[PATCH] D42449: [MachineVerifier] Add check that renamable operands aren't reserved registers.

[Geoff Berry via Phabricator via llvm-commits]

gberry added inline comments.


================
Comment at: lib/CodeGen/MachineVerifier.cpp:1111
+        if (MRI->isReserved(Reg))
+          report("isRenamable set on reserved register", MO, MONum);
         return;
----------------
qcolombet wrote:
> I know we already touched that in the other thread, but are we definitely giving up on all optimizations as soon as a MachineOperand has been assigned a reserved register?
> 
> I guess we could say that given the liveness of reserved registers is not necessarily properly modeled, when such register is assigned to an operand, it won't be safe later on to change it, so that's expected.
Yes, I agree with your answer to your question here :)


Repository:
  rL LLVM

https://reviews.llvm.org/D42449