The current version of the code overwrites the source array. This isn't ideal (it doesn't hurt the NPB benchmark because of the way it is structured), but could quickly become expensive since it requires a temporary copy in addition to the work array.
It's probably worth a ~small~ hit to performance to fix this.
The current version of the code overwrites the source array. This isn't ideal (it doesn't hurt the NPB benchmark because of the way it is structured), but could quickly become expensive since it requires a temporary copy in addition to the work array.
It's probably worth a ~small~ hit to performance to fix this.