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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
|
-- -----------------------------------------------------------------------------
--
-- (c) The University of Glasgow 1993-2004
--
-- This is the top-level module in the native code generator.
--
-- -----------------------------------------------------------------------------
{-# LANGUAGE BangPatterns, CPP, GADTs, ScopedTypeVariables, PatternSynonyms,
DeriveFunctor #-}
#if !defined(GHC_LOADED_INTO_GHCI)
{-# LANGUAGE UnboxedTuples #-}
#endif
{-# OPTIONS_GHC -Wno-incomplete-record-updates #-}
module GHC.CmmToAsm (
-- * Module entry point
nativeCodeGen
-- * Test-only exports: see trac #12744
-- used by testGraphNoSpills, which needs to access
-- the register allocator intermediate data structures
-- cmmNativeGen emits
, cmmNativeGen
, NcgImpl(..)
, x86NcgImpl
) where
#include "HsVersions.h"
import GhcPrelude
import qualified GHC.CmmToAsm.X86.CodeGen as X86.CodeGen
import qualified GHC.CmmToAsm.X86.Regs as X86.Regs
import qualified GHC.CmmToAsm.X86.Instr as X86.Instr
import qualified GHC.CmmToAsm.X86.Ppr as X86.Ppr
import qualified GHC.CmmToAsm.SPARC.CodeGen as SPARC.CodeGen
import qualified GHC.CmmToAsm.SPARC.Regs as SPARC.Regs
import qualified GHC.CmmToAsm.SPARC.Instr as SPARC.Instr
import qualified GHC.CmmToAsm.SPARC.Ppr as SPARC.Ppr
import qualified GHC.CmmToAsm.SPARC.ShortcutJump as SPARC.ShortcutJump
import qualified GHC.CmmToAsm.SPARC.CodeGen.Expand as SPARC.CodeGen.Expand
import qualified GHC.CmmToAsm.PPC.CodeGen as PPC.CodeGen
import qualified GHC.CmmToAsm.PPC.Regs as PPC.Regs
import qualified GHC.CmmToAsm.PPC.RegInfo as PPC.RegInfo
import qualified GHC.CmmToAsm.PPC.Instr as PPC.Instr
import qualified GHC.CmmToAsm.PPC.Ppr as PPC.Ppr
import GHC.CmmToAsm.Reg.Liveness
import qualified GHC.CmmToAsm.Reg.Linear as Linear
import qualified GraphColor as Color
import qualified GHC.CmmToAsm.Reg.Graph as Color
import qualified GHC.CmmToAsm.Reg.Graph.Stats as Color
import qualified GHC.CmmToAsm.Reg.Graph.TrivColorable as Color
import AsmUtils
import GHC.CmmToAsm.Reg.Target
import GHC.Platform
import GHC.CmmToAsm.BlockLayout as BlockLayout
import Config
import GHC.CmmToAsm.Instr
import GHC.CmmToAsm.PIC
import GHC.Platform.Reg
import GHC.CmmToAsm.Monad
import GHC.CmmToAsm.CFG
import GHC.CmmToAsm.Dwarf
import GHC.CmmToAsm.Config
import GHC.Cmm.DebugBlock
import GHC.Cmm.BlockId
import GHC.StgToCmm.CgUtils ( fixStgRegisters )
import GHC.Cmm
import GHC.Cmm.Utils
import GHC.Cmm.Dataflow.Collections
import GHC.Cmm.Dataflow.Label
import GHC.Cmm.Dataflow.Block
import GHC.Cmm.Opt ( cmmMachOpFold )
import GHC.Cmm.Ppr
import GHC.Cmm.CLabel
import GHC.Types.Unique.FM
import GHC.Types.Unique.Supply
import GHC.Driver.Session
import Util
import GHC.Types.Basic ( Alignment )
import qualified Pretty
import BufWrite
import Outputable
import FastString
import GHC.Types.Unique.Set
import ErrUtils
import GHC.Types.Module
import Stream (Stream)
import qualified Stream
-- DEBUGGING ONLY
--import OrdList
import Data.List
import Data.Maybe
import Data.Ord ( comparing )
import Control.Exception
import Control.Monad
import System.IO
{-
The native-code generator has machine-independent and
machine-dependent modules.
This module ("AsmCodeGen") is the top-level machine-independent
module. Before entering machine-dependent land, we do some
machine-independent optimisations (defined below) on the
'CmmStmts's.
We convert to the machine-specific 'Instr' datatype with
'cmmCodeGen', assuming an infinite supply of registers. We then use
a machine-independent register allocator ('regAlloc') to rejoin
reality. Obviously, 'regAlloc' has machine-specific helper
functions (see about "RegAllocInfo" below).
