Move the complex address expression out of DIVariable and into an extra
argument of the llvm.dbg.declare/llvm.dbg.value intrinsics.

Previously, DIVariable was a variable-length field that has an optional
reference to a Metadata array consisting of a variable number of
complex address expressions. In the case of OpPiece expressions this is
wasting a lot of storage in IR, because when an aggregate type is, e.g.,
SROA'd into all of its n individual members, the IR will contain n copies
of the DIVariable, all alike, only differing in the complex address
reference at the end.

By making the complex address into an extra argument of the
dbg.value/dbg.declare intrinsics, all of the pieces can reference the
same variable and the complex address expressions can be uniqued across
the CU, too.
Down the road, this will allow us to move other flags, such as
"indirection" out of the DIVariable, too.

The new intrinsics look like this:
declare void @llvm.dbg.declare(metadata %storage, metadata %var, metadata %expr)
declare void @llvm.dbg.value(metadata %storage, i64 %offset, metadata %var, metadata %expr)

This patch adds a new LLVM-local tag to DIExpressions, so we can detect
and pretty-print DIExpression metadata nodes.

What this patch doesn't do:

This patch does not touch the "Indirect" field in DIVariable; but moving
that into the expression would be a natural next step.

http://reviews.llvm.org/D4919
rdar://problem/17994491

Thanks to dblaikie and dexonsmith for reviewing this patch!

Note: I accidentally committed a bogus older version of this patch previously.
llvm-svn: 218787
diff --git a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
index 7b656bf..72390a9 100644
--- a/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/FastISel.cpp
@@ -1125,13 +1125,14 @@
         Op->setIsDebug(true);
         BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                 TII.get(TargetOpcode::DBG_VALUE), false, Op->getReg(), 0,
-                DI->getVariable());
+                DI->getVariable(), DI->getExpression());
       } else
         BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                 TII.get(TargetOpcode::DBG_VALUE))
             .addOperand(*Op)
             .addImm(0)
-            .addMetadata(DI->getVariable());
+            .addMetadata(DI->getVariable())
+            .addMetadata(DI->getExpression());
     } else {
       // We can't yet handle anything else here because it would require
       // generating code, thus altering codegen because of debug info.
@@ -1150,28 +1151,32 @@
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, II)
           .addReg(0U)
           .addImm(DI->getOffset())
-          .addMetadata(DI->getVariable());
+          .addMetadata(DI->getVariable())
+          .addMetadata(DI->getExpression());
     } else if (const auto *CI = dyn_cast<ConstantInt>(V)) {
       if (CI->getBitWidth() > 64)
         BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, II)
             .addCImm(CI)
             .addImm(DI->getOffset())
-            .addMetadata(DI->getVariable());
+            .addMetadata(DI->getVariable())
+            .addMetadata(DI->getExpression());
       else
         BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, II)
             .addImm(CI->getZExtValue())
             .addImm(DI->getOffset())
-            .addMetadata(DI->getVariable());
+            .addMetadata(DI->getVariable())
+            .addMetadata(DI->getExpression());
     } else if (const auto *CF = dyn_cast<ConstantFP>(V)) {
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, II)
           .addFPImm(CF)
           .addImm(DI->getOffset())
-          .addMetadata(DI->getVariable());
+          .addMetadata(DI->getVariable())
+          .addMetadata(DI->getExpression());
     } else if (unsigned Reg = lookUpRegForValue(V)) {
       // FIXME: This does not handle register-indirect values at offset 0.
       bool IsIndirect = DI->getOffset() != 0;
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, II, IsIndirect, Reg,
-              DI->getOffset(), DI->getVariable());
+              DI->getOffset(), DI->getVariable(), DI->getExpression());
     } else {
       // We can't yet handle anything else here because it would require
       // generating code, thus altering codegen because of debug info.
@@ -2181,4 +2186,4 @@
   }
 
   return Predicate;
-}
\ No newline at end of file
+}
diff --git a/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp b/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
index c06612d..d1fcbe5 100644
--- a/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
@@ -197,7 +197,7 @@
                 StaticAllocaMap.find(AI);
               if (SI != StaticAllocaMap.end()) { // Check for VLAs.
                 int FI = SI->second;
-                MMI.setVariableDbgInfo(DI->getVariable(),
+                MMI.setVariableDbgInfo(DI->getVariable(), DI->getExpression(),
                                        FI, DI->getDebugLoc());
               }
             }
diff --git a/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp b/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
index 4d6de12..32d5ddc 100644
--- a/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
@@ -649,14 +649,18 @@
 InstrEmitter::EmitDbgValue(SDDbgValue *SD,
                            DenseMap<SDValue, unsigned> &VRBaseMap) {
   uint64_t Offset = SD->getOffset();
-  MDNode* MDPtr = SD->getMDPtr();
+  MDNode *Var = SD->getVariable();
+  MDNode *Expr = SD->getExpression();
   DebugLoc DL = SD->getDebugLoc();
 
