From a10b311aeca09090228493700352687e878de245 Mon Sep 17 00:00:00 2001
From: Nadav Rotem <nrotem@apple.com>
Date: Tue, 4 Dec 2012 18:17:33 +0000
Subject: [PATCH] Add support for reduction variables when IF-conversion is
 enabled.

llvm-svn: 169288
---
 llvm/lib/Transforms/Vectorize/LoopVectorize.cpp    | 33 ++++++++++-----
 .../test/Transforms/LoopVectorize/if-conversion.ll | 48 ++++++++++++++++++++++
 2 files changed, 71 insertions(+), 10 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 1de5b30..3502e9e 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1133,8 +1133,8 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
     // Reductions do not have to start at zero. They can start with
     // any loop invariant values.
     VecRdxPhi->addIncoming(VectorStart, VecPreheader);
-    unsigned SelfEdgeIdx = (RdxPhi)->getBasicBlockIndex(LoopScalarBody);
-    Value *Val = getVectorValue(RdxPhi->getIncomingValue(SelfEdgeIdx));
+    Value *Val =
+    getVectorValue(RdxPhi->getIncomingValueForBlock(OrigLoop->getLoopLatch()));
     VecRdxPhi->addIncoming(Val, LoopVectorBody);
 
     // Before each round, move the insertion point right between
@@ -1201,8 +1201,11 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
 
     // Fix the scalar loop reduction variable with the incoming reduction sum
     // from the vector body and from the backedge value.
-    int IncomingEdgeBlockIdx = (RdxPhi)->getBasicBlockIndex(LoopScalarBody);
-    int SelfEdgeBlockIdx = (IncomingEdgeBlockIdx ? 0 : 1); // The other block.
+    int IncomingEdgeBlockIdx =
+    (RdxPhi)->getBasicBlockIndex(OrigLoop->getLoopLatch());
+    assert(IncomingEdgeBlockIdx >= 0 && "Invalid block index");
+    // Pick the other block.
+    int SelfEdgeBlockIdx = (IncomingEdgeBlockIdx ? 0 : 1);
     (RdxPhi)->setIncomingValue(SelfEdgeBlockIdx, Scalar0);
     (RdxPhi)->setIncomingValue(IncomingEdgeBlockIdx, RdxDesc.LoopExitInstr);
   }// end of for each redux variable.
@@ -1961,11 +1964,13 @@ bool LoopVectorizationLegality::AddReductionVar(PHINode *Phi,
   if (Phi->getNumIncomingValues() != 2)
     return false;
 
-  // Find the possible incoming reduction variable.
-  BasicBlock *BB = Phi->getParent();
-  int SelfEdgeIdx = Phi->getBasicBlockIndex(BB);
-  int InEdgeBlockIdx = (SelfEdgeIdx ? 0 : 1); // The other entry.
-  Value *RdxStart = Phi->getIncomingValue(InEdgeBlockIdx);
+  // Reduction variables are only found in the loop header block.
+  if (Phi->getParent() != TheLoop->getHeader())
+    return false;
+
+  // Obtain the reduction start value from the value that comes from the loop
+  // preheader.
+  Value *RdxStart = Phi->getIncomingValueForBlock(TheLoop->getLoopPreheader());
 
   // ExitInstruction is the single value which is used outside the loop.
   // We only allow for a single reduction value to be used outside the loop.
@@ -2003,9 +2008,17 @@ bool LoopVectorizationLegality::AddReductionVar(PHINode *Phi,
         FoundStartPHI = true;
         continue;
       }
+
+      // We allow in-loop PHINodes which are not the original reduction PHI
+      // node. If this PHI is the only user of Iter (happens in IF w/ no ELSE
+      // structure) then don't skip this PHI.
+      if (isa<PHINode>(U) && U->getParent() != TheLoop->getHeader() &&
+          TheLoop->contains(U->getParent()) && Iter->getNumUses() > 1)
+        continue;
+
       // Check if we found the exit user.
       BasicBlock *Parent = U->getParent();
-      if (Parent != BB) {
+      if (!TheLoop->contains(Parent)) {
         // We must have a single exit instruction.
         if (ExitInstruction != 0)
           return false;
diff --git a/llvm/test/Transforms/LoopVectorize/if-conversion.ll b/llvm/test/Transforms/LoopVectorize/if-conversion.ll
index 35c549c..b4701b9 100644
--- a/llvm/test/Transforms/LoopVectorize/if-conversion.ll
+++ b/llvm/test/Transforms/LoopVectorize/if-conversion.ll
@@ -58,3 +58,51 @@ if.end:
 for.end:
   ret i32 undef
 }
+
+
+
+; int func(int *A, int n) {
+;   unsigned sum = 0;
+;   for (int i = 0; i < n; ++i)
+;     if (A[i] > 30)
+;       sum += A[i] + 2;
+;
+;   return sum;
+; }
+
+;CHECK: @reduction_func
+;CHECK: load <4 x i32>
+;CHECK: icmp sgt <4 x i32>
+;CHECK: add <4 x i32>
+;CHECK: select <4 x i1>
+;CHECK: ret i32
+define i32 @reduction_func(i32* nocapture %A, i32 %n) nounwind uwtable readonly ssp {
+entry:
+  %cmp10 = icmp sgt i32 %n, 0
+  br i1 %cmp10, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.inc
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ]
+  %sum.011 = phi i32 [ %sum.1, %for.inc ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32* %A, i64 %indvars.iv
+  %0 = load i32* %arrayidx, align 4
+  %cmp1 = icmp sgt i32 %0, 30
+  br i1 %cmp1, label %if.then, label %for.inc
+
+if.then:                                          ; preds = %for.body
+  %add = add i32 %sum.011, 2
+  %add4 = add i32 %add, %0
+  br label %for.inc
+
+for.inc:                                          ; preds = %for.body, %if.then
+  %sum.1 = phi i32 [ %add4, %if.then ], [ %sum.011, %for.body ]
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.inc, %entry
+  %sum.0.lcssa = phi i32 [ 0, %entry ], [ %sum.1, %for.inc ]
+  ret i32 %sum.0.lcssa
+}
+
-- 
2.7.4