1;; DFA-based pipeline description for MIPS32 model 74k.
2;; Contributed by MIPS Technologies and CodeSourcery.
3;;
4;; Reference:
5;;   "MIPS32 74K Microarchitecure Specification Rev. 01.02 Jun 15, 2006"
6;;   "MIPS32 74Kf Processor Core Datasheet Jun 2, 2006"
7;;
8;; Copyright (C) 2007-2015 Free Software Foundation, Inc.
9;;
10;; This file is part of GCC.
11;;
12;; GCC is free software; you can redistribute it and/or modify it
13;; under the terms of the GNU General Public License as published
14;; by the Free Software Foundation; either version 3, or (at your
15;; option) any later version.
16
17;; GCC is distributed in the hope that it will be useful, but WITHOUT
18;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
19;; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
20;; License for more details.
21
22;; You should have received a copy of the GNU General Public License
23;; along with GCC; see the file COPYING3.  If not see
24;; <http://www.gnu.org/licenses/>.
25
26(define_automaton "r74k_mdu_pipe, r74k_alu_pipe, r74k_agen_pipe, r74k_fpu")
27(define_cpu_unit "r74k_mul" "r74k_mdu_pipe")
28(define_cpu_unit "r74k_alu" "r74k_alu_pipe")
29(define_cpu_unit "r74k_agen" "r74k_agen_pipe")
30(define_cpu_unit "r74k_fpu_arith" "r74k_fpu")
31(define_cpu_unit "r74k_fpu_ldst" "r74k_fpu")
32
33;; --------------------------------------------------------------
34;; Producers
35;; --------------------------------------------------------------
36
37;; ALU: Logicals/Arithmetics
38;; - Logicals, move (addu/addiu with rt = 0), Set less than, 
39;;   sign extend - 1 cycle
40(define_insn_reservation "r74k_int_logical" 1
41  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
42       (eq_attr "type" "logical,move,signext,slt"))
43  "r74k_alu")
44
45;; - Arithmetics - 2 cycles
46(define_insn_reservation "r74k_int_arith" 2
47  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
48       (eq_attr "type" "arith,const,shift,clz"))
49  "r74k_alu")
50
51(define_insn_reservation "r74k_int_nop" 0
52  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
53       (eq_attr "type" "nop"))
54  "nothing")
55
56(define_insn_reservation "r74k_int_cmove" 4
57  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
58       (eq_attr "type" "condmove"))
59  "r74k_agen*2")
60
61;; MDU: fully pipelined multiplier
62;; mult - delivers result to hi/lo in 4 cycle (pipelined)
63(define_insn_reservation "r74k_int_mult" 4
64  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
65       (eq_attr "type" "imul"))
66  "r74k_alu+r74k_mul")
67
68;; madd, msub - delivers result to hi/lo in 4 cycle (pipelined)
69(define_insn_reservation "r74k_int_madd" 4
70  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
71       (eq_attr "type" "imadd"))
72  "r74k_alu+r74k_mul")
73
74;; mul - delivers result to general register in 7 cycles
75(define_insn_reservation "r74k_int_mul3" 7
76  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
77       (eq_attr "type" "imul3"))
78  "r74k_alu+r74k_mul")
79
80;; mfhi, mflo, mflhxu - deliver result to gpr in 7 cycles
81(define_insn_reservation "r74k_int_mfhilo" 7
82  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
83       (eq_attr "type" "mfhi,mflo"))
84  "r74k_alu+r74k_mul")
85
86;; mthi, mtlo, mtlhx - deliver result to hi/lo, thence madd, handled as bypass
87(define_insn_reservation "r74k_int_mthilo" 7
88  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
89       (eq_attr "type" "mthi,mtlo"))
90  "r74k_alu+r74k_mul")
91
92;; div - default to 50 cycles for 32bit operands.  Faster for 8 bit,
93;; but is tricky to identify.
94(define_insn_reservation "r74k_int_div" 50
95  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
96       (eq_attr "type" "idiv"))
97  "r74k_alu+r74k_mul*50")
98
99;; call
100(define_insn_reservation "r74k_int_call" 1
101  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
102       (eq_attr "type" "call"))
103  "r74k_agen")
104
105;; branch/jump
106(define_insn_reservation "r74k_int_jump" 1
107  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
108       (eq_attr "type" "branch,jump"))
109  "r74k_agen")
110
111;; loads: lb, lbu, lh, lhu, ll, lw, lwl, lwr, lwpc, lwxs
112;; prefetch: prefetch, prefetchx
113(define_insn_reservation "r74k_int_load" 3 
114  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
115       (eq_attr "type" "load,prefetch,prefetchx"))
116  "r74k_agen")
117
118;; stores
119(define_insn_reservation "r74k_int_store" 1
120  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
121       (eq_attr "type" "store"))
122  "r74k_agen")
123
124
125;; Unknowns - Currently these include blockage, consttable and alignment
126;;            rtls.  They do not really affect scheduling latency, (blockage
127;;            affects scheduling via log links, but not used here).
