1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=arm-eabi -mattr=+neon | FileCheck %s
4 define <8 x i8> @vcges8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
7 ; CHECK-NEXT: vldr d16, [r1]
8 ; CHECK-NEXT: vldr d17, [r0]
9 ; CHECK-NEXT: vcge.s8 d16, d17, d16
10 ; CHECK-NEXT: vmov r0, r1, d16
11 ; CHECK-NEXT: mov pc, lr
12 %tmp1 = load <8 x i8>, <8 x i8>* %A
13 %tmp2 = load <8 x i8>, <8 x i8>* %B
14 %tmp3 = icmp sge <8 x i8> %tmp1, %tmp2
15 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8>
19 define <4 x i16> @vcges16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
20 ; CHECK-LABEL: vcges16:
22 ; CHECK-NEXT: vldr d16, [r1]
23 ; CHECK-NEXT: vldr d17, [r0]
24 ; CHECK-NEXT: vcge.s16 d16, d17, d16
25 ; CHECK-NEXT: vmov r0, r1, d16
26 ; CHECK-NEXT: mov pc, lr
27 %tmp1 = load <4 x i16>, <4 x i16>* %A
28 %tmp2 = load <4 x i16>, <4 x i16>* %B
29 %tmp3 = icmp sge <4 x i16> %tmp1, %tmp2
30 %tmp4 = sext <4 x i1> %tmp3 to <4 x i16>
34 define <2 x i32> @vcges32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
35 ; CHECK-LABEL: vcges32:
37 ; CHECK-NEXT: vldr d16, [r1]
38 ; CHECK-NEXT: vldr d17, [r0]
39 ; CHECK-NEXT: vcge.s32 d16, d17, d16
40 ; CHECK-NEXT: vmov r0, r1, d16
41 ; CHECK-NEXT: mov pc, lr
42 %tmp1 = load <2 x i32>, <2 x i32>* %A
43 %tmp2 = load <2 x i32>, <2 x i32>* %B
44 %tmp3 = icmp sge <2 x i32> %tmp1, %tmp2
45 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
49 define <8 x i8> @vcgeu8(<8 x i8>* %A, <8 x i8>* %B) nounwind {
50 ; CHECK-LABEL: vcgeu8:
52 ; CHECK-NEXT: vldr d16, [r1]
53 ; CHECK-NEXT: vldr d17, [r0]
54 ; CHECK-NEXT: vcge.u8 d16, d17, d16
55 ; CHECK-NEXT: vmov r0, r1, d16
56 ; CHECK-NEXT: mov pc, lr
57 %tmp1 = load <8 x i8>, <8 x i8>* %A
58 %tmp2 = load <8 x i8>, <8 x i8>* %B
59 %tmp3 = icmp uge <8 x i8> %tmp1, %tmp2
60 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8>
64 define <4 x i16> @vcgeu16(<4 x i16>* %A, <4 x i16>* %B) nounwind {
65 ; CHECK-LABEL: vcgeu16:
67 ; CHECK-NEXT: vldr d16, [r1]
68 ; CHECK-NEXT: vldr d17, [r0]
69 ; CHECK-NEXT: vcge.u16 d16, d17, d16
70 ; CHECK-NEXT: vmov r0, r1, d16
71 ; CHECK-NEXT: mov pc, lr
72 %tmp1 = load <4 x i16>, <4 x i16>* %A
73 %tmp2 = load <4 x i16>, <4 x i16>* %B
74 %tmp3 = icmp uge <4 x i16> %tmp1, %tmp2
75 %tmp4 = sext <4 x i1> %tmp3 to <4 x i16>
79 define <2 x i32> @vcgeu32(<2 x i32>* %A, <2 x i32>* %B) nounwind {
80 ; CHECK-LABEL: vcgeu32:
82 ; CHECK-NEXT: vldr d16, [r1]
83 ; CHECK-NEXT: vldr d17, [r0]
84 ; CHECK-NEXT: vcge.u32 d16, d17, d16
85 ; CHECK-NEXT: vmov r0, r1, d16
86 ; CHECK-NEXT: mov pc, lr
87 %tmp1 = load <2 x i32>, <2 x i32>* %A
88 %tmp2 = load <2 x i32>, <2 x i32>* %B
89 %tmp3 = icmp uge <2 x i32> %tmp1, %tmp2
90 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
94 define <2 x i32> @vcgef32(<2 x float>* %A, <2 x float>* %B) nounwind {
95 ; CHECK-LABEL: vcgef32:
97 ; CHECK-NEXT: vldr d16, [r1]
98 ; CHECK-NEXT: vldr d17, [r0]
99 ; CHECK-NEXT: vcge.f32 d16, d17, d16
100 ; CHECK-NEXT: vmov r0, r1, d16
101 ; CHECK-NEXT: mov pc, lr
102 %tmp1 = load <2 x float>, <2 x float>* %A
103 %tmp2 = load <2 x float>, <2 x float>* %B
104 %tmp3 = fcmp oge <2 x float> %tmp1, %tmp2
105 %tmp4 = sext <2 x i1> %tmp3 to <2 x i32>
109 define <16 x i8> @vcgeQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
110 ; CHECK-LABEL: vcgeQs8:
112 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
