0
|
1 ;; DFA-based pipeline descriptions for MIPS Technologies 24K core.
|
|
2 ;; Contributed by Chao-ying Fu (fu@mips.com), Nigel Stephens (nigel@mips.com)
|
|
3 ;; and David Ung (davidu@mips.com)
|
|
4 ;;
|
|
5 ;; The 24kf2_1 is a single-issue processor with a half-clocked fpu.
|
|
6 ;; The 24kf1_1 is 24k with 1:1 clocked fpu.
|
|
7 ;;
|
|
8 ;; References:
|
|
9 ;; "MIPS32 24K Processor Core Family Software User's Manual, Rev 3.04."
|
|
10 ;;
|
131
|
11 ;; Copyright (C) 2005-2018 Free Software Foundation, Inc.
|
0
|
12 ;;
|
|
13 ;; This file is part of GCC.
|
|
14 ;;
|
|
15 ;; GCC is free software; you can redistribute it and/or modify it
|
|
16 ;; under the terms of the GNU General Public License as published
|
|
17 ;; by the Free Software Foundation; either version 3, or (at your
|
|
18 ;; option) any later version.
|
|
19
|
|
20 ;; GCC is distributed in the hope that it will be useful, but WITHOUT
|
|
21 ;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
22 ;; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
|
23 ;; License for more details.
|
|
24
|
|
25 ;; You should have received a copy of the GNU General Public License
|
|
26 ;; along with GCC; see the file COPYING3. If not see
|
|
27 ;; <http://www.gnu.org/licenses/>.
|
|
28
|
|
29 (define_automaton "r24k_cpu, r24k_mdu, r24k_fpu")
|
|
30
|
|
31 ;; Integer execution unit.
|
|
32 (define_cpu_unit "r24k_iss" "r24k_cpu")
|
|
33 (define_cpu_unit "r24k_ixu_arith" "r24k_cpu")
|
|
34 (define_cpu_unit "r24k_mul3a" "r24k_mdu")
|
|
35 (define_cpu_unit "r24k_mul3b" "r24k_mdu")
|
|
36 (define_cpu_unit "r24k_mul3c" "r24k_mdu")
|
|
37
|
|
38 ;; --------------------------------------------------------------
|
|
39 ;; Producers
|
|
40 ;; --------------------------------------------------------------
|
|
41
|
|
42 ;; 1. Loads: lb, lbu, lh, lhu, ll, lw, lwl, lwr, lwpc, lwxs
|
|
43 (define_insn_reservation "r24k_int_load" 2
|
|
44 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
45 (eq_attr "type" "load"))
|
|
46 "r24k_iss+r24k_ixu_arith")
|
|
47
|
|
48
|
|
49 ;; 2. Arithmetic: add, addi, addiu, addiupc, addu, and, andi, clo, clz,
|
|
50 ;; ext, ins, lui, movn, movz, nor, or, ori, rotr, rotrv, seb, seh, sll,
|
|
51 ;; sllv, slt, slti, sltiu, sltu, sra, srav, srl, srlv, sub, subu, wsbh,
|
|
52 ;; xor, xori
|
|
53 ;; (movn/movz is not matched, we'll need to split condmov to
|
|
54 ;; differentiate between integer/float moves)
|
|
55 (define_insn_reservation "r24k_int_arith" 1
|
|
56 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
57 (eq_attr "type" "arith,const,logical,move,nop,shift,signext,slt"))
|
|
58 "r24k_iss+r24k_ixu_arith")
|
|
59
|
|
60
|
|
61 ;; 3. Links: bgezal, bgezall, bltzal, bltzall, jal, jalr, jalx
|
|
62 ;; 3a. jr/jalr consumer
|
|
63 (define_insn_reservation "r24k_int_jump" 1
|
|
64 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
65 (eq_attr "type" "call,jump"))
|
|
66 "r24k_iss+r24k_ixu_arith")
|
|
67
|
|
68 ;; 3b. branch consumer
|
|
