diff --git a/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h b/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h
--- a/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h
+++ b/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.h
@@ -200,6 +200,7 @@
   const Function *F;
   std::string CurrentFnName;
 
+  void emitStartOfAsmFile(Module &M) override;
   void emitBasicBlockStart(const MachineBasicBlock &MBB) override;
   void emitFunctionEntryLabel() override;
   void emitFunctionBodyStart() override;
diff --git a/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp b/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
--- a/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
+++ b/llvm/lib/Target/NVPTX/NVPTXAsmPrinter.cpp
@@ -762,13 +762,21 @@
   return InitList->getNumOperands() == 0;
 }
 
-bool NVPTXAsmPrinter::doInitialization(Module &M) {
+void NVPTXAsmPrinter::emitStartOfAsmFile(Module &M) {
   // Construct a default subtarget off of the TargetMachine defaults. The
   // rest of NVPTX isn't friendly to change subtargets per function and
   // so the default TargetMachine will have all of the options.
   const NVPTXTargetMachine &NTM = static_cast<const NVPTXTargetMachine &>(TM);
   const auto* STI = static_cast<const NVPTXSubtarget*>(NTM.getSubtargetImpl());
+  SmallString<128> Str1;
+  raw_svector_ostream OS1(Str1);
 
+  // Emit header before any dwarf directives are emitted below.
+  emitHeader(M, OS1, *STI);
+  OutStreamer->emitRawText(OS1.str());
+}
+
+bool NVPTXAsmPrinter::doInitialization(Module &M) {
   if (M.alias_size()) {
     report_fatal_error("Module has aliases, which NVPTX does not support.");
     return true; // error
@@ -784,26 +792,9 @@
     return true;  // error
   }
 
-  SmallString<128> Str1;
-  raw_svector_ostream OS1(Str1);
-
   // We need to call the parent's one explicitly.
   bool Result = AsmPrinter::doInitialization(M);
 
-  // Emit header before any dwarf directives are emitted below.
-  emitHeader(M, OS1, *STI);
-  OutStreamer->emitRawText(OS1.str());
-
-  // Emit module-level inline asm if it exists.
-  if (!M.getModuleInlineAsm().empty()) {
-    OutStreamer->AddComment("Start of file scope inline assembly");
-    OutStreamer->AddBlankLine();
-    OutStreamer->emitRawText(StringRef(M.getModuleInlineAsm()));
-    OutStreamer->AddBlankLine();
-    OutStreamer->AddComment("End of file scope inline assembly");
-    OutStreamer->AddBlankLine();
-  }
-
   GlobalsEmitted = false;
 
   return Result;
diff --git a/llvm/test/CodeGen/NVPTX/module-inline-asm.ll b/llvm/test/CodeGen/NVPTX/module-inline-asm.ll
--- a/llvm/test/CodeGen/NVPTX/module-inline-asm.ll
+++ b/llvm/test/CodeGen/NVPTX/module-inline-asm.ll
@@ -2,9 +2,19 @@
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64"
 
+; module asm must come after PTX version/target directives.
+; CHECK-NOT: .global .b32 val;
+
+; CHECK-DAG: .version
+; CHECK-DAG: .target
+
 ; CHECK: .global .b32 val;
 module asm ".global .b32 val;"
 
+; module asm must happen before we emit other things.
+; CHECK-LABEL: .visible .func foo
 define void @foo() {
   ret void
 }
+; Make sure it does not show up anywhere else in the output.
+; CHECK-NOT: .global .b32 val;