OpenCores
URL https://opencores.org/ocsvn/openrisc/openrisc/trunk

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [gnu-old/] [gcc-4.2.2/] [gcc/] [config/] [rs6000/] [power4.md] - Blame information for rev 856

Go to most recent revision | Details | Compare with Previous | View Log

Line No. Rev Author Line
1 38 julius 
;; Scheduling description for IBM Power4 and PowerPC 970 processors.
2 
;;   Copyright (C) 2003, 2004, 2007 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 
;; .
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+iu2_power4)\
45 
                    |(du2_power4+du3_power4,lsu2_power4+iu2_power4)\
46 
                    |(du3_power4+du4_power4,lsu2_power4+iu1_power4)")
47 
 
48 
(define_reservation "iq_power4"
49 
                    "(du1_power4,iu1_power4)\
50 
                    |(du2_power4,iu2_power4)\
51 
                    |(du3_power4,iu2_power4)\
52 
                    |(du4_power4,iu1_power4)")
53 
 
54 
(define_reservation "fpq_power4"
55 
                    "(du1_power4,fpu1_power4)\
56 
                    |(du2_power4,fpu2_power4)\
57 
                    |(du3_power4,fpu2_power4)\
58 
                    |(du4_power4,fpu1_power4)")
59 
 
60 
(define_reservation "vq_power4"
61 
                    "(du1_power4,vec_power4)\
62 
                    |(du2_power4,vec_power4)\
63 
                    |(du3_power4,vec_power4)\
64 
                    |(du4_power4,vec_power4)")
65 
 
66 
(define_reservation "vpq_power4"
67 
                    "(du1_power4,vecperm_power4)\
68 
                    |(du2_power4,vecperm_power4)\
69 
                    |(du3_power4,vecperm_power4)\
70 
                    |(du4_power4,vecperm_power4)")
71 
 
72 
 
73 
; Dispatch slots are allocated in order conforming to program order.
74 
(absence_set "du1_power4" "du2_power4,du3_power4,du4_power4,du5_power4")
75 
(absence_set "du2_power4" "du3_power4,du4_power4,du5_power4")
76 
(absence_set "du3_power4" "du4_power4,du5_power4")
77 
(absence_set "du4_power4" "du5_power4")
78 
 
79 
 
80 
; Load/store
81 
(define_insn_reservation "power4-load" 4 ; 3
82 
  (and (eq_attr "type" "load")
83 
       (eq_attr "cpu" "power4"))
84 
  "lsq_power4")
85 
 
86 
(define_insn_reservation "power4-load-ext" 5
87 
  (and (eq_attr "type" "load_ext")
88 
       (eq_attr "cpu" "power4"))
89 
  "(du1_power4+du2_power4,lsu1_power4,nothing,nothing,iu2_power4)\
90 
  |(du2_power4+du3_power4,lsu2_power4,nothing,nothing,iu2_power4)\
91 
  |(du3_power4+du4_power4,lsu2_power4,nothing,nothing,iu1_power4)")
92 
 
93 
(define_insn_reservation "power4-load-ext-update" 5
94 
  (and (eq_attr "type" "load_ext_u")
95 
       (eq_attr "cpu" "power4"))
96 
  "du1_power4+du2_power4+du3_power4+du4_power4,\
97 
   lsu1_power4+iu2_power4,nothing,nothing,iu2_power4")
98 
 
99 
(define_insn_reservation "power4-load-ext-update-indexed" 5
100 
  (and (eq_attr "type" "load_ext_ux")
101 
       (eq_attr "cpu" "power4"))
102 
  "du1_power4+du2_power4+du3_power4+du4_power4,\
103 
   iu1_power4,lsu2_power4+iu1_power4,nothing,nothing,iu2_power4")
104 
 
105 
(define_insn_reservation "power4-load-update-indexed" 3
106 
  (and (eq_attr "type" "load_ux")
107 
       (eq_attr "cpu" "power4"))
108 
  "du1_power4+du2_power4+du3_power4+du4_power4,\
109 
   iu1_power4,lsu2_power4+iu2_power4")
110 
 
