[llvm] [LAA] Fix off-by-EltSize in negative-step deref bounds check (PR #211964)

Aleksandr Popov via llvm-commits llvm-commits at lists.llvm.org
Sun Aug 2 16:08:58 PDT 2026


https://github.com/aleks-tmb updated https://github.com/llvm/llvm-project/pull/211964

>From d8a8fb54108eec8f47bfc98235eaed887cb68e29 Mon Sep 17 00:00:00 2001
From: Aleksandr Popov <apopov at azul.com>
Date: Mon, 27 Jul 2026 00:08:52 +0000
Subject: [PATCH] [LAA] Fix off-by-EltSize bounds in reverse-loop deref no-wrap
 check

evaluatePtrAddRecAtMaxBTCWillNotWrap used AR->getStart() as the
lowest accessed address; for a negative step it is the *highest*.
Both safety checks on the reverse-loop branch inherited this and
were off by EltSize:

  * No-underflow: added an extra EltSize of slack, rejecting
    reverse loops whose last iteration lands on the base pointer.
  * Deref-end: dropped the size of the top access, accepting
    loops whose top read spills past DerefBytes.

Pick LowestAddr per step direction (AR->getStart() vs.
AR->evaluateAtIteration(MaxBTC, SE)). The range algebra downstream
becomes direction-agnostic and the per-direction MaxOffset branch
collapses into a single expression.
---
 llvm/lib/Analysis/LoopAccessAnalysis.cpp      | 55 +++++++++----------
 .../negative-step-deref-off-by-eltsize.ll     | 12 ++--
 2 files changed, 31 insertions(+), 36 deletions(-)

diff --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index c99d43dd1ccc2..9c1c79089c43d 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -208,8 +208,18 @@ static const SCEV *mulSCEVNoOverflow(const SCEV *A, const SCEV *B,
   return SE.getMulExpr(A, B);
 }
 
-/// Return true, if evaluating \p AR at \p MaxBTC cannot wrap, because \p AR at
+/// Return true if evaluating \p AR at \p MaxBTC cannot wrap, because \p AR at
 /// \p MaxBTC is guaranteed inbounds of the accessed object.
+///
+/// The accessed byte range is [LowestOffset, LowestOffset + AccessedBytes),
+/// where
+///   AccessedBytes = MaxBTC * |Step| + EltSize,
+///   LowestOffset  = smallest byte offset from StartPtr any iteration reaches.
+///
+/// The function returns true only when both safety invariants hold, regardless
+/// of step direction:
+///   1. LowestOffset >= 0                          (no access below StartPtr)
+///   2. LowestOffset + AccessedBytes <= DerefBytes (no access past the region)
 static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
     const SCEVAddRecExpr *AR, const SCEV *MaxBTC, const SCEV *EltSize,
     ScalarEvolution &SE, const DataLayout &DL, DominatorTree *DT,
@@ -267,15 +277,15 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
   Step = SE.getNoopOrSignExtend(Step, WiderTy);
   MaxBTC = SE.getNoopOrZeroExtend(MaxBTC, WiderTy);
 
-  // For the computations below, make sure they don't unsigned wrap.
-  // FIXME: for a negative step the lowest accessed address is not
-  // AR->getStart() but AR->evaluateAtIteration(MaxBTC, SE); the check below
-  // therefore compares StartPtr against the highest accessed address instead
-  // of the lowest.
-  if (!SE.isKnownPredicate(CmpInst::ICMP_UGE, AR->getStart(), StartPtr))
+  const SCEV *LowestAddr = IsKnownNonNegative
+                               ? static_cast<const SCEV *>(AR->getStart())
+                               : AR->evaluateAtIteration(MaxBTC, SE);
+  // Lower-bound safety check: the lowest accessed address must not fall below
+  // StartPtr.
+  if (!SE.isKnownPredicate(CmpInst::ICMP_UGE, LowestAddr, StartPtr))
     return false;
-  const SCEV *StartOffset = SE.getNoopOrZeroExtend(
-      SE.getMinusSCEV(AR->getStart(), StartPtr), WiderTy);
+  const SCEV *LowestOffset =
+      SE.getNoopOrZeroExtend(SE.getMinusSCEV(LowestAddr, StartPtr), WiderTy);
 
   if (!LoopGuards)
     LoopGuards.emplace(ScalarEvolution::LoopGuards::collect(AR->getLoop(), SE));
@@ -304,27 +314,12 @@ static bool evaluatePtrAddRecAtMaxBTCWillNotWrap(
   if (!AccessedBytes)
     return false;
 
