xref: /llvm-project/llvm/test/CodeGen/AArch64/fast-isel-cmp-vec.ll (revision b124295ef60f4ea5bc77b38d12da608e10d7d34e)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=aarch64-apple-darwin -fast-isel -verify-machineinstrs \
3; RUN:   -aarch64-enable-atomic-cfg-tidy=0 -disable-cgp -disable-branch-fold \
4; RUN:   < %s | FileCheck %s
5
6;
7; Verify that we don't mess up vector comparisons in fast-isel.
8;
9
10define <2 x i32> @icmp_v2i32(<2 x i32> %a) {
11; CHECK-LABEL: icmp_v2i32:
12; CHECK:       ; %bb.0:
13; CHECK-NEXT:    cmeq.2s v0, v0, #0
14; CHECK-NEXT:  ; %bb.1: ; %bb2
15; CHECK-NEXT:    movi.2s v1, #1
16; CHECK-NEXT:    and.8b v0, v0, v1
17; CHECK-NEXT:    ret
18  %c = icmp eq <2 x i32> %a, zeroinitializer
19  br label %bb2
20bb2:
21  %z = zext <2 x i1> %c to <2 x i32>
22  ret <2 x i32> %z
23}
24
25define <2 x i32> @icmp_constfold_v2i32(<2 x i32> %a) {
26; CHECK-LABEL: icmp_constfold_v2i32:
27; CHECK:       ; %bb.0:
28; CHECK-NEXT:    movi.2s v0, #1
29; CHECK-NEXT:    and.8b v0, v0, v0
30; CHECK-NEXT:    ret
31  %1 = icmp eq <2 x i32> %a, %a
32  br label %bb2
33bb2:
34  %2 = zext <2 x i1> %1 to <2 x i32>
35  ret <2 x i32> %2
36}
37
38define <4 x i32> @icmp_v4i32(<4 x i32> %a) {
39; CHECK-LABEL: icmp_v4i32:
40; CHECK:       ; %bb.0:
41; CHECK-NEXT:    cmeq.4s v0, v0, #0
42; CHECK-NEXT:    xtn.4h v0, v0
43; CHECK-NEXT:  ; %bb.1: ; %bb2
44; CHECK-NEXT:    movi.4h v1, #1
45; CHECK-NEXT:    and.8b v0, v0, v1
46; CHECK-NEXT:    ushll.4s v0, v0, #0
47; CHECK-NEXT:    ret
48  %c = icmp eq <4 x i32> %a, zeroinitializer
49  br label %bb2
50bb2:
51  %z = zext <4 x i1> %c to <4 x i32>
52  ret <4 x i32> %z
53}
54
55define <4 x i32> @icmp_constfold_v4i32(<4 x i32> %a) {
56; CHECK-LABEL: icmp_constfold_v4i32:
57; CHECK:       ; %bb.0:
58; CHECK-NEXT:    movi.4h v0, #1
59; CHECK-NEXT:  ; %bb.1: ; %bb2
60; CHECK-NEXT:    and.8b v0, v0, v0
61; CHECK-NEXT:    ushll.4s v0, v0, #0
62; CHECK-NEXT:    ret
63  %1 = icmp eq <4 x i32> %a, %a
64  br label %bb2
65bb2:
66  %2 = zext <4 x i1> %1 to <4 x i32>
67  ret <4 x i32> %2
68}
69
70define <16 x i8> @icmp_v16i8(<16 x i8> %a) {
71; CHECK-LABEL: icmp_v16i8:
72; CHECK:       ; %bb.0:
73; CHECK-NEXT:    cmeq.16b v0, v0, #0
74; CHECK-NEXT:  ; %bb.1: ; %bb2
75; CHECK-NEXT:    movi.16b v1, #1
76; CHECK-NEXT:    and.16b v0, v0, v1
77; CHECK-NEXT:    ret
78  %c = icmp eq <16 x i8> %a, zeroinitializer
79  br label %bb2
80bb2:
81  %z = zext <16 x i1> %c to <16 x i8>
82  ret <16 x i8> %z
83}
84
85define <16 x i8> @icmp_constfold_v16i8(<16 x i8> %a) {
86; CHECK-LABEL: icmp_constfold_v16i8:
87; CHECK:       ; %bb.0:
88; CHECK-NEXT:    movi.16b v0, #1
89; CHECK-NEXT:    ret
90  %1 = icmp eq <16 x i8> %a, %a
91  br label %bb2
92bb2:
93  %2 = zext <16 x i1> %1 to <16 x i8>
94  ret <16 x i8> %2
95}
96