1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
3
4;
5; LD1B, LD1W, LD1H, LD1D: vector base + scalar offset (index)
6;   e.g. ld1b { z0.d }, p0/z, [x0, z0.d]
7;
8
9; LD1B
10define <vscale x 4 x i32> @gld1b_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {
11; CHECK-LABEL: gld1b_s_scalar_offset:
12; CHECK:       // %bb.0:
13; CHECK-NEXT:    ld1b { z0.s }, p0/z, [x0, z0.s, uxtw]
14; CHECK-NEXT:    ret
15  %load = call <vscale x 4 x i8> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1> %pg,
16                                                                                           <vscale x 4 x i32> %base,
17                                                                                           i64 %offset)
18  %res = zext <vscale x 4 x i8> %load to <vscale x 4 x i32>
19  ret <vscale x 4 x i32> %res
20}
21
22define <vscale x 2 x i64> @gld1b_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
23; CHECK-LABEL: gld1b_d_scalar_offset:
24; CHECK:       // %bb.0:
25; CHECK-NEXT:    ld1b { z0.d }, p0/z, [x0, z0.d]
26; CHECK-NEXT:    ret
27  %load = call <vscale x 2 x i8> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1> %pg,
28                                                                                           <vscale x 2 x i64> %base,
29                                                                                           i64 %offset)
30  %res = zext <vscale x 2 x i8> %load to <vscale x 2 x i64>
31  ret <vscale x 2 x i64> %res
32}
33
34; LD1H
35define <vscale x 4 x i32> @gld1h_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {
36; CHECK-LABEL: gld1h_s_scalar_offset:
37; CHECK:       // %bb.0:
38; CHECK-NEXT:    ld1h { z0.s }, p0/z, [x0, z0.s, uxtw]
39; CHECK-NEXT:    ret
40  %load = call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> %pg,
41                                                                                             <vscale x 4 x i32> %base,
42                                                                                             i64 %offset)
43  %res = zext <vscale x 4 x i16> %load to <vscale x 4 x i32>
44  ret <vscale x 4 x i32> %res
45}
46
47define <vscale x 2 x i64> @gld1h_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
48; CHECK-LABEL: gld1h_d_scalar_offset:
49; CHECK:       // %bb.0:
50; CHECK-NEXT:    ld1h { z0.d }, p0/z, [x0, z0.d]
51; CHECK-NEXT:    ret
52  %load = call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> %pg,
53                                                                                             <vscale x 2 x i64> %base,
54                                                                                             i64 %offset)
55  %res = zext <vscale x 2 x i16> %load to <vscale x 2 x i64>
56  ret <vscale x 2 x i64> %res
57}
58
59; LD1W
60define <vscale x 4 x i32> @gld1w_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {
61; CHECK-LABEL: gld1w_s_scalar_offset:
62; CHECK:       // %bb.0:
63; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0, z0.s, uxtw]
64; CHECK-NEXT:    ret
65  %load = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i1> %pg,
66                                                                                             <vscale x 4 x i32> %base,
67                                                                                             i64 %offset)
68  ret <vscale x 4 x i32> %load
69}
70
71define <vscale x 2 x i64> @gld1w_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
72; CHECK-LABEL: gld1w_d_scalar_offset:
73; CHECK:       // %bb.0:
74; CHECK-NEXT:    ld1w { z0.d }, p0/z, [x0, z0.d]
75; CHECK-NEXT:    ret
76  %load = call <vscale x 2 x i32> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1> %pg,
77                                                                                             <vscale x 2 x i64> %base,
78                                                                                             i64 %offset)
79  %res = zext <vscale x 2 x i32> %load to <vscale x 2 x i64>
80  ret <vscale x 2 x i64> %res
81}
82
83define <vscale x 4 x float> @gld1w_s_scalar_offset_float(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {
84; CHECK-LABEL: gld1w_s_scalar_offset_float:
85; CHECK:       // %bb.0:
86; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0, z0.s, uxtw]
87; CHECK-NEXT:    ret
88  %load = call <vscale x 4 x float> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x i1> %pg,
89                                                                                               <vscale x 4 x i32> %base,
90                                                                                               i64 %offset)
91  ret <vscale x 4 x float> %load
92}
93
94; LD1D
95define <vscale x 2 x i64> @gld1d_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
96; CHECK-LABEL: gld1d_d_scalar_offset:
97; CHECK:       // %bb.0:
98; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0, z0.d]
99; CHECK-NEXT:    ret
100  %load = call <vscale x 2 x i64> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i1> %pg,
101                                                                                             <vscale x 2 x i64> %base,
102                                                                                             i64 %offset)
103  ret <vscale x 2 x i64> %load
104}
105
106define <vscale x 2 x double> @gld1d_d_scalar_offset_double(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
