When analyzing functions known to only access argument pointees,

only check arguments with pointer types. Update the documentation
of IntrReadArgMem reflect this.

While here, add support for TBAA tags on intrinsic calls.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130317 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 79 insertions, 10 deletions

diff --git a/include/llvm/Intrinsics.td b/include/llvm/Intrinsics.td
index d92ab4c..a63cd6a 100644
--- a/include/llvm/Intrinsics.td
+++ b/include/llvm/Intrinsics.td
@@ -30,7 +30,7 @@ class IntrinsicProperty;
 def IntrNoMem : IntrinsicProperty;
 
 // IntrReadArgMem - This intrinsic reads only from memory that one of its
-// arguments points to, but may read an unspecified amount.
+// pointer-typed arguments points to, but may read an unspecified amount.
 def IntrReadArgMem : IntrinsicProperty;
 
 // IntrReadMem - This intrinsic reads from unspecified memory, so it cannot be
diff --git a/lib/Analysis/AliasAnalysis.cpp b/lib/Analysis/AliasAnalysis.cpp
index be02ddb..c189a00 100644
--- a/lib/Analysis/AliasAnalysis.cpp
+++ b/lib/Analysis/AliasAnalysis.cpp
@@ -86,14 +86,20 @@ AliasAnalysis::getModRefInfo(ImmutableCallSite CS,
 
   if (onlyAccessesArgPointees(MRB)) {
     bool doesAlias = false;
-    if (doesAccessArgPointees(MRB))
+    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)
-        if (!isNoAlias(Location(*AI), Loc)) {
+           AI != AE; ++AI) {
+        const Value *Arg = *AI;
+        if (!Arg->getType()->isPointerTy())
+          continue;
+        Location CSLoc(Arg, UnknownSize, CSTag);
+        if (!isNoAlias(CSLoc, Loc)) {
           doesAlias = true;
           break;
         }
-
+      }
+    }
     if (!doesAlias)
       return NoModRef;
   }
@@ -138,13 +144,19 @@ AliasAnalysis::getModRefInfo(ImmutableCallSite CS1, ImmutableCallSite CS2) {
   // CS2's arguments.
   if (onlyAccessesArgPointees(CS2B)) {
     AliasAnalysis::ModRefResult R = NoModRef;
-    if (doesAccessArgPointees(CS2B))
+    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) {
-        R = ModRefResult((R | getModRefInfo(CS1, *I, UnknownSize)) & Mask);
+        const Value *Arg = *I;
+        if (!Arg->getType()->isPointerTy())
+          continue;
+        Location CS2Loc(Arg, UnknownSize, CS2Tag);
+        R = ModRefResult((R | getModRefInfo(CS1, CS2Loc)) & Mask);
         if (R == Mask)
           break;
       }
+    }
     return R;
   }
 
@@ -152,13 +164,20 @@ AliasAnalysis::getModRefInfo(ImmutableCallSite CS1, ImmutableCallSite CS2) {
   // any of the memory referenced by CS1's arguments. If not, return NoModRef.
   if (onlyAccessesArgPointees(CS1B)) {
     AliasAnalysis::ModRefResult R = NoModRef;
-    if (doesAccessArgPointees(CS1B))
+    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)
-        if (getModRefInfo(CS2, *I, UnknownSize) != NoModRef) {
+           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) {
           R = Mask;
           break;
         }
+      }
+    }
     if (R == NoModRef)
       return R;
   }
diff --git a/test/Analysis/BasicAA/intrinsics.ll b/test/Analysis/BasicAA/intrinsics.ll
new file mode 100644
index 0000000..bec878a
--- /dev/null
+++ b/test/Analysis/BasicAA/intrinsics.ll
@@ -0,0 +1,23 @@
+; RUN: opt -basicaa -gvn -S < %s | 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"
+
+; BasicAA should prove that these calls don't interfere, since they are
+; IntrArgReadMem and have noalias pointers.
+
+; CHECK:      define <8 x i16> @test0(i8* noalias %p, i8* noalias %q, <8 x i16> %y) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) nounwind
+; CHECK-NEXT:   call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16)
+; CHECK-NEXT:   %c = add <8 x i16> %a, %a
+define <8 x i16> @test0(i8* noalias %p, i8* noalias %q, <8 x i16> %y) {
+entry:
+  %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
+}
+
+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
diff --git a/test/Analysis/TypeBasedAliasAnalysis/intrinsics.ll b/test/Analysis/TypeBasedAliasAnalysis/intrinsics.ll
new file mode 100644
index 0000000..8f080e2
--- /dev/null
+++ b/test/Analysis/TypeBasedAliasAnalysis/intrinsics.ll
@@ -0,0 +1,27 @@
+; RUN: opt -tbaa -basicaa -gvn -S < %s | 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"
+
+; TBAA should prove that these calls don't interfere, since they are
+; IntrArgReadMem and have TBAA metadata.
+
+; CHECK:      define <8 x i16> @test0(i8* %p, i8* %q, <8 x i16> %y) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) nounwind
+; CHECK-NEXT:   call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16)
+; CHECK-NEXT:   %c = add <8 x i16> %a, %a
+define <8 x i16> @test0(i8* %p, i8* %q, <8 x i16> %y) {
+entry:
+  %a = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) nounwind, !tbaa !2
+  call void @llvm.arm.neon.vst1.v8i16(i8* %q, <8 x i16> %y, i32 16), !tbaa !1
+  %b = call <8 x i16> @llvm.arm.neon.vld1.v8i16(i8* %p, i32 16) nounwind, !tbaa !2
+  %c = add <8 x i16> %a, %b
+  ret <8 x i16> %c
+}
+
+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
+
+!0 = metadata !{metadata !"tbaa root", null}
+!1 = metadata !{metadata !"A", metadata !0}
+!2 = metadata !{metadata !"B", metadata !0}