-
Notifications
You must be signed in to change notification settings - Fork 86
/
Copy pathcmm_helpers.ml
4108 lines (3568 loc) · 134 KB
/
cmm_helpers.ml
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
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
(**************************************************************************)
(* *)
(* OCaml *)
(* *)
(* Xavier Leroy, projet Cristal, INRIA Rocquencourt *)
(* *)
(* Copyright 1996 Institut National de Recherche en Informatique et *)
(* en Automatique. *)
(* *)
(* All rights reserved. This file is distributed under the terms of *)
(* the GNU Lesser General Public License version 2.1, with the *)
(* special exception on linking described in the file LICENSE. *)
(* *)
(**************************************************************************)
[@@@ocaml.warning "+a-4-9-40-41-42-44-45"]
module V = Backend_var
module VP = Backend_var.With_provenance
open Cmm
open Arch
type arity =
{ function_kind : Lambda.function_kind;
params_layout : Lambda.layout list;
return_layout : Lambda.layout
}
(* Local binding of complex expressions *)
let bind name arg fn =
match arg with
| Cvar _ | Cconst_int _ | Cconst_natint _ | Cconst_symbol _ -> fn arg
| _ ->
let id = V.create_local name in
Clet (VP.create id, arg, fn (Cvar id))
let bind_list name args fn =
let rec aux bound_args = function
| [] -> fn bound_args
| arg :: args ->
bind name arg (fun bound_arg -> aux (bound_arg :: bound_args) args)
in
aux [] (List.rev args)
let caml_black = Nativeint.shift_left (Nativeint.of_int 3) 8
let caml_local =
Nativeint.shift_left (Nativeint.of_int (if Config.runtime5 then 3 else 2)) 8
(* cf. runtime/caml/gc.h *)
(* Loads *)
let mk_load_immut memory_chunk =
Cload { memory_chunk; mutability = Immutable; is_atomic = false }
let mk_load_mut memory_chunk =
Cload { memory_chunk; mutability = Mutable; is_atomic = false }
let mk_load_atomic memory_chunk =
Cload { memory_chunk; mutability = Mutable; is_atomic = true }
(* Block headers. Meaning of the tag field: see stdlib/obj.ml *)
let floatarray_tag dbg = Cconst_int (Obj.double_array_tag, dbg)
type t =
| Scan_all
| Scan_prefix of int
module Mixed_block_support : sig
val assert_mixed_block_support : unit -> unit
val make_header : Nativeint.t -> scannable_prefix:int -> Nativeint.t
end = struct
(* CR mixed blocks v1: This "8" is duplicated in [typedecl.ml]. We should fix
up this duplication when we make the "8" configurable. *)
let required_reserved_header_bits = 8
let required_addr_size_bits = 64
(* Many of these checks are duplicated *)
(* CR mixed blocks v1: This is also duplicated in [typedecl.ml]. *)
(* Why 2? We'd subtract 1 if the mixed block encoding could use all 8 bits of
the prefix. But the all-0 prefix means "not a mixed block", so we can't use
the all-0 pattern, and we must subtract 2 instead. *)
let max_scannable_prefix = (1 lsl required_reserved_header_bits) - 2
let max_header =
(1 lsl (required_addr_size_bits - required_reserved_header_bits)) - 1
|> Nativeint.of_int
let assert_mixed_block_support =
lazy
(if not Config.native_compiler
then Misc.fatal_error "Mixed blocks are only supported in native code";
let reserved_header_bits = Config.reserved_header_bits in
let addr_size_bits = Arch.size_addr * 8 in
match
( reserved_header_bits = required_reserved_header_bits,
addr_size_bits = required_addr_size_bits )
with
| true, true -> ()
| false, true ->
Misc.fatal_errorf
"Need %d reserved header bits for mixed blocks; got %d"
required_reserved_header_bits reserved_header_bits
| _, false ->
Misc.fatal_errorf
"Mixed blocks only supported on %d bit platforms; got %d"
required_addr_size_bits addr_size_bits)
let assert_mixed_block_support () = Lazy.force assert_mixed_block_support
let make_header header ~scannable_prefix =
assert_mixed_block_support ();
if scannable_prefix > max_scannable_prefix
then
Misc.fatal_errorf "Scannable prefix too big (%d > %d)" scannable_prefix
max_scannable_prefix;
(* This means we crash the compiler if someone tries to write a mixed record
with too many fields, but you effectively can't: you'd need something
like 2^46 fields. *)
if header > max_header
then
Misc.fatal_errorf
