[flang-commits] [PATCH] D129296: Lower mask intrinsics

[Peter Klausler via Phabricator via flang-commits]

klausler added a comment.

I haven't tested the mask intrinsic functions, but I did note on x86 & power that all compilers return zero from SHIFTR(x, count) when "count" is a negative value not known at compilation time -- except nvfortran which lacks SHIFTR.

The general rule should be to follow precedents for nonconforming behavior f they are common and there isn't much of a performance penalty, since we have to make it easy to port to f18.  We don't know whether there are users that depend on some behavior or not in their existing codes.  Another way to put it is that a code that compiles and runs the same way on multiple existing compilers is portable by definition, whether it's conforming or not, and if it doesn't work with f18 we should at least have a good intentional reason for it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D129296/new/

https://reviews.llvm.org/D129296