1 ; RUN: llc -aarch64-sve-vector-bits-min=128 -asm-verbose=0 < %s | FileCheck %s -check-prefix=NO_SVE
2 ; RUN: llc -aarch64-sve-vector-bits-min=256 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK
3 ; RUN: llc -aarch64-sve-vector-bits-min=384 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK
4 ; RUN: llc -aarch64-sve-vector-bits-min=512 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
5 ; RUN: llc -aarch64-sve-vector-bits-min=640 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
6 ; RUN: llc -aarch64-sve-vector-bits-min=768 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
7 ; RUN: llc -aarch64-sve-vector-bits-min=896 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512
8 ; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
9 ; RUN: llc -aarch64-sve-vector-bits-min=1152 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
10 ; RUN: llc -aarch64-sve-vector-bits-min=1280 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
11 ; RUN: llc -aarch64-sve-vector-bits-min=1408 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
12 ; RUN: llc -aarch64-sve-vector-bits-min=1536 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
13 ; RUN: llc -aarch64-sve-vector-bits-min=1664 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
14 ; RUN: llc -aarch64-sve-vector-bits-min=1792 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
15 ; RUN: llc -aarch64-sve-vector-bits-min=1920 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
16 ; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_GE_2048
18 target triple = "aarch64-unknown-linux-gnu"
20 ; Don't use SVE when its registers are no bigger than NEON.
27 define <16 x i8> @trunc_v16i16_v16i8(<16 x i16>* %in) #0 {
28 ; CHECK-LABEL: trunc_v16i16_v16i8:
29 ; CHECK: ptrue [[PG:p[0-9]+]].h, vl16
30 ; CHECK-NEXT: ld1h { [[A_HALFS:z[0-9]+]].h }, [[PG]]/z, [x0]
31 ; CHECK-NEXT: uzp1 z0.b, [[A_HALFS]].b, [[A_HALFS]].b
33 %a = load <16 x i16>, <16 x i16>* %in
34 %b = trunc <16 x i16> %a to <16 x i8>
38 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
39 define void @trunc_v32i16_v32i8(<32 x i16>* %in, <32 x i8>* %out) #0 {
40 ; CHECK-LABEL: trunc_v32i16_v32i8:
41 ; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl32
42 ; VBITS_GE_512: ld1h { [[A_HALFS:z[0-9]+]].h }, [[PG]]/z, [x0]
43 ; VBITS_GE_512: uzp1 [[A_BYTES:z[0-9]+]].b, [[A_HALFS]].b, [[A_HALFS]].b
44 ; VBITS_GE_512: add [[A_BYTES]].b, [[PG]]/m, [[A_BYTES]].b, [[A_BYTES]].b
45 %a = load <32 x i16>, <32 x i16>* %in
46 %b = trunc <32 x i16> %a to <32 x i8>
47 %c = add <32 x i8> %b, %b
48 store <32 x i8> %c, <32 x i8>* %out
52 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
53 define void @trunc_v64i16_v64i8(<64 x i16>* %in, <64 x i8>* %out) #0 {
54 ; CHECK-LABEL: trunc_v64i16_v64i8:
55 ; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl64
56 ; VBITS_GE_1024: ld1h { [[A_HALFS:z[0-9]+]].h }, [[PG]]/z, [x0]
57 ; VBITS_GE_1024: uzp1 [[A_BYTES:z[0-9]+]].b, [[A_HALFS]].b, [[A_HALFS]].b
58 ; VBITS_GE_1024: add [[A_BYTES]].b, [[PG]]/m, [[A_BYTES]].b, [[A_BYTES]].b
59 %a = load <64 x i16>, <64 x i16>* %in
60 %b = trunc <64 x i16> %a to <64 x i8>
61 %c = add <64 x i8> %b, %b
62 store <64 x i8> %c, <64 x i8>* %out
66 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
67 define void @trunc_v128i16_v128i8(<128 x i16>* %in, <128 x i8>* %out) #0 {
68 ; CHECK-LABEL: trunc_v128i16_v128i8:
69 ; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl128
70 ; VBITS_GE_2048: ld1h { [[A_HALFS:z[0-9]+]].h }, [[PG]]/z, [x0]
71 ; VBITS_GE_2048: uzp1 [[A_BYTES:z[0-9]+]].b, [[A_HALFS]].b, [[A_HALFS]].b
72 ; VBITS_GE_2048: add [[A_BYTES]].b, [[PG]]/m, [[A_BYTES]].b, [[A_BYTES]].b
73 %a = load <128 x i16>, <128 x i16>* %in
74 %b = trunc <128 x i16> %a to <128 x i8>
75 %c = add <128 x i8> %b, %b
76 store <128 x i8> %c, <128 x i8>* %out
84 define <8 x i8> @trunc_v8i32_v8i8(<8 x i32>* %in) #0 {
85 ; CHECK-LABEL: trunc_v8i32_v8i8:
86 ; CHECK: ptrue [[PG:p[0-9]+]].s, vl8
87 ; CHECK-NEXT: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
88 ; CHECK-NEXT: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
89 ; CHECK-NEXT: uzp1 z0.b, [[A_HALFS]].b, [[A_HALFS]].b
91 %a = load <8 x i32>, <8 x i32>* %in
92 %b = trunc <8 x i32> %a to <8 x i8>
