mve-v2i2v.ll
10.8 KB
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
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -instcombine -S -o - %s | FileCheck %s
target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
declare i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1>)
declare i32 @llvm.arm.mve.pred.v2i.v8i1(<8 x i1>)
declare i32 @llvm.arm.mve.pred.v2i.v16i1(<16 x i1>)
declare <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32)
declare <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32)
declare <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32)
; Round-trip conversions from predicate vector to i32 back to the same
; size of vector should be eliminated.
define <4 x i1> @v2i2v_4(<4 x i1> %vin) {
; CHECK-LABEL: @v2i2v_4(
; CHECK-NEXT: entry:
; CHECK-NEXT: ret <4 x i1> [[VIN:%.*]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %int)
ret <4 x i1> %vout
}
define <8 x i1> @v2i2v_8(<8 x i1> %vin) {
; CHECK-LABEL: @v2i2v_8(
; CHECK-NEXT: entry:
; CHECK-NEXT: ret <8 x i1> [[VIN:%.*]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v8i1(<8 x i1> %vin)
%vout = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 %int)
ret <8 x i1> %vout
}
define <16 x i1> @v2i2v_16(<16 x i1> %vin) {
; CHECK-LABEL: @v2i2v_16(
; CHECK-NEXT: entry:
; CHECK-NEXT: ret <16 x i1> [[VIN:%.*]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v16i1(<16 x i1> %vin)
%vout = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 %int)
ret <16 x i1> %vout
}
; Conversions from a predicate vector to i32 and then to a _different_
; size of predicate vector should be left alone.
define <16 x i1> @v2i2v_4_16(<4 x i1> %vin) {
; CHECK-LABEL: @v2i2v_4_16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[INT:%.*]] = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> [[VIN:%.*]]), !range !0
; CHECK-NEXT: [[VOUT:%.*]] = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 [[INT]])
; CHECK-NEXT: ret <16 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%vout = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 %int)
ret <16 x i1> %vout
}
define <4 x i1> @v2i2v_8_4(<8 x i1> %vin) {
; CHECK-LABEL: @v2i2v_8_4(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[INT:%.*]] = call i32 @llvm.arm.mve.pred.v2i.v8i1(<8 x i1> [[VIN:%.*]]), !range !0
; CHECK-NEXT: [[VOUT:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[INT]])
; CHECK-NEXT: ret <4 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v8i1(<8 x i1> %vin)
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %int)
ret <4 x i1> %vout
}
define <8 x i1> @v2i2v_16_8(<16 x i1> %vin) {
; CHECK-LABEL: @v2i2v_16_8(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[INT:%.*]] = call i32 @llvm.arm.mve.pred.v2i.v16i1(<16 x i1> [[VIN:%.*]]), !range !0
; CHECK-NEXT: [[VOUT:%.*]] = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 [[INT]])
; CHECK-NEXT: ret <8 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v16i1(<16 x i1> %vin)
%vout = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 %int)
ret <8 x i1> %vout
}
; Round-trip conversions from i32 to predicate vector back to i32
; should be eliminated.
define i32 @i2v2i_4(i32 %iin) {
; CHECK-LABEL: @i2v2i_4(
; CHECK-NEXT: entry:
; CHECK-NEXT: ret i32 [[IIN:%.*]]
;
entry:
%vec = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %iin)
%iout = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vec)
ret i32 %iout
}
define i32 @i2v2i_8(i32 %iin) {
; CHECK-LABEL: @i2v2i_8(
; CHECK-NEXT: entry:
; CHECK-NEXT: ret i32 [[IIN:%.*]]
;
entry:
%vec = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 %iin)
%iout = call i32 @llvm.arm.mve.pred.v2i.v8i1(<8 x i1> %vec)
ret i32 %iout
}
define i32 @i2v2i_16(i32 %iin) {
; CHECK-LABEL: @i2v2i_16(
; CHECK-NEXT: entry:
; CHECK-NEXT: ret i32 [[IIN:%.*]]
;
entry:
%vec = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 %iin)
%iout = call i32 @llvm.arm.mve.pred.v2i.v16i1(<16 x i1> %vec)
ret i32 %iout
}
; v2i leaves the top 16 bits clear. So a trunc/zext pair applied to
; its output, going via i16, can be completely eliminated - but not
; one going via i8. Similarly with other methods of clearing the top
; bits, like bitwise and.
define i32 @v2i_truncext_i16(<4 x i1> %vin) {
; CHECK-LABEL: @v2i_truncext_i16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[WIDE1:%.*]] = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> [[VIN:%.*]]), !range !0
; CHECK-NEXT: ret i32 [[WIDE1]]
;
entry:
%wide1 = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%narrow = trunc i32 %wide1 to i16
%wide2 = zext i16 %narrow to i32
ret i32 %wide2
}
define i32 @v2i_truncext_i8(<4 x i1> %vin) {
; CHECK-LABEL: @v2i_truncext_i8(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[WIDE1:%.*]] = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> [[VIN:%.*]]), !range !0
; CHECK-NEXT: [[WIDE2:%.*]] = and i32 [[WIDE1]], 255
; CHECK-NEXT: ret i32 [[WIDE2]]
;
entry:
%wide1 = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%narrow = trunc i32 %wide1 to i8
%wide2 = zext i8 %narrow to i32
ret i32 %wide2
}
define i32 @v2i_and_16(<4 x i1> %vin) {
; CHECK-LABEL: @v2i_and_16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[WIDE1:%.*]] = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> [[VIN:%.*]]), !range !0
; CHECK-NEXT: ret i32 [[WIDE1]]
;
entry:
%wide1 = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%wide2 = and i32 %wide1, 65535
ret i32 %wide2
}
define i32 @v2i_and_15(<4 x i1> %vin) {
; CHECK-LABEL: @v2i_and_15(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[WIDE1:%.*]] = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> [[VIN:%.*]]), !range !0
; CHECK-NEXT: [[WIDE2:%.*]] = and i32 [[WIDE1]], 32767
; CHECK-NEXT: ret i32 [[WIDE2]]
;
entry:
%wide1 = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%wide2 = and i32 %wide1, 32767
ret i32 %wide2
}
; i2v doesn't use the top bits of its input. So the same operations
; on a value that's about to be passed to i2v can be eliminated.