69 (define_insn_reservation "r24k_int_branch" 1
|
|
70 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
71 (eq_attr "type" "branch"))
|
|
72 "r24k_iss+r24k_ixu_arith")
|
|
73
|
|
74
|
|
75 ;; 4. MDU: fully pipelined multiplier
|
|
76 ;; mult - delivers result to hi/lo in 1 cycle (pipelined)
|
|
77 (define_insn_reservation "r24k_int_mult" 1
|
|
78 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
79 (eq_attr "type" "imul"))
|
|
80 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
81
|
|
82 ;; madd, msub - delivers result to hi/lo in 1 cycle (pipelined)
|
|
83 (define_insn_reservation "r24k_int_madd" 1
|
|
84 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
85 (eq_attr "type" "imadd"))
|
|
86 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
87
|
|
88 ;; mul - delivers result to gpr in 5 cycles
|
|
89 (define_insn_reservation "r24k_int_mul3" 5
|
|
90 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
91 (eq_attr "type" "imul3"))
|
|
92 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)*5")
|
|
93
|
|
94 ;; mfhi, mflo, mflhxu - deliver result to gpr in 5 cycles
|
|
95 (define_insn_reservation "r24k_int_mfhilo" 5
|
|
96 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
111
|
97 (eq_attr "type" "mfhi,mflo"))
|
0
|
98 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
99
|
|
100 ;; mthi, mtlo, mtlhx - deliver result to hi/lo, thence madd, handled as bypass
|
|
101 (define_insn_reservation "r24k_int_mthilo" 1
|
|
102 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
111
|
103 (eq_attr "type" "mthi,mtlo"))
|
0
|
104 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
105
|
|
106 ;; div - default to 36 cycles for 32bit operands. Faster for 24bit, 16bit and
|
|
107 ;; 8bit, but is tricky to identify.
|
|
108 (define_insn_reservation "r24k_int_div" 36
|
|
109 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
110 (eq_attr "type" "idiv"))
|
|
111 "r24k_iss+(r24k_mul3a+r24k_mul3b+r24k_mul3c)*36")
|
|
112
|
|
113
|
|
114 ;; 5. Cop: cfc1, di, ei, mfc0, mtc0
|
|
115 ;; (Disabled until we add proper cop0 support)
|
|
116 ;;(define_insn_reservation "r24k_int_cop" 3
|
|
117 ;; (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
118 ;; (eq_attr "type" "cop0"))
|
|
119 ;; "r24k_iss+r24k_ixu_arith")
|
|
120
|
|
121
|
|
122 ;; 6. Store
|
|
123 (define_insn_reservation "r24k_int_store" 1
|
|
124 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
111
|
125 (eq_attr "type" "store"))
|
0
|
126 "r24k_iss+r24k_ixu_arith")
|
|
127
|
|
128
|
|
129 ;; 7. Multiple instructions
|
|
130 (define_insn_reservation "r24k_int_multi" 1
|
|
131 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
132 (eq_attr "type" "multi"))
|
|
133 "r24k_iss+r24k_ixu_arith+r24k_fpu_arith+(r24k_mul3a+r24k_mul3b+r24k_mul3c)")
|
|
134
|
|
135
|
|
136 ;; 8. Unknowns - Currently these include blockage, consttable and alignment
|
|
137 ;; rtls. They do not really affect scheduling latency, (blockage affects
|
|
138 ;; scheduling via log links, but not used here).
