From 975b1ddf60387139357c8cbbaeb613de5a39294f Mon Sep 17 00:00:00 2001
From: Nadav Rotem <nrotem@apple.com>
Date: Thu, 25 Apr 2013 19:55:03 +0000
Subject: LoopVectorizer: No need to generate pointer disambiguation checks
 between readonly pointers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@180570 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Transforms/Vectorize/LoopVectorize.cpp         | 16 +++++++---
 .../LoopVectorize/runtime-check-readonly.ll        | 36 ++++++++++++++++++++++
 2 files changed, 48 insertions(+), 4 deletions(-)
 create mode 100644 test/Transforms/LoopVectorize/runtime-check-readonly.ll

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 8372bd07239..0faab7a3c0e 100644
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -419,7 +419,7 @@ public:
     }
 
     /// Insert a pointer and calculate the start and end SCEVs.
-    void insert(ScalarEvolution *SE, Loop *Lp, Value *Ptr);
+    void insert(ScalarEvolution *SE, Loop *Lp, Value *Ptr, bool WritePtr);
 
     /// This flag indicates if we need to add the runtime check.
     bool Need;
@@ -429,6 +429,8 @@ public:
     SmallVector<const SCEV*, 2> Starts;
     /// Holds the pointer value at the end of the loop.
     SmallVector<const SCEV*, 2> Ends;
+    /// Holds the information if this pointer is used for writing to memory.
+    SmallVector<bool, 2> IsWritePtr;
   };
 
   /// A POD for saving information about induction variables.
@@ -787,7 +789,8 @@ struct LoopVectorize : public LoopPass {
 
 void
 LoopVectorizationLegality::RuntimePointerCheck::insert(ScalarEvolution *SE,
-                                                       Loop *Lp, Value *Ptr) {
+                                                       Loop *Lp, Value *Ptr,
+                                                       bool WritePtr) {
   const SCEV *Sc = SE->getSCEV(Ptr);
   const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Sc);
   assert(AR && "Invalid addrec expression");
@@ -796,6 +799,7 @@ LoopVectorizationLegality::RuntimePointerCheck::insert(ScalarEvolution *SE,
   Pointers.push_back(Ptr);
   Starts.push_back(AR->getStart());
   Ends.push_back(ScEnd);
+  IsWritePtr.push_back(WritePtr);
 }
 
 Value *InnerLoopVectorizer::getBroadcastInstrs(Value *V) {
@@ -1166,6 +1170,10 @@ InnerLoopVectorizer::addRuntimeCheck(LoopVectorizationLegality *Legal,
 
   for (unsigned i = 0; i < NumPointers; ++i) {
     for (unsigned j = i+1; j < NumPointers; ++j) {
+      // No need to check if two readonly pointers intersect.
+      if (!PtrRtCheck->IsWritePtr[i] && !PtrRtCheck->IsWritePtr[j])
+        continue;
+
       Value *Start0 = ChkBuilder.CreateBitCast(Starts[i], PtrArithTy, "bc");
       Value *Start1 = ChkBuilder.CreateBitCast(Starts[j], PtrArithTy, "bc");
       Value *End0 =   ChkBuilder.CreateBitCast(Ends[i],   PtrArithTy, "bc");
@@ -2678,7 +2686,7 @@ bool LoopVectorizationLegality::canVectorizeMemory() {
   for (MI = ReadWrites.begin(), ME = ReadWrites.end(); MI != ME; ++MI) {
     Value *V = (*MI).first;
     if (hasComputableBounds(V)) {
-      PtrRtCheck.insert(SE, TheLoop, V);
+      PtrRtCheck.insert(SE, TheLoop, V, true);
       DEBUG(dbgs() << "LV: Found a runtime check ptr:" << *V <<"\n");
     } else {
       CanDoRT = false;
@@ -2688,7 +2696,7 @@ bool LoopVectorizationLegality::canVectorizeMemory() {
   for (MI = Reads.begin(), ME = Reads.end(); MI != ME; ++MI) {
     Value *V = (*MI).first;
     if (hasComputableBounds(V)) {
-      PtrRtCheck.insert(SE, TheLoop, V);
+      PtrRtCheck.insert(SE, TheLoop, V, false);
       DEBUG(dbgs() << "LV: Found a runtime check ptr:" << *V <<"\n");
     } else {
       CanDoRT = false;
diff --git a/test/Transforms/LoopVectorize/runtime-check-readonly.ll b/test/Transforms/LoopVectorize/runtime-check-readonly.ll
new file mode 100644
index 00000000000..4145d134fd7
--- /dev/null
+++ b/test/Transforms/LoopVectorize/runtime-check-readonly.ll
@@ -0,0 +1,36 @@
+; RUN: opt < %s  -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s
+
+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-apple-macosx10.8.0"
+
+;CHECK: add_ints
+;CHECK: br
+;CHECK: getelementptr
+;CHECK-NEXT: getelementptr
+;CHECK-NEXT: icmp uge
+;CHECK-NEXT: icmp uge
+;CHECK-NEXT: icmp uge
+;CHECK-NEXT: icmp uge
+;CHECK-NEXT: and
+;CHECK: ret
+define void @add_ints(i32* nocapture %A, i32* nocapture %B, i32* nocapture %C) {
+entry:
+  br label %for.body
+
+for.body:
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds i32* %B, i64 %indvars.iv
+  %0 = load i32* %arrayidx, align 4
+  %arrayidx2 = getelementptr inbounds i32* %C, i64 %indvars.iv
+  %1 = load i32* %arrayidx2, align 4
+  %add = add nsw i32 %1, %0
+  %arrayidx4 = getelementptr inbounds i32* %A, i64 %indvars.iv
+  store i32 %add, i32* %arrayidx4, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 200
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret void
+}
-- 
cgit v1.2.3