107; CHECK-LABEL: gld1d_d_scalar_offset_double:
108; CHECK:       // %bb.0:
109; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0, z0.d]
110; CHECK-NEXT:    ret
111  %load = call <vscale x 2 x double> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x i1> %pg,
112                                                                                                <vscale x 2 x i64> %base,
113                                                                                                i64 %offset)
114  ret <vscale x 2 x double> %load
115}
116
117; LD1SB, LD1SW, LD1SH: vector base + scalar offset (index)
118;   e.g. ld1b { z0.d }, p0/z, [x0, z0.d]
119;
120
121; LD1SB
122define <vscale x 4 x i32> @gld1sb_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {
123; CHECK-LABEL: gld1sb_s_scalar_offset:
124; CHECK:       // %bb.0:
125; CHECK-NEXT:    ld1sb { z0.s }, p0/z, [x0, z0.s, uxtw]
126; CHECK-NEXT:    ret
127  %load = call <vscale x 4 x i8> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1> %pg,
128                                                                                           <vscale x 4 x i32> %base,
129                                                                                           i64 %offset)
130  %res = sext <vscale x 4 x i8> %load to <vscale x 4 x i32>
131  ret <vscale x 4 x i32> %res
132}
133
134define <vscale x 2 x i64> @gld1sb_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
135; CHECK-LABEL: gld1sb_d_scalar_offset:
136; CHECK:       // %bb.0:
137; CHECK-NEXT:    ld1sb { z0.d }, p0/z, [x0, z0.d]
138; CHECK-NEXT:    ret
139  %load = call <vscale x 2 x i8> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1> %pg,
140                                                                                           <vscale x 2 x i64> %base,
141                                                                                           i64 %offset)
142  %res = sext <vscale x 2 x i8> %load to <vscale x 2 x i64>
143  ret <vscale x 2 x i64> %res
144}
145
146; LD1SH
147define <vscale x 4 x i32> @gld1sh_s_scalar_offset(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %base, i64 %offset) {
148; CHECK-LABEL: gld1sh_s_scalar_offset:
149; CHECK:       // %bb.0:
150; CHECK-NEXT:    ld1sh { z0.s }, p0/z, [x0, z0.s, uxtw]
151; CHECK-NEXT:    ret
152  %load = call <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1> %pg,
153                                                                                             <vscale x 4 x i32> %base,
154                                                                                             i64 %offset)
155  %res = sext <vscale x 4 x i16> %load to <vscale x 4 x i32>
156  ret <vscale x 4 x i32> %res
157}
158
159define <vscale x 2 x i64> @gld1sh_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
160; CHECK-LABEL: gld1sh_d_scalar_offset:
161; CHECK:       // %bb.0:
162; CHECK-NEXT:    ld1sh { z0.d }, p0/z, [x0, z0.d]
163; CHECK-NEXT:    ret
164  %load = call <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1> %pg,
165                                                                                             <vscale x 2 x i64> %base,
166                                                                                             i64 %offset)
167  %res = sext <vscale x 2 x i16> %load to <vscale x 2 x i64>
168  ret <vscale x 2 x i64> %res
169}
170
171; LD1SW
172define <vscale x 2 x i64> @gld1sw_d_scalar_offset(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %base, i64 %offset) {
173; CHECK-LABEL: gld1sw_d_scalar_offset:
174; CHECK:       // %bb.0:
175; CHECK-NEXT:    ld1sw { z0.d }, p0/z, [x0, z0.d]
176; CHECK-NEXT:    ret
177  %load = call <vscale x 2 x i32> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1> %pg,
178                                                                                             <vscale x 2 x i64> %base,
179                                                                                             i64 %offset)
180  %res = sext <vscale x 2 x i32> %load to <vscale x 2 x i64>
181  ret <vscale x 2 x i64> %res
182}
183
184; LD1B/LD1SB
185declare <vscale x 4 x i8> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i8.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)
186declare <vscale x 2 x i8> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i8.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)
187
188; LD1H/LD1SH
189declare <vscale x 4 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i16.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)
190declare <vscale x 2 x i16> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i16.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)
191
192; LD1W/LD1SW
193declare <vscale x 4 x i32> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4i32.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)
194declare <vscale x 2 x i32> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i32.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)
195
196declare <vscale x 4 x float> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv4f32.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, i64)
197
198; LD1D
199declare <vscale x 2 x i64> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2i64.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)
200
201declare <vscale x 2 x double> @llvm.aarch64.sve.ld1.gather.scalar.offset.nxv2f64.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, i64)
202