1; RUN: llc < %s -mtriple=armv8 -mattr=+v8.1a | FileCheck %s
2
3;-----------------------------------------------------------------------------
4; RDMA Vector
5
6declare <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16>, <4 x i16>)
7declare <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16>, <8 x i16>)
8declare <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32>, <2 x i32>)
9declare <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32>, <4 x i32>)
10
11declare <4 x i16> @llvm.arm.neon.vqadds.v4i16(<4 x i16>, <4 x i16>)
12declare <8 x i16> @llvm.arm.neon.vqadds.v8i16(<8 x i16>, <8 x i16>)
13declare <2 x i32> @llvm.arm.neon.vqadds.v2i32(<2 x i32>, <2 x i32>)
14declare <4 x i32> @llvm.arm.neon.vqadds.v4i32(<4 x i32>, <4 x i32>)
15
16declare <4 x i16> @llvm.arm.neon.vqsubs.v4i16(<4 x i16>, <4 x i16>)
17declare <8 x i16> @llvm.arm.neon.vqsubs.v8i16(<8 x i16>, <8 x i16>)
18declare <2 x i32> @llvm.arm.neon.vqsubs.v2i32(<2 x i32>, <2 x i32>)
19declare <4 x i32> @llvm.arm.neon.vqsubs.v4i32(<4 x i32>, <4 x i32>)
20
21define <4 x i16> @test_vqrdmlah_v4i16(<4 x i16> %acc, <4 x i16> %mhs, <4 x i16> %rhs) {
22; CHECK-LABEL: test_vqrdmlah_v4i16:
23   %prod = call <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16> %mhs,  <4 x i16> %rhs)
24   %retval =  call <4 x i16> @llvm.arm.neon.vqadds.v4i16(<4 x i16> %acc,  <4 x i16> %prod)
25; CHECK: vqrdmlah.s16 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}
26   ret <4 x i16> %retval
27}
28
29define <8 x i16> @test_vqrdmlah_v8i16(<8 x i16> %acc, <8 x i16> %mhs, <8 x i16> %rhs) {
30; CHECK-LABEL: test_vqrdmlah_v8i16:
31   %prod = call <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16> %mhs, <8 x i16> %rhs)
32   %retval =  call <8 x i16> @llvm.arm.neon.vqadds.v8i16(<8 x i16> %acc, <8 x i16> %prod)
33; CHECK: vqrdmlah.s16 {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
34   ret <8 x i16> %retval
35}
36
37define <2 x i32> @test_vqrdmlah_v2i32(<2 x i32> %acc, <2 x i32> %mhs, <2 x i32> %rhs) {
38; CHECK-LABEL: test_vqrdmlah_v2i32:
39   %prod = call <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32> %mhs, <2 x i32> %rhs)
40   %retval =  call <2 x i32> @llvm.arm.neon.vqadds.v2i32(<2 x i32> %acc, <2 x i32> %prod)
41; CHECK: vqrdmlah.s32 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}
42   ret <2 x i32> %retval
43}
44
45define <4 x i32> @test_vqrdmlah_v4i32(<4 x i32> %acc, <4 x i32> %mhs, <4 x i32> %rhs) {
46; CHECK-LABEL: test_vqrdmlah_v4i32:
47   %prod = call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> %mhs, <4 x i32> %rhs)
48   %retval =  call <4 x i32> @llvm.arm.neon.vqadds.v4i32(<4 x i32> %acc, <4 x i32> %prod)
49; CHECK: vqrdmlah.s32 {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
50   ret <4 x i32> %retval
51}
52
53define <4 x i16> @test_vqrdmlsh_v4i16(<4 x i16> %acc, <4 x i16> %mhs, <4 x i16> %rhs) {
54; CHECK-LABEL: test_vqrdmlsh_v4i16:
55   %prod = call <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16> %mhs,  <4 x i16> %rhs)