"Header too big for the mixed block encoding to be added (%nd > %nd)"
header max_header;
Nativeint.add
(Nativeint.shift_left
(Nativeint.of_int (scannable_prefix + 1))
(required_addr_size_bits - required_reserved_header_bits))
header
end
(* CR mshinwell: update to use NOT_MARKABLE terminology *)
let block_header ?(scannable_prefix = Scan_all) tag sz =
let hdr =
Nativeint.add
(Nativeint.shift_left (Nativeint.of_int sz) 10)
(Nativeint.of_int tag)
in
match scannable_prefix with
| Scan_all -> hdr
| Scan_prefix scannable_prefix ->
Mixed_block_support.make_header hdr ~scannable_prefix
(* Static data corresponding to "value"s must be marked black in case we are in
no-naked-pointers mode. See [caml_darken] and the code below that emits
structured constants and static module definitions. *)
let black_block_header tag sz = Nativeint.logor (block_header tag sz) caml_black
let local_block_header ?scannable_prefix tag sz =
Nativeint.logor (block_header ?scannable_prefix tag sz) caml_local
let white_closure_header sz = block_header Obj.closure_tag sz
let black_closure_header sz = black_block_header Obj.closure_tag sz
let local_closure_header sz = local_block_header Obj.closure_tag sz
let infix_header ofs = block_header Obj.infix_tag ofs
let boxedfloat32_header = block_header Obj.custom_tag 2
let boxedfloat32_local_header = local_block_header Obj.custom_tag 2
let float_header = block_header Obj.double_tag (size_float / size_addr)
let float_local_header =
local_block_header Obj.double_tag (size_float / size_addr)
let boxedvec128_header = block_header Obj.abstract_tag (size_vec128 / size_addr)
let boxedvec128_local_header =
local_block_header Obj.abstract_tag (size_vec128 / size_addr)
let floatarray_header len =
(* Zero-sized float arrays have tag zero for consistency with
[caml_alloc_float_array]. *)
assert (len >= 0);
if len = 0
then block_header 0 0
else block_header Obj.double_array_tag (len * size_float / size_addr)
let string_header len =
block_header Obj.string_tag ((len + size_addr) / size_addr)
let boxedint32_header = block_header Obj.custom_tag 2
let boxedint64_header = block_header Obj.custom_tag (1 + (8 / size_addr))
let boxedintnat_header = block_header Obj.custom_tag 2
let boxedint32_local_header = local_block_header Obj.custom_tag 2
let boxedint64_local_header =
local_block_header Obj.custom_tag (1 + (8 / size_addr))
let boxedintnat_local_header = local_block_header Obj.custom_tag 2
let black_custom_header ~size = black_block_header Obj.custom_tag size
let custom_header ~size = block_header Obj.custom_tag size
let custom_local_header ~size = local_block_header Obj.custom_tag size
let caml_float32_ops = "caml_float32_ops"
let caml_nativeint_ops = "caml_nativeint_ops"
let caml_int32_ops = "caml_int32_ops"
let caml_int64_ops = "caml_int64_ops"
let pos_arity_in_closinfo = (8 * size_addr) - 8
(* arity = the top 8 bits of the closinfo word *)
let pack_closure_info ~arity ~startenv ~is_last =
assert (-128 <= arity && arity <= 127);
assert (0 <= startenv && startenv < 1 lsl (pos_arity_in_closinfo - 2));
Nativeint.(
add
(shift_left (of_int arity) pos_arity_in_closinfo)
(add
(shift_left
(Bool.to_int is_last |> Nativeint.of_int)
(pos_arity_in_closinfo - 1))
(add (shift_left (of_int startenv) 1) 1n)))
let closure_info' ~arity ~startenv ~is_last =
let arity =
match arity with
| Lambda.Tupled, l -> -List.length l
| Lambda.Curried _, l -> List.length l
in
pack_closure_info ~arity ~startenv ~is_last
let closure_info ~(arity : arity) ~startenv ~is_last =
closure_info'
~arity:(arity.function_kind, arity.params_layout)
~startenv ~is_last
let alloc_boxedfloat32_header mode dbg =
match mode with
| Lambda.Alloc_heap -> Cconst_natint (boxedfloat32_header, dbg)
| Lambda.Alloc_local -> Cconst_natint (boxedfloat32_local_header, dbg)
let alloc_float_header mode dbg =
match mode with
| Lambda.Alloc_heap -> Cconst_natint (float_header, dbg)
| Lambda.Alloc_local -> Cconst_natint (float_local_header, dbg)
let alloc_boxedvec128_header mode dbg =
match mode with
| Lambda.Alloc_heap -> Cconst_natint (boxedvec128_header, dbg)
| Lambda.Alloc_local -> Cconst_natint (boxedvec128_local_header, dbg)
