Improve MachineMemOperand handling.
 - Allocate MachineMemOperands and MachineMemOperand lists in MachineFunctions.
   This eliminates MachineInstr's std::list member and allows the data to be
   created by isel and live for the remainder of codegen, avoiding a lot of
   copying and unnecessary translation. This also shrinks MemSDNode.
 - Delete MemOperandSDNode. Introduce MachineSDNode which has dedicated
   fields for MachineMemOperands.
 - Change MemSDNode to have a MachineMemOperand member instead of its own
   fields with the same information. This introduces some redundancy, but
   it's more consistent with what MachineInstr will eventually want.
 - Ignore alignment when searching for redundant loads for CSE, but remember
   the greatest alignment.

Target-specific code which previously used MemOperandSDNodes with generic
SDNodes now use MemIntrinsicSDNodes, with opcodes in a designated range
so that the SelectionDAG framework knows that MachineMemOperand information
is available.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@82794 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index addfa9c..71b4062 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -1465,11 +1465,14 @@
   SDValue Tmp0, Tmp1, Tmp2, Tmp3, Tmp4;
   if (!SelectAddr(In1, In1, Tmp0, Tmp1, Tmp2, Tmp3, Tmp4))
     return NULL;
-  SDValue LSI = Node->getOperand(4);    // MemOperand
-  const SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, In2L, In2H, LSI, Chain};
-  return CurDAG->getMachineNode(Opc, Node->getDebugLoc(),
-                                MVT::i32, MVT::i32, MVT::Other, Ops,
-                                array_lengthof(Ops));
+  MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
+  MemOp[0] = cast<MemSDNode>(Node)->getMemOperand();
+  const SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, In2L, In2H, Chain};
+  SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(),
+                                           MVT::i32, MVT::i32, MVT::Other, Ops,
+                                           array_lengthof(Ops));
+  cast<MachineSDNode>(ResNode)->setMemRefs(MemOp, MemOp + 1);
+  return ResNode;
 }
 
 SDNode *X86DAGToDAGISel::SelectAtomicLoadAdd(SDNode *Node, EVT NVT) {
@@ -1605,15 +1608,18 @@
   DebugLoc dl = Node->getDebugLoc();
   SDValue Undef = SDValue(CurDAG->getMachineNode(TargetInstrInfo::IMPLICIT_DEF,
                                                  dl, NVT), 0);
-  SDValue MemOp = CurDAG->getMemOperand(cast<MemSDNode>(Node)->getMemOperand());
+  MachineSDNode::mmo_iterator MemOp = MF->allocateMemRefsArray(1);
+  MemOp[0] = cast<MemSDNode>(Node)->getMemOperand();
   if (isInc || isDec) {
-    SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, MemOp, Chain };
-    SDValue Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops, 7), 0);
+    SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, Chain };
+    SDValue Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops, 6), 0);
+    cast<MachineSDNode>(Ret)->setMemRefs(MemOp, MemOp + 1);
     SDValue RetVals[] = { Undef, Ret };
     return CurDAG->getMergeValues(RetVals, 2, dl).getNode();
   } else {
-    SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, Val, MemOp, Chain };
-    SDValue Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops, 8), 0);
+    SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, Val, Chain };
+    SDValue Ret = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Other, Ops, 7), 0);
+    cast<MachineSDNode>(Ret)->setMemRefs(MemOp, MemOp + 1);
     SDValue RetVals[] = { Undef, Ret };
     return CurDAG->getMergeValues(RetVals, 2, dl).getNode();
   }
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 7ca5ee5..de44adf 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -6983,12 +6983,11 @@
                              Node->getOperand(2), DAG.getIntPtrConstant(0));
   SDValue In2H = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::i32,
                              Node->getOperand(2), DAG.getIntPtrConstant(1));
-  // This is a generalized SDNode, not an AtomicSDNode, so it doesn't
-  // have a MemOperand.  Pass the info through as a normal operand.
-  SDValue LSI = DAG.getMemOperand(cast<MemSDNode>(Node)->getMemOperand());
-  SDValue Ops[] = { Chain, In1, In2L, In2H, LSI };
+  SDValue Ops[] = { Chain, In1, In2L, In2H };
   SDVTList Tys = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other);
-  SDValue Result = DAG.getNode(NewOp, dl, Tys, Ops, 5);
+  SDValue Result =
+    DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops, 4, MVT::i64,
+                            cast<MemSDNode>(Node)->getMemOperand());
   SDValue OpsF[] = { Result.getValue(0), Result.getValue(1)};
   Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, OpsF, 2));
   Results.push_back(Result.getValue(2));
@@ -7396,7 +7395,8 @@
     (*MIB).addOperand(*argOpers[i]);
   MIB.addReg(t2);
   assert(bInstr->hasOneMemOperand() && "Unexpected number of memoperand");
-  (*MIB).addMemOperand(*F, *bInstr->memoperands_begin());
+  (*MIB).setMemRefs(bInstr->memoperands_begin(),
+                    bInstr->memoperands_end());
 
