AMDGPU: Move SILowerControlFlowPass to be post-regalloc

[arsenm]

Summary

Currently this is inserting a prologue into the function.
Move this before PEI so we can emit the prologue there
for flat_scratch and m0 initialization.

[• tstellarAMD]

LGTM.

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

void GCNPassConfig::addOptimizedRegAlloc(FunctionPass *RegAllocPass) {
   // FIXME: We shouldn't disable the verifier here. r249087 introduced a failure
   // that needs to be fixed.
   insertPass(&LiveVariablesID, &SIFixSGPRLiveRangesID, /*VerifyAfter=*/false);
   TargetPassConfig::addOptimizedRegAlloc(RegAllocPass);
 }
 
 void GCNPassConfig::addPostRegAlloc() {
   addPass(createSIShrinkInstructionsPass(), false);
+  addPass(createSILowerControlFlowPass(*TM), false);
 }
 
 void GCNPassConfig::addPreSched2() {
 }
 
 void GCNPassConfig::addPreEmitPass() {
   addPass(createSIInsertWaitsPass(), false);
-  addPass(createSILowerControlFlowPass(*TM), false);
 }
 
 TargetPassConfig *GCNTargetMachine::createPassConfig(PassManagerBase &PM) {
   return new GCNPassConfig(this, PM);
 }