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
| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=riscv32 -mattr=+d -verify-machineinstrs < %s \
; RUN: | FileCheck %s -check-prefix=RV32I
; RUN: llc -mtriple=riscv64 -mattr=+d -verify-machineinstrs < %s \
; RUN: | FileCheck %s -check-prefix=RV64I
; Selects of wide values are split into two selects, which can easily cause
; unnecessary control flow. Here we check some cases where we can currently
; emit a sequence of selects with shared control flow.
define i64 @cmovcc64(i32 signext %a, i64 %b, i64 %c) nounwind {
; RV32I-LABEL: cmovcc64:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: addi a5, zero, 123
; RV32I-NEXT: beq a0, a5, .LBB0_2
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: mv a1, a3
; RV32I-NEXT: mv a2, a4
; RV32I-NEXT: .LBB0_2: # %entry
; RV32I-NEXT: mv a0, a1
; RV32I-NEXT: mv a1, a2
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmovcc64:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: addi a3, zero, 123
; RV64I-NEXT: beq a0, a3, .LBB0_2
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: mv a1, a2
; RV64I-NEXT: .LBB0_2: # %entry
; RV64I-NEXT: mv a0, a1
; RV64I-NEXT: ret
entry:
%cmp = icmp eq i32 %a, 123
%cond = select i1 %cmp, i64 %b, i64 %c
ret i64 %cond
}
define i128 @cmovcc128(i64 signext %a, i128 %b, i128 %c) nounwind {
; RV32I-LABEL: cmovcc128:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: xori a1, a1, 123
; RV32I-NEXT: or a2, a1, a2
; RV32I-NEXT: mv a1, a3
; RV32I-NEXT: beqz a2, .LBB1_2
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: mv a1, a4
; RV32I-NEXT: .LBB1_2: # %entry
; RV32I-NEXT: lw a6, 0(a1)
; RV32I-NEXT: beqz a2, .LBB1_6
; RV32I-NEXT: # %bb.3: # %entry
; RV32I-NEXT: addi a1, a4, 4
; RV32I-NEXT: lw a5, 0(a1)
; RV32I-NEXT: bnez a2, .LBB1_7
; RV32I-NEXT: .LBB1_4:
; RV32I-NEXT: addi a1, a3, 8
; RV32I-NEXT: lw a1, 0(a1)
; RV32I-NEXT: bnez a2, .LBB1_8
; RV32I-NEXT: .LBB1_5:
; RV32I-NEXT: addi a2, a3, 12
; RV32I-NEXT: j .LBB1_9
; RV32I-NEXT: .LBB1_6:
; RV32I-NEXT: addi a1, a3, 4
; RV32I-NEXT: lw a5, 0(a1)
; RV32I-NEXT: beqz a2, .LBB1_4
; RV32I-NEXT: .LBB1_7: # %entry
; RV32I-NEXT: addi a1, a4, 8
; RV32I-NEXT: lw a1, 0(a1)
; RV32I-NEXT: beqz a2, .LBB1_5
; RV32I-NEXT: .LBB1_8: # %entry
; RV32I-NEXT: addi a2, a4, 12
; RV32I-NEXT: .LBB1_9: # %entry
; RV32I-NEXT: lw a2, 0(a2)
; RV32I-NEXT: sw a2, 12(a0)
; RV32I-NEXT: sw a1, 8(a0)
; RV32I-NEXT: sw a5, 4(a0)
; RV32I-NEXT: sw a6, 0(a0)
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmovcc128:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: addi a5, zero, 123
; RV64I-NEXT: beq a0, a5, .LBB1_2
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: mv a1, a3
; RV64I-NEXT: mv a2, a4
; RV64I-NEXT: .LBB1_2: # %entry
; RV64I-NEXT: mv a0, a1
; RV64I-NEXT: mv a1, a2
; RV64I-NEXT: ret
entry:
%cmp = icmp eq i64 %a, 123
%cond = select i1 %cmp, i128 %b, i128 %c
ret i128 %cond
}
define i64 @cmov64(i1 %a, i64 %b, i64 %c) nounwind {
; RV32I-LABEL: cmov64:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: andi a5, a0, 1
; RV32I-NEXT: mv a0, a1
; RV32I-NEXT: bnez a5, .LBB2_2
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: mv a0, a3
; RV32I-NEXT: mv a2, a4
; RV32I-NEXT: .LBB2_2: # %entry
; RV32I-NEXT: mv a1, a2
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmov64:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: andi a3, a0, 1
; RV64I-NEXT: mv a0, a1
; RV64I-NEXT: bnez a3, .LBB2_2
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: mv a0, a2
; RV64I-NEXT: .LBB2_2: # %entry
; RV64I-NEXT: ret
entry:
%cond = select i1 %a, i64 %b, i64 %c
ret i64 %cond
}
define i128 @cmov128(i1 %a, i128 %b, i128 %c) nounwind {
; RV32I-LABEL: cmov128:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: andi a4, a1, 1
