xref: /llvm-project/llvm/test/CodeGen/AArch64/arm64-AdvSIMD-Scalar.ll (revision db158c7c830807caeeb0691739c41f1d522029e9)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -verify-machineinstrs -mtriple=arm64-eabi -aarch64-neon-syntax=apple -aarch64-enable-simd-scalar=true | FileCheck %s -check-prefix=CHECK
3; RUN: llc < %s -verify-machineinstrs -mtriple=arm64-eabi -aarch64-neon-syntax=generic -aarch64-enable-simd-scalar=true | FileCheck %s -check-prefix=GENERIC
4
5define <2 x i64> @bar(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
6; CHECK-LABEL: bar:
7; CHECK:       // %bb.0:
8; CHECK-NEXT:    add.2d v0, v0, v1
9; CHECK-NEXT:    sub d2, d0, d1
10; CHECK-NEXT:    add d0, d0, d1
11; CHECK-NEXT:    fmov x8, d2
12; CHECK-NEXT:    mov.d v0[1], x8
13; CHECK-NEXT:    ret
14;
15; GENERIC-LABEL: bar:
16; GENERIC:       // %bb.0:
17; GENERIC-NEXT:    add v0.2d, v0.2d, v1.2d
18; GENERIC-NEXT:    sub d2, d0, d1
19; GENERIC-NEXT:    add d0, d0, d1
20; GENERIC-NEXT:    fmov x8, d2
21; GENERIC-NEXT:    mov v0.d[1], x8
22; GENERIC-NEXT:    ret
23  %add = add <2 x i64> %a, %b
24  %vgetq_lane = extractelement <2 x i64> %add, i32 0
25  %vgetq_lane2 = extractelement <2 x i64> %b, i32 0
26  %add3 = add i64 %vgetq_lane, %vgetq_lane2
27  %sub = sub i64 %vgetq_lane, %vgetq_lane2
28  %vecinit = insertelement <2 x i64> undef, i64 %add3, i32 0
29  %vecinit8 = insertelement <2 x i64> %vecinit, i64 %sub, i32 1
30  ret <2 x i64> %vecinit8
31}
32
33define double @subdd_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
34; CHECK-LABEL: subdd_su64:
35; CHECK:       // %bb.0:
36; CHECK-NEXT:    sub d0, d1, d0
37; CHECK-NEXT:    ret
38;
39; GENERIC-LABEL: subdd_su64:
40; GENERIC:       // %bb.0:
41; GENERIC-NEXT:    sub d0, d1, d0
42; GENERIC-NEXT:    ret
43  %vecext = extractelement <2 x i64> %a, i32 0
44  %vecext1 = extractelement <2 x i64> %b, i32 0
45  %sub.i = sub nsw i64 %vecext1, %vecext
46  %retval = bitcast i64 %sub.i to double
47  ret double %retval
48}
49
50define double @vaddd_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
51; CHECK-LABEL: vaddd_su64:
52; CHECK:       // %bb.0:
53; CHECK-NEXT:    add d0, d1, d0
54; CHECK-NEXT:    ret
55;
56; GENERIC-LABEL: vaddd_su64:
57; GENERIC:       // %bb.0:
58; GENERIC-NEXT:    add d0, d1, d0
59; GENERIC-NEXT:    ret
60  %vecext = extractelement <2 x i64> %a, i32 0
61  %vecext1 = extractelement <2 x i64> %b, i32 0
62  %add.i = add nsw i64 %vecext1, %vecext
63  %retval = bitcast i64 %add.i to double
64  ret double %retval
65}
66
67; sub MI doesn't access dsub register.
68define double @add_sub_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
69; CHECK-LABEL: add_sub_su64:
70; CHECK:       // %bb.0:
71; CHECK-NEXT:    add d0, d1, d0
72; CHECK-NEXT:    fmov d1, xzr
73; CHECK-NEXT:    sub d0, d1, d0
74; CHECK-NEXT:    ret
75;
76; GENERIC-LABEL: add_sub_su64:
77; GENERIC:       // %bb.0:
78; GENERIC-NEXT:    add d0, d1, d0
79; GENERIC-NEXT:    fmov d1, xzr
80; GENERIC-NEXT:    sub d0, d1, d0
81; GENERIC-NEXT:    ret
82  %vecext = extractelement <2 x i64> %a, i32 0
83  %vecext1 = extractelement <2 x i64> %b, i32 0
84  %add.i = add i64 %vecext1, %vecext
85  %sub.i = sub i64 0, %add.i
86  %retval = bitcast i64 %sub.i to double
87  ret double %retval
88}
89define double @and_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
90; CHECK-LABEL: and_su64:
91; CHECK:       // %bb.0:
92; CHECK-NEXT:    and.8b v0, v1, v0
93; CHECK-NEXT:    ret
94;
95; GENERIC-LABEL: and_su64:
96; GENERIC:       // %bb.0:
97; GENERIC-NEXT:    and v0.8b, v1.8b, v0.8b
98; GENERIC-NEXT:    ret
99  %vecext = extractelement <2 x i64> %a, i32 0
100  %vecext1 = extractelement <2 x i64> %b, i32 0
101  %or.i = and i64 %vecext1, %vecext
102  %retval = bitcast i64 %or.i to double
103  ret double %retval
104}
105
106define double @orr_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
107; CHECK-LABEL: orr_su64:
108; CHECK:       // %bb.0:
109; CHECK-NEXT:    orr.8b v0, v1, v0
110; CHECK-NEXT:    ret
111;
112; GENERIC-LABEL: orr_su64:
113; GENERIC:       // %bb.0:
114; GENERIC-NEXT:    orr v0.8b, v1.8b, v0.8b
115; GENERIC-NEXT:    ret
116  %vecext = extractelement <2 x i64> %a, i32 0
117  %vecext1 = extractelement <2 x i64> %b, i32 0
118  %or.i = or i64 %vecext1, %vecext
119  %retval = bitcast i64 %or.i to double
120  ret double %retval
121}
122
123define double @xorr_su64(<2 x i64> %a, <2 x i64> %b) nounwind readnone {
124; CHECK-LABEL: xorr_su64:
125; CHECK:       // %bb.0:
126; CHECK-NEXT:    eor.8b v0, v1, v0
127; CHECK-NEXT:    ret
128;
129; GENERIC-LABEL: xorr_su64:
130; GENERIC:       // %bb.0:
131; GENERIC-NEXT:    eor v0.8b, v1.8b, v0.8b
132; GENERIC-NEXT:    ret
133  %vecext = extractelement <2 x i64> %a, i32 0
134  %vecext1 = extractelement <2 x i64> %b, i32 0
135  %xor.i = xor i64 %vecext1, %vecext
136  %retval = bitcast i64 %xor.i to double
137  ret double %retval
138}
139