[clang] [llvm] [inlineasm] Add special support for "rm" inline asm constraints (PR #181973)

[Nick Desaulniers via cfe-commits]

================
@@ -207,28 +636,83 @@ static bool processCallBrInst(Function &F, CallBrInst *CBR, DominatorTree *DT) {
   return Changed;
 }
 
-static SmallVector<CallBrInst *, 2> findCallBrs(Function &F) {
-  SmallVector<CallBrInst *, 2> CBRs;
-  for (BasicBlock &BB : F)
-    if (auto *CBR = dyn_cast<CallBrInst>(BB.getTerminator()))
-      if (!CBR->getType()->isVoidTy() && !CBR->use_empty())
-        CBRs.push_back(CBR);
-  return CBRs;
-}
-
-static bool runImpl(Function &F, ArrayRef<CallBrInst *> CBRs,
-                    DominatorTree *DT) {
+static bool runImpl(Function &F, ArrayRef<CallBase *> IAs, DominatorTree *DT,
+                    const TargetMachine *TM) {
   bool Changed = false;
+  bool isOptLevelNone = TM->getOptLevel() == CodeGenOptLevel::None;
----------------
nickdesaulniers wrote:

The approach of rewriting inline asms (`call`, `callbr`) is getting messy; having to handle them separately and differently between the two subclasses of `CallBase` appears messy _and_ error prone.  I'm hesitant to accept that this is the best approach. (I think it's important that this PR prototyped this approach, which gave me a better sense for how difficult/complicated doing this BEFORE ISEL would be).

So this `TM->getOptLevel()` is what I didn't know existed.  I think it might be a much much much smaller incision to use `TM->getOptLevel()` in `getConstraintPiority` used by `TargetLowering::getConstraintPreferences`, which in turn is used by BOTH ISEL frameworks.

See also:
330fa7d2a4e0cfbb4b078c9f0ec83f176c331afa

So the idea is that _today_ we basically have a switch statement that returns a relative ordering of preferences when we see `"irm"` constraints. Today we pick `"r"`, which _may fail_ (I.e. experience unrecoverable register exhaustion) in fastregalloc.  But I think the proper, smallest, lowest risk fix is to peek at `TM->getOptLevel()` and return a different relative ordering that prefers `"m"` when `TM->getOptLevel() == CodeGenOptLevel::None`.

https://github.com/llvm/llvm-project/pull/181973