Index: include/llvm/CodeGen/AsmPrinter.h
===================================================================
--- include/llvm/CodeGen/AsmPrinter.h
+++ include/llvm/CodeGen/AsmPrinter.h
@@ -223,6 +223,9 @@
 
   void EmitToStreamer(MCStreamer &S, const MCInst &Inst);
 
+  /// Emits inital debug location directive.
+  void emitInitialRawDwarfLocDirective(const MachineFunction &MF);
+
   /// Return the current section we are emitting to.
   const MCSection *getCurrentSection() const;
 
Index: lib/CodeGen/AsmPrinter/AsmPrinter.cpp
===================================================================
--- lib/CodeGen/AsmPrinter/AsmPrinter.cpp
+++ lib/CodeGen/AsmPrinter/AsmPrinter.cpp
@@ -229,6 +229,12 @@
   S.EmitInstruction(Inst, getSubtargetInfo());
 }
 
+void AsmPrinter::emitInitialRawDwarfLocDirective(const MachineFunction &MF) {
+  assert(DD && "Dwarf debug file is not defined.");
+  assert(OutStreamer->hasRawTextSupport() && "Expected assembly output mode.");
+  (void)DD->emitInitialLocDirective(MF);
+}
+
 /// getCurrentSection() - Return the current section we are emitting to.
 const MCSection *AsmPrinter::getCurrentSection() const {
   return OutStreamer->getCurrentSectionOnly();
Index: lib/CodeGen/AsmPrinter/DwarfDebug.h
===================================================================
--- lib/CodeGen/AsmPrinter/DwarfDebug.h
+++ lib/CodeGen/AsmPrinter/DwarfDebug.h
@@ -466,7 +466,7 @@
   /// label that was emitted and which provides correspondence to the
   /// source line list.
   void recordSourceLine(unsigned Line, unsigned Col, const MDNode *Scope,
-                        unsigned Flags);
+                        unsigned Flags, Optional<unsigned> CUID = None);
 
   /// Populate LexicalScope entries with variables' info.
   void collectVariableInfo(DwarfCompileUnit &TheCU, const DISubprogram *SP,
@@ -508,6 +508,10 @@
   /// Emit all Dwarf sections that should come after the content.
   void endModule() override;
 
+  /// Emits inital debug location directive.
+  DebugLoc emitInitialLocDirective(const MachineFunction &MF,
+                                   unsigned CUID = 0);
+
   /// Process beginning of an instruction.
   void beginInstruction(const MachineInstr *MI) override;
 
Index: lib/CodeGen/AsmPrinter/DwarfDebug.cpp
===================================================================
--- lib/CodeGen/AsmPrinter/DwarfDebug.cpp
+++ lib/CodeGen/AsmPrinter/DwarfDebug.cpp
@@ -1366,13 +1366,31 @@
   return DebugLoc();
 }
 
+DebugLoc DwarfDebug::emitInitialLocDirective(const MachineFunction &MF,
+                                             unsigned CUID) {
+  // Get beginning of function.
+  if (DebugLoc PrologEndLoc = findPrologueEndLoc(&MF)) {
+    // Ensure the compile unit is created if the function is called before
+    // beginFunction().
+    (void)getOrCreateDwarfCompileUnit(
+        MF.getFunction().getSubprogram()->getUnit());
+    // We'd like to list the prologue as "not statements" but GDB behaves
+    // poorly if we do that. Revisit this with caution/GDB (7.5+) testing.
+    const DISubprogram *SP = PrologEndLoc->getInlinedAtScope()->getSubprogram();
+    recordSourceLine(SP->getScopeLine(), 0, SP, DWARF2_FLAG_IS_STMT, CUID);
+    return PrologEndLoc;
+  }
+  return DebugLoc();
+}
+
 // Gather pre-function debug information.  Assumes being called immediately
 // after the function entry point has been emitted.
 void DwarfDebug::beginFunctionImpl(const MachineFunction *MF) {
   CurFn = MF;
 
-  auto *SP = MF->getFunction().getSubprogram();
-  assert(LScopes.empty() || SP == LScopes.getCurrentFunctionScope()->getScopeNode());
+  const DISubprogram *SP = MF->getFunction().getSubprogram();
+  assert(LScopes.empty() ||
+         SP == LScopes.getCurrentFunctionScope()->getScopeNode());
   if (SP->getUnit()->getEmissionKind() == DICompileUnit::NoDebug)
     return;
 
@@ -1388,13 +1406,8 @@
     Asm->OutStreamer->getContext().setDwarfCompileUnitID(CU.getUniqueID());
 
