1 |
709 |
jeremybenn |
;; Copyright (C) 2007 Free Software Foundation, Inc.
|
2 |
|
|
;;
|
3 |
|
|
;; This file is part of GCC.
|
4 |
|
|
;;
|
5 |
|
|
;; GCC is free software; you can redistribute it and/or modify
|
6 |
|
|
;; it under the terms of the GNU General Public License as published by
|
7 |
|
|
;; the Free Software Foundation; either version 3, or (at your option)
|
8 |
|
|
;; any later version.
|
9 |
|
|
;;
|
10 |
|
|
;; GCC is distributed in the hope that it will be useful,
|
11 |
|
|
;; but WITHOUT ANY WARRANTY; without even the implied warranty of
|
12 |
|
|
;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
13 |
|
|
;; GNU General Public License for more details.
|
14 |
|
|
;;
|
15 |
|
|
;; You should have received a copy of the GNU General Public License
|
16 |
|
|
;; along with GCC; see the file COPYING3. If not see
|
17 |
|
|
;; . */
|
18 |
|
|
;;
|
19 |
|
|
;; .........................
|
20 |
|
|
;;
|
21 |
|
|
;; DFA-based pipeline description for MIPS64 model R20Kc.
|
22 |
|
|
;; Contributed by Jason Eckhardt (jle@cygnus.com).
|
23 |
|
|
;;
|
24 |
|
|
;; The R20Kc is a dual-issue processor that can generally bundle
|
25 |
|
|
;; instructions as follows:
|
26 |
|
|
;; 1. integer with integer
|
27 |
|
|
;; 2. integer with fp
|
28 |
|
|
;; 3. fp with fpload/fpstore
|
29 |
|
|
;;
|
30 |
|
|
;; Of course, there are various restrictions.
|
31 |
|
|
;; Reference:
|
32 |
|
|
;; "Ruby (R20K) Technical Specification Rev. 1.2, December 28, 1999."
|
33 |
|
|
;;
|
34 |
|
|
;; .........................
|
35 |
|
|
|
36 |
|
|
;; Use three automata to isolate long latency operations, reducing space.
|
37 |
|
|
(define_automaton "r20kc_other, r20kc_fdiv, r20kc_idiv")
|
38 |
|
|
|
39 |
|
|
;;
|
40 |
|
|
;; Describe the resources.
|
41 |
|
|
;;
|
42 |
|
|
|
43 |
|
|
;; Global.
|
44 |
|
|
(define_cpu_unit "r20kc_iss0, r20kc_iss1" "r20kc_other")
|
45 |
|
|
|
46 |
|
|
;; Integer execution unit (pipeline A).
|
47 |
|
|
(define_cpu_unit "r20kc_ixua_addsub_agen" "r20kc_other")
|
48 |
|
|
(define_cpu_unit "r20kc_ixua_shift" "r20kc_other")
|
49 |
|
|
|
50 |
|
|
(exclusion_set "r20kc_ixua_addsub_agen" "r20kc_ixua_shift")
|
51 |
|
|
|
52 |
|
|
;; Integer execution unit (pipeline B).
|
53 |
|
|
(define_cpu_unit "r20kc_ixub_addsub" "r20kc_other")
|
54 |
|
|
(define_cpu_unit "r20kc_ixub_branch" "r20kc_other")
|
55 |
|
|
(define_cpu_unit "r20kc_ixub_mpydiv" "r20kc_other")
|
56 |
|
|
(define_cpu_unit "r20kc_ixub_mpydiv_iter" "r20kc_idiv")
|
57 |
|
|
|
58 |
|
|
(exclusion_set "r20kc_ixub_addsub" "r20kc_ixub_branch, r20kc_ixub_mpydiv")
|
59 |
|
|
(exclusion_set "r20kc_ixub_branch" "r20kc_ixub_mpydiv")
|
60 |
|
|
|
61 |
|
|
;; Cache / memory interface.
