comparison gcc/config/powerpcspe/power4.md @ 111:04ced10e8804

gcc 7
author kono
date Fri, 27 Oct 2017 22:46:09 +0900
parents
children 84e7813d76e9
comparison
equal deleted inserted replaced
68:561a7518be6b 111:04ced10e8804
1 ;; Scheduling description for IBM Power4 and PowerPC 970 processors.
2 ;; Copyright (C) 2003-2017 Free Software Foundation, Inc.
3 ;;
4 ;; This file is part of GCC.
5 ;;
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 3, or (at your
9 ;; option) any later version.
10 ;;
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.
15 ;;
16 ;; You should have received a copy of the GNU General Public License
17 ;; along with GCC; see the file COPYING3. If not see
18 ;; <http://www.gnu.org/licenses/>.
19
20 ;; Sources: IBM Red Book and White Paper on POWER4
21
22 ;; The POWER4 has 2 iu, 2 fpu, 2 lsu per engine (2 engines per chip).
23 ;; Instructions that update more than one register get broken into two
24 ;; (split) or more internal ops. The chip can issue up to 5
25 ;; internal ops per cycle.
26
27 (define_automaton "power4iu,power4fpu,power4vec,power4misc")
28
29 (define_cpu_unit "iu1_power4,iu2_power4" "power4iu")
30 (define_cpu_unit "lsu1_power4,lsu2_power4" "power4misc")
31 (define_cpu_unit "fpu1_power4,fpu2_power4" "power4fpu")
32 (define_cpu_unit "bpu_power4,cru_power4" "power4misc")
33 (define_cpu_unit "vec_power4,vecperm_power4" "power4vec")
34 (define_cpu_unit "du1_power4,du2_power4,du3_power4,du4_power4,du5_power4"
35 "power4misc")
36
37 (define_reservation "lsq_power4"
38 "(du1_power4,lsu1_power4)\
39 |(du2_power4,lsu2_power4)\
40 |(du3_power4,lsu2_power4)\
41 |(du4_power4,lsu1_power4)")
42
43 (define_reservation "lsuq_power4"
44 "((du1_power4+du2_power4,lsu1_power4)\
45 |(du2_power4+du3_power4,lsu2_power4)\
46 |(du3_power4+du4_power4,lsu2_power4))\
47 +(nothing,iu2_power4|nothing,iu1_power4)")
48
49 (define_reservation "iq_power4"
50 "(du1_power4|du2_power4|du3_power4|du4_power4),\
51 (iu1_power4|iu2_power4)")
52
53 (define_reservation "fpq_power4"
54 "(du1_power4|du2_power4|du3_power4|du4_power4),\
55 (fpu1_power4|fpu2_power4)")
56
57 (define_reservation "vq_power4"
58 "(du1_power4,vec_power4)\
59 |(du2_power4,vec_power4)\
60 |(du3_power4,vec_power4)\
61 |(du4_power4,vec_power4)")
62
63 (define_reservation "vpq_power4"
64 "(du1_power4,vecperm_power4)\
65 |(du2_power4,vecperm_power4)\
66 |(du3_power4,vecperm_power4)\
67 |(du4_power4,vecperm_power4)")
68
69
70 ; Dispatch slots are allocated in order conforming to program order.