   if (SD->getKind() == SDDbgValue::FRAMEIX) {
     // Stack address; this needs to be lowered in target-dependent fashion.
     // EmitTargetCodeForFrameDebugValue is responsible for allocation.
     return BuildMI(*MF, DL, TII->get(TargetOpcode::DBG_VALUE))
-        .addFrameIndex(SD->getFrameIx()).addImm(Offset).addMetadata(MDPtr);
+        .addFrameIndex(SD->getFrameIx())
+        .addImm(Offset)
+        .addMetadata(Var)
+        .addMetadata(Expr);
   }
   // Otherwise, we're going to create an instruction here.
   const MCInstrDesc &II = TII->get(TargetOpcode::DBG_VALUE);
@@ -702,7 +706,8 @@
     MIB.addReg(0U, RegState::Debug);
   }
 
-  MIB.addMetadata(MDPtr);
+  MIB.addMetadata(Var);
+  MIB.addMetadata(Expr);
 
   return &*MIB;
 }
diff --git a/llvm/lib/CodeGen/SelectionDAG/SDNodeDbgValue.h b/llvm/lib/CodeGen/SelectionDAG/SDNodeDbgValue.h
index 307808e..c8bb3d8 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SDNodeDbgValue.h
+++ b/llvm/lib/CodeGen/SelectionDAG/SDNodeDbgValue.h
@@ -44,7 +44,8 @@
     const Value *Const;     // valid for constants
     unsigned FrameIx;       // valid for stack objects
   } u;
-  MDNode *mdPtr;
+  MDNode *Var;
+  MDNode *Expr;
   bool IsIndirect;
   uint64_t Offset;
   DebugLoc DL;
@@ -52,29 +53,29 @@
   bool Invalid;
 public:
   // Constructor for non-constants.
-  SDDbgValue(MDNode *mdP, SDNode *N, unsigned R,
-	     bool indir, uint64_t off, DebugLoc dl,
-             unsigned O) : mdPtr(mdP), IsIndirect(indir),
-			   Offset(off), DL(dl), Order(O),
-                           Invalid(false) {
+  SDDbgValue(MDNode *Var, MDNode *Expr, SDNode *N, unsigned R, bool indir,
+             uint64_t off, DebugLoc dl, unsigned O)
+      : Var(Var), Expr(Expr), IsIndirect(indir), Offset(off), DL(dl), Order(O),
+        Invalid(false) {
     kind = SDNODE;
     u.s.Node = N;
     u.s.ResNo = R;
   }
 
   // Constructor for constants.
-  SDDbgValue(MDNode *mdP, const Value *C, uint64_t off, DebugLoc dl,
-             unsigned O) : 
-    mdPtr(mdP), IsIndirect(false), Offset(off), DL(dl), Order(O),
-    Invalid(false) {
+  SDDbgValue(MDNode *Var, MDNode *Expr, const Value *C, uint64_t off,
+             DebugLoc dl, unsigned O)
+      : Var(Var), Expr(Expr), IsIndirect(false), Offset(off), DL(dl), Order(O),
+        Invalid(false) {
     kind = CONST;
     u.Const = C;
   }
 
   // Constructor for frame indices.
-  SDDbgValue(MDNode *mdP, unsigned FI, uint64_t off, DebugLoc dl, unsigned O) : 
-    mdPtr(mdP), IsIndirect(false), Offset(off), DL(dl), Order(O),
-    Invalid(false) {
+  SDDbgValue(MDNode *Var, MDNode *Expr, unsigned FI, uint64_t off, DebugLoc dl,
+             unsigned O)
+      : Var(Var), Expr(Expr), IsIndirect(false), Offset(off), DL(dl), Order(O),
+        Invalid(false) {
     kind = FRAMEIX;
     u.FrameIx = FI;
   }
@@ -82,8 +83,11 @@
   // Returns the kind.
   DbgValueKind getKind() { return kind; }
 