96 define <16 x i8> @trunc_v16i32_v16i8(<16 x i32>* %in) #0 {
97 ; CHECK-LABEL: trunc_v16i32_v16i8:
98 ; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl16
99 ; VBITS_GE_512-NEXT: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
100 ; VBITS_GE_512-NEXT: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
101 ; VBITS_GE_512-NEXT: uzp1 z0.b, [[A_HALFS]].b, [[A_HALFS]].b
102 ; VBITS_GE_512-NEXT: ret
103 %a = load <16 x i32>, <16 x i32>* %in
104 %b = trunc <16 x i32> %a to <16 x i8>
108 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
109 define void @trunc_v32i32_v32i8(<32 x i32>* %in, <32 x i8>* %out) #0 {
110 ; CHECK-LABEL: trunc_v32i32_v32i8:
111 ; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].s, vl32
112 ; VBITS_GE_1024: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
113 ; VBITS_GE_1024: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
114 ; VBITS_GE_1024: uzp1 [[A_BYTES:z[0-9]+]].b, [[A_HALFS]].b, [[A_HALFS]].b
115 ; VBITS_GE_1024: add [[A_BYTES]].b, [[PG]]/m, [[A_BYTES]].b, [[A_BYTES]].b
116 %a = load <32 x i32>, <32 x i32>* %in
117 %b = trunc <32 x i32> %a to <32 x i8>
118 %c = add <32 x i8> %b, %b
119 store <32 x i8> %c, <32 x i8>* %out
123 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
124 define void @trunc_v64i32_v64i8(<64 x i32>* %in, <64 x i8>* %out) #0 {
125 ; CHECK-LABEL: trunc_v64i32_v64i8:
126 ; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].s, vl64
127 ; VBITS_GE_2048: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
128 ; VBITS_GE_2048: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
129 ; VBITS_GE_2048: uzp1 [[A_BYTES:z[0-9]+]].b, [[A_HALFS]].b, [[A_HALFS]].b
130 ; VBITS_GE_2048: add [[A_BYTES]].b, [[PG]]/m, [[A_BYTES]].b, [[A_BYTES]].b
131 %a = load <64 x i32>, <64 x i32>* %in
132 %b = trunc <64 x i32> %a to <64 x i8>
133 %c = add <64 x i8> %b, %b
134 store <64 x i8> %c, <64 x i8>* %out
139 ; truncate i32 -> i16
142 define <8 x i16> @trunc_v8i32_v8i16(<8 x i32>* %in) #0 {
143 ; CHECK-LABEL: trunc_v8i32_v8i16:
144 ; CHECK: ptrue [[PG:p[0-9]+]].s, vl8
145 ; CHECK-NEXT: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
146 ; CHECK-NEXT: uzp1 z0.h, [[A_WORDS]].h, [[A_WORDS]].h
148 %a = load <8 x i32>, <8 x i32>* %in
149 %b = trunc <8 x i32> %a to <8 x i16>
153 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
154 define void @trunc_v16i32_v16i16(<16 x i32>* %in, <16 x i16>* %out) #0 {
155 ; CHECK-LABEL: trunc_v16i32_v16i16:
156 ; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl16
157 ; VBITS_GE_512: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
158 ; VBITS_GE_512: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
159 ; VBITS_GE_512: add [[A_HALFS]].h, [[PG]]/m, [[A_HALFS]].h, [[A_HALFS]].h
160 %a = load <16 x i32>, <16 x i32>* %in
161 %b = trunc <16 x i32> %a to <16 x i16>
162 %c = add <16 x i16> %b, %b
163 store <16 x i16> %c, <16 x i16>* %out
167 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
168 define void @trunc_v32i32_v32i16(<32 x i32>* %in, <32 x i16>* %out) #0 {
169 ; CHECK-LABEL: trunc_v32i32_v32i16:
170 ; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].s, vl32
171 ; VBITS_GE_1024: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
172 ; VBITS_GE_1024: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
173 ; VBITS_GE_1024: add [[A_HALFS]].h, [[PG]]/m, [[A_HALFS]].h, [[A_HALFS]].h
174 %a = load <32 x i32>, <32 x i32>* %in
175 %b = trunc <32 x i32> %a to <32 x i16>
176 %c = add <32 x i16> %b, %b
177 store <32 x i16> %c, <32 x i16>* %out
181 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
182 define void @trunc_v64i32_v64i16(<64 x i32>* %in, <64 x i16>* %out) #0 {
183 ; CHECK-LABEL: trunc_v64i32_v64i16:
184 ; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].s, vl64
185 ; VBITS_GE_2048: ld1w { [[A_WORDS:z[0-9]+]].s }, [[PG]]/z, [x0]
186 ; VBITS_GE_2048: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
187 ; VBITS_GE_2048: add [[A_HALFS]].h, [[PG]]/m, [[A_HALFS]].h, [[A_HALFS]].h
188 %a = load <64 x i32>, <64 x i32>* %in
189 %b = trunc <64 x i32> %a to <64 x i16>
190 %c = add <64 x i16> %b, %b
191 store <64 x i16> %c, <64 x i16>* %out
199 ; NOTE: v4i8 is not legal so result i8 elements are held within i16 containers.
