[Clang][BasicAA][AIE2] Enable full PHI AA for AIE

Also increase the max limit of the search depth in DecomposeGEPExpression.
With this change we can allow LICM to hoist more instructions.

clang/lib/Driver/ToolChains/AIE.cpp

@@ -113,6 +113,12 @@ void AIEToolChain::addClangTargetOptions(
   // Make sure to perform most optimizations before mandatory inlinings,
   // otherwise noalias attributes can get lost and hurt AA results.
   CC1Args.append({"-mllvm", "-mandatory-inlining-before-opt=false"});
+
+  // Perform complete AA analysis on phi nodes.
+  CC1Args.append({"-mllvm", "-basic-aa-full-phi-analysis=true"});
+
+  // Extend the max limit of the search depth in BasicAA
+  CC1Args.append({"-mllvm", "-basic-aa-max-lookup-search-depth=10"});
 }
 
 // Avoid using newer dwarf versions, as the simulator doesn't understand newer

llvm/lib/Analysis/BasicAliasAnalysis.cpp

@@ -68,20 +68,26 @@ using namespace llvm;
 static cl::opt<bool> EnableRecPhiAnalysis("basic-aa-recphi", cl::Hidden,
                                           cl::init(true));
 
+/// Enable full analysis of PHI nodes.
+static cl::opt<bool> EnableFullPHIAnalysis("basic-aa-full-phi-analysis",
+                                           cl::Hidden, cl::init(false));
+
 static cl::opt<bool> EnableSeparateStorageAnalysis("basic-aa-separate-storage",
                                                    cl::Hidden, cl::init(true));
 
+// The max limit of the search depth in DecomposeGEPExpression() and
+// getUnderlyingObject().
+static cl::opt<unsigned>
+    MaxLookupSearchDepth("basic-aa-max-lookup-search-depth", cl::Hidden,
+                         cl::init(6));
+
 /// SearchLimitReached / SearchTimes shows how often the limit of
 /// to decompose GEPs is reached. It will affect the precision
 /// of basic alias analysis.
 STATISTIC(SearchLimitReached, "Number of times the limit to "
                               "decompose GEPs is reached");
 STATISTIC(SearchTimes, "Number of times a GEP is decomposed");
 
-// The max limit of the search depth in DecomposeGEPExpression() and
-// getUnderlyingObject().
-static const unsigned MaxLookupSearchDepth = 6;
-
 bool BasicAAResult::invalidate(Function &Fn, const PreservedAnalyses &PA,
                                FunctionAnalysisManager::Invalidator &Inv) {
   // We don't care if this analysis itself is preserved, it has no state. But
@@ -1389,7 +1395,7 @@ AliasResult BasicAAResult::aliasPHI(const PHINode *PN, LocationSize PNSize,
     if (PV1 == PN)
       continue;
 
-    if (isa<PHINode>(PV1)) {
+    if (!EnableFullPHIAnalysis && isa<PHINode>(PV1)) {
       if (OnePhi && OnePhi != PV1) {
         // To control potential compile time explosion, we choose to be
         // conserviate when we have more than one Phi input.  It is important

llvm/test/Analysis/BasicAA/phi-aa.ll

@@ -1,4 +1,7 @@
 ; RUN: opt < %s -aa-pipeline=basic-aa -passes=aa-eval -print-all-alias-modref-info -disable-output 2>&1 | FileCheck %s
+; RUN: opt < %s -aa-pipeline=basic-aa -passes=aa-eval -print-all-alias-modref-info -basic-aa-full-phi-analysis \
+; RUN:       -disable-output 2>&1 | FileCheck %s --check-prefix=FULL-PHI
+
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
@@ -85,6 +88,9 @@ declare void @inc(ptr)
 ; CHECK: MayAlias: i32* %val1, i32* @Y
 ; CHECK: MayAlias: i32* %val2, i32* @Y
 ; CHECK: MayAlias: i32* %val3, i32* @Y
+; FULL-PHI: NoAlias: i32* %val1, i32* @Y
+; FULL-PHI: NoAlias: i32* %val2, i32* @Y
+; FULL-PHI: NoAlias: i32* %val3, i32* @Y
 define void @loop_phi_chain(i32 %a, i32 %b, i32 %c) {
 entry:
   br label %loop1

llvm/test/Analysis/BasicAA/phi-full-aa-diamond.ll

@@ -0,0 +1,62 @@
+; RUN: opt < %s -aa-pipeline=basic-aa -passes=aa-eval -print-all-alias-modref-info -disable-output 2>&1 | FileCheck %s
+; RUN: opt < %s -aa-pipeline=basic-aa -passes=aa-eval -print-all-alias-modref-info -basic-aa-full-phi-analysis \
+; RUN:       -disable-output 2>&1 | FileCheck %s --check-prefix=FULL-PHI
+
+
+; The goal of this test is to test the differences in the behavior of
+; the BasicAA with and without full phi analysis in a dual diamond scenario.
+; The basic shape of the test is the following (simplified without phis):
+; if (...) {
+;    if (...) {
+;      %P = getelementptr @X ...
+;    } else {
+;      %P = getelementptr @X ...
+;    }    
+; } else {
+;   %P = getelementptr @X ...
+; }
+;
+; // Use of %P and @Y 
+;
+
+@X = common global i32 0
+@Y = common global i32 0
+
+; CHECK: MayAlias: i32* %P, i32* @Y
+; FULL-PHI: NoAlias: i32* %P, i32* @Y
+define void @foo(i32 %cond) nounwind {
+entry:
+  %"alloca point" = bitcast i32 0 to i32
+  %tmp = icmp ne i32 %cond, 0
+  br i1 %tmp, label %bbtrue, label %bbfalse
+
+bbtrue:
+  %tmp1 = icmp ne i32 %cond, 2
+  br i1 %tmp, label %bbtruetrue, label %bbtruefalse
+
+bbtruetrue:
+  %p1 = getelementptr i32, ptr @X, i64 3
+  br label %bbtrueend
+
+bbtruefalse:
+  %p2 = getelementptr i32, ptr @X, i64 4
+  br label %bbtrueend
+
+bbtrueend:
+  %p3 = phi ptr [ %p1, %bbtruetrue ], [ %p2, %bbtruefalse ]
+  br label %bblast
+
+bbfalse:
+  %p4 = getelementptr i32, ptr @X, i64 2
+  br label %bblast
+
+bblast:
+  %P = phi ptr [ %p3, %bbtrueend ], [ %p4, %bbfalse ]
+  %tmp2 = load i32, ptr @Y, align 4
+  store i32 123, ptr %P, align 4
+  %tmp3 = load i32, ptr @Y, align 4
+  br label %return
+
+return:
+  ret void
+}