113 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
114 ; CHECK-NEXT: vcge.s8 q8, q9, q8
115 ; CHECK-NEXT: vmov r0, r1, d16
116 ; CHECK-NEXT: vmov r2, r3, d17
117 ; CHECK-NEXT: mov pc, lr
118 %tmp1 = load <16 x i8>, <16 x i8>* %A
119 %tmp2 = load <16 x i8>, <16 x i8>* %B
120 %tmp3 = icmp sge <16 x i8> %tmp1, %tmp2
121 %tmp4 = sext <16 x i1> %tmp3 to <16 x i8>
125 define <8 x i16> @vcgeQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
126 ; CHECK-LABEL: vcgeQs16:
128 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
129 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
130 ; CHECK-NEXT: vcge.s16 q8, q9, q8
131 ; CHECK-NEXT: vmov r0, r1, d16
132 ; CHECK-NEXT: vmov r2, r3, d17
133 ; CHECK-NEXT: mov pc, lr
134 %tmp1 = load <8 x i16>, <8 x i16>* %A
135 %tmp2 = load <8 x i16>, <8 x i16>* %B
136 %tmp3 = icmp sge <8 x i16> %tmp1, %tmp2
137 %tmp4 = sext <8 x i1> %tmp3 to <8 x i16>
141 define <4 x i32> @vcgeQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
142 ; CHECK-LABEL: vcgeQs32:
144 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
145 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
146 ; CHECK-NEXT: vcge.s32 q8, q9, q8
147 ; CHECK-NEXT: vmov r0, r1, d16
148 ; CHECK-NEXT: vmov r2, r3, d17
149 ; CHECK-NEXT: mov pc, lr
150 %tmp1 = load <4 x i32>, <4 x i32>* %A
151 %tmp2 = load <4 x i32>, <4 x i32>* %B
152 %tmp3 = icmp sge <4 x i32> %tmp1, %tmp2
153 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32>
157 define <16 x i8> @vcgeQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind {
158 ; CHECK-LABEL: vcgeQu8:
160 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
161 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
162 ; CHECK-NEXT: vcge.u8 q8, q9, q8
163 ; CHECK-NEXT: vmov r0, r1, d16
164 ; CHECK-NEXT: vmov r2, r3, d17
165 ; CHECK-NEXT: mov pc, lr
166 %tmp1 = load <16 x i8>, <16 x i8>* %A
167 %tmp2 = load <16 x i8>, <16 x i8>* %B
168 %tmp3 = icmp uge <16 x i8> %tmp1, %tmp2
169 %tmp4 = sext <16 x i1> %tmp3 to <16 x i8>
173 define <8 x i16> @vcgeQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind {
174 ; CHECK-LABEL: vcgeQu16:
176 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
177 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
178 ; CHECK-NEXT: vcge.u16 q8, q9, q8
179 ; CHECK-NEXT: vmov r0, r1, d16
180 ; CHECK-NEXT: vmov r2, r3, d17
181 ; CHECK-NEXT: mov pc, lr
182 %tmp1 = load <8 x i16>, <8 x i16>* %A
183 %tmp2 = load <8 x i16>, <8 x i16>* %B
184 %tmp3 = icmp uge <8 x i16> %tmp1, %tmp2
185 %tmp4 = sext <8 x i1> %tmp3 to <8 x i16>
189 define <4 x i32> @vcgeQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind {
190 ; CHECK-LABEL: vcgeQu32:
192 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
193 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
194 ; CHECK-NEXT: vcge.u32 q8, q9, q8
195 ; CHECK-NEXT: vmov r0, r1, d16
196 ; CHECK-NEXT: vmov r2, r3, d17
197 ; CHECK-NEXT: mov pc, lr
198 %tmp1 = load <4 x i32>, <4 x i32>* %A
199 %tmp2 = load <4 x i32>, <4 x i32>* %B
200 %tmp3 = icmp uge <4 x i32> %tmp1, %tmp2
201 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32>
205 define <4 x i32> @vcgeQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
206 ; CHECK-LABEL: vcgeQf32:
208 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
209 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
