1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=aarch64-- | FileCheck %s
3
4declare <1 x i8> @llvm.usub.sat.v1i8(<1 x i8>, <1 x i8>)
5declare <2 x i8> @llvm.usub.sat.v2i8(<2 x i8>, <2 x i8>)
6declare <4 x i8> @llvm.usub.sat.v4i8(<4 x i8>, <4 x i8>)
7declare <8 x i8> @llvm.usub.sat.v8i8(<8 x i8>, <8 x i8>)
8declare <12 x i8> @llvm.usub.sat.v12i8(<12 x i8>, <12 x i8>)
9declare <16 x i8> @llvm.usub.sat.v16i8(<16 x i8>, <16 x i8>)
10declare <32 x i8> @llvm.usub.sat.v32i8(<32 x i8>, <32 x i8>)
11declare <64 x i8> @llvm.usub.sat.v64i8(<64 x i8>, <64 x i8>)
12
13declare <1 x i16> @llvm.usub.sat.v1i16(<1 x i16>, <1 x i16>)
14declare <2 x i16> @llvm.usub.sat.v2i16(<2 x i16>, <2 x i16>)
15declare <4 x i16> @llvm.usub.sat.v4i16(<4 x i16>, <4 x i16>)
16declare <8 x i16> @llvm.usub.sat.v8i16(<8 x i16>, <8 x i16>)
17declare <12 x i16> @llvm.usub.sat.v12i16(<12 x i16>, <12 x i16>)
18declare <16 x i16> @llvm.usub.sat.v16i16(<16 x i16>, <16 x i16>)
19declare <32 x i16> @llvm.usub.sat.v32i16(<32 x i16>, <32 x i16>)
20
21declare <16 x i1> @llvm.usub.sat.v16i1(<16 x i1>, <16 x i1>)
22declare <16 x i4> @llvm.usub.sat.v16i4(<16 x i4>, <16 x i4>)
23
24declare <2 x i32> @llvm.usub.sat.v2i32(<2 x i32>, <2 x i32>)
25declare <4 x i32> @llvm.usub.sat.v4i32(<4 x i32>, <4 x i32>)
26declare <8 x i32> @llvm.usub.sat.v8i32(<8 x i32>, <8 x i32>)
27declare <16 x i32> @llvm.usub.sat.v16i32(<16 x i32>, <16 x i32>)
28declare <2 x i64> @llvm.usub.sat.v2i64(<2 x i64>, <2 x i64>)
29declare <4 x i64> @llvm.usub.sat.v4i64(<4 x i64>, <4 x i64>)
30declare <8 x i64> @llvm.usub.sat.v8i64(<8 x i64>, <8 x i64>)
31
32declare <4 x i24> @llvm.usub.sat.v4i24(<4 x i24>, <4 x i24>)
33declare <2 x i128> @llvm.usub.sat.v2i128(<2 x i128>, <2 x i128>)
34
35
36define <16 x i8> @v16i8(<16 x i8> %x, <16 x i8> %y) nounwind {
37; CHECK-LABEL: v16i8:
38; CHECK:       // %bb.0:
39; CHECK-NEXT:    uqsub v0.16b, v0.16b, v1.16b
40; CHECK-NEXT:    ret
41  %z = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> %x, <16 x i8> %y)
42  ret <16 x i8> %z
43}
44
45define <32 x i8> @v32i8(<32 x i8> %x, <32 x i8> %y) nounwind {
46; CHECK-LABEL: v32i8:
47; CHECK:       // %bb.0:
48; CHECK-NEXT:    uqsub v0.16b, v0.16b, v2.16b
49; CHECK-NEXT:    uqsub v1.16b, v1.16b, v3.16b
50; CHECK-NEXT:    ret
51  %z = call <32 x i8> @llvm.usub.sat.v32i8(<32 x i8> %x, <32 x i8> %y)
52  ret <32 x i8> %z
53}
54
55define <64 x i8> @v64i8(<64 x i8> %x, <64 x i8> %y) nounwind {