|
|
139 (define_insn_reservation "r24k_int_unknown" 0
|
|
140 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
111
|
141 (eq_attr "type" "unknown,atomic,syncloop"))
|
0
|
142 "r24k_iss")
|
|
143
|
|
144
|
|
145 ;; 9. Prefetch
|
|
146 (define_insn_reservation "r24k_int_prefetch" 1
|
|
147 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
148 (eq_attr "type" "prefetch,prefetchx"))
|
|
149 "r24k_iss+r24k_ixu_arith")
|
|
150
|
|
151
|
|
152 ;; --------------------------------------------------------------
|
|
153 ;; Bypass to Consumer
|
|
154 ;; --------------------------------------------------------------
|
|
155
|
|
156 ;; load->next use : 2 cycles (Default)
|
|
157 ;; load->load base: 3 cycles
|
|
158 ;; load->store base: 3 cycles
|
|
159 ;; load->prefetch: 3 cycles
|
|
160 (define_bypass 3 "r24k_int_load" "r24k_int_load")
|
111
|
161 (define_bypass 3 "r24k_int_load" "r24k_int_store" "!mips_store_data_bypass_p")
|
0
|
162 (define_bypass 3 "r24k_int_load" "r24k_int_prefetch")
|
|
163
|
|
164 ;; arith->next use : 1 cycles (Default)
|
|
165 ;; arith->load base: 2 cycles
|
|
166 ;; arith->store base: 2 cycles
|
|
167 ;; arith->prefetch: 2 cycles
|
|
168 (define_bypass 2 "r24k_int_arith" "r24k_int_load")
|
111
|
169 (define_bypass 2 "r24k_int_arith" "r24k_int_store" "!mips_store_data_bypass_p")
|
0
|
170 (define_bypass 2 "r24k_int_arith" "r24k_int_prefetch")
|
|
171
|
|
172 ;; mul3->next use : 5 cycles (default)
|
|
173 ;; mul3->l/s base : 6 cycles
|
|
174 ;; mul3->prefetch : 6 cycles
|
|
175 (define_bypass 6 "r24k_int_mul3" "r24k_int_load")
|
111
|
176 (define_bypass 6 "r24k_int_mul3" "r24k_int_store" "!mips_store_data_bypass_p")
|
0
|
177 (define_bypass 6 "r24k_int_mul3" "r24k_int_prefetch")
|
|
178
|
|
179 ;; mul3->madd/msub : 1 cycle
|
|
180 (define_bypass 1 "r24k_int_mul3" "r24k_int_madd" "mips_linked_madd_p")
|
|
181
|
|
182 ;; mfhilo->next use : 5 cycles (default)
|
|
183 ;; mfhilo->l/s base : 6 cycles
|
|
184 ;; mfhilo->prefetch : 6 cycles
|
|
185 ;; mthilo->madd/msub : 2 cycle (only for mthi/lo not mfhi/lo)
|
|
186 (define_bypass 6 "r24k_int_mfhilo" "r24k_int_load")
|
111
|
187 (define_bypass 6 "r24k_int_mfhilo" "r24k_int_store"
|
|
188 "!mips_store_data_bypass_p")
|
0
|
189 (define_bypass 6 "r24k_int_mfhilo" "r24k_int_prefetch")
|
|
190 (define_bypass 2 "r24k_int_mthilo" "r24k_int_madd")
|
|
191
|
|
192 ;; cop->next use : 3 cycles (Default)
|
|
193 ;; cop->l/s base : 4 cycles
|
|
194 ;; (define_bypass 4 "r24k_int_cop" "r24k_int_load")
|
111
|
195 ;; (define_bypass 4 "r24k_int_cop" "r24k_int_store"
|
|
196 ;; "!mips_store_data_bypass_p")
|
0
|
197
|
|
198 ;; multi->next use : 1 cycles (Default)
|
|
199 ;; multi->l/s base : 2 cycles
|
|
200 ;; multi->prefetch : 2 cycles
|
|
201 (define_bypass 2 "r24k_int_multi" "r24k_int_load")
