Do not vectorize loops with subtraction reductions

Since subtraction does not commute the loop vectorizer incorrectly vectorizes
reductions such as x = A[i] - x.

Disabling for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171537 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 4e508988e33b25db661f890303b3592172a2e3b8..5e2d7971f84da8eb58d01674bcc50a18211102f8 100644 (file)
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1986,7 +1986,6 @@ LoopVectorizationLegality::isReductionInstr(Instruction *I,
     // possibly.
     return true;
   case Instruction::Add:
-  case Instruction::Sub:
     return Kind == IntegerAdd;
   case Instruction::Mul:
     return Kind == IntegerMult;

diff --git a/test/Transforms/LoopVectorize/gcc-examples.ll b/test/Transforms/LoopVectorize/gcc-examples.ll
index 652c2a0d025ece33afd6eef25d7e0a300e534bc2..e52a3f164a3fe618e0388d8da13f5a39918dbf91 100644 (file)
--- a/test/Transforms/LoopVectorize/gcc-examples.ll
+++ b/test/Transforms/LoopVectorize/gcc-examples.ll
@@ -241,7 +241,7 @@ define void @example8(i32 %x) nounwind uwtable ssp {
 }
 
 ;CHECK: @example9
-;CHECK: phi <4 x i32>
+;CHECK-NOT: phi <4 x i32>
 ;CHECK: ret i32
 define i32 @example9() nounwind uwtable readonly ssp {
   br label %1

diff --git a/test/Transforms/LoopVectorize/reduction.ll b/test/Transforms/LoopVectorize/reduction.ll
index bc1c0290dec89fc4fcfe64e0b67fa75397d967e9..91fc9538772090665bd467690d9446d8105c8510 100644 (file)
--- a/test/Transforms/LoopVectorize/reduction.ll
+++ b/test/Transforms/LoopVectorize/reduction.ll
@@ -270,3 +270,53 @@ for.end:                                          ; preds = %for.body, %entry
   %result.0.lcssa = phi i32 [ 0, %entry ], [ %xor, %for.body ]
   ret i32 %result.0.lcssa
 }
+
+;CHECK: @reduction_sub_rhs
+;CHECK-NOT: phi <4 x i32>
+;CHECK-NOT: sub nsw <4 x i32>
+;CHECK: ret i32
+define i32 @reduction_sub_rhs(i32 %n, i32* noalias nocapture %A) nounwind uwtable readonly {
+entry:
+  %cmp3 = icmp sgt i32 %n, 0
+  br i1 %cmp3, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+  %x.04 = phi i32 [ %sub, %for.body ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32* %A, i64 %indvars.iv
+  %0 = load i32* %arrayidx, align 4
+  %sub = sub nsw i32 %x.04, %0
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  %x.0.lcssa = phi i32 [ 0, %entry ], [ %sub, %for.body ]
+  ret i32 %x.0.lcssa
+}
+
+;CHECK: @reduction_sub_lhs
+;CHECK-NOT: phi <4 x i32>
+;CHECK-NOT: sub nsw <4 x i32>
+;CHECK: ret i32
+define i32 @reduction_sub_lhs(i32 %n, i32* noalias nocapture %A) nounwind uwtable readonly {
+entry:
+  %cmp4 = icmp sgt i32 %n, 0
+  br i1 %cmp4, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+  %x.05 = phi i32 [ %sub, %for.body ], [ 0, %entry ]
+  %arrayidx = getelementptr inbounds i32* %A, i64 %indvars.iv
+  %0 = load i32* %arrayidx, align 4
+  %sub = sub nsw i32 %0, %x.05
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  %x.0.lcssa = phi i32 [ 0, %entry ], [ %sub, %for.body ]
+  ret i32 %x.0.lcssa
+}