[LAA] Always require non-wrapping pointers for runtime checks.

Currently we only check if the pointers involved in runtime checks do
not wrap if we need to perform dependency checks. If that's not the
case, we generate runtime checks, even if the pointers may wrap (see
test/Analysis/LoopAccessAnalysis/runtime-checks-may-wrap.ll).

I might be missing something, but at least for the test it doesn't seem
correct to ignore wrapping pointers. If the pointer wraps, then we swap
start and end of the runtime check.

An Alive2 proof of what the runtime checks are checking conceptually (on
i4 to have it complete in reasonable time) showing the incorrect result
should be https://alive2.llvm.org/ce/z/KsHzn8

Depends on llvm#127410 to avoid
more regressions.

Author: fhahn

llvm/lib/Analysis/LoopAccessAnalysis.cpp

@@ -677,16 +677,17 @@ class AccessAnalysis {
                             const DenseMap<Value *, const SCEV *> &Strides,
                             DenseMap<Value *, unsigned> &DepSetId,
                             Loop *TheLoop, unsigned &RunningDepId,
-                            unsigned ASId, bool ShouldCheckStride, bool Assume);
+                            unsigned ASId, bool Assume);
 
   /// Check whether we can check the pointers at runtime for
   /// non-intersection.
   ///
   /// Returns true if we need no check or if we do and we can generate them
   /// (i.e. the pointers have computable bounds).
   bool canCheckPtrAtRT(RuntimePointerChecking &RtCheck, ScalarEvolution *SE,
-                       Loop *TheLoop, const DenseMap<Value *, const SCEV *> &Strides,
-                       Value *&UncomputablePtr, bool ShouldCheckWrap = false);
+                       Loop *TheLoop,
+                       const DenseMap<Value *, const SCEV *> &Strides,
+                       Value *&UncomputablePtr);
 
   /// Goes over all memory accesses, checks whether a RT check is needed
   /// and builds sets of dependent accesses.
@@ -1115,13 +1116,11 @@ findForkedPointer(PredicatedScalarEvolution &PSE,
   return {{replaceSymbolicStrideSCEV(PSE, StridesMap, Ptr), false}};
 }
 
-bool AccessAnalysis::createCheckForAccess(RuntimePointerChecking &RtCheck,
-                                          MemAccessInfo Access, Type *AccessTy,
-                                          const DenseMap<Value *, const SCEV *> &StridesMap,
-                                          DenseMap<Value *, unsigned> &DepSetId,
-                                          Loop *TheLoop, unsigned &RunningDepId,
-                                          unsigned ASId, bool ShouldCheckWrap,
-                                          bool Assume) {
+bool AccessAnalysis::createCheckForAccess(
+    RuntimePointerChecking &RtCheck, MemAccessInfo Access, Type *AccessTy,
+    const DenseMap<Value *, const SCEV *> &StridesMap,
+    DenseMap<Value *, unsigned> &DepSetId, Loop *TheLoop,
+    unsigned &RunningDepId, unsigned ASId, bool Assume) {
   Value *Ptr = Access.getPointer();
 
   SmallVector<PointerIntPair<const SCEV *, 1, bool>> TranslatedPtrs =
@@ -1152,8 +1151,7 @@ bool AccessAnalysis::createCheckForAccess(RuntimePointerChecking &RtCheck,
 
     // When we run after a failing dependency check we have to make sure
     // we don't have wrapping pointers.
-    if (ShouldCheckWrap &&
-        !isNoWrap(PSE, AR, TranslatedPtrs.size() == 1 ? Ptr : nullptr, AccessTy,
+    if (!isNoWrap(PSE, AR, TranslatedPtrs.size() == 1 ? Ptr : nullptr, AccessTy,
                   TheLoop, Assume)) {
       return false;
     }
@@ -1182,10 +1180,10 @@ bool AccessAnalysis::createCheckForAccess(RuntimePointerChecking &RtCheck,
   return true;
 }
 
