[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
Re: correctly rounded transcendentals
The goals of improving portability and interval arithmetic in the paper by Zimmermann and Muller are certainly worthy, but I have some questions.
To ensure portability, would it be more effective to bind the libraries to the application before distribution? This seems possible, but I'm not informed on this issue.
While it would be worthwhile to improve interval arithmetic, aren't the majority of computation cycles spent in round-to-nearest? I'm in favor of requiring compliance in all rounding modes, but library writers for some architectures claim that merely detecting and responding to the rounding mode would significantly slow their implementation.
Would it be worthwhile to specify (I'm riding my hobbyhorse) the requirements for a function to be IEEE754R compliant without requiring an implementation to to provide an entire compliant library in order for the system to IEEE754R compliant?