[PATCH] D40554: [PowerPC] Fix bugs in sign-/zero-extension elimination

[Nemanja Ivanovic via Phabricator via llvm-commits]

nemanjai added inline comments.


================
Comment at: lib/Target/PowerPC/PPCInstrInfo.h:343
 
-  bool isSignOrZeroExtended(const MachineInstr &MI, bool SignExt,
-                            const unsigned PhiDepth) const;
+  bool isSignOrZeroExtended(const unsigned Reg, bool SignExt,
+                            const unsigned PhiDepth,
----------------
inouehrs wrote:
> nemanjai wrote:
> > I still think this setup with separate queries for sign/zero/any extend seems unnecessarily complex. Seems like it would be clearer if we had an enum for the extension type (sign, zero, none) and we just have a single query that will return this. Therefore no need for Boolean parameters, no need for multiple functions, etc.
> I agree that the single query approach is cleaner. But I am afraid that it will increase the cost of compilation slightly since it always checks for both sign- and zero-extensions. Do you think the potential increase in compilation cost does not matter too much here?
It seems to me that the checks are not costly except in chained operations cases (which we traverse the chain as it is). Or am I missing something. What is the worst case complexity of the computations for each?


https://reviews.llvm.org/D40554