[LSR] canonicalize Prod*(1<<C) to Prod<<C

Summary:
Because LSR happens at a late stage where mul of a power of 2 is
typically canonicalized to shl, this canonicalization emits code that
can be better CSE'ed.

Test Plan:
Transforms/LoopStrengthReduce/shl.ll shows how this change makes GVN more
powerful. Fixes some existing tests due to this change.

Reviewers: sanjoy, majnemer, atrick

Reviewed By: majnemer, atrick

Subscribers: majnemer, llvm-commits

Differential Revision: http://reviews.llvm.org/D10448

llvm-svn: 240573

6 files changed, 56 insertions, 11 deletions

diff --git a/llvm/lib/Analysis/ScalarEvolutionExpander.cpp b/llvm/lib/Analysis/ScalarEvolutionExpander.cpp
index 8274374621c..fee2a2d0d18 100644
--- a/llvm/lib/Analysis/ScalarEvolutionExpander.cpp
+++ b/llvm/lib/Analysis/ScalarEvolutionExpander.cpp
@@ -24,10 +24,12 @@
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Module.h"
+#include "llvm/IR/PatternMatch.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 
 using namespace llvm;
+using namespace PatternMatch;
 
 /// ReuseOrCreateCast - Arrange for there to be a cast of V to Ty at IP,
 /// reusing an existing cast if a suitable one exists, moving an existing
@@ -751,25 +753,30 @@ Value *SCEVExpander::visitMulExpr(const SCEVMulExpr *S) {
   // out of loops.
   Value *Prod = nullptr;
   for (SmallVectorImpl<std::pair<const Loop *, const SCEV *> >::iterator
-       I = OpsAndLoops.begin(), E = OpsAndLoops.end(); I != E; ) {
+       I = OpsAndLoops.begin(), E = OpsAndLoops.end(); I != E; ++I) {
     const SCEV *Op = I->second;
     if (!Prod) {
       // This is the first operand. Just expand it.
       Prod = expand(Op);
-      ++I;
     } else if (Op->isAllOnesValue()) {
       // Instead of doing a multiply by negative one, just do a negate.
       Prod = InsertNoopCastOfTo(Prod, Ty);
       Prod = InsertBinop(Instruction::Sub, Constant::getNullValue(Ty), Prod);
-      ++I;
     } else {
       // A simple mul.
       Value *W = expandCodeFor(Op, Ty);
       Prod = InsertNoopCastOfTo(Prod, Ty);
       // Canonicalize a constant to the RHS.
       if (isa<Constant>(Prod)) std::swap(Prod, W);
-      Prod = InsertBinop(Instruction::Mul, Prod, W);
-      ++I;
+      const APInt *RHS;
+      if (match(W, m_Power2(RHS))) {
+        // Canonicalize Prod*(1<<C) to Prod<<C.
+        assert(!Ty->isVectorTy() && "vector types are not SCEVable");
+        Prod = InsertBinop(Instruction::Shl, Prod,
+                           ConstantInt::get(Ty, RHS->logBase2()));
+      } else {
+        Prod = InsertBinop(Instruction::Mul, Prod, W);
+      }
     }
   }
 
