[LAA] Drop alias scope metadata that is not valid across iterations (l…

…lvm#79161)

LAA currently adds memory locations with their original AATags to AST.
However, scoped alias AATags may be valid only within one loop
iteration, while LAA reasons across iterations.

Fix this by determining which alias scopes are defined inside the loop,
and drop AATags that reference these scopes.

Fixes llvm#79137.

(cherry picked from commit cd7ea4e)

llvm/lib/Analysis/LoopAccessAnalysis.cpp

@@ -625,14 +625,17 @@ class AccessAnalysis {
 
   AccessAnalysis(Loop *TheLoop, AAResults *AA, LoopInfo *LI,
                  MemoryDepChecker::DepCandidates &DA,
-                 PredicatedScalarEvolution &PSE)
-      : TheLoop(TheLoop), BAA(*AA), AST(BAA), LI(LI), DepCands(DA), PSE(PSE) {
+                 PredicatedScalarEvolution &PSE,
+                 SmallPtrSetImpl<MDNode *> &LoopAliasScopes)
+      : TheLoop(TheLoop), BAA(*AA), AST(BAA), LI(LI), DepCands(DA), PSE(PSE),
+        LoopAliasScopes(LoopAliasScopes) {
     // We're analyzing dependences across loop iterations.
     BAA.enableCrossIterationMode();
   }
 
   /// Register a load  and whether it is only read from.
-  void addLoad(MemoryLocation &Loc, Type *AccessTy, bool IsReadOnly) {
+  void addLoad(MemoryLocation Loc, Type *AccessTy, bool IsReadOnly) {
+    Loc = adjustLoc(Loc);
     Value *Ptr = const_cast<Value*>(Loc.Ptr);
     AST.add(Ptr, LocationSize::beforeOrAfterPointer(), Loc.AATags);
     Accesses[MemAccessInfo(Ptr, false)].insert(AccessTy);
@@ -641,7 +644,8 @@ class AccessAnalysis {
   }
 
   /// Register a store.
-  void addStore(MemoryLocation &Loc, Type *AccessTy) {
+  void addStore(MemoryLocation Loc, Type *AccessTy) {
+    Loc = adjustLoc(Loc);
     Value *Ptr = const_cast<Value*>(Loc.Ptr);
     AST.add(Ptr, LocationSize::beforeOrAfterPointer(), Loc.AATags);
     Accesses[MemAccessInfo(Ptr, true)].insert(AccessTy);
@@ -694,6 +698,32 @@ class AccessAnalysis {
 private:
   typedef MapVector<MemAccessInfo, SmallSetVector<Type *, 1>> PtrAccessMap;
 
+  /// Adjust the MemoryLocation so that it represents accesses to this
+  /// location across all iterations, rather than a single one.
+  MemoryLocation adjustLoc(MemoryLocation Loc) const {
+    // The accessed location varies within the loop, but remains within the
+    // underlying object.
+    Loc.Size = LocationSize::beforeOrAfterPointer();
+    Loc.AATags.Scope = adjustAliasScopeList(Loc.AATags.Scope);
+    Loc.AATags.NoAlias = adjustAliasScopeList(Loc.AATags.NoAlias);
+    return Loc;
+  }
+
+  /// Drop alias scopes that are only valid within a single loop iteration.
+  MDNode *adjustAliasScopeList(MDNode *ScopeList) const {
+    if (!ScopeList)
+      return nullptr;
+
+    // For the sake of simplicity, drop the whole scope list if any scope is
+    // iteration-local.
+    if (any_of(ScopeList->operands(), [&](Metadata *Scope) {
+          return LoopAliasScopes.contains(cast<MDNode>(Scope));
+        }))
+      return nullptr;
+
+    return ScopeList;
+  }
+
   /// Go over all memory access and check whether runtime pointer checks
   /// are needed and build sets of dependency check candidates.
   void processMemAccesses();
@@ -736,6 +766,10 @@ class AccessAnalysis {
 
   /// The SCEV predicate containing all the SCEV-related assumptions.
   PredicatedScalarEvolution &PSE;
+
+  /// Alias scopes that are declared inside the loop, and as such not valid
+  /// across iterations.
+  SmallPtrSetImpl<MDNode *> &LoopAliasScopes;
 };
 
 } // end anonymous namespace
@@ -2149,6 +2183,7 @@ void LoopAccessInfo::analyzeLoop(AAResults *AA, LoopInfo *LI,
   // Holds the Load and Store instructions.
   SmallVector<LoadInst *, 16> Loads;
   SmallVector<StoreInst *, 16> Stores;
+  SmallPtrSet<MDNode *, 8> LoopAliasScopes;
 
   // Holds all the different accesses in the loop.
   unsigned NumReads = 0;
@@ -2192,6 +2227,11 @@ void LoopAccessInfo::analyzeLoop(AAResults *AA, LoopInfo *LI,
       if (HasComplexMemInst)
         continue;
 
