GNU MP assertion failed: T(1)+std::numeric_limits<double>::denorm_min()>1
marc.glisse at inria.fr
Tue Sep 9 15:39:24 UTC 2014
On Tue, 9 Sep 2014, Jens Timmerman wrote:
> I tried building GMP 6.0.0a with the intel compiler icc/2015.0.090
> and got the above error when running make check.
> relevant lines in the output of make check.
> I do believe this might be a compiler bug, since I can build it fine
> with older versions of the intel compiler,
> however it is not entirely clear how to further debug this.
Yes, something like that was reported to the list a few months ago. By
default, Intel's compiler uses the equivalent to gcc's -ffast-math. You
need to give it some option so it doesn't ignore denormals.
I believe this only affects the testsuite (that is, you can safely use
the libgmp you built) but I never checked to be sure.
More information about the gmp-bugs