Loop Vectorize: optimize the vectorization of trunc(induction_var). The truncation is now done on scalars.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169904 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 46 insertions, 5 deletions

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index a69124032b..9550e8a04c 100644
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1204,8 +1204,20 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
     case Instruction::Trunc:
     case Instruction::FPTrunc:
     case Instruction::BitCast: {
-      /// Vectorize bitcasts.
       CastInst *CI = dyn_cast<CastInst>(it);
+      /// Optimize the special case where the source is the induction
+      /// variable. Notice that we can only optimize the 'trunc' case
+      /// because: a. FP conversions lose precision, b. sext/zext may wrap,
+      /// c. other casts depend on pointer size.
+      if (CI->getOperand(0) == OldInduction &&
+          it->getOpcode() == Instruction::Trunc) {
+        Value *ScalarCast = Builder.CreateCast(CI->getOpcode(), Induction,
+                                               CI->getType());
+        Value *Broadcasted = getBroadcastInstrs(ScalarCast);
+        WidenMap[it] = getConsecutiveVector(Broadcasted);
+        break;
+      }
+      /// Vectorize casts.
       Value *A = getVectorValue(it->getOperand(0));
       Type *DestTy = VectorType::get(CI->getType()->getScalarType(), VF);
       WidenMap[it] = Builder.CreateCast(CI->getOpcode(), A, DestTy);
diff --git a/test/Transforms/LoopVectorize/cast-induction.ll b/test/Transforms/LoopVectorize/cast-induction.ll
new file mode 100644
index 0000000000..fc8281ba8a
--- /dev/null
+++ b/test/Transforms/LoopVectorize/cast-induction.ll
@@ -0,0 +1,30 @@
+; RUN: opt < %s  -loop-vectorize -force-vector-width=4 -dce -instcombine -licm -S | FileCheck %s
+
+; rdar://problem/12848162
+
+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"
+
+@a = common global [2048 x i32] zeroinitializer, align 16
+
+;CHECK: @example12
+;CHECK: trunc i64
+;CHECK: store <4 x i32>
+;CHECK: ret void
+define void @example12() nounwind uwtable ssp {
+  br label %1
+
+; <label>:1                                       ; preds = %1, %0
+  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
+  %2 = getelementptr inbounds [2048 x i32]* @a, i64 0, i64 %indvars.iv
+  %3 = trunc i64 %indvars.iv to i32
+  store i32 %3, i32* %2, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 1024
+  br i1 %exitcond, label %4, label %1
+
+; <label>:4                                       ; preds = %1
+  ret void
+}
+
diff --git a/test/Transforms/LoopVectorize/cpp-new-array.ll b/test/Transforms/LoopVectorize/cpp-new-array.ll
index 26902eba9e..af59963899 100644
--- a/test/Transforms/LoopVectorize/cpp-new-array.ll
+++ b/test/Transforms/LoopVectorize/cpp-new-array.ll
@@ -4,7 +4,7 @@ target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f3
 target triple = "x86_64-apple-macosx10.8.0"
 
 ;CHECK: @cpp_new_arrays
-;CHECK: insertelement <4 x i32>
+;CHECK: sext i32
 ;CHECK: load <4 x float>
 ;CHECK: fadd <4 x float>
 ;CHECK: ret i32
diff --git a/test/Transforms/LoopVectorize/gcc-examples.ll b/test/Transforms/LoopVectorize/gcc-examples.ll
index f1bf6cb6d8..4fabc8e2d3 100644
--- a/test/Transforms/LoopVectorize/gcc-examples.ll
+++ b/test/Transforms/LoopVectorize/gcc-examples.ll
@@ -329,7 +329,7 @@ define void @example11() nounwind uwtable ssp {
 }
 
 ;CHECK: @example12
-;CHECK: trunc <4 x i64>
+;CHECK: trunc i64
 ;CHECK: store <4 x i32>
 ;CHECK: ret void
 define void @example12() nounwind uwtable ssp {
diff --git a/test/Transforms/LoopVectorize/induction_plus.ll b/test/Transforms/LoopVectorize/induction_plus.ll
index b31bceb50d..a25845f1c1 100644
--- a/test/Transforms/LoopVectorize/induction_plus.ll
+++ b/test/Transforms/LoopVectorize/induction_plus.ll
@@ -6,8 +6,7 @@ target triple = "x86_64-apple-macosx10.8.0"
 @array = common global [1024 x i32] zeroinitializer, align 16
 
 ;CHECK: @array_at_plus_one
-;CHECK: add <4 x i64>
-;CHECK: trunc <4 x i64>
+;CHECK: trunc i64
 ;CHECK: add i64 %index, 12
 ;CHECK: ret i32
 define i32 @array_at_plus_one(i32 %n) nounwind uwtable ssp {