128;;
129(define_insn_reservation "r74k_unknown" 1 
130  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
131       (eq_attr "type" "unknown,atomic,syncloop"))
132  "r74k_alu")
133
134(define_insn_reservation "r74k_multi" 10
135  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
136       (eq_attr "type" "multi"))
137  "(r74k_alu+r74k_agen)*10")
138
139;; --------------------------------------------------------------
140;; Bypass to Consumer
141;; --------------------------------------------------------------
142
143;; load->next use :  3 cycles (Default)
144;; load->load base:  4 cycles
145;; load->store base: 4 cycles
146(define_bypass 4 "r74k_int_load" "r74k_int_load")
147(define_bypass 4 "r74k_int_load" "r74k_int_store" "!mips_store_data_bypass_p")
148
149;; logical/move/slt/signext->next use :  1 cycles (Default)
150;; logical/move/slt/signext->load base:  2 cycles
151;; logical/move/slt/signext->store base: 2 cycles
152(define_bypass 2 "r74k_int_logical" "r74k_int_load")
153(define_bypass 2 "r74k_int_logical" "r74k_int_store"
154  "!mips_store_data_bypass_p")
155
156;; arith->next use :  2 cycles (Default)
157;; arith->load base:  3 cycles
158;; arith->store base: 3 cycles
159(define_bypass 3 "r74k_int_arith" "r74k_int_load")
160(define_bypass 3 "r74k_int_arith" "r74k_int_store" "!mips_store_data_bypass_p")
161
162;; cmove->next use :  4 cycles (Default)
163;; cmove->load base:  5 cycles
164;; cmove->store base: 5 cycles
165(define_bypass 5 "r74k_int_cmove"  "r74k_int_load")
166(define_bypass 5 "r74k_int_cmove"  "r74k_int_store"
167  "!mips_store_data_bypass_p")
168
169;; mult/madd/msub->int_mfhilo  : 4 cycles (default)
170;; mult->madd/msub             : 1 cycles
171;; madd/msub->madd/msub        : 1 cycles
172(define_bypass 1 "r74k_int_mult,r74k_int_mul3" "r74k_int_madd"
173  "mips_linked_madd_p")
174(define_bypass 1 "r74k_int_madd" "r74k_int_madd"
175  "mips_linked_madd_p")
176
177
178;; --------------------------------------------------------------
179;; DSP instructions
180;; --------------------------------------------------------------
181
182;; Non-saturating insn have the same latency as normal ALU operations,
183(define_insn_reservation "r74k_dsp_alu" 2
184  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
185       (eq_attr "type" "dspalu"))
186  "r74k_alu")
187
188;; Saturating insn takes an extra cycle.
189(define_insn_reservation "r74k_dsp_alu_sat" 3
190  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
191       (eq_attr "type" "dspalusat"))
192  "r74k_alu")
193
194;; dpaq_s, dpau, dpsq_s, dpsu, maq_s, mulsaq
195;; - delivers result to hi/lo in 6 cycle (bypass at M4)
196(define_insn_reservation "r74k_dsp_mac" 6
197  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
198       (eq_attr "type" "dspmac"))
199  "r74k_alu+r74k_mul")
200
201;; dpaq_sa, dpsq_sa, maq_sa
202;; - delivers result to hi/lo in 7 cycle (bypass at WB)
203(define_insn_reservation "r74k_dsp_mac_sat" 7
204  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
205       (eq_attr "type" "dspmacsat"))
206  "r74k_alu+r74k_mul")
207
208;; extp, extpdp, extpdpv, extpv, extr, extrv
209;; - same latency as "mul"
210(define_insn_reservation "r74k_dsp_acc_ext" 7
211  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
212       (eq_attr "type" "accext"))
213  "r74k_alu+r74k_mul")
214
215;; mthlip, shilo, shilov
216;; - same latency as "mul"
217(define_insn_reservation "r74k_dsp_acc_mod" 7
218  (and (eq_attr "cpu" "74kc,74kf2_1,74kf1_1,74kf3_2")
219       (eq_attr "type" "accmod"))
220  "r74k_alu+r74k_mul")
221
222;; dspalu   ->load/store base
223;; dspalusat->load/store base
224;; - we should never see these in real life.
