1 ;; Pipeline description for Motorola PowerPC 8540 processor.
2 ;; Copyright (C) 2003, 2004 Free Software Foundation, Inc.
4 ;; This file is part of GCC.
6 ;; GCC is free software; you can redistribute it and/or modify it
7 ;; under the terms of the GNU General Public License as published
8 ;; by the Free Software Foundation; either version 2, or (at your
9 ;; option) any later version.
11 ;; GCC is distributed in the hope that it will be useful, but WITHOUT
12 ;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
13 ;; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
14 ;; License for more details.
16 ;; You should have received a copy of the GNU General Public License
17 ;; along with GCC; see the file COPYING. If not, write to the
18 ;; Free Software Foundation, 51 Franklin Street, Fifth Floor, Boston,
19 ;; MA 02110-1301, USA.
21 (define_automaton "ppc8540_most,ppc8540_long,ppc8540_retire")
22 (define_cpu_unit "ppc8540_decode_0,ppc8540_decode_1" "ppc8540_most")
24 ;; We don't simulate general issue queue (GIC). If we have SU insn
25 ;; and then SU1 insn, they cannot be issued on the same cycle
26 ;; (although SU1 insn and then SU insn can be issued) because the SU
27 ;; insn will go to SU1 from GIC0 entry. Fortunately, the first cycle
28 ;; multipass insn scheduling will find the situation and issue the SU1
29 ;; insn and then the SU insn.
30 (define_cpu_unit "ppc8540_issue_0,ppc8540_issue_1" "ppc8540_most")
32 ;; We could describe completion buffers slots in combination with the
33 ;; retirement units and the order of completion but the result
34 ;; automaton would behave in the same way because we cannot describe
35 ;; real latency time with taking in order completion into account.
36 ;; Actually we could define the real latency time by querying reserved
37 ;; automaton units but the current scheduler uses latency time before
38 ;; issuing insns and making any reservations.
40 ;; So our description is aimed to achieve a insn schedule in which the
41 ;; insns would not wait in the completion buffer.
42 (define_cpu_unit "ppc8540_retire_0,ppc8540_retire_1" "ppc8540_retire")
45 (define_cpu_unit "ppc8540_bu" "ppc8540_most")
48 (define_cpu_unit "ppc8540_su0_stage0,ppc8540_su1_stage0" "ppc8540_most")
50 ;; We could describe here MU subunits for float multiply, float add
51 ;; etc. But the result automaton would behave the same way as the
52 ;; described one pipeline below because MU can start only one insn
53 ;; per cycle. Actually we could simplify the automaton more not
54 ;; describing stages 1-3, the result automata would be the same.
55 (define_cpu_unit "ppc8540_mu_stage0,ppc8540_mu_stage1" "ppc8540_most")
56 (define_cpu_unit "ppc8540_mu_stage2,ppc8540_mu_stage3" "ppc8540_most")
58 ;; The following unit is used to describe non-pipelined division.
59 (define_cpu_unit "ppc8540_mu_div" "ppc8540_long")
61 ;; Here we simplified LSU unit description not describing the stages.
62 (define_cpu_unit "ppc8540_lsu" "ppc8540_most")
64 ;; The following units are used to make automata deterministic
65 (define_cpu_unit "present_ppc8540_decode_0" "ppc8540_most")
66 (define_cpu_unit "present_ppc8540_issue_0" "ppc8540_most")
67 (define_cpu_unit "present_ppc8540_retire_0" "ppc8540_retire")
68 (define_cpu_unit "present_ppc8540_su0_stage0" "ppc8540_most")
70 ;; The following sets to make automata deterministic when option ndfa is used.
71 (presence_set "present_ppc8540_decode_0" "ppc8540_decode_0")
72 (presence_set "present_ppc8540_issue_0" "ppc8540_issue_0")
73 (presence_set "present_ppc8540_retire_0" "ppc8540_retire_0")
74 (presence_set "present_ppc8540_su0_stage0" "ppc8540_su0_stage0")
76 ;; Some useful abbreviations.
77 (define_reservation "ppc8540_decode"
78 "ppc8540_decode_0|ppc8540_decode_1+present_ppc8540_decode_0")
79 (define_reservation "ppc8540_issue"
80 "ppc8540_issue_0|ppc8540_issue_1+present_ppc8540_issue_0")
81 (define_reservation "ppc8540_retire"
82 "ppc8540_retire_0|ppc8540_retire_1+present_ppc8540_retire_0")
83 (define_reservation "ppc8540_su_stage0"
84 "ppc8540_su0_stage0|ppc8540_su1_stage0+present_ppc8540_su0_stage0")
87 (define_insn_reservation "ppc8540_su" 1
88 (and (eq_attr "type" "integer,insert_word,cmp,compare,delayed_compare,fast_compare")
89 (eq_attr "cpu" "ppc8540"))
90 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
92 (define_insn_reservation "ppc8540_two" 1
93 (and (eq_attr "type" "two")
94 (eq_attr "cpu" "ppc8540"))
95 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire,\
96 ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
98 (define_insn_reservation "ppc8540_three" 1
99 (and (eq_attr "type" "three")
100 (eq_attr "cpu" "ppc8540"))
101 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire,\
102 ppc8540_issue+ppc8540_su_stage0+ppc8540_retire,\
103 ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
105 ;; Branch. Actually this latency time is not used by the scheduler.