-  // Returns the MDNode pointer.
-  MDNode *getMDPtr() { return mdPtr; }
+  // Returns the MDNode pointer for the variable.
+  MDNode *getVariable() { return Var; }
+
+  // Returns the MDNode pointer for the expression.
+  MDNode *getExpression() { return Expr; }
 
   // Returns the SDNode* for a register ref
   SDNode *getSDNode() { assert (kind==SDNODE); return u.s.Node; }
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index e2ead47..2322ce0 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5764,26 +5764,24 @@
 /// getDbgValue - Creates a SDDbgValue node.
 ///
 /// SDNode
-SDDbgValue *
-SelectionDAG::getDbgValue(MDNode *MDPtr, SDNode *N, unsigned R,
-			  bool IsIndirect, uint64_t Off,
-                          DebugLoc DL, unsigned O) {
-  return new (Allocator) SDDbgValue(MDPtr, N, R, IsIndirect, Off, DL, O);
+SDDbgValue *SelectionDAG::getDbgValue(MDNode *Var, MDNode *Expr, SDNode *N,
+                                      unsigned R, bool IsIndirect, uint64_t Off,
+                                      DebugLoc DL, unsigned O) {
+  return new (Allocator) SDDbgValue(Var, Expr, N, R, IsIndirect, Off, DL, O);
 }
 
 /// Constant
-SDDbgValue *
-SelectionDAG::getConstantDbgValue(MDNode *MDPtr, const Value *C,
-				  uint64_t Off,
-				  DebugLoc DL, unsigned O) {
-  return new (Allocator) SDDbgValue(MDPtr, C, Off, DL, O);
+SDDbgValue *SelectionDAG::getConstantDbgValue(MDNode *Var, MDNode *Expr,
+                                              const Value *C, uint64_t Off,
+                                              DebugLoc DL, unsigned O) {
+  return new (Allocator) SDDbgValue(Var, Expr, C, Off, DL, O);
 }
 
 /// FrameIndex
-SDDbgValue *
-SelectionDAG::getFrameIndexDbgValue(MDNode *MDPtr, unsigned FI, uint64_t Off,
-				    DebugLoc DL, unsigned O) {
-  return new (Allocator) SDDbgValue(MDPtr, FI, Off, DL, O);
+SDDbgValue *SelectionDAG::getFrameIndexDbgValue(MDNode *Var, MDNode *Expr,
+                                                unsigned FI, uint64_t Off,
+                                                DebugLoc DL, unsigned O) {
+  return new (Allocator) SDDbgValue(Var, Expr, FI, Off, DL, O);
 }
 
 namespace {
@@ -6189,10 +6187,10 @@
        I != E; ++I) {
     SDDbgValue *Dbg = *I;
     if (Dbg->getKind() == SDDbgValue::SDNODE) {
-      SDDbgValue *Clone = getDbgValue(Dbg->getMDPtr(), ToNode, To.getResNo(),
-				      Dbg->isIndirect(),
-                                      Dbg->getOffset(), Dbg->getDebugLoc(),
-                                      Dbg->getOrder());
+      SDDbgValue *Clone =
+          getDbgValue(Dbg->getVariable(), Dbg->getExpression(), ToNode,
+                      To.getResNo(), Dbg->isIndirect(), Dbg->getOffset(),
+                      Dbg->getDebugLoc(), Dbg->getOrder());
       ClonedDVs.push_back(Clone);
     }
   }
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 6a8cd2e..9bd5dfea 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -993,15 +993,16 @@
     DebugLoc dl = DDI.getdl();
     unsigned DbgSDNodeOrder = DDI.getSDNodeOrder();
     MDNode *Variable = DI->getVariable();
+    MDNode *Expr = DI->getExpression();
     uint64_t Offset = DI->getOffset();
     // A dbg.value for an alloca is always indirect.
     bool IsIndirect = isa<AllocaInst>(V) || Offset != 0;
     SDDbgValue *SDV;
     if (Val.getNode()) {
-      if (!EmitFuncArgumentDbgValue(V, Variable, Offset, IsIndirect, Val)) {
-        SDV = DAG.getDbgValue(Variable, Val.getNode(),
-                              Val.getResNo(), IsIndirect,
-			      Offset, dl, DbgSDNodeOrder);
+      if (!EmitFuncArgumentDbgValue(V, Variable, Expr, Offset, IsIndirect,
+                                    Val)) {
+        SDV = DAG.getDbgValue(Variable, Expr, Val.getNode(), Val.getResNo(),
+                              IsIndirect, Offset, dl, DbgSDNodeOrder);
         DAG.AddDbgValue(SDV, Val.getNode(), false);
       }
     } else
@@ -4603,10 +4604,11 @@
 /// EmitFuncArgumentDbgValue - If the DbgValueInst is a dbg_value of a function
 /// argument, create the corresponding DBG_VALUE machine instruction for it now.
 /// At the end of instruction selection, they will be inserted to the entry BB.
-bool
-SelectionDAGBuilder::EmitFuncArgumentDbgValue(const Value *V, MDNode *Variable,
-                                              int64_t Offset, bool IsIndirect,
-                                              const SDValue &N) {
+bool SelectionDAGBuilder::EmitFuncArgumentDbgValue(const Value *V,
+                                                   MDNode *Variable,
+                                                   MDNode *Expr, int64_t Offset,
+                                                   bool IsIndirect,
+                                                   const SDValue &N) {
   const Argument *Arg = dyn_cast<Argument>(V);
   if (!Arg)
     return false;
@@ -4658,14 +4660,16 @@
     return false;
 