210 ; CHECK-NEXT: vcge.f32 q8, q9, q8
211 ; CHECK-NEXT: vmov r0, r1, d16
212 ; CHECK-NEXT: vmov r2, r3, d17
213 ; CHECK-NEXT: mov pc, lr
214 %tmp1 = load <4 x float>, <4 x float>* %A
215 %tmp2 = load <4 x float>, <4 x float>* %B
216 %tmp3 = fcmp oge <4 x float> %tmp1, %tmp2
217 %tmp4 = sext <4 x i1> %tmp3 to <4 x i32>
221 define <2 x i32> @vacgef32(<2 x float>* %A, <2 x float>* %B) nounwind {
222 ; CHECK-LABEL: vacgef32:
224 ; CHECK-NEXT: vldr d16, [r1]
225 ; CHECK-NEXT: vldr d17, [r0]
226 ; CHECK-NEXT: vacge.f32 d16, d17, d16
227 ; CHECK-NEXT: vmov r0, r1, d16
228 ; CHECK-NEXT: mov pc, lr
229 %tmp1 = load <2 x float>, <2 x float>* %A
230 %tmp2 = load <2 x float>, <2 x float>* %B
231 %tmp3 = call <2 x i32> @llvm.arm.neon.vacge.v2i32.v2f32(<2 x float> %tmp1, <2 x float> %tmp2)
235 define <4 x i32> @vacgeQf32(<4 x float>* %A, <4 x float>* %B) nounwind {
236 ; CHECK-LABEL: vacgeQf32:
238 ; CHECK-NEXT: vld1.64 {d16, d17}, [r1]
239 ; CHECK-NEXT: vld1.64 {d18, d19}, [r0]
240 ; CHECK-NEXT: vacge.f32 q8, q9, q8
241 ; CHECK-NEXT: vmov r0, r1, d16
242 ; CHECK-NEXT: vmov r2, r3, d17
243 ; CHECK-NEXT: mov pc, lr
244 %tmp1 = load <4 x float>, <4 x float>* %A
245 %tmp2 = load <4 x float>, <4 x float>* %B
246 %tmp3 = call <4 x i32> @llvm.arm.neon.vacge.v4i32.v4f32(<4 x float> %tmp1, <4 x float> %tmp2)
250 declare <2 x i32> @llvm.arm.neon.vacge.v2i32.v2f32(<2 x float>, <2 x float>) nounwind readnone
251 declare <4 x i32> @llvm.arm.neon.vacge.v4i32.v4f32(<4 x float>, <4 x float>) nounwind readnone
253 define <8 x i8> @vcgei8Z(<8 x i8>* %A) nounwind {
254 ; CHECK-LABEL: vcgei8Z:
256 ; CHECK-NEXT: vldr d16, [r0]
257 ; CHECK-NEXT: vcge.s8 d16, d16, #0
258 ; CHECK-NEXT: vmov r0, r1, d16
259 ; CHECK-NEXT: mov pc, lr
260 %tmp1 = load <8 x i8>, <8 x i8>* %A
261 %tmp3 = icmp sge <8 x i8> %tmp1, <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>
262 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8>
266 define <8 x i8> @vclei8Z(<8 x i8>* %A) nounwind {
267 ; CHECK-LABEL: vclei8Z:
269 ; CHECK-NEXT: vldr d16, [r0]
270 ; CHECK-NEXT: vcle.s8 d16, d16, #0
271 ; CHECK-NEXT: vmov r0, r1, d16
272 ; CHECK-NEXT: mov pc, lr
273 %tmp1 = load <8 x i8>, <8 x i8>* %A
274 %tmp3 = icmp sle <8 x i8> %tmp1, <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>
275 %tmp4 = sext <8 x i1> %tmp3 to <8 x i8>
280 ; Floating-point comparisons against zero produce results with integer
281 ; elements, not floating-point elements.
282 define void @test_vclez_fp(<4 x float>* %A) nounwind optsize {
283 ; CHECK-LABEL: test_vclez_fp:
284 ; CHECK: @ %bb.0: @ %entry
285 ; CHECK-NEXT: vld1.64 {d16, d17}, [r0]
286 ; CHECK-NEXT: vcle.f32 q8, q8, #0
287 ; CHECK-NEXT: vmovn.i32 d16, q8
288 ; CHECK-NEXT: vmov.i8 d17, #0x1
289 ; CHECK-NEXT: vuzp.8 d16, d18
290 ; CHECK-NEXT: vadd.i8 d16, d16, d17
291 ; CHECK-NEXT: vst1.8 {d16}, [r0]
293 %ld = load <4 x float>, <4 x float>* %A
294 %0 = fcmp ole <4 x float> %ld, zeroinitializer
295 %1 = sext <4 x i1> %0 to <4 x i16>
296 %2 = add <4 x i16> %1, zeroinitializer
297 %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
298 %4 = add <8 x i16> %3, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
299 %5 = trunc <8 x i16> %4 to <8 x i8>
300 tail call void @llvm.arm.neon.vst1.p0i8.v8i8(i8* undef, <8 x i8> %5, i32 1)
304 declare void @llvm.arm.neon.vst1.p0i8.v8i8(i8*, <8 x i8>, i32) nounwind