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
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -passes=instcombine -S < %s | FileCheck %s
; Tests for select to scmp
define i32 @scmp_x_0_inverted(i32 %x) {
; CHECK-LABEL: define i32 @scmp_x_0_inverted(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.scmp.i32.i32(i32 [[X]], i32 0)
; CHECK-NEXT: ret i32 [[TMP1]]
;
%2 = icmp ne i32 %x, 0
%3 = zext i1 %2 to i32
%4 = icmp sgt i32 %x, -1
%5 = select i1 %4, i32 %3, i32 -1
ret i32 %5
}
; y = -10
define i32 @scmp_x_0_inverted_const_neg10(i32 %x) {
; CHECK-LABEL: define i32 @scmp_x_0_inverted_const_neg10(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.scmp.i32.i32(i32 [[X]], i32 -10)
; CHECK-NEXT: ret i32 [[TMP1]]
;
%1 = icmp ne i32 %x, -10
%2 = zext i1 %1 to i32
%3 = icmp sgt i32 %x, -11
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; y = 7 (i8)
define i8 @scmp_x_0_inverted_i8(i8 %x) {
; CHECK-LABEL: define i8 @scmp_x_0_inverted_i8(
; CHECK-SAME: i8 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.scmp.i8.i8(i8 [[X]], i8 7)
; CHECK-NEXT: ret i8 [[TMP1]]
;
%1 = icmp ne i8 %x, 7
%2 = zext i1 %1 to i8
%3 = icmp sgt i8 %x, 6
%4 = select i1 %3, i8 %2, i8 -1
ret i8 %4
}
; scmp using ints of two kinds- i32 and i64
define i32 @scmp_x_0_inverted_i64_neq(i32 %x) {
; CHECK-LABEL: define i32 @scmp_x_0_inverted_i64_neq(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[SEL:%.*]] = call i64 @llvm.scmp.i64.i32(i32 [[X]], i32 0)
; CHECK-NEXT: [[RET:%.*]] = trunc i64 [[SEL]] to i32
; CHECK-NEXT: ret i32 [[RET]]
;
%x64 = sext i32 %x to i64
%cmp1 = icmp ne i64 %x64, 0
%zext = zext i1 %cmp1 to i64
%cmp2 = icmp sgt i64 %x64, -1
%sel = select i1 %cmp2, i64 %zext, i64 -1
%ret = trunc i64 %sel to i32
ret i32 %ret
}
; Same example as previous but with inequality
define i32 @scmp_x_0_inverted_i64_sgt(i32 %x) {
; CHECK-LABEL: define i32 @scmp_x_0_inverted_i64_sgt(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[SEL:%.*]] = call i64 @llvm.scmp.i64.i32(i32 [[X]], i32 0)
; CHECK-NEXT: [[RET:%.*]] = trunc i64 [[SEL]] to i32
; CHECK-NEXT: ret i32 [[RET]]
;
%x64 = sext i32 %x to i64
%cmp1 = icmp sgt i64 %x64, 0
%zext = zext i1 %cmp1 to i64
%cmp2 = icmp sgt i64 %x64, -1
%sel = select i1 %cmp2, i64 %zext, i64 -1
%ret = trunc i64 %sel to i32
ret i32 %ret
}
; y = -1000
define i32 @scmp_x_0_inverted_const_neg1000(i32 %x) {
; CHECK-LABEL: define i32 @scmp_x_0_inverted_const_neg1000(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.scmp.i32.i32(i32 [[X]], i32 -1000)
; CHECK-NEXT: ret i32 [[TMP1]]
;
%1 = icmp sgt i32 %x, -1000
%2 = zext i1 %1 to i32
%3 = icmp sgt i32 %x, -1001
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; y = 1729
define i32 @scmp_x_0_inverted_const_1729_sgt(i32 %x) {
; CHECK-LABEL: define i32 @scmp_x_0_inverted_const_1729_sgt(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.scmp.i32.i32(i32 [[X]], i32 1729)
; CHECK-NEXT: ret i32 [[TMP1]]
;
%1 = icmp sgt i32 %x, 1729
%2 = zext i1 %1 to i32
%3 = icmp sgt i32 %x, 1728
