error in tgamma

       Hi,

while testing Newlib 4.3.0, I also noticed the following regression
with respect to 4.2.0 for tgammaf:

zimmerma@tartine:~/svn/tbd/20/src/binary32_exhaustive$ VERBOSE=-v ./doit2.newlib tgamma
Checking tgamma with newlib-4.3.0.20230120
Using RedHat newlib
MPFR library: 4.2.0       
MPFR header:  4.2.0 (based on 4.2.0)
Checking function tgammaf with MPFR_RNDN
...
libm wrong by up to nan ulp(s) [9223372036854775808] for x=-0x1p+0
tgamma      gives inf
mpfr_tgamma gives -nan

For a negative integer, tgamma should return NaN.

Best regards,
Paul Zimmermann