1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
|
;; AMD Athlon Scheduling
;;
;; The Athlon does contain three pipelined FP units, three integer units and
;; three address generation units.
;;
;; The predecode logic is determining boundaries of instructions in the 64
;; byte cache line. So the cache line straddling problem of K6 might be issue
;; here as well, but it is not noted in the documentation.
;;
;; Three DirectPath instructions decoders and only one VectorPath decoder
;; is available. They can decode three DirectPath instructions or one VectorPath
;; instruction per cycle.
;; Decoded macro instructions are then passed to 72 entry instruction control
;; unit, that passes
;; it to the specialized integer (18 entry) and fp (36 entry) schedulers.
;;
;; The load/store queue unit is not attached to the schedulers but
;; communicates with all the execution units separately instead.
(define_attr "athlon_decode" "direct,vector,double"
(cond [(eq_attr "type" "call,imul,idiv,other,multi,fcmov,fpspc,str,pop,cld,leave")
(const_string "vector")
(and (eq_attr "type" "push")
(match_operand 1 "memory_operand" ""))
(const_string "vector")
(and (eq_attr "type" "fmov")
(and (eq_attr "memory" "load,store")
(eq_attr "mode" "XF")))
(const_string "vector")]
(const_string "direct")))
;;
;; decode0 decode1 decode2
;; \ | /
;; instruction control unit (72 entry scheduler)
;; | |
;; integer scheduler (18) stack map
;; / | | | | \ stack rename
;; ieu0 agu0 ieu1 agu1 ieu2 agu2 scheduler
;; | agu0 | agu1 agu2 register file
;; | \ | | / | | |
;; \ /\ | / fadd fmul fstore
;; \ / \ | / fadd fmul fstore
;; imul load/store (2x) fadd fmul fstore
(define_automaton "athlon,athlon_load,athlon_mult,athlon_fp")
(define_cpu_unit "athlon-decode0" "athlon")
(define_cpu_unit "athlon-decode1" "athlon")
(define_cpu_unit "athlon-decode2" "athlon")
(define_cpu_unit "athlon-decodev" "athlon")
;; Model the fact that double decoded instruction may take 2 cycles
;; to decode when decoder2 and decoder0 in next cycle
;; is used (this is needed to allow troughput of 1.5 double decoded
;; instructions per cycle).
;;
;; In order to avoid dependence between reservation of decoder
;; and other units, we model decoder as two stage fully pipelined unit
;; and only double decoded instruction may occupy unit in the first cycle.
;; With this scheme however two double instructions can be issued cycle0.
;;
;; Avoid this by using presence set requiring decoder0 to be allocated
;; too. Vector decoded instructions then can't be issued when
;; modeled as consuming decoder0+decoder1+decoder2.
;; We solve that by specialized vector decoder unit and exclusion set.
(presence_set "athlon-decode2" "athlon-decode0")
(exclusion_set "athlon-decodev" "athlon-decode0,athlon-decode1,athlon-decode2")
(define_reservation "athlon-vector" "nothing,athlon-decodev")
(define_reservation "athlon-direct0" "nothing,athlon-decode0")
(define_reservation "athlon-direct" "nothing,
(athlon-decode0 | athlon-decode1
| athlon-decode2)")
;; Double instructions behaves like two direct instructions.
(define_reservation "athlon-double" "((athlon-decode2, athlon-decode0)
| (nothing,(athlon-decode0 + athlon-decode1))
| (nothing,(athlon-decode1 + athlon-decode2)))")
;; Agu and ieu unit results in extremely large automatons and
;; in our approximation they are hardly filled in. Only ieu
;; unit can, as issue rate is 3 and agu unit is always used
;; first in the insn reservations. Skip the models.
