[NVPTX] Move NVVMReflect pass out of codegen passes.

[jlebar]

Summary

Currently the NVVMReflect pass is run at the beginning of our backend
passes. But really, it should be run as early as possible, as it's
simply resolving an "if" statement in code. So move it into
TargetMachine::addEarlyAsPossiblePasses.

(Specifically, NVVMReflect changes each call to the __nvvm_reflect
function or llvm.nvvm.reflect intrinsic into an integer constant, based
on the pass's configuration. Clearly we miss many optimization
opportunities if we perform this transformation at the beginning of
codegen.)

[jlebar]

Still run NVVMReflect at the start of the target-specific passes, since it's necessary for correctness.

llvm/trunk/lib/Target/NVPTX/NVPTXTargetMachine.h

public:
   bool addPassesToEmitMC(PassManagerBase &, MCContext *&, raw_pwrite_stream &,
                          bool = true) override {
     return true;
   }
   TargetLoweringObjectFile *getObjFileLowering() const override {
     return TLOF.get();
   }
 
+  void addEarlyAsPossiblePasses(PassManagerBase &PM) override;
   TargetIRAnalysis getTargetIRAnalysis() override;
 
 }; // NVPTXTargetMachine.
 
 class NVPTXTargetMachine32 : public NVPTXTargetMachine {
   virtual void anchor();
 public:
   NVPTXTargetMachine32(const Target &T, const Triple &TT, StringRef CPU,

llvm/trunk/lib/Target/NVPTX/NVPTXTargetMachine.cpp

private:
   void addStraightLineScalarOptimizationPasses();
 };
 } // end anonymous namespace
 
 TargetPassConfig *NVPTXTargetMachine::createPassConfig(PassManagerBase &PM) {
   return new NVPTXPassConfig(this, PM);
 }
 
+void NVPTXTargetMachine::addEarlyAsPossiblePasses(PassManagerBase &PM) {
+  PM.add(createNVVMReflectPass());
+}
+
 TargetIRAnalysis NVPTXTargetMachine::getTargetIRAnalysis() {
   return TargetIRAnalysis([this](const Function &F) {
     return TargetTransformInfo(NVPTXTTIImpl(this, F));
   });
 }
 
 void NVPTXPassConfig::addEarlyCSEOrGVNPass() {
   if (getOptLevel() == CodeGenOpt::Aggressive)
void NVPTXPassConfig::addIRPasses() {
   disablePass(&MachineCopyPropagationID);
   disablePass(&TailDuplicateID);
   disablePass(&StackMapLivenessID);
   disablePass(&LiveDebugValuesID);
   disablePass(&PostRASchedulerID);
   disablePass(&FuncletLayoutID);
   disablePass(&PatchableFunctionID);
 
+  // NVVMReflectPass is added in addEarlyAsPossiblePasses, so hopefully running
+  // it here does nothing.  But since we need it for correctness when lowering
+  // to NVPTX, run it here too, in case whoever built our pass pipeline didn't
+  // call addEarlyAsPossiblePasses.
   addPass(createNVVMReflectPass());
+
   if (getOptLevel() != CodeGenOpt::None)
     addPass(createNVPTXImageOptimizerPass());
   addPass(createNVPTXAssignValidGlobalNamesPass());
   addPass(createGenericToNVVMPass());
 
   // NVPTXLowerKernelArgs is required for correctness and should be run right
   // before the address space inference passes.
   addPass(createNVPTXLowerKernelArgsPass(&getNVPTXTargetMachine()));