56; CHECK-LABEL: v64i8:
57; CHECK:       // %bb.0:
58; CHECK-NEXT:    uqsub v0.16b, v0.16b, v4.16b
59; CHECK-NEXT:    uqsub v1.16b, v1.16b, v5.16b
60; CHECK-NEXT:    uqsub v2.16b, v2.16b, v6.16b
61; CHECK-NEXT:    uqsub v3.16b, v3.16b, v7.16b
62; CHECK-NEXT:    ret
63  %z = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> %x, <64 x i8> %y)
64  ret <64 x i8> %z
65}
66
67define <8 x i16> @v8i16(<8 x i16> %x, <8 x i16> %y) nounwind {
68; CHECK-LABEL: v8i16:
69; CHECK:       // %bb.0:
70; CHECK-NEXT:    uqsub v0.8h, v0.8h, v1.8h
71; CHECK-NEXT:    ret
72  %z = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> %x, <8 x i16> %y)
73  ret <8 x i16> %z
74}
75
76define <16 x i16> @v16i16(<16 x i16> %x, <16 x i16> %y) nounwind {
77; CHECK-LABEL: v16i16:
78; CHECK:       // %bb.0:
79; CHECK-NEXT:    uqsub v0.8h, v0.8h, v2.8h
80; CHECK-NEXT:    uqsub v1.8h, v1.8h, v3.8h
81; CHECK-NEXT:    ret
82  %z = call <16 x i16> @llvm.usub.sat.v16i16(<16 x i16> %x, <16 x i16> %y)
83  ret <16 x i16> %z
84}
85
86define <32 x i16> @v32i16(<32 x i16> %x, <32 x i16> %y) nounwind {
87; CHECK-LABEL: v32i16:
88; CHECK:       // %bb.0:
89; CHECK-NEXT:    uqsub v0.8h, v0.8h, v4.8h
90; CHECK-NEXT:    uqsub v1.8h, v1.8h, v5.8h
91; CHECK-NEXT:    uqsub v2.8h, v2.8h, v6.8h
92; CHECK-NEXT:    uqsub v3.8h, v3.8h, v7.8h
93; CHECK-NEXT:    ret
94  %z = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> %x, <32 x i16> %y)
95  ret <32 x i16> %z
96}
97
98define void @v8i8(<8 x i8>* %px, <8 x i8>* %py, <8 x i8>* %pz) nounwind {
99; CHECK-LABEL: v8i8:
100; CHECK:       // %bb.0:
101; CHECK-NEXT:    ldr d0, [x0]
102; CHECK-NEXT:    ldr d1, [x1]
103; CHECK-NEXT:    uqsub v0.8b, v0.8b, v1.8b
104; CHECK-NEXT:    str d0, [x2]
105; CHECK-NEXT:    ret
106  %x = load <8 x i8>, <8 x i8>* %px
107  %y = load <8 x i8>, <8 x i8>* %py
108  %z = call <8 x i8> @llvm.usub.sat.v8i8(<8 x i8> %x, <8 x i8> %y)
109  store <8 x i8> %z, <8 x i8>* %pz
110  ret void
111}
112
113define void @v4i8(<4 x i8>* %px, <4 x i8>* %py, <4 x i8>* %pz) nounwind {
114; CHECK-LABEL: v4i8:
115; CHECK:       // %bb.0:
116; CHECK-NEXT:    ldrb w8, [x0]
117; CHECK-NEXT:    ldrb w9, [x1]
118; CHECK-NEXT:    ldrb w10, [x0, #1]
119; CHECK-NEXT:    ldrb w11, [x1, #1]
120; CHECK-NEXT:    fmov s0, w8
121; CHECK-NEXT:    fmov s1, w9
122; CHECK-NEXT:    ldrb w8, [x0, #2]
123; CHECK-NEXT:    ldrb w9, [x1, #2]
124; CHECK-NEXT:    mov v0.h[1], w10
125; CHECK-NEXT:    mov v1.h[1], w11