-bool AccessAnalysis::canCheckPtrAtRT(RuntimePointerChecking &RtCheck,
-                                     ScalarEvolution *SE, Loop *TheLoop,
-                                     const DenseMap<Value *, const SCEV *> &StridesMap,
-                                     Value *&UncomputablePtr, bool ShouldCheckWrap) {
+bool AccessAnalysis::canCheckPtrAtRT(
+    RuntimePointerChecking &RtCheck, ScalarEvolution *SE, Loop *TheLoop,
+    const DenseMap<Value *, const SCEV *> &StridesMap,
+    Value *&UncomputablePtr) {
   // Find pointers with computable bounds. We are going to use this information
   // to place a runtime bound check.
   bool CanDoRT = true;
@@ -1245,7 +1243,7 @@ bool AccessAnalysis::canCheckPtrAtRT(RuntimePointerChecking &RtCheck,
       for (const auto &AccessTy : Accesses[Access]) {
         if (!createCheckForAccess(RtCheck, Access, AccessTy, StridesMap,
                                   DepSetId, TheLoop, RunningDepId, ASId,
-                                  ShouldCheckWrap, false)) {
+                                  false)) {
           LLVM_DEBUG(dbgs() << "LAA: Can't find bounds for ptr:"
                             << *Access.getPointer() << '\n');
           Retries.emplace_back(Access, AccessTy);
@@ -1275,7 +1273,7 @@ bool AccessAnalysis::canCheckPtrAtRT(RuntimePointerChecking &RtCheck,
       for (const auto &[Access, AccessTy] : Retries) {
         if (!createCheckForAccess(RtCheck, Access, AccessTy, StridesMap,
                                   DepSetId, TheLoop, RunningDepId, ASId,
-                                  ShouldCheckWrap, /*Assume=*/true)) {
+                                  /*Assume=*/true)) {
           CanDoAliasSetRT = false;
           UncomputablePtr = Access.getPointer();
           break;
@@ -2645,9 +2643,8 @@ bool LoopAccessInfo::analyzeLoop(AAResults *AA, const LoopInfo *LI,
   // Find pointers with computable bounds. We are going to use this information
   // to place a runtime bound check.
   Value *UncomputablePtr = nullptr;
-  bool CanDoRTIfNeeded =
-      Accesses.canCheckPtrAtRT(*PtrRtChecking, PSE->getSE(), TheLoop,
-                               SymbolicStrides, UncomputablePtr, false);
+  bool CanDoRTIfNeeded = Accesses.canCheckPtrAtRT(
+      *PtrRtChecking, PSE->getSE(), TheLoop, SymbolicStrides, UncomputablePtr);
   if (!CanDoRTIfNeeded) {
     const auto *I = dyn_cast_or_null<Instruction>(UncomputablePtr);
     recordAnalysis("CantIdentifyArrayBounds", I)
@@ -2678,7 +2675,7 @@ bool LoopAccessInfo::analyzeLoop(AAResults *AA, const LoopInfo *LI,
       auto *SE = PSE->getSE();
       UncomputablePtr = nullptr;
       CanDoRTIfNeeded = Accesses.canCheckPtrAtRT(
-          *PtrRtChecking, SE, TheLoop, SymbolicStrides, UncomputablePtr, true);
+          *PtrRtChecking, SE, TheLoop, SymbolicStrides, UncomputablePtr);
 
       // Check that we found the bounds for the pointer.
       if (!CanDoRTIfNeeded) {

