[llvm] AtomicExpand: Allow incrementally legalizing atomicrmw (PR #103371)

[Vitaly Buka via llvm-commits]

================
@@ -351,17 +351,30 @@ bool AtomicExpandImpl::run(Function &F, const TargetMachine *TM) {
 
   bool MadeChange = false;
 
-  SmallVector<Instruction *, 1> AtomicInsts;
-
-  // Changing control-flow while iterating through it is a bad idea, so gather a
-  // list of all atomic instructions before we start.
-  for (Instruction &I : instructions(F))
-    if (I.isAtomic() && !isa<FenceInst>(&I))
-      AtomicInsts.push_back(&I);
-
-  for (auto *I : AtomicInsts) {
-    if (processAtomicInstr(I))
-      MadeChange = true;
+  for (Function::iterator BBI = F.begin(), BBE = F.end(); BBI != BBE;) {
+    BasicBlock *BB = &*BBI;
+    ++BBI;
+
+    BasicBlock::iterator Next;
+
+    for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E;
+         I = Next) {
+      Instruction &Inst = *I;
+      Next = std::next(I);
+
+      if (processAtomicInstr(&Inst)) {
+        MadeChange = true;
----------------
vitalybuka wrote:

Heads up, I reverted the patch again
`processAtomicInstr` may erase `Next`

https://github.com/llvm/llvm-project/pull/103371