rtoy/crlibm

A clone of crlibm from Inria: https://gforge.inria.fr/anonscm/git/metalibm/crlibm.git

crlibm's Issues

jakes.fdlibm prints out incorrect results

jakes (using the system libm on OSX) and jakes.crlibm print out

   0: -0.220863 5.83353
   1: -0.220863 5.83353
   2: -0.220863 5.83352
   3: -0.220864 5.83352
   4: -0.220865 5.83352
...
1249995: 0.659915   -2.04764
1249996: 0.659962   -2.04758
1249997: 0.660009   -2.04753
1249998: 0.660057   -2.04748
1249999: 0.660104   -2.04743

but jakes.fdlibm gives something totally different:

   0: 2.9168    4.6186
   1: 2.9168    4.6186
   2: 2.9168    4.6186
   3: 2.27267   2.76334
   4: 2.47508   4.6186
...
1249995: 1.745e+74  -1.42599e+25
1249996: -1.5558e+24    -1.42601e+25
1249997: -1.24162e+19   2.76334
1249998: -1.55583e+24   -1.42604e+25
1249999: -1.24162e+19   4.6186

Use unions in fdlibm to access the parts of a double

See #1 for an example.

This is fixed for sin and cos, but the remaining functions need this fixed.