Implement an x86-64 ABI detail of passing structs by hidden first
argument. The x86-64 ABI requires the incoming value of %rdi to
be copied to %rax on exit from a function that is returning a
large C struct.

Also, add a README-X86-64 entry detailing the missed optimization
opportunity and proposing an alternative approach.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@50075 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/README-X86-64.txt b/lib/Target/X86/README-X86-64.txt
index 359b83d..594a66f 100644
--- a/lib/Target/X86/README-X86-64.txt
+++ b/lib/Target/X86/README-X86-64.txt
@@ -236,3 +236,24 @@
 
 //===---------------------------------------------------------------------===//
 
+The x86-64 ABI for hidden-argument struct returns requires that the
+incoming value of %rdi be copied into %rax by the callee upon return.
+
+The idea is that it saves callers from having to remember this value,
+which would often require a callee-saved register. Callees usually
+need to keep this value live for most of their body anyway, so it
+doesn't add a significant burden on them.
+
+We currently implement this in codegen, however this is suboptimal
+because it means that it would be quite awkward to implement the
+optimization for callers.
+
+A better implementation would be to relax the LLVM IR rules for sret
+arguments to allow a function with an sret argument to have a non-void
+return type, and to have the front-end to set up the sret argument value
+as the return value of the function. The front-end could more easily
+emit uses of the returned struct value to be in terms of the function's
+lowered return value, and it would free non-C frontends from a
+complication only required by a C-based ABI.
+
+//===---------------------------------------------------------------------===//
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 3a4dc23..9ba1f53 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -875,6 +875,25 @@
     Chain = DAG.getCopyToReg(Chain, VA.getLocReg(), ValToCopy, Flag);
     Flag = Chain.getValue(1);
   }
+
+  // The x86-64 ABI for returning structs by value requires that we copy
+  // the sret argument into %rax for the return. We saved the argument into
+  // a virtual register in the entry block, so now we copy the value out
+  // and into %rax.
+  if (Subtarget->is64Bit() &&
+      DAG.getMachineFunction().getFunction()->hasStructRetAttr()) {
+    MachineFunction &MF = DAG.getMachineFunction();
+    X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
+    unsigned Reg = FuncInfo->getSRetReturnReg();
+    if (!Reg) {
+      Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(MVT::i64));
+      FuncInfo->setSRetReturnReg(Reg);
+    }
+    SDOperand Val = DAG.getCopyFromReg(Chain, Reg, getPointerTy());
+
+    Chain = DAG.getCopyToReg(Chain, X86::RAX, Val, Flag);
+    Flag = Chain.getValue(1);
+  }
   
   RetOps[0] = Chain;  // Update chain.
 
@@ -1225,6 +1244,21 @@
     }
   }
 
+  // The x86-64 ABI for returning structs by value requires that we copy
+  // the sret argument into %rax for the return. Save the argument into
+  // a virtual register so that we can access it from the return points.
+  if (Is64Bit && DAG.getMachineFunction().getFunction()->hasStructRetAttr()) {
+    MachineFunction &MF = DAG.getMachineFunction();
+    X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
+    unsigned Reg = FuncInfo->getSRetReturnReg();
+    if (!Reg) {
+      Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(MVT::i64));
+      FuncInfo->setSRetReturnReg(Reg);
+    }
+    SDOperand Copy = DAG.getCopyToReg(DAG.getEntryNode(), Reg, ArgValues[0]);
+    Root = DAG.getNode(ISD::TokenFactor, MVT::Other, Copy, Root);
+  }
+
   unsigned StackSize = CCInfo.getNextStackOffset();
   // align stack specially for tail calls
   if (CC == CallingConv::Fast)
diff --git a/lib/Target/X86/X86MachineFunctionInfo.h b/lib/Target/X86/X86MachineFunctionInfo.h
index fcdeb05..b5c9caf 100644
--- a/lib/Target/X86/X86MachineFunctionInfo.h
+++ b/lib/Target/X86/X86MachineFunctionInfo.h
@@ -53,20 +53,27 @@
   /// the returnaddr can be savely move to this area
   int TailCallReturnAddrDelta;
 
+  /// SRetReturnReg - Some subtargets require that sret lowering includes
+  /// returning the value of the returned struct in a register. This field
+  /// holds the virtual register into which the sret argument is passed.
+  unsigned SRetReturnReg;
+
 public:
   X86MachineFunctionInfo() : ForceFramePointer(false),
                              CalleeSavedFrameSize(0),
                              BytesToPopOnReturn(0),
                              DecorationStyle(None),
                              ReturnAddrIndex(0),
-                             TailCallReturnAddrDelta(0) {}
+                             TailCallReturnAddrDelta(0),
+                             SRetReturnReg(0) {}
   
   X86MachineFunctionInfo(MachineFunction &MF) : ForceFramePointer(false),
                                                 CalleeSavedFrameSize(0),
                                                 BytesToPopOnReturn(0),
                                                 DecorationStyle(None),
                                                 ReturnAddrIndex(0),
-                                                TailCallReturnAddrDelta(0) {}
+                                                TailCallReturnAddrDelta(0),
+                                                SRetReturnReg(0) {}
   
   bool getForceFramePointer() const { return ForceFramePointer;} 
   void setForceFramePointer(bool forceFP) { ForceFramePointer = forceFP; }
@@ -85,6 +92,9 @@
 
   int getTCReturnAddrDelta() const { return TailCallReturnAddrDelta; }
   void setTCReturnAddrDelta(int delta) {TailCallReturnAddrDelta = delta;}
+
+  unsigned getSRetReturnReg() const { return SRetReturnReg; }
+  void setSRetReturnReg(unsigned Reg) { SRetReturnReg = Reg; }
 };
 } // End llvm namespace