xref: /llvm-project/llvm/test/Analysis/UniformityAnalysis/AMDGPU/hidden_diverge.ll (revision ae77aceba5ad6ee575d3d79eb0259624322b19f4)
1; RUN: opt -mtriple amdgcn-unknown-amdhsa -passes='print<uniformity>' -disable-output %s 2>&1 | FileCheck %s
2
3define amdgpu_kernel void @hidden_diverge(i32 %n, i32 %a, i32 %b) #0 {
4; CHECK-LABEL: for function 'hidden_diverge'
5entry:
6  %tid = call i32 @llvm.amdgcn.workitem.id.x()
7  %cond.var = icmp slt i32 %tid, 0
8  br i1 %cond.var, label %B, label %C ; divergent
9; CHECK:  DIVERGENT: %cond.var =
10; CHECK: DIVERGENT: br i1 %cond.var,
11B:
12  %cond.uni = icmp slt i32 %n, 0
13  br i1 %cond.uni, label %C, label %merge ; uniform
14; CHECK-NOT: DIVERGENT: br i1 %cond.uni,
15C:
16  %phi.var.hidden = phi i32 [ 1, %entry ], [ 2, %B  ]
17; CHECK: DIVERGENT: %phi.var.hidden = phi i32
18  br label %merge
19merge:
20  %phi.ipd = phi i32 [ %a, %B ], [ %b, %C ]
21; CHECK: DIVERGENT: %phi.ipd = phi i32
22  ret void
23}
24
25define amdgpu_kernel void @hidden_loop_ipd(i32 %n, i32 %a, i32 %b) #0 {
26; CHECK-LABEL: for function 'hidden_loop_ipd'
27entry:
28  %tid = call i32 @llvm.amdgcn.workitem.id.x()
29  %cond.var = icmp slt i32 %tid, 0
30; CHECK: DIVERGENT: %cond.var = icmp
31  %cond.uni = icmp slt i32 %n, 0
32; CHECK-NOT: DIVERGENT: %cond.uni = icmp
33  br label %for.header
34for.header:
35  br i1 %cond.var, label %A, label %B
36A:
37  br label %C
38B:
39  br label %C
40C:
41  br i1 %cond.uni, label %E, label %D
42D:
43  br i1 %cond.var, label %for.header, label %F
44
45E:
46  %e.lcssa.uni = phi i32 [ 0, %C ]
47; CHECK-NOT: DIVERGENT: %e.lcssa.uni = phi i32
48  br label %G
49
50F:
51  %f.lcssa.uni = phi i32 [ 1, %D ]
52; CHECK-NOT: DIVERGENT: %f.lcssa.uni = phi i32
53  br label %G
54
55G:
56  %g.join.var = phi i32 [ %e.lcssa.uni, %E ], [ %f.lcssa.uni, %F ]
57; CHECK: DIVERGENT: %g.join.var = phi i32
58  ret void
59}
60
61
62declare i32 @llvm.amdgcn.workitem.id.x() #0
63
64attributes #0 = { nounwind readnone }
65