define <4 x i1> @i2v_truncext_i16(i32 %wide1) {
; CHECK-LABEL: @i2v_truncext_i16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[WIDE1:%.*]])
; CHECK-NEXT: ret <4 x i1> [[VOUT]]
;
entry:
%narrow = trunc i32 %wide1 to i16
%wide2 = zext i16 %narrow to i32
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %wide2)
ret <4 x i1> %vout
}
define <4 x i1> @i2v_truncext_i8(i32 %wide1) {
; CHECK-LABEL: @i2v_truncext_i8(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[WIDE2:%.*]] = and i32 [[WIDE1:%.*]], 255
; CHECK-NEXT: [[VOUT:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[WIDE2]])
; CHECK-NEXT: ret <4 x i1> [[VOUT]]
;
entry:
%narrow = trunc i32 %wide1 to i8
%wide2 = zext i8 %narrow to i32
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %wide2)
ret <4 x i1> %vout
}
define <4 x i1> @i2v_and_16(i32 %wide1) {
; CHECK-LABEL: @i2v_and_16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[WIDE1:%.*]])
; CHECK-NEXT: ret <4 x i1> [[VOUT]]
;
entry:
%wide2 = and i32 %wide1, 65535
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %wide2)
ret <4 x i1> %vout
}
define <4 x i1> @i2v_and_15(i32 %wide1) {
; CHECK-LABEL: @i2v_and_15(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[WIDE2:%.*]] = and i32 [[WIDE1:%.*]], 32767
; CHECK-NEXT: [[VOUT:%.*]] = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 [[WIDE2]])
; CHECK-NEXT: ret <4 x i1> [[VOUT]]
;
entry:
%wide2 = and i32 %wide1, 32767
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %wide2)
ret <4 x i1> %vout
}
; If a predicate vector is round-tripped to an integer and back, and
; complemented while it's in integer form, we should collapse that to
; a complement of the vector itself. (Rationale: this is likely to
; allow it to be code-generated as MVE VPNOT.)
define <4 x i1> @vpnot_4(<4 x i1> %vin) {
; CHECK-LABEL: @vpnot_4(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = xor <4 x i1> [[VIN:%.*]], <i1 true, i1 true, i1 true, i1 true>
; CHECK-NEXT: ret <4 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%flipped = xor i32 %int, 65535
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %flipped)
ret <4 x i1> %vout
}
define <8 x i1> @vpnot_8(<8 x i1> %vin) {
; CHECK-LABEL: @vpnot_8(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = xor <8 x i1> [[VIN:%.*]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
; CHECK-NEXT: ret <8 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v8i1(<8 x i1> %vin)
%flipped = xor i32 %int, 65535
%vout = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 %flipped)
ret <8 x i1> %vout
}
define <16 x i1> @vpnot_16(<16 x i1> %vin) {
; CHECK-LABEL: @vpnot_16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = xor <16 x i1> [[VIN:%.*]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
; CHECK-NEXT: ret <16 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v16i1(<16 x i1> %vin)
%flipped = xor i32 %int, 65535
%vout = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 %flipped)
ret <16 x i1> %vout
}
; And this still works even if the i32 is narrowed to i16 and back on
; opposite sides of the xor.
define <4 x i1> @vpnot_narrow_4(<4 x i1> %vin) {
; CHECK-LABEL: @vpnot_narrow_4(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = xor <4 x i1> [[VIN:%.*]], <i1 true, i1 true, i1 true, i1 true>
; CHECK-NEXT: ret <4 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v4i1(<4 x i1> %vin)
%narrow = trunc i32 %int to i16
%flipped_narrow = xor i16 %narrow, -1
%flipped = zext i16 %flipped_narrow to i32
%vout = call <4 x i1> @llvm.arm.mve.pred.i2v.v4i1(i32 %flipped)
ret <4 x i1> %vout
}
define <8 x i1> @vpnot_narrow_8(<8 x i1> %vin) {
; CHECK-LABEL: @vpnot_narrow_8(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = xor <8 x i1> [[VIN:%.*]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
; CHECK-NEXT: ret <8 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v8i1(<8 x i1> %vin)
%narrow = trunc i32 %int to i16
%flipped_narrow = xor i16 %narrow, -1
%flipped = zext i16 %flipped_narrow to i32
%vout = call <8 x i1> @llvm.arm.mve.pred.i2v.v8i1(i32 %flipped)
ret <8 x i1> %vout
}
define <16 x i1> @vpnot_narrow_16(<16 x i1> %vin) {
; CHECK-LABEL: @vpnot_narrow_16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VOUT:%.*]] = xor <16 x i1> [[VIN:%.*]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
; CHECK-NEXT: ret <16 x i1> [[VOUT]]
;
entry:
%int = call i32 @llvm.arm.mve.pred.v2i.v16i1(<16 x i1> %vin)
%narrow = trunc i32 %int to i16
%flipped_narrow = xor i16 %narrow, -1
%flipped = zext i16 %flipped_narrow to i32
%vout = call <16 x i1> @llvm.arm.mve.pred.i2v.v16i1(i32 %flipped)
ret <16 x i1> %vout
}