Improve BasicAA CS-CS queries

[hfinkel]

Summary

BasicAA contains knowledge of certain intrinsics, such as memcpy and memset, and uses that information to form more-accurate answers to CallSite vs. Loc ModRef queries. Unfortunately, it does not use this information when answering CallSite vs. CallSite queries.

Generically, when an intrinsic takes one or more pointers and the intrinsic is marked only to read/write from its arguments, the offset/size is unknown. As a result, the generic code that answers CallSite vs. CallSite (and CallSite vs. Loc) queries in AA uses UnknownSize when forming Locs from an intrinsic's arguments. While BasicAA's CallSite vs. Loc override could use more-accurate size information for some intrinsics, it did not do the same for CallSite vs. CallSite queries.

This patch refactors the intrinsic-specific logic in BasicAA into a generic AA query function: getArgLocation, which is overridden by BasicAA to supply the intrinsic-specific knowledge, and used by AA's generic implementation. This allows the intrinsic-specific knowledge to be used by both CallSite vs. Loc and CallSite vs. CallSite queries, and simplifies the BasicAA implementation.

Currently, only one function, Mac's memset_pattern16, is handled by BasicAA (all the rest are intrinsics). As a side-effect of this refactoring, BasicAA's getModRefBehavior override now also returns OnlyAccessesArgumentPointees for this function (which is an improvement).

[aschwaighofer]

LGTM. Sorry for the delay.

[hfinkel]

Committed r212572. Thanks!

[hfinkel]

Committed r212572.

include/llvm/Analysis/AliasAnalysis.h

enum ModRefBehavior {
     /// This property corresponds to the IntrReadMem LLVM intrinsic flag.
     OnlyReadsMemory = Anywhere | Ref,
 
     /// UnknownModRefBehavior - This indicates that the function could not be
     /// classified into one of the behaviors above.
     UnknownModRefBehavior = Anywhere | ModRef
   };
 
+  /// Get the location associated with a pointer argument of a callsite.
+  /// The mask bits are set to indicate the allowed aliasing ModRef kinds.
+  /// Note that these mask bits do not necessarily account for the overall
+  /// behavior of the function, but rather only provide additional
+  /// per-argument information.
+  virtual Location getArgLocation(ImmutableCallSite CS, unsigned ArgIdx,
+                                  ModRefResult &Mask);
+
   /// getModRefBehavior - Return the behavior when calling the given call site.
   virtual ModRefBehavior getModRefBehavior(ImmutableCallSite CS);
 
   /// getModRefBehavior - Return the behavior when calling the given function.
   /// For use when the call site is not known.
   virtual ModRefBehavior getModRefBehavior(const Function *F);
 
   /// doesNotAccessMemory - If the specified call is known to never read or

lib/Analysis/AliasAnalysis.cpp

 }
 
 bool AliasAnalysis::pointsToConstantMemory(const Location &Loc,
                                            bool OrLocal) {
   assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
   return AA->pointsToConstantMemory(Loc, OrLocal);
 }
 
+AliasAnalysis::Location
+AliasAnalysis::getArgLocation(ImmutableCallSite CS, unsigned ArgIdx,
+                              AliasAnalysis::ModRefResult &Mask) {
+  assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
+  return AA->getArgLocation(CS, ArgIdx, Mask);
+}
+
 void AliasAnalysis::deleteValue(Value *V) {
   assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
   AA->deleteValue(V);
 }
 
 void AliasAnalysis::copyValue(Value *From, Value *To) {
   assert(AA && "AA didn't call InitializeAliasAnalysis in its run method!");
   AA->copyValue(From, To);
if (MRB == DoesNotAccessMemory)
     return NoModRef;
 
