[PATCH] D11817: Enable Shrink Wrapping for PPC64.

[hfinkel@anl.gov via llvm-commits]

hfinkel added inline comments.

================
Comment at: lib/Target/PowerPC/PPCFrameLowering.cpp:1002
@@ -1011,20 +1001,3 @@
 
-  bool UsesTCRet =  RetOpcode == PPC::TCRETURNri ||
-    RetOpcode == PPC::TCRETURNdi ||
-    RetOpcode == PPC::TCRETURNai ||
-    RetOpcode == PPC::TCRETURNri8 ||
-    RetOpcode == PPC::TCRETURNdi8 ||
-    RetOpcode == PPC::TCRETURNai8;
-
-  if (UsesTCRet) {
-    int MaxTCRetDelta = FI->getTailCallSPDelta();
-    MachineOperand &StackAdjust = MBBI->getOperand(1);
-    assert(StackAdjust.isImm() && "Expecting immediate value.");
-    // Adjust stack pointer.
-    int StackAdj = StackAdjust.getImm();
-    int Delta = StackAdj - MaxTCRetDelta;
-    assert((Delta >= 0) && "Delta must be positive");
-    if (MaxTCRetDelta>0)
-      FrameSize += (StackAdj +Delta);
-    else
-      FrameSize += StackAdj;
+  bool IsRB = MBBI->isReturn();
+  
----------------
Does RB stand for Return Block? Please use a more verbose name.

================
Comment at: lib/Target/PowerPC/PPCTargetMachine.cpp:388
@@ +387,3 @@
+bool PPCPassConfig::runShrinkWrap(const MachineFunction &Fn) const {
+  return (EnableShrinkWrapOpt != cl::BOU_FALSE && 
+          Fn.getSubtarget<PPCSubtarget>().isSVR4ABI() &&
----------------
Don't add a dependency on EnableShrinkWrapOpt here. We would need to duplicate this checking logic in every target, and that's not a good design. (and the command-line argument itself is an internal detail of ShrinkWrap.cpp). Please have the code in ShrinkWrap.cpp (or Passes.cpp) check this before calling the target specific callback.



http://reviews.llvm.org/D11817