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

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [gcc/] [config/] [rs6000/] [power4.md] - Blame information for rev 713

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

Line No. Rev Author Line
1 709 jeremybenn
;; Scheduling description for IBM Power4 and PowerPC 970 processors.
2
;;   Copyright (C) 2003, 2004, 2007, 2009 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)\
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 "cpu" "power4"))
81
  "lsq_power4")
82
 
83
(define_insn_reservation "power4-load-ext" 5
84
  (and (eq_attr "type" "load_ext")
85
       (eq_attr "cpu" "power4"))
86
  "(du1_power4+du2_power4,lsu1_power4\
87
    |du2_power4+du3_power4,lsu2_power4\
88
    |du3_power4+du4_power4,lsu2_power4),\
89
   nothing,nothing,\
90
   (iu2_power4|iu1_power4)")
91
 
92
(define_insn_reservation "power4-load-ext-update" 5
93
  (and (eq_attr "type" "load_ext_u")
94
       (eq_attr "cpu" "power4"))
95
  "du1_power4+du2_power4+du3_power4+du4_power4,\
96
   lsu1_power4+iu2_power4,nothing,nothing,iu2_power4")
97
 
98
(define_insn_reservation "power4-load-ext-update-indexed" 5
99
  (and (eq_attr "type" "load_ext_ux")
100
       (eq_attr "cpu" "power4"))
101
  "du1_power4+du2_power4+du3_power4+du4_power4,\
102
   iu1_power4,lsu2_power4+iu1_power4,nothing,nothing,iu2_power4")
103
 
104
(define_insn_reservation "power4-load-update-indexed" 3
105
  (and (eq_attr "type" "load_ux")
106
       (eq_attr "cpu" "power4"))
107
  "du1_power4+du2_power4+du3_power4+du4_power4,\
108
   iu1_power4,lsu2_power4+iu2_power4")
109
 
110
(define_insn_reservation "power4-load-update" 4 ; 3
111
  (and (eq_attr "type" "load_u")
112
       (eq_attr "cpu" "power4"))
113
  "lsuq_power4")
114
 
115
(define_insn_reservation "power4-fpload" 6 ; 5
116
  (and (eq_attr "type" "fpload")
117
       (eq_attr "cpu" "power4"))
118
  "lsq_power4")
119
 
120
(define_insn_reservation "power4-fpload-update" 6 ; 5
121
  (and (eq_attr "type" "fpload_u,fpload_ux")
122
       (eq_attr "cpu" "power4"))
123
  "lsuq_power4")
124
 
125
(define_insn_reservation "power4-vecload" 6 ; 5
126
  (and (eq_attr "type" "vecload")
127
       (eq_attr "cpu" "power4"))
128
  "lsq_power4")
129
 
130
(define_insn_reservation "power4-store" 12
131
  (and (eq_attr "type" "store")
132
       (eq_attr "cpu" "power4"))
133
  "((du1_power4,lsu1_power4)\
134
    |(du2_power4,lsu2_power4)\
135
    |(du3_power4,lsu2_power4)\
136
    |(du4_power4,lsu1_power4)),\
137
   (iu1_power4|iu2_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)\
143
    |(du2_power4+du3_power4,lsu2_power4)\
144
    |(du3_power4+du4_power4,lsu2_power4)\
145
    |(du3_power4+du4_power4,lsu2_power4))+\
146
   ((nothing,iu2_power4,iu1_power4)\
147
    |(nothing,iu2_power4,iu2_power4)\
148
    |(nothing,iu1_power4,iu2_power4)\
149
    |(nothing,iu1_power4,iu2_power4))")
150
 
151
(define_insn_reservation "power4-store-update-indexed" 12
152
  (and (eq_attr "type" "store_ux")
153
       (eq_attr "cpu" "power4"))
154
   "du1_power4+du2_power4+du3_power4+du4_power4,\
155
    iu1_power4,lsu2_power4+iu2_power4,iu2_power4")
156
 