;(define_cpu_unit "athlon-ieu0" "athlon_ieu")
;(define_cpu_unit "athlon-ieu1" "athlon_ieu")
;(define_cpu_unit "athlon-ieu2" "athlon_ieu")
;(define_reservation "athlon-ieu" "(athlon-ieu0 | athlon-ieu1 | athlon-ieu2)")
(define_reservation "athlon-ieu" "nothing")
(define_cpu_unit "athlon-ieu0" "athlon")
;(define_cpu_unit "athlon-agu0" "athlon_agu")
;(define_cpu_unit "athlon-agu1" "athlon_agu")
;(define_cpu_unit "athlon-agu2" "athlon_agu")
;(define_reservation "athlon-agu" "(athlon-agu0 | athlon-agu1 | athlon-agu2)")
(define_reservation "athlon-agu" "nothing")
(define_cpu_unit "athlon-mult" "athlon_mult")
(define_cpu_unit "athlon-load0" "athlon_load")
(define_cpu_unit "athlon-load1" "athlon_load")
(define_reservation "athlon-load" "athlon-agu,
(athlon-load0 | athlon-load1),nothing")
;; 128bit SSE instructions issue two loads at once
(define_reservation "athlon-load2" "athlon-agu,
(athlon-load0 + athlon-load1),nothing")
(define_reservation "athlon-store" "(athlon-load0 | athlon-load1)")
;; 128bit SSE instructions issue two stores at once
(define_reservation "athlon-store2" "(athlon-load0 + athlon-load1)")
;; The FP operations start to execute at stage 12 in the pipeline, while
;; integer operations start to execute at stage 9 for Athlon and 11 for K8
;; Compensate the difference for Athlon because it results in significantly
;; smaller automata.
(define_reservation "athlon-fpsched" "nothing,nothing,nothing")
;; The floating point loads.
(define_reservation "athlon-fpload" "(athlon-fpsched + athlon-load)")
(define_reservation "athlon-fpload2" "(athlon-fpsched + athlon-load2)")
(define_reservation "athlon-fploadk8" "(athlon-fpsched + athlon-load)")
(define_reservation "athlon-fpload2k8" "(athlon-fpsched + athlon-load2)")
;; The three fp units are fully pipelined with latency of 3
(define_cpu_unit "athlon-fadd" "athlon_fp")
(define_cpu_unit "athlon-fmul" "athlon_fp")
(define_cpu_unit "athlon-fstore" "athlon_fp")
(define_reservation "athlon-fany" "(athlon-fstore | athlon-fmul | athlon-fadd)")
(define_reservation "athlon-faddmul" "(athlon-fmul | athlon-fadd)")
;; Vector operations usually consume many of pipes.
(define_reservation "athlon-fvector" "(athlon-fadd + athlon-fmul + athlon-fstore)")
;; Jump instructions are executed in the branch unit completely transparent to us
(define_insn_reservation "athlon_branch" 0
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "ibr"))
"athlon-direct,athlon-ieu")
(define_insn_reservation "athlon_call" 0
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "call,callv"))
"athlon-vector,athlon-ieu")
;; Latency of push operation is 3 cycles, but ESP value is available
;; earlier
(define_insn_reservation "athlon_push" 2
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "push"))
"athlon-direct,athlon-agu,athlon-store")
(define_insn_reservation "athlon_pop" 4
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "pop"))
"athlon-vector,athlon-load,athlon-ieu")
(define_insn_reservation "athlon_pop_k8" 3
(and (eq_attr "cpu" "k8")
(eq_attr "type" "pop"))
"athlon-double,(athlon-ieu+athlon-load)")
(define_insn_reservation "athlon_leave" 3
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "leave"))
"athlon-vector,(athlon-ieu+athlon-load)")
(define_insn_reservation "athlon_leave_k8" 3
(and (eq_attr "cpu" "k8")
(eq_attr "type" "leave"))
"athlon-double,(athlon-ieu+athlon-load)")
;; Lea executes in AGU unit with 2 cycles latency.
(define_insn_reservation "athlon_lea" 2
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "lea"))
"athlon-direct,athlon-agu,nothing")
;; Mul executes in special multiplier unit attached to IEU0
(define_insn_reservation "athlon_imul" 5
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "imul")
(eq_attr "memory" "none,unknown")))
"athlon-vector,athlon-ieu0,athlon-mult,nothing,nothing,athlon-ieu0")
;; ??? Widening multiply is vector or double.