   ModRefResult Mask = ModRef;
   if (onlyReadsMemory(MRB))
     Mask = Ref;
 
   if (onlyAccessesArgPointees(MRB)) {
     bool doesAlias = false;
+    ModRefResult AllArgsMask = NoModRef;
     if (doesAccessArgPointees(MRB)) {
-      MDNode *CSTag = CS.getInstruction()->getMetadata(LLVMContext::MD_tbaa);
       for (ImmutableCallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end();
            AI != AE; ++AI) {
         const Value *Arg = *AI;
         if (!Arg->getType()->isPointerTy())
           continue;
-        Location CSLoc(Arg, UnknownSize, CSTag);
+        ModRefResult ArgMask;
+        Location CSLoc =
+          getArgLocation(CS, (unsigned) std::distance(CS.arg_begin(), AI),
+                         ArgMask);
         if (!isNoAlias(CSLoc, Loc)) {
           doesAlias = true;
-          break;
+          AllArgsMask = ModRefResult(AllArgsMask | ArgMask);
         }
       }
     }
     if (!doesAlias)
       return NoModRef;
+    Mask = ModRefResult(Mask & AllArgsMask);
   }
 
   // If Loc is a constant memory location, the call definitely could not
   // modify the memory location.
   if ((Mask & Mod) && pointsToConstantMemory(Loc))
     Mask = ModRefResult(Mask & ~Mod);
 
   // If this is the end of the chain, don't forward.
if (onlyReadsMemory(CS1B))
     Mask = ModRefResult(Mask & Ref);
 
   // If CS2 only access memory through arguments, accumulate the mod/ref
   // information from CS1's references to the memory referenced by
   // CS2's arguments.
   if (onlyAccessesArgPointees(CS2B)) {
     AliasAnalysis::ModRefResult R = NoModRef;
     if (doesAccessArgPointees(CS2B)) {
-      MDNode *CS2Tag = CS2.getInstruction()->getMetadata(LLVMContext::MD_tbaa);
       for (ImmutableCallSite::arg_iterator
            I = CS2.arg_begin(), E = CS2.arg_end(); I != E; ++I) {
         const Value *Arg = *I;
         if (!Arg->getType()->isPointerTy())
           continue;
-        Location CS2Loc(Arg, UnknownSize, CS2Tag);
-        R = ModRefResult((R | getModRefInfo(CS1, CS2Loc)) & Mask);
+        ModRefResult ArgMask;
+        Location CS2Loc =
+          getArgLocation(CS2, (unsigned) std::distance(CS2.arg_begin(), I),
+                         ArgMask);
+        // ArgMask indicates what CS2 might do to CS2Loc, and the dependence of
+        // CS1 on that location is the inverse.
+        if (ArgMask == Mod)
+          ArgMask = ModRef;
+        else if (ArgMask == Ref)
+          ArgMask = Mod;
+
+        R = ModRefResult((R | (getModRefInfo(CS1, CS2Loc) & ArgMask)) & Mask);
         if (R == Mask)
           break;
       }
     }
     return R;
   }
 
   // If CS1 only accesses memory through arguments, check if CS2 references
   // any of the memory referenced by CS1's arguments. If not, return NoModRef.
   if (onlyAccessesArgPointees(CS1B)) {
     AliasAnalysis::ModRefResult R = NoModRef;
     if (doesAccessArgPointees(CS1B)) {
-      MDNode *CS1Tag = CS1.getInstruction()->getMetadata(LLVMContext::MD_tbaa);
       for (ImmutableCallSite::arg_iterator
            I = CS1.arg_begin(), E = CS1.arg_end(); I != E; ++I) {
         const Value *Arg = *I;
         if (!Arg->getType()->isPointerTy())
           continue;
-        Location CS1Loc(Arg, UnknownSize, CS1Tag);
-        if (getModRefInfo(CS2, CS1Loc) != NoModRef) {
+        ModRefResult ArgMask;
+        Location CS1Loc =
+          getArgLocation(CS1, (unsigned) std::distance(CS1.arg_begin(), I),
+                         ArgMask);
+        if ((getModRefInfo(CS2, CS1Loc) & ArgMask) != NoModRef) {
           R = Mask;
           break;
         }
       }
     }
     if (R == NoModRef)
       return R;
   }

lib/Analysis/BasicAliasAnalysis.cpp

ModRefResult getModRefInfo(ImmutableCallSite CS1,
       // The AliasAnalysis base class has some smarts, lets use them.
       return AliasAnalysis::getModRefInfo(CS1, CS2);
     }
 