   if (Op->isReg())
-    FuncInfo.ArgDbgValues.push_back(BuildMI(MF, getCurDebugLoc(),
-                                            TII->get(TargetOpcode::DBG_VALUE),
-                                            IsIndirect,
-                                            Op->getReg(), Offset, Variable));
+    FuncInfo.ArgDbgValues.push_back(
+        BuildMI(MF, getCurDebugLoc(), TII->get(TargetOpcode::DBG_VALUE),
+                IsIndirect, Op->getReg(), Offset, Variable, Expr));
   else
     FuncInfo.ArgDbgValues.push_back(
-      BuildMI(MF, getCurDebugLoc(), TII->get(TargetOpcode::DBG_VALUE))
-          .addOperand(*Op).addImm(Offset).addMetadata(Variable));
+        BuildMI(MF, getCurDebugLoc(), TII->get(TargetOpcode::DBG_VALUE))
+            .addOperand(*Op)
+            .addImm(Offset)
+            .addMetadata(Variable)
+            .addMetadata(Expr));
 
   return true;
 }
@@ -4785,6 +4789,7 @@
   case Intrinsic::dbg_declare: {
     const DbgDeclareInst &DI = cast<DbgDeclareInst>(I);
     MDNode *Variable = DI.getVariable();
+    MDNode *Expression = DI.getExpression();
     const Value *Address = DI.getAddress();
     DIVariable DIVar(Variable);
     assert((!DIVar || DIVar.isVariable()) &&
@@ -4820,16 +4825,16 @@
         FrameIndexSDNode *FINode = dyn_cast<FrameIndexSDNode>(N.getNode());
         if (FINode)
           // Byval parameter.  We have a frame index at this point.
-          SDV = DAG.getFrameIndexDbgValue(Variable, FINode->getIndex(),
-					  0, dl, SDNodeOrder);
+          SDV = DAG.getFrameIndexDbgValue(
+              Variable, Expression, FINode->getIndex(), 0, dl, SDNodeOrder);
         else {
           // Address is an argument, so try to emit its dbg value using
           // virtual register info from the FuncInfo.ValueMap.
-          EmitFuncArgumentDbgValue(Address, Variable, 0, false, N);
+          EmitFuncArgumentDbgValue(Address, Variable, Expression, 0, false, N);
           return nullptr;
         }
       } else if (AI)
-        SDV = DAG.getDbgValue(Variable, N.getNode(), N.getResNo(),
+        SDV = DAG.getDbgValue(Variable, Expression, N.getNode(), N.getResNo(),
                               true, 0, dl, SDNodeOrder);
       else {
         // Can't do anything with other non-AI cases yet.
@@ -4842,7 +4847,8 @@
     } else {
       // If Address is an argument then try to emit its dbg value using
       // virtual register info from the FuncInfo.ValueMap.
-      if (!EmitFuncArgumentDbgValue(Address, Variable, 0, false, N)) {
+      if (!EmitFuncArgumentDbgValue(Address, Variable, Expression, 0, false,
+                                    N)) {
         // If variable is pinned by a alloca in dominating bb then
         // use StaticAllocaMap.
         if (const AllocaInst *AI = dyn_cast<AllocaInst>(Address)) {
@@ -4850,7 +4856,7 @@
             DenseMap<const AllocaInst*, int>::iterator SI =
               FuncInfo.StaticAllocaMap.find(AI);
             if (SI != FuncInfo.StaticAllocaMap.end()) {
-              SDV = DAG.getFrameIndexDbgValue(Variable, SI->second,
+              SDV = DAG.getFrameIndexDbgValue(Variable, Expression, SI->second,
                                               0, dl, SDNodeOrder);
               DAG.AddDbgValue(SDV, nullptr, false);
               return nullptr;
@@ -4871,6 +4877,7 @@
       return nullptr;
 
