[llvm] r243666 - [SLP vectorizer]: Choose the best consecutive candidate to pair with a store instruction.

[Wei Mi]

Author: wmi
Date: Thu Jul 30 12:40:39 2015
New Revision: 243666

URL: http://llvm.org/viewvc/llvm-project?rev=243666&view=rev
Log:
[SLP vectorizer]: Choose the best consecutive candidate to pair with a store instruction.

The patch changes the SLPVectorizer::vectorizeStores to choose the immediate
succeeding or preceding candidate for a store instruction when it has multiple
consecutive candidates. In this way it has better chance to find more slp
vectorization opportunities.

Differential Revision: http://reviews.llvm.org/D10445

Added:
    llvm/trunk/test/Transforms/SLPVectorizer/X86/pr23510.ll
Modified:
    llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp

Modified: llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp?rev=243666&r1=243665&r2=243666&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp (original)
+++ llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp Thu Jul 30 12:40:39 2015
@@ -3262,15 +3262,26 @@ bool SLPVectorizer::vectorizeStores(Arra
 
   // Do a quadratic search on all of the given stores and find
   // all of the pairs of stores that follow each other.
+  SmallVector<unsigned, 16> IndexQueue;
   for (unsigned i = 0, e = Stores.size(); i < e; ++i) {
-    for (unsigned j = 0; j < e; ++j) {
-      if (i == j)
-        continue;
-      const DataLayout &DL = Stores[i]->getModule()->getDataLayout();
-      if (R.isConsecutiveAccess(Stores[i], Stores[j], DL)) {
-        Tails.insert(Stores[j]);
+    const DataLayout &DL = Stores[i]->getModule()->getDataLayout();
+    IndexQueue.clear();
+    // If a store has multiple consecutive store candidates, search Stores
+    // array according to the sequence: from i+1 to e, then from i-1 to 0.
+    // This is because usually pairing with immediate succeeding or preceding
+    // candidate create the best chance to find slp vectorization opportunity.
+    unsigned j = 0;
+    for (j = i + 1; j < e; ++j)
+      IndexQueue.push_back(j);
+    for (j = i; j > 0; --j)
+      IndexQueue.push_back(j - 1);
+
+    for (auto &k : IndexQueue) {
+      if (R.isConsecutiveAccess(Stores[i], Stores[k], DL)) {
+        Tails.insert(Stores[k]);
         Heads.insert(Stores[i]);
-        ConsecutiveChain[Stores[i]] = Stores[j];
+        ConsecutiveChain[Stores[i]] = Stores[k];
+        break;
       }
     }
   }

Added: llvm/trunk/test/Transforms/SLPVectorizer/X86/pr23510.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SLPVectorizer/X86/pr23510.ll?rev=243666&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/SLPVectorizer/X86/pr23510.ll (added)
+++ llvm/trunk/test/Transforms/SLPVectorizer/X86/pr23510.ll Thu Jul 30 12:40:39 2015
@@ -0,0 +1,38 @@
+; PR23510
+; RUN: opt < %s -basicaa -slp-vectorizer -S | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; CHECK-LABEL: @_Z3fooPml(
+; CHECK: lshr <2 x i64>
+; CHECK: lshr <2 x i64>
+
+ at total = global i64 0, align 8
+
+define void @_Z3fooPml(i64* nocapture %a, i64 %i) {
+entry:
+  %tmp = load i64, i64* %a, align 8
+  %shr = lshr i64 %tmp, 4
+  store i64 %shr, i64* %a, align 8
+  %arrayidx1 = getelementptr inbounds i64, i64* %a, i64 1
+  %tmp1 = load i64, i64* %arrayidx1, align 8
+  %shr2 = lshr i64 %tmp1, 4
+  store i64 %shr2, i64* %arrayidx1, align 8
+  %arrayidx3 = getelementptr inbounds i64, i64* %a, i64 %i
+  %tmp2 = load i64, i64* %arrayidx3, align 8
+  %tmp3 = load i64, i64* @total, align 8
+  %add = add i64 %tmp3, %tmp2
+  store i64 %add, i64* @total, align 8
+  %tmp4 = load i64, i64* %a, align 8
+  %shr5 = lshr i64 %tmp4, 4
+  store i64 %shr5, i64* %a, align 8
+  %tmp5 = load i64, i64* %arrayidx1, align 8
+  %shr7 = lshr i64 %tmp5, 4
+  store i64 %shr7, i64* %arrayidx1, align 8
+  %tmp6 = load i64, i64* %arrayidx3, align 8
+  %tmp7 = load i64, i64* @total, align 8
+  %add9 = add i64 %tmp7, %tmp6
+  store i64 %add9, i64* @total, align 8
+  ret void
+}