1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-LE
; RUN: llc -mtriple=thumbebv8.1m.main-none-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-BE
define void @load_load_add_store(<4 x i32> *%src1, <4 x i32> *%src2) {
; CHECK-LABEL: load_load_add_store:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: vldrw.u32 q0, [r1]
; CHECK-NEXT: vldrw.u32 q1, [r0]
; CHECK-NEXT: vadd.i32 q0, q1, q0
; CHECK-NEXT: vstrw.32 q0, [r0]
; CHECK-NEXT: bx lr
entry:
%l1 = load <4 x i32>, <4 x i32>* %src1, align 4
%l2 = load <4 x i32>, <4 x i32>* %src2, align 4
%a = add <4 x i32> %l1, %l2
store <4 x i32> %a, <4 x i32>* %src1, align 4
ret void
}
define void @load_load_add_store_align1(<4 x i32> *%src1, <4 x i32> *%src2) {
; CHECK-LE-LABEL: load_load_add_store_align1:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vldrb.u8 q0, [r1]
; CHECK-LE-NEXT: vldrb.u8 q1, [r0]
; CHECK-LE-NEXT: vadd.i32 q0, q1, q0
; CHECK-LE-NEXT: vstrb.8 q0, [r0]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_load_add_store_align1:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vldrb.u8 q0, [r1]
; CHECK-BE-NEXT: vldrb.u8 q1, [r0]
; CHECK-BE-NEXT: vrev32.8 q0, q0
; CHECK-BE-NEXT: vrev32.8 q1, q1
; CHECK-BE-NEXT: vadd.i32 q0, q1, q0
; CHECK-BE-NEXT: vrev32.8 q0, q0
; CHECK-BE-NEXT: vstrb.8 q0, [r0]
; CHECK-BE-NEXT: bx lr
entry:
%l1 = load <4 x i32>, <4 x i32>* %src1, align 1
%l2 = load <4 x i32>, <4 x i32>* %src2, align 1
%a = add <4 x i32> %l1, %l2
store <4 x i32> %a, <4 x i32>* %src1, align 1
ret void
}
define arm_aapcs_vfpcc void @load_arg_add_store(<4 x i32> *%src1, <4 x i32> %src2) {
; CHECK-LE-LABEL: load_arg_add_store:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vldrw.u32 q1, [r0]
; CHECK-LE-NEXT: vadd.i32 q0, q1, q0
; CHECK-LE-NEXT: vstrw.32 q0, [r0]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: load_arg_add_store:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vldrw.u32 q0, [r0]
; CHECK-BE-NEXT: vadd.i32 q0, q0, q1
; CHECK-BE-NEXT: vstrw.32 q0, [r0]
; CHECK-BE-NEXT: bx lr
entry:
%l1 = load <4 x i32>, <4 x i32>* %src1, align 4
%a = add <4 x i32> %l1, %src2
store <4 x i32> %a, <4 x i32>* %src1, align 4
ret void
}
define <4 x i32> @add_soft(<4 x i32> %src1, <4 x i32> %src2) {
; CHECK-LE-LABEL: add_soft:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vmov d0, r0, r1
; CHECK-LE-NEXT: mov r0, sp
; CHECK-LE-NEXT: vldrw.u32 q1, [r0]
; CHECK-LE-NEXT: vmov d1, r2, r3
; CHECK-LE-NEXT: vadd.i32 q0, q0, q1
; CHECK-LE-NEXT: vmov r0, r1, d0
; CHECK-LE-NEXT: vmov r2, r3, d1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: add_soft:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vmov d0, r1, r0
; CHECK-BE-NEXT: mov r0, sp
; CHECK-BE-NEXT: vmov d1, r3, r2
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vldrw.u32 q0, [r0]
; CHECK-BE-NEXT: vadd.i32 q0, q1, q0
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov r1, r0, d2
; CHECK-BE-NEXT: vmov r3, r2, d3
; CHECK-BE-NEXT: bx lr
entry:
%0 = add <4 x i32> %src1, %src2
ret <4 x i32> %0
}
define arm_aapcs_vfpcc <4 x i32> @add_hard(<4 x i32> %src1, <4 x i32> %src2) {
; CHECK-LE-LABEL: add_hard:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vadd.i32 q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: add_hard:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.32 q2, q1