   // Record beginning of function.
-  PrologEndLoc = findPrologueEndLoc(MF);
-  if (PrologEndLoc) {
-    // We'd like to list the prologue as "not statements" but GDB behaves
-    // poorly if we do that. Revisit this with caution/GDB (7.5+) testing.
-    auto *SP = PrologEndLoc->getInlinedAtScope()->getSubprogram();
-    recordSourceLine(SP->getScopeLine(), 0, SP, DWARF2_FLAG_IS_STMT);
-  }
+  PrologEndLoc = emitInitialLocDirective(
+      *MF, Asm->OutStreamer->getContext().getDwarfCompileUnitID());
 }
 
 void DwarfDebug::skippedNonDebugFunction() {
@@ -1481,7 +1494,7 @@
 // Register a source line with debug info. Returns the  unique label that was
 // emitted and which provides correspondence to the source line list.
 void DwarfDebug::recordSourceLine(unsigned Line, unsigned Col, const MDNode *S,
-                                  unsigned Flags) {
+                                  unsigned Flags, Optional<unsigned> CUID) {
   StringRef Fn;
   unsigned FileNo = 1;
   unsigned Discriminator = 0;
@@ -1491,9 +1504,10 @@
       if (auto *LBF = dyn_cast<DILexicalBlockFile>(Scope))
         Discriminator = LBF->getDiscriminator();
 
-    unsigned CUID = Asm->OutStreamer->getContext().getDwarfCompileUnitID();
-    FileNo = static_cast<DwarfCompileUnit &>(*InfoHolder.getUnits()[CUID])
-              .getOrCreateSourceID(Scope->getFile());
+    unsigned ID =
+        CUID.getValueOr(Asm->OutStreamer->getContext().getDwarfCompileUnitID());
+    FileNo = static_cast<DwarfCompileUnit &>(*InfoHolder.getUnits()[ID])
+                 .getOrCreateSourceID(Scope->getFile());
   }
   Asm->OutStreamer->EmitDwarfLocDirective(FileNo, Line, Col, Flags, 0,
                                           Discriminator, Fn);
Index: lib/Target/NVPTX/NVPTXAsmPrinter.cpp
===================================================================
--- lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -470,6 +470,9 @@
   // Emit open brace for function body.
   OutStreamer->EmitRawText(StringRef("{\n"));
   setAndEmitFunctionVirtualRegisters(*MF);
+  // Emit initial .loc debug directive for correct relocation symbol data.
+  if (MMI && MMI->hasDebugInfo())
+    emitInitialRawDwarfLocDirective(*MF);
 }
 
 bool NVPTXAsmPrinter::runOnMachineFunction(MachineFunction &F) {
Index: test/DebugInfo/NVPTX/cu-range-hole.ll
===================================================================
--- test/DebugInfo/NVPTX/cu-range-hole.ll
+++ test/DebugInfo/NVPTX/cu-range-hole.ll
@@ -6,6 +6,7 @@
 ; CHECK: .param .b32 b_param_0
 ; CHECK: )
 ; CHECK: {
+; CHECK: .loc 1 1 0
 ; CHECK: Lfunc_begin0:
 ; CHECK: .loc 1 1 0
 ; CHECK: .loc 1 1 0
@@ -27,6 +28,7 @@
 ; CHECK: .param .b32 d_param_0
 ; CHECK: )
 ; CHECK: {
+; CHECK: .loc 1 3 0
 ; CHECK: Lfunc_begin2:
 ; CHECK: .loc 1 3 0
 ; CHECK: ret;
Index: test/DebugInfo/NVPTX/debug-loc-offset.ll
===================================================================
--- test/DebugInfo/NVPTX/debug-loc-offset.ll
+++ test/DebugInfo/NVPTX/debug-loc-offset.ll
@@ -11,8 +11,9 @@
 
 ; CHECK: .visible .func  (.param .b32 func_retval0) _Z3bari(
 ; CHECK: {
-; CHECK: Lfunc_begin0:
 ; CHECK: .loc [[CU1:[0-9]+]] 1 0
+; CHECK: Lfunc_begin0:
+; CHECK: .loc [[CU1]] 1 0
 
 ; CHECK: //DEBUG_VALUE: bar:b <- {{[0-9]+}}
 ; CHECK: //DEBUG_VALUE: bar:b <- {{[0-9]+}}
@@ -39,8 +40,9 @@
 
 ; CHECK: .visible .func _Z3baz1A(
 ; CHECK: {
-; CHECK: Lfunc_begin1:
 ; CHECK: .loc [[CU2:[0-9]+]] 6 0
+; CHECK: Lfunc_begin1:
+; CHECK: .loc [[CU2]] 6 0
 ; CHECK: //DEBUG_VALUE: baz:z <- {{[0-9]+}}
 ; CHECK: //DEBUG_VALUE: baz:z <- {{[0-9]+}}
 ; CHECK: .loc [[CU2]] 10 0