(define_insn_reservation "athlon_imul_k8_DI" 4
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "imul")
(and (eq_attr "mode" "DI")
(eq_attr "memory" "none,unknown"))))
"athlon-direct0,athlon-ieu0,athlon-mult,nothing,athlon-ieu0")
(define_insn_reservation "athlon_imul_k8" 3
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "imul")
(eq_attr "memory" "none,unknown")))
"athlon-direct0,athlon-ieu0,athlon-mult,athlon-ieu0")
(define_insn_reservation "athlon_imul_mem" 8
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "imul")
(eq_attr "memory" "load,both")))
"athlon-vector,athlon-load,athlon-ieu,athlon-mult,nothing,nothing,athlon-ieu")
(define_insn_reservation "athlon_imul_mem_k8_DI" 7
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "imul")
(and (eq_attr "mode" "DI")
(eq_attr "memory" "load,both"))))
"athlon-vector,athlon-load,athlon-ieu,athlon-mult,nothing,athlon-ieu")
(define_insn_reservation "athlon_imul_mem_k8" 6
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "imul")
(eq_attr "memory" "load,both")))
"athlon-vector,athlon-load,athlon-ieu,athlon-mult,athlon-ieu")
;; Idiv can not execute in parallel with other instructions. Dealing with it
;; as with short latency vector instruction is good approximation avoiding
;; scheduler from trying too hard to can hide it's latency by overlap with
;; other instructions.
;; ??? Experiments show that the idiv can overlap with roughly 6 cycles
;; of the other code
(define_insn_reservation "athlon_idiv" 6
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "idiv")
(eq_attr "memory" "none,unknown")))
"athlon-vector,(athlon-ieu0*6+(athlon-fpsched,athlon-fvector))")
(define_insn_reservation "athlon_idiv_mem" 9
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "idiv")
(eq_attr "memory" "load,both")))
"athlon-vector,((athlon-load,athlon-ieu0*6)+(athlon-fpsched,athlon-fvector))")
;; The parallelism of string instructions is not documented. Model it same way
;; as idiv to create smaller automata. This probably does not matter much.
(define_insn_reservation "athlon_str" 6
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "str")
(eq_attr "memory" "load,both,store")))
"athlon-vector,athlon-load,athlon-ieu0*6")
(define_insn_reservation "athlon_idirect" 1
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "direct")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "none,unknown"))))
"athlon-direct,athlon-ieu")
(define_insn_reservation "athlon_ivector" 2
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "none,unknown"))))
"athlon-vector,athlon-ieu,athlon-ieu")
(define_insn_reservation "athlon_idirect_loadmov" 3
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "imov")
(eq_attr "memory" "load")))
"athlon-direct,athlon-load")
(define_insn_reservation "athlon_idirect_load" 4
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "direct")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-load,athlon-ieu")
(define_insn_reservation "athlon_ivector_load" 6
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "load"))))
"athlon-vector,athlon-load,athlon-ieu,athlon-ieu")
(define_insn_reservation "athlon_idirect_movstore" 1
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "imov")
(eq_attr "memory" "store")))
"athlon-direct,athlon-agu,athlon-store")
(define_insn_reservation "athlon_idirect_both" 4
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "direct")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "both"))))
"athlon-direct,athlon-load,
athlon-ieu,athlon-store,
athlon-store")
(define_insn_reservation "athlon_ivector_both" 6
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "both"))))
"athlon-vector,athlon-load,
athlon-ieu,
athlon-ieu,
athlon-store")
(define_insn_reservation "athlon_idirect_store" 1
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "direct")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "store"))))
"athlon-direct,(athlon-ieu+athlon-agu),
athlon-store")
(define_insn_reservation "athlon_ivector_store" 2
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "unit" "integer,unknown")
(eq_attr "memory" "store"))))
"athlon-vector,(athlon-ieu+athlon-agu),athlon-ieu,
athlon-store")
;; Athlon floatin point unit
(define_insn_reservation "athlon_fldxf" 12
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fmov")
(and (eq_attr "memory" "load")
(eq_attr "mode" "XF"))))
"athlon-vector,athlon-fpload2,athlon-fvector*9")
(define_insn_reservation "athlon_fldxf_k8" 13
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fmov")
(and (eq_attr "memory" "load")
(eq_attr "mode" "XF"))))
"athlon-vector,athlon-fpload2k8,athlon-fvector*9")
;; Assume superforwarding to take place so effective latency of fany op is 0.
