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
|
; RUN: llc -debug-only=isel -o /dev/null < %s 2>&1 | FileCheck %s
; REQUIRES: asserts
; NOTE: Due to their nature the expected inserts and extracts often emit no
; instructions and so these tests verify the output of DAGCombiner directly.
target triple = "aarch64-unknown-linux-gnu"
; CHECK: Initial selection DAG: %bb.0 'insert_small_fixed_into_big_fixed:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v8i8,ch = CopyFromReg t0, Register:v8i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: v16i8 = insert_subvector poison:v16i8, t4, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v16i8 $q0, t6
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v16i8 $q0, t8:1
; CHECK: Optimized lowered selection DAG: %bb.0 'insert_small_fixed_into_big_fixed:'
; CHECK: SelectionDAG has 9 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v8i8,ch = CopyFromReg t0, Register:v8i8 %0
; CHECK: t10: v16i8 = insert_subvector poison:v16i8, t2, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v16i8 $q0, t10
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v16i8 $q0, t8:1
define <16 x i8> @insert_small_fixed_into_big_fixed(<8 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<8 x i8> %a, i64 0)
%insert = call <16 x i8> @llvm.vector.insert(<16 x i8> poison, <4 x i8> %extract, i64 0)
ret <16 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'insert_small_fixed_into_big_scalable:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v8i8,ch = CopyFromReg t0, Register:v8i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: nxv16i8 = insert_subvector poison:nxv16i8, t4, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t6
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:nxv16i8 $z0, t8:1
; CHECK: Optimized lowered selection DAG: %bb.0 'insert_small_fixed_into_big_scalable:'
; CHECK: SelectionDAG has 9 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v8i8,ch = CopyFromReg t0, Register:v8i8 %0
; CHECK: t10: nxv16i8 = insert_subvector poison:nxv16i8, t2, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t10
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:nxv16i8 $z0, t8:1
define <vscale x 16 x i8> @insert_small_fixed_into_big_scalable(<8 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<8 x i8> %a, i64 0)
%insert = call <vscale x 16 x i8> @llvm.vector.insert(<vscale x 16 x i8> poison, <4 x i8> %extract, i64 0)
ret <vscale x 16 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'insert_small_scalable_into_big_fixed:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv8i16,ch = CopyFromReg t0, Register:nxv8i16 %0
; CHECK: t3: nxv8i8 = truncate t2
; CHECK: t5: v4i8 = extract_subvector t3, Constant:i64<0>
; CHECK: t7: v16i8 = insert_subvector poison:v16i8, t5, Constant:i64<0>
; CHECK: t9: ch,glue = CopyToReg t0, Register:v16i8 $q0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:v16i8 $q0, t9:1
; CHECK: Optimized lowered selection DAG: %bb.0 'insert_small_scalable_into_big_fixed:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv8i16,ch = CopyFromReg t0, Register:nxv8i16 %0
; CHECK: t3: nxv8i8 = truncate t2
; CHECK: t5: v4i8 = extract_subvector t3, Constant:i64<0>
; CHECK: t7: v16i8 = insert_subvector poison:v16i8, t5, Constant:i64<0>
; CHECK: t9: ch,glue = CopyToReg t0, Register:v16i8 $q0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:v16i8 $q0, t9:1
; Resulting insert would not be legal, so there's no transformation.