|
111
|
202 (define_bypass 2 "r24k_int_multi" "r24k_int_store" "!mips_store_data_bypass_p")
|
0
|
203 (define_bypass 2 "r24k_int_multi" "r24k_int_prefetch")
|
|
204
|
|
205
|
|
206 ;; --------------------------------------------------------------
|
111
|
207 ;; DSP instructions
|
|
208 ;; --------------------------------------------------------------
|
|
209
|
|
210 ;; absq, addq, addsc, addu, addwc, bitrev, cmp, cmpgu, cmpu, insv, modsub,
|
|
211 ;; packrl, pick, preceq, preceu, precequ, precrq, precrqu, raddu, rddsp, repl,
|
|
212 ;; replv, shll, shllv, shra, shrav, shrl, shrlv, subq, subu, wrdsp
|
|
213 (define_insn_reservation "r24k_dsp_alu" 2
|
|
214 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
215 (eq_attr "type" "dspalu,dspalusat"))
|
|
216 "r24k_iss+r24k_ixu_arith")
|
|
217
|
|
218 ;; dpaq_s, dpau, dpsq_s, dpsu, maq_s, mulsaq
|
|
219 (define_insn_reservation "r24k_dsp_mac" 1
|
|
220 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
221 (eq_attr "type" "dspmac"))
|
|
222 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
223
|
|
224 ;; dpaq_sa, dpsq_sa, maq_sa
|
|
225 (define_insn_reservation "r24k_dsp_mac_sat" 1
|
|
226 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
227 (eq_attr "type" "dspmacsat"))
|
|
228 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
229
|
|
230 ;; extp, extpdp, extpdpv, extpv, extr, extrv
|
|
231 (define_insn_reservation "r24k_dsp_acc_ext" 5
|
|
232 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
233 (eq_attr "type" "accext"))
|
|
234 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
235
|
|
236 ;; mthlip, shilo, shilov
|
|
237 (define_insn_reservation "r24k_dsp_acc_mod" 1
|
|
238 (and (eq_attr "cpu" "24kc,24kf2_1,24kf1_1")
|
|
239 (eq_attr "type" "accmod"))
|
|
240 "r24k_iss+(r24k_mul3a|r24k_mul3b|r24k_mul3c)")
|
|
241
|
|
242
|
|
243 ;; mult/madd->dsp_acc_ext : 4 cycles
|
|
244 ;; mult/madd->dsp_acc_mod : 4 cycles
|
|
245 (define_bypass 4 "r24k_int_mult" "r24k_dsp_acc_ext")
|
|
246 (define_bypass 4 "r24k_int_mult" "r24k_dsp_acc_mod")
|
|
247
|
|
248 ;; mthilo->dsp_acc_ext : 4 cycles
|
|
249 ;; mthilo->dsp_acc_ext : 4 cycles
|
|
250 (define_bypass 4 "r24k_int_mthilo" "r24k_dsp_acc_ext")
|
|
251 (define_bypass 4 "r24k_int_mthilo" "r24k_dsp_acc_mod")
|
|
252
|
|
253 ;; dsp_mac->next use : 1 cycles (default)
|
|
254 ;; dsp_mac->dsp_acc_ext : 4 cycles
|
|
255 ;; dsp_mac->dsp_acc_mod : 4 cycles
|
|
256 (define_bypass 4 "r24k_dsp_mac" "r24k_dsp_acc_ext")
|
|
257 (define_bypass 4 "r24k_dsp_mac" "r24k_dsp_acc_mod")
|
|
258
|
|
259 ;; dsp_mac_sat->next use : 1 cycles (default)
|
|
260 ;; dsp_mac_sat->mult/madd : 2 cycles
|
|
261 ;; dsp_mac_sat->dsp_mac : 2 cycles
|
|
262 ;; dsp_mac_sat->dsp_mac_sat : 2 cycles
|
|
263 ;; dsp_mac_sat->dsp_acc_ext : 4 cycles
|