126; CHECK-NEXT:    ldrb w10, [x0, #3]
127; CHECK-NEXT:    ldrb w11, [x1, #3]
128; CHECK-NEXT:    mov v0.h[2], w8
129; CHECK-NEXT:    mov v1.h[2], w9
130; CHECK-NEXT:    mov v0.h[3], w10
131; CHECK-NEXT:    mov v1.h[3], w11
132; CHECK-NEXT:    shl v1.4h, v1.4h, #8
133; CHECK-NEXT:    shl v0.4h, v0.4h, #8
134; CHECK-NEXT:    uqsub v0.4h, v0.4h, v1.4h
135; CHECK-NEXT:    ushr v0.4h, v0.4h, #8
136; CHECK-NEXT:    xtn v0.8b, v0.8h
137; CHECK-NEXT:    str s0, [x2]
138; CHECK-NEXT:    ret
139  %x = load <4 x i8>, <4 x i8>* %px
140  %y = load <4 x i8>, <4 x i8>* %py
141  %z = call <4 x i8> @llvm.usub.sat.v4i8(<4 x i8> %x, <4 x i8> %y)
142  store <4 x i8> %z, <4 x i8>* %pz
143  ret void
144}
145
146define void @v2i8(<2 x i8>* %px, <2 x i8>* %py, <2 x i8>* %pz) nounwind {
147; CHECK-LABEL: v2i8:
148; CHECK:       // %bb.0:
149; CHECK-NEXT:    ldrb w8, [x0]
150; CHECK-NEXT:    ldrb w9, [x1]
151; CHECK-NEXT:    ldrb w10, [x0, #1]
152; CHECK-NEXT:    ldrb w11, [x1, #1]
153; CHECK-NEXT:    fmov s0, w8
154; CHECK-NEXT:    fmov s1, w9
155; CHECK-NEXT:    mov v0.s[1], w10
156; CHECK-NEXT:    mov v1.s[1], w11
157; CHECK-NEXT:    shl v1.2s, v1.2s, #24
158; CHECK-NEXT:    shl v0.2s, v0.2s, #24
159; CHECK-NEXT:    uqsub v0.2s, v0.2s, v1.2s
160; CHECK-NEXT:    ushr v0.2s, v0.2s, #24
161; CHECK-NEXT:    mov w8, v0.s[1]
162; CHECK-NEXT:    fmov w9, s0
163; CHECK-NEXT:    strb w8, [x2, #1]
164; CHECK-NEXT:    strb w9, [x2]
165; CHECK-NEXT:    ret
166  %x = load <2 x i8>, <2 x i8>* %px
167  %y = load <2 x i8>, <2 x i8>* %py
168  %z = call <2 x i8> @llvm.usub.sat.v2i8(<2 x i8> %x, <2 x i8> %y)
169  store <2 x i8> %z, <2 x i8>* %pz
170  ret void
171}
172
173define void @v4i16(<4 x i16>* %px, <4 x i16>* %py, <4 x i16>* %pz) nounwind {
174; CHECK-LABEL: v4i16:
175; CHECK:       // %bb.0:
176; CHECK-NEXT:    ldr d0, [x0]
177; CHECK-NEXT:    ldr d1, [x1]
178; CHECK-NEXT:    uqsub v0.4h, v0.4h, v1.4h
179; CHECK-NEXT:    str d0, [x2]
180; CHECK-NEXT:    ret
181  %x = load <4 x i16>, <4 x i16>* %px
182  %y = load <4 x i16>, <4 x i16>* %py
183  %z = call <4 x i16> @llvm.usub.sat.v4i16(<4 x i16> %x, <4 x i16> %y)
184  store <4 x i16> %z, <4 x i16>* %pz
185  ret void
186}
187
188define void @v2i16(<2 x i16>* %px, <2 x i16>* %py, <2 x i16>* %pz) nounwind {
189; CHECK-LABEL: v2i16:
190; CHECK:       // %bb.0:
191; CHECK-NEXT:    ldrh w8, [x0]
192; CHECK-NEXT:    ldrh w9, [x1]
193; CHECK-NEXT:    ldrh w10, [x0, #2]
194; CHECK-NEXT:    ldrh w11, [x1, #2]