define <16 x i8> @insert_small_scalable_into_big_fixed(<vscale x 8 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<vscale x 8 x i8> %a, i64 0)
%insert = call <16 x i8> @llvm.vector.insert(<16 x i8> poison, <4 x i8> %extract, i64 0)
ret <16 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'insert_small_scalable_into_big_scalable_1:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv8i16,ch = CopyFromReg t0, Register:nxv8i16 %0
; CHECK: t3: nxv8i8 = truncate t2
; CHECK: t5: v4i8 = extract_subvector t3, Constant:i64<0>
; CHECK: t7: nxv16i8 = insert_subvector poison:nxv16i8, t5, Constant:i64<0>
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv16i8 $z0, t9:1
; CHECK: Optimized lowered selection DAG: %bb.0 'insert_small_scalable_into_big_scalable_1:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv8i16,ch = CopyFromReg t0, Register:nxv8i16 %0
; CHECK: t3: nxv8i8 = truncate t2
; CHECK: t11: nxv16i8 = insert_subvector poison:nxv16i8, t3, Constant:i64<0>
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t11
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv16i8 $z0, t9:1
define <vscale x 16 x i8> @insert_small_scalable_into_big_scalable_1(<vscale x 8 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<vscale x 8 x i8> %a, i64 0)
%insert = call <vscale x 16 x i8> @llvm.vector.insert(<vscale x 16 x i8> poison, <4 x i8> %extract, i64 0)
ret <vscale x 16 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'insert_small_scalable_into_big_scalable_2:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv8i16,ch = CopyFromReg t0, Register:nxv8i16 %0
; CHECK: t3: nxv8i8 = truncate t2
; CHECK: t5: nxv4i8 = extract_subvector t3, Constant:i64<0>
; CHECK: t7: nxv16i8 = insert_subvector poison:nxv16i8, t5, Constant:i64<0>
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv16i8 $z0, t9:1
; CHECK: Optimized lowered selection DAG: %bb.0 'insert_small_scalable_into_big_scalable_2:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv8i16,ch = CopyFromReg t0, Register:nxv8i16 %0
; CHECK: t3: nxv8i8 = truncate t2
; CHECK: t11: nxv16i8 = insert_subvector poison:nxv16i8, t3, Constant:i64<0>
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t11
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv16i8 $z0, t9:1
define <vscale x 16 x i8> @insert_small_scalable_into_big_scalable_2(<vscale x 8 x i8> %a) #0 {
%extract = call <vscale x 4 x i8> @llvm.vector.extract(<vscale x 8 x i8> %a, i64 0)
%insert = call <vscale x 16 x i8> @llvm.vector.insert(<vscale x 16 x i8> poison, <vscale x 4 x i8> %extract, i64 0)
ret <vscale x 16 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'extract_small_fixed_from_big_fixed:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v16i8,ch = CopyFromReg t0, Register:v16i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: v8i8 = insert_subvector poison:v8i8, t4, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v8i8 $d0, t6
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v8i8 $d0, t8:1
; CHECK: Optimized lowered selection DAG: %bb.0 'extract_small_fixed_from_big_fixed:'
; CHECK: SelectionDAG has 8 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v16i8,ch = CopyFromReg t0, Register:v16i8 %0
; CHECK: t10: v8i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v8i8 $d0, t10
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v8i8 $d0, t8:1
define <8 x i8> @extract_small_fixed_from_big_fixed(<16 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<16 x i8> %a, i64 0)
%insert = call <8 x i8> @llvm.vector.insert(<8 x i8> poison, <4 x i8> %extract, i64 0)
ret <8 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'extract_small_scalable_from_big_fixed:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v16i8,ch = CopyFromReg t0, Register:v16i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: nxv8i8 = insert_subvector poison:nxv8i8, t4, Constant:i64<0>
; CHECK: t7: nxv8i16 = any_extend t6
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv8i16 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv8i16 $z0, t9:1
; CHECK: Optimized lowered selection DAG: %bb.0 'extract_small_scalable_from_big_fixed:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v16i8,ch = CopyFromReg t0, Register:v16i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: nxv8i8 = insert_subvector poison:nxv8i8, t4, Constant:i64<0>
; CHECK: t7: nxv8i16 = any_extend t6
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv8i16 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv8i16 $z0, t9:1
; Resulting insert would not be legal, so there's no transformation.
define <vscale x 8 x i8> @extract_small_scalable_from_big_fixed(<16 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<16 x i8> %a, i64 0)
%insert = call <vscale x 8 x i8> @llvm.vector.insert(<vscale x 8 x i8> poison, <4 x i8> %extract, i64 0)
ret <vscale x 8 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'extract_small_fixed_from_big_scalable:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: v8i8 = insert_subvector poison:v8i8, t4, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v8i8 $d0, t6
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v8i8 $d0, t8:1
; CHECK: Optimized lowered selection DAG: %bb.0 'extract_small_fixed_from_big_scalable:'
; CHECK: SelectionDAG has 8 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t10: v8i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v8i8 $d0, t10
