[AsmPrinter] Allow .cfi_restore_state to be put at the end of a function

[sinan]

Summary

If cfi_restore_state is inserted into an empty block placed at the end of the function, this directive would be ignored when generating asm (because of this patch https://reviews.llvm.org/D32246, but cfi_restore_state looks fine, since it does not describe any target address), resulting in a cfi_remember_state/restore_ state mismatch and then crashes in applications like bolt(assertion in BinaryBasicBlock::getCFIStateAtInstr). This patch will allow .cfi_restore_state to be put at the end of the function.

reproduce(mismatch.ll at https://reviews.llvm.org/F26307823):
llc -mtriple=aarch64 mismatch.ll -o mismatch.s
then you can find there are four cfi_remember_state but only three cfi_restore_state.

So, the whole story for this test case is that CFIFixup inserts directives

1. insert .cfi_remember_state at bb.0, and insert .cfi_restore_state at bb.4
2. insert .cfi_remember_state at bb.4, and insert .cfi_restore_state at bb.6
3. insert .cfi_remember_state at bb.6, and insert .cfi_restore_state at bb.7
4. insert .cfi_remember_state at bb.7, and insert .cfi_restore_state at bb.8

and, .cfi_restore_state at bb.8 is ignored when emitting asm and then the mismatch happens.

[sinan]

Another way to avoid this mismatch is to set such blocks as 'Reachable' in CFIFixup, but it looks a bit different from the 'Unreachable block' case in CFIFixup to me and I think we still need a restore_state in the empty block. Any idea?

[sinan]

gentle ping ...

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

     return;
 
   // If there is no "real" instruction following this CFI instruction, skip
   // emitting it; it would be beyond the end of the function's FDE range.
   auto *MBB = MI.getParent();
   auto I = std::next(MI.getIterator());
   while (I != MBB->end() && I->isTransient())
     ++I;
-  if (I == MBB->instr_end() &&
-      MBB->getReverseIterator() == MBB->getParent()->rbegin())
-    return;
 
   const std::vector<MCCFIInstruction> &Instrs = MF->getFrameInstructions();
   unsigned CFIIndex = MI.getOperand(0).getCFIIndex();
   const MCCFIInstruction &CFI = Instrs[CFIIndex];
+
+  if (I == MBB->instr_end() &&
+      MBB->getReverseIterator() == MBB->getParent()->rbegin() &&
+      CFI.getOperation() != MCCFIInstruction::OpRestoreState)
+    return;
+
   emitCFIInstruction(CFI);
 }
 
 void AsmPrinter::emitFrameAlloc(const MachineInstr &MI) {
   // The operands are the MCSymbol and the frame offset of the allocation.
   MCSymbol *FrameAllocSym = MI.getOperand(0).getMCSymbol();
   int FrameOffset = MI.getOperand(1).getImm();