111 
(define_insn_reservation "power4-load-update" 4 ; 3
112 
  (and (eq_attr "type" "load_u")
113 
       (eq_attr "cpu" "power4"))
114 
  "lsuq_power4")
115 
 
116 
(define_insn_reservation "power4-fpload" 6 ; 5
117 
  (and (eq_attr "type" "fpload")
118 
       (eq_attr "cpu" "power4"))
119 
  "lsq_power4")
120 
 
121 
(define_insn_reservation "power4-fpload-update" 6 ; 5
122 
  (and (eq_attr "type" "fpload_u,fpload_ux")
123 
       (eq_attr "cpu" "power4"))
124 
  "lsuq_power4")
125 
 
126 
(define_insn_reservation "power4-vecload" 6 ; 5
127 
  (and (eq_attr "type" "vecload")
128 
       (eq_attr "cpu" "power4"))
129 
  "lsq_power4")
130 
 
131 
(define_insn_reservation "power4-store" 12
132 
  (and (eq_attr "type" "store")
133 
       (eq_attr "cpu" "power4"))
134 
  "(du1_power4,lsu1_power4,iu1_power4)\
135 
  |(du2_power4,lsu2_power4,iu2_power4)\
136 
  |(du3_power4,lsu2_power4,iu2_power4)\
137 
  |(du4_power4,lsu1_power4,iu1_power4)")
138 
 
139 
(define_insn_reservation "power4-store-update" 12
140 
  (and (eq_attr "type" "store_u")
141 
       (eq_attr "cpu" "power4"))
142 
  "(du1_power4+du2_power4,lsu1_power4+iu2_power4,iu1_power4)\
143 
  |(du2_power4+du3_power4,lsu2_power4+iu2_power4,iu2_power4)\
144 
  |(du3_power4+du4_power4,lsu2_power4+iu1_power4,iu2_power4)\
145 
  |(du3_power4+du4_power4,lsu2_power4,iu1_power4,iu2_power4)")
146 
 
147 
(define_insn_reservation "power4-store-update-indexed" 12
148 
  (and (eq_attr "type" "store_ux")
149 
       (eq_attr "cpu" "power4"))
150 
   "du1_power4+du2_power4+du3_power4+du4_power4,\
151 
    iu1_power4,lsu2_power4+iu2_power4,iu2_power4")
152 
 
153 
(define_insn_reservation "power4-fpstore" 12
154 
  (and (eq_attr "type" "fpstore")
155 
       (eq_attr "cpu" "power4"))
156 
  "(du1_power4,lsu1_power4,fpu1_power4)\
157 
  |(du2_power4,lsu2_power4,fpu2_power4)\
158 
  |(du3_power4,lsu2_power4,fpu2_power4)\
159 
  |(du4_power4,lsu1_power4,fpu1_power4)")
160 
 
161 
(define_insn_reservation "power4-fpstore-update" 12
162 
  (and (eq_attr "type" "fpstore_u,fpstore_ux")
163 
       (eq_attr "cpu" "power4"))
164 
  "(du1_power4+du2_power4,lsu1_power4+iu2_power4,fpu1_power4)\
165 
  |(du2_power4+du3_power4,lsu2_power4+iu2_power4,fpu2_power4)\
166 
  |(du3_power4+du4_power4,lsu2_power4+iu1_power4,fpu2_power4)")
167 
 
168 
(define_insn_reservation "power4-vecstore" 12
169 
  (and (eq_attr "type" "vecstore")
170 
       (eq_attr "cpu" "power4"))
171 
  "(du1_power4,lsu1_power4,vec_power4)\
172 
  |(du2_power4,lsu2_power4,vec_power4)\
173 
  |(du3_power4,lsu2_power4,vec_power4)\
174 
  |(du4_power4,lsu1_power4,vec_power4)")
175 
 