195; CHECK-NEXT:    fmov s0, w8
196; CHECK-NEXT:    fmov s1, w9
197; CHECK-NEXT:    mov v0.s[1], w10
198; CHECK-NEXT:    mov v1.s[1], w11
199; CHECK-NEXT:    shl v1.2s, v1.2s, #16
200; CHECK-NEXT:    shl v0.2s, v0.2s, #16
201; CHECK-NEXT:    uqsub v0.2s, v0.2s, v1.2s
202; CHECK-NEXT:    ushr v0.2s, v0.2s, #16
203; CHECK-NEXT:    mov w8, v0.s[1]
204; CHECK-NEXT:    fmov w9, s0
205; CHECK-NEXT:    strh w8, [x2, #2]
206; CHECK-NEXT:    strh w9, [x2]
207; CHECK-NEXT:    ret
208  %x = load <2 x i16>, <2 x i16>* %px
209  %y = load <2 x i16>, <2 x i16>* %py
210  %z = call <2 x i16> @llvm.usub.sat.v2i16(<2 x i16> %x, <2 x i16> %y)
211  store <2 x i16> %z, <2 x i16>* %pz
212  ret void
213}
214
215define <12 x i8> @v12i8(<12 x i8> %x, <12 x i8> %y) nounwind {
216; CHECK-LABEL: v12i8:
217; CHECK:       // %bb.0:
218; CHECK-NEXT:    uqsub v0.16b, v0.16b, v1.16b
219; CHECK-NEXT:    ret
220  %z = call <12 x i8> @llvm.usub.sat.v12i8(<12 x i8> %x, <12 x i8> %y)
221  ret <12 x i8> %z
222}
223
224define void @v12i16(<12 x i16>* %px, <12 x i16>* %py, <12 x i16>* %pz) nounwind {
225; CHECK-LABEL: v12i16:
226; CHECK:       // %bb.0:
227; CHECK-NEXT:    ldp q0, q1, [x0]
228; CHECK-NEXT:    ldp q3, q2, [x1]
229; CHECK-NEXT:    uqsub v1.8h, v1.8h, v2.8h
230; CHECK-NEXT:    uqsub v0.8h, v0.8h, v3.8h
231; CHECK-NEXT:    str q0, [x2]
232; CHECK-NEXT:    str d1, [x2, #16]
233; CHECK-NEXT:    ret
234  %x = load <12 x i16>, <12 x i16>* %px
235  %y = load <12 x i16>, <12 x i16>* %py
236  %z = call <12 x i16> @llvm.usub.sat.v12i16(<12 x i16> %x, <12 x i16> %y)
237  store <12 x i16> %z, <12 x i16>* %pz
238  ret void
239}
240
241define void @v1i8(<1 x i8>* %px, <1 x i8>* %py, <1 x i8>* %pz) nounwind {
242; CHECK-LABEL: v1i8:
243; CHECK:       // %bb.0:
244; CHECK-NEXT:    ldr b0, [x0]
245; CHECK-NEXT:    ldr b1, [x1]
246; CHECK-NEXT:    uqsub v0.8b, v0.8b, v1.8b
247; CHECK-NEXT:    st1 { v0.b }[0], [x2]
248; CHECK-NEXT:    ret
249  %x = load <1 x i8>, <1 x i8>* %px
250  %y = load <1 x i8>, <1 x i8>* %py
251  %z = call <1 x i8> @llvm.usub.sat.v1i8(<1 x i8> %x, <1 x i8> %y)
252  store <1 x i8> %z, <1 x i8>* %pz
253  ret void
254}
255
256define void @v1i16(<1 x i16>* %px, <1 x i16>* %py, <1 x i16>* %pz) nounwind {
257; CHECK-LABEL: v1i16:
258; CHECK:       // %bb.0:
259; CHECK-NEXT:    ldr h0, [x0]
260; CHECK-NEXT:    ldr h1, [x1]
261; CHECK-NEXT:    uqsub v0.4h, v0.4h, v1.4h
262; CHECK-NEXT:    str h0, [x2]