|
62 |
|
|
(define_cpu_unit "r20kc_cache" "r20kc_other")
|
63 |
|
|
|
64 |
|
|
;; Floating-point unit.
|
65 |
|
|
(define_cpu_unit "r20kc_fpu_add" "r20kc_other")
|
66 |
|
|
(define_cpu_unit "r20kc_fpu_mpy" "r20kc_other")
|
67 |
|
|
(define_cpu_unit "r20kc_fpu_mpy_iter" "r20kc_fdiv")
|
68 |
|
|
(define_cpu_unit "r20kc_fpu_divsqrt" "r20kc_other")
|
69 |
|
|
(define_cpu_unit "r20kc_fpu_divsqrt_iter" "r20kc_fdiv")
|
70 |
|
|
|
71 |
|
|
(exclusion_set "r20kc_fpu_add" "r20kc_fpu_mpy, r20kc_fpu_divsqrt")
|
72 |
|
|
(exclusion_set "r20kc_fpu_mpy" "r20kc_fpu_divsqrt")
|
73 |
|
|
|
74 |
|
|
;; After branch any insn can not be issued.
|
75 |
|
|
(absence_set "r20kc_iss0,r20kc_iss1" "r20kc_ixub_branch")
|
76 |
|
|
|
77 |
|
|
;;
|
78 |
|
|
;; Define reservations for unit name mnemonics or combinations.
|
79 |
|
|
;;
|
80 |
|
|
|
81 |
|
|
(define_reservation "r20kc_iss"
|
82 |
|
|
"r20kc_iss0|r20kc_iss1")
|
83 |
|
|
(define_reservation "r20kc_single_dispatch"
|
84 |
|
|
"r20kc_iss0+r20kc_iss1")
|
85 |
|
|
(define_reservation "r20kc_iaddsub"
|
86 |
|
|
"r20kc_iss+(r20kc_ixua_addsub_agen|r20kc_ixub_addsub)")
|
87 |
|
|
(define_reservation "r20kc_ishift"
|
88 |
|
|
"r20kc_iss+r20kc_ixua_shift")
|
89 |
|
|
(define_reservation "r20kc_fpmove"
|
90 |
|
|
"r20kc_iss+r20kc_ixua_addsub_agen")
|
91 |
|
|
(define_reservation "r20kc_imem"
|
92 |
|
|
"r20kc_iss+r20kc_ixua_addsub_agen+r20kc_cache")
|
93 |
|
|
(define_reservation "r20kc_icache"
|
94 |
|
|
"r20kc_cache")
|
95 |
|
|
(define_reservation "r20kc_impydiv"
|
96 |
|
|
"r20kc_iss+r20kc_ixub_mpydiv")
|
97 |
|
|
(define_reservation "r20kc_impydiv_iter"
|
98 |
|
|
"r20kc_ixub_mpydiv_iter")
|
99 |
|
|
(define_reservation "r20kc_ibranch"
|
100 |
|
|
"r20kc_iss+r20kc_ixub_branch")
|
101 |
|
|
|
102 |
|
|
(define_reservation "r20kc_fpadd"
|
103 |
|
|
"r20kc_iss+r20kc_fpu_add")
|
104 |
|
|
(define_reservation "r20kc_fpmpy"
|
105 |
|
|
"r20kc_iss+r20kc_fpu_mpy")
|
106 |
|
|
(define_reservation "r20kc_fpmpy_iter"
|
107 |
|
|
"r20kc_fpu_mpy_iter")
|
108 |
|
|
(define_reservation "r20kc_fpdivsqrt"
|
109 |
|
|
"r20kc_iss+r20kc_fpu_divsqrt")
|
110 |
|
|
(define_reservation "r20kc_fpdivsqrt_iter"
|
111 |
|
|
"r20kc_fpu_divsqrt_iter")
|
112 |
|
|
|
113 |
|
|
;;
|
114 |
|
|
;; Describe instruction reservations for integer operations.