     /// pointsToConstantMemory - Chase pointers until we find a (constant
     /// global) or not.
     bool pointsToConstantMemory(const Location &Loc, bool OrLocal) override;
 
+    /// Get the location associated with a pointer argument of a callsite.
+    Location getArgLocation(ImmutableCallSite CS, unsigned ArgIdx,
+                            ModRefResult &Mask) override;
+
     /// getModRefBehavior - Return the behavior when calling the given
     /// call site.
     ModRefBehavior getModRefBehavior(ImmutableCallSite CS) override;
 
     /// getModRefBehavior - Return the behavior when calling the given function.
     /// For use when the call site is not known.
     ModRefBehavior getModRefBehavior(const Function *F) override;
 
do {
     return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
 
   } while (!Worklist.empty() && --MaxLookup);
 
   Visited.clear();
   return Worklist.empty();
 }
 
+static bool isMemsetPattern16(const Function *MS,
+                              const TargetLibraryInfo &TLI) {
+  if (TLI.has(LibFunc::memset_pattern16) &&
+      MS->getName() == "memset_pattern16") {
+    FunctionType *MemsetType = MS->getFunctionType();
+    if (!MemsetType->isVarArg() && MemsetType->getNumParams() == 3 &&
+        isa<PointerType>(MemsetType->getParamType(0)) &&
+        isa<PointerType>(MemsetType->getParamType(1)) &&
+        isa<IntegerType>(MemsetType->getParamType(2)))
+      return true;
+  }
+
+  return false;
+}
+
 /// getModRefBehavior - Return the behavior when calling the given call site.
 AliasAnalysis::ModRefBehavior
 BasicAliasAnalysis::getModRefBehavior(ImmutableCallSite CS) {
   if (CS.doesNotAccessMemory())
     // Can't do better than this.
     return DoesNotAccessMemory;
 
   ModRefBehavior Min = UnknownModRefBehavior;
#undef GET_INTRINSIC_MODREF_BEHAVIOR
   }
 
   ModRefBehavior Min = UnknownModRefBehavior;
 
   // If the function declares it only reads memory, go with that.
   if (F->onlyReadsMemory())
     Min = OnlyReadsMemory;
 
+  const TargetLibraryInfo &TLI = getAnalysis<TargetLibraryInfo>();
+  if (isMemsetPattern16(F, TLI))
+    Min = OnlyAccessesArgumentPointees;
+
   // Otherwise be conservative.
   return ModRefBehavior(AliasAnalysis::getModRefBehavior(F) & Min);
 }
 