157
(define_insn_reservation "power4-fpstore" 12
158
  (and (eq_attr "type" "fpstore")
159
       (eq_attr "cpu" "power4"))
160
  "((du1_power4,lsu1_power4)\
161
    |(du2_power4,lsu2_power4)\
162
    |(du3_power4,lsu2_power4)\
163
    |(du4_power4,lsu1_power4)),\
164
   (fpu1_power4|fpu2_power4)")
165
 
166
(define_insn_reservation "power4-fpstore-update" 12
167
  (and (eq_attr "type" "fpstore_u,fpstore_ux")
168
       (eq_attr "cpu" "power4"))
169
  "((du1_power4+du2_power4,lsu1_power4)\
170
    |(du2_power4+du3_power4,lsu2_power4)\
171
    |(du3_power4+du4_power4,lsu2_power4))\
172
   +(nothing,(iu1_power4|iu2_power4),(fpu1_power4|fpu2_power4))")
173
 
174
(define_insn_reservation "power4-vecstore" 12
175
  (and (eq_attr "type" "vecstore")
176
       (eq_attr "cpu" "power4"))
177
  "(du1_power4,lsu1_power4,vec_power4)\
178
  |(du2_power4,lsu2_power4,vec_power4)\
179
  |(du3_power4,lsu2_power4,vec_power4)\
180
  |(du4_power4,lsu1_power4,vec_power4)")
181
 
182
(define_insn_reservation "power4-llsc" 11
183
  (and (eq_attr "type" "load_l,store_c,sync")
184
       (eq_attr "cpu" "power4"))
185
  "du1_power4+du2_power4+du3_power4+du4_power4,lsu1_power4")
186
 
187
 
188
; Integer latency is 2 cycles
189
(define_insn_reservation "power4-integer" 2
190
  (and (eq_attr "type" "integer,insert_dword,shift,trap,\
191
                        var_shift_rotate,cntlz,exts,isel")
192
       (eq_attr "cpu" "power4"))
193
  "iq_power4")
194
 
195
(define_insn_reservation "power4-two" 2
196
  (and (eq_attr "type" "two")
197
       (eq_attr "cpu" "power4"))
198
  "((du1_power4+du2_power4)\
199
    |(du2_power4+du3_power4)\
200
    |(du3_power4+du4_power4)\
201
    |(du4_power4+du1_power4)),\
202
    ((iu1_power4,nothing,iu2_power4)\
203
     |(iu2_power4,nothing,iu2_power4)\
204
     |(iu2_power4,nothing,iu1_power4)\
205
     |(iu1_power4,nothing,iu1_power4))")
206
 
207
(define_insn_reservation "power4-three" 2
208
  (and (eq_attr "type" "three")
209
       (eq_attr "cpu" "power4"))
210
  "(du1_power4+du2_power4+du3_power4|du2_power4+du3_power4+du4_power4\
211
    |du3_power4+du4_power4+du1_power4|du4_power4+du1_power4+du2_power4),\
212
   ((iu1_power4,nothing,iu2_power4,nothing,iu2_power4)\
213
    |(iu2_power4,nothing,iu2_power4,nothing,iu1_power4)\
214
    |(iu2_power4,nothing,iu1_power4,nothing,iu1_power4)\
215
    |(iu1_power4,nothing,iu2_power4,nothing,iu2_power4))")
216
 
217
(define_insn_reservation "power4-insert" 4
218
  (and (eq_attr "type" "insert_word")
219
       (eq_attr "cpu" "power4"))
220
  "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
221
   ((iu1_power4,nothing,iu2_power4)\
222
    |(iu2_power4,nothing,iu2_power4)\
223
    |(iu2_power4,nothing,iu1_power4))")
224
 
225
(define_insn_reservation "power4-cmp" 3
226
  (and (eq_attr "type" "cmp,fast_compare")
227
       (eq_attr "cpu" "power4"))
228
  "iq_power4")
229
 
230
(define_insn_reservation "power4-compare" 2
231
  (and (eq_attr "type" "compare,delayed_compare,var_delayed_compare")
232
       (eq_attr "cpu" "power4"))
233
  "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
234
   ((iu1_power4,iu2_power4)\
235
    |(iu2_power4,iu2_power4)\
236
    |(iu2_power4,iu1_power4))")
237
 
