1; RUN: llc -march=amdgcn -mcpu=verde < %s | FileCheck %s
2
3; Test a simple uniform loop that lives inside non-uniform control flow.
4
5; CHECK-LABEL: {{^}}test1:
6; CHECK: v_cmp_ne_u32_e32 vcc, 0
7; CHECK: s_and_saveexec_b64
8; CHECK-NEXT: s_xor_b64
9; CHECK-NEXT: ; mask branch
10
11; CHECK-NEXT: BB{{[0-9]+_[0-9]+}}: ; %loop_body.preheader
12
13; CHECK: [[LOOP_BODY_LABEL:BB[0-9]+_[0-9]+]]:
14; CHECK: s_cbranch_scc0 [[LOOP_BODY_LABEL]]
15
16; CHECK: s_endpgm
17define amdgpu_ps void @test1(<8 x i32> inreg %rsrc, <2 x i32> %addr.base, i32 %y, i32 %p) {
18main_body:
19  %cc = icmp eq i32 %p, 0
20  br i1 %cc, label %out, label %loop_body
21
22loop_body:
23  %counter = phi i32 [ 0, %main_body ], [ %incr, %loop_body ]
24
25  ; Prevent the loop from being optimized out
26  call void asm sideeffect "", "" ()
27
28  %incr = add i32 %counter, 1
29  %lc = icmp sge i32 %incr, 1000
30  br i1 %lc, label %out, label %loop_body
31
32out:
33  ret void
34}
35
36; CHECK-LABEL: {{^}}test2:
37; CHECK: s_and_saveexec_b64
38; CHECK-NEXT: s_xor_b64
39; CHECK-NEXT: ; mask branch
40; CHECK-NEXT: s_cbranch_execz
41define void @test2(i32 addrspace(1)* %out, i32 %a, i32 %b) {
42main_body:
43  %tid = call i32 @llvm.amdgcn.workitem.id.x() #1
44  %cc = icmp eq i32 %tid, 0
45  br i1 %cc, label %done1, label %if
46
47if:
48  %cmp = icmp eq i32 %a, 0
49  br i1 %cmp, label %done0, label %loop_body
50
51loop_body:
52  %counter = phi i32 [ 0, %if ], [0, %done0], [ %incr, %loop_body ]
53
54  ; Prevent the loop from being optimized out
55  call void asm sideeffect "", "" ()
56
57  %incr = add i32 %counter, 1
58  %lc = icmp sge i32 %incr, 1000
59  br i1 %lc, label %done1, label %loop_body
60
61done0:
62  %cmp0 = icmp eq i32 %b, 0
63  br i1 %cmp0, label %done1, label %loop_body
64
65done1:
66  ret void
67}
68
69declare i32 @llvm.amdgcn.workitem.id.x() #1
70
71attributes #1 = { nounwind readonly }
72