[llvm] [Loads] Check if Ptr can be freed between Assume and CtxI. (PR #161255)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 29 11:25:53 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-analysis
Author: Florian Hahn (fhahn)
<details>
<summary>Changes</summary>
When using information from dereferenceable assumptions, we need to make sure that the memory is not freed between the assume and the specified context instruction. Instead of just checking canBeFreed, check if there any calls that may free between the assume and the context instruction.
Note that this also adjusts the context instruction to be the terminator in the loop predecessor, if there is one and it is a branch (to avoid things like invoke).
---
Full diff: https://github.com/llvm/llvm-project/pull/161255.diff
2 Files Affected:
- (modified) llvm/lib/Analysis/Loads.cpp (+41-7)
- (modified) llvm/test/Transforms/LoopVectorize/dereferenceable-info-from-assumption-constant-size.ll (+5-23)
``````````diff
diff --git a/llvm/lib/Analysis/Loads.cpp b/llvm/lib/Analysis/Loads.cpp
index 0c4e3a2e3b233..50fa4264ffcb1 100644
--- a/llvm/lib/Analysis/Loads.cpp
+++ b/llvm/lib/Analysis/Loads.cpp
@@ -32,22 +32,48 @@ static bool isAligned(const Value *Base, Align Alignment,
return Base->getPointerAlignment(DL) >= Alignment;
}
+/// Maximum number of instructions to check between assume and context
+/// instruction.
+static constexpr unsigned MaxInstrsBetweenCheck = 16;
+
+/// Returns true, if no instruction between \p Assume and \p CtxI may free
+/// memory and the function is marked as NoSync. The latter ensures the current
+/// function cannot arrange for another thread to free on its behalf.
+static bool willNotFreeBetween(const Instruction *Assume,
+ const Instruction *CtxI) {
+ if (CtxI->getParent() != Assume->getParent() || !Assume->comesBefore(CtxI))
+ return false;
+ // Make sure the current function cannot arrange for another thread to free on
+ // its behalf.
+ if (!CtxI->getFunction()->hasNoSync())
+ return false;
+
+ // Check if there are any calls between the assume and CtxI that may
+ // free memory.
+ for (const auto &[Idx, I] :
+ enumerate(make_range(Assume->getIterator(), CtxI->getIterator()))) {
+ // Limit number of instructions to walk.
+ if (Idx > MaxInstrsBetweenCheck)
+ return false;
+ if (const auto *CB = dyn_cast<CallBase>(&I))
+ if (!CB->hasFnAttr(Attribute::NoFree))
+ return false;
+ }
+ return true;
+}
+
static bool isDereferenceableAndAlignedPointerViaAssumption(
const Value *Ptr, Align Alignment,
function_ref<bool(const RetainedKnowledge &RK)> CheckSize,
const DataLayout &DL, const Instruction *CtxI, AssumptionCache *AC,
const DominatorTree *DT) {
- // Dereferenceable information from assumptions is only valid if the value
- // cannot be freed between the assumption and use. For now just use the
- // information for values that cannot be freed in the function.
- // TODO: More precisely check if the pointer can be freed between assumption
- // and use.
- if (!CtxI || Ptr->canBeFreed())
+ if (!CtxI)
return false;
/// Look through assumes to see if both dereferencability and alignment can
/// be proven by an assume if needed.
RetainedKnowledge AlignRK;
RetainedKnowledge DerefRK;
+ bool PtrCanBeFreed = Ptr->canBeFreed();
bool IsAligned = Ptr->getPointerAlignment(DL) >= Alignment;
return getKnowledgeForValue(
Ptr, {Attribute::Dereferenceable, Attribute::Alignment}, *AC,
@@ -56,7 +82,11 @@ static bool isDereferenceableAndAlignedPointerViaAssumption(
return false;
if (RK.AttrKind == Attribute::Alignment)
AlignRK = std::max(AlignRK, RK);
- if (RK.AttrKind == Attribute::Dereferenceable)
+
+ // Dereferenceable information from assumptions is only valid if the
+ // value cannot be freed between the assumption and use.