71 (absence_set "du1_power4" "du2_power4,du3_power4,du4_power4,du5_power4")
72 (absence_set "du2_power4" "du3_power4,du4_power4,du5_power4")
73 (absence_set "du3_power4" "du4_power4,du5_power4")
74 (absence_set "du4_power4" "du5_power4")
75
76
77 ; Load/store
78 (define_insn_reservation "power4-load" 4 ; 3
79 (and (eq_attr "type" "load")
80 (eq_attr "sign_extend" "no")
81 (eq_attr "update" "no")
82 (eq_attr "cpu" "power4"))
83 "lsq_power4")
84
85 (define_insn_reservation "power4-load-ext" 5
86 (and (eq_attr "type" "load")
87 (eq_attr "sign_extend" "yes")
88 (eq_attr "update" "no")
89 (eq_attr "cpu" "power4"))
90 "(du1_power4+du2_power4,lsu1_power4\
91 |du2_power4+du3_power4,lsu2_power4\
92 |du3_power4+du4_power4,lsu2_power4),\
93 nothing,nothing,\
94 (iu2_power4|iu1_power4)")
95
96 (define_insn_reservation "power4-load-ext-update" 5
97 (and (eq_attr "type" "load")
98 (eq_attr "sign_extend" "yes")
99 (eq_attr "update" "yes")
100 (eq_attr "indexed" "no")
101 (eq_attr "cpu" "power4"))
102 "du1_power4+du2_power4+du3_power4+du4_power4,\
103 lsu1_power4+iu2_power4,nothing,nothing,iu2_power4")
104
105 (define_insn_reservation "power4-load-ext-update-indexed" 5
106 (and (eq_attr "type" "load")
107 (eq_attr "sign_extend" "yes")
108 (eq_attr "update" "yes")
109 (eq_attr "indexed" "yes")
110 (eq_attr "cpu" "power4"))
111 "du1_power4+du2_power4+du3_power4+du4_power4,\
112 iu1_power4,lsu2_power4+iu1_power4,nothing,nothing,iu2_power4")
113
114 (define_insn_reservation "power4-load-update-indexed" 3
115 (and (eq_attr "type" "load")
116 (eq_attr "sign_extend" "no")
117 (eq_attr "update" "yes")
118 (eq_attr "indexed" "yes")
119 (eq_attr "cpu" "power4"))
120 "du1_power4+du2_power4+du3_power4+du4_power4,\
121 iu1_power4,lsu2_power4+iu2_power4")
122
123 (define_insn_reservation "power4-load-update" 4 ; 3
124 (and (eq_attr "type" "load")
125 (eq_attr "sign_extend" "no")
126 (eq_attr "update" "yes")
127 (eq_attr "indexed" "no")
128 (eq_attr "cpu" "power4"))
129 "lsuq_power4")
130
131 (define_insn_reservation "power4-fpload" 6 ; 5
132 (and (eq_attr "type" "fpload")
133 (eq_attr "update" "no")
134 (eq_attr "cpu" "power4"))
135 "lsq_power4")
136
137 (define_insn_reservation "power4-fpload-update" 6 ; 5
138 (and (eq_attr "type" "fpload")
139 (eq_attr "update" "yes")
140 (eq_attr "cpu" "power4"))
141 "lsuq_power4")
142
143 (define_insn_reservation "power4-vecload" 6 ; 5
144 (and (eq_attr "type" "vecload")
145 (eq_attr "cpu" "power4"))
146 "lsq_power4")
147
148 (define_insn_reservation "power4-store" 12
149 (and (eq_attr "type" "store")
150 (eq_attr "update" "no")
151 (eq_attr "cpu" "power4"))
152 "((du1_power4,lsu1_power4)\
153 |(du2_power4,lsu2_power4)\
154 |(du3_power4,lsu2_power4)\
155 |(du4_power4,lsu1_power4)),\
156 (iu1_power4|iu2_power4)")
157
158 (define_insn_reservation "power4-store-update" 12
159 (and (eq_attr "type" "store")
160 (eq_attr "update" "yes")
161 (eq_attr "indexed" "no")
162 (eq_attr "cpu" "power4"))
163 "((du1_power4+du2_power4,lsu1_power4)\
164 |(du2_power4+du3_power4,lsu2_power4)\