   MIB = BuildMI(newMBB, dl, TII->get(copyOpc), destOper.getReg());
   MIB.addReg(EAXreg);
@@ -7548,7 +7548,8 @@
     (*MIB).addOperand(*argOpers[i]);
 
   assert(bInstr->hasOneMemOperand() && "Unexpected number of memoperand");
-  (*MIB).addMemOperand(*F, *bInstr->memoperands_begin());
+  (*MIB).setMemRefs(bInstr->memoperands_begin(),
+                    bInstr->memoperands_end());
 
   MIB = BuildMI(newMBB, dl, TII->get(copyOpc), t3);
   MIB.addReg(X86::EAX);
@@ -7652,7 +7653,8 @@
     (*MIB).addOperand(*argOpers[i]);
   MIB.addReg(t3);
   assert(mInstr->hasOneMemOperand() && "Unexpected number of memoperand");
-  (*MIB).addMemOperand(*F, *mInstr->memoperands_begin());
+  (*MIB).setMemRefs(mInstr->memoperands_begin(),
+                    mInstr->memoperands_end());
 
   MIB = BuildMI(newMBB, dl, TII->get(X86::MOV32rr), destOper.getReg());
   MIB.addReg(X86::EAX);
@@ -7747,6 +7749,11 @@
   // In the XMM save block, save all the XMM argument registers.
   for (int i = 3, e = MI->getNumOperands(); i != e; ++i) {
     int64_t Offset = (i - 3) * 16 + VarArgsFPOffset;
+    MachineMemOperand *MMO =
+      F->getMachineMemOperand(
+        PseudoSourceValue::getFixedStack(RegSaveFrameIndex),
+        MachineMemOperand::MOStore, Offset,
+        /*Size=*/16, /*Align=*/16);
     BuildMI(XMMSaveMBB, DL, TII->get(X86::MOVAPSmr))
       .addFrameIndex(RegSaveFrameIndex)
       .addImm(/*Scale=*/1)
@@ -7754,10 +7761,7 @@
       .addImm(/*Disp=*/Offset)
       .addReg(/*Segment=*/0)
       .addReg(MI->getOperand(i).getReg())
-      .addMemOperand(MachineMemOperand(
-                       PseudoSourceValue::getFixedStack(RegSaveFrameIndex),
-                       MachineMemOperand::MOStore, Offset,
-                       /*Size=*/16, /*Align=*/16));
+      .addMemOperand(MMO);
   }
 
   F->DeleteMachineInstr(MI);   // The pseudo instruction is gone now.
diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 440742d..2f7b8ba 100644
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -204,17 +204,6 @@
       LCMPXCHG_DAG,
       LCMPXCHG8_DAG,
 