llvm/test/Analysis/LoopAccessAnalysis/nusw-predicates.ll

@@ -15,6 +15,12 @@ define void @int_and_pointer_predicate(ptr %v, i32 %N) {
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Run-time memory checks:
 ; CHECK-NEXT:      Grouped accesses:
+; CHECK-NEXT:        Group [[GRP1:0x[0-9a-f]+]]:
+; CHECK-NEXT:          (Low: %v High: (2 + %v))
+; CHECK-NEXT:            Member: %v
+; CHECK-NEXT:        Group [[GRP2:0x[0-9a-f]+]]:
+; CHECK-NEXT:          (Low: %v High: (6 + (4 * (trunc i32 %N to i16)) + %v))
+; CHECK-NEXT:            Member: {%v,+,4}<%loop>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions:
@@ -57,36 +63,36 @@ define void @int_and_multiple_pointer_predicates(ptr %v, ptr %w, i32 %N) {
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Run-time memory checks:
 ; CHECK-NEXT:      Check 0:
-; CHECK-NEXT:        Comparing group ([[GRP1:0x[0-9a-f]+]]):
+; CHECK-NEXT:        Comparing group ([[GRP3:0x[0-9a-f]+]]):
 ; CHECK-NEXT:        ptr %v
-; CHECK-NEXT:        Against group ([[GRP2:0x[0-9a-f]+]]):
+; CHECK-NEXT:        Against group ([[GRP4:0x[0-9a-f]+]]):
 ; CHECK-NEXT:        ptr %w
 ; CHECK-NEXT:      Check 1:
-; CHECK-NEXT:        Comparing group ([[GRP1]]):
+; CHECK-NEXT:        Comparing group ([[GRP3]]):
 ; CHECK-NEXT:        ptr %v
-; CHECK-NEXT:        Against group ([[GRP3:0x[0-9a-f]+]]):
+; CHECK-NEXT:        Against group ([[GRP5:0x[0-9a-f]+]]):
 ; CHECK-NEXT:          %gep.w = getelementptr { i16, i16 }, ptr %w, i16 %iv.i16
 ; CHECK-NEXT:      Check 2:
-; CHECK-NEXT:        Comparing group ([[GRP4:0x[0-9a-f]+]]):
+; CHECK-NEXT:        Comparing group ([[GRP6:0x[0-9a-f]+]]):
 ; CHECK-NEXT:          %gep.v = getelementptr { i16, i16 }, ptr %v, i16 %iv.i16
-; CHECK-NEXT:        Against group ([[GRP2]]):
+; CHECK-NEXT:        Against group ([[GRP4]]):
 ; CHECK-NEXT:        ptr %w
 ; CHECK-NEXT:      Check 3:
-; CHECK-NEXT:        Comparing group ([[GRP4]]):
+; CHECK-NEXT:        Comparing group ([[GRP6]]):
 ; CHECK-NEXT:          %gep.v = getelementptr { i16, i16 }, ptr %v, i16 %iv.i16
-; CHECK-NEXT:        Against group ([[GRP3]]):
+; CHECK-NEXT:        Against group ([[GRP5]]):
 ; CHECK-NEXT:          %gep.w = getelementptr { i16, i16 }, ptr %w, i16 %iv.i16
 ; CHECK-NEXT:      Grouped accesses:
-; CHECK-NEXT:        Group [[GRP1]]:
+; CHECK-NEXT:        Group [[GRP3]]:
 ; CHECK-NEXT:          (Low: %v High: (2 + %v))
 ; CHECK-NEXT:            Member: %v
-; CHECK-NEXT:        Group [[GRP4]]:
+; CHECK-NEXT:        Group [[GRP6]]:
 ; CHECK-NEXT:          (Low: %v High: (6 + (4 * (trunc i32 %N to i16)) + %v))
 ; CHECK-NEXT:            Member: {%v,+,4}<%loop>
-; CHECK-NEXT:        Group [[GRP2]]:
+; CHECK-NEXT:        Group [[GRP4]]:
 ; CHECK-NEXT:          (Low: %w High: (2 + %w))
 ; CHECK-NEXT:            Member: %w
-; CHECK-NEXT:        Group [[GRP3]]:
+; CHECK-NEXT:        Group [[GRP5]]:
 ; CHECK-NEXT:          (Low: %w High: (6 + (4 * (trunc i32 %N to i16)) + %w))
 ; CHECK-NEXT:            Member: {%w,+,4}<%loop>
 ; CHECK-EMPTY:

llvm/test/Analysis/LoopAccessAnalysis/retry-runtime-checks-after-dependence-analysis-forked-pointers.ll

