From 46357438baefbdcf630abc5d74565afcbf1c48dd Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo@fhahn.com>
Date: Thu, 17 Jul 2025 16:47:54 +0200
Subject: [PATCH] [SCEV] Try to re-use existing LCSSA phis when expanding
 SCEVAddRecExpr. (#147214)

If an AddRec is expanded outside a loop with a single exit block, check
if any of the (lcssa) phi nodes in the exit block match the AddRec. If
that's the case, simply use the existing lcssa phi.

This can reduce the number of instruction created for SCEV expansions,
mainly for runtime checks generated by the loop vectorizer.

Compile-time impact should be mostly neutral

https://llvm-compile-time-tracker.com/compare.php?from=48c7a3187f9831304a38df9bdb3b4d5bf6b6b1a2&to=cf9d039a7b0db5d0d912e0e2c01b19c2a653273a&stat=instructions:u

PR: https://github.com/llvm/llvm-project/pull/147214
---
 .../Utils/ScalarEvolutionExpander.h           |  1 +
 .../Utils/ScalarEvolutionExpander.cpp         | 23 +++++++++++++++++++
 .../reuse-lcssa-phi-scev-expansion.ll         |  4 +---
 3 files changed, 25 insertions(+), 3 deletions(-)

diff --git a/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h b/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h
index a101151eed7c..39fef921a959 100644
--- a/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h
+++ b/llvm/include/llvm/Transforms/Utils/ScalarEvolutionExpander.h
@@ -530,6 +530,7 @@ private:
 
   bool isExpandedAddRecExprPHI(PHINode *PN, Instruction *IncV, const Loop *L);
 
+  Value *tryToReuseLCSSAPhi(const SCEVAddRecExpr *S);
   Value *expandAddRecExprLiterally(const SCEVAddRecExpr *);
   PHINode *getAddRecExprPHILiterally(const SCEVAddRecExpr *Normalized,
                                      const Loop *L, Type *&TruncTy,
diff --git a/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp b/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
index 739ac00ba47c..ed08c0bfa2e7 100644
--- a/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
+++ b/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
@@ -1223,6 +1223,24 @@ Value *SCEVExpander::expandAddRecExprLiterally(const SCEVAddRecExpr *S) {
   return Result;
 }
 
+Value *SCEVExpander::tryToReuseLCSSAPhi(const SCEVAddRecExpr *S) {
+  const Loop *L = S->getLoop();
+  BasicBlock *EB = L->getExitBlock();
+  if (!EB || !EB->getSinglePredecessor() ||
+      !SE.DT.dominates(EB, Builder.GetInsertBlock()))
+    return nullptr;
+
+  for (auto &PN : EB->phis()) {
+    if (!SE.isSCEVable(PN.getType()) || PN.getType() != S->getType())
+      continue;
+    auto *ExitV = SE.getSCEV(&PN);
+    if (S == ExitV)
+      return &PN;
+  }
+
+  return nullptr;
+}
+
 Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) {
   // In canonical mode we compute the addrec as an expression of a canonical IV
   // using evaluateAtIteration and expand the resulting SCEV expression. This
@@ -1262,6 +1280,11 @@ Value *SCEVExpander::visitAddRecExpr(const SCEVAddRecExpr *S) {
     return V;
   }
 
+  // If S is expanded outside the defining loop, check if there is a
+  // matching LCSSA phi node for it.
+  if (Value *V = tryToReuseLCSSAPhi(S))
+    return V;
+
   // {X,+,F} --> X + {0,+,F}
   if (!S->getStart()->isZero()) {
     if (isa<PointerType>(S->getType())) {
diff --git a/llvm/test/Transforms/LoopVectorize/reuse-lcssa-phi-scev-expansion.ll b/llvm/test/Transforms/LoopVectorize/reuse-lcssa-phi-scev-expansion.ll
index 2747895f06a7..ce4270dc4b7f 100644
--- a/llvm/test/Transforms/LoopVectorize/reuse-lcssa-phi-scev-expansion.ll
+++ b/llvm/test/Transforms/LoopVectorize/reuse-lcssa-phi-scev-expansion.ll
@@ -18,11 +18,9 @@ define void @reuse_lcssa_phi_for_add_rec1(ptr %head) {
 ; CHECK-NEXT:    [[IV_NEXT]] = add nuw i64 [[IV]], 1
 ; CHECK-NEXT:    br i1 [[EC_1]], label %[[PH:.*]], label %[[LOOP_1]]
 ; CHECK:       [[PH]]:
-; CHECK-NEXT:    [[IV_2_LCSSA:%.*]] = phi i32 [ [[IV_2]], %[[LOOP_1]] ]
 ; CHECK-NEXT:    [[IV_LCSSA:%.*]] = phi i64 [ [[IV]], %[[LOOP_1]] ]
-; CHECK-NEXT:    [[IV_2_NEXT_LCSSA:%.*]] = phi i32 [ [[IV_2_NEXT]], %[[LOOP_1]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = phi i32 [ [[IV_2_NEXT]], %[[LOOP_1]] ]
 ; CHECK-NEXT:    [[SRC_2:%.*]] = tail call noalias noundef dereferenceable_or_null(8) ptr @calloc(i64 1, i64 8)
-; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[IV_2_LCSSA]], 1
 ; CHECK-NEXT:    [[SMIN:%.*]] = call i32 @llvm.smin.i32(i32 [[TMP0]], i32 1)
 ; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 [[TMP0]], [[SMIN]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP1]] to i64