham/sourcecode/mpir-2.1.0/cleo/tests'
>>>>> make[1]: *** [check-recursive] Error 1
>>>>> make[1]: Leaving directory `/home/jasonmoxham/sourcecode/mpir-2.1.0/cleo'
>>>>> make: *** [check] Error 2
>>>>> cleo
>>>>>
>>>>> PASSED CC=gcc CXX=g++ configure=
>>>>> PASSED CC=gcc CXX=g++ configure= --enable-cxx --enable-gmpcompat
>>>>> PASSED CC=gcc CXX=g++ configure= --enable-cxx --enable-gmpcompat --enable-
>>>>> assert --enable-alloca=debug
>>>>> PASSED CC=gcc CXX=g++ configure= --enable-cxx --enable-gmpcompat --enable-
>>>>> assert --enable-alloca=debug --build=none-unknown-linux-gnu
>>>>> PASSED CC=cc CXX=c++ configure=
>>>>> PASSED CC=cc CXX=c++ configure= --enable-cxx --enable-gmpcompat
>>>>> PASSED CC=cc CXX=c++ configure= --enable-cxx --enable-gmpcompat
>>>>> --enable-assert --enable-alloca=debug
>>>>> PASSED CC=cc CXX=c++ configure= --enable-cxx --enable-gmpcompat
>>>>> --enable-assert --enable-alloca=debug --build=none-unknown-linux-gnu
>>>>> PASSED CC=icc CXX=g++ configure=
>>>>> PASSED CC=icc CXX=g++ configure= --enable-cxx --enable-gmpcompat
>>>>> PASSED CC=icc CXX=g++ configure= --enable-cxx --enable-gmpcompat --enable-
>>>>> assert --enable-alloca=debug
>>>>> FAILED CC=icc CXX=g++ configure= --enable-cxx --enable-gmpcompat --enable-
>>>>> assert --enable-alloca=debug --build=none-unknown-linux-gnu
>>>>>
>>>>> This looks easy to fix , I'll do it tomorrow
>>>>> Jason
>>>>
>>>> This is caused by LONGLONG_STANDALONE not being defined ANYWHERE???? which
>>>> makes invert_limb depend on udiv_qrnnd_preinv which depends on invert_limb
>>>> which depends on .....
>>>
>>> LONGLONG_STANDALONE shouldn't be defined anywhere. The longlong.h file
>>> used to be maintained as a standalone file, but has not been for many
>>> years. This define is left over from the days when it used to be used
>>> that way.
>>>
>>>>
>>>> As this is just to run some more tests and not a production mpir lib , I
>>>> suggest we forget about it for now (plus it's only on icc not gcc)
>>>>
>>>> I've run the full mpirtests on my K8 and atom both with 64 and 32bit OSes 
>>>> as
>>>> described before with no errors , and also on mark.skynet (although the 
>>>> test
>>>> script burped at the end , but as the run took a full day , I debug the 
>>>> test
>>>> script later)
>>>>
>>>> Are there any outstanding tests?
>>>> I assume building sage with it went OK?
>>>
>>> The last I heard about sage was the ecm issue. I don't know if they've
>>> updated ECM yet or not. It's not an upstream issue, as ECM have fixed
>>> this in a more recent release.
>>>
>>>>
>>>> Jason
>>>>
>>>>
>>>>
>>>>>
>>>>> > for lena and flavius gcc34 doesn't have a g++
>>>>> >
>>>>> > gcc54 has a broken c++
>>>>> >
>>>>> > gcc42 has a broken c++ and there is no way to get  a timer for tuning
>>>>> > (same for cato)
>>>>> >
>>>>> > fulvia gcc4.5.0 is not installed right
>>>>> >
>>>>> > fulvia cc , 1 real error , suns assembler is old , we should force it to
>>>>> > use yasm is all cases , I see if I can do it (I have to remove my brain
>>>>> > first , to get in the autotools mood) , but it may have to wait for next
>>>>> > release.
>>>>> >
>>>>> > bsd.math  we dont support fat builds on a mac (PIC code is different) 
>>>>> > and
>>>>> > they arn't needed that much
>>>>> >
>>>>> > gcc101 is a bsd system and I think we dont support fat builds (same as
>>>>> > above?) so possibly a real error
>>>>> >
>>>>> > rest of the machines use gcc-4.3.2 which is broken
>>>>> >
>>>>> > I have updated the test script to reflect some of changes , but there 
>>>>> > are
>>>>> > a lot of broken systems , so we will still get some errors from these.
>>>>> >
>>>>> > Jason
>>>>> >
>>>>> > On Wednesday 02 June 2010 16:47:22 Minh Nguyen wrote:
>>>>> > > Hi folks,
>>>>> > >
>>>>> > > Build and test results for MPIR 2.1.0-rc2 on the build farm are
>>>>> > > available at
>>>>> > >
>>>>> > > http://wiki.sagemath.org/mpir/BuildFarm/mpir-2.1.0#MPIR2.1.0-rc2
>>>>> > >
>>>>> > > Each build was done using one thread, i.e. first exporting
>>>>> > >
>>>>> > > $ export MAKE='make'
>>>>> > >
>>>>> > > The test suite was ran using one CPU. On some machines, doing a
>>>>> > > parallel build would result in a failure to tune, i.e. doing
>>>>> > >
>>>>> > > $ make tune
>>>>> > >
>>>>> > > after a parallel build would fail. So for each reported machine, I
>>>>> > > only used one thread/CPU for build, check, tune, and test suite.
>>>>> > >
>>>>> > > --
>>>>> > > Regards
>>>>> > > Minh Van Nguyen
>>>>
>>>> --
>>>> You received this message because you are subscribed to the Google Groups 
>>>> "mpir-devel" group.
>>>> To post to this group, send email to [email protected].
>>>> To unsubscribe from this group, send email to 
>>>> [email protected].
>>>> For more options, visit this group at 
>>>> http://groups.google.com/group/mpir-devel?hl=en.
>>>>
>>>>
>>>
>>
>
> --
> You received this message because you are subscribed to the Google Groups 
> "mpir-devel" group.
> To post to this group, send email to [email protected].
> To unsubscribe from this group, send email to 
> [email protected].
> For more options, visit this group at 
> http://groups.google.com/group/mpir-devel?hl=en.
>
>



-- 
William Stein
Professor of Mathematics
University of Washington
http://wstein.org

-- 
You received this message because you are subscribed to the Google Groups 
"mpir-devel" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/mpir-devel?hl=en.

Reply via email to