1; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s 2>%t | FileCheck %s
2; RUN: FileCheck --check-prefix=WARN --allow-empty %s <%t
3
4; If this check fails please read test/CodeGen/AArch64/README for instructions on how to resolve it.
5; WARN-NOT: warning
6
7;
8; ST1B
9;
10
11define void @st1b_upper_bound(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %a) {
12; CHECK-LABEL: st1b_upper_bound:
13; CHECK: st1b { z0.b }, p0, [x0, #7, mul vl]
14; CHECK-NEXT: ret
15  %base_scalable = bitcast i8* %a to <vscale x 16 x i8>*
16  %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %base_scalable, i64 7
17  %base_scalar = bitcast <vscale x 16 x i8>* %base to i8*
18  call void @llvm.aarch64.sve.st1.nxv16i8(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %base_scalar)
19  ret void
20}
21
22define void @st1b_inbound(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %a) {
23; CHECK-LABEL: st1b_inbound:
24; CHECK: st1b { z0.b }, p0, [x0, #1, mul vl]
25; CHECK-NEXT: ret
26  %base_scalable = bitcast i8* %a to <vscale x 16 x i8>*
27  %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %base_scalable, i64 1
28  %base_scalar = bitcast <vscale x 16 x i8>* %base to i8*
29  call void @llvm.aarch64.sve.st1.nxv16i8(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %base_scalar)
30  ret void
31}
32
33define void @st1b_lower_bound(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %a) {
34; CHECK-LABEL: st1b_lower_bound:
35; CHECK: st1b { z0.b }, p0, [x0, #-8, mul vl]
36; CHECK-NEXT: ret
37  %base_scalable = bitcast i8* %a to <vscale x 16 x i8>*
38  %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %base_scalable, i64 -8
39  %base_scalar = bitcast <vscale x 16 x i8>* %base to i8*
40  call void @llvm.aarch64.sve.st1.nxv16i8(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %base_scalar)
41  ret void
42}
43
44define void @st1b_out_of_upper_bound(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %a) {
45; CHECK-LABEL: st1b_out_of_upper_bound:
46; CHECK: rdvl x[[OFFSET:[0-9]+]], #8
47; CHECK: st1b { z0.b }, p0, [x0, x[[OFFSET]]]
48; CHECK-NEXT: ret
49  %base_scalable = bitcast i8* %a to <vscale x 16 x i8>*
50  %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %base_scalable, i64 8
51  %base_scalar = bitcast <vscale x 16 x i8>* %base to i8*
52  call void @llvm.aarch64.sve.st1.nxv16i8(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %base_scalar)
53  ret void
54}
55
56define void @st1b_out_of_lower_bound(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %a) {
57; CHECK-LABEL: st1b_out_of_lower_bound:
58; CHECK: rdvl x[[OFFSET:[0-9]+]], #-9
59; CHECK: st1b { z0.b }, p0, [x0, x[[OFFSET]]]
60; CHECK-NEXT: ret
61  %base_scalable = bitcast i8* %a to <vscale x 16 x i8>*
62  %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %base_scalable, i64 -9
63  %base_scalar = bitcast <vscale x 16 x i8>* %base to i8*
64  call void @llvm.aarch64.sve.st1.nxv16i8(<vscale x 16 x i8> %data, <vscale x 16 x i1> %pg, i8* %base_scalar)
65  ret void
66}
67
68define void @st1b_s_inbound(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i8* %a) {
69; CHECK-LABEL: st1b_s_inbound:
70; CHECK: st1b { z0.s }, p0, [x0, #7, mul vl]
71; CHECK-NEXT: ret
72  %base_scalable = bitcast i8* %a to <vscale x 4 x i8>*
73  %base = getelementptr <vscale x 4 x i8>, <vscale x 4 x i8>* %base_scalable, i64 7
74  %base_scalar = bitcast <vscale x 4 x i8>* %base to i8*
75  %trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i8>
76  call void @llvm.aarch64.sve.st1.nxv4i8(<vscale x 4 x i8> %trunc, <vscale x 4 x i1> %pg, i8* %base_scalar)
77  ret void
78}
79
80define void @st1b_h_inbound(<vscale x 8 x i16> %data, <vscale x 8 x i1> %pg, i8* %a) {
81; CHECK-LABEL: st1b_h_inbound:
82; CHECK: st1b { z0.h }, p0, [x0, #1, mul vl]
83; CHECK-NEXT: ret
84  %base_scalable = bitcast i8* %a to <vscale x 8 x i8>*
85  %base = getelementptr <vscale x 8 x i8>, <vscale x 8 x i8>* %base_scalable, i64 1
86  %base_scalar = bitcast <vscale x 8 x i8>* %base to i8*