|
264 ;; dsp_mac_sat->dsp_acc_mod : 4 cycles
|
|
265 (define_bypass 2 "r24k_dsp_mac_sat" "r24k_int_mult")
|
|
266 (define_bypass 2 "r24k_dsp_mac_sat" "r24k_dsp_mac")
|
|
267 (define_bypass 2 "r24k_dsp_mac_sat" "r24k_dsp_mac_sat")
|
|
268 (define_bypass 4 "r24k_dsp_mac_sat" "r24k_dsp_acc_ext")
|
|
269 (define_bypass 4 "r24k_dsp_mac_sat" "r24k_dsp_acc_mod")
|
|
270
|
|
271 ;; dsp_acc_ext->next use : 5 cycles (default)
|
|
272 ;; dsp_acc_ext->l/s base : 6 cycles
|
|
273 ;; dsp_acc_ext->prefetch : 6 cycles
|
|
274 (define_bypass 6 "r24k_dsp_acc_ext" "r24k_int_load")
|
|
275 (define_bypass 6 "r24k_dsp_acc_ext" "r24k_int_store"
|
|
276 "!mips_store_data_bypass_p")
|
|
277 (define_bypass 6 "r24k_dsp_acc_ext" "r24k_int_prefetch")
|
|
278
|
|
279 ;; dsp_acc_mod->next use : 1 cycles (default)
|
|
280 ;; dsp_acc_mod->mult/madd : 2 cycles
|
|
281 ;; dsp_acc_mod->dsp_mac : 2 cycles
|
|
282 ;; dsp_acc_mod->dsp_mac_sat : 2 cycles
|
|
283 ;; dsp_acc_mod->dsp_acc_ext : 4 cycles
|
|
284 ;; dsp_acc_mod->dsp_acc_mod : 4 cycles
|
|
285 (define_bypass 2 "r24k_dsp_acc_mod" "r24k_int_mult")
|
|
286 (define_bypass 2 "r24k_dsp_acc_mod" "r24k_dsp_mac")
|
|
287 (define_bypass 2 "r24k_dsp_acc_mod" "r24k_dsp_mac_sat")
|
|
288 (define_bypass 4 "r24k_dsp_acc_mod" "r24k_dsp_acc_ext")
|
|
289 (define_bypass 4 "r24k_dsp_acc_mod" "r24k_dsp_acc_mod")
|
|
290
|
|
291 ;; dspalu->next use : 2 cycles (default)
|
|
292 ;; dspalu->l/s base : 3 cycles
|
|
293 ;; dspalu->prefetch : 3 cycles
|
|
294 ;; some pairs of dspalu (addsc/addwc, cmp/pick, wrdsp/insv) : 1 cycle
|
|
295 (define_bypass 3 "r24k_dsp_alu" "r24k_int_load")
|
|
296 (define_bypass 3 "r24k_dsp_alu" "r24k_int_store" "!mips_store_data_bypass_p")
|
|
297 (define_bypass 3 "r24k_dsp_alu" "r24k_int_prefetch")
|
|
298 (define_bypass 1 "r24k_dsp_alu" "r24k_dsp_alu" "mips_dspalu_bypass_p")
|
|
299
|
|
300
|
|
301 ;; --------------------------------------------------------------
|
0
|
302 ;; Floating Point Instructions
|
|
303 ;; --------------------------------------------------------------
|
|
304
|
|
305 (define_cpu_unit "r24k_fpu_arith" "r24k_fpu")
|
|
306
|
|
307 ;; The 24k is a single issue cpu, and the fpu runs at half clock speed,
|
|
308 ;; so each fpu instruction ties up the shared instruction scheduler for
|
|
309 ;; 1 cycle, and the fpu scheduler for 2 cycles.
|
|
310 ;;
|
|
311 ;; These timings are therefore twice the values in the 24K manual,
|
|
312 ;; which are quoted in fpu clocks.
|
|
313 ;;
|
|
314 ;; The 24kf1_1 is a 24k configured with 1:1 cpu and fpu, so use
|
|
315 ;; the unscaled timings
|
|
316
|
|
317 (define_reservation "r24kf2_1_fpu_iss" "r24k_iss+(r24k_fpu_arith*2)")
|
|
318
|
|
319 ;; fadd, fabs, fneg
|
|
320 (define_insn_reservation "r24kf2_1_fadd" 8
|
|
321 (and (eq_attr "cpu" "24kf2_1")