225
226;; dsp_mac->dsp_mac     : 1 cycles (repeat rate of 1)
227;; dsp_mac->dsp_mac_sat : 1 cycles (repeat rate of 1)
228(define_bypass 1 "r74k_dsp_mac" "r74k_dsp_mac")
229(define_bypass 1 "r74k_dsp_mac" "r74k_dsp_mac_sat")
230
231;; dsp_mac_sat->dsp_mac_sat : 2 cycles (repeat rate of 2)
232;; dsp_mac_sat->dsp_mac     : 2 cycles (repeat rate of 2)
233(define_bypass 2 "r74k_dsp_mac_sat" "r74k_dsp_mac_sat")
234(define_bypass 2 "r74k_dsp_mac_sat" "r74k_dsp_mac")
235
236(define_bypass 1 "r74k_int_mult" "r74k_dsp_mac")
237(define_bypass 1 "r74k_int_mult" "r74k_dsp_mac_sat")
238
239(define_bypass 1 "r74k_int_mul3" "r74k_dsp_mac" "mips_linked_madd_p")
240(define_bypass 1 "r74k_int_mul3" "r74k_dsp_mac_sat" "mips_linked_madd_p")
241
242;; Assuming the following is true (bypass at M4)
243;;  AP AF AM MB M1 M2 M3 M4 WB GR GC
244;;              AP AF AM MB M1 M2 M3 M4 WB GR GC
245;; dsp_mac->dsp_acc_ext : 4 cycles
246;; dsp_mac->dsp_acc_mod : 4 cycles
247(define_bypass 4 "r74k_dsp_mac" "r74k_dsp_acc_ext")
248(define_bypass 4 "r74k_dsp_mac" "r74k_dsp_acc_mod")
249
250;; Assuming the following is true (bypass at WB)
251;;  AP AF AM MB M1 M2 M3 M4 WB GR GC
252;;                 AP AF AM MB M1 M2 M3 M4 WB GR GC
253;; dsp_mac_sat->dsp_acc_ext : 5 cycles
254;; dsp_mac_sat->dsp_acc_mod : 5 cycles
255(define_bypass 5 "r74k_dsp_mac_sat" "r74k_dsp_acc_ext")
256(define_bypass 5 "r74k_dsp_mac_sat" "r74k_dsp_acc_mod")
257
258
259;; --------------------------------------------------------------
260;; Floating Point Instructions
261;; --------------------------------------------------------------
262
263;; 74Kf FPU runs at 1:1 or 2:1 core/FPU clock ratio.
264
265;; fadd, fabs, fneg, 
266(define_insn_reservation "r74kf1_1_fadd" 4
267  (and (eq_attr "cpu" "74kf1_1")
268       (eq_attr "type" "fadd,fabs,fneg"))
269  "r74k_fpu_arith")
270
271(define_insn_reservation "r74kf2_1_fadd" 8
272  (and (eq_attr "cpu" "74kf2_1")
273       (eq_attr "type" "fadd,fabs,fneg"))
274  "r74k_fpu_arith*2")
275
276(define_insn_reservation "r74kf3_2_fadd" 6
277  (and (eq_attr "cpu" "74kf3_2")
278       (eq_attr "type" "fadd,fabs,fneg"))
279  "r74k_fpu_arith")
280
281;; fmove, fcmove
282(define_insn_reservation "r74kf1_1_fmove" 4
283  (and (eq_attr "cpu" "74kf1_1")
284       (eq_attr "type" "fmove"))
285  "r74k_fpu_arith")
286
287(define_insn_reservation "r74kf2_1_fmove" 8
288  (and (eq_attr "cpu" "74kf2_1")
289       (eq_attr "type" "fmove"))
290  "r74k_fpu_arith*2")
291
292(define_insn_reservation "r74kf3_2_fmove" 6
293  (and (eq_attr "cpu" "74kf3_2")
294       (eq_attr "type" "fmove"))
295  "r74k_fpu_arith")
296
297;; fload
298(define_insn_reservation "r74kf1_1_fload" 4
299  (and (eq_attr "cpu" "74kf1_1")
300       (eq_attr "type" "fpload,fpidxload"))
301  "r74k_agen+r74k_fpu_ldst")
302
303(define_insn_reservation "r74kf2_1_fload" 8
304  (and (eq_attr "cpu" "74kf2_1")
305       (eq_attr "type" "fpload,fpidxload"))
306  "r74k_agen+(r74k_fpu_ldst*2)")
307
308(define_insn_reservation "r74kf3_2_fload" 6
