xref: /llvm-project/llvm/test/CodeGen/Thumb2/LowOverheadLoops/inlineasm.ll (revision 7b3bbd83c0c24087072ec5b22a76799ab31f87d5) 
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(ptr nocapture readonly %x, ptr nocapture readonly %y, i32 %n) {
5; CHECK-LABEL: test:
6; CHECK:       @ %bb.0: @ %entry
7; CHECK-NEXT:    cmp r2, #1
8; CHECK-NEXT:    itt lt
9; CHECK-NEXT:    movlt r0, #0
10; CHECK-NEXT:    bxlt lr
11; CHECK-NEXT:  .LBB0_1: @ %for.body.preheader
12; CHECK-NEXT:    .save {r7, lr}
13; CHECK-NEXT:    push {r7, lr}
14; CHECK-NEXT:    mov lr, r0
15; CHECK-NEXT:    movs r0, #0
16; CHECK-NEXT:  .LBB0_2: @ %for.body
17; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
18; CHECK-NEXT:    ldrh r3, [r1], #2
19; CHECK-NEXT:    subs r2, #1
20; CHECK-NEXT:    ldrh r12, [lr], #2
21; CHECK-NEXT:    @APP
22; CHECK-NEXT:    add r3, r12
23; CHECK-NEXT:    @NO_APP
24; CHECK-NEXT:    add r0, r3
25; CHECK-NEXT:    bne .LBB0_2
26; CHECK-NEXT:  @ %bb.3:
27; CHECK-NEXT:    pop {r7, pc}
28entry:
29  %cmp9 = icmp sgt i32 %n, 0
30  br i1 %cmp9, label %for.body, label %for.cond.cleanup
31
32for.cond.cleanup:                                 ; preds = %for.body, %entry
33  %s.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]
34  ret i32 %s.0.lcssa
35
36for.body:                                         ; preds = %entry, %for.body
37  %s.011 = phi i32 [ %add, %for.body ], [ 0, %entry ]
38  %i.010 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
39  %arrayidx = getelementptr inbounds i16, ptr %x, i32 %i.010
40  %0 = load i16, ptr %arrayidx, align 2
41  %arrayidx1 = getelementptr inbounds i16, ptr %y, i32 %i.010
42  %1 = load i16, ptr %arrayidx1, align 2
43  %2 = tail call i32 asm "add $0, $1, $2", "=r,r,r"(i16 %0, i16 %1) #1
44  %add = add nsw i32 %2, %s.011
45  %inc = add nuw nsw i32 %i.010, 1
46  %exitcond.not = icmp eq i32 %inc, %n
47  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
48}
49
50define i32 @testlr(ptr nocapture readonly %x, ptr nocapture readonly %y, i32 %n) {
51; CHECK-LABEL: testlr:
52; CHECK:       @ %bb.0: @ %entry
53; CHECK-NEXT:    cmp r2, #1
54; CHECK-NEXT:    itt lt
55; CHECK-NEXT:    movlt r0, #0
56; CHECK-NEXT:    bxlt lr
57; CHECK-NEXT:  .LBB1_1: @ %for.body.preheader
58; CHECK-NEXT:    .save {r4, lr}
59; CHECK-NEXT:    push {r4, lr}
60; CHECK-NEXT:    mov r3, r0
61; CHECK-NEXT:    movs r0, #0
62; CHECK-NEXT:  .LBB1_2: @ %for.body
63; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
64; CHECK-NEXT:    ldrh r4, [r1], #2
65; CHECK-NEXT:    subs r2, #1
66; CHECK-NEXT:    ldrh r12, [r3], #2
67; CHECK-NEXT:    @APP
68; CHECK-NEXT:    add r4, r12
69; CHECK-NEXT:    @NO_APP
70; CHECK-NEXT:    add r0, r4
71; CHECK-NEXT:    bne .LBB1_2
72; CHECK-NEXT:  @ %bb.3:
73; CHECK-NEXT:    pop {r4, pc}
74entry:
75  %cmp9 = icmp sgt i32 %n, 0
76  br i1 %cmp9, label %for.body, label %for.cond.cleanup
77
78for.cond.cleanup:                                 ; preds = %for.body, %entry
79  %s.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]
80  ret i32 %s.0.lcssa
81
82for.body:                                         ; preds = %entry, %for.body
83  %s.011 = phi i32 [ %add, %for.body ], [ 0, %entry ]
84  %i.010 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
85  %arrayidx = getelementptr inbounds i16, ptr %x, i32 %i.010
86  %0 = load i16, ptr %arrayidx, align 2
87  %arrayidx1 = getelementptr inbounds i16, ptr %y, i32 %i.010
88  %1 = load i16, ptr %arrayidx1, align 2
89  %2 = tail call i32 asm "add $0, $1, $2", "=r,r,r,~{lr}"(i16 %0, i16 %1) #1
90  %add = add nsw i32 %2, %s.011
91  %inc = add nuw nsw i32 %i.010, 1
92  %exitcond.not = icmp eq i32 %inc, %n
93  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
94}
95