GNU MP assertion failed: T(1)+std::numeric_limits<double>::denorm_min()>1

Jens Timmerman jens.timmerman at
Tue Sep 9 16:00:11 UTC 2014

On 09/09/14 17:41, roger golliver wrote:
> Jens,
> This is a known issue for those trying to develop high quality numeric codes using the Intel Compiler Suite.
> You must ask the Intel Compiler to conform to the C language standard for floating-point
> expression evaluation; via the "-fp-model precise" compiler switch.  By default, the Intel Compiler developers chose to ignore the C language expression evaluation rules.  Presumably in hope for gains in floating-point performance.
> Try:
>     CC="icc -fp-model precise" CXX="icpc -fp-model precise" ./configure --enable-cxx
>     make
>     make check
> Roger

 thank you for your fast reply,
I had set fp-model source, since this Rounds intermediate results to
source-defined precision. [0]

gmp builds correctly with precise.

It seems gmp does not correctly define the precision then?
(or the new compiler version ignores it, in which case someone should
open a bug report with intel)

Jens Timmerman


More information about the gmp-bugs mailing list