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v8i8 $d0, t8:1
define <8 x i8> @extract_small_fixed_from_big_scalable(<vscale x 16 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<vscale x 16 x i8> %a, i64 0)
%insert = call <8 x i8> @llvm.vector.insert(<8 x i8> poison, <4 x i8> %extract, i64 0)
ret <8 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'extract_small_scalable_from_big_scalable_1:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: nxv8i8 = insert_subvector poison:nxv8i8, t4, Constant:i64<0>
; CHECK: t7: nxv8i16 = any_extend t6
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv8i16 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv8i16 $z0, t9:1
; CHECK: Optimized lowered selection DAG: %bb.0 'extract_small_scalable_from_big_scalable_1:'
; CHECK: SelectionDAG has 9 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t11: nxv8i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t7: nxv8i16 = any_extend t11
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv8i16 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv8i16 $z0, t9:1
define <vscale x 8 x i8> @extract_small_scalable_from_big_scalable_1(<vscale x 16 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<vscale x 16 x i8> %a, i64 0)
%insert = call <vscale x 8 x i8> @llvm.vector.insert(<vscale x 8 x i8> poison, <4 x i8> %extract, i64 0)
ret <vscale x 8 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'extract_small_scalable_from_big_scalable_2:'
; CHECK: SelectionDAG has 11 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t4: nxv4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: nxv8i8 = insert_subvector poison:nxv8i8, t4, Constant:i64<0>
; CHECK: t7: nxv8i16 = any_extend t6
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv8i16 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv8i16 $z0, t9:1
; CHECK: Optimized lowered selection DAG: %bb.0 'extract_small_scalable_from_big_scalable_2:'
; CHECK: SelectionDAG has 9 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t11: nxv8i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t7: nxv8i16 = any_extend t11
; CHECK: t9: ch,glue = CopyToReg t0, Register:nxv8i16 $z0, t7
; CHECK: t10: ch = AArch64ISD::RET_GLUE t9, Register:nxv8i16 $z0, t9:1
define <vscale x 8 x i8> @extract_small_scalable_from_big_scalable_2(<vscale x 16 x i8> %a) #0 {
%extract = call <vscale x 4 x i8> @llvm.vector.extract(<vscale x 16 x i8> %a, i64 0)
%insert = call <vscale x 8 x i8> @llvm.vector.insert(<vscale x 8 x i8> poison, <vscale x 4 x i8> %extract, i64 0)
ret <vscale x 8 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'extract_fixed_from_scalable:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: v16i8 = insert_subvector poison:v16i8, t4, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v16i8 $q0, t6
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v16i8 $q0, t8:1
; CHECK: Optimized lowered selection DAG: %bb.0 'extract_fixed_from_scalable:'
; CHECK: SelectionDAG has 8 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: nxv16i8,ch = CopyFromReg t0, Register:nxv16i8 %0
; CHECK: t10: v16i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:v16i8 $q0, t10
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:v16i8 $q0, t8:1
; A variant of insert_small_scalable_into_big_fixed whose vector types prevent
; the expected transformation because the resulting insert would not be legal.
; In this instance their matching minimum vector lengths allow us to perform the
; opposite transformation and emit an extract instead.
define <16 x i8> @extract_fixed_from_scalable(<vscale x 16 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<vscale x 16 x i8> %a, i64 0)
%insert = call <16 x i8> @llvm.vector.insert(<16 x i8> poison, <4 x i8> %extract, i64 0)
ret <16 x i8> %insert
}
; CHECK: Initial selection DAG: %bb.0 'insert_fixed_into_scalable:'
; CHECK: SelectionDAG has 10 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v16i8,ch = CopyFromReg t0, Register:v16i8 %0
; CHECK: t4: v4i8 = extract_subvector t2, Constant:i64<0>
; CHECK: t6: nxv16i8 = insert_subvector poison:nxv16i8, t4, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t6
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:nxv16i8 $z0, t8:1
; CHECK: Optimized lowered selection DAG: %bb.0 'insert_fixed_into_scalable:'
; CHECK: SelectionDAG has 9 nodes:
; CHECK: t0: ch,glue = EntryToken
; CHECK: t2: v16i8,ch = CopyFromReg t0, Register:v16i8 %0
; CHECK: t10: nxv16i8 = insert_subvector poison:nxv16i8, t2, Constant:i64<0>
; CHECK: t8: ch,glue = CopyToReg t0, Register:nxv16i8 $z0, t10
; CHECK: t9: ch = AArch64ISD::RET_GLUE t8, Register:nxv16i8 $z0, t8:1
; A variant of extract_small_scalable_from_big_fixed whose vector types prevent
; the expected transformation because the resulting extract would not be legal.
; In this instance their matching minimum vector lengths allow us to perform the
; opposite transformation and emit an insert instead.
define <vscale x 16 x i8> @insert_fixed_into_scalable(<16 x i8> %a) #0 {
%extract = call <4 x i8> @llvm.vector.extract(<16 x i8> %a, i64 0)
%insert = call <vscale x 16 x i8> @llvm.vector.insert(<vscale x 16 x i8> poison, <4 x i8> %extract, i64 0)
ret <vscale x 16 x i8> %insert
}
attributes #0 = { "target-features"="+sve" }
|
