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
|
;; ARM Cortex-R4 scheduling description.
;; Copyright (C) 2007, 2008 Free Software Foundation, Inc.
;; Contributed by CodeSourcery.
;; This file is part of GCC.
;; GCC is free software; you can redistribute it and/or modify it
;; under the terms of the GNU General Public License as published
;; by the Free Software Foundation; either version 3, or (at your
;; option) any later version.
;; GCC is distributed in the hope that it will be useful, but WITHOUT
;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
;; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
;; License for more details.
;; You should have received a copy of the GNU General Public License
;; along with GCC; see the file COPYING3. If not see
;; <http://www.gnu.org/licenses/>.
(define_automaton "cortex_r4")
;; We approximate the dual-issue constraints of this core using four
;; "issue units" and a reservation matrix as follows. The numbers indicate
;; the instruction groups' preferences in order. Multiple entries for
;; the same numbered preference indicate units that must be reserved
;; together.
;;
;; Issue unit: A B C ALU
;;
;; ALU w/o reg shift 1st 2nd 1st and 2nd
;; ALU w/ reg shift 1st 2nd 2nd 1st and 2nd
;; Moves 1st 2nd 2nd
;; Multiplication 1st 1st
;; Division 1st 1st
;; Load/store single 1st 1st
;; Other load/store 1st 1st
;; Branches 1st
(define_cpu_unit "cortex_r4_issue_a" "cortex_r4")
(define_cpu_unit "cortex_r4_issue_b" "cortex_r4")
(define_cpu_unit "cortex_r4_issue_c" "cortex_r4")
(define_cpu_unit "cortex_r4_issue_alu" "cortex_r4")
(define_reservation "cortex_r4_alu"
"(cortex_r4_issue_a+cortex_r4_issue_alu)|\
(cortex_r4_issue_b+cortex_r4_issue_alu)")
(define_reservation "cortex_r4_alu_shift_reg"
"(cortex_r4_issue_a+cortex_r4_issue_alu)|\
(cortex_r4_issue_b+cortex_r4_issue_c+\
cortex_r4_issue_alu)")
(define_reservation "cortex_r4_mov"
"cortex_r4_issue_a|(cortex_r4_issue_b+\
cortex_r4_issue_alu)")
(define_reservation "cortex_r4_mul" "cortex_r4_issue_a+cortex_r4_issue_alu")
(define_reservation "cortex_r4_mul_2"
"(cortex_r4_issue_a+cortex_r4_issue_alu)*2")
;; Division instructions execute out-of-order with respect to the
;; rest of the pipeline and only require reservations on their first and
;; final cycles.
(define_reservation "cortex_r4_div_9"
"cortex_r4_issue_a+cortex_r4_issue_alu,\
nothing*7,\
cortex_r4_issue_a+cortex_r4_issue_alu")
(define_reservation "cortex_r4_div_10"
"cortex_r4_issue_a+cortex_r4_issue_alu,\
nothing*8,\
cortex_r4_issue_a+cortex_r4_issue_alu")
(define_reservation "cortex_r4_load_store"
"cortex_r4_issue_a+cortex_r4_issue_c")
(define_reservation "cortex_r4_load_store_2"
"(cortex_r4_issue_a+cortex_r4_issue_b)*2")
(define_reservation "cortex_r4_branch" "cortex_r4_issue_b")
;; We assume that all instructions are unconditional.
;; Data processing instructions. Moves without shifts are kept separate
;; for the purposes of the dual-issue constraints above.
(define_insn_reservation "cortex_r4_alu" 2
(and (eq_attr "tune_cortexr4" "yes")
(and (eq_attr "type" "alu")
(not (eq_attr "insn" "mov"))))
"cortex_r4_alu")
(define_insn_reservation "cortex_r4_mov" 2
(and (eq_attr "tune_cortexr4" "yes")
(and (eq_attr "type" "alu")
(eq_attr "insn" "mov")))
"cortex_r4_mov")
(define_insn_reservation "cortex_r4_alu_shift" 2
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "alu_shift"))
"cortex_r4_alu")
(define_insn_reservation "cortex_r4_alu_shift_reg" 2
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "alu_shift_reg"))
"cortex_r4_alu_shift_reg")
;; An ALU instruction followed by an ALU instruction with no early dep.