309  (and (eq_attr "cpu" "74kf3_2")
310       (eq_attr "type" "fpload,fpidxload"))
311  "r74k_agen+r74k_fpu_ldst")
312
313;; fstore
314(define_insn_reservation "r74kf1_1_fstore" 1
315  (and (eq_attr "cpu" "74kf1_1")
316       (eq_attr "type" "fpstore,fpidxstore"))
317  "r74k_agen+r74k_fpu_ldst")
318
319(define_insn_reservation "r74kf2_1_fstore" 2
320  (and (eq_attr "cpu" "74kf2_1")
321       (eq_attr "type" "fpstore,fpidxstore"))
322  "r74k_agen+(r74k_fpu_ldst*2)")
323
324(define_insn_reservation "r74kf3_2_fstore" 1
325  (and (eq_attr "cpu" "74kf3_2")
326       (eq_attr "type" "fpstore,fpidxstore"))
327  "r74k_agen+r74k_fpu_ldst")
328
329;; fmul, fmadd
330(define_insn_reservation "r74kf1_1_fmul_sf" 4
331  (and (eq_attr "cpu" "74kf1_1")
332       (and (eq_attr "type" "fmul,fmadd")
333	    (eq_attr "mode" "SF")))
334  "r74k_fpu_arith")
335
336(define_insn_reservation "r74kf2_1_fmul_sf" 8
337  (and (eq_attr "cpu" "74kf2_1")
338       (and (eq_attr "type" "fmul,fmadd")
339	    (eq_attr "mode" "SF")))
340  "r74k_fpu_arith*2")
341
342(define_insn_reservation "r74kf3_2_fmul_sf" 6
343  (and (eq_attr "cpu" "74kf3_2")
344       (and (eq_attr "type" "fmul,fmadd")
345	    (eq_attr "mode" "SF")))
346  "r74k_fpu_arith")
347
348(define_insn_reservation "r74kf1_1_fmul_df" 5
349  (and (eq_attr "cpu" "74kf1_1")
350       (and (eq_attr "type" "fmul,fmadd")
351	    (eq_attr "mode" "DF")))
352  "r74k_fpu_arith*2")
353
354(define_insn_reservation "r74kf2_1_fmul_df" 10
355  (and (eq_attr "cpu" "74kf2_1")
356       (and (eq_attr "type" "fmul,fmadd")
357	    (eq_attr "mode" "DF")))
358  "r74k_fpu_arith*4")
359
360(define_insn_reservation "r74kf3_2_fmul_df" 7
361  (and (eq_attr "cpu" "74kf3_2")
362       (and (eq_attr "type" "fmul,fmadd")
363	    (eq_attr "mode" "DF")))
364  "r74k_fpu_arith*2")
365
366;; fdiv, fsqrt
367(define_insn_reservation "r74kf1_1_fdiv_sf" 17
368  (and (eq_attr "cpu" "74kf1_1")
369       (and (eq_attr "type" "fdiv,fsqrt")
370	    (eq_attr "mode" "SF")))
371  "r74k_fpu_arith*14")
372
373(define_insn_reservation "r74kf2_1_fdiv_sf" 34
374  (and (eq_attr "cpu" "74kf2_1")
375       (and (eq_attr "type" "fdiv,fsqrt")
376	    (eq_attr "mode" "SF")))
377  "r74k_fpu_arith*28")
378
379(define_insn_reservation "r74kf3_2_fdiv_sf" 25
380  (and (eq_attr "cpu" "74kf3_2")
381       (and (eq_attr "type" "fdiv,fsqrt")
382	    (eq_attr "mode" "SF")))
383  "r74k_fpu_arith*14")
384
385(define_insn_reservation "r74kf1_1_fdiv_df" 32
386  (and (eq_attr "cpu" "74kf1_1")
387       (and (eq_attr "type" "fdiv,fsqrt")
388	    (eq_attr "mode" "DF")))
389  "r74k_fpu_arith*29")
390
391(define_insn_reservation "r74kf2_1_fdiv_df" 64
392  (and (eq_attr "cpu" "74kf2_1")
393       (and (eq_attr "type" "fdiv,fsqrt")
394	    (eq_attr "mode" "DF")))
395  "r74k_fpu_arith*58")
396
397(define_insn_reservation "r74kf3_2_fdiv_df" 48
398  (and (eq_attr "cpu" "74kf3_2")
399       (and (eq_attr "type" "fdiv,fsqrt")
400	    (eq_attr "mode" "DF")))
401  "r74k_fpu_arith*29")
402
403;; frsqrt
