By default, use Early-CSE instead of GVN for vectorization cleanup.

As has been suggested by Duncan and others, Early-CSE and GVN should
do similar redundancy elimination, but Early-CSE is much less expensive.
Most of my autovectorization benchmarks show a performance regresion, but
all of these are < 0.1%, and so I think that it is still worth using
the less expensive pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@154673 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/IPO/PassManagerBuilder.cpp b/lib/Transforms/IPO/PassManagerBuilder.cpp
index a1b0a4580bf508ffa76e3fd7a6c34c47f4b51ed9..43b4ab5efa4df369781c02d2f9f08eeafd795861 100644 (file)
--- a/lib/Transforms/IPO/PassManagerBuilder.cpp
+++ b/lib/Transforms/IPO/PassManagerBuilder.cpp
@@ -35,6 +35,11 @@ using namespace llvm;
 static cl::opt<bool>
 RunVectorization("vectorize", cl::desc("Run vectorization passes"));
 
+static cl::opt<bool>
+UseGVNAfterVectorization("use-gvn-after-vectorization",
+  cl::init(false), cl::Hidden,
+  cl::desc("Run GVN instead of Early CSE after vectorization passes"));
+
 PassManagerBuilder::PassManagerBuilder() {
     OptLevel = 2;
     SizeLevel = 0;
@@ -182,8 +187,10 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) {
   if (Vectorize) {
     MPM.add(createBBVectorizePass());
     MPM.add(createInstructionCombiningPass());
-    if (OptLevel > 1)
-      MPM.add(createGVNPass());                 // Remove redundancies
+    if (OptLevel > 1 && UseGVNAfterVectorization)
+      MPM.add(createGVNPass());                   // Remove redundancies
+    else
+      MPM.add(createEarlyCSEPass());              // Catch trivial redundancies
   }
 
   MPM.add(createAggressiveDCEPass());         // Delete dead instructions