|
|
322 (eq_attr "type" "fadd,fabs,fneg"))
|
|
323 "r24kf2_1_fpu_iss")
|
|
324
|
|
325 ;; fmove, fcmove
|
|
326 (define_insn_reservation "r24kf2_1_fmove" 8
|
|
327 (and (eq_attr "cpu" "24kf2_1")
|
|
328 (eq_attr "type" "fmove,condmove"))
|
|
329 "r24kf2_1_fpu_iss")
|
|
330
|
|
331 ;; fload
|
|
332 (define_insn_reservation "r24kf2_1_fload" 6
|
|
333 (and (eq_attr "cpu" "24kf2_1")
|
|
334 (eq_attr "type" "fpload,fpidxload"))
|
|
335 "r24kf2_1_fpu_iss")
|
|
336
|
|
337 ;; fstore
|
|
338 (define_insn_reservation "r24kf2_1_fstore" 2
|
|
339 (and (eq_attr "cpu" "24kf2_1")
|
|
340 (eq_attr "type" "fpstore"))
|
|
341 "r24kf2_1_fpu_iss")
|
|
342
|
|
343 ;; fmul, fmadd
|
|
344 (define_insn_reservation "r24kf2_1_fmul_sf" 8
|
|
345 (and (eq_attr "cpu" "24kf2_1")
|
|
346 (and (eq_attr "type" "fmul,fmadd")
|
|
347 (eq_attr "mode" "SF")))
|
|
348 "r24kf2_1_fpu_iss")
|
|
349
|
|
350 (define_insn_reservation "r24kf2_1_fmul_df" 10
|
|
351 (and (eq_attr "cpu" "24kf2_1")
|
|
352 (and (eq_attr "type" "fmul,fmadd")
|
|
353 (eq_attr "mode" "DF")))
|
|
354 "r24kf2_1_fpu_iss,(r24k_fpu_arith*2)")
|
|
355
|
|
356
|
|
357 ;; fdiv, fsqrt, frsqrt
|
|
358 (define_insn_reservation "r24kf2_1_fdiv_sf" 34
|
|
359 (and (eq_attr "cpu" "24kf2_1")
|
|
360 (and (eq_attr "type" "fdiv,fsqrt,frsqrt")
|
|
361 (eq_attr "mode" "SF")))
|
|
362 "r24kf2_1_fpu_iss,(r24k_fpu_arith*26)")
|
|
363
|
|
364 (define_insn_reservation "r24kf2_1_fdiv_df" 64
|
|
365 (and (eq_attr "cpu" "24kf2_1")
|
|
366 (and (eq_attr "type" "fdiv,fsqrt")
|
|
367 (eq_attr "mode" "DF")))
|
|
368 "r24kf2_1_fpu_iss,(r24k_fpu_arith*56)")
|
|
369
|
|
370 ;; frsqrt
|
|
371 (define_insn_reservation "r24kf2_1_frsqrt_df" 70
|
|
372 (and (eq_attr "cpu" "24kf2_1")
|
|
373 (and (eq_attr "type" "frsqrt")
|
|
374 (eq_attr "mode" "DF")))
|
|
375 "r24kf2_1_fpu_iss,(r24k_fpu_arith*60)")
|
|
376
|
|
377 ;; fcmp
|
|
378 (define_insn_reservation "r24kf2_1_fcmp" 4
|
|
379 (and (eq_attr "cpu" "24kf2_1")
|
|
380 (eq_attr "type" "fcmp"))
|
|
381 "r24kf2_1_fpu_iss")
|
|
382
|
|
383 ;; fcmp -> movf.fmt & movt.fmt bypass (dependency must be on the condition)
|
|
384 (define_bypass 2 "r24kf2_1_fcmp" "r24kf2_1_fmove")
|
|
385
|
|
386 ;; fcvt (cvt.d.s, cvt.[sd].[wl])
|
|
387 (define_insn_reservation "r24kf2_1_fcvt_i2f_s2d" 8
|
|
388 (and (eq_attr "cpu" "24kf2_1")
|
|
389 (and (eq_attr "type" "fcvt")
|
|
390 (eq_attr "cnv_mode" "I2S,I2D,S2D")))
|
|
391 "r24kf2_1_fpu_iss")
|
|
392
|
|
393 ;; fcvt (cvt.s.d)
|
|
394 (define_insn_reservation "r24kf2_1_fcvt_s2d" 12
|
|
395 (and (eq_attr "cpu" "24kf2_1")
|
|
396 (and (eq_attr "type" "fcvt")
|
|
397 (eq_attr "cnv_mode" "D2S")))
|
|
398 "r24kf2_1_fpu_iss")
|
|
399
|
|
400 ;; fcvt (cvt.[wl].[sd], etc)
|
|
401 (define_insn_reservation "r24kf2_1_fcvt_f2i" 10
|
|
402 (and (eq_attr "cpu" "24kf2_1")
