I guess that as exponent size increases for a given size FFT, the probability of error rises so rapidly form say 2% to 98% that there is little to be gained by allowing a few maverick exponents with a 30% chance of error.
There is much to be said for not allowing the discrete sizes of the FFT to introduce inconsistency in the error rates for exponents over the range. Note that the existence of cosmic ray etc errors is an argument in favour of a bold approach to FP errors, not to be conservative. David Eddy _________________________________________________________________ Be one of the first to try Windows Live Mail. http://ideas.live.com/programpage.aspx?versionId=5d21c51a-b161-4314-9b0e-4911fb2b2e6d _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