diff --git a/llvm/test/Transforms/LoopIdiom/basic.ll b/llvm/test/Transforms/LoopIdiom/basic.ll
index a8a2c8efa38..c633ae95d16 100644
--- a/llvm/test/Transforms/LoopIdiom/basic.ll
+++ b/llvm/test/Transforms/LoopIdiom/basic.ll
@@ -69,7 +69,7 @@ for.end:                                          ; preds = %for.body, %entry
   ret void
 ; CHECK-LABEL: @test2(
 ; CHECK: br i1 %cmp10,
-; CHECK: %0 = mul i64 %Size, 4
+; CHECK: %0 = shl i64 %Size, 2
 ; CHECK: call void @llvm.memset.p0i8.i64(i8* %Base1, i8 1, i64 %0, i32 4, i1 false)
 ; CHECK-NOT: store
 }
diff --git a/llvm/test/Transforms/LoopReroll/nonconst_lb.ll b/llvm/test/Transforms/LoopReroll/nonconst_lb.ll
index 5effa42d07c..96090e8dc0b 100644
--- a/llvm/test/Transforms/LoopReroll/nonconst_lb.ll
+++ b/llvm/test/Transforms/LoopReroll/nonconst_lb.ll
@@ -52,7 +52,7 @@ for.end:                                          ; preds = %for.body, %entry
 ; CHECK:   %0 = add i32 %n, -1
 ; CHECK:   %1 = sub i32 %0, %m
 ; CHECK:   %2 = lshr i32 %1, 2
-; CHECK:   %3 = mul i32 %2, 4
+; CHECK:   %3 = shl i32 %2, 2
 ; CHECK:   %4 = add i32 %m, %3
 ; CHECK:   %5 = add i32 %4, 3
 ; CHECK:   br label %for.body
@@ -132,7 +132,7 @@ for.end:                                          ; preds = %for.body, %entry
 ; CHECK:   %0 = add i32 %n, -1
 ; CHECK:   %1 = sub i32 %0, %rem
 ; CHECK:   %2 = lshr i32 %1, 2
-; CHECK:   %3 = mul i32 %2, 4
+; CHECK:   %3 = shl i32 %2, 2
 ; CHECK:   %4 = add i32 %rem, %3
 ; CHECK:   %5 = add i32 %4, 3
 ; CHECK:   br label %for.body
diff --git a/llvm/test/Transforms/LoopStrengthReduce/2011-10-06-ReusePhi.ll b/llvm/test/Transforms/LoopStrengthReduce/2011-10-06-ReusePhi.ll
index cc8226e1258..5923a42fa65 100644
--- a/llvm/test/Transforms/LoopStrengthReduce/2011-10-06-ReusePhi.ll
+++ b/llvm/test/Transforms/LoopStrengthReduce/2011-10-06-ReusePhi.ll
@@ -12,8 +12,8 @@ target datalayout = "n8:16:32:64"
 ; CHECK-LABEL: @test(
 ; multiplies are hoisted out of the loop
 ; CHECK: while.body.lr.ph:
-; CHECK: mul i64
-; CHECK: mul i64
+; CHECK: shl i64
+; CHECK: shl i64
 ; GEPs are ugly
 ; CHECK: while.body:
 ; CHECK: phi
diff --git a/llvm/test/Transforms/LoopStrengthReduce/post-inc-icmpzero.ll b/llvm/test/Transforms/LoopStrengthReduce/post-inc-icmpzero.ll
index 092b274bfc0..466566ed8a0 100644
--- a/llvm/test/Transforms/LoopStrengthReduce/post-inc-icmpzero.ll
+++ b/llvm/test/Transforms/LoopStrengthReduce/post-inc-icmpzero.ll
@@ -6,7 +6,7 @@
 
 ; CHECK:   [[r1:%[a-z0-9]+]] = sub i64 %sub.ptr.lhs.cast, %sub.ptr.rhs.cast
 ; CHECK:   [[r2:%[a-z0-9]+]] = lshr i64 [[r1]], 1
-; CHECK:   [[r3:%[a-z0-9]+]] = mul i64 [[r2]], 2
+; CHECK:   [[r3:%[a-z0-9]+]] = shl i64 [[r2]], 1
 ; CHECK:   br label %for.body
 ; CHECK: for.body:
 ; CHECK:   %lsr.iv2 = phi i64 [ %lsr.iv.next, %for.body ], [ [[r3]], %for.body.lr.ph ]
diff --git a/llvm/test/Transforms/LoopStrengthReduce/shl.ll b/llvm/test/Transforms/LoopStrengthReduce/shl.ll
new file mode 100644
index 00000000000..bb9cb39f437
--- /dev/null
+++ b/llvm/test/Transforms/LoopStrengthReduce/shl.ll
@@ -0,0 +1,38 @@
+; RUN: opt < %s -loop-reduce -gvn -S | FileCheck %s
+
+target datalayout = "e-i64:64-v16:16-v32:32-n16:32:64"
+
+define void @_Z3fooPfll(float* nocapture readonly %input, i64 %n, i64 %s) {
+; CHECK-LABEL: @_Z3fooPfll(
+entry:
+  %mul = shl nsw i64 %s, 2
+; CHECK: %mul = shl i64 %s, 2
+  tail call void @_Z3bazl(i64 %mul) #2
+; CHECK-NEXT: call void @_Z3bazl(i64 %mul)
+  %cmp.5 = icmp sgt i64 %n, 0
+  br i1 %cmp.5, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:                               ; preds = %entry
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %for.body
+  br label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret void
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %i.06 = phi i64 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds float, float* %input, i64 %i.06
+; LoopStrengthReduce should reuse %mul as the stride.
+; CHECK: getelementptr i1, i1* {{[^,]+}}, i64 %mul
+  %0 = load float, float* %arrayidx, align 4
+  tail call void @_Z3barf(float %0) #2
+  %add = add nsw i64 %i.06, %s
+  %cmp = icmp slt i64 %add, %n
+  br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit
+}
+
+declare void @_Z3bazl(i64)
+
+declare void @_Z3barf(float)