[llvm] [AArch64][SME] Propagate desired ZA states in the MachineSMEABIPass (PR #149510)

[Benjamin Maxwell via llvm-commits]

================
@@ -422,12 +438,69 @@ FunctionInfo MachineSMEABI::collectNeededZAStates(SMEAttrs SMEFnAttrs) {
 
     // Reverse vector (as we had to iterate backwards for liveness).
     std::reverse(Block.Insts.begin(), Block.Insts.end());
+
+    // Record the desired states on entry/exit of this block. These are the
+    // states that would not incur a state transition.
+    if (!Block.Insts.empty()) {
+      Block.DesiredIncomingState = Block.Insts.front().NeededState;
+      Block.DesiredOutgoingState = Block.Insts.back().NeededState;
+    }
   }
 
   return FunctionInfo{std::move(Blocks), AfterSMEProloguePt,
                       PhysLiveRegsAfterSMEPrologue};
 }
 
+void MachineSMEABI::propagateDesiredStates(FunctionInfo &FnInfo,
+                                           bool Forwards) {
+  // If `Forwards`, this propagates desired states from predecessors to
+  // successors, otherwise, this propagates states from successors to
+  // predecessors.
+  auto GetBlockState = [](BlockInfo &Block, bool Incoming) -> ZAState & {
+    return Incoming ? Block.DesiredIncomingState : Block.DesiredOutgoingState;
+  };
+
+  SmallVector<MachineBasicBlock *> Worklist;
+  for (auto [BlockID, BlockInfo] : enumerate(FnInfo.Blocks)) {
+    if (!isLegalEdgeBundleZAState(GetBlockState(BlockInfo, Forwards)))
+      Worklist.push_back(MF->getBlockNumbered(BlockID));
+  }
+
+  while (!Worklist.empty()) {
+    MachineBasicBlock *MBB = Worklist.pop_back_val();
+    BlockInfo &Block = FnInfo.Blocks[MBB->getNumber()];
+
+    // Pick a legal edge bundle state that matches the majority of
+    // predecessors/successors.
+    int StateCounts[ZAState::NUM_ZA_STATE] = {0};
+    for (MachineBasicBlock *PredOrSucc :
+         Forwards ? predecessors(MBB) : successors(MBB)) {
----------------
MacDue wrote:

I don't think so. `MBB` is the block we're propagating state to. If we're going "forwards" (in control flow), we're picking a state that matches the majority of our predecessors. I think this counts as "forwards", as it's propagating the state from earlier in the function to later in the function. 

Backwards propagation is picking a state that matches the majority of the successors, which I think of as "backwards" as it's propagating state from later in the function to an earlier point. 

https://github.com/llvm/llvm-project/pull/149510