[llvm] [BOLT][AArch64] Implement PLTCall optimization (PR #93584)

[Pavel Samolysov via llvm-commits]

================
@@ -61,18 +61,20 @@ Error PLTCall::runOnFunctions(BinaryContext &BC) {
       if (opts::PLT == OT_HOT && !BB.getKnownExecutionCount())
         continue;
 
-      for (MCInst &Instr : BB) {
-        if (!BC.MIB->isCall(Instr))
+      for (auto II = BB.begin(); II != BB.end(); II++) {
+        if (!BC.MIB->isCall(*II))
           continue;
-        const MCSymbol *CallSymbol = BC.MIB->getTargetSymbol(Instr);
+        const MCSymbol *CallSymbol = BC.MIB->getTargetSymbol(*II);
         if (!CallSymbol)
           continue;
         const BinaryFunction *CalleeBF = BC.getFunctionForSymbol(CallSymbol);
         if (!CalleeBF || !CalleeBF->isPLTFunction())
           continue;
-        BC.MIB->convertCallToIndirectCall(Instr, CalleeBF->getPLTSymbol(),
-                                          BC.Ctx.get());
-        BC.MIB->addAnnotation(Instr, "PLTCall", true);
+        const InstructionListType NewCode = BC.MIB->createIndirectPltCall(
+            *II, CalleeBF->getPLTSymbol(), BC.Ctx.get());
+        II = BB.replaceInstruction(II, NewCode);
----------------
samolisov wrote:

`std::advance()` after this line uses `NewCode.size() - 1` as the second argument. I think it may make sense to ensure that the size is not zero adding a corresponding `assert`.

https://github.com/llvm/llvm-project/pull/93584