let alloc_floatarray_header len dbg = Cconst_natint (floatarray_header len, dbg)
let alloc_closure_header ~mode sz dbg =
match (mode : Lambda.alloc_mode) with
| Alloc_heap -> Cconst_natint (white_closure_header sz, dbg)
| Alloc_local -> Cconst_natint (local_closure_header sz, dbg)
let alloc_infix_header ofs dbg = Cconst_natint (infix_header ofs, dbg)
let alloc_boxedint32_header mode dbg =
match mode with
| Lambda.Alloc_heap -> Cconst_natint (boxedint32_header, dbg)
| Lambda.Alloc_local -> Cconst_natint (boxedint32_local_header, dbg)
let alloc_boxedint64_header mode dbg =
match mode with
| Lambda.Alloc_heap -> Cconst_natint (boxedint64_header, dbg)
| Lambda.Alloc_local -> Cconst_natint (boxedint64_local_header, dbg)
let alloc_boxedintnat_header mode dbg =
match mode with
| Lambda.Alloc_heap -> Cconst_natint (boxedintnat_header, dbg)
| Lambda.Alloc_local -> Cconst_natint (boxedintnat_local_header, dbg)
(* Integers *)
let max_repr_int = max_int asr 1
let min_repr_int = min_int asr 1
let int_const dbg n =
if n <= max_repr_int && n >= min_repr_int
then Cconst_int ((n lsl 1) + 1, dbg)
else
Cconst_natint
(Nativeint.add (Nativeint.shift_left (Nativeint.of_int n) 1) 1n, dbg)
let natint_const_untagged dbg n =
if n > Nativeint.of_int max_int || n < Nativeint.of_int min_int
then Cconst_natint (n, dbg)
else Cconst_int (Nativeint.to_int n, dbg)
let cint_const n =
Cint (Nativeint.add (Nativeint.shift_left (Nativeint.of_int n) 1) 1n)
let add_no_overflow n x c dbg =
let d = n + x in
if d = 0 then c else Cop (Caddi, [c; Cconst_int (d, dbg)], dbg)
let rec add_const c n dbg =
if n = 0
then c
else
match c with
| Cconst_int (x, _) when Misc.no_overflow_add x n -> Cconst_int (x + n, dbg)
| Cop (Caddi, [Cconst_int (x, _); c], _) when Misc.no_overflow_add n x ->
add_no_overflow n x c dbg
| Cop (Caddi, [c; Cconst_int (x, _)], _) when Misc.no_overflow_add n x ->
add_no_overflow n x c dbg
| Cop (Csubi, [Cconst_int (x, _); c], _) when Misc.no_overflow_add n x ->
Cop (Csubi, [Cconst_int (n + x, dbg); c], dbg)
| Cop (Csubi, [c; Cconst_int (x, _)], _) when Misc.no_overflow_sub n x ->
add_const c (n - x) dbg
| c -> Cop (Caddi, [c; Cconst_int (n, dbg)], dbg)
let incr_int c dbg = add_const c 1 dbg
let decr_int c dbg = add_const c (-1) dbg
let rec add_int c1 c2 dbg =
match c1, c2 with
| Cconst_int (n, _), c | c, Cconst_int (n, _) -> add_const c n dbg
| Cop (Caddi, [c1; Cconst_int (n1, _)], _), c2 ->
add_const (add_int c1 c2 dbg) n1 dbg
| c1, Cop (Caddi, [c2; Cconst_int (n2, _)], _) ->
add_const (add_int c1 c2 dbg) n2 dbg
| _, _ -> Cop (Caddi, [c1; c2], dbg)
let rec sub_int c1 c2 dbg =
match c1, c2 with
| c1, Cconst_int (n2, _) when n2 <> min_int -> add_const c1 (-n2) dbg
| c1, Cop (Caddi, [c2; Cconst_int (n2, _)], _) when n2 <> min_int ->
add_const (sub_int c1 c2 dbg) (-n2) dbg
| Cop (Caddi, [c1; Cconst_int (n1, _)], _), c2 ->
add_const (sub_int c1 c2 dbg) n1 dbg
| c1, c2 -> Cop (Csubi, [c1; c2], dbg)
let neg_int c dbg = sub_int (Cconst_int (0, dbg)) c dbg
let rec lsl_int c1 c2 dbg =
match c1, c2 with
| Cop (Clsl, [c; Cconst_int (n1, _)], _), Cconst_int (n2, _)
when n1 > 0 && n2 > 0 && n1 + n2 < size_int * 8 ->
Cop (Clsl, [c; Cconst_int (n1 + n2, dbg)], dbg)
| Cop (Caddi, [c1; Cconst_int (n1, _)], _), Cconst_int (n2, _)
when Misc.no_overflow_lsl n1 n2 ->
add_const (lsl_int c1 c2 dbg) (n1 lsl n2) dbg
| _, _ -> Cop (Clsl, [c1; c2], dbg)
let is_power2 n = n = 1 lsl Misc.log2 n
and mult_power2 c n dbg = lsl_int c (Cconst_int (Misc.log2 n, dbg)) dbg
let rec mul_int c1 c2 dbg =
match c1, c2 with
| c, Cconst_int (0, _) | Cconst_int (0, _), c ->
Csequence (c, Cconst_int (0, dbg))
| c, Cconst_int (1, _) | Cconst_int (1, _), c -> c
| c, Cconst_int (-1, _) | Cconst_int (-1, _), c ->
sub_int (Cconst_int (0, dbg)) c dbg
| c, Cconst_int (n, _) when is_power2 n -> mult_power2 c n dbg
| Cconst_int (n, _), c when is_power2 n -> mult_power2 c n dbg
| Cop (Caddi, [c; Cconst_int (n, _)], _), Cconst_int (k, _)
| Cconst_int (k, _), Cop (Caddi, [c; Cconst_int (n, _)], _)
when Misc.no_overflow_mul n k ->
add_const (mul_int c (Cconst_int (k, dbg)) dbg) (n * k) dbg
| c1, c2 -> Cop (Cmuli, [c1; c2], dbg)
(* identify cmm operations whose result is guaranteed to be small integers (e.g.