|
115 |
|
|
;;
|
116 |
|
|
|
117 |
|
|
;; Conditional moves always force single-dispatch.
|
118 |
|
|
(define_insn_reservation "r20kc_cond_move_int" 1
|
119 |
|
|
(and (eq_attr "cpu" "20kc")
|
120 |
|
|
(and (eq_attr "type" "condmove")
|
121 |
|
|
(eq_attr "mode" "!SF,DF")))
|
122 |
|
|
"r20kc_single_dispatch")
|
123 |
|
|
|
124 |
|
|
(define_insn_reservation "r20kc_cond_move_fp" 4
|
125 |
|
|
(and (eq_attr "cpu" "20kc")
|
126 |
|
|
(and (eq_attr "type" "condmove")
|
127 |
|
|
(eq_attr "mode" "SF,DF")))
|
128 |
|
|
"r20kc_single_dispatch")
|
129 |
|
|
|
130 |
|
|
(define_insn_reservation "r20kc_int_other" 1
|
131 |
|
|
(and (eq_attr "cpu" "20kc")
|
132 |
|
|
(eq_attr "type" "move,arith,const,nop"))
|
133 |
|
|
"r20kc_iaddsub")
|
134 |
|
|
|
135 |
|
|
;; Shifts can only execute on ixu pipeline A.
|
136 |
|
|
(define_insn_reservation "r20kc_int_shift" 1
|
137 |
|
|
(and (eq_attr "cpu" "20kc")
|
138 |
|
|
(eq_attr "type" "shift"))
|
139 |
|
|
"r20kc_ishift")
|
140 |
|
|
|
141 |
|
|
(define_insn_reservation "r20kc_ld" 2
|
142 |
|
|
(and (eq_attr "cpu" "20kc")
|
143 |
|
|
(eq_attr "type" "load,prefetch,prefetchx"))
|
144 |
|
|
"r20kc_imem")
|
145 |
|
|
|
146 |
|
|
|
147 |
|
|
;; A load immediately following a store will stall, so
|
148 |
|
|
;; say that a store uses the cache for an extra cycle.
|
149 |
|
|
(define_insn_reservation "r20kc_st" 2
|
150 |
|
|
(and (eq_attr "cpu" "20kc")
|
151 |
|
|
(eq_attr "type" "store"))
|
152 |
|
|
"r20kc_imem,r20kc_icache")
|
153 |
|
|
|
154 |
|
|
(define_insn_reservation "r20kc_fld" 3
|
155 |
|
|
(and (eq_attr "cpu" "20kc")
|
156 |
|
|
(eq_attr "type" "fpload"))
|
157 |
|
|
"r20kc_imem")
|
158 |
|
|
|
159 |
|
|
(define_insn_reservation "r20kc_ffst" 3
|
160 |
|
|
(and (eq_attr "cpu" "20kc")
|
161 |
|
|
(eq_attr "type" "fpstore"))
|
162 |
|
|
"r20kc_imem,r20kc_icache*2")
|
163 |
|
|
|
164 |
|
|
;; Integer divide latency is between 13 and 42 cycles for DIV[U] and between
|
165 |
|
|
;; 13 and 72 cycles for DDIV[U]. This depends on the value of the inputs
|
166 |
|
|
;; so we just choose the worst case latency.
|
167 |
|
|
(define_insn_reservation "r20kc_idiv_si" 42
|
168 |
|
|
(and (eq_attr "cpu" "20kc")
|
169 |
|
|
(and (eq_attr "type" "idiv")
|
170 |
|
|
(eq_attr "mode" "SI")))
|
171 |
|
|
"r20kc_impydiv+(r20kc_impydiv_iter*42)")
|
172 |
|
|
|
173 |
|
|
(define_insn_reservation "r20kc_idiv_di" 72
|
174 |
|
|
(and (eq_attr "cpu" "20kc")
|
175 |
|
|
(and (eq_attr "type" "idiv")
|
176 |
|
|
(eq_attr "mode" "DI")))
|
177 |
|
|
"r20kc_impydiv+(r20kc_impydiv_iter*72)")
|
178 |
|
|
|
179 |
|
|
;; Integer multiply latency is 4 or 7 cycles for word and double-word
|
180 |
|
|
;; respectively.