; RV32I-NEXT: mv a1, a2
; RV32I-NEXT: bnez a4, .LBB3_2
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: mv a1, a3
; RV32I-NEXT: .LBB3_2: # %entry
; RV32I-NEXT: lw a6, 0(a1)
; RV32I-NEXT: bnez a4, .LBB3_6
; RV32I-NEXT: # %bb.3: # %entry
; RV32I-NEXT: addi a1, a3, 4
; RV32I-NEXT: lw a5, 0(a1)
; RV32I-NEXT: beqz a4, .LBB3_7
; RV32I-NEXT: .LBB3_4:
; RV32I-NEXT: addi a1, a2, 8
; RV32I-NEXT: lw a1, 0(a1)
; RV32I-NEXT: beqz a4, .LBB3_8
; RV32I-NEXT: .LBB3_5:
; RV32I-NEXT: addi a2, a2, 12
; RV32I-NEXT: j .LBB3_9
; RV32I-NEXT: .LBB3_6:
; RV32I-NEXT: addi a1, a2, 4
; RV32I-NEXT: lw a5, 0(a1)
; RV32I-NEXT: bnez a4, .LBB3_4
; RV32I-NEXT: .LBB3_7: # %entry
; RV32I-NEXT: addi a1, a3, 8
; RV32I-NEXT: lw a1, 0(a1)
; RV32I-NEXT: bnez a4, .LBB3_5
; RV32I-NEXT: .LBB3_8: # %entry
; RV32I-NEXT: addi a2, a3, 12
; RV32I-NEXT: .LBB3_9: # %entry
; RV32I-NEXT: lw a2, 0(a2)
; RV32I-NEXT: sw a2, 12(a0)
; RV32I-NEXT: sw a1, 8(a0)
; RV32I-NEXT: sw a5, 4(a0)
; RV32I-NEXT: sw a6, 0(a0)
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmov128:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: andi a5, a0, 1
; RV64I-NEXT: mv a0, a1
; RV64I-NEXT: bnez a5, .LBB3_2
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: mv a0, a3
; RV64I-NEXT: mv a2, a4
; RV64I-NEXT: .LBB3_2: # %entry
; RV64I-NEXT: mv a1, a2
; RV64I-NEXT: ret
entry:
%cond = select i1 %a, i128 %b, i128 %c
ret i128 %cond
}
define float @cmovfloat(i1 %a, float %b, float %c, float %d, float %e) nounwind {
; RV32I-LABEL: cmovfloat:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: andi a0, a0, 1
; RV32I-NEXT: bnez a0, .LBB4_2
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: fmv.w.x ft0, a4
; RV32I-NEXT: fmv.w.x ft1, a2
; RV32I-NEXT: j .LBB4_3
; RV32I-NEXT: .LBB4_2:
; RV32I-NEXT: fmv.w.x ft0, a3
; RV32I-NEXT: fmv.w.x ft1, a1
; RV32I-NEXT: .LBB4_3: # %entry
; RV32I-NEXT: fadd.s ft0, ft1, ft0
; RV32I-NEXT: fmv.x.w a0, ft0
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmovfloat:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: andi a0, a0, 1
; RV64I-NEXT: bnez a0, .LBB4_2
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: fmv.w.x ft0, a4
; RV64I-NEXT: fmv.w.x ft1, a2
; RV64I-NEXT: j .LBB4_3
; RV64I-NEXT: .LBB4_2:
; RV64I-NEXT: fmv.w.x ft0, a3
; RV64I-NEXT: fmv.w.x ft1, a1
; RV64I-NEXT: .LBB4_3: # %entry
; RV64I-NEXT: fadd.s ft0, ft1, ft0
; RV64I-NEXT: fmv.x.w a0, ft0
; RV64I-NEXT: ret
entry:
%cond1 = select i1 %a, float %b, float %c
%cond2 = select i1 %a, float %d, float %e
%ret = fadd float %cond1, %cond2
ret float %ret
}
define double @cmovdouble(i1 %a, double %b, double %c) nounwind {
; RV32I-LABEL: cmovdouble:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: addi sp, sp, -16
; RV32I-NEXT: sw a3, 8(sp)
; RV32I-NEXT: sw a4, 12(sp)
; RV32I-NEXT: fld ft0, 8(sp)
; RV32I-NEXT: sw a1, 8(sp)
; RV32I-NEXT: sw a2, 12(sp)
; RV32I-NEXT: fld ft1, 8(sp)
; RV32I-NEXT: andi a0, a0, 1
; RV32I-NEXT: bnez a0, .LBB5_2
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: fmv.d ft1, ft0
; RV32I-NEXT: .LBB5_2: # %entry
; RV32I-NEXT: fsd ft1, 8(sp)
; RV32I-NEXT: lw a0, 8(sp)
; RV32I-NEXT: lw a1, 12(sp)
; RV32I-NEXT: addi sp, sp, 16
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmovdouble:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: andi a0, a0, 1
; RV64I-NEXT: bnez a0, .LBB5_2
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: fmv.d.x ft0, a2
; RV64I-NEXT: fmv.x.d a0, ft0
; RV64I-NEXT: ret
; RV64I-NEXT: .LBB5_2:
; RV64I-NEXT: fmv.d.x ft0, a1
; RV64I-NEXT: fmv.x.d a0, ft0
; RV64I-NEXT: ret
entry:
%cond = select i1 %a, double %b, double %c
ret double %cond
}
; Check that selects with dependencies on previous ones aren't incorrectly
; optimized.