+      // Record alias scopes defined inside the loop.
+      if (auto *Decl = dyn_cast<NoAliasScopeDeclInst>(&I))
+        for (Metadata *Op : Decl->getScopeList()->operands())
+          LoopAliasScopes.insert(cast<MDNode>(Op));
+
       // If this is a load, save it. If this instruction can read from memory
       // but is not a load, then we quit. Notice that we don't handle function
       // calls that read or write.
@@ -2273,7 +2313,8 @@ void LoopAccessInfo::analyzeLoop(AAResults *AA, LoopInfo *LI,
   }
 
   MemoryDepChecker::DepCandidates DependentAccesses;
-  AccessAnalysis Accesses(TheLoop, AA, LI, DependentAccesses, *PSE);
+  AccessAnalysis Accesses(TheLoop, AA, LI, DependentAccesses, *PSE,
+                          LoopAliasScopes);
 
   // Holds the analyzed pointers. We don't want to call getUnderlyingObjects
   // multiple times on the same object. If the ptr is accessed twice, once

llvm/test/Analysis/LoopAccessAnalysis/noalias-scope-decl.ll

@@ -7,8 +7,17 @@
 define void @test_scope_in_loop(ptr %arg, i64 %num) {
 ; CHECK-LABEL: 'test_scope_in_loop'
 ; CHECK-NEXT:    loop:
-; CHECK-NEXT:      Memory dependences are safe
+; CHECK-NEXT:      Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
+; CHECK-NEXT:  Backward loop carried data dependence.
 ; CHECK-NEXT:      Dependences:
+; CHECK-NEXT:        Backward:
+; CHECK-NEXT:            %load.prev = load i8, ptr %prev.ptr, align 1, !alias.scope !0, !noalias !3 ->
+; CHECK-NEXT:            store i8 %add, ptr %cur.ptr, align 1, !alias.scope !3
+; CHECK-EMPTY:
+; CHECK-NEXT:        Forward:
+; CHECK-NEXT:            %load.cur = load i8, ptr %cur.ptr, align 1, !alias.scope !3 ->
+; CHECK-NEXT:            store i8 %add, ptr %cur.ptr, align 1, !alias.scope !3
+; CHECK-EMPTY:
 ; CHECK-NEXT:      Run-time memory checks:
 ; CHECK-NEXT:      Grouped accesses:
 ; CHECK-EMPTY:

