diff --git a/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp b/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp
index c7b8aa49370..d5be0e42ab0 100644
--- a/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp
+++ b/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.cpp
@@ -18,6 +18,7 @@
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCSymbol.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FormattedStream.h"
@@ -277,3 +278,12 @@ void NVPTXInstPrinter::printMemOperand(const MCInst *MI, int OpNum,
     printOperand(MI, OpNum + 1, O);
   }
 }
+
+void NVPTXInstPrinter::printProtoIdent(const MCInst *MI, int OpNum,
+                                       raw_ostream &O, const char *Modifier) {
+  const MCOperand &Op = MI->getOperand(OpNum);
+  assert(Op.isExpr() && "Call prototype is not an MCExpr?");
+  const MCExpr *Expr = Op.getExpr();
+  const MCSymbol &Sym = cast<MCSymbolRefExpr>(Expr)->getSymbol();
+  O << Sym.getName();
+}
diff --git a/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h b/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h
index e0f44da643b..93029ae8d39 100644
--- a/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h
+++ b/lib/Target/NVPTX/InstPrinter/NVPTXInstPrinter.h
@@ -44,7 +44,8 @@ public:
                      raw_ostream &O, const char *Modifier = 0);
   void printMemOperand(const MCInst *MI, int OpNum,
                        raw_ostream &O, const char *Modifier = 0);
-
+  void printProtoIdent(const MCInst *MI, int OpNum,
+                       raw_ostream &O, const char *Modifier = 0);
 };
 
 }
diff --git a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
index 0f8649f8139..7552fe70411 100644
--- a/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ b/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -314,6 +314,14 @@ void NVPTXAsmPrinter::EmitInstruction(const MachineInstr *MI) {
 void NVPTXAsmPrinter::lowerToMCInst(const MachineInstr *MI, MCInst &OutMI) {
   OutMI.setOpcode(MI->getOpcode());
 
+  // Special: Do not mangle symbol operand of CALL_PROTOTYPE
+  if (MI->getOpcode() == NVPTX::CALL_PROTOTYPE) {
+    const MachineOperand &MO = MI->getOperand(0);
+    OutMI.addOperand(GetSymbolRef(MO,
+      OutContext.GetOrCreateSymbol(Twine(MO.getSymbolName()))));
+    return;
+  }
+
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     const MachineOperand &MO = MI->getOperand(i);
 
diff --git a/lib/Target/NVPTX/NVPTXISelLowering.cpp b/lib/Target/NVPTX/NVPTXISelLowering.cpp
index 98b9ce6e268..7ff43bf2310 100644
--- a/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -310,6 +310,8 @@ const char *NVPTXTargetLowering::getTargetNodeName(unsigned Opcode) const {
     return "NVPTXISD::CallSeqBegin";
   case NVPTXISD::CallSeqEnd:
     return "NVPTXISD::CallSeqEnd";
+  case NVPTXISD::CallPrototype:
+    return "NVPTXISD::CallPrototype";
   case NVPTXISD::LoadV2:
     return "NVPTXISD::LoadV2";
   case NVPTXISD::LoadV4:
@@ -885,18 +887,16 @@ SDValue NVPTXTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
     // proto_0 : .callprototype(.param .b32 _) _ (.param .b32 _);
     // to be emitted, and the label has to used as the last arg of call
     // instruction.
-    // The prototype is embedded in a string and put as the operand for an
-    // INLINEASM SDNode.
-    SDVTList InlineAsmVTs = DAG.getVTList(MVT::Other, MVT::Glue);
-    std::string proto_string =
-        getPrototype(retTy, Args, Outs, retAlignment, CS);
-    const char *asmstr = nvTM->getManagedStrPool()
-        ->getManagedString(proto_string.c_str())->c_str();
-    SDValue InlineAsmOps[] = {
-      Chain, DAG.getTargetExternalSymbol(asmstr, getPointerTy()),
-      DAG.getMDNode(0), DAG.getTargetConstant(0, MVT::i32), InFlag
+    // The prototype is embedded in a string and put as the operand for a
+    // CallPrototype SDNode which will print out to the value of the string.
+    SDVTList ProtoVTs = DAG.getVTList(MVT::Other, MVT::Glue);
+    std::string Proto = getPrototype(retTy, Args, Outs, retAlignment, CS);
+    const char *ProtoStr =
+      nvTM->getManagedStrPool()->getManagedString(Proto.c_str())->c_str();
+    SDValue ProtoOps[] = {
+      Chain, DAG.getTargetExternalSymbol(ProtoStr, MVT::i32), InFlag,
     };
-    Chain = DAG.getNode(ISD::INLINEASM, dl, InlineAsmVTs, InlineAsmOps, 5);
+    Chain = DAG.getNode(NVPTXISD::CallPrototype, dl, ProtoVTs, &ProtoOps[0], 3);
     InFlag = Chain.getValue(1);
   }
   // Op to just print "call"
diff --git a/lib/Target/NVPTX/NVPTXISelLowering.h b/lib/Target/NVPTX/NVPTXISelLowering.h
index 34184372c76..66e708fcea0 100644
--- a/lib/Target/NVPTX/NVPTXISelLowering.h
+++ b/lib/Target/NVPTX/NVPTXISelLowering.h
@@ -49,6 +49,7 @@ enum NodeType {
   RETURN,
   CallSeqBegin,
   CallSeqEnd,
+  CallPrototype,
   Dummy,
 
   LoadV2 = ISD::FIRST_TARGET_MEMORY_OPCODE,
diff --git a/lib/Target/NVPTX/NVPTXInstrInfo.td b/lib/Target/NVPTX/NVPTXInstrInfo.td
index 3e430bf723c..b23f1e4b458 100644
--- a/lib/Target/NVPTX/NVPTXInstrInfo.td
+++ b/lib/Target/NVPTX/NVPTXInstrInfo.td
@@ -2607,6 +2607,20 @@ def trapinst : NVPTXInst<(outs), (ins),
                          "trap;",
                          [(trap)]>;
 
+// Call prototype wrapper
+def SDTCallPrototype : SDTypeProfile<0, 1, [SDTCisInt<0>]>;
+def CallPrototype
+  : SDNode<"NVPTXISD::CallPrototype", SDTCallPrototype,
+           [SDNPHasChain, SDNPOutGlue, SDNPInGlue, SDNPSideEffect]>;
+def ProtoIdent : Operand<i32> {
+  let PrintMethod = "printProtoIdent";
+}
+def CALL_PROTOTYPE
+  : NVPTXInst<(outs), (ins ProtoIdent:$ident),
+              "$ident", [(CallPrototype (i32 texternalsym:$ident))]>;
+
+
+
 include "NVPTXIntrinsics.td"
 
 
diff --git a/test/CodeGen/NVPTX/callchain.ll b/test/CodeGen/NVPTX/callchain.ll
new file mode 100644
index 00000000000..60b118b6a19
--- /dev/null
+++ b/test/CodeGen/NVPTX/callchain.ll
@@ -0,0 +1,10 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
+
+target triple = "nvptx"
+
+define void @foo(i8* %ptr) {
+  %fnptr = bitcast i8* %ptr to void ()*
+; CHECK: prototype_0 : .callprototype ()_ ()
+  tail call void %fnptr()
+  ret void
+}