long double test failures due to underflow exception getting set?

[Carlos O'Donell]

On 07/19/2017 05:34 PM, Steve Ellcey wrote:
> Is anyone else seeing these failures with the latest glibc sources?
> I am getting them on aarch64 (standard ToT sources) and was wondering
> if this is an aarch64 specific problem or a more generic bug.  Since
> all the failures involve long double I don't think they are using 
> aarch64 specific code, but I could be wrong.
> 
> FAIL: math/test-ldouble-cexp
> FAIL: math/test-ldouble-exp
> FAIL: math/test-ldouble-finite-cexp
> FAIL: math/test-ldouble-finite-exp
> 
> I think these failures are new and all of the failures listed are
> about the Underflow exception being set:
> 
> I.e.
> 
> testing long double (without inline functions)
> Failure: exp (-0x1p-10000): Exception "Underflow" set
> Failure: exp (-0x2p-16384): Exception "Underflow" set
> Failure: exp (-0x4p-16384): Exception "Underflow" set
> Failure: exp (-0x4p-16448): Exception "Underflow" set
> Failure: exp (-0x4p-16496): Exception "Underflow" set
> Failure: exp (-0x8p-16448): Exception "Underflow" set

Neither DJ nor Szabolcs's glibc 2.26 builds showed those failures:
https://sourceware.org/glibc/wiki/Release/2.26

-- 
Cheers,
Carlos.