(define_insn_reservation "athlon_fld" 0
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fmov")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-fany")
(define_insn_reservation "athlon_fld_k8" 2
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fmov")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fploadk8,athlon-fstore")
(define_insn_reservation "athlon_fstxf" 10
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fmov")
(and (eq_attr "memory" "store,both")
(eq_attr "mode" "XF"))))
"athlon-vector,(athlon-fpsched+athlon-agu),(athlon-store2+(athlon-fvector*7))")
(define_insn_reservation "athlon_fstxf_k8" 8
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fmov")
(and (eq_attr "memory" "store,both")
(eq_attr "mode" "XF"))))
"athlon-vector,(athlon-fpsched+athlon-agu),(athlon-store2+(athlon-fvector*6))")
(define_insn_reservation "athlon_fst" 4
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fmov")
(eq_attr "memory" "store,both")))
"athlon-direct,(athlon-fpsched+athlon-agu),(athlon-fstore+athlon-store)")
(define_insn_reservation "athlon_fst_k8" 2
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fmov")
(eq_attr "memory" "store,both")))
"athlon-direct,(athlon-fpsched+athlon-agu),(athlon-fstore+athlon-store)")
(define_insn_reservation "athlon_fist" 4
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "fistp"))
"athlon-direct,(athlon-fpsched+athlon-agu),(athlon-fstore+athlon-store)")
(define_insn_reservation "athlon_fmov" 2
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "fmov"))
"athlon-direct,athlon-fpsched,athlon-faddmul")
(define_insn_reservation "athlon_fadd_load" 4
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fop")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-fadd")
(define_insn_reservation "athlon_fadd_load_k8" 6
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fop")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fploadk8,athlon-fadd")
(define_insn_reservation "athlon_fadd" 4
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "fop"))
"athlon-direct,athlon-fpsched,athlon-fadd")
(define_insn_reservation "athlon_fmul_load" 4
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fmul")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-fmul")
(define_insn_reservation "athlon_fmul_load_k8" 6
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fmul")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fploadk8,athlon-fmul")
(define_insn_reservation "athlon_fmul" 4
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "fmul"))
"athlon-direct,athlon-fpsched,athlon-fmul")
(define_insn_reservation "athlon_fsgn" 2
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "fsgn"))
"athlon-direct,athlon-fpsched,athlon-fmul")
(define_insn_reservation "athlon_fdiv_load" 24
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fdiv")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-fmul")
(define_insn_reservation "athlon_fdiv_load_k8" 13
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fdiv")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fploadk8,athlon-fmul")
(define_insn_reservation "athlon_fdiv" 24
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "fdiv"))
"athlon-direct,athlon-fpsched,athlon-fmul")
(define_insn_reservation "athlon_fdiv_k8" 11
(and (eq_attr "cpu" "k8")
(eq_attr "type" "fdiv"))
"athlon-direct,athlon-fpsched,athlon-fmul")
(define_insn_reservation "athlon_fpspc_load" 103
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "fpspc")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload,athlon-fvector")
(define_insn_reservation "athlon_fpspc" 100
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "fpspc"))
"athlon-vector,athlon-fpsched,athlon-fvector")
(define_insn_reservation "athlon_fcmov_load" 7
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fcmov")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload,athlon-fvector")
(define_insn_reservation "athlon_fcmov" 7
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "fcmov"))
"athlon-vector,athlon-fpsched,athlon-fvector")
(define_insn_reservation "athlon_fcmov_load_k8" 17
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fcmov")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fploadk8,athlon-fvector")
(define_insn_reservation "athlon_fcmov_k8" 15
(and (eq_attr "cpu" "k8")
(eq_attr "type" "fcmov"))
"athlon-vector,athlon-fpsched,athlon-fvector")
;; fcomi is vector decoded by uses only one pipe.
