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,insert_dword,cmp,compare,\
89 delayed_compare,var_delayed_compare,fast_compare,\
90 shift,trap,var_shift_rotate,cntlz,exts")
91 (eq_attr "cpu" "ppc8540"))
92 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
94 (define_insn_reservation "ppc8540_two" 1
95 (and (eq_attr "type" "two")
96 (eq_attr "cpu" "ppc8540"))
97 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire,\
98 ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
100 (define_insn_reservation "ppc8540_three" 1
101 (and (eq_attr "type" "three")
102 (eq_attr "cpu" "ppc8540"))
103 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire,\
104 ppc8540_issue+ppc8540_su_stage0+ppc8540_retire,\
105 ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
107 ;; Branch. Actually this latency time is not used by the scheduler.
108 (define_insn_reservation "ppc8540_branch" 1
109 (and (eq_attr "type" "jmpreg,branch,isync")
110 (eq_attr "cpu" "ppc8540"))
111 "ppc8540_decode,ppc8540_bu,ppc8540_retire")
114 (define_insn_reservation "ppc8540_multiply" 4
115 (and (eq_attr "type" "imul,imul2,imul3,imul_compare")
116 (eq_attr "cpu" "ppc8540"))
117 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
118 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
120 ;; Divide. We use the average latency time here. We omit reserving a
121 ;; retire unit because of the result automata will be huge. We ignore
122 ;; reservation of miu_stage3 here because we use the average latency
124 (define_insn_reservation "ppc8540_divide" 14
125 (and (eq_attr "type" "idiv")
126 (eq_attr "cpu" "ppc8540"))
127 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0+ppc8540_mu_div,\
131 (define_insn_reservation "ppc8540_cr_logical" 1
132 (and (eq_attr "type" "cr_logical,delayed_cr")
133 (eq_attr "cpu" "ppc8540"))
134 "ppc8540_decode,ppc8540_bu,ppc8540_retire")
137 (define_insn_reservation "ppc8540_mfcr" 1
138 (and (eq_attr "type" "mfcr")
139 (eq_attr "cpu" "ppc8540"))
140 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
143 (define_insn_reservation "ppc8540_mtcrf" 1
144 (and (eq_attr "type" "mtcr")
145 (eq_attr "cpu" "ppc8540"))
146 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
149 (define_insn_reservation "ppc8540_mtjmpr" 1
150 (and (eq_attr "type" "mtjmpr,mfjmpr")
151 (eq_attr "cpu" "ppc8540"))
152 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
155 (define_insn_reservation "ppc8540_load" 3
156 (and (eq_attr "type" "load,load_ext,load_ext_u,load_ext_ux,load_ux,load_u,\
158 (eq_attr "cpu" "ppc8540"))
159 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")
162 (define_insn_reservation "ppc8540_store" 3
163 (and (eq_attr "type" "store,store_ux,store_u,store_c")
164 (eq_attr "cpu" "ppc8540"))
165 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")
168 (define_insn_reservation "ppc8540_simple_float" 1
169 (and (eq_attr "type" "fpsimple")
170 (eq_attr "cpu" "ppc8540"))
171 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
174 (define_insn_reservation "ppc8540_float" 4
175 (and (eq_attr "type" "fp")
176 (eq_attr "cpu" "ppc8540"))
177 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
178 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
180 ;; float divides. We omit reserving a retire unit and miu_stage3
181 ;; because of the result automata will be huge.
182 (define_insn_reservation "ppc8540_float_vector_divide" 29
183 (and (eq_attr "type" "vecfdiv")
184 (eq_attr "cpu" "ppc8540"))
185 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0+ppc8540_mu_div,\
189 (define_insn_reservation "ppc8540_brinc" 1
190 (and (eq_attr "type" "brinc")
191 (eq_attr "cpu" "ppc8540"))
192 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
195 (define_insn_reservation "ppc8540_simple_vector" 1
196 (and (eq_attr "type" "vecsimple")
197 (eq_attr "cpu" "ppc8540"))
198 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
200 ;; Simple vector compare
201 (define_insn_reservation "ppc8540_simple_vector_compare" 1
202 (and (eq_attr "type" "veccmpsimple")
203 (eq_attr "cpu" "ppc8540"))
204 "ppc8540_decode,ppc8540_issue+ppc8540_su_stage0+ppc8540_retire")
207 (define_insn_reservation "ppc8540_vector_compare" 1
208 (and (eq_attr "type" "veccmp")
209 (eq_attr "cpu" "ppc8540"))
210 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
212 ;; evsplatfi evsplati
213 (define_insn_reservation "ppc8540_vector_perm" 1
214 (and (eq_attr "type" "vecperm")
215 (eq_attr "cpu" "ppc8540"))
216 "ppc8540_decode,ppc8540_issue+ppc8540_su1_stage0+ppc8540_retire")
219 (define_insn_reservation "ppc8540_float_vector" 4
220 (and (eq_attr "type" "vecfloat")
221 (eq_attr "cpu" "ppc8540"))
222 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
223 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
225 ;; Vector divides: Use the average. We omit reserving a retire unit
226 ;; because of the result automata will be huge. We ignore reservation
227 ;; of miu_stage3 here because we use the average latency time.
228 (define_insn_reservation "ppc8540_vector_divide" 14
229 (and (eq_attr "type" "vecdiv")
230 (eq_attr "cpu" "ppc8540"))
231 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0+ppc8540_mu_div,\
235 (define_insn_reservation "ppc8540_complex_vector" 4
236 (and (eq_attr "type" "veccomplex")
237 (eq_attr "cpu" "ppc8540"))
238 "ppc8540_decode,ppc8540_issue+ppc8540_mu_stage0,ppc8540_mu_stage1,\
239 ppc8540_mu_stage2,ppc8540_mu_stage3+ppc8540_retire")
242 (define_insn_reservation "ppc8540_vector_load" 3
243 (and (eq_attr "type" "vecload")
244 (eq_attr "cpu" "ppc8540"))
245 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")
248 (define_insn_reservation "ppc8540_vector_store" 3
249 (and (eq_attr "type" "vecstore")
250 (eq_attr "cpu" "ppc8540"))
251 "ppc8540_decode,ppc8540_issue+ppc8540_lsu,nothing,ppc8540_retire")