+ if ((!PtrCanBeFreed || willNotFreeBetween(Assume, CtxI)) &&
+ RK.AttrKind == Attribute::Dereferenceable)
DerefRK = std::max(DerefRK, RK);
IsAligned |= AlignRK && AlignRK.ArgValue >= Alignment.value();
if (IsAligned && DerefRK && CheckSize(DerefRK))
@@ -391,6 +421,10 @@ bool llvm::isDereferenceableAndAlignedInLoop(
return false;
Instruction *HeaderFirstNonPHI = &*L->getHeader()->getFirstNonPHIIt();
+ if (BasicBlock *LoopPred = L->getLoopPredecessor()) {
+ if (isa<BranchInst>(LoopPred->getTerminator()))
+ HeaderFirstNonPHI = LoopPred->getTerminator();
+ }
return isDereferenceableAndAlignedPointerViaAssumption(
Base, Alignment,
[&SE, AccessSizeSCEV, &LoopGuards](const RetainedKnowledge &RK) {
diff --git a/llvm/test/Transforms/LoopVectorize/dereferenceable-info-from-assumption-constant-size.ll b/llvm/test/Transforms/LoopVectorize/dereferenceable-info-from-assumption-constant-size.ll
index 75420d40f2aad..bcea03a6b9ba8 100644
--- a/llvm/test/Transforms/LoopVectorize/dereferenceable-info-from-assumption-constant-size.ll
+++ b/llvm/test/Transforms/LoopVectorize/dereferenceable-info-from-assumption-constant-size.ll
@@ -1182,31 +1182,13 @@ define void @deref_assumption_in_header_constant_trip_count_nofree_via_context(p
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2:.*]] ]
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]]
; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x i32>, ptr [[TMP0]], align 4
-; CHECK-NEXT: [[TMP2:%.*]] = icmp slt <2 x i32> [[WIDE_LOAD1]], zeroinitializer
-; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i1> [[TMP2]], i32 0
-; CHECK-NEXT: br i1 [[TMP13]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
-; CHECK: [[PRED_LOAD_IF]]:
-; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP3]]
-; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4
-; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i32> poison, i32 [[TMP5]], i32 0
-; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE]]
-; CHECK: [[PRED_LOAD_CONTINUE]]:
-; CHECK-NEXT: [[TMP7:%.*]] = phi <2 x i32> [ poison, %[[VECTOR_BODY]] ], [ [[TMP6]], %[[PRED_LOAD_IF]] ]
-; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[TMP2]], i32 1
-; CHECK-NEXT: br i1 [[TMP8]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2]]
-; CHECK: [[PRED_LOAD_IF1]]:
-; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP9]]
-; CHECK-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
-; CHECK-NEXT: [[TMP12:%.*]] = insertelement <2 x i32> [[TMP7]], i32 [[TMP11]], i32 1
-; CHECK-NEXT: br label %[[PRED_LOAD_CONTINUE2]]
-; CHECK: [[PRED_LOAD_CONTINUE2]]:
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = phi <2 x i32> [ [[TMP7]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP12]], %[[PRED_LOAD_IF1]] ]
-; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i32> [[WIDE_LOAD]], <2 x i32> [[WIDE_LOAD1]]
+; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <2 x i32> [[WIDE_LOAD1]], zeroinitializer
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i32>, ptr [[TMP1]], align 4
+; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i32> [[WIDE_LOAD1]], <2 x i32> [[WIDE_LOAD2]]
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX]]
; CHECK-NEXT: store <2 x i32> [[PREDPHI]], ptr [[TMP14]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
``````````
</details>
https://github.com/llvm/llvm-project/pull/161255
More information about the llvm-commits
mailing list