(define_insn_reservation "athlon_fcomi_load" 3
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fcmp")
(and (eq_attr "athlon_decode" "vector")
(eq_attr "memory" "load"))))
"athlon-vector,athlon-fpload,athlon-fadd")
(define_insn_reservation "athlon_fcomi_load_k8" 5
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fcmp")
(and (eq_attr "athlon_decode" "vector")
(eq_attr "memory" "load"))))
"athlon-vector,athlon-fploadk8,athlon-fadd")
(define_insn_reservation "athlon_fcomi" 3
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "athlon_decode" "vector")
(eq_attr "type" "fcmp")))
"athlon-vector,athlon-fpsched,athlon-fadd")
(define_insn_reservation "athlon_fcom_load" 2
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "fcmp")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-fadd")
(define_insn_reservation "athlon_fcom_load_k8" 4
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "fcmp")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fploadk8,athlon-fadd")
(define_insn_reservation "athlon_fcom" 2
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "fcmp"))
"athlon-direct,athlon-fpsched,athlon-fadd")
;; Never seen by the scheduler because we still don't do post reg-stack
;; scheduling.
;(define_insn_reservation "athlon_fxch" 2
; (and (eq_attr "cpu" "athlon,k8")
; (eq_attr "type" "fxch"))
; "athlon-direct,athlon-fpsched,athlon-fany")
;; Athlon handle MMX operations in the FPU unit with shorter latencies
(define_insn_reservation "athlon_movlpd_load" 0
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssemov")
(match_operand:DF 1 "memory_operand" "")))
"athlon-direct,athlon-fpload,athlon-fany")
(define_insn_reservation "athlon_movlpd_load_k8" 2
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssemov")
(match_operand:DF 1 "memory_operand" "")))
"athlon-direct,athlon-fploadk8,athlon-fstore")
(define_insn_reservation "athlon_movaps_load_k8" 2
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssemov")
(and (eq_attr "mode" "V4SF,V2DF,TI")
(eq_attr "memory" "load"))))
"athlon-double,athlon-fpload2k8,athlon-fstore,athlon-fstore")
(define_insn_reservation "athlon_movaps_load" 0
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssemov")
(and (eq_attr "mode" "V4SF,V2DF,TI")
(eq_attr "memory" "load"))))
"athlon-vector,athlon-fpload2,(athlon-fany+athlon-fany)")
(define_insn_reservation "athlon_movss_load" 1
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssemov")
(and (eq_attr "mode" "SF,DI")
(eq_attr "memory" "load"))))
"athlon-vector,athlon-fpload,(athlon-fany*2)")
(define_insn_reservation "athlon_movss_load_k8" 1
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssemov")
(and (eq_attr "mode" "SF,DI")
(eq_attr "memory" "load"))))
"athlon-double,athlon-fploadk8,(athlon-fstore+athlon-fany)")
(define_insn_reservation "athlon_mmxsseld" 0
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "mmxmov,ssemov")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-fany")
(define_insn_reservation "athlon_mmxsseld_k8" 2
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "mmxmov,ssemov")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fploadk8,athlon-fstore")
(define_insn_reservation "athlon_mmxssest" 3
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "mmxmov,ssemov")
(and (eq_attr "mode" "V4SF,V2DF,TI")
(eq_attr "memory" "store,both"))))
"athlon-vector,(athlon-fpsched+athlon-agu),((athlon-fstore+athlon-store2)*2)")
(define_insn_reservation "athlon_mmxssest_k8" 3
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "mmxmov,ssemov")
(and (eq_attr "mode" "V4SF,V2DF,TI")
(eq_attr "memory" "store,both"))))
"athlon-double,(athlon-fpsched+athlon-agu),((athlon-fstore+athlon-store2)*2)")
(define_insn_reservation "athlon_mmxssest_short" 2
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "mmxmov,ssemov")
(eq_attr "memory" "store,both")))
"athlon-direct,(athlon-fpsched+athlon-agu),(athlon-fstore+athlon-store)")
(define_insn_reservation "athlon_movaps" 2
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssemov")
(eq_attr "mode" "V4SF,V2DF,TI")))
"athlon-double,athlon-fpsched,(athlon-faddmul+athlon-faddmul)")
(define_insn_reservation "athlon_movaps_k8" 2
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssemov")
(eq_attr "mode" "V4SF,V2DF,TI")))
"athlon-vector,athlon-fpsched,(athlon-faddmul+athlon-faddmul)")
(define_insn_reservation "athlon_mmxssemov" 2
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "mmxmov,ssemov"))
"athlon-direct,athlon-fpsched,athlon-faddmul")
(define_insn_reservation "athlon_mmxmul_load" 4
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "mmxmul")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-fmul")