@@ -163,7 +163,7 @@ exit:
 define void @dependency_check_and_runtime_checks_needed_select_of_ptr_add_recs_may_wrap_1(ptr %a, ptr %b, ptr %c, i64 %offset, i64 %n) {
 ; CHECK-LABEL: 'dependency_check_and_runtime_checks_needed_select_of_ptr_add_recs_may_wrap_1'
 ; CHECK-NEXT:    loop:
-; CHECK-NEXT:      Report: cannot check memory dependencies at runtime
+; CHECK-NEXT:      Report: cannot identify array bounds
 ; CHECK-NEXT:      Dependences:
 ; CHECK-NEXT:      Run-time memory checks:
 ; CHECK-NEXT:      Grouped accesses:
@@ -204,7 +204,7 @@ exit:
 define void @dependency_check_and_runtime_checks_needed_select_of_ptr_add_recs_may_wrap_2(ptr %a, ptr %b, ptr %c, i64 %offset, i64 %n) {
 ; CHECK-LABEL: 'dependency_check_and_runtime_checks_needed_select_of_ptr_add_recs_may_wrap_2'
 ; CHECK-NEXT:    loop:
-; CHECK-NEXT:      Report: cannot check memory dependencies at runtime
+; CHECK-NEXT:      Report: cannot identify array bounds
 ; CHECK-NEXT:      Dependences:
 ; CHECK-NEXT:      Run-time memory checks:
 ; CHECK-NEXT:      Grouped accesses:

llvm/test/Analysis/LoopAccessAnalysis/retry-runtime-checks-after-dependence-analysis.ll

@@ -72,27 +72,27 @@ define void @dependency_check_and_runtime_checks_needed_gepb_not_inbounds_iv2_st
 ; CHECK-NEXT:        Comparing group ([[GRP4:0x[0-9a-f]+]]):
 ; CHECK-NEXT:          %gep.a.iv = getelementptr inbounds float, ptr %a, i64 %iv
 ; CHECK-NEXT:        Against group ([[GRP5:0x[0-9a-f]+]]):
-; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
+; CHECK-NEXT:          %gep.b = getelementptr i8, ptr %b, i64 %iv2
 ; CHECK-NEXT:      Check 1:
 ; CHECK-NEXT:        Comparing group ([[GRP4]]):
 ; CHECK-NEXT:          %gep.a.iv = getelementptr inbounds float, ptr %a, i64 %iv
 ; CHECK-NEXT:        Against group ([[GRP6:0x[0-9a-f]+]]):
-; CHECK-NEXT:          %gep.b = getelementptr i8, ptr %b, i64 %iv2
+; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
 ; CHECK-NEXT:      Check 2:
 ; CHECK-NEXT:        Comparing group ([[GRP5]]):
-; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
-; CHECK-NEXT:        Against group ([[GRP6]]):
 ; CHECK-NEXT:          %gep.b = getelementptr i8, ptr %b, i64 %iv2
+; CHECK-NEXT:        Against group ([[GRP6]]):
+; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
 ; CHECK-NEXT:      Grouped accesses:
 ; CHECK-NEXT:        Group [[GRP4]]:
 ; CHECK-NEXT:          (Low: %a High: ((4 * %n) + %a))
 ; CHECK-NEXT:            Member: {%a,+,4}<nuw><%loop>
 ; CHECK-NEXT:        Group [[GRP5]]:
-; CHECK-NEXT:          (Low: ((4 * %offset) + %a) High: ((4 * %offset) + (4 * %n) + %a))
-; CHECK-NEXT:            Member: {((4 * %offset) + %a),+,4}<%loop>
-; CHECK-NEXT:        Group [[GRP6]]:
 ; CHECK-NEXT:          (Low: %b High: (-1 + (5 * %n) + %b))
 ; CHECK-NEXT:            Member: {%b,+,5}<%loop>
+; CHECK-NEXT:        Group [[GRP6]]:
+; CHECK-NEXT:          (Low: ((4 * %offset) + %a) High: ((4 * %offset) + (4 * %n) + %a))
+; CHECK-NEXT:            Member: {((4 * %offset) + %a),+,4}<%loop>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions:
@@ -265,27 +265,27 @@ define void @dependency_check_and_runtime_checks_needed_gepb_may_wrap(ptr %a, pt
 ; CHECK-NEXT:        Comparing group ([[GRP13:0x[0-9a-f]+]]):
 ; CHECK-NEXT:          %gep.a.iv = getelementptr inbounds float, ptr %a, i64 %iv
 ; CHECK-NEXT:        Against group ([[GRP14:0x[0-9a-f]+]]):
-; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
+; CHECK-NEXT:          %gep.b = getelementptr float, ptr %b, i64 %iv2
 ; CHECK-NEXT:      Check 1:
 ; CHECK-NEXT:        Comparing group ([[GRP13]]):
 ; CHECK-NEXT:          %gep.a.iv = getelementptr inbounds float, ptr %a, i64 %iv
 ; CHECK-NEXT:        Against group ([[GRP15:0x[0-9a-f]+]]):
-; CHECK-NEXT:          %gep.b = getelementptr float, ptr %b, i64 %iv2
+; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
 ; CHECK-NEXT:      Check 2:
 ; CHECK-NEXT:        Comparing group ([[GRP14]]):
-; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
-; CHECK-NEXT:        Against group ([[GRP15]]):
 ; CHECK-NEXT:          %gep.b = getelementptr float, ptr %b, i64 %iv2
+; CHECK-NEXT:        Against group ([[GRP15]]):
+; CHECK-NEXT:          %gep.a.iv.off = getelementptr inbounds float, ptr %a, i64 %iv.offset
 ; CHECK-NEXT:      Grouped accesses:
 ; CHECK-NEXT:        Group [[GRP13]]:
 ; CHECK-NEXT:          (Low: %a High: ((4 * %n) + %a))
 ; CHECK-NEXT:            Member: {%a,+,4}<nuw><%loop>
 ; CHECK-NEXT:        Group [[GRP14]]:
-; CHECK-NEXT:          (Low: ((4 * %offset) + %a) High: ((4 * %offset) + (4 * %n) + %a))
-; CHECK-NEXT:            Member: {((4 * %offset) + %a),+,4}<%loop>
-; CHECK-NEXT:        Group [[GRP15]]:
 ; CHECK-NEXT:          (Low: %b High: (-4 + (8 * %n) + %b))
 ; CHECK-NEXT:            Member: {%b,+,8}<%loop>
+; CHECK-NEXT:        Group [[GRP15]]:
+; CHECK-NEXT:          (Low: ((4 * %offset) + %a) High: ((4 * %offset) + (4 * %n) + %a))
+; CHECK-NEXT:            Member: {((4 * %offset) + %a),+,4}<%loop>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions:

