[LSR] Return nullptr from getExpr if the result isn't invertible.

getExpr is missing a check to make sure the result is invertible. This can lead to incorrect results, so return nullptr in those cases like in other places in IVUsers. Fixes #62660. Reviewed By: qcolombet Differential Revision: https://reviews.llvm.org/D153202
llvm · Jun 22, 2023 · 3a57152 · 3a57152
1 parent 1d56510
commit 3a57152
Show file tree

Hide file tree

Showing 5 changed files with 31 additions and 11 deletions.
diff --git a/llvm/include/llvm/Analysis/IVUsers.h b/llvm/include/llvm/Analysis/IVUsers.h
@@ -131,7 +131,8 @@ class IVUsers {
   /// value of the OperandValToReplace of the given IVStrideUse.
   const SCEV *getReplacementExpr(const IVStrideUse &IU) const;
 
-  /// getExpr - Return the expression for the use.
+  /// getExpr - Return the expression for the use. Returns nullptr if the result
+  /// is not invertible.
   const SCEV *getExpr(const IVStrideUse &IU) const;
 
   const SCEV *getStride(const IVStrideUse &IU, const Loop *L) const;

diff --git a/llvm/lib/Analysis/IVUsers.cpp b/llvm/lib/Analysis/IVUsers.cpp
@@ -334,8 +334,15 @@ const SCEV *IVUsers::getReplacementExpr(const IVStrideUse &IU) const {
 
 /// getExpr - Return the expression for the use.
 const SCEV *IVUsers::getExpr(const IVStrideUse &IU) const {
-  return normalizeForPostIncUse(getReplacementExpr(IU), IU.getPostIncLoops(),
-                                *SE);
+  const SCEV *Replacement = getReplacementExpr(IU);
+  const SCEV *Normalized =
+      normalizeForPostIncUse(Replacement, IU.getPostIncLoops(), *SE);
+  const SCEV *Denormalized =
+      denormalizeForPostIncUse(Normalized, IU.getPostIncLoops(), *SE);
+  // If the normalized expression isn't invertible.
+  if (Denormalized != Replacement)
+    return nullptr;
+  return Normalized;
 }
 
 static const SCEVAddRecExpr *findAddRecForLoop(const SCEV *S, const Loop *L) {

diff --git a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -2707,6 +2707,8 @@ void LSRInstance::CollectInterestingTypesAndFactors() {
   SmallVector<const SCEV *, 4> Worklist;
   for (const IVStrideUse &U : IU) {
     const SCEV *Expr = IU.getExpr(U);
+    if (!Expr)
+      continue;
 
     // Collect interesting types.
     Types.insert(SE.getEffectiveSCEVType(Expr->getType()));
@@ -3325,6 +3327,8 @@ void LSRInstance::CollectFixupsAndInitialFormulae() {
     }
 
     const SCEV *S = IU.getExpr(U);
+    if (!S)
+      continue;
     PostIncLoopSet TmpPostIncLoops = U.getPostIncLoops();
 
     // Equality (== and !=) ICmps are special. We can rewrite (i == N) as

diff --git a/llvm/test/Transforms/LoopStrengthReduce/X86/pr62660-normalization-failure.ll b/llvm/test/Transforms/LoopStrengthReduce/X86/pr62660-normalization-failure.ll
@@ -1,9 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
 ; RUN: opt -loop-reduce -verify-scev -S %s | FileCheck %s
 
-; XFAIL: *
-; REQUIRES: asserts
-
 target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macos"
 
@@ -12,11 +9,16 @@ define i64 @test_pr62660() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi i32 [ [[LSR_IV_NEXT2:%.*]], [[LOOP]] ], [ 65533, [[ENTRY:%.*]] ]
-; CHECK-NEXT:    [[LSR_IV:%.*]] = phi i64 [ [[LSR_IV_NEXT:%.*]], [[LOOP]] ], [ -1, [[ENTRY]] ]
+; CHECK-NEXT:    [[LSR_IV:%.*]] = phi i64 [ [[LSR_IV_NEXT:%.*]], [[LOOP]] ], [ -1, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[LSR_IV]], 1
+; CHECK-NEXT:    [[TMP:%.*]] = trunc i64 [[TMP0]] to i32
+; CHECK-NEXT:    [[CONV1:%.*]] = and i32 [[TMP]], 65535
+; CHECK-NEXT:    [[ADD:%.*]] = add nsw i32 [[IV]], -1
+; CHECK-NEXT:    [[SUB:%.*]] = add i32 [[ADD]], [[CONV1]]
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
 ; CHECK-NEXT:    [[LSR_IV_NEXT]] = add nsw i64 [[LSR_IV]], 1
-; CHECK-NEXT:    [[LSR_IV_NEXT2]] = add nuw nsw i32 [[LSR_IV1]], 2
-; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[LSR_IV_NEXT2]], 8
+; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[SUB]], 8
 ; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret i64 [[LSR_IV_NEXT]]

diff --git a/llvm/test/Transforms/LoopStrengthReduce/scaling-factor-incompat-type.ll b/llvm/test/Transforms/LoopStrengthReduce/scaling-factor-incompat-type.ll
@@ -17,9 +17,15 @@ define void @foo() {
 ; CHECK:       bb4:
 ; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi i16 [ [[LSR_IV_NEXT2:%.*]], [[BB13:%.*]] ], [ 6, [[BB:%.*]] ]
 ; CHECK-NEXT:    [[LSR_IV:%.*]] = phi i64 [ [[LSR_IV_NEXT]], [[BB13]] ], [ 8589934593, [[BB]] ]
+; CHECK-NEXT:    [[T5:%.*]] = phi i64 [ 2, [[BB]] ], [ [[T14:%.*]], [[BB13]] ]
+; CHECK-NEXT:    [[T6:%.*]] = add i64 [[T5]], 4
+; CHECK-NEXT:    [[T7:%.*]] = trunc i64 [[T6]] to i16
+; CHECK-NEXT:    [[T8:%.*]] = urem i16 [[T7]], 3
+; CHECK-NEXT:    [[T9:%.*]] = mul i16 [[T8]], 2
 ; CHECK-NEXT:    [[LSR_IV_NEXT]] = add nuw nsw i64 [[LSR_IV]], 25769803776
 ; CHECK-NEXT:    [[LSR_IV_NEXT2]] = add nuw nsw i16 [[LSR_IV1]], 6
-; CHECK-NEXT:    [[T10:%.*]] = icmp eq i16 1, 0
+; CHECK-NEXT:    [[T14]] = add nuw nsw i64 [[T5]], 6
+; CHECK-NEXT:    [[T10:%.*]] = icmp eq i16 [[T9]], 1
 ; CHECK-NEXT:    br i1 [[T10]], label [[BB11:%.*]], label [[BB13]]
 ; CHECK:       bb11:
 ; CHECK-NEXT:    [[T12:%.*]] = udiv i16 1, [[LSR_IV1]]