87  %trunc = trunc <vscale x 8 x i16> %data to <vscale x 8 x i8>
88  call void @llvm.aarch64.sve.st1.nxv8i8(<vscale x 8 x i8> %trunc, <vscale x 8 x i1> %pg, i8* %base_scalar)
89  ret void
90}
91
92define void @st1b_d_inbound(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i8* %a) {
93; CHECK-LABEL: st1b_d_inbound:
94; CHECK: st1b { z0.d }, p0, [x0, #-7, mul vl]
95; CHECK-NEXT: ret
96  %base_scalable = bitcast i8* %a to <vscale x 2 x i8>*
97  %base = getelementptr <vscale x 2 x i8>, <vscale x 2 x i8>* %base_scalable, i64 -7
98  %base_scalar = bitcast <vscale x 2 x i8>* %base to i8*
99  %trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i8>
100  call void @llvm.aarch64.sve.st1.nxv2i8(<vscale x 2 x i8> %trunc, <vscale x 2 x i1> %pg, i8* %base_scalar)
101  ret void
102}
103
104;
105; ST1H
106;
107
108define void @st1h_inbound(<vscale x 8 x i16> %data, <vscale x 8 x i1> %pg, i16* %a) {
109; CHECK-LABEL: st1h_inbound:
110; CHECK: st1h { z0.h }, p0, [x0, #-1, mul vl]
111; CHECK-NEXT: ret
112  %base_scalable = bitcast i16* %a to <vscale x 8 x i16>*
113  %base = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* %base_scalable, i64 -1
114  %base_scalar = bitcast <vscale x 8 x i16>* %base to i16*
115  call void @llvm.aarch64.sve.st1.nxv8i16(<vscale x 8 x i16> %data, <vscale x 8 x i1> %pg, i16* %base_scalar)
116  ret void
117}
118
119define void @st1h_f16_inbound(<vscale x 8 x half> %data, <vscale x 8 x i1> %pg, half* %a) {
120; CHECK-LABEL: st1h_f16_inbound:
121; CHECK: st1h { z0.h }, p0, [x0, #-5, mul vl]
122; CHECK-NEXT: ret
123  %base_scalable = bitcast half* %a to <vscale x 8 x half>*
124  %base = getelementptr <vscale x 8 x half>, <vscale x 8 x half>* %base_scalable, i64 -5
125  %base_scalar = bitcast <vscale x 8 x half>* %base to half*
126  call void @llvm.aarch64.sve.st1.nxv8f16(<vscale x 8 x half> %data, <vscale x 8 x i1> %pg, half* %base_scalar)
127  ret void
128}
129
130define void @st1h_bf16_inbound(<vscale x 8 x bfloat> %data, <vscale x 8 x i1> %pg, bfloat* %a) #0 {
131; CHECK-LABEL: st1h_bf16_inbound:
132; CHECK: st1h { z0.h }, p0, [x0, #-5, mul vl]
133; CHECK-NEXT: ret
134  %base_scalable = bitcast bfloat* %a to <vscale x 8 x bfloat>*
135  %base = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* %base_scalable, i64 -5
136  %base_scalar = bitcast <vscale x 8 x bfloat>* %base to bfloat*
137  call void @llvm.aarch64.sve.st1.nxv8bf16(<vscale x 8 x bfloat> %data, <vscale x 8 x i1> %pg, bfloat* %base_scalar)
138  ret void
139}
140
141define void @st1h_s_inbound(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i16* %a) {
142; CHECK-LABEL: st1h_s_inbound:
143; CHECK: st1h { z0.s }, p0, [x0, #2, mul vl]
144; CHECK-NEXT: ret
145  %base_scalable = bitcast i16* %a to <vscale x 4 x i16>*
146  %base = getelementptr <vscale x 4 x i16>, <vscale x 4 x i16>* %base_scalable, i64 2
147  %base_scalar = bitcast <vscale x 4 x i16>* %base to i16*
148  %trunc = trunc <vscale x 4 x i32> %data to <vscale x 4 x i16>
149  call void @llvm.aarch64.sve.st1.nxv4i16(<vscale x 4 x i16> %trunc, <vscale x 4 x i1> %pg, i16* %base_scalar)
150  ret void
151}
152
153define void @st1h_d_inbound(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i16* %a) {
154; CHECK-LABEL: st1h_d_inbound:
155; CHECK: st1h { z0.d }, p0, [x0, #-4, mul vl]
156; CHECK-NEXT: ret
157  %base_scalable = bitcast i16* %a to <vscale x 2 x i16>*
158  %base = getelementptr <vscale x 2 x i16>, <vscale x 2 x i16>* %base_scalable, i64 -4
159  %base_scalar = bitcast <vscale x 2 x i16>* %base to i16*
160  %trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i16>
161  call void @llvm.aarch64.sve.st1.nxv2i16(<vscale x 2 x i16> %trunc, <vscale x 2 x i1> %pg, i16* %base_scalar)
162  ret void
163}
164
165;
166; ST1W
167;
168
169define void @st1w_inbound(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i32* %a) {
170; CHECK-LABEL: st1w_inbound:
171; CHECK: st1w { z0.s }, p0, [x0, #6, mul vl]
172; CHECK-NEXT: ret
173  %base_scalable = bitcast i32* %a to <vscale x 4 x i32>*
