1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs -o - %s | FileCheck %s
3
4define i32 @test(i16* nocapture readonly %x, i16* nocapture readonly %y, i32 %n) {
5; CHECK-LABEL: test:
6; CHECK:       @ %bb.0: @ %entry
7; CHECK-NEXT:    .save {r7, lr}
8; CHECK-NEXT:    push {r7, lr}
9; CHECK-NEXT:    cmp r2, #1
10; CHECK-NEXT:    blt .LBB0_4
11; CHECK-NEXT:  @ %bb.1: @ %for.body.preheader
12; CHECK-NEXT:    mov lr, r0
13; CHECK-NEXT:    movs r0, #0
14; CHECK-NEXT:  .LBB0_2: @ %for.body
15; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
16; CHECK-NEXT:    ldrh r3, [r1], #2
17; CHECK-NEXT:    subs r2, #1
18; CHECK-NEXT:    ldrh r12, [lr], #2
19; CHECK-NEXT:    @APP
20; CHECK-NEXT:    add r3, r12
21; CHECK-NEXT:    @NO_APP
22; CHECK-NEXT:    add r0, r3
23; CHECK-NEXT:    bne .LBB0_2
24; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup
25; CHECK-NEXT:    pop {r7, pc}
26; CHECK-NEXT:  .LBB0_4:
27; CHECK-NEXT:    movs r0, #0
28; CHECK-NEXT:    pop {r7, pc}
29entry:
30  %cmp9 = icmp sgt i32 %n, 0
31  br i1 %cmp9, label %for.body, label %for.cond.cleanup
32
33for.cond.cleanup:                                 ; preds = %for.body, %entry
34  %s.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]
35  ret i32 %s.0.lcssa
36
37for.body:                                         ; preds = %entry, %for.body
38  %s.011 = phi i32 [ %add, %for.body ], [ 0, %entry ]
39  %i.010 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
40  %arrayidx = getelementptr inbounds i16, i16* %x, i32 %i.010
41  %0 = load i16, i16* %arrayidx, align 2
42  %arrayidx1 = getelementptr inbounds i16, i16* %y, i32 %i.010
43  %1 = load i16, i16* %arrayidx1, align 2
44  %2 = tail call i32 asm "add $0, $1, $2", "=r,r,r"(i16 %0, i16 %1) #1
45  %add = add nsw i32 %2, %s.011
46  %inc = add nuw nsw i32 %i.010, 1
47  %exitcond.not = icmp eq i32 %inc, %n
48  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
49}
50
51define i32 @testlr(i16* nocapture readonly %x, i16* nocapture readonly %y, i32 %n) {
52; CHECK-LABEL: testlr:
53; CHECK:       @ %bb.0: @ %entry
54; CHECK-NEXT:    .save {r4, lr}
55; CHECK-NEXT:    push {r4, lr}
56; CHECK-NEXT:    cmp r2, #1
57; CHECK-NEXT:    blt .LBB1_4
58; CHECK-NEXT:  @ %bb.1: @ %for.body.preheader
59; CHECK-NEXT:    mov r3, r0
60; CHECK-NEXT:    movs r0, #0
61; CHECK-NEXT:  .LBB1_2: @ %for.body
62; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
63; CHECK-NEXT:    ldrh r4, [r1], #2
64; CHECK-NEXT:    subs r2, #1
65; CHECK-NEXT:    ldrh r12, [r3], #2
66; CHECK-NEXT:    @APP
67; CHECK-NEXT:    add r4, r12
68; CHECK-NEXT:    @NO_APP
69; CHECK-NEXT:    add r0, r4
70; CHECK-NEXT:    bne .LBB1_2
71; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup
72; CHECK-NEXT:    pop {r4, pc}
73; CHECK-NEXT:  .LBB1_4:
74; CHECK-NEXT:    movs r0, #0
75; CHECK-NEXT:    pop {r4, pc}
76entry:
77  %cmp9 = icmp sgt i32 %n, 0
78  br i1 %cmp9, label %for.body, label %for.cond.cleanup
79
80for.cond.cleanup:                                 ; preds = %for.body, %entry
81  %s.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]
82  ret i32 %s.0.lcssa
83
84for.body:                                         ; preds = %entry, %for.body
85  %s.011 = phi i32 [ %add, %for.body ], [ 0, %entry ]
86  %i.010 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
87  %arrayidx = getelementptr inbounds i16, i16* %x, i32 %i.010
88  %0 = load i16, i16* %arrayidx, align 2
89  %arrayidx1 = getelementptr inbounds i16, i16* %y, i32 %i.010
90  %1 = load i16, i16* %arrayidx1, align 2
91  %2 = tail call i32 asm "add $0, $1, $2", "=r,r,r,~{lr}"(i16 %0, i16 %1) #1
92  %add = add nsw i32 %2, %s.011
93  %inc = add nuw nsw i32 %i.010, 1
94  %exitcond.not = icmp eq i32 %inc, %n
95  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
96}
97