165 |(du3_power4+du4_power4,lsu2_power4))+\
166 ((nothing,iu1_power4,iu2_power4)\
167 |(nothing,iu2_power4,iu2_power4)\
168 |(nothing,iu2_power4,iu1_power4))")
169
170 (define_insn_reservation "power4-store-update-indexed" 12
171 (and (eq_attr "type" "store")
172 (eq_attr "update" "yes")
173 (eq_attr "indexed" "yes")
174 (eq_attr "cpu" "power4"))
175 "du1_power4+du2_power4+du3_power4+du4_power4,\
176 iu1_power4,lsu2_power4+iu2_power4,iu2_power4")
177
178 (define_insn_reservation "power4-fpstore" 12
179 (and (eq_attr "type" "fpstore")
180 (eq_attr "update" "no")
181 (eq_attr "cpu" "power4"))
182 "((du1_power4,lsu1_power4)\
183 |(du2_power4,lsu2_power4)\
184 |(du3_power4,lsu2_power4)\
185 |(du4_power4,lsu1_power4)),\
186 (fpu1_power4|fpu2_power4)")
187
188 (define_insn_reservation "power4-fpstore-update" 12
189 (and (eq_attr "type" "fpstore")
190 (eq_attr "update" "yes")
191 (eq_attr "cpu" "power4"))
192 "((du1_power4+du2_power4,lsu1_power4)\
193 |(du2_power4+du3_power4,lsu2_power4)\
194 |(du3_power4+du4_power4,lsu2_power4))\
195 +(nothing,(iu1_power4|iu2_power4),(fpu1_power4|fpu2_power4))")
196
197 (define_insn_reservation "power4-vecstore" 12
198 (and (eq_attr "type" "vecstore")
199 (eq_attr "cpu" "power4"))
200 "(du1_power4,lsu1_power4,vec_power4)\
201 |(du2_power4,lsu2_power4,vec_power4)\
202 |(du3_power4,lsu2_power4,vec_power4)\
203 |(du4_power4,lsu1_power4,vec_power4)")
204
205 (define_insn_reservation "power4-llsc" 11
206 (and (eq_attr "type" "load_l,store_c,sync")
207 (eq_attr "cpu" "power4"))
208 "du1_power4+du2_power4+du3_power4+du4_power4,lsu1_power4")
209
210
211 ; Integer latency is 2 cycles
212 (define_insn_reservation "power4-integer" 2
213 (and (ior (eq_attr "type" "integer,trap,cntlz,isel")
214 (and (eq_attr "type" "add,logical,shift,exts")
215 (eq_attr "dot" "no"))
216 (and (eq_attr "type" "insert")
217 (eq_attr "size" "64")))
218 (eq_attr "cpu" "power4"))
219 "iq_power4")
220
221 (define_insn_reservation "power4-two" 2
222 (and (eq_attr "type" "two")
223 (eq_attr "cpu" "power4"))
224 "((du1_power4+du2_power4)\
225 |(du2_power4+du3_power4)\
226 |(du3_power4+du4_power4)\
227 |(du4_power4+du1_power4)),\
228 ((iu1_power4,nothing,iu2_power4)\
229 |(iu2_power4,nothing,iu2_power4)\
230 |(iu2_power4,nothing,iu1_power4)\
231 |(iu1_power4,nothing,iu1_power4))")
232
233 (define_insn_reservation "power4-three" 2
234 (and (eq_attr "type" "three")
235 (eq_attr "cpu" "power4"))
236 "(du1_power4+du2_power4+du3_power4|du2_power4+du3_power4+du4_power4\
237 |du3_power4+du4_power4+du1_power4|du4_power4+du1_power4+du2_power4),\
238 ((iu1_power4,nothing,iu2_power4,nothing,iu2_power4)\
239 |(iu2_power4,nothing,iu2_power4,nothing,iu1_power4)\
240 |(iu2_power4,nothing,iu1_power4,nothing,iu1_power4)\
241 |(iu1_power4,nothing,iu1_power4,nothing,iu2_power4))")
242
243 (define_insn_reservation "power4-insert" 4
244 (and (eq_attr "type" "insert")
245 (eq_attr "size" "32")
246 (eq_attr "cpu" "power4"))
247 "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
