llvm.noalias - GetStringLength should look through them too

[hfinkel]

Summary

This is part of the series started by D9375, and teaches GetStringLength to look through calls to llvm.noalias.

[reames]

Resigning as a reviewer to get a very stale review off my list of blocking tasks in phabricator. Please readd when desired.

[hfinkel]

This change should not be needed if the intrinsics are stripped by default.

lib/Analysis/ValueTracking.cpp

 /// This function computes the length of a null-terminated C string pointed to
 /// by V. If successful, it returns true and returns the string in Str.
 /// If unsuccessful, it returns false.
 bool llvm::getConstantStringInfo(const Value *V, StringRef &Str,
                                  uint64_t Offset, bool TrimAtNul) {
   assert(V);
 
   // Look through bitcast instructions and geps.
-  V = V->stripPointerCasts();
+  V = V->stripPointerCasts(/*LookThroughNoAlias*/true);
 
   // If the value is a GEP instruction or constant expression, treat it as an
   // offset.
   if (const GEPOperator *GEP = dyn_cast<GEPOperator>(V)) {
     // Make sure the GEP has exactly three arguments.
     if (GEP->getNumOperands() != 3)
       return false;
 

 // These next two are very similar to the above, but also look through PHI
 // nodes.
 // TODO: See if we can integrate these two together.
 
 /// If we can compute the length of the string pointed to by
 /// the specified pointer, return 'len+1'.  If we can't, return 0.
 static uint64_t GetStringLengthH(Value *V, SmallPtrSetImpl<PHINode*> &PHIs) {
   // Look through noop bitcast instructions.
-  V = V->stripPointerCasts();
+  V = V->stripPointerCasts(/*LookThroughNoAlias*/true);
 
   // If this is a PHI node, there are two cases: either we have already seen it
   // or we haven't.
   if (PHINode *PN = dyn_cast<PHINode>(V)) {
     if (!PHIs.insert(PN).second)
       return ~0ULL;  // already in the set.
 
     // If it was new, see if all the input strings are the same length.

test/Transforms/InstCombine/strlen-1.ll

 define i32 @test_simplify1() {
 ; CHECK-LABEL: @test_simplify1(
   %hello_p = getelementptr [6 x i8], [6 x i8]* @hello, i32 0, i32 0
   %hello_l = call i32 @strlen(i8* %hello_p)
   ret i32 %hello_l
 ; CHECK-NEXT: ret i32 5
 }
 
+define i32 @test_simplify1a() {
+; CHECK-LABEL: @test_simplify1a(
+  %hello_p = getelementptr [6 x i8], [6 x i8]* @hello, i32 0, i32 0
+  %hello_pa = call i8* @llvm.noalias.p0i8(i8* %hello_p, metadata !1)
+  %hello_l = call i32 @strlen(i8* %hello_pa)
+  ret i32 %hello_l
+; CHECK-NEXT: ret i32 5
+}
+
 define i32 @test_simplify2() {
 ; CHECK-LABEL: @test_simplify2(
   %null_p = getelementptr [1 x i8], [1 x i8]* @null, i32 0, i32 0
   %null_l = call i32 @strlen(i8* %null_p)
   ret i32 %null_l
 ; CHECK-NEXT: ret i32 0
 }
 

 define i32 @test_no_simplify1() {
 ; CHECK-LABEL: @test_no_simplify1(
   %a_p = getelementptr [32 x i8], [32 x i8]* @a, i32 0, i32 0
   %a_l = call i32 @strlen(i8* %a_p)
 ; CHECK-NEXT: %a_l = call i32 @strlen
   ret i32 %a_l
 ; CHECK-NEXT: ret i32 %a_l
 }
+
+declare i8* @llvm.noalias.p0i8(i8*, metadata) nounwind
+
+!0 = !{!0, !"some domain"}
+!1 = !{!1, !0, !"some scope"}
+