174  %base = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* %base_scalable, i64 6
175  %base_scalar = bitcast <vscale x 4 x i32>* %base to i32*
176  call void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32> %data, <vscale x 4 x i1> %pg, i32* %base_scalar)
177  ret void
178}
179
180define void @st1w_f32_inbound(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, float* %a) {
181; CHECK-LABEL: st1w_f32_inbound:
182; CHECK: st1w { z0.s }, p0, [x0, #-1, mul vl]
183; CHECK-NEXT: ret
184  %base_scalable = bitcast float* %a to <vscale x 4 x float>*
185  %base = getelementptr <vscale x 4 x float>, <vscale x 4 x float>* %base_scalable, i64 -1
186  %base_scalar = bitcast <vscale x 4 x float>* %base to float*
187  call void @llvm.aarch64.sve.st1.nxv4f32(<vscale x 4 x float> %data, <vscale x 4 x i1> %pg, float* %base_scalar)
188  ret void
189}
190
191define void @st1w_d_inbound(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i32* %a) {
192; CHECK-LABEL: st1w_d_inbound:
193; CHECK: st1w { z0.d }, p0, [x0, #1, mul vl]
194; CHECK-NEXT: ret
195  %base_scalable = bitcast i32* %a to <vscale x 2 x i32>*
196  %base = getelementptr <vscale x 2 x i32>, <vscale x 2 x i32>* %base_scalable, i64 1
197  %base_scalar = bitcast <vscale x 2 x i32>* %base to i32*
198  %trunc = trunc <vscale x 2 x i64> %data to <vscale x 2 x i32>
199  call void @llvm.aarch64.sve.st1.nxv2i32(<vscale x 2 x i32> %trunc, <vscale x 2 x i1> %pg, i32* %base_scalar)
200  ret void
201}
202
203;
204; ST1D
205;
206
207define void @st1d_inbound(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %a) {
208; CHECK-LABEL: st1d_inbound:
209; CHECK: st1d { z0.d }, p0, [x0, #5, mul vl]
210; CHECK-NEXT: ret
211  %base_scalable = bitcast i64* %a to <vscale x 2 x i64>*
212  %base = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* %base_scalable, i64 5
213  %base_scalar = bitcast <vscale x 2 x i64>* %base to i64*
214  call void @llvm.aarch64.sve.st1.nxv2i64(<vscale x 2 x i64> %data, <vscale x 2 x i1> %pg, i64* %base_scalar)
215  ret void
216}
217
218define void @st1d_f64_inbound(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %a) {
219; CHECK-LABEL: st1d_f64_inbound:
220; CHECK: st1d { z0.d }, p0, [x0, #-8, mul vl]
221; CHECK-NEXT: ret
222  %base_scalable = bitcast double* %a to <vscale x 2 x double>*
223  %base = getelementptr <vscale x 2 x double>, <vscale x 2 x double>* %base_scalable, i64 -8
224  %base_scalar = bitcast <vscale x 2 x double>* %base to double*
225  call void @llvm.aarch64.sve.st1.nxv2f64(<vscale x 2 x double> %data, <vscale x 2 x i1> %pg, double* %base_scalar)
226  ret void
227}
228
229declare void @llvm.aarch64.sve.st1.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i1>, i8*)
230
231declare void @llvm.aarch64.sve.st1.nxv8i8(<vscale x 8 x i8>, <vscale x 8 x i1>, i8*)
232declare void @llvm.aarch64.sve.st1.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i1>, i16*)
233declare void @llvm.aarch64.sve.st1.nxv8f16(<vscale x 8 x half>, <vscale x 8 x i1>, half*)
234declare void @llvm.aarch64.sve.st1.nxv8bf16(<vscale x 8 x bfloat>, <vscale x 8 x i1>, bfloat*)
235
236declare void @llvm.aarch64.sve.st1.nxv4i8(<vscale x 4 x i8>, <vscale x 4 x i1>, i8*)
237declare void @llvm.aarch64.sve.st1.nxv4i16(<vscale x 4 x i16>, <vscale x 4 x i1>, i16*)
238declare void @llvm.aarch64.sve.st1.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i1>, i32*)
239declare void @llvm.aarch64.sve.st1.nxv4f32(<vscale x 4 x float>, <vscale x 4 x i1>, float*)
240
241declare void @llvm.aarch64.sve.st1.nxv2i8(<vscale x 2 x i8>, <vscale x 2 x i1>, i8*)
242declare void @llvm.aarch64.sve.st1.nxv2i16(<vscale x 2 x i16>, <vscale x 2 x i1>, i16*)
243declare void @llvm.aarch64.sve.st1.nxv2i32(<vscale x 2 x i32>, <vscale x 2 x i1>, i32*)
244declare void @llvm.aarch64.sve.st1.nxv2i64(<vscale x 2 x i64>, <vscale x 2 x i1>, i64*)
245declare void @llvm.aarch64.sve.st1.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double*)
246
247; +bf16 is required for the bfloat version.
248attributes #0 = { "target-features"="+sve,+bf16" }
249