SLP Vectorizer: Extract cost will only be added once even if the scalar has multiple...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195406 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/SLPVectorizer.cpp b/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 2b498a88f523dc8a04f03bf6587ed72451c7319d..91ccac7c2e61be7dc3b08f4713b5fd6b7c5321be 100644 (file)
--- a/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -1082,9 +1082,13 @@ int BoUpSLP::getTreeCost() {
     Cost += C;
   }
 
+  SmallSet<Value *, 16> ExtractCostCalculated; 
   int ExtractCost = 0;
   for (UserList::iterator I = ExternalUses.begin(), E = ExternalUses.end();
        I != E; ++I) {
+    // We only add extract cost once for the same scalar. 
+    if (!ExtractCostCalculated.insert(I->Scalar))
+      continue; 
 
     VectorType *VecTy = VectorType::get(I->Scalar->getType(), BundleWidth);
     ExtractCost += TTI->getVectorInstrCost(Instruction::ExtractElement, VecTy,

diff --git a/test/Transforms/SLPVectorizer/X86/extractcost.ll b/test/Transforms/SLPVectorizer/X86/extractcost.ll
new file mode 100644 (file)
index 0000000..01baf66
--- /dev/null
+++ b/test/Transforms/SLPVectorizer/X86/extractcost.ll
@@ -0,0 +1,30 @@
+; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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-LABEL: @foo(
+;CHECK: store <4 x i32>
+;CHECK: ret
+define i32 @foo(i32* nocapture %A, i32 %n, i32 %m) {
+entry:
+  %mul = mul nsw i32 %n, 5
+  %add = add nsw i32 %mul, 9
+  store i32 %add, i32* %A, align 4
+  %mul1 = mul nsw i32 %n, 9
+  %add2 = add nsw i32 %mul1, 9
+  %arrayidx3 = getelementptr inbounds i32* %A, i64 1
+  store i32 %add2, i32* %arrayidx3, align 4
+  %mul4 = shl i32 %n, 3
+  %add5 = add nsw i32 %mul4, 9
+  %arrayidx6 = getelementptr inbounds i32* %A, i64 2
+  store i32 %add5, i32* %arrayidx6, align 4
+  %mul7 = mul nsw i32 %n, 10
+  %add8 = add nsw i32 %mul7, 9
+  %arrayidx9 = getelementptr inbounds i32* %A, i64 3
+  store i32 %add8, i32* %arrayidx9, align 4
+  %externaluse1 = add nsw i32 %add, %m  
+  %externaluse2 = mul nsw i32 %add, %m  ; we should add the extract cost only once and the store will be vectorized
+  %add10 = add nsw i32 %externaluse1, %externaluse2
+  ret i32 %add10
+}