On 3/11/11 6:18 AM, Barton Willis wrote:
> By the way: In GCL (expt 10.0d0 6.0d0) --> 1000000.0000000013, but both CCL and Allegro give
> 1000000.0. For GCL, the relative difference between the true and floating point value is
> about 10 times the machine epsilon. I don't know what the IEEE standard says about the
> accuracy of floating point exponentiation, but ten times the machine epsilon seems too big.
In cmucl, (exp (* 6d0 (log 10d0))) -> 1000000.0000000013. I think we
can guess how gcl implement expt.
Ray