[llvm-commits] [llvm] r97797 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/tailcall2.ll

[Evan Cheng]

Author: evancheng
Date: Fri Mar  5 02:38:04 2010
New Revision: 97797

URL: http://llvm.org/viewvc/llvm-project?rev=97797&view=rev
Log:
Fix an oops in x86 sibcall optimization. If the ByVal callee argument is itself passed as a pointer, then it's obviously not safe to do a tail call.

Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
    llvm/trunk/test/CodeGen/X86/tailcall2.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=97797&r1=97796&r2=97797&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Fri Mar  5 02:38:04 2010
@@ -2235,7 +2235,8 @@
 bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags,
                          MachineFrameInfo *MFI, const MachineRegisterInfo *MRI,
                          const X86InstrInfo *TII) {
-  int FI;
+  unsigned Bytes = Arg.getValueType().getSizeInBits() / 8;
+  int FI = INT_MAX;
   if (Arg.getOpcode() == ISD::CopyFromReg) {
     unsigned VR = cast<RegisterSDNode>(Arg.getOperand(1))->getReg();
     if (!VR || TargetRegisterInfo::isPhysicalRegister(VR))
@@ -2251,25 +2252,30 @@
       if ((Opcode == X86::LEA32r || Opcode == X86::LEA64r) &&
           Def->getOperand(1).isFI()) {
         FI = Def->getOperand(1).getIndex();
-        if (MFI->getObjectSize(FI) != Flags.getByValSize())
-          return false;
+        Bytes = Flags.getByValSize();
       } else
         return false;
     }
-  } else {
-    LoadSDNode *Ld = dyn_cast<LoadSDNode>(Arg);
-    if (!Ld)
+  } else if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Arg)) {
+    if (Flags.isByVal())
+      // ByVal argument is passed in as a pointer but it's now being
+      // derefernced. e.g.
+      // define @foo(%struct.X* %A) {
+      //   tail call @bar(%struct.X* byval %A)
+      // }
       return false;
     SDValue Ptr = Ld->getBasePtr();
     FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(Ptr);
     if (!FINode)
       return false;
     FI = FINode->getIndex();
-  }
+  } else
+    return false;
 
+  assert(FI != INT_MAX);
   if (!MFI->isFixedObjectIndex(FI))
     return false;
-  return Offset == MFI->getObjectOffset(FI);
+  return Offset == MFI->getObjectOffset(FI) && Bytes == MFI->getObjectSize(FI);
 }
 
 /// IsEligibleForTailCallOptimization - Check whether the call is eligible

Modified: llvm/trunk/test/CodeGen/X86/tailcall2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/tailcall2.ll?rev=97797&r1=97796&r2=97797&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/tailcall2.ll (original)
+++ llvm/trunk/test/CodeGen/X86/tailcall2.ll Fri Mar  5 02:38:04 2010
@@ -195,3 +195,24 @@
 }
 
 declare i32 @foo6(i32, i32, %struct.t* byval align 4)
+
+; rdar://r7717598
+%struct.ns = type { i32, i32 }
+%struct.cp = type { float, float }
+
+define %struct.ns* @t13(%struct.cp* %yy) nounwind ssp {
+; 32: t13:
+; 32-NOT: jmp
+; 32: call
+; 32: ret
+
+; 64: t13:
+; 64-NOT: jmp
+; 64: call
+; 64: ret
+entry:
+  %0 = tail call fastcc %struct.ns* @foo7(%struct.cp* byval align 4 %yy, i8 signext 0) nounwind
+  ret %struct.ns* %0
+}
+
+declare fastcc %struct.ns* @foo7(%struct.cp* byval align 4, i8 signext) nounwind ssp