Re: --jit-enable

[Max Brister]

On Tue, Sep 11, 2012 at 12:15 AM, Philip Nienhuis <address@hidden> wrote:
> Rik wrote:
>>
>>   On 09/10/2012 02:37 PM, Philip Nienhuis wrote:
>>>
>>>  Rik wrote:
>>> > On 09/10/2012 12:50 PM, address@hidden wrote:
>>> >> Message: 7
>>> >> Date: Mon, 10 Sep 2012 21:34:17 +0200
>>> >> From: Philip Nienhuis<address@hidden>
>>> >> To: Max Brister<address@hidden>
>>> >> Cc: Philip Nienhuis<address@hidden>,
>>> >> address@hidden
>>> >> Subject: Re: LLVM not picked up - too old version?
>>> >> Message-ID:<address@hidden>
>>> >> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>>> >>
>>> >> Max Brister wrote:
>>> >>>> On Mon, Sep 10, 2012 at 1:05 PM, Philip Nienhuis
>>> >>>> <address@hidden> wrote:
>>> >>>>>> What version of LLVM is needed these days?
>>> >>>>>>
>>> >>>>>> I have llvm 3.0; configure output at the end doesn't mention the
>>
>> llvm
>>>
>>> >>>>>> libs/flags/include but only says something like:
>>> >>>>>> "JIT compiler for loops: false"
>>> >>>>>>
>>> >>>>>> AFAICR a week or two weeks ago it did pick up the llvm stuff
>>
>> (that is, I saw
>>>
>>> >>>>>> the llvm libs/include/flags mentioned in configure's output).
>>> >>>>>>
>>> >>>>>> linux Mageia 2, gcc 4.6.3, Pentium M
>>> >>>>>>
>>> >>>>>> Philip
>>> >>>>
>>> >>>> LLVM 3.0 should still work, I haven't tested it recently though.
>>> >>>> Recently there was a change that requires "--enable-jit" in order
>>> >>>> for
>>> >>>> JIT to be enabled.
>>> >> Thanks, yes --enable-jit=yes did the trick.
>>> >> I must have missed that announcement somehow.
>>> >>
>>> >> BTW I now see messages (using sources updated two hours ago):
>>> >> "cc1plus: warning: /usr/bin/llvm-config: not a directory {enabled by
>>> >> default]"
>>> >> repeated twice after every compile step. Should I start worrying?
>>> > Yes. That is unlikely to work. We can wait until the end of the
>>> > compile,
>>> > but I think I want to see your config.log file.
>>>
>>>  Hmmm, you were right. It just errored out. Messages attached below in
>>
>> msg body. Sorry for line wrap, that's how I copied it from a Konsole
>> terminal.
>>>
>>>  I'll attach the bzip2'd config.log (78 KB) anyway.
>>
>> 9/10/12
>>
>> Philip,
>>
>> The problem is that the location for the llvm-config program is being
>> set to "which llvm-config".  It looks like maybe there was an attempt to
>> set the LLVM_CONFIG environment variable with LLVM_CONFIG="which
>> llvm-config" instead of LLVM_CONFIG=`which llvm-config`.  Notice the
>> backticks in the second expression which will force it to be executed in
>> a subshell and return the result of the command.
>>
>> Offhand, unless you have llvm-config in a strange place, I would let
>> configure search for it rather than specifying it in the LLVM_CONFIG
>> environment variable.  Try something like
>>
>> unset LLVM_CONFIG
>> ./configure {OPTIONS} | tee myconfig.log
>>
>> Look inside myconfig.log and you should have something like the following:
>>
>> checking for llvm-config... /usr/local/bin/llvm-config
>>
>> Otherwise, I would locate llvm-config by hand on your system and then use
>>
>> LLVM_CONFIG='full_path_to_llvm-config'
>> export LLVM_CONFIG
>> ./configure {OPTIONS} | tee myconfig.log
>>
>> And again check myconfig.log to see that the build system correctly
>> found llvm-config.
>
>
> Thanks.
>
> My build went succesfully this might; but now make check ends here:
>
> ===================================================
> address@hidden oct-3.7]$ make check
> Makefile:2655: warning: overriding recipe for target `check'
> :
> make[3]: Leaving directory `/home/philip/devel/oct-3.7/doc/interpreter'
>
> Octave successfully built.  Now choose from the following:
>
>    ./run-octave    - to run in place to test before installing
>    make check      - to run the tests
>    make install    - to install (PREFIX=/usr/local)
>
> make[2]: Leaving directory `/home/philip/devel/oct-3.7'
>
> make[1]: Leaving directory `/home/philip/devel/oct-3.7'
> make -C test check
> make[1]: Entering directory `/home/philip/devel/oct-3.7/test'
> ../../octave/test/build_sparse_tests.sh
> ../../octave/test/build_bc_overload_tests.sh
> ../../octave/test/bc_overloads_expected
> ../run-octave --norc --silent --no-history ../../octave/test/fntests.m
> ../../octave/test
>
> Integrated test scripts:
>
>   libinterp/corefcn/__contourc__.cc ...................... PASS    1/1
>   libinterp/corefcn/__dispatch__.cc ...................... PASS    1/1
>   libinterp/corefcn/__lin_interpn__.cc ................... PASS    1/1
>   libinterp/corefcn/__pchip_deriv__.cc ................... PASS    1/1
>   libinterp/corefcn/__qp__.cc ............................ PASS    1/1
>   libinterp/corefcn/besselj.cc ........................... PASS  191/191
>   libinterp/corefcn/betainc.cc ........................... PASS   23/23
>   libinterp/corefcn/bsxfun.cc ............................ PASS   73/73
>   libinterp/corefcn/cellfun.cc ...........................Stack dump:
> 0.      Running pass 'CallGraph Pass Manager' on module 'octave'.
> panic: Segmentation fault -- stopping myself...
> make[1]: *** [check] Segmentation fault
> make[1]: Leaving directory `/home/philip/devel/oct-3.7/test'
> make: *** [check] Error 2
> address@hidden oct-3.7]$
>
>
> ===================================================
> Running Jordi's debug command from
> https://mailman.cae.wisc.edu/pipermail/help-octave/2012-September/053816.html
> (empty lines in output removed):
>
> address@hidden oct-3.7]$ ./run-octave -q --norc --no-history
> ../octave/test/fntests.m
> Integrated test scripts:
> Fixed test scripts:
>   test_bc_overloads.m .................................... PASS  451/451
>   test_sparse.m .......................................... PASS 1204/1204
> Summary:
>   PASS      1655
>   FAIL         0
> See the file test/fntests.log for additional details.
> 0 (of 2) .m files have no tests.
> 0 (of 0) .cc files have no tests.
>
> ===================================================
>
>
> ...for some reason that stops after the first two tests and never reaches
> the offending one
>
> I think I'll get a fresh checkout and start over completely. When I get back
> from work tonight I'll see if that helped.
>
> Philip

Philip, you aren't alone. It looks like my latest patch to JIT has
caused problems for anyone who has a setup significantly different
then me. I'm starting to reproduce the problem, so hopefully I should
have a fix soon.

For reference: https://savannah.gnu.org/bugs/?37308

-- 
Max Brister