virtualization-lms-core icon indicating copy to clipboard operation
virtualization-lms-core copied to clipboard

Published 20 hours ago • verified publisher icon TiarkRompf

Dealing with NaN in NumericOpsExpOpt

Open mdashti opened this issue 11 years ago • 0 comments

Currently we have trivial optimization of converting "X = Y * 0" to "X = 0" in NumericOpsExpOpt trait and it works fine for integers, but it is not a correct optimization considering IEEE standard for floating point operations (NaN * 0 = NaN). I think, disabling this optimization for Double and Float numbers is the correct solution. Does anyone have any comments?

mdashti avatar May 25 '13 19:05 mdashti