+AliasAnalysis::Location
+BasicAliasAnalysis::getArgLocation(ImmutableCallSite CS, unsigned ArgIdx,
+                                   ModRefResult &Mask) {
+  Location Loc = AliasAnalysis::getArgLocation(CS, ArgIdx, Mask);
+  const TargetLibraryInfo &TLI = getAnalysis<TargetLibraryInfo>();
+  const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction());
+  if (II != nullptr)
+    switch (II->getIntrinsicID()) {
+    default: break;
+    case Intrinsic::memset:
+    case Intrinsic::memcpy:
+    case Intrinsic::memmove: {
+      assert((ArgIdx == 0 || ArgIdx == 1) &&
+             "Invalid argument index for memory intrinsic");
+      if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getArgOperand(2)))
+        Loc.Size = LenCI->getZExtValue();
+      assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+             "Memory intrinsic location pointer not argument?");
+      Mask = ArgIdx ? Ref : Mod;
+      break;
+    }
+    case Intrinsic::lifetime_start:
+    case Intrinsic::lifetime_end:
+    case Intrinsic::invariant_start: {
+      assert(ArgIdx == 1 && "Invalid argument index");
+      assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+             "Intrinsic location pointer not argument?");
+      Loc.Size = cast<ConstantInt>(II->getArgOperand(0))->getZExtValue();
+      break;
+    }
+    case Intrinsic::invariant_end: {
+      assert(ArgIdx == 2 && "Invalid argument index");
+      assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+             "Intrinsic location pointer not argument?");
+      Loc.Size = cast<ConstantInt>(II->getArgOperand(1))->getZExtValue();
+      break;
+    }
+    case Intrinsic::arm_neon_vld1: {
+      assert(ArgIdx == 0 && "Invalid argument index");
+      assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+             "Intrinsic location pointer not argument?");
+      // LLVM's vld1 and vst1 intrinsics currently only support a single
+      // vector register.
+      if (DL)
+        Loc.Size = DL->getTypeStoreSize(II->getType());
+      break;
+    }
+    case Intrinsic::arm_neon_vst1: {
+      assert(ArgIdx == 0 && "Invalid argument index");
+      assert(Loc.Ptr == II->getArgOperand(ArgIdx) &&
+             "Intrinsic location pointer not argument?");
+      if (DL)
+        Loc.Size = DL->getTypeStoreSize(II->getArgOperand(1)->getType());
+      break;
+    }
+    }
+
+  // We can bound the aliasing properties of memset_pattern16 just as we can
+  // for memcpy/memset.  This is particularly important because the
+  // LoopIdiomRecognizer likes to turn loops into calls to memset_pattern16
+  // whenever possible.
+  else if (CS.getCalledFunction() &&
+           isMemsetPattern16(CS.getCalledFunction(), TLI)) {
+    assert((ArgIdx == 0 || ArgIdx == 1) &&
+           "Invalid argument index for memset_pattern16");
+    if (ArgIdx == 1)
+      Loc.Size = 16;
+    else if (const ConstantInt *LenCI =
+             dyn_cast<ConstantInt>(CS.getArgument(2)))
+      Loc.Size = LenCI->getZExtValue();
+    assert(Loc.Ptr == CS.getArgument(ArgIdx) &&
+           "memset_pattern16 location pointer not argument?");
+    Mask = ArgIdx ? Ref : Mod;
+  }
+  // FIXME: Handle memset_pattern4 and memset_pattern8 also.
+
+  return Loc;
+}
+
 /// getModRefInfo - Check to see if the specified callsite can clobber the
 /// specified memory object.  Since we only look at local properties of this
 /// function, we really can't say much about this query.  We do, however, use
 /// simple "address taken" analysis on local objects.
 AliasAnalysis::ModRefResult
 BasicAliasAnalysis::getModRefInfo(ImmutableCallSite CS,
                                   const Location &Loc) {
   assert(notDifferentParent(CS.getInstruction(), Loc.Ptr) &&
for (ImmutableCallSite::arg_iterator CI = CS.arg_begin(), CE = CS.arg_end();
         break;
       }
     }
 