248 ((iu1_power4,nothing,iu2_power4)\
249 |(iu2_power4,nothing,iu2_power4)\
250 |(iu2_power4,nothing,iu1_power4))")
251
252 (define_insn_reservation "power4-cmp" 3
253 (and (ior (eq_attr "type" "cmp")
254 (and (eq_attr "type" "add,logical")
255 (eq_attr "dot" "yes")))
256 (eq_attr "cpu" "power4"))
257 "iq_power4")
258
259 (define_insn_reservation "power4-compare" 2
260 (and (eq_attr "type" "shift,exts")
261 (eq_attr "dot" "yes")
262 (eq_attr "cpu" "power4"))
263 "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
264 ((iu1_power4,iu2_power4)\
265 |(iu2_power4,iu2_power4)\
266 |(iu2_power4,iu1_power4))")
267
268 (define_bypass 4 "power4-compare" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
269
270 (define_insn_reservation "power4-lmul-cmp" 7
271 (and (eq_attr "type" "mul")
272 (eq_attr "dot" "yes")
273 (eq_attr "size" "64")
274 (eq_attr "cpu" "power4"))
275 "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
276 ((iu1_power4*6,iu2_power4)\
277 |(iu2_power4*6,iu2_power4)\
278 |(iu2_power4*6,iu1_power4))")
279
280 (define_bypass 10 "power4-lmul-cmp" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
281
282 (define_insn_reservation "power4-imul-cmp" 5
283 (and (eq_attr "type" "mul")
284 (eq_attr "dot" "yes")
285 (eq_attr "size" "32")
286 (eq_attr "cpu" "power4"))
287 "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
288 ((iu1_power4*4,iu2_power4)\
289 |(iu2_power4*4,iu2_power4)\
290 |(iu2_power4*4,iu1_power4))")
291
292 (define_bypass 8 "power4-imul-cmp" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
293
294 (define_insn_reservation "power4-lmul" 7
295 (and (eq_attr "type" "mul")
296 (eq_attr "dot" "no")
297 (eq_attr "size" "64")
298 (eq_attr "cpu" "power4"))
299 "(du1_power4|du2_power4|du3_power4|du4_power4),\
300 (iu1_power4*6|iu2_power4*6)")
301
302 (define_insn_reservation "power4-imul" 5
303 (and (eq_attr "type" "mul")
304 (eq_attr "dot" "no")
305 (eq_attr "size" "32")
306 (eq_attr "cpu" "power4"))
307 "(du1_power4|du2_power4|du3_power4|du4_power4),\
308 (iu1_power4*4|iu2_power4*4)")
309
310 (define_insn_reservation "power4-imul3" 4
311 (and (eq_attr "type" "mul")
312 (eq_attr "size" "8,16")
313 (eq_attr "cpu" "power4"))
314 "(du1_power4|du2_power4|du3_power4|du4_power4),\
315 (iu1_power4*3|iu2_power4*3)")
316
317
318 ; SPR move only executes in first IU.
319 ; Integer division only executes in second IU.
320 (define_insn_reservation "power4-idiv" 36
321 (and (eq_attr "type" "div")
322 (eq_attr "size" "32")
323 (eq_attr "cpu" "power4"))
324 "du1_power4+du2_power4,iu2_power4*35")
325
326 (define_insn_reservation "power4-ldiv" 68
327 (and (eq_attr "type" "div")
328 (eq_attr "size" "64")
329 (eq_attr "cpu" "power4"))
330 "du1_power4+du2_power4,iu2_power4*67")
331
332
333 (define_insn_reservation "power4-mtjmpr" 3
334 (and (eq_attr "type" "mtjmpr,mfjmpr")
335 (eq_attr "cpu" "power4"))
336 "du1_power4,bpu_power4")
337
338
339 ; Branches take dispatch Slot 4. The presence_sets prevent other insn from
340 ; grabbing previous dispatch slots once this is assigned.
