1; Test vector population-count instruction
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
4
5declare <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a)
6declare <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a)
7declare <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a)
8declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
9
10define <16 x i8> @f1(<16 x i8> %a) {
11; CHECK-LABEL: f1:
12; CHECK: vpopct  %v24, %v24, 0
13; CHECK: br      %r14
14
15  %popcnt = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a)
16  ret <16 x i8> %popcnt
17}
18
19define <8 x i16> @f2(<8 x i16> %a) {
20; CHECK-LABEL: f2:
21; CHECK: vpopct  [[T1:%v[0-9]+]], %v24, 0
22; CHECK: veslh   [[T2:%v[0-9]+]], [[T1]], 8
23; CHECK: vah     [[T3:%v[0-9]+]], [[T1]], [[T2]]
24; CHECK: vesrlh  %v24, [[T3]], 8
25; CHECK: br      %r14
26
27  %popcnt = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a)
28  ret <8 x i16> %popcnt
29}
30
31define <4 x i32> @f3(<4 x i32> %a) {
32; CHECK-LABEL: f3:
33; CHECK: vpopct  [[T1:%v[0-9]+]], %v24, 0
34; CHECK: vgbm    [[T2:%v[0-9]+]], 0
35; CHECK: vsumb   %v24, [[T1]], [[T2]]
36; CHECK: br      %r14
37
38  %popcnt = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a)
39  ret <4 x i32> %popcnt
40}
41
42define <2 x i64> @f4(<2 x i64> %a) {
43; CHECK-LABEL: f4:
44; CHECK: vpopct  [[T1:%v[0-9]+]], %v24, 0
45; CHECK: vgbm    [[T2:%v[0-9]+]], 0
46; CHECK: vsumb   [[T3:%v[0-9]+]], [[T1]], [[T2]]
47; CHECK: vsumgf  %v24, [[T3]], [[T2]]
48; CHECK: br      %r14
49
50  %popcnt = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
51  ret <2 x i64> %popcnt
52}
53
54