|
181 |
|
|
(define_insn_reservation "r20kc_impy_si" 4
|
182 |
|
|
(and (eq_attr "cpu" "20kc")
|
183 |
|
|
(and (eq_attr "type" "imadd,imul,imul3")
|
184 |
|
|
(eq_attr "mode" "SI")))
|
185 |
|
|
"r20kc_impydiv+(r20kc_impydiv_iter*2)")
|
186 |
|
|
|
187 |
|
|
(define_insn_reservation "r20kc_impy_di" 7
|
188 |
|
|
(and (eq_attr "cpu" "20kc")
|
189 |
|
|
(and (eq_attr "type" "imadd,imul,imul3")
|
190 |
|
|
(eq_attr "mode" "DI")))
|
191 |
|
|
"r20kc_impydiv+(r20kc_impydiv_iter*7)")
|
192 |
|
|
|
193 |
|
|
;; Move to/from HI/LO.
|
194 |
|
|
;; Moving to HI/LO has a 3 cycle latency while moving from only has a 1
|
195 |
|
|
;; cycle latency. Repeat rate is 3 for both.
|
196 |
|
|
(define_insn_reservation "r20kc_imthilo" 3
|
197 |
|
|
(and (eq_attr "cpu" "20kc")
|
198 |
|
|
(eq_attr "type" "mthilo"))
|
199 |
|
|
"r20kc_impydiv+(r20kc_impydiv_iter*3)")
|
200 |
|
|
|
201 |
|
|
(define_insn_reservation "r20kc_imfhilo" 1
|
202 |
|
|
(and (eq_attr "cpu" "20kc")
|
203 |
|
|
(eq_attr "type" "mfhilo"))
|
204 |
|
|
"r20kc_impydiv+(r20kc_impydiv_iter*3)")
|
205 |
|
|
|
206 |
|
|
;; Move to fp coprocessor.
|
207 |
|
|
(define_insn_reservation "r20kc_ixfer_mt" 3
|
208 |
|
|
(and (eq_attr "cpu" "20kc")
|
209 |
|
|
(eq_attr "type" "mtc"))
|
210 |
|
|
"r20kc_fpmove")
|
211 |
|
|
|
212 |
|
|
;; Move from fp coprocessor.
|
213 |
|
|
(define_insn_reservation "r20kc_ixfer_mf" 2
|
214 |
|
|
(and (eq_attr "cpu" "20kc")
|
215 |
|
|
(eq_attr "type" "mfc"))
|
216 |
|
|
"r20kc_fpmove")
|
217 |
|
|
|
218 |
|
|
;; Assume branch predicted correctly.
|
219 |
|
|
(define_insn_reservation "r20kc_ibr" 1
|
220 |
|
|
(and (eq_attr "cpu" "20kc")
|
221 |
|
|
(eq_attr "type" "branch,jump,call"))
|
222 |
|
|
"r20kc_ibranch")
|
223 |
|
|
|
224 |
|
|
;;
|
225 |
|
|
;; Describe instruction reservations for the floating-point operations.
