1; Test signaling vector floating-point comparisons on z13.
2; Note that these must be scalarized as we do not have native instructions.
3;
4; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
5
6; Test v4f32.
7define <4 x i32> @f1(<4 x float> %val1, <4 x float> %val2) #0 {
8; CHECK-LABEL: f1:
9; CHECK: kebr
10; CHECK: kebr
11; CHECK: kebr
12; CHECK: kebr
13; CHECK: br %r14
14  %cmp = call <4 x i1> @llvm.experimental.constrained.fcmps.v4f32(
15                                               <4 x float> %val1, <4 x float> %val2,
16                                               metadata !"oeq",
17                                               metadata !"fpexcept.strict") #0
18  %ret = sext <4 x i1> %cmp to <4 x i32>
19  ret <4 x i32> %ret
20}
21
22; Test v2f64.
23define <2 x i64> @f2(<2 x i64> %dummy, <2 x double> %val1, <2 x double> %val2) #0 {
24; CHECK-LABEL: f2:
25; CHECK: {{kdbr|wfkdb}}
26; CHECK: {{kdbr|wfkdb}}
27; CHECK: br %r14
28  %cmp = call <2 x i1> @llvm.experimental.constrained.fcmps.v2f64(
29                                               <2 x double> %val1, <2 x double> %val2,
30                                               metadata !"oeq",
31                                               metadata !"fpexcept.strict") #0
32  %ret = sext <2 x i1> %cmp to <2 x i64>
33  ret <2 x i64> %ret
34}
35
36; Test an f64 comparison that uses vector registers.
37define i64 @f3(i64 %a, i64 %b, double %f1, <2 x double> %vec) #0 {
38; CHECK-LABEL: f3:
39; CHECK: wfkdb %f0, %v24
40; CHECK-NEXT: locgrne %r2, %r3
41; CHECK: br %r14
42  %f2 = extractelement <2 x double> %vec, i32 0
43  %cond = call i1 @llvm.experimental.constrained.fcmps.f64(
44                                               double %f1, double %f2,
45                                               metadata !"oeq",
46                                               metadata !"fpexcept.strict") #0
47  %res = select i1 %cond, i64 %a, i64 %b
48  ret i64 %res
49}
50
51attributes #0 = { strictfp }
52
53declare <4 x i1> @llvm.experimental.constrained.fcmps.v4f32(<4 x float>, <4 x float>, metadata, metadata)
54declare <2 x i1> @llvm.experimental.constrained.fcmps.v2f64(<2 x double>, <2 x double>, metadata, metadata)
55declare i1 @llvm.experimental.constrained.fcmps.f64(double, double, metadata, metadata)
56
57