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; ucmp with 10
define i32 @ucmp_x_10_inverted(i32 %x) {
; CHECK-LABEL: define i32 @ucmp_x_10_inverted(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.ucmp.i32.i32(i32 [[X]], i32 10)
; CHECK-NEXT: ret i32 [[TMP1]]
;
%1 = icmp ne i32 %x, 10
%2 = zext i1 %1 to i32
%3 = icmp ugt i32 %x, 9
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; ucmp with -3, wraps around
define i32 @ucmp_x_neg1_inverted(i32 %x) {
; CHECK-LABEL: define i32 @ucmp_x_neg1_inverted(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.ucmp.i32.i32(i32 [[X]], i32 -3)
; CHECK-NEXT: ret i32 [[TMP1]]
;
%1 = icmp ne i32 %x, -3
%2 = zext i1 %1 to i32
%3 = icmp ugt i32 %x, -4
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; ucmp with -4, wraps around
define i8 @ucmp_x_neg4_i8_ugt(i8 %x) {
; CHECK-LABEL: define i8 @ucmp_x_neg4_i8_ugt(
; CHECK-SAME: i8 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.ucmp.i8.i8(i8 [[X]], i8 -4)
; CHECK-NEXT: ret i8 [[TMP1]]
;
%1 = icmp ugt i8 %x, -4
%2 = zext i1 %1 to i8
%3 = icmp ugt i8 %x, -5
%4 = select i1 %3, i8 %2, i8 -1
ret i8 %4
}
; Vector tests
; Test with splat vec
define <4 x i32> @scmp_x_0_inverted_splat_vec(<4 x i32> %x) {
; CHECK-LABEL: define <4 x i32> @scmp_x_0_inverted_splat_vec(
; CHECK-SAME: <4 x i32> [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> @llvm.scmp.v4i32.v4i32(<4 x i32> [[X]], <4 x i32> zeroinitializer)
; CHECK-NEXT: ret <4 x i32> [[TMP1]]
;
%2 = icmp ne <4 x i32> %x, zeroinitializer
%3 = zext <4 x i1> %2 to <4 x i32>
%4 = icmp sgt <4 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1>
%5 = select <4 x i1> %4, <4 x i32> %3, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
ret <4 x i32> %5
}
; Test with non-splat vector and different bitwidth
define <4 x i32> @non_splat_vec_scmp_diff_bitwidth(<4 x i32> %x) {
; CHECK-LABEL: define <4 x i32> @non_splat_vec_scmp_diff_bitwidth(
; CHECK-SAME: <4 x i32> [[X:%.*]]) {
; CHECK-NEXT: [[SEL:%.*]] = call <4 x i64> @llvm.scmp.v4i64.v4i32(<4 x i32> [[X]], <4 x i32> <i32 0, i32 1, i32 -1, i32 5>)
; CHECK-NEXT: [[RET:%.*]] = trunc <4 x i64> [[SEL]] to <4 x i32>
; CHECK-NEXT: ret <4 x i32> [[RET]]
;
%x64 = sext <4 x i32> %x to <4 x i64>
%cmp1 = icmp slt <4 x i64> %x64, <i64 0, i64 1, i64 -1, i64 5>
%sext = sext <4 x i1> %cmp1 to <4 x i64>
%cmp2 = icmp slt <4 x i64> %x64, <i64 1, i64 2, i64 0, i64 6>
%sel = select <4 x i1> %cmp2, <4 x i64> %sext, <4 x i64> <i64 1, i64 1, i64 1, i64 1>
%ret = trunc <4 x i64> %sel to <4 x i32>
ret <4 x i32> %ret
}
; Negative examples
; Not scmp due to wrong RHS of the predicate
define i32 @scmp_ne_0(i32 %0) {
; CHECK-LABEL: define i32 @scmp_ne_0(
; CHECK-SAME: i32 [[TMP0:%.*]]) {
; CHECK-NEXT: [[TMP2:%.*]] = icmp ne i32 [[TMP0]], 0
; CHECK-NEXT: [[TMP3:%.*]] = zext i1 [[TMP2]] to i32
; CHECK-NEXT: [[TMP4:%.*]] = icmp sgt i32 [[TMP0]], 1
; CHECK-NEXT: [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP3]], i32 -1
; CHECK-NEXT: ret i32 [[TMP5]]
;
%2 = icmp ne i32 %0, 0
%3 = zext i1 %2 to i32
%4 = icmp sgt i32 %0, 1