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vadd.i32 q1, q1, q2
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%0 = add <4 x i32> %src1, %src2
ret <4 x i32> %0
}
define <4 x i32> @call_soft(<4 x i32> %src1, <4 x i32> %src2) {
; CHECK-LE-LABEL: call_soft:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: .save {r7, lr}
; CHECK-LE-NEXT: push {r7, lr}
; CHECK-LE-NEXT: .pad #16
; CHECK-LE-NEXT: sub sp, #16
; CHECK-LE-NEXT: add.w r12, sp, #24
; CHECK-LE-NEXT: vldrw.u32 q0, [r12]
; CHECK-LE-NEXT: vstrw.32 q0, [sp]
; CHECK-LE-NEXT: vmov d1, r2, r3
; CHECK-LE-NEXT: vmov d0, r0, r1
; CHECK-LE-NEXT: vshr.u32 q0, q0, #1
; CHECK-LE-NEXT: vmov r0, r1, d0
; CHECK-LE-NEXT: vmov r2, r3, d1
; CHECK-LE-NEXT: bl add_soft
; CHECK-LE-NEXT: vmov d1, r2, r3
; CHECK-LE-NEXT: vmov d0, r0, r1
; CHECK-LE-NEXT: vshr.u32 q0, q0, #1
; CHECK-LE-NEXT: vmov r0, r1, d0
; CHECK-LE-NEXT: vmov r2, r3, d1
; CHECK-LE-NEXT: add sp, #16
; CHECK-LE-NEXT: pop {r7, pc}
;
; CHECK-BE-LABEL: call_soft:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: .save {r7, lr}
; CHECK-BE-NEXT: push {r7, lr}
; CHECK-BE-NEXT: .pad #16
; CHECK-BE-NEXT: sub sp, #16
; CHECK-BE-NEXT: add.w r12, sp, #24
; CHECK-BE-NEXT: vldrw.u32 q0, [r12]
; CHECK-BE-NEXT: vstrw.32 q0, [sp]
; CHECK-BE-NEXT: vmov d1, r3, r2
; CHECK-BE-NEXT: vmov d0, r1, r0
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vshr.u32 q0, q1, #1
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov r1, r0, d2
; CHECK-BE-NEXT: vmov r3, r2, d3
; CHECK-BE-NEXT: bl add_soft
; CHECK-BE-NEXT: vmov d1, r3, r2
; CHECK-BE-NEXT: vmov d0, r1, r0
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vshr.u32 q0, q1, #1
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov r1, r0, d2
; CHECK-BE-NEXT: vmov r3, r2, d3
; CHECK-BE-NEXT: add sp, #16
; CHECK-BE-NEXT: pop {r7, pc}
entry:
%0 = lshr <4 x i32> %src1, <i32 1, i32 1, i32 1, i32 1>
%1 = call <4 x i32> @add_soft(<4 x i32> %0, <4 x i32> %src2)
%2 = lshr <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
ret <4 x i32> %2
}
define arm_aapcs_vfpcc <4 x i32> @call_hard(<4 x i32> %src1, <4 x i32> %src2) {
; CHECK-LE-LABEL: call_hard:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: .save {r7, lr}
; CHECK-LE-NEXT: push {r7, lr}
; CHECK-LE-NEXT: vshr.u32 q0, q0, #1
; CHECK-LE-NEXT: bl add_hard
; CHECK-LE-NEXT: vshr.u32 q0, q0, #1
; CHECK-LE-NEXT: pop {r7, pc}
;
; CHECK-BE-LABEL: call_hard:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: .save {r7, lr}
; CHECK-BE-NEXT: push {r7, lr}
; CHECK-BE-NEXT: vrev64.32 q2, q0
; CHECK-BE-NEXT: vshr.u32 q2, q2, #1
; CHECK-BE-NEXT: vrev64.32 q0, q2
; CHECK-BE-NEXT: bl add_hard
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vshr.u32 q1, q1, #1
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: pop {r7, pc}
entry:
%0 = lshr <4 x i32> %src1, <i32 1, i32 1, i32 1, i32 1>
%1 = call arm_aapcs_vfpcc <4 x i32> @add_hard(<4 x i32> %0, <4 x i32> %src2)
%2 = lshr <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
ret <4 x i32> %2
}
define arm_aapcs_vfpcc <16 x i8> @and_v4i32(<4 x i32> %src) {
; CHECK-LE-LABEL: and_v4i32:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vmov.i32 q1, #0x1