(define_insn_reservation "athlon_mmxmul" 3
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "mmxmul"))
"athlon-direct,athlon-fpsched,athlon-fmul")
(define_insn_reservation "athlon_mmx_load" 3
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "unit" "mmx")
(eq_attr "memory" "load")))
"athlon-direct,athlon-fpload,athlon-faddmul")
(define_insn_reservation "athlon_mmx" 2
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "unit" "mmx"))
"athlon-direct,athlon-fpsched,athlon-faddmul")
;; SSE operations are handled by the i387 unit as well. The latency
;; is same as for i387 operations for scalar operations
(define_insn_reservation "athlon_sselog_load" 3
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "sselog")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload2,(athlon-fmul*2)")
(define_insn_reservation "athlon_sselog_load_k8" 5
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "sselog")
(eq_attr "memory" "load")))
"athlon-double,athlon-fpload2k8,(athlon-fmul*2)")
(define_insn_reservation "athlon_sselog" 3
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "sselog"))
"athlon-vector,athlon-fpsched,athlon-fmul*2")
(define_insn_reservation "athlon_sselog_k8" 3
(and (eq_attr "cpu" "k8")
(eq_attr "type" "sselog"))
"athlon-double,athlon-fpsched,athlon-fmul")
;; ??? pcmp executes in addmul, probably not wortwhile to brother about that.
(define_insn_reservation "athlon_ssecmp_load" 2
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssecmp")
(and (eq_attr "mode" "SF,DF,DI")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fpload,athlon-fadd")
(define_insn_reservation "athlon_ssecmp_load_k8" 4
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssecmp")
(and (eq_attr "mode" "SF,DF,DI,TI")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fploadk8,athlon-fadd")
(define_insn_reservation "athlon_ssecmp" 2
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "ssecmp")
(eq_attr "mode" "SF,DF,DI,TI")))
"athlon-direct,athlon-fpsched,athlon-fadd")
(define_insn_reservation "athlon_ssecmpvector_load" 3
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssecmp")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload2,(athlon-fadd*2)")
(define_insn_reservation "athlon_ssecmpvector_load_k8" 5
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssecmp")
(eq_attr "memory" "load")))
"athlon-double,athlon-fpload2k8,(athlon-fadd*2)")
(define_insn_reservation "athlon_ssecmpvector" 3
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "ssecmp"))
"athlon-vector,athlon-fpsched,(athlon-fadd*2)")
(define_insn_reservation "athlon_ssecmpvector_k8" 3
(and (eq_attr "cpu" "k8")
(eq_attr "type" "ssecmp"))
"athlon-double,athlon-fpsched,(athlon-fadd*2)")
(define_insn_reservation "athlon_ssecomi_load" 4
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssecomi")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload,athlon-fadd")
(define_insn_reservation "athlon_ssecomi_load_k8" 6
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssecomi")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fploadk8,athlon-fadd")
(define_insn_reservation "athlon_ssecomi" 4
(and (eq_attr "cpu" "athlon,k8")
(eq_attr "type" "ssecmp"))
"athlon-vector,athlon-fpsched,athlon-fadd")
(define_insn_reservation "athlon_sseadd_load" 4
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "sseadd")
(and (eq_attr "mode" "SF,DF,DI")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fpload,athlon-fadd")
(define_insn_reservation "athlon_sseadd_load_k8" 6
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "sseadd")
(and (eq_attr "mode" "SF,DF,DI")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fploadk8,athlon-fadd")
(define_insn_reservation "athlon_sseadd" 4
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "sseadd")
(eq_attr "mode" "SF,DF,DI")))
"athlon-direct,athlon-fpsched,athlon-fadd")
(define_insn_reservation "athlon_sseaddvector_load" 5
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "sseadd")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload2,(athlon-fadd*2)")
(define_insn_reservation "athlon_sseaddvector_load_k8" 7
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "sseadd")
(eq_attr "memory" "load")))
"athlon-double,athlon-fpload2k8,(athlon-fadd*2)")
(define_insn_reservation "athlon_sseaddvector" 5
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "sseadd"))
"athlon-vector,athlon-fpsched,(athlon-fadd*2)")
(define_insn_reservation "athlon_sseaddvector_k8" 5
(and (eq_attr "cpu" "k8")
(eq_attr "type" "sseadd"))
"athlon-double,athlon-fpsched,(athlon-fadd*2)")
;; Conversions behaves very irregularly and the scheduling is critical here.
