[llvm] [AMDGPU][SIPreEmitPeephole] mustRetainExeczBranch: use BranchProbability and TargetSchedmodel (PR #109818)

[Jay Foad via llvm-commits]

Juan Manuel Martinez =?utf-8?q?Caama=C3=B1o?= <juamarti at amd.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/109818 at github.com>


================
@@ -304,11 +299,67 @@ bool SIPreEmitPeephole::getBlockDestinations(
   return true;
 }
 
+namespace {
+class BranchWeightCostModel {
+  const SIInstrInfo &TII;
+  const TargetSchedModel &SchedModel;
+  BranchProbability BranchProb;
+  uint64_t BranchCost;
+  uint64_t ThenCyclesCost = 0;
+
+public:
+  BranchWeightCostModel(const SIInstrInfo &TII, const MachineInstr &Branch,
+                        const MachineBasicBlock &Succ)
+      : TII(TII), SchedModel(TII.getSchedModel()) {
+    assert(SchedModel.hasInstrSchedModelOrItineraries());
+
+    const MachineBasicBlock &Head = *Branch.getParent();
+    const auto *FromIt = find(Head.successors(), &Succ);
+    assert(FromIt != Head.succ_end());
+
+    BranchProb = Head.getSuccProbability(FromIt);
+    if (BranchProb.isUnknown())
+      return;
+
+    BranchCost = SchedModel.computeInstrLatency(&Branch, false);
+  }
+
+  bool isUnknown() const { return BranchProb.isUnknown(); }
+
+  bool isProfitable(const MachineInstr &MI) {
+    assert(!isUnknown());
+
+    if (TII.isWaitcnt(MI.getOpcode()))
+      return false;
+
+    ThenCyclesCost += SchedModel.computeInstrLatency(&MI, false);
+
+    // Consider `P = N/D` to be the probability of execnz being true
+    // The transformation is profitable if always executing the 'then' block
+    // is cheaper than executing sometimes 'then' and always
+    // executing s_cbranch_execnz:
+    // * ThenCost <= P*ThenCost + BranchCost
+    // * (1-P) * ThenCost <= BranchCost
+    // * (D-N)/D * ThenCost <= BranchCost
+    uint64_t Numerator = BranchProb.getNumerator();
+    uint64_t Denominator = BranchProb.getDenominator();
+    return (Denominator - Numerator) * ThenCyclesCost <=
+           Denominator * BranchCost;
+  }
+};
+
 bool SIPreEmitPeephole::mustRetainExeczBranch(
-    const MachineBasicBlock &From, const MachineBasicBlock &To) const {
-  unsigned NumInstr = 0;
-  const MachineFunction *MF = From.getParent();
+    const MachineBasicBlock &Head, const MachineBasicBlock &From,
+    const MachineBasicBlock &To) const {
+
+  const auto *FromIt = find(Head.successors(), &From);
+  assert(FromIt != Head.succ_end());
----------------
jayfoad wrote:

`FromIt` is not used below.

https://github.com/llvm/llvm-project/pull/109818