; CHECK-LE-NEXT: vand q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: and_v4i32:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.32 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x1
; CHECK-BE-NEXT: vand q1, q1, q0
; CHECK-BE-NEXT: vrev64.32 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%s1 = and <4 x i32> %src, <i32 1, i32 1, i32 1, i32 1>
%r = bitcast <4 x i32> %s1 to <16 x i8>
ret <16 x i8> %r
}
; Should be the same as and_v4i32 for LE
define arm_aapcs_vfpcc <16 x i8> @and_v16i8_le(<4 x i32> %src) {
; CHECK-LE-LABEL: and_v16i8_le:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vmov.i32 q1, #0x1
; CHECK-LE-NEXT: vand q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: and_v16i8_le:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.8 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x1
; CHECK-BE-NEXT: vrev32.8 q0, q0
; CHECK-BE-NEXT: vand q1, q1, q0
; CHECK-BE-NEXT: vrev64.8 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%0 = bitcast <4 x i32> %src to <16 x i8>
%r = and <16 x i8> %0, <i8 1, i8 0, i8 0, i8 0, i8 1, i8 0, i8 0, i8 0, i8 1, i8 0, i8 0, i8 0, i8 1, i8 0, i8 0, i8 0>
ret <16 x i8> %r
}
; Should be the same (or at least equivalent) as and_v4i32 for BE
define arm_aapcs_vfpcc <16 x i8> @and_v16i8_be(<4 x i32> %src) {
; CHECK-LE-LABEL: and_v16i8_be:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vmov.i32 q1, #0x1000000
; CHECK-LE-NEXT: vand q0, q0, q1
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: and_v16i8_be:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vrev64.8 q1, q0
; CHECK-BE-NEXT: vmov.i32 q0, #0x1000000
; CHECK-BE-NEXT: vrev32.8 q0, q0
; CHECK-BE-NEXT: vand q1, q1, q0
; CHECK-BE-NEXT: vrev64.8 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%0 = bitcast <4 x i32> %src to <16 x i8>
%r = and <16 x i8> %0, <i8 0, i8 0, i8 0, i8 1, i8 0, i8 0, i8 0, i8 1, i8 0, i8 0, i8 0, i8 1, i8 0, i8 0, i8 0, i8 1>
ret <16 x i8> %r
}
; FIXME: This looks wrong
define arm_aapcs_vfpcc <4 x i32> @test(i32* %data) {
; CHECK-LE-LABEL: test:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vldrw.u32 q0, [r0, #32]
; CHECK-LE-NEXT: movs r0, #1
; CHECK-LE-NEXT: vadd.i32 q1, q0, r0
; CHECK-LE-NEXT: @APP
; CHECK-LE-NEXT: vmullb.s32 q0, q1, q1
; CHECK-LE-NEXT: @NO_APP
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: test:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: movs r1, #1
; CHECK-BE-NEXT: vldrw.u32 q1, [r0, #32]
; CHECK-BE-NEXT: vdup.32 q0, r1
; CHECK-BE-NEXT: vadd.i32 q0, q1, q0
; CHECK-BE-NEXT: vrev32.8 q0, q0
; CHECK-BE-NEXT: @APP
; CHECK-BE-NEXT: vmullb.s32 q1, q0, q0
; CHECK-BE-NEXT: @NO_APP
; CHECK-BE-NEXT: vrev64.8 q0, q1
; CHECK-BE-NEXT: bx lr
entry:
%add.ptr = getelementptr inbounds i32, i32* %data, i32 8
%0 = bitcast i32* %add.ptr to <4 x i32>*
%1 = load <4 x i32>, <4 x i32>* %0, align 4
%2 = add <4 x i32> %1, <i32 1, i32 1, i32 1, i32 1>
%3 = tail call <4 x i32> asm sideeffect " VMULLB.s32 $0, $1, $1", "=&w,w"(<4 x i32> %2) #2
ret <4 x i32> %3
}
; Test case demonstrating that 'bitcast' reinterprets the memory format of a
; vector, as if stored and then loaded. So if it has to go between two
; operations treating a register as having different lane sizes, then in
; big-endian mode, it has to emit a vrev32.16, which is equivalent to the
; effect that vstrw.32 followed by vldrh.16 would have.
