My proposal would be two sets of typeclasses and operators for arithmetic. The usual set (+,-,*,/) would assume the usual field properties and give the optimizer some leeway. The other set would be for when you're prescribing order of operations (of course the optimizer is always free to optimize things in ways that don't change the result) . But both sets of operations would work for both exact and inexact numbers.
Matt --- On Tue, 3/10/09, Jonathan S. Shapiro <[email protected]> wrote: From: Jonathan S. Shapiro <[email protected]> Subject: [bitc-dev] Overloading arithmetic To: [email protected] Date: Tuesday, March 10, 2009, 1:08 AM An obvious example of an incorrect OpenMP program has just convinced me that FP arith ops should not share operators with integer arith ops: FP ops are not associative, which interacts with optimizability. Violent arguments? shap, from my phone -----Inline Attachment Follows----- _______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
_______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
