diff --git a/ctx.cpp b/ctx.cpp
index 3aee776a..c97da129 100644
--- a/ctx.cpp
+++ b/ctx.cpp
@@ -1404,15 +1404,33 @@ FunctionEmitContext::MasksAllEqual(llvm::Value *v1, llvm::Value *v2) {
 
 llvm::Value *
 FunctionEmitContext::ProgramIndexVector(bool is32bits) {
+  if (1) //g->target->getISA() != Target::NVPTX64)
+  {
     llvm::SmallVector<llvm::Constant*, 16> array;
     for (int i = 0; i < g->target->getVectorWidth() ; ++i) {
-        llvm::Constant *C = is32bits ? LLVMInt32(i) : LLVMInt64(i);
-        array.push_back(C);
+      llvm::Constant *C = is32bits ? LLVMInt32(i) : LLVMInt64(i);
+      array.push_back(C);
     }
 
     llvm::Constant* index = llvm::ConstantVector::get(array);
 
     return index;
+  }
+  else
+  {
+    std::vector<Symbol *> mm;
+    m->symbolTable->LookupFunction("__tid_x", &mm);
+    if (g->target->getMaskBitCount() == 1)
+      AssertPos(currentPos, mm.size() == 1);
+    else
+      // There should be one with signed int signature, one unsigned int.
+      AssertPos(currentPos, mm.size() == 2);
+    // We can actually call either one, since both are i32s as far as
+    // LLVM's type system is concerned...
+    llvm::Function *fmm = mm[0]->function;
+    std::vector<llvm::Value*> args;
+    return CallInst(fmm, NULL, args, "__tid_x");
+  }
 }
 
 
diff --git a/func.cpp b/func.cpp
index 532f92e6..62372a21 100644
--- a/func.cpp
+++ b/func.cpp
@@ -523,6 +523,11 @@ Function::GenerateIR() {
     }
 
     // And we can now go ahead and emit the code
+    { /* export function with NVPTX64 target should be emitted host architecture */
+      const FunctionType *type= CastType<FunctionType>(sym->type);
+      if (g->target->getISA() == Target::NVPTX64 && type->isExported)
+        return;
+    }
     {
         FunctionEmitContext ec(this, sym, function, firstStmtPos);
         emitCode(&ec, function, firstStmtPos);
@@ -540,7 +545,7 @@ Function::GenerateIR() {
         // the application can call it
         const FunctionType *type = CastType<FunctionType>(sym->type);
         Assert(type != NULL);
-        if (type->isExported) {
+        if (type->isExported && g->target->getISA() != Target::NVPTX64) {
             if (!type->isTask) {
                 llvm::FunctionType *ftype = type->LLVMFunctionType(g->ctx, true);
                 llvm::GlobalValue::LinkageTypes linkage = llvm::GlobalValue::ExternalLinkage;
diff --git a/ispc.cpp b/ispc.cpp
index 8bdcb0c4..00d94000 100644
--- a/ispc.cpp
+++ b/ispc.cpp
@@ -167,6 +167,7 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
 #endif
     m_valid(false),
     m_isa(SSE2),
+    m_isPTX(false),
     m_arch(""),
     m_is32Bit(true),
     m_cpu(""),
@@ -639,6 +640,7 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
 #endif
     else if (!strcasecmp(isa, "nvptx64")) {
         this->m_isa = Target::NVPTX64;
+        this->m_isPTX = true;
         this->m_nativeVectorWidth = 1;
         this->m_vectorWidth = 1;
         this->m_attributes = "+sm_35";
diff --git a/ispc.h b/ispc.h
index e2a58ba9..eea77348 100644
--- a/ispc.h
+++ b/ispc.h
@@ -244,6 +244,7 @@ public:
     bool isValid() const {return m_valid;}
 
     ISA getISA() const {return m_isa;}
+    bool isPTX() const {return m_isPTX;}
 
     std::string getArch() const {return m_arch;}
 
@@ -298,6 +299,7 @@ private:
 
     /** Instruction set being compiled to. */
     ISA m_isa;
+    bool m_isPTX;
 
     /** Target system architecture.  (e.g. "x86-64", "x86"). */
     std::string m_arch;
diff --git a/module.cpp b/module.cpp
index 28c922c8..07dc5e48 100644
--- a/module.cpp
+++ b/module.cpp
@@ -2316,6 +2316,61 @@ Module::CompileAndOutput(const char *srcFile,
                          const char *hostStubFileName,
                          const char *devStubFileName)
 {
+  if (target != NULL && !strcmp(target,"nvptx64"))
+  {
+    fprintf(stderr,  "compiling nvptx64 \n");
+    // We're only compiling to a single target
+    g->target = new Target(arch, cpu, target, generatePIC);
+    if (!g->target->isValid())
+      return 1;
+
+    m = new Module(srcFile);
+    if (m->CompileFile() == 0) {
+      if (outputType == CXX) {
+        if (target == NULL || strncmp(target, "generic-", 8) != 0) {
+          Error(SourcePos(), "When generating C++ output, one of the \"generic-*\" "
+              "targets must be used.");
+          return 1;
+        }
+      }
+      else if (outputType == Asm || outputType == Object) {
+        if (target != NULL && strncmp(target, "generic-", 8) == 0) {
+          Error(SourcePos(), "When using a \"generic-*\" compilation target, "
+              "%s output can not be used.",
+              (outputType == Asm) ? "assembly" : "object file");
+          return 1;
+        }
+      }
+
+      if (outFileName != NULL)
+        if (!m->writeOutput(outputType, outFileName, includeFileName))
+          return 1;
+      if (headerFileName != NULL)
+        if (!m->writeOutput(Module::Header, headerFileName))
+          return 1;
+      if (depsFileName != NULL)
+        if (!m->writeOutput(Module::Deps,depsFileName))
+          return 1;
+      if (hostStubFileName != NULL)
+        if (!m->writeOutput(Module::HostStub,hostStubFileName))
+          return 1;
+      if (devStubFileName != NULL)
+        if (!m->writeOutput(Module::DevStub,devStubFileName))
+          return 1;
+    }
+    else
+      ++m->errorCount;
+
+    int errorCount = m->errorCount;
+    delete m;
+    m = NULL;
+
+    delete g->target;
+    g->target = NULL;
+
+    return errorCount > 0;
+  }
+  else
     if (target == NULL || strchr(target, ',') == NULL) {
         // We're only compiling to a single target
         g->target = new Target(arch, cpu, target, generatePIC);