-  // Compute MaxOffset per direction: exclusive upper offset of the
-  // accessed range.
-  const SCEV *MaxOffset;
-  if (IsKnownNonNegative) {
-    MaxOffset = addSCEVNoOverflow(StartOffset, AccessedBytes, SE);
-    if (!MaxOffset)
-      return false;
-    DerefBytesSCEV = SE.applyLoopGuards(DerefBytesSCEV, *LoopGuards);
-  } else {
-    // FIXME: two independent off-by-EltSize bugs on this branch:
-    //  1. StartOffset here is actually the HIGHEST offset, because it is
-    //     computed from AR->getStart() rather than
-    //     AR->evaluateAtIteration(MaxBTC, SE) (see FIXME above).
-    //  2. The lower check is over-strict by EltSize and the upper is
-    //     under-counted by EltSize.
-    assert(SE.isKnownNegative(Step) && "must be known negative");
-    if (!SE.isKnownPredicate(CmpInst::ICMP_SGE, StartOffset, AccessedBytes))
-      return false;
-    MaxOffset = StartOffset;
-  }
-  // MaxOffset must not exceed the deref-region end.
+  // Exclusive upper offset of the accessed range.
+  const SCEV *MaxOffset = addSCEVNoOverflow(LowestOffset, AccessedBytes, SE);
+  if (!MaxOffset)
+    return false;
+  DerefBytesSCEV = SE.applyLoopGuards(DerefBytesSCEV, *LoopGuards);
+  // Upper-bound safety check: MaxOffset must not exceed the deref-region end.
   return SE.isKnownPredicate(CmpInst::ICMP_ULE, MaxOffset, DerefBytesSCEV);
 }
 
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/negative-step-deref-off-by-eltsize.ll b/llvm/test/Analysis/LoopAccessAnalysis/negative-step-deref-off-by-eltsize.ll
index 1a53fba949749..872a948de6cb5 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/negative-step-deref-off-by-eltsize.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/negative-step-deref-off-by-eltsize.ll
@@ -4,7 +4,7 @@
 ; Reverse loop loading 4 i32 elements whose access range exactly fills the
 ; dereferenceable region (deref(16), reads bytes [0, 16)).
 ;
-; TODO: LAA should recognise that this AR fits within the deref
+; LAA should recognise that this AR fits within the deref
 ; region and produce tight bounds (Low: %A, High: %A + 16).
 ;
 ; Pseudocode:
@@ -28,10 +28,10 @@ define void @reverse_reaches_base(ptr dereferenceable(16) %A, ptr dereferenceabl
 ; CHECK-NEXT:          %gep.A = getelementptr inbounds i32, ptr %A, i64 %iv
 ; CHECK-NEXT:      Grouped accesses:
 ; CHECK-NEXT:        Group GRP0:
-; CHECK-NEXT:          (Low: (-4 + inttoptr (i64 -1 to ptr))<nsw> High: (16 + %B)<nuw>)
+; CHECK-NEXT:          (Low: %B High: (16 + %B)<nuw>)
 ; CHECK-NEXT:            Member: {(12 + %B)<nuw>,+,-4}<nw><%loop>
 ; CHECK-NEXT:        Group GRP1:
-; CHECK-NEXT:          (Low: (-4 + inttoptr (i64 -1 to ptr))<nsw> High: (16 + %A)<nuw>)
+; CHECK-NEXT:          (Low: %A High: (16 + %A)<nuw>)
 ; CHECK-NEXT:            Member: {(12 + %A)<nuw>,+,-4}<nw><%loop>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
@@ -67,7 +67,7 @@ exit.done:
 ; The top i32 read at byte 13 covers [13, 17), but deref(16) only
 ; guarantees [0, 16) — bytes at/after 16 may or may not be dereferenceable.
 ;
-; TODO: LAA must not assume the AR fits in the deref region and should
+; LAA must not assume the AR fits in the deref region and should
 ; fall back to the wide low bound.
 ;
 ; Pseudocode:
@@ -90,10 +90,10 @@ define void @reverse_top_spills(ptr dereferenceable(16) %A, ptr dereferenceable(
 ; CHECK-NEXT:          %gep.A = getelementptr inbounds i8, ptr %A, i64 %iv
 ; CHECK-NEXT:      Grouped accesses:
 ; CHECK-NEXT:        Group GRP0:
-; CHECK-NEXT:          (Low: (5 + %B)<nuw> High: (17 + %B))
+; CHECK-NEXT:          (Low: (-4 + inttoptr (i64 -1 to ptr))<nsw> High: (17 + %B))
 ; CHECK-NEXT:            Member: {(13 + %B)<nuw>,+,-4}<nw><%loop>
 ; CHECK-NEXT:        Group GRP1:
-; CHECK-NEXT:          (Low: (5 + %A)<nuw> High: (17 + %A))
+; CHECK-NEXT:          (Low: (-4 + inttoptr (i64 -1 to ptr))<nsw> High: (17 + %A))
 ; CHECK-NEXT:            Member: {(13 + %A)<nuw>,+,-4}<nw><%loop>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.



More information about the llvm-commits mailing list