200 define <4 x i8> @trunc_v4i64_v4i8(<4 x i64>* %in) #0 {
201 ; CHECK-LABEL: trunc_v4i64_v4i8:
202 ; CHECK: ptrue [[PG:p[0-9]+]].d, vl4
203 ; CHECK-NEXT: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
204 ; CHECK-NEXT: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
205 ; CHECK-NEXT: uzp1 z0.h, [[A_WORDS]].h, [[A_WORDS]].h
207 %a = load <4 x i64>, <4 x i64>* %in
208 %b = trunc <4 x i64> %a to <4 x i8>
212 define <8 x i8> @trunc_v8i64_v8i8(<8 x i64>* %in) #0 {
213 ; CHECK-LABEL: trunc_v8i64_v8i8:
214 ; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl8
215 ; VBITS_GE_512-NEXT: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
216 ; VBITS_GE_512-NEXT: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
217 ; VBITS_GE_512-NEXT: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
218 ; VBITS_GE_512-NEXT: uzp1 z0.b, [[A_HALFS]].b, [[A_HALFS]].b
219 ; VBITS_GE_512-NEXT: ret
220 %a = load <8 x i64>, <8 x i64>* %in
221 %b = trunc <8 x i64> %a to <8 x i8>
225 define <16 x i8> @trunc_v16i64_v16i8(<16 x i64>* %in) #0 {
226 ; CHECK-LABEL: trunc_v16i64_v16i8:
227 ; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl16
228 ; VBITS_GE_1024-NEXT: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
229 ; VBITS_GE_1024-NEXT: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
230 ; VBITS_GE_1024-NEXT: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
231 ; VBITS_GE_1024-NEXT: uzp1 z0.b, [[A_HALFS]].b, [[A_HALFS]].b
232 ; VBITS_GE_1024-NEXT: ret
233 %a = load <16 x i64>, <16 x i64>* %in
234 %b = trunc <16 x i64> %a to <16 x i8>
238 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
239 define void @trunc_v32i64_v32i8(<32 x i64>* %in, <32 x i8>* %out) #0 {
240 ; CHECK-LABEL: trunc_v32i64_v32i8:
241 ; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl32
242 ; VBITS_GE_2048: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
243 ; VBITS_GE_2048: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
244 ; VBITS_GE_2048: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
245 ; VBITS_GE_2048: uzp1 [[A_BYTES:z[0-9]+]].b, [[A_HALFS]].b, [[A_HALFS]].b
246 ; VBITS_GE_2048: add [[A_BYTES]].b, [[PG]]/m, [[A_BYTES]].b, [[A_BYTES]].b
247 %a = load <32 x i64>, <32 x i64>* %in
248 %b = trunc <32 x i64> %a to <32 x i8>
249 %c = add <32 x i8> %b, %b
250 store <32 x i8> %c, <32 x i8>* %out
255 ; truncate i64 -> i16
258 define <4 x i16> @trunc_v4i64_v4i16(<4 x i64>* %in) #0 {
259 ; CHECK-LABEL: trunc_v4i64_v4i16:
260 ; CHECK: ptrue [[PG:p[0-9]+]].d, vl4
261 ; CHECK-NEXT: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
262 ; CHECK-NEXT: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
263 ; CHECK-NEXT: uzp1 z0.h, [[A_WORDS]].h, [[A_WORDS]].h
265 %a = load <4 x i64>, <4 x i64>* %in
266 %b = trunc <4 x i64> %a to <4 x i16>
270 define <8 x i16> @trunc_v8i64_v8i16(<8 x i64>* %in) #0 {
271 ; CHECK-LABEL: trunc_v8i64_v8i16:
272 ; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl8
273 ; VBITS_GE_512-NEXT: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
274 ; VBITS_GE_512-NEXT: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
275 ; VBITS_GE_512-NEXT: uzp1 z0.h, [[A_WORDS]].h, [[A_WORDS]].h
276 ; VBITS_GE_512-NEXT: ret
277 %a = load <8 x i64>, <8 x i64>* %in
278 %b = trunc <8 x i64> %a to <8 x i16>
282 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
283 define void @trunc_v16i64_v16i16(<16 x i64>* %in, <16 x i16>* %out) #0 {
284 ; CHECK-LABEL: trunc_v16i64_v16i16:
285 ; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl16