-      // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG, 
-      // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG - 
-      // Atomic 64-bit binary operations.
-      ATOMADD64_DAG,
-      ATOMSUB64_DAG,
-      ATOMOR64_DAG,
-      ATOMXOR64_DAG,
-      ATOMAND64_DAG,
-      ATOMNAND64_DAG,
-      ATOMSWAP64_DAG,
-
       // FNSTCW16m - Store FP control world into i16 memory.
       FNSTCW16m,
 
@@ -248,7 +237,18 @@
       // VASTART_SAVE_XMM_REGS - Save xmm argument registers to the stack,
       // according to %al. An operator is needed so that this can be expanded
       // with control flow.
-      VASTART_SAVE_XMM_REGS
+      VASTART_SAVE_XMM_REGS,
+
+      // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG, 
+      // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG - 
+      // Atomic 64-bit binary operations.
+      ATOMADD64_DAG = ISD::FIRST_TARGET_MEMORY_OPCODE,
+      ATOMSUB64_DAG,
+      ATOMOR64_DAG,
+      ATOMXOR64_DAG,
+      ATOMAND64_DAG,
+      ATOMNAND64_DAG,
+      ATOMSWAP64_DAG
     };
   }
 
diff --git a/lib/Target/X86/X86InstrBuilder.h b/lib/Target/X86/X86InstrBuilder.h
index 6aae2e8..c475b56 100644
--- a/lib/Target/X86/X86InstrBuilder.h
+++ b/lib/Target/X86/X86InstrBuilder.h
@@ -26,6 +26,7 @@
 
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/PseudoSourceValue.h"
 
 namespace llvm {
@@ -142,11 +143,11 @@
     Flags |= MachineMemOperand::MOLoad;
   if (TID.mayStore())
     Flags |= MachineMemOperand::MOStore;
-  MachineMemOperand MMO(PseudoSourceValue::getFixedStack(FI),
-                        Flags,
-                        Offset,
-                        MFI.getObjectSize(FI),
-                        MFI.getObjectAlignment(FI));
+  MachineMemOperand *MMO =
+    MF.getMachineMemOperand(PseudoSourceValue::getFixedStack(FI),
+                            Flags, Offset,
+                            MFI.getObjectSize(FI),
+                            MFI.getObjectAlignment(FI));
   return addOffset(MIB.addFrameIndex(FI), Offset)
             .addMemOperand(MMO);
 }
diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index d94b741..12c4b9c 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -2296,7 +2296,7 @@
   // Determine the alignment of the load.
   unsigned Alignment = 0;
   if (LoadMI->hasOneMemOperand())
-    Alignment = LoadMI->memoperands_begin()->getAlignment();
+    Alignment = (*LoadMI->memoperands_begin())->getAlignment();
   else
     switch (LoadMI->getOpcode()) {
     case X86::V_SET0:
@@ -2567,7 +2567,7 @@
   std::vector<SDValue> AfterOps;
   DebugLoc dl = N->getDebugLoc();
   unsigned NumOps = N->getNumOperands();
-  for (unsigned i = 0; i != NumOps-2; ++i) {
+  for (unsigned i = 0; i != NumOps-1; ++i) {
     SDValue Op = N->getOperand(i);
     if (i >= Index-NumDefs && i < Index-NumDefs + X86AddrNumOperands)
       AddrOps.push_back(Op);
@@ -2576,8 +2576,6 @@
     else if (i > Index-NumDefs)
       AfterOps.push_back(Op);
   }
-  SDValue MemOp = N->getOperand(NumOps-2);
-  AddrOps.push_back(MemOp);
   SDValue Chain = N->getOperand(NumOps-1);
   AddrOps.push_back(Chain);
 
@@ -2615,9 +2613,7 @@
   // Emit the store instruction.
   if (FoldedStore) {
     AddrOps.pop_back();
-    AddrOps.pop_back();
     AddrOps.push_back(SDValue(NewNode, 0));
-    AddrOps.push_back(MemOp);
     AddrOps.push_back(Chain);
     bool isAligned = (RI.getStackAlignment() >= 16) ||
       RI.needsStackRealignment(MF);