     if (!PassedAsArg)
       return NoModRef;
   }
 
-  const TargetLibraryInfo &TLI = getAnalysis<TargetLibraryInfo>();
-  ModRefResult Min = ModRef;
-
-  // Finally, handle specific knowledge of intrinsics.
-  const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction());
-  if (II != nullptr)
-    switch (II->getIntrinsicID()) {
-    default: break;
-    case Intrinsic::memcpy:
-    case Intrinsic::memmove: {
-      uint64_t Len = UnknownSize;
-      if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getArgOperand(2)))
-        Len = LenCI->getZExtValue();
-      Value *Dest = II->getArgOperand(0);
-      Value *Src = II->getArgOperand(1);
-      // If it can't overlap the source dest, then it doesn't modref the loc.
-      if (isNoAlias(Location(Dest, Len), Loc)) {
-        if (isNoAlias(Location(Src, Len), Loc))
-          return NoModRef;
-        // If it can't overlap the dest, then worst case it reads the loc.
-        Min = Ref;
-      } else if (isNoAlias(Location(Src, Len), Loc)) {
-        // If it can't overlap the source, then worst case it mutates the loc.
-        Min = Mod;
-      }
-      break;
-    }
-    case Intrinsic::memset:
-      // Since memset is 'accesses arguments' only, the AliasAnalysis base class
-      // will handle it for the variable length case.
-      if (ConstantInt *LenCI = dyn_cast<ConstantInt>(II->getArgOperand(2))) {
-        uint64_t Len = LenCI->getZExtValue();
-        Value *Dest = II->getArgOperand(0);
-        if (isNoAlias(Location(Dest, Len), Loc))
-          return NoModRef;
-      }
-      // We know that memset doesn't load anything.
-      Min = Mod;
-      break;
-    case Intrinsic::lifetime_start:
-    case Intrinsic::lifetime_end:
-    case Intrinsic::invariant_start: {
-      uint64_t PtrSize =
-        cast<ConstantInt>(II->getArgOperand(0))->getZExtValue();
-      if (isNoAlias(Location(II->getArgOperand(1),
-                             PtrSize,
-                             II->getMetadata(LLVMContext::MD_tbaa)),
-                    Loc))
-        return NoModRef;
-      break;
-    }
-    case Intrinsic::invariant_end: {
-      uint64_t PtrSize =
-        cast<ConstantInt>(II->getArgOperand(1))->getZExtValue();
-      if (isNoAlias(Location(II->getArgOperand(2),
-                             PtrSize,
-                             II->getMetadata(LLVMContext::MD_tbaa)),
-                    Loc))
-        return NoModRef;
-      break;
-    }
-    case Intrinsic::arm_neon_vld1: {
-      // LLVM's vld1 and vst1 intrinsics currently only support a single
-      // vector register.
-      uint64_t Size =
-        DL ? DL->getTypeStoreSize(II->getType()) : UnknownSize;
-      if (isNoAlias(Location(II->getArgOperand(0), Size,
-                             II->getMetadata(LLVMContext::MD_tbaa)),
-                    Loc))
-        return NoModRef;
-      break;
-    }
-    case Intrinsic::arm_neon_vst1: {
-      uint64_t Size =
-        DL ? DL->getTypeStoreSize(II->getArgOperand(1)->getType()) : UnknownSize;
-      if (isNoAlias(Location(II->getArgOperand(0), Size,
-                             II->getMetadata(LLVMContext::MD_tbaa)),
-                    Loc))
-        return NoModRef;
-      break;
-    }
-    }
-
-  // We can bound the aliasing properties of memset_pattern16 just as we can
-  // for memcpy/memset.  This is particularly important because the
-  // LoopIdiomRecognizer likes to turn loops into calls to memset_pattern16
-  // whenever possible.
-  else if (TLI.has(LibFunc::memset_pattern16) &&
-           CS.getCalledFunction() &&
-           CS.getCalledFunction()->getName() == "memset_pattern16") {
-    const Function *MS = CS.getCalledFunction();
-    FunctionType *MemsetType = MS->getFunctionType();
-    if (!MemsetType->isVarArg() && MemsetType->getNumParams() == 3 &&
-        isa<PointerType>(MemsetType->getParamType(0)) &&
-        isa<PointerType>(MemsetType->getParamType(1)) &&
-        isa<IntegerType>(MemsetType->getParamType(2))) {
-      uint64_t Len = UnknownSize;
-      if (const ConstantInt *LenCI = dyn_cast<ConstantInt>(CS.getArgument(2)))
-        Len = LenCI->getZExtValue();
-      const Value *Dest = CS.getArgument(0);
-      const Value *Src = CS.getArgument(1);
-      // If it can't overlap the source dest, then it doesn't modref the loc.
-      if (isNoAlias(Location(Dest, Len), Loc)) {
-        // Always reads 16 bytes of the source.
-        if (isNoAlias(Location(Src, 16), Loc))
-          return NoModRef;
-        // If it can't overlap the dest, then worst case it reads the loc.
-        Min = Ref;
-      // Always reads 16 bytes of the source.
-      } else if (isNoAlias(Location(Src, 16), Loc)) {
-        // If it can't overlap the source, then worst case it mutates the loc.
-        Min = Mod;
-      }
-    }
-  }
-
   // The AliasAnalysis base class has some smarts, lets use them.
-  return ModRefResult(AliasAnalysis::getModRefInfo(CS, Loc) & Min);
+  return AliasAnalysis::getModRefInfo(CS, Loc);
 }
 