286 ; VBITS_GE_1024: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
287 ; VBITS_GE_1024: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
288 ; VBITS_GE_1024: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
289 ; VBITS_GE_1024: add [[A_HALFS]].h, [[PG]]/m, [[A_HALFS]].h, [[A_HALFS]].h
290 %a = load <16 x i64>, <16 x i64>* %in
291 %b = trunc <16 x i64> %a to <16 x i16>
292 %c = add <16 x i16> %b, %b
293 store <16 x i16> %c, <16 x i16>* %out
297 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
298 define void @trunc_v32i64_v32i16(<32 x i64>* %in, <32 x i16>* %out) #0 {
299 ; CHECK-LABEL: trunc_v32i64_v32i16:
300 ; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl32
301 ; VBITS_GE_2048: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
302 ; VBITS_GE_2048: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
303 ; VBITS_GE_2048: uzp1 [[A_HALFS:z[0-9]+]].h, [[A_WORDS]].h, [[A_WORDS]].h
304 ; VBITS_GE_2048: add [[A_HALFS]].h, [[PG]]/m, [[A_HALFS]].h, [[A_HALFS]].h
305 %a = load <32 x i64>, <32 x i64>* %in
306 %b = trunc <32 x i64> %a to <32 x i16>
307 %c = add <32 x i16> %b, %b
308 store <32 x i16> %c, <32 x i16>* %out
313 ; truncate i64 -> i32
316 define <4 x i32> @trunc_v4i64_v4i32(<4 x i64>* %in) #0 {
317 ; CHECK-LABEL: trunc_v4i64_v4i32:
318 ; CHECK: ptrue [[PG:p[0-9]+]].d, vl4
319 ; CHECK-NEXT: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
320 ; CHECK-NEXT: uzp1 z0.s, [[A_DWORDS]].s, [[A_DWORDS]].s
322 %a = load <4 x i64>, <4 x i64>* %in
323 %b = trunc <4 x i64> %a to <4 x i32>
327 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
328 define void @trunc_v8i64_v8i32(<8 x i64>* %in, <8 x i32>* %out) #0 {
329 ; CHECK-LABEL: trunc_v8i64_v8i32:
330 ; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl8
331 ; VBITS_GE_512: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
332 ; VBITS_GE_512: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
333 ; VBITS_GE_512: add [[A_WORDS]].s, [[PG]]/m, [[A_WORDS]].s, [[A_WORDS]].s
334 %a = load <8 x i64>, <8 x i64>* %in
335 %b = trunc <8 x i64> %a to <8 x i32>
336 %c = add <8 x i32> %b, %b
337 store <8 x i32> %c, <8 x i32>* %out
341 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
342 define void @trunc_v16i64_v16i32(<16 x i64>* %in, <16 x i32>* %out) #0 {
343 ; CHECK-LABEL: trunc_v16i64_v16i32:
344 ; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl16
345 ; VBITS_GE_1024: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
346 ; VBITS_GE_1024: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
347 ; VBITS_GE_1024: add [[A_WORDS]].s, [[PG]]/m, [[A_WORDS]].s, [[A_WORDS]].s
348 %a = load <16 x i64>, <16 x i64>* %in
349 %b = trunc <16 x i64> %a to <16 x i32>
350 %c = add <16 x i32> %b, %b
351 store <16 x i32> %c, <16 x i32>* %out
355 ; NOTE: Extra 'add' is to prevent the truncate being combined with the store.
356 define void @trunc_v32i64_v32i32(<32 x i64>* %in, <32 x i32>* %out) #0 {
357 ; CHECK-LABEL: trunc_v32i64_v32i32:
358 ; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl32
359 ; VBITS_GE_2048: ld1d { [[A_DWORDS:z[0-9]+]].d }, [[PG]]/z, [x0]
360 ; VBITS_GE_2048: uzp1 [[A_WORDS:z[0-9]+]].s, [[A_DWORDS]].s, [[A_DWORDS]].s
361 ; VBITS_GE_2048: add [[A_WORDS]].s, [[PG]]/m, [[A_WORDS]].s, [[A_WORDS]].s
362 %a = load <32 x i64>, <32 x i64>* %in
363 %b = trunc <32 x i64> %a to <32 x i32>
364 %c = add <32 x i32> %b, %b
365 store <32 x i32> %c, <32 x i32>* %out
369 attributes #0 = { nounwind "target-features"="+sve" }