define arm_aapcs_vfpcc void @test_bitcast(<4 x i32>* readonly %in, <8 x i16>* %out) {
; CHECK-LE-LABEL: test_bitcast:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vldrw.u32 q0, [r0]
; CHECK-LE-NEXT: vmul.i32 q0, q0, q0
; CHECK-LE-NEXT: vmul.i16 q0, q0, q0
; CHECK-LE-NEXT: vstrw.32 q0, [r1]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: test_bitcast:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vldrw.u32 q0, [r0]
; CHECK-BE-NEXT: vmul.i32 q0, q0, q0
; CHECK-BE-NEXT: vrev32.16 q0, q0
; CHECK-BE-NEXT: vmul.i16 q0, q0, q0
; CHECK-BE-NEXT: vstrh.16 q0, [r1]
; CHECK-BE-NEXT: bx lr
entry:
%vin = load <4 x i32>, <4 x i32>* %in, align 8
%vdbl = mul <4 x i32> %vin, %vin
%cast = bitcast <4 x i32> %vdbl to <8 x i16>
%cdbl = mul <8 x i16> %cast, %cast
store <8 x i16> %cdbl, <8 x i16>* %out, align 8
ret void
}
; Similar test case but using the arm.mve.vreinterpretq intrinsic instead,
; which is defined to reinterpret the in-register format, so it generates no
; instruction in either endianness.
define arm_aapcs_vfpcc void @test_vreinterpretq(<4 x i32>* readonly %in, <8 x i16>* %out) {
; CHECK-LE-LABEL: test_vreinterpretq:
; CHECK-LE: @ %bb.0: @ %entry
; CHECK-LE-NEXT: vldrw.u32 q0, [r0]
; CHECK-LE-NEXT: vmul.i32 q0, q0, q0
; CHECK-LE-NEXT: vmul.i16 q0, q0, q0
; CHECK-LE-NEXT: vstrw.32 q0, [r1]
; CHECK-LE-NEXT: bx lr
;
; CHECK-BE-LABEL: test_vreinterpretq:
; CHECK-BE: @ %bb.0: @ %entry
; CHECK-BE-NEXT: vldrw.u32 q0, [r0]
; CHECK-BE-NEXT: vmul.i32 q0, q0, q0
; CHECK-BE-NEXT: vmul.i16 q0, q0, q0
; CHECK-BE-NEXT: vstrh.16 q0, [r1]
; CHECK-BE-NEXT: bx lr
entry:
%vin = load <4 x i32>, <4 x i32>* %in, align 8
%vdbl = mul <4 x i32> %vin, %vin
%cast = call <8 x i16> @llvm.arm.mve.vreinterpretq.v8i16.v4i32(<4 x i32> %vdbl)
%cdbl = mul <8 x i16> %cast, %cast
store <8 x i16> %cdbl, <8 x i16>* %out, align 8
ret void
}
define arm_aapcs_vfpcc <8 x half> @undef_one() {
; CHECK-LABEL: undef_one:
; CHECK: @ %bb.0:
; CHECK-NEXT: bx lr
%c = call <8 x half> @llvm.arm.mve.vreinterpretq.v8f16.v4f32(<4 x float> undef)
ret <8 x half> %c
}
declare <8 x half> @llvm.arm.mve.vreinterpretq.v8f16.v4f32(<4 x float>)
declare <8 x i16> @llvm.arm.mve.vreinterpretq.v8i16.v4i32(<4 x i32>)
|