in the range [min_int / 4; max_int / 4]) *)
let guaranteed_to_be_small_int = function
| Cop ((Ccmpi _ | Ccmpf _), _, _) ->
(* integer/float comparisons return either [1] or [0]. *)
true
| _ -> false
let ignore_low_bit_int = function
| Cop
( Caddi,
[(Cop (Clsl, [_; Cconst_int (n, _)], _) as c); Cconst_int (1, _)],
_ )
when n > 0 ->
c
| Cop (Cor, [c; Cconst_int (1, _)], _) -> c
| c -> c
let lsr_int c1 c2 dbg =
match c2 with
| Cconst_int (0, _) -> c1
| Cconst_int (n, _) when n > 0 -> Cop (Clsr, [ignore_low_bit_int c1; c2], dbg)
| _ -> Cop (Clsr, [c1; c2], dbg)
let asr_int c1 c2 dbg =
match c2 with
| Cconst_int (0, _) -> c1
| Cconst_int (n, _) when n > 0 -> (
match ignore_low_bit_int c1 with
(* some operations always return small enough integers that it is safe and
correct to optimise [asr (lsl x 1) 1] into [x]. *)
| Cop (Clsl, [c; Cconst_int (1, _)], _)
when n = 1 && guaranteed_to_be_small_int c ->
c
| c1' -> Cop (Casr, [c1'; c2], dbg))
| _ -> Cop (Casr, [c1; c2], dbg)
let tag_int i dbg =
match i with
| Cconst_int (n, _) -> int_const dbg n
| Cop (Casr, [c; Cconst_int (n, _)], _) when n > 0 ->
Cop
(Cor, [asr_int c (Cconst_int (n - 1, dbg)) dbg; Cconst_int (1, dbg)], dbg)
| c -> incr_int (lsl_int c (Cconst_int (1, dbg)) dbg) dbg
let untag_int i dbg =
match i with
| Cconst_int (n, _) -> Cconst_int (n asr 1, dbg)
| Cop (Cor, [Cop (Casr, [c; Cconst_int (n, _)], _); Cconst_int (1, _)], _)
when n > 0 && n < (size_int * 8) - 1 ->
Cop (Casr, [c; Cconst_int (n + 1, dbg)], dbg)
| Cop (Cor, [Cop (Clsr, [c; Cconst_int (n, _)], _); Cconst_int (1, _)], _)
when n > 0 && n < (size_int * 8) - 1 ->
Cop (Clsr, [c; Cconst_int (n + 1, dbg)], dbg)
| c -> asr_int c (Cconst_int (1, dbg)) dbg
let mk_not dbg cmm =
match cmm with
| Cop (Caddi, [Cop (Clsl, [c; Cconst_int (1, _)], _); Cconst_int (1, _)], dbg')
-> (
match c with
| Cop (Ccmpi cmp, [c1; c2], dbg'') ->
tag_int
(Cop (Ccmpi (negate_integer_comparison cmp), [c1; c2], dbg''))
dbg'
| Cop (Ccmpa cmp, [c1; c2], dbg'') ->
tag_int
(Cop (Ccmpa (negate_integer_comparison cmp), [c1; c2], dbg''))
dbg'
| Cop (Ccmpf (w, cmp), [c1; c2], dbg'') ->
tag_int
(Cop (Ccmpf (w, negate_float_comparison cmp), [c1; c2], dbg''))
dbg'
| _ ->
(* 0 -> 3, 1 -> 1 *)
Cop
( Csubi,
[Cconst_int (3, dbg); Cop (Clsl, [c; Cconst_int (1, dbg)], dbg)],
dbg ))
| Cconst_int (3, _) -> Cconst_int (1, dbg)
| Cconst_int (1, _) -> Cconst_int (3, dbg)
| c ->
(* 1 -> 3, 3 -> 1 *)
Cop (Csubi, [Cconst_int (4, dbg); c], dbg)
let mk_compare_ints_untagged dbg a1 a2 =
bind "int_cmp" a2 (fun a2 ->
bind "int_cmp" a1 (fun a1 ->
let op1 = Cop (Ccmpi Cgt, [a1; a2], dbg) in
let op2 = Cop (Ccmpi Clt, [a1; a2], dbg) in
sub_int op1 op2 dbg))
let mk_compare_ints dbg a1 a2 =
match a1, a2 with
| Cconst_int (c1, _), Cconst_int (c2, _) -> int_const dbg (Int.compare c1 c2)
| Cconst_natint (c1, _), Cconst_natint (c2, _) ->
int_const dbg (Nativeint.compare c1 c2)
| Cconst_int (c1, _), Cconst_natint (c2, _) ->
int_const dbg Nativeint.(compare (of_int c1) c2)
| Cconst_natint (c1, _), Cconst_int (c2, _) ->
int_const dbg Nativeint.(compare c1 (of_int c2))
| a1, a2 -> tag_int (mk_compare_ints_untagged dbg a1 a2) dbg
let mk_compare_floats_gen ~tag_result ~width dbg a1 a2 =
bind "float_cmp" a2 (fun a2 ->
bind "float_cmp" a1 (fun a1 ->
let op1 = Cop (Ccmpf (width, CFgt), [a1; a2], dbg) in
let op2 = Cop (Ccmpf (width, CFlt), [a1; a2], dbg) in
let op3 = Cop (Ccmpf (width, CFeq), [a1; a1], dbg) in
let op4 = Cop (Ccmpf (width, CFeq), [a2; a2], dbg) in
(* If both operands a1 and a2 are not NaN, then op3 = op4 = 1, and the
result is op1 - op2.
