[racket] Fun with denormalized floating point numbers

On 08/01/2012 11:40 AM, Michael Wilber wrote:
> ...
>
> Fascinating. I wonder if your machine has better floating point hardware
> So maybe this is just something to watch out for on older chips.

9000ms vs 350ms is a surprising difference! That's got to be software 
floating-point.

I know my floating-point hardware is pretty good. For the math 
collection, I've written a test suite that checks Racket's flonum ops 
against MPFR, an arbitrary-precision floating-point library whose 
algorithms are proved correct in various research papers. My FPU gets 
everything I've tested right, while it's pretty common for the outputs 
for tricky inputs (like tan(HUGE_NUMBER)) to be a few bits off.

FWIW, /proc/cpuinfo says this is my processor:

   Intel(R) Core(TM) i7-2630QM CPU @ 2.00GHz

which is similar to Asumu's.

Neil ⊥