238
(define_bypass 4 "power4-compare" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
239
 
240
(define_insn_reservation "power4-lmul-cmp" 7
241
  (and (eq_attr "type" "lmul_compare")
242
       (eq_attr "cpu" "power4"))
243
  "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
244
   ((iu1_power4*6,iu2_power4)\
245
    |(iu2_power4*6,iu2_power4)\
246
    |(iu2_power4*6,iu1_power4))")
247
 
248
(define_bypass 10 "power4-lmul-cmp" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
249
 
250
(define_insn_reservation "power4-imul-cmp" 5
251
  (and (eq_attr "type" "imul_compare")
252
       (eq_attr "cpu" "power4"))
253
  "(du1_power4+du2_power4|du2_power4+du3_power4|du3_power4+du4_power4),\
254
   ((iu1_power4*4,iu2_power4)\
255
    |(iu2_power4*4,iu2_power4)\
256
    |(iu2_power4*4,iu1_power4))")
257
 
258
(define_bypass 8 "power4-imul-cmp" "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
259
 
260
(define_insn_reservation "power4-lmul" 7
261
  (and (eq_attr "type" "lmul")
262
       (eq_attr "cpu" "power4"))
263
  "(du1_power4|du2_power4|du3_power4|du4_power4),\
264
   (iu1_power4*6|iu2_power4*6)")
265
 
266
(define_insn_reservation "power4-imul" 5
267
  (and (eq_attr "type" "imul")
268
       (eq_attr "cpu" "power4"))
269
  "(du1_power4|du2_power4|du3_power4|du4_power4),\
270
   (iu1_power4*4|iu2_power4*4)")
271
 
272
(define_insn_reservation "power4-imul3" 4
273
  (and (eq_attr "type" "imul2,imul3")
274
       (eq_attr "cpu" "power4"))
275
  "(du1_power4|du2_power4|du3_power4|du4_power4),\
276
   (iu1_power4*3|iu2_power4*3)")
277
 
278
 
279
; SPR move only executes in first IU.
280
; Integer division only executes in second IU.
281
(define_insn_reservation "power4-idiv" 36
282
  (and (eq_attr "type" "idiv")
283
       (eq_attr "cpu" "power4"))
284
  "du1_power4+du2_power4,iu2_power4*35")
285
 
286
(define_insn_reservation "power4-ldiv" 68
287
  (and (eq_attr "type" "ldiv")
288
       (eq_attr "cpu" "power4"))
289
  "du1_power4+du2_power4,iu2_power4*67")
290
 
291
 
292
(define_insn_reservation "power4-mtjmpr" 3
293
  (and (eq_attr "type" "mtjmpr,mfjmpr")
294
       (eq_attr "cpu" "power4"))
295
  "du1_power4,bpu_power4")
296
 
297
 
298
; Branches take dispatch Slot 4.  The presence_sets prevent other insn from
299
; grabbing previous dispatch slots once this is assigned.
300
(define_insn_reservation "power4-branch" 2
301
  (and (eq_attr "type" "jmpreg,branch")
302
       (eq_attr "cpu" "power4"))
303
  "(du5_power4\
304
   |du4_power4+du5_power4\
305
   |du3_power4+du4_power4+du5_power4\
306
   |du2_power4+du3_power4+du4_power4+du5_power4\
307
   |du1_power4+du2_power4+du3_power4+du4_power4+du5_power4),bpu_power4")
308
 
309
 
310
; Condition Register logical ops are split if non-destructive (RT != RB)
311
(define_insn_reservation "power4-crlogical" 2
312
  (and (eq_attr "type" "cr_logical")
313
       (eq_attr "cpu" "power4"))
314
  "du1_power4,cru_power4")
315
 
316
(define_insn_reservation "power4-delayedcr" 4
317
  (and (eq_attr "type" "delayed_cr")
318
       (eq_attr "cpu" "power4"))
319
  "du1_power4+du2_power4,cru_power4,cru_power4")
320
 