llvm/test/Analysis/LoopAccessAnalysis/runtime-checks-may-wrap.ll

@@ -11,20 +11,21 @@ define void @geps_may_wrap(ptr %a, ptr %b, i64 %N) {
 ; CHECK-NEXT:      Run-time memory checks:
 ; CHECK-NEXT:      Check 0:
 ; CHECK-NEXT:        Comparing group ([[GRP1:0x[0-9a-f]+]]):
-; CHECK-NEXT:          %gep.iv = getelementptr i32, ptr %a, i64 %iv
-; CHECK-NEXT:        Against group ([[GRP2:0x[0-9a-f]+]]):
 ; CHECK-NEXT:        ptr %b
+; CHECK-NEXT:        Against group ([[GRP2:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %gep.iv = getelementptr i32, ptr %a, i64 %iv
 ; CHECK-NEXT:      Grouped accesses:
 ; CHECK-NEXT:        Group [[GRP1]]:
-; CHECK-NEXT:          (Low: %a High: (16 + (12 * (trunc i128 ((zext i64 %N to i128) /u 3) to i16)) + %a))
-; CHECK-NEXT:            Member: {%a,+,12}<%loop>
-; CHECK-NEXT:        Group [[GRP2]]:
 ; CHECK-NEXT:          (Low: %b High: (4 + %b))
 ; CHECK-NEXT:            Member: %b
+; CHECK-NEXT:        Group [[GRP2]]:
+; CHECK-NEXT:          (Low: %a High: (16 + (12 * (trunc i128 ((zext i64 %N to i128) /u 3) to i16)) + %a))
+; CHECK-NEXT:            Member: {%a,+,12}<%loop>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions:
 ; CHECK-NEXT:      {0,+,3}<%loop> Added Flags: <nusw>
+; CHECK-NEXT:      {%a,+,12}<%loop> Added Flags: <nusw>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Expressions re-written:
 ;