xref: /llvm-project/llvm/test/CodeGen/Hexagon/hexagon_vector_loop_carried_reuse_commutative.ll (revision ef57977f2aa32661a09fa6538f47ddee0a004d11)
1; RUN: opt -mtriple hexagon-- -passes='loop(hexagon-vlcr),adce' -S %s | FileCheck %s
2
3; CHECK: %v32.hexagon.vlcr = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B
4
5target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"
6target triple = "hexagon"
7
8@g0 = external local_unnamed_addr global i32, align 4
9
10; Function Attrs: nounwind
11define void @f0(ptr noalias nocapture readonly %a0, ptr noalias nocapture %a1, i32 %a2) local_unnamed_addr #0 {
12b0:
13  %v0 = getelementptr inbounds i8, ptr %a0, i32 %a2
14  %v1 = mul nsw i32 %a2, 2
15  %v2 = getelementptr inbounds i8, ptr %a0, i32 %v1
16  %v3 = load i32, ptr @g0, align 4, !tbaa !0
17  %v4 = icmp sgt i32 %v3, 0
18  br i1 %v4, label %b1, label %b4
19
20b1:                                               ; preds = %b0
21  %v6 = load <32 x i32>, ptr %v2, align 128, !tbaa !4
22  %v7 = getelementptr inbounds i8, ptr %v2, i32 128
23  %v10 = load <32 x i32>, ptr %v0, align 128, !tbaa !4
24  %v11 = getelementptr inbounds i8, ptr %v0, i32 128
25  %v14 = load <32 x i32>, ptr %a0, align 128, !tbaa !4
26  %v15 = getelementptr inbounds i8, ptr %a0, i32 128
27  br label %b2
28
29b2:                                               ; preds = %b2, %b1
30  %v18 = phi ptr [ %a1, %b1 ], [ %v37, %b2 ]
31  %v19 = phi ptr [ %v7, %b1 ], [ %v30, %b2 ]
32  %v20 = phi ptr [ %v11, %b1 ], [ %v28, %b2 ]
33  %v21 = phi ptr [ %v15, %b1 ], [ %v26, %b2 ]
34  %v22 = phi i32 [ 0, %b1 ], [ %v38, %b2 ]
35  %v23 = phi <32 x i32> [ %v14, %b1 ], [ %v27, %b2 ]
36  %v24 = phi <32 x i32> [ %v10, %b1 ], [ %v29, %b2 ]
37  %v25 = phi <32 x i32> [ %v6, %b1 ], [ %v31, %b2 ]
38  %v26 = getelementptr inbounds <32 x i32>, ptr %v21, i32 1
39  %v27 = load <32 x i32>, ptr %v21, align 128, !tbaa !4
40  %v28 = getelementptr inbounds <32 x i32>, ptr %v20, i32 1
41  %v29 = load <32 x i32>, ptr %v20, align 128, !tbaa !4
42  %v30 = getelementptr inbounds <32 x i32>, ptr %v19, i32 1
43  %v31 = load <32 x i32>, ptr %v19, align 128, !tbaa !4
44  %v32 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v23, <32 x i32> %v24)
45  %v33 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v32, <32 x i32> %v25)
46  %v34 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v29, <32 x i32> %v27)
47  %v35 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v34, <32 x i32> %v31)
48  %v36 = tail call <32 x i32> @llvm.hexagon.V6.valignbi.128B(<32 x i32> %v35, <32 x i32> %v33, i32 1)
49  %v37 = getelementptr inbounds <32 x i32>, ptr %v18, i32 1
50  store <32 x i32> %v36, ptr %v18, align 128, !tbaa !4
51  %v38 = add nuw nsw i32 %v22, 128
52  %v39 = icmp slt i32 %v38, %v3
53  br i1 %v39, label %b2, label %b3
54
55b3:                                               ; preds = %b2
56  br label %b4
57
58b4:                                               ; preds = %b3, %b0
59  ret void
60}
61
62; Function Attrs: nounwind readnone
63declare <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32>, <32 x i32>) #1
64
65; Function Attrs: nounwind readnone
66declare <32 x i32> @llvm.hexagon.V6.valignbi.128B(<32 x i32>, <32 x i32>, i32) #1
67
68attributes #0 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvx,+hvx-length128b,-long-calls" }
69attributes #1 = { nounwind readnone }
70
71!0 = !{!1, !1, i64 0}
72!1 = !{!"int", !2, i64 0}
73!2 = !{!"omnipotent char", !3, i64 0}
74!3 = !{!"Simple C/C++ TBAA"}
75!4 = !{!2, !2, i64 0}
76