1 ; RUN: llc -march=hexagon -enable-pipeliner -pipeliner-max-stages=1 < %s -pipeliner-experimental-cg=true | FileCheck %s
3 ; Test that we update the offset correctly for loads that are
4 ; moved past stores. In these cases, we change the dependences
5 ; to make it easier to move the instructions, and we have to update
6 ; the register/offsets correctly after the schedule is finalized.
8 @g0 = common global [400 x i32] zeroinitializer, align 8
9 @g1 = common global [400 x i32] zeroinitializer, align 8
11 ; Function Attrs: nounwind
12 define void @f0() #0 {
19 ; CHECK: loop0(.LBB0_[[LOOP:.]],
20 ; CHECK: .LBB0_[[LOOP]]:
21 ; CHECK: = memd([[REG1:(r[0-9]+)]]+#8)
22 ; CHECK: memd([[REG1]]++#8) =
23 ; CHECK: }{{[ \t]*}}:endloop
25 b2: ; preds = %b2, %b0
26 %v0 = phi i32* [ getelementptr inbounds ([400 x i32], [400 x i32]* @g0, i32 0, i32 0), %b0 ], [ %v11, %b2 ]
27 %v1 = phi i32* [ getelementptr inbounds ([400 x i32], [400 x i32]* @g1, i32 0, i32 0), %b0 ], [ %v12, %b2 ]
28 %v2 = phi i32 [ 0, %b0 ], [ %v9, %b2 ]
29 %v3 = bitcast i32* %v0 to <2 x i32>*
30 %v4 = load <2 x i32>, <2 x i32>* %v3, align 8
31 %v5 = mul <2 x i32> %v4, <i32 7, i32 7>
32 %v6 = bitcast i32* %v1 to <2 x i32>*
33 %v7 = load <2 x i32>, <2 x i32>* %v6, align 8
34 %v8 = add <2 x i32> %v7, %v5
35 store <2 x i32> %v8, <2 x i32>* %v6, align 8
36 %v9 = add nsw i32 %v2, 2
37 %v10 = icmp slt i32 %v2, 398
38 %v11 = getelementptr i32, i32* %v0, i32 2
39 %v12 = getelementptr i32, i32* %v1, i32 2
40 br i1 %v10, label %b2, label %b1
43 attributes #0 = { nounwind "target-cpu"="hexagonv55" }