[PATCH] D52785: [PseudoSourceValue] New category to represent floating-point status

[Ulrich Weigand via Phabricator via llvm-commits]

uweigand added a comment.

Ping?

How can we get to a decision on whether or not we can get to a place where we're no longer allowed to drop memoperands?

If we indeed cannot use memoperands for floating-point status, should I try looking for another way to encode this?  After looking into the options a bit more, the most natural way might be to add a new MIFlag bit, say FmExcept, that indicates the instruction may raise a floating-point exception flag and/or trap.  (Or maybe two bits to distinguish the two cases.)  In a first implementation, this bit could be used by MachineInstruction::hasUnmodeledSideEffects to return true if set.

Should I try implementing a solution along those lines?


Repository:
  rL LLVM

https://reviews.llvm.org/D52785