(define_bypass 1 "cortex_r4_alu,cortex_r4_alu_shift,cortex_r4_alu_shift_reg,\
cortex_r4_mov"
"cortex_r4_alu")
(define_bypass 1 "cortex_r4_alu,cortex_r4_alu_shift,cortex_r4_alu_shift_reg,\
cortex_r4_mov"
"cortex_r4_alu_shift"
"arm_no_early_alu_shift_dep")
(define_bypass 1 "cortex_r4_alu,cortex_r4_alu_shift,cortex_r4_alu_shift_reg,\
cortex_r4_mov"
"cortex_r4_alu_shift_reg"
"arm_no_early_alu_shift_value_dep")
;; In terms of availabilities, a consumer mov could theoretically be
;; issued together with a producer ALU instruction, without stalls.
;; In practice this cannot happen because mov;add (in that order) is not
;; eligible for dual issue and furthermore dual issue is not permitted
;; when a dependency is involved. We therefore note it as latency one.
;; A mov followed by another of the same is also latency one.
(define_bypass 1 "cortex_r4_alu,cortex_r4_alu_shift,cortex_r4_alu_shift_reg,\
cortex_r4_mov"
"cortex_r4_mov")
;; qadd, qdadd, qsub and qdsub are not currently emitted, and neither are
;; media data processing instructions nor sad instructions.
;; Multiplication instructions.
(define_insn_reservation "cortex_r4_mul_4" 4
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "mul,smmul"))
"cortex_r4_mul_2")
(define_insn_reservation "cortex_r4_mul_3" 3
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smulxy,smulwy,smuad,smusd"))
"cortex_r4_mul")
(define_insn_reservation "cortex_r4_mla_4" 4
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "mla,smmla"))
"cortex_r4_mul_2")
(define_insn_reservation "cortex_r4_mla_3" 3
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smlaxy,smlawy,smlad,smlsd"))
"cortex_r4_mul")
(define_insn_reservation "cortex_r4_smlald" 3
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smlald,smlsld"))
"cortex_r4_mul")
(define_insn_reservation "cortex_r4_mull" 4
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "smull,umull,umlal,umaal"))
"cortex_r4_mul_2")
;; A multiply or an MLA with a single-register result, followed by an
;; MLA with an accumulator dependency, has its result forwarded.
(define_bypass 2 "cortex_r4_mul_3,cortex_r4_mla_3"
"cortex_r4_mla_3,cortex_r4_mla_4"
"arm_mac_accumulator_is_mul_result")
(define_bypass 3 "cortex_r4_mul_4,cortex_r4_mla_4"
"cortex_r4_mla_3,cortex_r4_mla_4"
"arm_mac_accumulator_is_mul_result")
;; A multiply followed by an ALU instruction needing the multiply
;; result only at ALU has lower latency than one needing it at Shift.
(define_bypass 2 "cortex_r4_mul_3,cortex_r4_mla_3,cortex_r4_smlald"
"cortex_r4_alu")
(define_bypass 2 "cortex_r4_mul_3,cortex_r4_mla_3,cortex_r4_smlald"
"cortex_r4_alu_shift"
"arm_no_early_alu_shift_dep")
(define_bypass 2 "cortex_r4_mul_3,cortex_r4_mla_3,cortex_r4_smlald"
"cortex_r4_alu_shift_reg"
"arm_no_early_alu_shift_value_dep")
(define_bypass 3 "cortex_r4_mul_4,cortex_r4_mla_4,cortex_r4_mull"
"cortex_r4_alu")
(define_bypass 3 "cortex_r4_mul_4,cortex_r4_mla_4,cortex_r4_mull"
"cortex_r4_alu_shift"
"arm_no_early_alu_shift_dep")
(define_bypass 3 "cortex_r4_mul_4,cortex_r4_mla_4,cortex_r4_mull"
"cortex_r4_alu_shift_reg"
"arm_no_early_alu_shift_value_dep")
;; A multiply followed by a mov has one cycle lower latency again.