%5 = select i1 %4, i32 %3, i32 -1
ret i32 %5
}
; y = 0 with unsigned compare but RHS wraps
define i32 @ucmp_x_0_inverted(i32 %x) {
; CHECK-LABEL: define i32 @ucmp_x_0_inverted(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: ret i32 -1
;
%1 = icmp ne i32 %x, 0
%2 = zext i1 %1 to i32
%3 = icmp ugt i32 %x, -1
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; Don't fold with INT32_MIN
define i32 @scmp_x_0_inverted_const_min(i32 %x) {
; CHECK-LABEL: define i32 @scmp_x_0_inverted_const_min(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i32 [[X]], -2147483648
; CHECK-NEXT: [[TMP2:%.*]] = zext i1 [[TMP1]] to i32
; CHECK-NEXT: ret i32 [[TMP2]]
;
%1 = icmp ne i32 %x, -2147483648
%2 = zext i1 %1 to i32
%3 = icmp sge i32 %x, -2147483648
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; Unsigned cmp of zext of i32 with i64 -1 should always be -1
define i32 @ucmp_x_0_inverted_i64_ugt(i32 %x) {
; CHECK-LABEL: define i32 @ucmp_x_0_inverted_i64_ugt(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: ret i32 -1
;
%x64 = zext i32 %x to i64
%cmp1 = icmp ugt i64 %x64, 0
%zext = zext i1 %cmp1 to i64
%cmp2 = icmp ugt i64 %x64, -1
%sel = select i1 %cmp2, i64 %zext, i64 -1
%ret = trunc i64 %sel to i32
ret i32 %ret
}
; y = 4294967295 (UINT32_MAX), simply sign extend neq
define i32 @ucmp_x_const_u32max(i32 %x) {
; CHECK-LABEL: define i32 @ucmp_x_const_u32max(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = icmp ne i32 [[X]], -1
; CHECK-NEXT: [[TMP2:%.*]] = sext i1 [[TMP1]] to i32
; CHECK-NEXT: ret i32 [[TMP2]]
;
%1 = icmp ugt i32 %x, 4294967295
%2 = zext i1 %1 to i32
%3 = icmp ugt i32 %x, 4294967294
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; Don't fold with different signedness
define i32 @different_signedness_neg(i32 %x) {
; CHECK-LABEL: define i32 @different_signedness_neg(
; CHECK-SAME: i32 [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i32 [[X]], -10
; CHECK-NEXT: [[TMP2:%.*]] = zext i1 [[TMP1]] to i32
; CHECK-NEXT: [[TMP3:%.*]] = icmp sgt i32 [[X]], -11
; CHECK-NEXT: [[TMP4:%.*]] = select i1 [[TMP3]], i32 [[TMP2]], i32 -1
; CHECK-NEXT: ret i32 [[TMP4]]
;
%1 = icmp ugt i32 %x, -10
%2 = zext i1 %1 to i32
%3 = icmp sgt i32 %x, -11
%4 = select i1 %3, i32 %2, i32 -1
ret i32 %4
}
; Test with wrong false value
define <4 x i32> @scmp_x_0_inverted_vec(<4 x i32> %x) {
; CHECK-LABEL: define <4 x i32> @scmp_x_0_inverted_vec(
; CHECK-SAME: <4 x i32> [[X:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <4 x i32> [[X]], zeroinitializer
; CHECK-NEXT: [[TMP2:%.*]] = zext <4 x i1> [[TMP1]] to <4 x i32>
; CHECK-NEXT: [[TMP3:%.*]] = icmp sgt <4 x i32> [[X]], splat (i32 -1)
; CHECK-NEXT: [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x i32> [[TMP2]], <4 x i32> <i32 -1, i32 -2, i32 -1, i32 -1>
; CHECK-NEXT: ret <4 x i32> [[TMP4]]
;
%2 = icmp ne <4 x i32> %x, zeroinitializer
%3 = zext <4 x i1> %2 to <4 x i32>
%4 = icmp sgt <4 x i32> %x, <i32 -1, i32 -1, i32 -1, i32 -1>
%5 = select <4 x i1> %4, <4 x i32> %3, <4 x i32> <i32 -1, i32 -2, i32 -1, i32 -1>
ret <4 x i32> %5
}
|