263; CHECK-NEXT:    ret
264  %x = load <1 x i16>, <1 x i16>* %px
265  %y = load <1 x i16>, <1 x i16>* %py
266  %z = call <1 x i16> @llvm.usub.sat.v1i16(<1 x i16> %x, <1 x i16> %y)
267  store <1 x i16> %z, <1 x i16>* %pz
268  ret void
269}
270
271define <16 x i4> @v16i4(<16 x i4> %x, <16 x i4> %y) nounwind {
272; CHECK-LABEL: v16i4:
273; CHECK:       // %bb.0:
274; CHECK-NEXT:    movi v2.16b, #15
275; CHECK-NEXT:    and v0.16b, v0.16b, v2.16b
276; CHECK-NEXT:    and v1.16b, v1.16b, v2.16b
277; CHECK-NEXT:    shl v1.16b, v1.16b, #4
278; CHECK-NEXT:    shl v0.16b, v0.16b, #4
279; CHECK-NEXT:    uqsub v0.16b, v0.16b, v1.16b
280; CHECK-NEXT:    ushr v0.16b, v0.16b, #4
281; CHECK-NEXT:    ret
282  %z = call <16 x i4> @llvm.usub.sat.v16i4(<16 x i4> %x, <16 x i4> %y)
283  ret <16 x i4> %z
284}
285
286define <16 x i1> @v16i1(<16 x i1> %x, <16 x i1> %y) nounwind {
287; CHECK-LABEL: v16i1:
288; CHECK:       // %bb.0:
289; CHECK-NEXT:    movi v2.16b, #1
290; CHECK-NEXT:    and v0.16b, v0.16b, v2.16b
291; CHECK-NEXT:    and v1.16b, v1.16b, v2.16b
292; CHECK-NEXT:    shl v1.16b, v1.16b, #7
293; CHECK-NEXT:    shl v0.16b, v0.16b, #7
294; CHECK-NEXT:    uqsub v0.16b, v0.16b, v1.16b
295; CHECK-NEXT:    ushr v0.16b, v0.16b, #7
296; CHECK-NEXT:    ret
297  %z = call <16 x i1> @llvm.usub.sat.v16i1(<16 x i1> %x, <16 x i1> %y)
298  ret <16 x i1> %z
299}
300
301define <2 x i32> @v2i32(<2 x i32> %x, <2 x i32> %y) nounwind {
302; CHECK-LABEL: v2i32:
303; CHECK:       // %bb.0:
304; CHECK-NEXT:    uqsub v0.2s, v0.2s, v1.2s
305; CHECK-NEXT:    ret
306  %z = call <2 x i32> @llvm.usub.sat.v2i32(<2 x i32> %x, <2 x i32> %y)
307  ret <2 x i32> %z
308}
309
310define <4 x i32> @v4i32(<4 x i32> %x, <4 x i32> %y) nounwind {
311; CHECK-LABEL: v4i32:
312; CHECK:       // %bb.0:
313; CHECK-NEXT:    uqsub v0.4s, v0.4s, v1.4s
314; CHECK-NEXT:    ret
315  %z = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> %x, <4 x i32> %y)
316  ret <4 x i32> %z
317}
318
319define <8 x i32> @v8i32(<8 x i32> %x, <8 x i32> %y) nounwind {
320; CHECK-LABEL: v8i32:
321; CHECK:       // %bb.0:
322; CHECK-NEXT:    uqsub v0.4s, v0.4s, v2.4s
323; CHECK-NEXT:    uqsub v1.4s, v1.4s, v3.4s
324; CHECK-NEXT:    ret
325  %z = call <8 x i32> @llvm.usub.sat.v8i32(<8 x i32> %x, <8 x i32> %y)
326  ret <8 x i32> %z
327}
328
329define <16 x i32> @v16i32(<16 x i32> %x, <16 x i32> %y) nounwind {
330; CHECK-LABEL: v16i32:
331; CHECK:       // %bb.0:
332; CHECK-NEXT:    uqsub v0.4s, v0.4s, v4.4s