;; Take each instruction separately. Assume that the mode is always set to the
;; destination one and athlon_decode is set to the K8 versions.
;; cvtss2sd
(define_insn_reservation "athlon_ssecvt_cvtss2sd_load_k8" 4
(and (eq_attr "cpu" "k8,athlon")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "direct")
(and (eq_attr "mode" "DF")
(eq_attr "memory" "load")))))
"athlon-direct,athlon-fploadk8,athlon-fstore")
(define_insn_reservation "athlon_ssecvt_cvtss2sd" 2
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "direct")
(eq_attr "mode" "DF"))))
"athlon-direct,athlon-fpsched,athlon-fstore")
;; cvtps2pd. Model same way the other double decoded FP conversions.
(define_insn_reservation "athlon_ssecvt_cvtps2pd_load_k8" 5
(and (eq_attr "cpu" "k8,athlon")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "double")
(and (eq_attr "mode" "V2DF,V4SF,TI")
(eq_attr "memory" "load")))))
"athlon-double,athlon-fpload2k8,(athlon-fstore*2)")
(define_insn_reservation "athlon_ssecvt_cvtps2pd_k8" 3
(and (eq_attr "cpu" "k8,athlon")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "double")
(eq_attr "mode" "V2DF,V4SF,TI"))))
"athlon-double,athlon-fpsched,athlon-fstore,athlon-fstore")
;; cvtsi2sd mem,reg is directpath path (cvtsi2sd reg,reg is doublepath)
;; cvtsi2sd has troughput 1 and is executed in store unit with latency of 6
(define_insn_reservation "athlon_sseicvt_cvtsi2sd_load" 6
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "direct")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "load")))))
"athlon-direct,athlon-fploadk8,athlon-fstore")
;; cvtsi2ss mem, reg is doublepath
(define_insn_reservation "athlon_sseicvt_cvtsi2ss_load" 9
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "double")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "load")))))
"athlon-vector,athlon-fpload,(athlon-fstore*2)")
(define_insn_reservation "athlon_sseicvt_cvtsi2ss_load_k8" 9
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "double")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "load")))))
"athlon-double,athlon-fploadk8,(athlon-fstore*2)")
;; cvtsi2sd reg,reg is double decoded (vector on Athlon)
(define_insn_reservation "athlon_sseicvt_cvtsi2sd_k8" 11
(and (eq_attr "cpu" "k8,athlon")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "double")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "none")))))
"athlon-double,athlon-fploadk8,athlon-fstore")
;; cvtsi2ss reg, reg is doublepath
(define_insn_reservation "athlon_sseicvt_cvtsi2ss" 14
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "none")))))
"athlon-vector,athlon-fploadk8,(athlon-fvector*2)")
;; cvtsd2ss mem,reg is doublepath, troughput unknown, latency 9
(define_insn_reservation "athlon_ssecvt_cvtsd2ss_load_k8" 9
(and (eq_attr "cpu" "k8,athlon")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "double")
(and (eq_attr "mode" "SF")
(eq_attr "memory" "load")))))
"athlon-double,athlon-fploadk8,(athlon-fstore*3)")
;; cvtsd2ss reg,reg is vectorpath, troughput unknown, latency 12
(define_insn_reservation "athlon_ssecvt_cvtsd2ss" 12
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "mode" "SF")
(eq_attr "memory" "none")))))
"athlon-vector,athlon-fpsched,(athlon-fvector*3)")
(define_insn_reservation "athlon_ssecvt_cvtpd2ps_load_k8" 8
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "mode" "V4SF,V2DF,TI")
(eq_attr "memory" "load")))))
"athlon-double,athlon-fpload2k8,(athlon-fstore*3)")
;; cvtpd2ps mem,reg is vectorpath, troughput unknown, latency 10
;; ??? Why it is fater than cvtsd2ss?
