Help with testing 4.1.3

[Kevin Ryde]

Laurent Fousse <laurent at komite.net> writes:
>
> | libm.a differs from mpfr_sqrt for a=8.69381046010529202883e-303,
> | rnd_mode=GMP_RNDZ
> | libm.a gives 9.32406052109556613622e-152, mpfr_sqrt gives
> | 9.32406052109556444039e-152 (-1 ulp)
> | PASS: tsqrt
> | mpfr_sqrt_ui differs from sqrt for a=747735499, rnd_mode=GMP_RNDU
> | sqrt gives 2.73447526776162303577e+04, mpfr_sqrt_ui gives
> | 2.73447526776162339956e+04 (1 ulp)
> | FAIL: tsqrt_ui

We've had various troubles with unwise assumptions about libm in the
mpfr test programs, this might be one of those or it might be a real
problem.

You could change some of those printfs to %a to show the exact bits
coming out, which should show which answer is right.