 /// aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP instruction
 /// against another pointer.  We know that V1 is a GEP, but we don't know
 /// anything about V2.  UnderlyingV1 is GetUnderlyingObject(GEP1, DL),
 /// UnderlyingV2 is the same for V2.
 ///
 AliasAnalysis::AliasResult

lib/Analysis/NoAliasAnalysis.cpp

 // This file defines the default implementation of the Alias Analysis interface
 // that simply returns "I don't know" for all queries.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Analysis/Passes.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/IR/DataLayout.h"
+#include "llvm/IR/LLVMContext.h"
 #include "llvm/Pass.h"
 using namespace llvm;
 
 namespace {
   /// NoAA - This class implements the -no-aa pass, which always returns "I
   /// don't know" for alias queries.  NoAA is unlike other alias analysis
   /// implementations, in that it does not chain to a previous analysis.  As
   /// such it doesn't follow many of the rules that other alias analyses must.
struct NoAA : public ImmutablePass, public AliasAnalysis {
     }
     ModRefBehavior getModRefBehavior(const Function *F) override {
       return UnknownModRefBehavior;
     }
 
     bool pointsToConstantMemory(const Location &Loc, bool OrLocal) override {
       return false;
     }
+    Location getArgLocation(ImmutableCallSite CS, unsigned ArgIdx,
+                            ModRefResult &Mask) override {
+      Mask = ModRef;
+      return Location(CS.getArgument(ArgIdx), UnknownSize,
+                      CS.getInstruction()->getMetadata(LLVMContext::MD_tbaa));
+    }
+
     ModRefResult getModRefInfo(ImmutableCallSite CS,
                                const Location &Loc) override {
       return ModRef;
     }
     ModRefResult getModRefInfo(ImmutableCallSite CS1,
                                ImmutableCallSite CS2) override {
       return ModRef;
     }