321
; 4 mfcrf (each 3 cyc, 1/cyc) + 3 fxu
322
(define_insn_reservation "power4-mfcr" 6
323
  (and (eq_attr "type" "mfcr")
324
       (eq_attr "cpu" "power4"))
325
  "du1_power4+du2_power4+du3_power4+du4_power4,\
326
   du1_power4+du2_power4+du3_power4+du4_power4+cru_power4,\
327
   cru_power4,cru_power4,cru_power4")
328
 
329
; mfcrf (1 field)
330
(define_insn_reservation "power4-mfcrf" 3
331
  (and (eq_attr "type" "mfcrf")
332
       (eq_attr "cpu" "power4"))
333
  "du1_power4,cru_power4")
334
 
335
; mtcrf (1 field)
336
(define_insn_reservation "power4-mtcr" 4
337
  (and (eq_attr "type" "mtcr")
338
       (eq_attr "cpu" "power4"))
339
  "du1_power4,iu1_power4")
340
 
341
; Basic FP latency is 6 cycles
342
(define_insn_reservation "power4-fp" 6
343
  (and (eq_attr "type" "fp,dmul")
344
       (eq_attr "cpu" "power4"))
345
  "fpq_power4")
346
 
347
(define_insn_reservation "power4-fpcompare" 5
348
  (and (eq_attr "type" "fpcompare")
349
       (eq_attr "cpu" "power4"))
350
  "fpq_power4")
351
 
352
(define_insn_reservation "power4-sdiv" 33
353
  (and (eq_attr "type" "sdiv,ddiv")
354
       (eq_attr "cpu" "power4"))
355
  "(du1_power4|du2_power4|du3_power4|du4_power4),\
356
   (fpu1_power4*28|fpu2_power4*28)")
357
 
358
(define_insn_reservation "power4-sqrt" 40
359
  (and (eq_attr "type" "ssqrt,dsqrt")
360
       (eq_attr "cpu" "power4"))
361
  "(du1_power4|du2_power4|du3_power4|du4_power4),\
362
   (fpu1_power4*35|fpu2_power4*35)")
363
 
364
(define_insn_reservation "power4-isync" 2
365
  (and (eq_attr "type" "isync")
366
       (eq_attr "cpu" "power4"))
367
  "du1_power4+du2_power4+du3_power4+du4_power4,lsu1_power4")
368
 
369
 
370
; VMX
371
(define_insn_reservation "power4-vecsimple" 2
372
  (and (eq_attr "type" "vecsimple")
373
       (eq_attr "cpu" "power4"))
374
  "vq_power4")
375
 
376
(define_insn_reservation "power4-veccomplex" 5
377
  (and (eq_attr "type" "veccomplex")
378
       (eq_attr "cpu" "power4"))
379
  "vq_power4")
380
 
381
; vecfp compare
382
(define_insn_reservation "power4-veccmp" 8
383
  (and (eq_attr "type" "veccmp")
384
       (eq_attr "cpu" "power4"))
385
  "vq_power4")
386
 
387
(define_insn_reservation "power4-vecfloat" 8
388
  (and (eq_attr "type" "vecfloat")
389
       (eq_attr "cpu" "power4"))
390
  "vq_power4")
391
 
392
(define_insn_reservation "power4-vecperm" 2
393
  (and (eq_attr "type" "vecperm")
394
       (eq_attr "cpu" "power4"))
395
  "vpq_power4")
396
 
397
(define_bypass 4 "power4-vecload" "power4-vecperm")
398
 
399
(define_bypass 3 "power4-vecsimple" "power4-vecperm")
400
(define_bypass 6 "power4-veccomplex" "power4-vecperm")
401
(define_bypass 3 "power4-vecperm"
402
                 "power4-vecsimple,power4-veccomplex,power4-vecfloat")
403
(define_bypass 9 "power4-vecfloat" "power4-vecperm")
404
 
405
(define_bypass 5 "power4-vecsimple,power4-veccomplex"
406
                 "power4-branch,power4-crlogical,power4-delayedcr,power4-mfcr,power4-mfcrf")
407
 
408
(define_bypass 4 "power4-vecsimple,power4-vecperm" "power4-vecstore")
409
(define_bypass 7 "power4-veccomplex" "power4-vecstore")
410
(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.