106 (define_insn_reservation "ppc8540_branch" 1
107 (and (eq_attr "type" "jmpreg,branch,isync")
108 (eq_attr "cpu" "ppc8540"))
109 "ppc8540_decode,ppc8540_bu,ppc8540_retire")
112 (define_insn_reservation "ppc8540_multiply" 4
113 (and (eq_attr "type" "imul,imul2,imul3,imul_compare")
114 (eq_attr "cpu" "ppc8540"))
115 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
116 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
118 ;; Divide. We use the average latency time here. We omit reserving a
119 ;; retire unit because of the result automata will be huge. We ignore
120 ;; reservation of miu_stage3 here because we use the average latency
122 (define_insn_reservation "ppc8540_divide" 14
123 (and (eq_attr "type" "idiv")
124 (eq_attr "cpu" "ppc8540"))
125 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0+ppc8540_mu_div,\
129 (define_insn_reservation "ppc8540_cr_logical" 1
130 (and (eq_attr "type" "cr_logical,delayed_cr")
131 (eq_attr "cpu" "ppc8540"))
132 "ppc8540_decode,ppc8540_bu,ppc8540_retire")
135 (define_insn_reservation "ppc8540_mfcr" 1
136 (and (eq_attr "type" "mfcr")
137 (eq_attr "cpu" "ppc8540"))
138 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
141 (define_insn_reservation "ppc8540_mtcrf" 1
142 (and (eq_attr "type" "mtcr")
143 (eq_attr "cpu" "ppc8540"))
144 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
147 (define_insn_reservation "ppc8540_mtjmpr" 1
148 (and (eq_attr "type" "mtjmpr,mfjmpr")
149 (eq_attr "cpu" "ppc8540"))
150 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
153 (define_insn_reservation "ppc8540_load" 3
154 (and (eq_attr "type" "load,load_ext,load_ext_u,load_ext_ux,load_ux,load_u,\
156 (eq_attr "cpu" "ppc8540"))
157 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")
160 (define_insn_reservation "ppc8540_store" 3
161 (and (eq_attr "type" "store,store_ux,store_u,store_c")
162 (eq_attr "cpu" "ppc8540"))
163 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")
166 (define_insn_reservation "ppc8540_simple_float" 1
167 (and (eq_attr "type" "fpsimple")
168 (eq_attr "cpu" "ppc8540"))
169 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
172 (define_insn_reservation "ppc8540_float" 4
173 (and (eq_attr "type" "fp")
174 (eq_attr "cpu" "ppc8540"))
175 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
176 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
178 ;; float divides. We omit reserving a retire unit and miu_stage3
179 ;; because of the result automata will be huge.
180 (define_insn_reservation "ppc8540_float_vector_divide" 29
181 (and (eq_attr "type" "vecfdiv")
182 (eq_attr "cpu" "ppc8540"))
183 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0+ppc8540_mu_div,\
187 (define_insn_reservation "ppc8540_brinc" 1
188 (and (eq_attr "type" "brinc")
189 (eq_attr "cpu" "ppc8540"))
190 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
193 (define_insn_reservation "ppc8540_simple_vector" 1
194 (and (eq_attr "type" "vecsimple")
195 (eq_attr "cpu" "ppc8540"))
196 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
198 ;; Simple vector compare
199 (define_insn_reservation "ppc8540_simple_vector_compare" 1
200 (and (eq_attr "type" "veccmpsimple")
201 (eq_attr "cpu" "ppc8540"))
202 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
205 (define_insn_reservation "ppc8540_vector_compare" 1
206 (and (eq_attr "type" "veccmp")
207 (eq_attr "cpu" "ppc8540"))
208 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
210 ;; evsplatfi evsplati
211 (define_insn_reservation "ppc8540_vector_perm" 1
212 (and (eq_attr "type" "vecperm")
213 (eq_attr "cpu" "ppc8540"))
214 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
217 (define_insn_reservation "ppc8540_float_vector" 4
218 (and (eq_attr "type" "vecfloat")
219 (eq_attr "cpu" "ppc8540"))
220 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
221 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
223 ;; Vector divides: Use the average. We omit reserving a retire unit
224 ;; because of the result automata will be huge. We ignore reservation
225 ;; of miu_stage3 here because we use the average latency time.
226 (define_insn_reservation "ppc8540_vector_divide" 14
227 (and (eq_attr "type" "vecdiv")
228 (eq_attr "cpu" "ppc8540"))
229 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0+ppc8540_mu_div,\
233 (define_insn_reservation "ppc8540_complex_vector" 4
234 (and (eq_attr "type" "veccomplex")
235 (eq_attr "cpu" "ppc8540"))
236 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
237 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
240 (define_insn_reservation "ppc8540_vector_load" 3
241 (and (eq_attr "type" "vecload")
242 (eq_attr "cpu" "ppc8540"))
243 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")
246 (define_insn_reservation "ppc8540_vector_store" 3
247 (and (eq_attr "type" "vecstore")
248 (eq_attr "cpu" "ppc8540"))
249 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")