341 (define_insn_reservation "power4-branch" 2
342 (and (eq_attr "type" "jmpreg,branch")
343 (eq_attr "cpu" "power4"))
344 "(du5_power4\
345 |du4_power4+du5_power4\
346 |du3_power4+du4_power4+du5_power4\
347 |du2_power4+du3_power4+du4_power4+du5_power4\
348 |du1_power4+du2_power4+du3_power4+du4_power4+du5_power4),bpu_power4")
349
350
351 ; Condition Register logical ops are split if non-destructive (RT != RB)
352 (define_insn_reservation "power4-crlogical" 2
353 (and (eq_attr "type" "cr_logical")
354 (eq_attr "cpu" "power4"))
355 "du1_power4,cru_power4")
356
357 (define_insn_reservation "power4-delayedcr" 4
358 (and (eq_attr "type" "delayed_cr")
359 (eq_attr "cpu" "power4"))
360 "du1_power4+du2_power4,cru_power4,cru_power4")
361
362 ; 4 mfcrf (each 3 cyc, 1/cyc) + 3 fxu
363 (define_insn_reservation "power4-mfcr" 6
364 (and (eq_attr "type" "mfcr")
365 (eq_attr "cpu" "power4"))
366 "du1_power4+du2_power4+du3_power4+du4_power4,\
367 du1_power4+du2_power4+du3_power4+du4_power4+cru_power4,\
368 cru_power4,cru_power4,cru_power4")
369
370 ; mfcrf (1 field)
371 (define_insn_reservation "power4-mfcrf" 3
372 (and (eq_attr "type" "mfcrf")
373 (eq_attr "cpu" "power4"))
374 "du1_power4,cru_power4")
375
376 ; mtcrf (1 field)
377 (define_insn_reservation "power4-mtcr" 4
378 (and (eq_attr "type" "mtcr")
379 (eq_attr "cpu" "power4"))
380 "du1_power4,iu1_power4")
381
382 ; Basic FP latency is 6 cycles
383 (define_insn_reservation "power4-fp" 6
384 (and (eq_attr "type" "fp,fpsimple,dmul")
385 (eq_attr "cpu" "power4"))
386 "fpq_power4")
387
388 (define_insn_reservation "power4-fpcompare" 5
389 (and (eq_attr "type" "fpcompare")
390 (eq_attr "cpu" "power4"))
391 "fpq_power4")
392
393 (define_insn_reservation "power4-sdiv" 33
394 (and (eq_attr "type" "sdiv,ddiv")
395 (eq_attr "cpu" "power4"))
396 "(du1_power4|du2_power4|du3_power4|du4_power4),\
397 (fpu1_power4*28|fpu2_power4*28)")
398
399 (define_insn_reservation "power4-sqrt" 40
400 (and (eq_attr "type" "ssqrt,dsqrt")
401 (eq_attr "cpu" "power4"))
402 "(du1_power4|du2_power4|du3_power4|du4_power4),\
403 (fpu1_power4*35|fpu2_power4*35)")
404
405 (define_insn_reservation "power4-isync" 2
406 (and (eq_attr "type" "isync")
407 (eq_attr "cpu" "power4"))
408 "du1_power4+du2_power4+du3_power4+du4_power4,lsu1_power4")
409
410
411 ; VMX
412 (define_insn_reservation "power4-vecsimple" 2
413 (and (eq_attr "type" "vecsimple,veclogical,vecmove")
414 (eq_attr "cpu" "power4"))
415 "vq_power4")
416
417 (define_insn_reservation "power4-veccomplex" 5
418 (and (eq_attr "type" "veccomplex")
419 (eq_attr "cpu" "power4"))
420 "vq_power4")
421
422 ; vecfp compare
423 (define_insn_reservation "power4-veccmp" 8
424 (and (eq_attr "type" "veccmp,veccmpfx")
425 (eq_attr "cpu" "power4"))
426 "vq_power4")
427
428 (define_insn_reservation "power4-vecfloat" 8
429 (and (eq_attr "type" "vecfloat")
430 (eq_attr "cpu" "power4"))
431 "vq_power4")
432
433 (define_insn_reservation "power4-vecperm" 2
434 (and (eq_attr "type" "vecperm")
435 (eq_attr "cpu" "power4"))
436 "vpq_power4")
437
438 (define_bypass 4 "power4-vecload" "power4-vecperm")
439
440 (define_bypass 3 "power4-vecsimple" "power4-vecperm")
441 (define_bypass 6 "power4-veccomplex" "power4-vecperm")
442 (define_bypass 3 "power4-vecperm"
443 "power4-vecsimple,power4-veccomplex,power4-vecfloat")
444 (define_bypass 9 "power4-vecfloat" "power4-vecperm")
445
446 (define_bypass 5 "power4-vecsimple,power4-veccomplex"
447 "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
448
449 (define_bypass 4 "power4-vecsimple,power4-vecperm" "power4-vecstore")
450 (define_bypass 7 "power4-veccomplex" "power4-vecstore")
451 (define_bypass 10 "power4-vecfloat" "power4-vecstore")