333; CHECK-NEXT:    uqsub v1.4s, v1.4s, v5.4s
334; CHECK-NEXT:    uqsub v2.4s, v2.4s, v6.4s
335; CHECK-NEXT:    uqsub v3.4s, v3.4s, v7.4s
336; CHECK-NEXT:    ret
337  %z = call <16 x i32> @llvm.usub.sat.v16i32(<16 x i32> %x, <16 x i32> %y)
338  ret <16 x i32> %z
339}
340
341define <2 x i64> @v2i64(<2 x i64> %x, <2 x i64> %y) nounwind {
342; CHECK-LABEL: v2i64:
343; CHECK:       // %bb.0:
344; CHECK-NEXT:    uqsub v0.2d, v0.2d, v1.2d
345; CHECK-NEXT:    ret
346  %z = call <2 x i64> @llvm.usub.sat.v2i64(<2 x i64> %x, <2 x i64> %y)
347  ret <2 x i64> %z
348}
349
350define <4 x i64> @v4i64(<4 x i64> %x, <4 x i64> %y) nounwind {
351; CHECK-LABEL: v4i64:
352; CHECK:       // %bb.0:
353; CHECK-NEXT:    uqsub v0.2d, v0.2d, v2.2d
354; CHECK-NEXT:    uqsub v1.2d, v1.2d, v3.2d
355; CHECK-NEXT:    ret
356  %z = call <4 x i64> @llvm.usub.sat.v4i64(<4 x i64> %x, <4 x i64> %y)
357  ret <4 x i64> %z
358}
359
360define <8 x i64> @v8i64(<8 x i64> %x, <8 x i64> %y) nounwind {
361; CHECK-LABEL: v8i64:
362; CHECK:       // %bb.0:
363; CHECK-NEXT:    uqsub v0.2d, v0.2d, v4.2d
364; CHECK-NEXT:    uqsub v1.2d, v1.2d, v5.2d
365; CHECK-NEXT:    uqsub v2.2d, v2.2d, v6.2d
366; CHECK-NEXT:    uqsub v3.2d, v3.2d, v7.2d
367; CHECK-NEXT:    ret
368  %z = call <8 x i64> @llvm.usub.sat.v8i64(<8 x i64> %x, <8 x i64> %y)
369  ret <8 x i64> %z
370}
371
372define <2 x i128> @v2i128(<2 x i128> %x, <2 x i128> %y) nounwind {
373; CHECK-LABEL: v2i128:
374; CHECK:       // %bb.0:
375; CHECK-NEXT:    subs x8, x2, x6
376; CHECK-NEXT:    sbcs x9, x3, x7
377; CHECK-NEXT:    cmp x8, x2
378; CHECK-NEXT:    cset w10, hi
379; CHECK-NEXT:    cmp x9, x3
380; CHECK-NEXT:    cset w11, hi
381; CHECK-NEXT:    csel w10, w10, w11, eq
382; CHECK-NEXT:    cmp w10, #0 // =0
383; CHECK-NEXT:    csel x3, xzr, x9, ne
384; CHECK-NEXT:    csel x2, xzr, x8, ne
385; CHECK-NEXT:    subs x8, x0, x4
386; CHECK-NEXT:    sbcs x9, x1, x5
387; CHECK-NEXT:    cmp x8, x0
388; CHECK-NEXT:    cset w10, hi
389; CHECK-NEXT:    cmp x9, x1
390; CHECK-NEXT:    cset w11, hi
391; CHECK-NEXT:    csel w10, w10, w11, eq
392; CHECK-NEXT:    cmp w10, #0 // =0
393; CHECK-NEXT:    csel x8, xzr, x8, ne
394; CHECK-NEXT:    csel x1, xzr, x9, ne
395; CHECK-NEXT:    fmov d0, x8
396; CHECK-NEXT:    mov v0.d[1], x1
397; CHECK-NEXT:    fmov x0, d0
398; CHECK-NEXT:    ret
399  %z = call <2 x i128> @llvm.usub.sat.v2i128(<2 x i128> %x, <2 x i128> %y)
400  ret <2 x i128> %z
401}
402