[PPL-devel] PPL make check fails (OS X)

Enea Zaffanella zaffanella at cs.unipr.it
Sat May 12 08:35:30 CEST 2012


On 05/12/2012 12:14 AM, Sergey Oboguev wrote:
> See the attachments.

[...]

> Please send us your (compressed) config.log as well as the output of the
> following commands:
>
> # cd <ppl-build-dir>
> # cd tests/Concrete_Expression
> # PPL_NOISY_TESTS=1 ./polyhedron1
> # PPL_NOISY_TESTS=1 ./polyhedron2
>
> Enea.

 From the config log I see that your machine is using llvm-gcc:

> configure:5217: g++ --version >&5
> i686-apple-darwin11-llvm-g++-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2336.9.00)

LLVM does not provide support for -frounding-math and hence uncorrect 
results may be obtained when using floating-point based domains.
See the following threads:

http://www.cs.unipr.it/pipermail/ppl-devel/2011-August/017559.html
http://www.cs.unipr.it/pipermail/ppl-devel/2011-August/017560.html


Actually, we recently added some configure machinery in the PPL so as to 
detect "clang masqueraded as gcc" ... but that seems to be not enough:

> configure:5524: checking whether we are actually using clang++
[...]
> configure:5545: result: no

because your system is not using the clang frontend (but the problem is 
anyway rooted in the LLVM backend). We will probably have to refine this 
detection mechanism and make sure all FP based tests are disabled.

To summarize: afaict, those ppl tests are failing because LLVM does not 
support -frounding-math. The failure will only affects you if you are 
using floating-point based computations (in particular, when GCC uses 
the PPL it does *not* use floating-point based computations).

Regards,
Enea.



More information about the PPL-devel mailing list