define i32 @cmovccdep(i32 signext %a, i32 %b, i32 %c, i32 %d) nounwind {
; RV32I-LABEL: cmovccdep:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: addi a4, zero, 123
; RV32I-NEXT: bne a0, a4, .LBB6_3
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: mv a2, a1
; RV32I-NEXT: bne a0, a4, .LBB6_4
; RV32I-NEXT: .LBB6_2: # %entry
; RV32I-NEXT: add a0, a1, a2
; RV32I-NEXT: ret
; RV32I-NEXT: .LBB6_3: # %entry
; RV32I-NEXT: mv a1, a2
; RV32I-NEXT: mv a2, a1
; RV32I-NEXT: beq a0, a4, .LBB6_2
; RV32I-NEXT: .LBB6_4: # %entry
; RV32I-NEXT: mv a2, a3
; RV32I-NEXT: add a0, a1, a2
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmovccdep:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: addi a4, zero, 123
; RV64I-NEXT: bne a0, a4, .LBB6_3
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: mv a2, a1
; RV64I-NEXT: bne a0, a4, .LBB6_4
; RV64I-NEXT: .LBB6_2: # %entry
; RV64I-NEXT: addw a0, a1, a2
; RV64I-NEXT: ret
; RV64I-NEXT: .LBB6_3: # %entry
; RV64I-NEXT: mv a1, a2
; RV64I-NEXT: mv a2, a1
; RV64I-NEXT: beq a0, a4, .LBB6_2
; RV64I-NEXT: .LBB6_4: # %entry
; RV64I-NEXT: mv a2, a3
; RV64I-NEXT: addw a0, a1, a2
; RV64I-NEXT: ret
entry:
%cmp = icmp eq i32 %a, 123
%cond1 = select i1 %cmp, i32 %b, i32 %c
%cond2 = select i1 %cmp, i32 %cond1, i32 %d
%ret = add i32 %cond1, %cond2
ret i32 %ret
}
; Check that selects with different conditions aren't incorrectly optimized.
define i32 @cmovdiffcc(i1 %a, i1 %b, i32 %c, i32 %d, i32 %e, i32 %f) nounwind {
; RV32I-LABEL: cmovdiffcc:
; RV32I: # %bb.0: # %entry
; RV32I-NEXT: andi a0, a0, 1
; RV32I-NEXT: andi a1, a1, 1
; RV32I-NEXT: beqz a0, .LBB7_3
; RV32I-NEXT: # %bb.1: # %entry
; RV32I-NEXT: beqz a1, .LBB7_4
; RV32I-NEXT: .LBB7_2: # %entry
; RV32I-NEXT: add a0, a2, a4
; RV32I-NEXT: ret
; RV32I-NEXT: .LBB7_3: # %entry
; RV32I-NEXT: mv a2, a3
; RV32I-NEXT: bnez a1, .LBB7_2
; RV32I-NEXT: .LBB7_4: # %entry
; RV32I-NEXT: mv a4, a5
; RV32I-NEXT: add a0, a2, a4
; RV32I-NEXT: ret
;
; RV64I-LABEL: cmovdiffcc:
; RV64I: # %bb.0: # %entry
; RV64I-NEXT: andi a0, a0, 1
; RV64I-NEXT: andi a1, a1, 1
; RV64I-NEXT: beqz a0, .LBB7_3
; RV64I-NEXT: # %bb.1: # %entry
; RV64I-NEXT: beqz a1, .LBB7_4
; RV64I-NEXT: .LBB7_2: # %entry
; RV64I-NEXT: addw a0, a2, a4
; RV64I-NEXT: ret
; RV64I-NEXT: .LBB7_3: # %entry
; RV64I-NEXT: mv a2, a3
; RV64I-NEXT: bnez a1, .LBB7_2
; RV64I-NEXT: .LBB7_4: # %entry
; RV64I-NEXT: mv a4, a5
; RV64I-NEXT: addw a0, a2, a4
; RV64I-NEXT: ret
entry:
%cond1 = select i1 %a, i32 %c, i32 %d
%cond2 = select i1 %b, i32 %e, i32 %f
%ret = add i32 %cond1, %cond2
ret i32 %ret
}
|