|
| 1 | +; RUN: opt -loop-vectorize -force-vector-width=2 -S %s | FileCheck %s |
| 2 | + |
| 3 | +; Test cases to make sure LV & loop versioning can handle loops with |
| 4 | +; multiple exiting branches. |
| 5 | + |
| 6 | +; Multiple branches exiting the loop to a unique exit block. The loop should |
| 7 | +; be vectorized with versioning & noalias metadata should be added. |
| 8 | +define void @multiple_exits_unique_exit_block(i32* %A, i32* %B, i64 %N) { |
| 9 | +; CHECK-LABEL: @multiple_exits_unique_exit_block |
| 10 | +; CHECK: vector.memcheck: |
| 11 | +; CHECK-LABEL: vector.body: |
| 12 | +; CHECK: %wide.load = load <2 x i32>, <2 x i32>* {{.*}}, align 4, !alias.scope |
| 13 | +; CHECK: store <2 x i32> %wide.load, <2 x i32>* {{.*}}, align 4, !alias.scope |
| 14 | +; CHECK: br |
| 15 | +; |
| 16 | +entry: |
| 17 | + br label %loop.header |
| 18 | + |
| 19 | +loop.header: |
| 20 | + %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] |
| 21 | + %cond.0 = icmp eq i64 %iv, %N |
| 22 | + br i1 %cond.0, label %exit, label %for.body |
| 23 | + |
| 24 | +for.body: |
| 25 | + %A.gep = getelementptr inbounds i32, i32* %A, i64 %iv |
| 26 | + %lv = load i32, i32* %A.gep, align 4 |
| 27 | + %B.gep = getelementptr inbounds i32, i32* %B, i64 %iv |
| 28 | + store i32 %lv, i32* %B.gep, align 4 |
| 29 | + %iv.next = add nuw i64 %iv, 1 |
| 30 | + %cond.1 = icmp ult i64 %iv.next, 1000 |
| 31 | + br i1 %cond.1, label %loop.header, label %exit |
| 32 | + |
| 33 | +exit: |
| 34 | + ret void |
| 35 | +} |
| 36 | + |
| 37 | + |
| 38 | +; Multiple branches exiting the loop to different blocks. Currently this is not supported. |
| 39 | +define i32 @multiple_exits_multiple_exit_blocks(i32* %A, i32* %B, i64 %N) { |
| 40 | +; CHECK-LABEL: @multiple_exits_multiple_exit_blocks |
| 41 | +; CHECK-NEXT: entry: |
| 42 | +; CHECK: br label %loop.header |
| 43 | +; CHECK-NOT: <2 x i32> |
| 44 | +; |
| 45 | +entry: |
| 46 | + br label %loop.header |
| 47 | + |
| 48 | +loop.header: |
| 49 | + %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] |
| 50 | + %cond.0 = icmp eq i64 %iv, %N |
| 51 | + br i1 %cond.0, label %exit.0, label %for.body |
| 52 | + |
| 53 | +for.body: |
| 54 | + %A.gep = getelementptr inbounds i32, i32* %A, i64 %iv |
| 55 | + %lv = load i32, i32* %A.gep, align 4 |
| 56 | + %B.gep = getelementptr inbounds i32, i32* %B, i64 %iv |
| 57 | + store i32 %lv, i32* %B.gep, align 4 |
| 58 | + %iv.next = add nuw i64 %iv, 1 |
| 59 | + %cond.1 = icmp ult i64 %iv.next, 1000 |
| 60 | + br i1 %cond.1, label %loop.header, label %exit.1 |
| 61 | + |
| 62 | +exit.0: |
| 63 | + ret i32 1 |
| 64 | + |
| 65 | +exit.1: |
| 66 | + ret i32 2 |
| 67 | +} |
0 commit comments