test/Analysis/BasicAA/cs-cs.ll

+; RUN: opt < %s -basicaa -aa-eval -print-all-alias-modref-info -disable-output 2>&1 | FileCheck %s
+target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:32:64-v128:32:128-a0:0:32-n32"
+target triple = "arm-apple-ios"
+
+declare <8 x i16> @llvm.arm.neon.vld1.v8i16(i8*, i32) nounwind readonly
+declare void @llvm.arm.neon.vst1.v8i16(i8*, <8 x i16>, i32) nounwind
+
+declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1) nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
+
+define <8 x i16> @test1(i8* %p, <8 x i16> %y) {
+entry:
+  %q = getelementptr i8* %p, i64 16
+  %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) nounwind
+  call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16)
+  %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) nounwind
+  %c = add <8 x i16> %a, %b
+  ret <8 x i16> %c
+
+; CHECK-LABEL: Function: test1:
+
+; CHECK: NoAlias:      i8* %p, i8* %q
+; CHECK: Just Ref:  Ptr: i8* %p        <->  %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:  Ptr: i8* %q        <->  %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:  Ptr: i8* %p        <->  call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16)
+; CHECK: Both ModRef:  Ptr: i8* %q     <->  call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16)
+; CHECK: Just Ref:  Ptr: i8* %p        <->  %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:  Ptr: i8* %q        <->  %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:   %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1 <->   call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16)
+; CHECK: NoModRef:   %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1 <->   %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:   call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16) <->   %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:   call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16) <->   %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:   %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1 <->   %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1
+; CHECK: NoModRef:   %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) #1 <->   call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16)
+}
+
+define void @test2(i8* %P, i8* %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test2:
+
+; CHECK:   MayAlias:     i8* %P, i8* %Q
+; CHECK:   Both ModRef:  Ptr: i8* %P     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK:   Both ModRef:  Ptr: i8* %Q     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK:   Both ModRef:  Ptr: i8* %P     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK:   Both ModRef:  Ptr: i8* %Q     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK:   Both ModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK:   Both ModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+}
+
+define void @test2a(i8* noalias %P, i8* noalias %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test2a:
+
+; CHECK: NoAlias:      i8* %P, i8* %Q
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+}
+
+define void @test2b(i8* noalias %P, i8* noalias %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  %R = getelementptr i8* %P, i64 12
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test2b:
+
+; CHECK: NoAlias:      i8* %P, i8* %Q
+; CHECK: NoAlias:      i8* %P, i8* %R
+; CHECK: NoAlias:      i8* %Q, i8* %R
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+}
+
+define void @test2c(i8* noalias %P, i8* noalias %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  %R = getelementptr i8* %P, i64 11
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test2c:
+
+; CHECK: NoAlias:      i8* %P, i8* %Q
+; CHECK: NoAlias:      i8* %P, i8* %R
+; CHECK: NoAlias:      i8* %Q, i8* %R
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+}
+
+define void @test2d(i8* noalias %P, i8* noalias %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  %R = getelementptr i8* %P, i64 -12
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test2d:
+
+; CHECK: NoAlias:      i8* %P, i8* %Q
+; CHECK: NoAlias:      i8* %P, i8* %R
+; CHECK: NoAlias:      i8* %Q, i8* %R
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+}
+
+define void @test2e(i8* noalias %P, i8* noalias %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  %R = getelementptr i8* %P, i64 -11
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test2e:
+
+; CHECK: NoAlias:      i8* %P, i8* %Q
+; CHECK: NoAlias:      i8* %P, i8* %R
+; CHECK: NoAlias:      i8* %Q, i8* %R
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: NoModRef:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %R        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %R, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+}
+
+define void @test3(i8* %P, i8* %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test3:
+
+; CHECK: MayAlias:     i8* %P, i8* %Q
+; CHECK: Both ModRef:  Ptr: i8* %P     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %Q     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %P     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %Q     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+}
+
+define void @test3a(i8* noalias %P, i8* noalias %Q) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test3a:
+
+; CHECK: NoAlias:      i8* %P, i8* %Q
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 8, i32 1, i1 false)
+}
+
+define void @test4(i8* %P, i8* noalias %Q) nounwind ssp {
+  tail call void @llvm.memset.p0i8.i64(i8* %P, i8 42, i64 8, i32 1, i1 false)
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test4:
+
+; CHECK: NoAlias:      i8* %P, i8* %Q
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memset.p0i8.i64(i8* %P, i8 42, i64 8, i32 1, i1 false)
+; CHECK: NoModRef:  Ptr: i8* %Q        <->  tail call void @llvm.memset.p0i8.i64(i8* %P, i8 42, i64 8, i32 1, i1 false)
+; CHECK: Just Mod:  Ptr: i8* %P        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Ref:  Ptr: i8* %Q        <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memset.p0i8.i64(i8* %P, i8 42, i64 8, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Just Mod:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memset.p0i8.i64(i8* %P, i8 42, i64 8, i32 1, i1 false)
+}
+
+define void @test5(i8* %P, i8* %Q, i8* %R) nounwind ssp {
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %R, i64 12, i32 1, i1 false)
+  ret void
+
+; CHECK-LABEL: Function: test5:
+
+; CHECK: MayAlias:     i8* %P, i8* %Q
+; CHECK: MayAlias:     i8* %P, i8* %R
+; CHECK: MayAlias:     i8* %Q, i8* %R
+; CHECK: Both ModRef:  Ptr: i8* %P     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %Q     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %R     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %P     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %R, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %Q     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %R, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:  Ptr: i8* %R     <->  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %R, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %R, i64 12, i32 1, i1 false)
+; CHECK: Both ModRef:   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %R, i64 12, i32 1, i1 false) <->   tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %Q, i64 12, i32 1, i1 false)
+}
+
+attributes #0 = { nounwind }