(define_bypass 1 "cortex_r4_mul_3,cortex_r4_mla_3,cortex_r4_smlald"
"cortex_r4_mov")
(define_bypass 2 "cortex_r4_mul_4,cortex_r4_mla_4,cortex_r4_mull"
"cortex_r4_mov")
;; We guess that division of A/B using sdiv or udiv, on average,
;; is performed with B having ten more leading zeros than A.
;; This gives a latency of nine for udiv and ten for sdiv.
(define_insn_reservation "cortex_r4_udiv" 9
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "udiv"))
"cortex_r4_div_9")
(define_insn_reservation "cortex_r4_sdiv" 10
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "insn" "sdiv"))
"cortex_r4_div_10")
;; Branches. We assume correct prediction.
(define_insn_reservation "cortex_r4_branch" 0
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "branch"))
"cortex_r4_branch")
;; Call latencies are not predictable. A semi-arbitrary very large
;; number is used as "positive infinity" so that everything should be
;; finished by the time of return.
(define_insn_reservation "cortex_r4_call" 32
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "call"))
"nothing")
;; Status register access instructions are not currently emitted.
;; Load instructions.
;; We do not model the "addr_md_3cycle" cases and assume that
;; accesses following are correctly aligned.
(define_insn_reservation "cortex_r4_load_1_2" 3
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "load1,load2"))
"cortex_r4_load_store")
(define_insn_reservation "cortex_r4_load_3_4" 4
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "load3,load4"))
"cortex_r4_load_store_2")
;; If a producing load is followed by an instruction consuming only
;; as a Normal Reg, there is one fewer cycle of latency.
(define_bypass 2 "cortex_r4_load_1_2"
"cortex_r4_alu")
(define_bypass 2 "cortex_r4_load_1_2"
"cortex_r4_alu_shift"
"arm_no_early_alu_shift_dep")
(define_bypass 2 "cortex_r4_load_1_2"
"cortex_r4_alu_shift_reg"
"arm_no_early_alu_shift_value_dep")
(define_bypass 3 "cortex_r4_load_3_4"
"cortex_r4_alu")
(define_bypass 3 "cortex_r4_load_3_4"
"cortex_r4_alu_shift"
"arm_no_early_alu_shift_dep")
(define_bypass 3 "cortex_r4_load_3_4"
"cortex_r4_alu_shift_reg"
"arm_no_early_alu_shift_value_dep")
;; If a producing load is followed by an instruction consuming only
;; as a Late Reg, there are two fewer cycles of latency. Such consumer
;; instructions are moves and stores.
(define_bypass 1 "cortex_r4_load_1_2"
"cortex_r4_mov,cortex_r4_store_1_2,cortex_r4_store_3_4")
(define_bypass 2 "cortex_r4_load_3_4"
"cortex_r4_mov,cortex_r4_store_1_2,cortex_r4_store_3_4")
;; If a producer's result is required as the base or offset of a load,
;; there is an extra cycle latency.
(define_bypass 3 "cortex_r4_alu,cortex_r4_mov,cortex_r4_alu_shift,\
cortex_r4_alu_shift_reg"
"cortex_r4_load_1_2,cortex_r4_load_3_4")
(define_bypass 4 "cortex_r4_mul_3,cortex_r4_mla_3,cortex_r4_smlald"
"cortex_r4_load_1_2,cortex_r4_load_3_4")
(define_bypass 5 "cortex_r4_mul_4,cortex_r4_mla_4,cortex_r4_mull"
"cortex_r4_load_1_2,cortex_r4_load_3_4")
;; Store instructions.
(define_insn_reservation "cortex_r4_store_1_2" 0
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "store1,store2"))
"cortex_r4_load_store")
(define_insn_reservation "cortex_r4_store_3_4" 0
(and (eq_attr "tune_cortexr4" "yes")
(eq_attr "type" "store3,store4"))
"cortex_r4_load_store_2")
|