(define_insn_reservation "athlon_ssecvt_cvtpd2ps" 8
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "ssecvt")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "mode" "V4SF,V2DF,TI")
(eq_attr "memory" "none")))))
"athlon-vector,athlon-fpsched,athlon-fvector*2")
;; cvtsd2si mem,reg is doublepath, troughput 1, latency 9
(define_insn_reservation "athlon_secvt_cvtsX2si_load" 9
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "vector")
(and (eq_attr "mode" "SI,DI")
(eq_attr "memory" "load")))))
"athlon-vector,athlon-fploadk8,athlon-fvector")
;; cvtsd2si reg,reg is doublepath, troughput 1, latency 9
(define_insn_reservation "athlon_ssecvt_cvtsX2si" 9
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "double")
(and (eq_attr "mode" "SI,DI")
(eq_attr "memory" "none")))))
"athlon-vector,athlon-fpsched,athlon-fvector")
(define_insn_reservation "athlon_ssecvt_cvtsX2si_k8" 9
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "sseicvt")
(and (eq_attr "athlon_decode" "double")
(and (eq_attr "mode" "SI,DI")
(eq_attr "memory" "none")))))
"athlon-double,athlon-fpsched,athlon-fstore")
(define_insn_reservation "athlon_ssemul_load" 4
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssemul")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fpload,athlon-fmul")
(define_insn_reservation "athlon_ssemul_load_k8" 6
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssemul")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fploadk8,athlon-fmul")
(define_insn_reservation "athlon_ssemul" 4
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "ssemul")
(eq_attr "mode" "SF,DF")))
"athlon-direct,athlon-fpsched,athlon-fmul")
(define_insn_reservation "athlon_ssemulvector_load" 5
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssemul")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload2,(athlon-fmul*2)")
(define_insn_reservation "athlon_ssemulvector_load_k8" 7
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssemul")
(eq_attr "memory" "load")))
"athlon-double,athlon-fpload2k8,(athlon-fmul*2)")
(define_insn_reservation "athlon_ssemulvector" 5
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "ssemul"))
"athlon-vector,athlon-fpsched,(athlon-fmul*2)")
(define_insn_reservation "athlon_ssemulvector_k8" 5
(and (eq_attr "cpu" "k8")
(eq_attr "type" "ssemul"))
"athlon-double,athlon-fpsched,(athlon-fmul*2)")
;; divsd timings. divss is faster
(define_insn_reservation "athlon_ssediv_load" 20
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssediv")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fpload,athlon-fmul*17")
(define_insn_reservation "athlon_ssediv_load_k8" 22
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssediv")
(and (eq_attr "mode" "SF,DF")
(eq_attr "memory" "load"))))
"athlon-direct,athlon-fploadk8,athlon-fmul*17")
(define_insn_reservation "athlon_ssediv" 20
(and (eq_attr "cpu" "athlon,k8")
(and (eq_attr "type" "ssediv")
(eq_attr "mode" "SF,DF")))
"athlon-direct,athlon-fpsched,athlon-fmul*17")
(define_insn_reservation "athlon_ssedivvector_load" 39
(and (eq_attr "cpu" "athlon")
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "load")))
"athlon-vector,athlon-fpload2,athlon-fmul*34")
(define_insn_reservation "athlon_ssedivvector_load_k8" 35
(and (eq_attr "cpu" "k8")
(and (eq_attr "type" "ssediv")
(eq_attr "memory" "load")))
"athlon-double,athlon-fpload2k8,athlon-fmul*34")
(define_insn_reservation "athlon_ssedivvector" 39
(and (eq_attr "cpu" "athlon")
(eq_attr "type" "ssediv"))
"athlon-vector,athlon-fmul*34")
(define_insn_reservation "athlon_ssedivvector_k8" 39
(and (eq_attr "cpu" "k8")
(eq_attr "type" "ssediv"))
"athlon-double,athlon-fmul*34")
|
