26 Mar
2017
26 Mar
'17
10:34 p.m.
On Sun, Mar 26, 2017 at 7:08 AM, Henry Baker <hbaker1@pipeline.com> wrote:
If I understood Gustafson's talk correctly, IEEE doubles aren't good enough. He showed some examples where IEEE quads (128 bits) weren't good enough.
IEEE doubles are good enough to emulate 32-bit posits.
Is there any way you could convert to Maxima bigfloats, do the operation, and convert back?
Bigfloats can be set to any precision you want.
I could have used any of the available multi-precision libraries, but for the 32-bit posit experiment that wouldn't be necessary.
The question is, how many of the existing algorithms depend on the fact that the relative precision is stable across almost all representable range, Leo