|
|
403 (and (eq_attr "type" "fcvt")
|
|
404 (eq_attr "cnv_mode" "S2I,D2I")))
|
|
405 "r24kf2_1_fpu_iss")
|
|
406
|
|
407 ;; fxfer (mfc1, mfhc1, mtc1, mthc1)
|
|
408 (define_insn_reservation "r24kf2_1_fxfer" 4
|
|
409 (and (eq_attr "cpu" "24kf2_1")
|
|
410 (eq_attr "type" "mfc,mtc"))
|
|
411 "r24kf2_1_fpu_iss")
|
|
412
|
|
413 ;; --------------------------------------------------------------
|
|
414 ;; Bypass to Consumer
|
|
415 ;; --------------------------------------------------------------
|
|
416 ;; r24kf2_1_fcvt_f2i->l/s base : 11 cycles
|
|
417 ;; r24kf2_1_fcvt_f2i->prefetch : 11 cycles
|
|
418 (define_bypass 11 "r24kf2_1_fcvt_f2i" "r24k_int_load")
|
111
|
419 (define_bypass 11 "r24kf2_1_fcvt_f2i" "r24k_int_store"
|
|
420 "!mips_store_data_bypass_p")
|
0
|
421 (define_bypass 11 "r24kf2_1_fcvt_f2i" "r24k_int_prefetch")
|
|
422
|
|
423 ;; r24kf2_1_fxfer->l/s base : 5 cycles
|
|
424 ;; r24kf2_1_fxfer->prefetch : 5 cycles
|
|
425 (define_bypass 5 "r24kf2_1_fxfer" "r24k_int_load")
|
111
|
426 (define_bypass 5 "r24kf2_1_fxfer" "r24k_int_store" "!mips_store_data_bypass_p")
|
0
|
427 (define_bypass 5 "r24kf2_1_fxfer" "r24k_int_prefetch")
|
|
428
|
|
429 ;; --------------------------------------------------------------
|
|
430 ;; The 24kf1_1 is a 24k configured with 1:1 cpu and fpu, so use
|
|
431 ;; the unscaled timings
|
|
432 ;; --------------------------------------------------------------
|
|
433
|
|
434 (define_reservation "r24kf1_1_fpu_iss" "r24k_iss+r24k_fpu_arith")
|
|
435
|
|
436 ;; fadd, fabs, fneg
|
|
437 (define_insn_reservation "r24kf1_1_fadd" 4
|
|
438 (and (eq_attr "cpu" "24kf1_1")
|
|
439 (eq_attr "type" "fadd,fabs,fneg"))
|
|
440 "r24kf1_1_fpu_iss")
|
|
441
|
|
442 ;; fmove, fcmove
|
|
443 (define_insn_reservation "r24kf1_1_fmove" 4
|
|
444 (and (eq_attr "cpu" "24kf1_1")
|
|
445 (eq_attr "type" "fmove,condmove"))
|
|
446 "r24kf1_1_fpu_iss")
|
|
447
|
|
448 ;; fload
|
|
449 (define_insn_reservation "r24kf1_1_fload" 3
|
|
450 (and (eq_attr "cpu" "24kf1_1")
|
|
451 (eq_attr "type" "fpload,fpidxload"))
|
|
452 "r24kf1_1_fpu_iss")
|
|
453
|
|
454 ;; fstore
|
|
455 (define_insn_reservation "r24kf1_1_fstore" 1
|
|
456 (and (eq_attr "cpu" "24kf1_1")
|
|
457 (eq_attr "type" "fpstore"))
|
|
458 "r24kf1_1_fpu_iss")
|
|
459
|
|
460 ;; fmul, fmadd
|
|
461 (define_insn_reservation "r24kf1_1_fmul_sf" 4
|
|
462 (and (eq_attr "cpu" "24kf1_1")
|
|
463 (and (eq_attr "type" "fmul,fmadd")
|
|
464 (eq_attr "mode" "SF")))
|
|
465 "r24kf1_1_fpu_iss")
|
|
466
|
|
467 (define_insn_reservation "r24kf1_1_fmul_df" 5
|
|
468 (and (eq_attr "cpu" "24kf1_1")
|
|
469 (and (eq_attr "type" "fmul,fmadd")
|
|
470 (eq_attr "mode" "DF")))
|
|
471 "r24kf1_1_fpu_iss,r24k_fpu_arith")
|
|
472
|
|
473
|
|
474 ;; fdiv, fsqrt, frsqrt
|
|