56   %retval =  call <4 x i16> @llvm.arm.neon.vqsubs.v4i16(<4 x i16> %acc, <4 x i16> %prod)
57; CHECK: vqrdmlsh.s16 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}
58   ret <4 x i16> %retval
59}
60
61define <8 x i16> @test_vqrdmlsh_v8i16(<8 x i16> %acc, <8 x i16> %mhs, <8 x i16> %rhs) {
62; CHECK-LABEL: test_vqrdmlsh_v8i16:
63   %prod = call <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16> %mhs, <8 x i16> %rhs)
64   %retval =  call <8 x i16> @llvm.arm.neon.vqsubs.v8i16(<8 x i16> %acc, <8 x i16> %prod)
65; CHECK: vqrdmlsh.s16 {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
66   ret <8 x i16> %retval
67}
68
69define <2 x i32> @test_vqrdmlsh_v2i32(<2 x i32> %acc, <2 x i32> %mhs, <2 x i32> %rhs) {
70; CHECK-LABEL: test_vqrdmlsh_v2i32:
71   %prod = call <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32> %mhs, <2 x i32> %rhs)
72   %retval =  call <2 x i32> @llvm.arm.neon.vqsubs.v2i32(<2 x i32> %acc, <2 x i32> %prod)
73; CHECK: vqrdmlsh.s32 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}
74   ret <2 x i32> %retval
75}
76
77define <4 x i32> @test_vqrdmlsh_v4i32(<4 x i32> %acc, <4 x i32> %mhs, <4 x i32> %rhs) {
78; CHECK-LABEL: test_vqrdmlsh_v4i32:
79   %prod = call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> %mhs, <4 x i32> %rhs)
80   %retval =  call <4 x i32> @llvm.arm.neon.vqsubs.v4i32(<4 x i32> %acc, <4 x i32> %prod)
81; CHECK: vqrdmlsh.s32 {{q[0-9]+}}, {{q[0-9]+}}, {{q[0-9]+}}
82   ret <4 x i32> %retval
83}
84
85;-----------------------------------------------------------------------------
86; RDMA Scalar
87
88define <4 x i16> @test_vqrdmlah_lane_s16(<4 x i16> %acc, <4 x i16> %x, <4 x i16> %v) {
89; CHECK-LABEL: test_vqrdmlah_lane_s16:
90entry:
91  %shuffle = shufflevector <4 x i16> %v, <4 x i16> undef, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
92  %prod = call <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16> %x, <4 x i16> %shuffle)
93  %retval =  call <4 x i16> @llvm.arm.neon.vqadds.v4i16(<4 x i16> %acc, <4 x i16> %prod)
94; CHECK: vqrdmlah.s16 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}[3]
95  ret <4 x i16> %retval
96}
97
98define <8 x i16> @test_vqrdmlahq_lane_s16(<8 x i16> %acc, <8 x i16> %x, <4 x i16> %v) {
99; CHECK-LABEL: test_vqrdmlahq_lane_s16:
100entry:
101  %shuffle = shufflevector <4 x i16> %v, <4 x i16> undef, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
102  %prod = call <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16> %x, <8 x i16> %shuffle)
103  %retval =  call <8 x i16> @llvm.arm.neon.vqadds.v8i16(<8 x i16> %acc, <8 x i16> %prod)
104; CHECK: vqrdmlah.s16 {{q[0-9]+}}, {{q[0-9]+}}, {{d[0-9]+}}[2]
105  ret <8 x i16> %retval
106}
107
108define <2 x i32> @test_vqrdmlah_lane_s32(<2 x i32> %acc, <2 x i32> %x, <2 x i32> %v) {
109; CHECK-LABEL: test_vqrdmlah_lane_s32:
110entry:
111  %shuffle = shufflevector <2 x i32> %v, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