If at least one of the operands is NaN, then op1 = op2 = 0, and the
result is op3 - op4, which orders NaN before other values.
To detect if the operand is NaN, we use the property:
for all x, NaN is not equal to x, even if x is NaN.
Therefore, op3 is 0 if and only if a1 is NaN, and op4 is 0 if and
only if a2 is NaN. See also caml_float_compare_unboxed in
runtime/floats.c *)
let result =
add_int (sub_int op1 op2 dbg) (sub_int op3 op4 dbg) dbg
in
if tag_result then tag_int result dbg else result))
let mk_compare_floats = mk_compare_floats_gen ~tag_result:true ~width:Float64
let mk_compare_floats_untagged =
mk_compare_floats_gen ~tag_result:false ~width:Float64
let mk_compare_float32s_untagged =
mk_compare_floats_gen ~tag_result:false ~width:Float32
let create_loop body dbg =
let cont = Lambda.next_raise_count () in
let call_cont = Cexit (Lbl cont, [], []) in
let body = Csequence (body, call_cont) in
Ccatch (Recursive, [cont, [], body, dbg, false], call_cont, Any)
(* Turning integer divisions into multiply-high then shift. The
[division_parameters] function is used in module Emit for those target
platforms that support this optimization. *)
(* Unsigned comparison between native integers. *)
let ucompare x y = Nativeint.(compare (add x min_int) (add y min_int))
(* Unsigned division and modulus at type nativeint. Algorithm: Hacker's Delight
section 9.3 *)
let udivmod n d =
Nativeint.(
if d < 0n
then if ucompare n d < 0 then 0n, n else 1n, sub n d
else
let q = shift_left (div (shift_right_logical n 1) d) 1 in
let r = sub n (mul q d) in
if ucompare r d >= 0 then succ q, sub r d else q, r)
(* Compute division parameters. Algorithm: Hacker's Delight chapter 10, fig
10-1. *)
let divimm_parameters d =
Nativeint.(
assert (d > 0n);
let twopsm1 = min_int in
(* 2^31 for 32-bit archs, 2^63 for 64-bit archs *)
let nc = sub (pred twopsm1) (snd (udivmod twopsm1 d)) in
let rec loop p (q1, r1) (q2, r2) =
let p = p + 1 in
let q1 = shift_left q1 1 and r1 = shift_left r1 1 in
let q1, r1 = if ucompare r1 nc >= 0 then succ q1, sub r1 nc else q1, r1 in
let q2 = shift_left q2 1 and r2 = shift_left r2 1 in
let q2, r2 = if ucompare r2 d >= 0 then succ q2, sub r2 d else q2, r2 in
let delta = sub d r2 in
if ucompare q1 delta < 0 || (q1 = delta && r1 = 0n)
then loop p (q1, r1) (q2, r2)
else succ q2, p - size
in
loop (size - 1) (udivmod twopsm1 nc) (udivmod twopsm1 d))
(* The result [(m, p)] of [divimm_parameters d] satisfies the following
inequality:
2^(wordsize + p) < m * d <= 2^(wordsize + p) + 2^(p + 1) (i)
from which it follows that
floor(n / d) = floor(n * m / 2^(wordsize+p)), if 0 <= n < 2^(wordsize-1)
ceil(n / d) = floor(n * m / 2^(wordsize+p)) + 1, if -2^(wordsize-1) <= n < 0
The correctness condition (i) above can be checked by the code below. It was
exhaustively tested for values of d from 2 to 10^9 in the wordsize = 64
case.