404(define_insn_reservation "r74kf1_1_frsqrt_sf" 17
405  (and (eq_attr "cpu" "74kf1_1")
406       (and (eq_attr "type" "frsqrt")
407	    (eq_attr "mode" "SF")))
408  "r74k_fpu_arith*14")
409
410(define_insn_reservation "r74kf2_1_frsqrt_sf" 34
411  (and (eq_attr "cpu" "74kf2_1")
412       (and (eq_attr "type" "frsqrt")
413	    (eq_attr "mode" "SF")))
414  "r74k_fpu_arith*28")
415
416(define_insn_reservation "r74kf3_2_frsqrt_sf" 25
417  (and (eq_attr "cpu" "74kf3_2")
418       (and (eq_attr "type" "frsqrt")
419	    (eq_attr "mode" "SF")))
420  "r74k_fpu_arith*14")
421
422(define_insn_reservation "r74kf1_1_frsqrt_df" 36
423  (and (eq_attr "cpu" "74kf1_1")
424       (and (eq_attr "type" "frsqrt")
425	    (eq_attr "mode" "DF")))
426  "r74k_fpu_arith*31")
427
428(define_insn_reservation "r74kf2_1_frsqrt_df" 72
429  (and (eq_attr "cpu" "74kf2_1")
430       (and (eq_attr "type" "frsqrt")
431	    (eq_attr "mode" "DF")))
432  "r74k_fpu_arith*62")
433
434(define_insn_reservation "r74kf3_2_frsqrt_df" 54
435  (and (eq_attr "cpu" "74kf3_2")
436       (and (eq_attr "type" "frsqrt")
437	    (eq_attr "mode" "DF")))
438  "r74k_fpu_arith*31")
439
440;; fcmp
441(define_insn_reservation "r74kf1_1_fcmp" 4
442  (and (eq_attr "cpu" "74kf1_1")
443       (eq_attr "type" "fcmp"))
444  "r74k_fpu_arith")
445
446(define_insn_reservation "r74kf2_1_fcmp" 8
447  (and (eq_attr "cpu" "74kf2_1")
448       (eq_attr "type" "fcmp"))
449  "r74k_fpu_arith*2")
450
451(define_insn_reservation "r74kf3_2_fcmp" 6
452  (and (eq_attr "cpu" "74kf3_2")
453       (eq_attr "type" "fcmp"))
454  "r74k_fpu_arith")
455
456;; fcvt
457(define_insn_reservation "r74kf1_1_fcvt" 4
458  (and (eq_attr "cpu" "74kf1_1")
459       (eq_attr "type" "fcvt"))
460  "r74k_fpu_arith")
461
462(define_insn_reservation "r74kf2_1_fcvt" 8
463  (and (eq_attr "cpu" "74kf2_1")
464       (eq_attr "type" "fcvt"))
465  "r74k_fpu_arith*2")
466
467(define_insn_reservation "r74kf3_2_fcvt" 6
468  (and (eq_attr "cpu" "74kf3_2")
469       (eq_attr "type" "fcvt"))
470  "r74k_fpu_arith")
471
472;; fxfer (MTC1, DMTC1: latency is 4) (MFC1, DMFC1: latency is 1)
473(define_insn_reservation "r74kf1_1_fxfer_to_c1" 4
474  (and (eq_attr "cpu" "74kf1_1")
475       (eq_attr "type" "mtc"))
476  "r74k_fpu_arith")
477
478(define_insn_reservation "r74kf2_1_fxfer_to_c1" 8
479  (and (eq_attr "cpu" "74kf2_1")
480       (eq_attr "type" "mtc"))
481  "r74k_fpu_arith*2")
482
483(define_insn_reservation "r74kf3_2_fxfer_to_c1" 6
484  (and (eq_attr "cpu" "74kf3_2")
485       (eq_attr "type" "mtc"))
486  "r74k_fpu_arith")
487
488(define_insn_reservation "r74kf1_1_fxfer_from_c1" 1
489  (and (eq_attr "cpu" "74kf1_1")
490       (eq_attr "type" "mfc"))
491  "r74k_fpu_arith")
492
493(define_insn_reservation "r74kf2_1_fxfer_from_c1" 2
494  (and (eq_attr "cpu" "74kf2_1")
495       (eq_attr "type" "mfc"))
496  "r74k_fpu_arith*2")
497
498(define_insn_reservation "r74kf3_2_fxfer_from_c1" 1
499  (and (eq_attr "cpu" "74kf3_2")
500       (eq_attr "type" "mfc"))
501  "r74k_fpu_arith")
502