176 
(define_insn_reservation "power4-llsc" 11
177 
  (and (eq_attr "type" "load_l,store_c,sync")
178 
       (eq_attr "cpu" "power4"))
179 
  "du1_power4+du2_power4+du3_power4+du4_power4,\
180 
  lsu1_power4")
181 
 
182 
 
183 
; Integer latency is 2 cycles
184 
(define_insn_reservation "power4-integer" 2
185 
  (and (eq_attr "type" "integer")
186 
       (eq_attr "cpu" "power4"))
187 
  "iq_power4")
188 
 
189 
(define_insn_reservation "power4-two" 2
190 
  (and (eq_attr "type" "two")
191 
       (eq_attr "cpu" "power4"))
192 
  "(du1_power4+du2_power4,iu1_power4,nothing,iu2_power4)\
193 
  |(du2_power4+du3_power4,iu2_power4,nothing,iu2_power4)\
194 
  |(du3_power4+du4_power4,iu2_power4,nothing,iu1_power4)\
195 
  |(du4_power4+du1_power4,iu1_power4,nothing,iu1_power4)")
196 
 
197 
(define_insn_reservation "power4-three" 2
198 
  (and (eq_attr "type" "three")
199 
       (eq_attr "cpu" "power4"))
200 
  "(du1_power4+du2_power4+du3_power4,\
201 
    iu1_power4,nothing,iu2_power4,nothing,iu2_power4)\
202 
  |(du2_power4+du3_power4+du4_power4,\
203 
    iu2_power4,nothing,iu2_power4,nothing,iu1_power4)\
204 
  |(du3_power4+du4_power4+du1_power4,\
205 
    iu2_power4,nothing,iu1_power4,nothing,iu1_power4)\
206 
  |(du4_power4+du1_power4+du2_power4,\
207 
    iu1_power4,nothing,iu2_power4,nothing,iu2_power4)")
208 
 
209 
(define_insn_reservation "power4-insert" 4
210 
  (and (eq_attr "type" "insert_word")
211 
       (eq_attr "cpu" "power4"))
212 
  "(du1_power4+du2_power4,iu1_power4,nothing,iu2_power4)\
213 
  |(du2_power4+du3_power4,iu2_power4,nothing,iu2_power4)\
214 
  |(du3_power4+du4_power4,iu2_power4,nothing,iu1_power4)")
215 
 
216 
(define_insn_reservation "power4-cmp" 3
217 
  (and (eq_attr "type" "cmp,fast_compare")
218 
       (eq_attr "cpu" "power4"))
219 
  "iq_power4")
220 
 
221 
(define_insn_reservation "power4-compare" 2
222 
  (and (eq_attr "type" "compare,delayed_compare")
223 
       (eq_attr "cpu" "power4"))
224 
  "(du1_power4+du2_power4,iu1_power4,iu2_power4)\
225 
  |(du2_power4+du3_power4,iu2_power4,iu2_power4)\
226 
  |(du3_power4+du4_power4,iu2_power4,iu1_power4)")
227 
 
228 
(define_bypass 4 "power4-compare" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
229 
 
230 
(define_insn_reservation "power4-lmul-cmp" 7
231 
  (and (eq_attr "type" "lmul_compare")
232 
       (eq_attr "cpu" "power4"))
233 
  "(du1_power4+du2_power4,iu1_power4*6,iu2_power4)\
234 
  |(du2_power4+du3_power4,iu2_power4*6,iu2_power4)\
235 
  |(du3_power4+du4_power4,iu2_power4*6,iu1_power4)")
236 
 
237 
(define_bypass 10 "power4-lmul-cmp" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
238 
 
239 
(define_insn_reservation "power4-imul-cmp" 5
240 
  (and (eq_attr "type" "imul_compare")
241 
       (eq_attr "cpu" "power4"))
242 
  "(du1_power4+du2_power4,iu1_power4*4,iu2_power4)\
243 
  |(du2_power4+du3_power4,iu2_power4*4,iu2_power4)\
244 
  |(du3_power4+du4_power4,iu2_power4*4,iu1_power4)")
245 
 
246 
(define_bypass 8 "power4-imul-cmp" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
247 
 