* let add2 (xh, xl) (yh, yl) =
* let zl = add xl yl and zh = add xh yh in
* (if ucompare zl xl < 0 then succ zh else zh), zl
*
* let shl2 (xh, xl) n =
* assert (0 < n && n < size + size);
* if n < size
* then
* logor (shift_left xh n) (shift_right_logical xl (size - n)),
* shift_left xl n
* else shift_left xl (n - size), 0n
*
* let mul2 x y =
* let halfsize = size / 2 in
* let halfmask = pred (shift_left 1n halfsize) in
* let xl = logand x halfmask and xh = shift_right_logical x halfsize in
* let yl = logand y halfmask and yh = shift_right_logical y halfsize in
* add2
* (mul xh yh, 0n)
* (add2
* (shl2 (0n, mul xl yh) halfsize)
* (add2 (shl2 (0n, mul xh yl) halfsize) (0n, mul xl yl)))
*
* let ucompare2 (xh, xl) (yh, yl) =
* let c = ucompare xh yh in
* if c = 0 then ucompare xl yl else c
*
* let validate d m p =
* let md = mul2 m d in
* let one2 = 0n, 1n in
* let twoszp = shl2 one2 (size + p) in
* let twop1 = shl2 one2 (p + 1) in
* ucompare2 twoszp md < 0 && ucompare2 md (add2 twoszp twop1) <= 0
*)
let raise_symbol dbg symb =
Cop
(Craise Lambda.Raise_regular, [Cconst_symbol (global_symbol symb, dbg)], dbg)
let rec div_int c1 c2 is_safe dbg =
match c1, c2 with
| c1, Cconst_int (0, _) ->
Csequence (c1, raise_symbol dbg "caml_exn_Division_by_zero")
| c1, Cconst_int (1, _) -> c1
| Cconst_int (n1, _), Cconst_int (n2, _) -> Cconst_int (n1 / n2, dbg)
| c1, Cconst_int (n, _) when n <> min_int ->
let l = Misc.log2 n in
if n = 1 lsl l
then
(* Algorithm:
t = shift-right-signed(c1, l - 1)
t = shift-right(t, W - l)
t = c1 + t res = shift-right-signed(c1 + t, l) *)
Cop
( Casr,
[ bind "dividend" c1 (fun c1 ->
assert (l >= 1);
let t = asr_int c1 (Cconst_int (l - 1, dbg)) dbg in
let t = lsr_int t (Cconst_int (Nativeint.size - l, dbg)) dbg in
add_int c1 t dbg);
Cconst_int (l, dbg) ],
dbg )
else if n < 0
then
sub_int
(Cconst_int (0, dbg))
(div_int c1 (Cconst_int (-n, dbg)) is_safe dbg)
dbg
else
let m, p = divimm_parameters (Nativeint.of_int n) in
(* Algorithm:
t = multiply-high-signed(c1, m) if m < 0,
t = t + c1 if p > 0,
t = shift-right-signed(t, p)
res = t + sign-bit(c1) *)
bind "dividend" c1 (fun c1 ->
let t =
Cop
(Cmulhi { signed = true }, [c1; natint_const_untagged dbg m], dbg)
in
let t = if m < 0n then Cop (Caddi, [t; c1], dbg) else t in
let t =
if p > 0 then Cop (Casr, [t; Cconst_int (p, dbg)], dbg) else t
in
add_int t (lsr_int c1 (Cconst_int (Nativeint.size - 1, dbg)) dbg) dbg)
| c1, c2 when !Clflags.unsafe || is_safe = Lambda.Unsafe ->
Cop (Cdivi, [c1; c2], dbg)
| c1, c2 ->
bind "divisor" c2 (fun c2 ->
bind "dividend" c1 (fun c1 ->
Cifthenelse
( c2,
dbg,
Cop (Cdivi, [c1; c2], dbg),
dbg,
raise_symbol dbg "caml_exn_Division_by_zero",
dbg,
Any )))
let mod_int c1 c2 is_safe dbg =
match c1, c2 with
| c1, Cconst_int (0, _) ->
Csequence (c1, raise_symbol dbg "caml_exn_Division_by_zero")
| c1, Cconst_int ((1 | -1), _) -> Csequence (c1, Cconst_int (0, dbg))
| Cconst_int (n1, _), Cconst_int (n2, _) -> Cconst_int (n1 mod n2, dbg)
| c1, (Cconst_int (n, _) as c2) when n <> min_int ->
let l = Misc.log2 n in
if n = 1 lsl l
then
(* Algorithm:
t = shift-right-signed(c1, l - 1)
t = shift-right(t, W - l)
t = c1 + t
t = bit-and(t, -n)
res = c1 - t *)
bind "dividend" c1 (fun c1 ->
assert (l >= 1);
let t = asr_int c1 (Cconst_int (l - 1, dbg)) dbg in
let t = lsr_int t (Cconst_int (Nativeint.size - l, dbg)) dbg in
let t = add_int c1 t dbg in
let t = Cop (Cand, [t; Cconst_int (-n, dbg)], dbg) in
sub_int c1 t dbg)
else
bind "dividend" c1 (fun c1 ->
sub_int c1 (mul_int (div_int c1 c2 is_safe dbg) c2 dbg) dbg)
| c1, c2 when !Clflags.unsafe || is_safe = Lambda.Unsafe ->
(* Flambda already generates that test *)
Cop (Cmodi, [c1; c2], dbg)
| c1, c2 ->
bind "divisor" c2 (fun c2 ->
bind "dividend" c1 (fun c1 ->
Cifthenelse
( c2,
dbg,
Cop (Cmodi, [c1; c2], dbg),
dbg,
raise_symbol dbg "caml_exn_Division_by_zero",
dbg,
Any )))
(* Division or modulo on boxed integers. The overflow case min_int / -1 can