112  %prod = tail call <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32> %x, <2 x i32> %shuffle)
113  %retval =  call <2 x i32> @llvm.arm.neon.vqadds.v2i32(<2 x i32> %acc, <2 x i32> %prod)
114; CHECK: vqrdmlah.s32 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}[1]
115  ret <2 x i32> %retval
116}
117
118define <4 x i32> @test_vqrdmlahq_lane_s32(<4 x i32> %acc,<4 x i32> %x, <2 x i32> %v) {
119; CHECK-LABEL: test_vqrdmlahq_lane_s32:
120entry:
121  %shuffle = shufflevector <2 x i32> %v, <2 x i32> undef, <4 x i32> zeroinitializer
122  %prod = tail call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> %x, <4 x i32> %shuffle)
123  %retval =  call <4 x i32> @llvm.arm.neon.vqadds.v4i32(<4 x i32> %acc, <4 x i32> %prod)
124; CHECK: vqrdmlah.s32 {{q[0-9]+}}, {{q[0-9]+}}, {{d[0-9]+}}[0]
125  ret <4 x i32> %retval
126}
127
128define <4 x i16> @test_vqrdmlsh_lane_s16(<4 x i16> %acc, <4 x i16> %x, <4 x i16> %v) {
129; CHECK-LABEL: test_vqrdmlsh_lane_s16:
130entry:
131  %shuffle = shufflevector <4 x i16> %v, <4 x i16> undef, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
132  %prod = call <4 x i16> @llvm.arm.neon.vqrdmulh.v4i16(<4 x i16> %x, <4 x i16> %shuffle)
133  %retval =  call <4 x i16> @llvm.arm.neon.vqsubs.v4i16(<4 x i16> %acc, <4 x i16> %prod)
134; CHECK: vqrdmlsh.s16 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}[3]
135  ret <4 x i16> %retval
136}
137
138define <8 x i16> @test_vqrdmlshq_lane_s16(<8 x i16> %acc, <8 x i16> %x, <4 x i16> %v) {
139; CHECK-LABEL: test_vqrdmlshq_lane_s16:
140entry:
141  %shuffle = shufflevector <4 x i16> %v, <4 x i16> undef, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
142  %prod = call <8 x i16> @llvm.arm.neon.vqrdmulh.v8i16(<8 x i16> %x, <8 x i16> %shuffle)
143  %retval =  call <8 x i16> @llvm.arm.neon.vqsubs.v8i16(<8 x i16> %acc, <8 x i16> %prod)
144; CHECK: vqrdmlsh.s16 {{q[0-9]+}}, {{q[0-9]+}}, {{d[0-9]+}}[2]
145  ret <8 x i16> %retval
146}
147
148define <2 x i32> @test_vqrdmlsh_lane_s32(<2 x i32> %acc, <2 x i32> %x, <2 x i32> %v) {
149; CHECK-LABEL: test_vqrdmlsh_lane_s32:
150entry:
151  %shuffle = shufflevector <2 x i32> %v, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
152  %prod = tail call <2 x i32> @llvm.arm.neon.vqrdmulh.v2i32(<2 x i32> %x, <2 x i32> %shuffle)
153  %retval =  call <2 x i32> @llvm.arm.neon.vqsubs.v2i32(<2 x i32> %acc, <2 x i32> %prod)
154; CHECK: vqrdmlsh.s32  {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}[1]
155  ret <2 x i32> %retval
156}
157
158define <4 x i32> @test_vqrdmlshq_lane_s32(<4 x i32> %acc,<4 x i32> %x, <2 x i32> %v) {
159; CHECK-LABEL: test_vqrdmlshq_lane_s32:
160entry:
161  %shuffle = shufflevector <2 x i32> %v, <2 x i32> undef, <4 x i32> zeroinitializer
162  %prod = tail call <4 x i32> @llvm.arm.neon.vqrdmulh.v4i32(<4 x i32> %x, <4 x i32> %shuffle)
163  %retval =  call <4 x i32> @llvm.arm.neon.vqsubs.v4i32(<4 x i32> %acc, <4 x i32> %prod)
164; CHECK: vqrdmlsh.s32 {{q[0-9]+}}, {{q[0-9]+}}, {{d[0-9]+}}[0]
165  ret <4 x i32> %retval
166}
167