248 
(define_insn_reservation "power4-lmul" 7
249 
  (and (eq_attr "type" "lmul")
250 
       (eq_attr "cpu" "power4"))
251 
  "(du1_power4,iu1_power4*6)\
252 
  |(du2_power4,iu2_power4*6)\
253 
  |(du3_power4,iu2_power4*6)\
254 
  |(du4_power4,iu1_power4*6)")
255 
 
256 
(define_insn_reservation "power4-imul" 5
257 
  (and (eq_attr "type" "imul")
258 
       (eq_attr "cpu" "power4"))
259 
  "(du1_power4,iu1_power4*4)\
260 
  |(du2_power4,iu2_power4*4)\
261 
  |(du3_power4,iu2_power4*4)\
262 
  |(du4_power4,iu1_power4*4)")
263 
 
264 
(define_insn_reservation "power4-imul3" 4
265 
  (and (eq_attr "type" "imul2,imul3")
266 
       (eq_attr "cpu" "power4"))
267 
  "(du1_power4,iu1_power4*3)\
268 
  |(du2_power4,iu2_power4*3)\
269 
  |(du3_power4,iu2_power4*3)\
270 
  |(du4_power4,iu1_power4*3)")
271 
 
272 
 
273 
; SPR move only executes in first IU.
274 
; Integer division only executes in second IU.
275 
(define_insn_reservation "power4-idiv" 36
276 
  (and (eq_attr "type" "idiv")
277 
       (eq_attr "cpu" "power4"))
278 
  "du1_power4+du2_power4,iu2_power4*35")
279 
 
280 
(define_insn_reservation "power4-ldiv" 68
281 
  (and (eq_attr "type" "ldiv")
282 
       (eq_attr "cpu" "power4"))
283 
  "du1_power4+du2_power4,iu2_power4*67")
284 
 
285 
 
286 
(define_insn_reservation "power4-mtjmpr" 3
287 
  (and (eq_attr "type" "mtjmpr,mfjmpr")
288 
       (eq_attr "cpu" "power4"))
289 
  "du1_power4,bpu_power4")
290 
 
291 
 
292 
; Branches take dispatch Slot 4.  The presence_sets prevent other insn from
293 
; grabbing previous dispatch slots once this is assigned.
294 
(define_insn_reservation "power4-branch" 2
295 
  (and (eq_attr "type" "jmpreg,branch")
296 
       (eq_attr "cpu" "power4"))
297 
  "(du5_power4\
298 
   |du4_power4+du5_power4\
299 
   |du3_power4+du4_power4+du5_power4\
300 
   |du2_power4+du3_power4+du4_power4+du5_power4\
301 
   |du1_power4+du2_power4+du3_power4+du4_power4+du5_power4),bpu_power4")
302 
 
303 
 
304 
; Condition Register logical ops are split if non-destructive (RT != RB)
305 
(define_insn_reservation "power4-crlogical" 2
306 
  (and (eq_attr "type" "cr_logical")
307 
       (eq_attr "cpu" "power4"))
308 
  "du1_power4,cru_power4")
309 
 
310 
(define_insn_reservation "power4-delayedcr" 4
311 
  (and (eq_attr "type" "delayed_cr")
312 
       (eq_attr "cpu" "power4"))
313 
  "du1_power4+du2_power4,cru_power4,cru_power4")
314 
 
315 
; 4 mfcrf (each 3 cyc, 1/cyc) + 3 fxu
316 
(define_insn_reservation "power4-mfcr" 6
317 
  (and (eq_attr "type" "mfcr")
318 
       (eq_attr "cpu" "power4"))
319 
  "du1_power4+du2_power4+du3_power4+du4_power4,\
320 
   du1_power4+du2_power4+du3_power4+du4_power4+cru_power4,\
321 
   cru_power4,cru_power4,cru_power4")
322 
 
323 
; mfcrf (1 field)
324 
(define_insn_reservation "power4-mfcrf" 3
325 
  (and (eq_attr "type" "mfcrf")
326 
       (eq_attr "cpu" "power4"))
327 
  "du1_power4,cru_power4")
328 
 