     MDNode *Variable = DI.getVariable();
+    MDNode *Expression = DI.getExpression();
     uint64_t Offset = DI.getOffset();
     const Value *V = DI.getValue();
     if (!V)
@@ -4878,7 +4885,8 @@
 
     SDDbgValue *SDV;
     if (isa<ConstantInt>(V) || isa<ConstantFP>(V) || isa<UndefValue>(V)) {
-      SDV = DAG.getConstantDbgValue(Variable, V, Offset, dl, SDNodeOrder);
+      SDV = DAG.getConstantDbgValue(Variable, Expression, V, Offset, dl,
+                                    SDNodeOrder);
       DAG.AddDbgValue(SDV, nullptr, false);
     } else {
       // Do not use getValue() in here; we don't want to generate code at
@@ -4890,10 +4898,10 @@
       if (N.getNode()) {
         // A dbg.value for an alloca is always indirect.
         bool IsIndirect = isa<AllocaInst>(V) || Offset != 0;
-        if (!EmitFuncArgumentDbgValue(V, Variable, Offset, IsIndirect, N)) {
-          SDV = DAG.getDbgValue(Variable, N.getNode(),
-                                N.getResNo(), IsIndirect,
-				Offset, dl, SDNodeOrder);
+        if (!EmitFuncArgumentDbgValue(V, Variable, Expression, Offset,
+                                      IsIndirect, N)) {
+          SDV = DAG.getDbgValue(Variable, Expression, N.getNode(), N.getResNo(),
+                                IsIndirect, Offset, dl, SDNodeOrder);
           DAG.AddDbgValue(SDV, N.getNode(), false);
         }
       } else if (!V->use_empty() ) {
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
index 45dd554..6ecddc1 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
@@ -791,7 +791,7 @@
   /// EmitFuncArgumentDbgValue - If V is an function argument then create
   /// corresponding DBG_VALUE machine instruction for it now. At the end of
   /// instruction selection, they will be inserted to the entry BB.
-  bool EmitFuncArgumentDbgValue(const Value *V, MDNode *Variable,
+  bool EmitFuncArgumentDbgValue(const Value *V, MDNode *Variable, MDNode *Expr,
                                 int64_t Offset, bool IsIndirect,
                                 const SDValue &N);
 };
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
index 4ef3e6f..df2c6b8 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
@@ -486,15 +486,14 @@
                        "- add if needed");
       MachineInstr *Def = RegInfo->getVRegDef(LDI->second);
       MachineBasicBlock::iterator InsertPos = Def;
-      const MDNode *Variable =
-        MI->getOperand(MI->getNumOperands()-1).getMetadata();
+      const MDNode *Variable = MI->getDebugVariable();
+      const MDNode *Expr = MI->getDebugExpression();
       bool IsIndirect = MI->isIndirectDebugValue();
       unsigned Offset = IsIndirect ? MI->getOperand(1).getImm() : 0;
       // Def is never a terminator here, so it is ok to increment InsertPos.
       BuildMI(*EntryMBB, ++InsertPos, MI->getDebugLoc(),
-              TII.get(TargetOpcode::DBG_VALUE),
-              IsIndirect,
-              LDI->second, Offset, Variable);
+              TII.get(TargetOpcode::DBG_VALUE), IsIndirect, LDI->second, Offset,
+              Variable, Expr);
 
       // If this vreg is directly copied into an exported register then
       // that COPY instructions also need DBG_VALUE, if it is the only
@@ -513,11 +512,9 @@
       }
       if (CopyUseMI) {
         MachineInstr *NewMI =
-          BuildMI(*MF, CopyUseMI->getDebugLoc(),
-                  TII.get(TargetOpcode::DBG_VALUE),
-                  IsIndirect,
-                  CopyUseMI->getOperand(0).getReg(),
-                  Offset, Variable);
+            BuildMI(*MF, CopyUseMI->getDebugLoc(),
+                    TII.get(TargetOpcode::DBG_VALUE), IsIndirect,
+                    CopyUseMI->getOperand(0).getReg(), Offset, Variable, Expr);
         MachineBasicBlock::iterator Pos = CopyUseMI;
         EntryMBB->insertAfter(Pos, NewMI);
       }