475 (define_insn_reservation "r24kf1_1_fdiv_sf" 17
|
|
476 (and (eq_attr "cpu" "24kf1_1")
|
|
477 (and (eq_attr "type" "fdiv,fsqrt,frsqrt")
|
|
478 (eq_attr "mode" "SF")))
|
|
479 "r24kf1_1_fpu_iss,(r24k_fpu_arith*13)")
|
|
480
|
|
481 (define_insn_reservation "r24kf1_1_fdiv_df" 32
|
|
482 (and (eq_attr "cpu" "24kf1_1")
|
|
483 (and (eq_attr "type" "fdiv,fsqrt")
|
|
484 (eq_attr "mode" "DF")))
|
|
485 "r24kf1_1_fpu_iss,(r24k_fpu_arith*28)")
|
|
486
|
|
487 ;; frsqrt
|
|
488 (define_insn_reservation "r24kf1_1_frsqrt_df" 35
|
|
489 (and (eq_attr "cpu" "24kf1_1")
|
|
490 (and (eq_attr "type" "frsqrt")
|
|
491 (eq_attr "mode" "DF")))
|
|
492 "r24kf1_1_fpu_iss,(r24k_fpu_arith*30)")
|
|
493
|
|
494 ;; fcmp
|
|
495 (define_insn_reservation "r24kf1_1_fcmp" 2
|
|
496 (and (eq_attr "cpu" "24kf1_1")
|
|
497 (eq_attr "type" "fcmp"))
|
|
498 "r24kf1_1_fpu_iss")
|
|
499
|
|
500 ;; fcmp -> movf.fmt & movt.fmt bypass (dependency must be on the condition)
|
|
501 (define_bypass 1 "r24kf1_1_fcmp" "r24kf1_1_fmove")
|
|
502
|
|
503 ;; fcvt (cvt.d.s, cvt.[sd].[wl])
|
|
504 (define_insn_reservation "r24kf1_1_fcvt_i2f_s2d" 4
|
|
505 (and (eq_attr "cpu" "24kf1_1")
|
|
506 (and (eq_attr "type" "fcvt")
|
|
507 (eq_attr "cnv_mode" "I2S,I2D,S2D")))
|
|
508 "r24kf1_1_fpu_iss")
|
|
509
|
|
510 ;; fcvt (cvt.s.d)
|
|
511 (define_insn_reservation "r24kf1_1_fcvt_s2d" 6
|
|
512 (and (eq_attr "cpu" "24kf1_1")
|
|
513 (and (eq_attr "type" "fcvt")
|
|
514 (eq_attr "cnv_mode" "D2S")))
|
|
515 "r24kf1_1_fpu_iss")
|
|
516
|
|
517 ;; fcvt (cvt.[wl].[sd], etc)
|
|
518 (define_insn_reservation "r24kf1_1_fcvt_f2i" 5
|
|
519 (and (eq_attr "cpu" "24kf1_1")
|
|
520 (and (eq_attr "type" "fcvt")
|
|
521 (eq_attr "cnv_mode" "S2I,D2I")))
|
|
522 "r24kf1_1_fpu_iss")
|
|
523
|
|
524 ;; fxfer (mfc1, mfhc1, mtc1, mthc1)
|
|
525 (define_insn_reservation "r24kf1_1_fxfer" 2
|
|
526 (and (eq_attr "cpu" "24kf1_1")
|
|
527 (eq_attr "type" "mfc,mtc"))
|
|
528 "r24kf1_1_fpu_iss")
|
|
529
|
|
530 ;; --------------------------------------------------------------
|
|
531 ;; Bypass to Consumer
|
|
532 ;; --------------------------------------------------------------
|
|
533 ;; r24kf1_1_fcvt_f2i->l/s base : 6 cycles
|
|
534 ;; r24kf1_1_fcvt_f2i->prefetch : 6 cycles
|
|
535 (define_bypass 6 "r24kf1_1_fcvt_f2i" "r24k_int_load")
|
111
|
536 (define_bypass 6 "r24kf1_1_fcvt_f2i" "r24k_int_store"
|
|
537 "!mips_store_data_bypass_p")
|
0
|
538 (define_bypass 6 "r24kf1_1_fcvt_f2i" "r24k_int_prefetch")
|
|
539
|
|
540 ;; r24kf1_1_fxfer->l/s base : 3 cycles
|
|
541 ;; r24kf1_1_fxfer->prefetch : 3 cycles
|
|
542 (define_bypass 3 "r24kf1_1_fxfer" "r24k_int_load")
|
111
|
543 (define_bypass 3 "r24kf1_1_fxfer" "r24k_int_store" "!mips_store_data_bypass_p")
|
0
|
544 (define_bypass 3 "r24kf1_1_fxfer" "r24k_int_prefetch")
|
|
545
|