329 
; mtcrf (1 field)
330 
(define_insn_reservation "power4-mtcr" 4
331 
  (and (eq_attr "type" "mtcr")
332 
       (eq_attr "cpu" "power4"))
333 
  "du1_power4,iu1_power4")
334 
 
335 
; Basic FP latency is 6 cycles
336 
(define_insn_reservation "power4-fp" 6
337 
  (and (eq_attr "type" "fp,dmul")
338 
       (eq_attr "cpu" "power4"))
339 
  "fpq_power4")
340 
 
341 
(define_insn_reservation "power4-fpcompare" 5
342 
  (and (eq_attr "type" "fpcompare")
343 
       (eq_attr "cpu" "power4"))
344 
  "fpq_power4")
345 
 
346 
(define_insn_reservation "power4-sdiv" 33
347 
  (and (eq_attr "type" "sdiv,ddiv")
348 
       (eq_attr "cpu" "power4"))
349 
  "(du1_power4,fpu1_power4*28)\
350 
  |(du2_power4,fpu2_power4*28)\
351 
  |(du3_power4,fpu2_power4*28)\
352 
  |(du4_power4,fpu1_power4*28)")
353 
 
354 
(define_insn_reservation "power4-sqrt" 40
355 
  (and (eq_attr "type" "ssqrt,dsqrt")
356 
       (eq_attr "cpu" "power4"))
357 
  "(du1_power4,fpu1_power4*35)\
358 
  |(du2_power4,fpu2_power4*35)\
359 
  |(du3_power4,fpu2_power4*35)\
360 
  |(du4_power4,fpu2_power4*35)")
361 
 
362 
(define_insn_reservation "power4-isync" 2
363 
  (and (eq_attr "type" "isync")
364 
       (eq_attr "cpu" "power4"))
365 
  "du1_power4+du2_power4+du3_power4+du4_power4,\
366 
  lsu1_power4")
367 
 
368 
 
369 
; VMX
370 
(define_insn_reservation "power4-vecsimple" 2
371 
  (and (eq_attr "type" "vecsimple")
372 
       (eq_attr "cpu" "power4"))
373 
  "vq_power4")
374 
 
375 
(define_insn_reservation "power4-veccomplex" 5
376 
  (and (eq_attr "type" "veccomplex")
377 
       (eq_attr "cpu" "power4"))
378 
  "vq_power4")
379 
 
380 
; vecfp compare
381 
(define_insn_reservation "power4-veccmp" 8
382 
  (and (eq_attr "type" "veccmp")
383 
       (eq_attr "cpu" "power4"))
384 
  "vq_power4")
385 
 
386 
(define_insn_reservation "power4-vecfloat" 8
387 
  (and (eq_attr "type" "vecfloat")
388 
       (eq_attr "cpu" "power4"))
389 
  "vq_power4")
390 
 
391 
(define_insn_reservation "power4-vecperm" 2
392 
  (and (eq_attr "type" "vecperm")
393 
       (eq_attr "cpu" "power4"))
394 
  "vpq_power4")
395 
 
396 
(define_bypass 4 "power4-vecload" "power4-vecperm")
397 
 
398 
(define_bypass 3 "power4-vecsimple" "power4-vecperm")
399 
(define_bypass 6 "power4-veccomplex" "power4-vecperm")
400 
(define_bypass 3 "power4-vecperm"
401 
                 "power4-vecsimple,power4-veccomplex,power4-vecfloat")
402 
(define_bypass 9 "power4-vecfloat" "power4-vecperm")
403 
 
404 
(define_bypass 5 "power4-vecsimple,power4-veccomplex"
405 
                 "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
406 
 
407 
(define_bypass 4 "power4-vecsimple,power4-vecperm" "power4-vecstore")
408 
(define_bypass 7 "power4-veccomplex" "power4-vecstore")
409 
(define_bypass 10 "power4-vecfloat" "power4-vecstore")

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.