llvm/test/Transforms/PhaseOrdering/X86/loop-vectorizer-noalias.ll

@@ -14,89 +14,25 @@ entry:
 }
 
 ; This loop should not get vectorized.
-; FIXME: This is a miscompile.
 define void @accsum(ptr noundef %vals, i64 noundef %num) #0 {
 ; CHECK-LABEL: define void @accsum(
 ; CHECK-SAME: ptr nocapture noundef [[VALS:%.*]], i64 noundef [[NUM:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CMP1:%.*]] = icmp ugt i64 [[NUM]], 1
-; CHECK-NEXT:    br i1 [[CMP1]], label [[ITER_CHECK:%.*]], label [[FOR_END:%.*]]
-; CHECK:       iter.check:
-; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[NUM]], -1
-; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[NUM]], 9
-; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY_PREHEADER:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
-; CHECK:       vector.main.loop.iter.check:
-; CHECK-NEXT:    [[MIN_ITERS_CHECK3:%.*]] = icmp ult i64 [[NUM]], 33
-; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK3]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
-; CHECK:       vector.ph:
-; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP0]], -32
-; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
-; CHECK:       vector.body:
-; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = or disjoint i64 [[INDEX]], 1
-; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[VALS]], i64 [[OFFSET_IDX]]
-; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 -1
-; CHECK-NEXT:    tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0:![0-9]+]])
-; CHECK-NEXT:    tail call void @llvm.experimental.noalias.scope.decl(metadata [[META3:![0-9]+]])
-; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 15
-; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP2]], align 1, !alias.scope [[META3]], !noalias [[META0]]
-; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP3]], align 1, !alias.scope [[META3]], !noalias [[META0]]
-; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 16
-; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1, !alias.scope [[META0]], !noalias [[META3]]
-; CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP4]], align 1, !alias.scope [[META0]], !noalias [[META3]]
-; CHECK-NEXT:    [[TMP5:%.*]] = add <16 x i8> [[WIDE_LOAD5]], [[WIDE_LOAD]]
-; CHECK-NEXT:    [[TMP6:%.*]] = add <16 x i8> [[WIDE_LOAD6]], [[WIDE_LOAD4]]
-; CHECK-NEXT:    store <16 x i8> [[TMP5]], ptr [[TMP1]], align 1, !alias.scope [[META0]], !noalias [[META3]]
-; CHECK-NEXT:    store <16 x i8> [[TMP6]], ptr [[TMP4]], align 1, !alias.scope [[META0]], !noalias [[META3]]
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
-; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
-; CHECK:       middle.block:
-; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
-; CHECK:       vec.epilog.iter.check:
-; CHECK-NEXT:    [[IND_END9:%.*]] = or disjoint i64 [[N_VEC]], 1
-; CHECK-NEXT:    [[N_VEC_REMAINING:%.*]] = and i64 [[TMP0]], 24
-; CHECK-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0
-; CHECK-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[FOR_BODY_PREHEADER]], label [[VEC_EPILOG_PH]]
-; CHECK:       vec.epilog.ph:
-; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
-; CHECK-NEXT:    [[N_VEC8:%.*]] = and i64 [[TMP0]], -8
-; CHECK-NEXT:    [[IND_END:%.*]] = or disjoint i64 [[N_VEC8]], 1
-; CHECK-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
-; CHECK:       vec.epilog.vector.body:
-; CHECK-NEXT:    [[INDEX11:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT15:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; CHECK-NEXT:    [[OFFSET_IDX12:%.*]] = or disjoint i64 [[INDEX11]], 1
-; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[VALS]], i64 [[OFFSET_IDX12]]
-; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr i8, ptr [[TMP8]], i64 -1
-; CHECK-NEXT:    tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0]])
-; CHECK-NEXT:    tail call void @llvm.experimental.noalias.scope.decl(metadata [[META3]])
-; CHECK-NEXT:    [[WIDE_LOAD13:%.*]] = load <8 x i8>, ptr [[TMP9]], align 1, !alias.scope [[META3]], !noalias [[META0]]
-; CHECK-NEXT:    [[WIDE_LOAD14:%.*]] = load <8 x i8>, ptr [[TMP8]], align 1, !alias.scope [[META0]], !noalias [[META3]]
-; CHECK-NEXT:    [[TMP10:%.*]] = add <8 x i8> [[WIDE_LOAD14]], [[WIDE_LOAD13]]
-; CHECK-NEXT:    store <8 x i8> [[TMP10]], ptr [[TMP8]], align 1, !alias.scope [[META0]], !noalias [[META3]]
-; CHECK-NEXT:    [[INDEX_NEXT15]] = add nuw i64 [[INDEX11]], 8
-; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT15]], [[N_VEC8]]
-; CHECK-NEXT:    br i1 [[TMP11]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
-; CHECK:       vec.epilog.middle.block:
-; CHECK-NEXT:    [[CMP_N10:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC8]]
-; CHECK-NEXT:    br i1 [[CMP_N10]], label [[FOR_END]], label [[FOR_BODY_PREHEADER]]
+; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
 ; CHECK:       for.body.preheader:
-; CHECK-NEXT:    [[I_02_PH:%.*]] = phi i64 [ 1, [[ITER_CHECK]] ], [ [[IND_END9]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    [[LOAD_INITIAL:%.*]] = load i8, ptr [[VALS]], align 1
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
-; CHECK-NEXT:    [[I_02:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_BODY]] ], [ [[I_02_PH]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[STORE_FORWARDED:%.*]] = phi i8 [ [[LOAD_INITIAL]], [[FOR_BODY_PREHEADER]] ], [ [[ADD_I:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[I_02:%.*]] = phi i64 [ 1, [[FOR_BODY_PREHEADER]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[VALS]], i64 [[I_02]]
-; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr i8, ptr [[ARRAYIDX]], i64 -1
-; CHECK-NEXT:    tail call void @llvm.experimental.noalias.scope.decl(metadata [[META0]])
-; CHECK-NEXT:    tail call void @llvm.experimental.noalias.scope.decl(metadata [[META3]])
-; CHECK-NEXT:    [[TMP12:%.*]] = load i8, ptr [[ARRAYIDX1]], align 1, !alias.scope [[META3]], !noalias [[META0]]
-; CHECK-NEXT:    [[TMP13:%.*]] = load i8, ptr [[ARRAYIDX]], align 1, !alias.scope [[META0]], !noalias [[META3]]
-; CHECK-NEXT:    [[ADD_I:%.*]] = add i8 [[TMP13]], [[TMP12]]
+; CHECK-NEXT:    [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1, !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]]
+; CHECK-NEXT:    [[ADD_I]] = add i8 [[TMP0]], [[STORE_FORWARDED]]
 ; CHECK-NEXT:    store i8 [[ADD_I]], ptr [[ARRAYIDX]], align 1, !alias.scope [[META0]], !noalias [[META3]]
 ; CHECK-NEXT:    [[INC]] = add nuw i64 [[I_02]], 1
 ; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[NUM]]
-; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]]
 ; CHECK:       for.end:
 ; CHECK-NEXT:    ret void
 ;
@@ -139,9 +75,4 @@ attributes #0 = { "target-features"="+cmov,+cx8,+fxsr,+mmx,+sse,+sse2,+x87"}
 ; CHECK: [[META2]] = distinct !{[[META2]], !"acc"}
 ; CHECK: [[META3]] = !{[[META4:![0-9]+]]}
 ; CHECK: [[META4]] = distinct !{[[META4]], [[META2]], !"acc: %prev"}
-; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META6:![0-9]+]], [[META7:![0-9]+]]}
-; CHECK: [[META6]] = !{!"llvm.loop.isvectorized", i32 1}
-; CHECK: [[META7]] = !{!"llvm.loop.unroll.runtime.disable"}
-; CHECK: [[LOOP8]] = distinct !{[[LOOP8]], [[META6]], [[META7]]}
-; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META7]], [[META6]]}
 ;.