occur, in which case we force x / -1 = -x and x mod -1 = 0. (PR#5513). *)
let is_different_from x = function
| Cconst_int (n, _) -> n <> x
| Cconst_natint (n, _) -> n <> Nativeint.of_int x
| _ -> false
let safe_divmod_bi mkop kind is_safe mkm1 c1 c2 bi dbg =
bind "divisor" c2 (fun c2 ->
bind "dividend" c1 (fun c1 ->
let c = mkop c1 c2 is_safe dbg in
if Arch.division_crashes_on_overflow && bi <> Primitive.Pint32
&& not (is_different_from (-1) c2)
then
Cifthenelse
( Cop (Ccmpi Cne, [c2; Cconst_int (-1, dbg)], dbg),
dbg,
c,
dbg,
mkm1 c1 dbg,
dbg,
kind )
else c))
let safe_div_bi is_safe =
safe_divmod_bi div_int Any is_safe (fun c1 dbg ->
Cop (Csubi, [Cconst_int (0, dbg); c1], dbg))
let safe_mod_bi is_safe =
safe_divmod_bi mod_int Any is_safe (fun _ dbg -> Cconst_int (0, dbg))
(* Bool *)
let test_bool dbg cmm =
match cmm with
| Cop (Caddi, [Cop (Clsl, [c; Cconst_int (1, _)], _); Cconst_int (1, _)], _)
->
c
| Cconst_int (n, dbg) ->
if n = 1 then Cconst_int (0, dbg) else Cconst_int (1, dbg)
| c -> Cop (Ccmpi Cne, [c; Cconst_int (1, dbg)], dbg)
(* Float *)
let box_float32 dbg mode exp =
Cop
( Calloc mode,
[ alloc_boxedfloat32_header mode dbg;
Cconst_symbol (global_symbol caml_float32_ops, dbg);
exp ],
dbg )
let unbox_float32 dbg =
map_tail ~kind:Any (function
| Cop (Calloc _, [Cconst_natint (hdr, _); _ops; c], _)
when Nativeint.equal hdr boxedfloat32_header
|| Nativeint.equal hdr boxedfloat32_local_header ->
c
| Cconst_symbol (s, _dbg) as cmm -> (
match Cmmgen_state.structured_constant_of_sym s.sym_name with
| Some (Const_float32 x) -> Cconst_float32 (x, dbg) (* or keep _dbg? *)
| _ ->
Cop
( mk_load_immut (Single { reg = Float32 }),
[Cop (Cadda, [cmm; Cconst_int (size_addr, dbg)], dbg)],
dbg ))
| cmm ->
Cop
( mk_load_immut (Single { reg = Float32 }),
[Cop (Cadda, [cmm; Cconst_int (size_addr, dbg)], dbg)],
dbg ))
let box_float dbg m c = Cop (Calloc m, [alloc_float_header m dbg; c], dbg)
let unbox_float dbg =
map_tail ~kind:Any (function
| Cop (Calloc _, [Cconst_natint (hdr, _); c], _)
when Nativeint.equal hdr float_header
|| Nativeint.equal hdr float_local_header ->
c
| Cconst_symbol (s, _dbg) as cmm -> (
match Cmmgen_state.structured_constant_of_sym s.sym_name with
| Some (Const_float x) -> Cconst_float (x, dbg) (* or keep _dbg? *)
| _ -> Cop (mk_load_immut Double, [cmm], dbg))
| cmm -> Cop (mk_load_immut Double, [cmm], dbg))
(* Vectors *)
let box_vec128 dbg m c = Cop (Calloc m, [alloc_boxedvec128_header m dbg; c], dbg)
let unbox_vec128 dbg =
(* Boxed vectors are not 16-byte aligned by the GC, so we must use an
unaligned load. *)
map_tail ~kind:Any (function
| Cop (Calloc _, [Cconst_natint (hdr, _); c], _)
when Nativeint.equal hdr boxedvec128_header
|| Nativeint.equal hdr boxedvec128_local_header ->
c
| Cconst_symbol (s, _dbg) as cmm -> (
match Cmmgen_state.structured_constant_of_sym s.sym_name with
| Some (Const_vec128 { low; high }) ->
Cconst_vec128 ({ low; high }, dbg) (* or keep _dbg? *)
| _ -> Cop (mk_load_immut Onetwentyeight_unaligned, [cmm], dbg))
| cmm -> Cop (mk_load_immut Onetwentyeight_unaligned, [cmm], dbg))
(* Complex *)
let box_complex dbg c_re c_im =
Cop
(Calloc Lambda.alloc_heap, [alloc_floatarray_header 2 dbg; c_re; c_im], dbg)
let complex_re c dbg = Cop (mk_load_immut Double, [c], dbg)
let complex_im c dbg =
Cop
( mk_load_immut Double,
[Cop (Cadda, [c; Cconst_int (size_float, dbg)], dbg)],
dbg )
(* Unit *)
let return_unit dbg c = Csequence (c, Cconst_int (1, dbg))
let strided_field_address ptr ~index ~stride dbg =
if index * stride = 0
then ptr
else Cop (Cadda, [ptr; Cconst_int (index * stride, dbg)], dbg)
let field_address ?(memory_chunk = Word_val) ptr n dbg =
if n = 0
then ptr
else
let field_size_in_bytes =
match memory_chunk with
| Byte_unsigned | Byte_signed -> 1
| Sixteen_unsigned | Sixteen_signed -> 2
| Thirtytwo_unsigned | Thirtytwo_signed -> 4
| Single { reg = Float64 | Float32 } ->
assert (size_float = 8);
(* unclear what to do if this is false *)