|
226 |
|
|
;;
|
227 |
|
|
(define_insn_reservation "r20kc_fp_other" 4
|
228 |
|
|
(and (eq_attr "cpu" "20kc")
|
229 |
|
|
(eq_attr "type" "fmove,fadd,fabs,fneg,fcmp"))
|
230 |
|
|
"r20kc_fpadd")
|
231 |
|
|
|
232 |
|
|
(define_insn_reservation "r20kc_fp_cvt_a" 4
|
233 |
|
|
(and (eq_attr "cpu" "20kc")
|
234 |
|
|
(and (eq_attr "type" "fcvt")
|
235 |
|
|
(eq_attr "cnv_mode" "I2S,I2D,S2D")))
|
236 |
|
|
"r20kc_fpadd")
|
237 |
|
|
|
238 |
|
|
(define_insn_reservation "r20kc_fp_cvt_b" 5
|
239 |
|
|
(and (eq_attr "cpu" "20kc")
|
240 |
|
|
(and (eq_attr "type" "fcvt")
|
241 |
|
|
(eq_attr "cnv_mode" "D2S,S2I")))
|
242 |
|
|
"r20kc_fpadd")
|
243 |
|
|
|
244 |
|
|
(define_insn_reservation "r20kc_fp_divsqrt_df" 32
|
245 |
|
|
(and (eq_attr "cpu" "20kc")
|
246 |
|
|
(and (eq_attr "type" "fdiv,fsqrt")
|
247 |
|
|
(eq_attr "mode" "DF")))
|
248 |
|
|
"r20kc_fpdivsqrt+(r20kc_fpdivsqrt_iter*32)")
|
249 |
|
|
|
250 |
|
|
(define_insn_reservation "r20kc_fp_divsqrt_sf" 17
|
251 |
|
|
(and (eq_attr "cpu" "20kc")
|
252 |
|
|
(and (eq_attr "type" "fdiv,fsqrt")
|
253 |
|
|
(eq_attr "mode" "SF")))
|
254 |
|
|
"r20kc_fpdivsqrt+(r20kc_fpdivsqrt_iter*17)")
|
255 |
|
|
|
256 |
|
|
(define_insn_reservation "r20kc_fp_rsqrt_df" 35
|
257 |
|
|
(and (eq_attr "cpu" "20kc")
|
258 |
|
|
(and (eq_attr "type" "frsqrt")
|
259 |
|
|
(eq_attr "mode" "DF")))
|
260 |
|
|
"r20kc_fpdivsqrt+(r20kc_fpdivsqrt_iter*35)")
|
261 |
|
|
|
262 |
|
|
(define_insn_reservation "r20kc_fp_rsqrt_sf" 17
|
263 |
|
|
(and (eq_attr "cpu" "20kc")
|
264 |
|
|
(and (eq_attr "type" "frsqrt")
|
265 |
|
|
(eq_attr "mode" "SF")))
|
266 |
|
|
"r20kc_fpdivsqrt+(r20kc_fpdivsqrt_iter*17)")
|
267 |
|
|
|
268 |
|
|
(define_insn_reservation "r20kc_fp_mpy_sf" 4
|
269 |
|
|
(and (eq_attr "cpu" "20kc")
|
270 |
|
|
(and (eq_attr "type" "fmul,fmadd")
|
271 |
|
|
(eq_attr "mode" "SF")))
|
272 |
|
|
"r20kc_fpmpy+r20kc_fpmpy_iter")
|
273 |
|
|
|
274 |
|
|
(define_insn_reservation "r20kc_fp_mpy_df" 5
|
275 |
|
|
(and (eq_attr "cpu" "20kc")
|
276 |
|
|
(and (eq_attr "type" "fmul,fmadd")
|
277 |
|
|
(eq_attr "mode" "DF")))
|
278 |
|
|
"r20kc_fpmpy+(r20kc_fpmpy_iter*2)")
|
279 |
|
|
|
280 |
|
|
;; Force single-dispatch for unknown or multi.
|
281 |
|
|
(define_insn_reservation "r20kc_unknown" 1
|
282 |
|
|
(and (eq_attr "cpu" "20kc")
|
283 |
|
|
(eq_attr "type" "unknown,multi"))
|
284 |
|
|
"r20kc_single_dispatch")
|