[PATCH] D63590: [PowerPC] Sign extend the select instr operands if it is any_extend

[Hal Finkel via Phabricator via llvm-commits]

hfinkel added inline comments.


================
Comment at: llvm/lib/Target/PowerPC/PPCISelLowering.cpp:12325
       else
-        Ops[C+i] = DAG.getAnyExtOrTrunc(Ops[C+i], dl, N->getValueType(0));
+        llvm_unreachable("Unexpected opcode!");
     }
----------------
This seems like we're working around a problem elsewhere. Why does forcing any_extend to become a sign_extend produce better code than just producing an any_extend?


================
Comment at: llvm/test/CodeGen/PowerPC/bool-math.ll:9
-; CHECK-NEXT:    ori 3, 3, 65508
-; CHECK-NEXT:    oris 3, 3, 65535
 ; CHECK-NEXT:    blr
----------------
Is this a regression - it looks like we're going from two instructions to three?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63590/new/

https://reviews.llvm.org/D63590