size_float / 2
| Word_int -> size_int
| Word_val -> size_addr
| Double -> size_float
| Onetwentyeight_unaligned | Onetwentyeight_aligned -> size_vec128
in
Cop (Cadda, [ptr; Cconst_int (n * field_size_in_bytes, dbg)], dbg)
let get_field_gen_given_memory_chunk memory_chunk mutability ptr n dbg =
Cop
( Cload { memory_chunk; mutability; is_atomic = false },
[field_address ptr n dbg],
dbg )
let get_field_gen mut ptr n dbg =
get_field_gen_given_memory_chunk Word_val mut ptr n dbg
let get_field_codepointer mut ptr n dbg =
get_field_gen_given_memory_chunk Word_int mut ptr n dbg
let set_field ptr n newval init dbg =
Cop (Cstore (Word_val, init), [field_address ptr n dbg; newval], dbg)
let get_header ptr dbg =
(* Headers can be mutated when forcing a lazy value. However, for all purposes
that the mutability tag currently serves in the compiler, header loads can
be marked as [Immutable], since the runtime should ensure that there is no
data race on headers. This saves performance with ThreadSanitizer
instrumentation by avoiding to instrument header loads. *)
Cop
( mk_load_mut Word_int,
(* CR xclerc: consider whether that could be changed to mk_load_immut *)
[Cop (Cadda, [ptr; Cconst_int (-size_int, dbg)], dbg)],
dbg )
let get_header_masked ptr dbg =
if Config.reserved_header_bits > 0
then
let header_mask = (1 lsl (64 - Config.reserved_header_bits)) - 1 in
Cop (Cand, [get_header ptr dbg; Cconst_int (header_mask, dbg)], dbg)
else get_header ptr dbg
let tag_offset = if big_endian then -1 else -size_int
let get_tag ptr dbg =
if Proc.word_addressed
then
(* If byte loads are slow *)
Cop (Cand, [get_header ptr dbg; Cconst_int (255, dbg)], dbg)
else
(* If byte loads are efficient *)
(* Same comment as [get_header] above *)
Cop
( (if Config.runtime5
then mk_load_immut Byte_unsigned
else mk_load_mut Byte_unsigned),
[Cop (Cadda, [ptr; Cconst_int (tag_offset, dbg)], dbg)],
dbg )
let get_size ptr dbg =
Cop (Clsr, [get_header_masked ptr dbg; Cconst_int (10, dbg)], dbg)
(* Array indexing *)
let log2_size_addr = Misc.log2 size_addr
let log2_size_float = Misc.log2 size_float
let wordsize_shift = 9
let numfloat_shift = 9 + log2_size_float - log2_size_addr
let is_addr_array_hdr hdr dbg =
Cop
( Ccmpi Cne,
[Cop (Cand, [hdr; Cconst_int (255, dbg)], dbg); floatarray_tag dbg],
dbg )
let addr_array_length_shifted hdr dbg =
Cop (Clsr, [hdr; Cconst_int (wordsize_shift, dbg)], dbg)
let float_array_length_shifted hdr dbg =
Cop (Clsr, [hdr; Cconst_int (numfloat_shift, dbg)], dbg)
let lsl_const c n dbg =
if n = 0 then c else Cop (Clsl, [c; Cconst_int (n, dbg)], dbg)
(* Produces a pointer to the element of the array [ptr] on the position [ofs]
with the given element [log2size] log2 element size.
[ofs] is given as a tagged int expression.
The optional ?typ argument is the C-- type of the result. By default, it is
Addr, meaning we are constructing a derived pointer into the heap. If we know
the pointer is outside the heap (this is the case for bigarray indexing), we
give type Int instead. *)
let array_indexing ?typ log2size ptr ofs dbg =
let add =
match typ with
| None | Some Addr -> Cadda
| Some Int -> Caddi
| _ -> assert false
in
match ofs with
| Cconst_int (n, _) ->
let i = n asr 1 in
if i = 0
then ptr
else Cop (add, [ptr; Cconst_int (i lsl log2size, dbg)], dbg)
| Cop (Caddi, [Cop (Clsl, [c; Cconst_int (1, _)], _); Cconst_int (1, _)], dbg')
->
Cop (add, [ptr; lsl_const c log2size dbg], dbg')
| Cop (Caddi, [c; Cconst_int (n, _)], dbg') when log2size = 0 ->
Cop
( add,
[Cop (add, [ptr; untag_int c dbg], dbg); Cconst_int (n asr 1, dbg)],
dbg' )
| Cop (Caddi, [c; Cconst_int (n, _)], _) ->
Cop
( add,
[ Cop (add, [ptr; lsl_const c (log2size - 1) dbg], dbg);
Cconst_int ((n - 1) lsl (log2size - 1), dbg) ],
dbg )
| _ when log2size = 0 -> Cop (add, [ptr; untag_int ofs dbg], dbg)
| _ ->
Cop
( add,
[ Cop (add, [ptr; lsl_const ofs (log2size - 1) dbg], dbg);
Cconst_int (-1 lsl (log2size - 1), dbg) ],