1 ; RUN: llc -march=hexagon < %s -pipeliner-experimental-cg=true | FileCheck %s
3 ; Test that when we order instructions in a packet we check for
4 ; order dependences so that the source of an order dependence
5 ; appears before the destination.
7 ; CHECK: loop0(.LBB0_[[LOOP:.]],
8 ; CHECK: .LBB0_[[LOOP]]:
11 ; CHECK: memw({{.*}}) =
16 @g0 = external hidden unnamed_addr constant [19 x i8], align 1
18 ; Function Attrs: nounwind optsize
19 declare i32 @f0(i8* nocapture readonly, ...) #0
21 ; Function Attrs: nounwind optsize
22 declare void @f1(i32*, i32*, i32* nocapture readnone) #0
24 ; Function Attrs: argmemonly nounwind
25 declare i8* @llvm.hexagon.circ.stw(i8*, i32, i32, i32) #1
27 ; Function Attrs: nounwind optsize
28 define void @f2(i32* %a0, i32* %a1, i32* %a2) #0 {
30 %v0 = alloca i32, align 4
31 call void @f1(i32* %a2, i32* %a0, i32* %v0) #2
32 %v1 = bitcast i32* %a1 to i8*
35 b1: ; preds = %b1, %b0
36 %v2 = phi i32 [ 0, %b0 ], [ %v13, %b1 ]
37 %v3 = phi i32* [ %a2, %b0 ], [ %v16, %b1 ]
38 %v4 = phi i32 [ 0, %b0 ], [ %v14, %b1 ]
39 %v5 = load i32, i32* %a1, align 4, !tbaa !0
40 %v6 = add nsw i32 %v2, %v5
41 %v7 = load i32, i32* %v3, align 4, !tbaa !0
42 %v8 = tail call i8* @llvm.hexagon.circ.stw(i8* %v1, i32 %v7, i32 150995968, i32 4) #3
43 %v9 = bitcast i8* %v8 to i32*
44 %v10 = load i32, i32* %v3, align 4, !tbaa !0
45 %v11 = add nsw i32 %v6, %v10
46 %v12 = load i32, i32* %v9, align 4, !tbaa !0
47 %v13 = add nsw i32 %v11, %v12
48 %v14 = add nsw i32 %v4, 1
49 %v15 = icmp eq i32 %v14, 2
50 %v16 = getelementptr i32, i32* %v3, i32 1
51 br i1 %v15, label %b2, label %b1
54 %v17 = tail call i32 (i8*, ...) @f0(i8* getelementptr inbounds ([19 x i8], [19 x i8]* @g0, i32 0, i32 0), i32 %v13) #4
58 attributes #0 = { nounwind optsize "target-cpu"="hexagonv55" }
59 attributes #1 = { argmemonly nounwind }
60 attributes #2 = { optsize }
61 attributes #3 = { nounwind }
62 attributes #4 = { nounwind optsize }
65 !1 = !{!"int", !2, i64 0}
66 !2 = !{!"omnipotent char", !3, i64 0}
67 !3 = !{!"Simple C/C++ TBAA"}