Finally, we order the basic blocks of the function so as to minimise
the number of jumps between blocks, by utilising fallthrough wherever
possible.
The machine-dependent bits break down as follows:
* ["MachRegs"] Everything about the target platform's machine
registers (and immediate operands, and addresses, which tend to
intermingle/interact with registers).
* ["MachInstrs"] Includes the 'Instr' datatype (possibly should
have a module of its own), plus a miscellany of other things
(e.g., 'targetDoubleSize', 'smStablePtrTable', ...)
* ["MachCodeGen"] is where 'Cmm' stuff turns into
machine instructions.
* ["PprMach"] 'pprInstr' turns an 'Instr' into text (well, really
a 'SDoc').
* ["RegAllocInfo"] In the register allocator, we manipulate
'MRegsState's, which are 'BitSet's, one bit per machine register.
When we want to say something about a specific machine register
(e.g., ``it gets clobbered by this instruction''), we set/unset
its bit. Obviously, we do this 'BitSet' thing for efficiency
reasons.
The 'RegAllocInfo' module collects together the machine-specific
info needed to do register allocation.
* ["RegisterAlloc"] The (machine-independent) register allocator.
-}
--------------------
nativeCodeGen :: forall a . DynFlags -> Module -> ModLocation -> Handle -> UniqSupply
-> Stream IO RawCmmGroup a
-> IO a
nativeCodeGen dflags this_mod modLoc h us cmms
= let platform = targetPlatform dflags
nCG' :: ( Outputable statics, Outputable instr
, Outputable jumpDest, Instruction instr)
=> NcgImpl statics instr jumpDest -> IO a
nCG' ncgImpl = nativeCodeGen' dflags this_mod modLoc ncgImpl h us cmms
in case platformArch platform of
ArchX86 -> nCG' (x86NcgImpl dflags)
ArchX86_64 -> nCG' (x86_64NcgImpl dflags)
ArchPPC -> nCG' (ppcNcgImpl dflags)
ArchS390X -> panic "nativeCodeGen: No NCG for S390X"
ArchSPARC -> nCG' (sparcNcgImpl dflags)
ArchSPARC64 -> panic "nativeCodeGen: No NCG for SPARC64"
ArchARM {} -> panic "nativeCodeGen: No NCG for ARM"
ArchARM64 -> panic "nativeCodeGen: No NCG for ARM64"
ArchPPC_64 _ -> nCG' (ppcNcgImpl dflags)
ArchAlpha -> panic "nativeCodeGen: No NCG for Alpha"
ArchMipseb -> panic "nativeCodeGen: No NCG for mipseb"
ArchMipsel -> panic "nativeCodeGen: No NCG for mipsel"
ArchUnknown -> panic "nativeCodeGen: No NCG for unknown arch"
ArchJavaScript-> panic "nativeCodeGen: No NCG for JavaScript"
x86NcgImpl :: DynFlags -> NcgImpl (Alignment, RawCmmStatics)
X86.Instr.Instr X86.Instr.JumpDest
x86NcgImpl dflags
= (x86_64NcgImpl dflags)
x86_64NcgImpl :: DynFlags -> NcgImpl (Alignment, RawCmmStatics)
X86.Instr.Instr X86.Instr.JumpDest
x86_64NcgImpl dflags
= NcgImpl {
ncgConfig = config
,cmmTopCodeGen = X86.CodeGen.cmmTopCodeGen
,generateJumpTableForInstr = X86.CodeGen.generateJumpTableForInstr config
,getJumpDestBlockId = X86.Instr.getJumpDestBlockId
,canShortcut = X86.Instr.canShortcut
,shortcutStatics = X86.Instr.shortcutStatics
,shortcutJump = X86.Instr.shortcutJump
,pprNatCmmDecl = X86.Ppr.pprNatCmmDecl config
,maxSpillSlots = X86.Instr.maxSpillSlots config
,allocatableRegs = X86.Regs.allocatableRegs platform
,ncgAllocMoreStack = X86.Instr.allocMoreStack platform
,ncgExpandTop = id
,ncgMakeFarBranches = const id
,extractUnwindPoints = X86.CodeGen.extractUnwindPoints
,invertCondBranches = X86.CodeGen.invertCondBranches
}
where
config = initConfig dflags
platform = ncgPlatform config
ppcNcgImpl :: DynFlags -> NcgImpl RawCmmStatics PPC.Instr.Instr PPC.RegInfo.JumpDest
ppcNcgImpl dflags
= NcgImpl {
ncgConfig = config
,cmmTopCodeGen = PPC.CodeGen.cmmTopCodeGen
,generateJumpTableForInstr = PPC.CodeGen.generateJumpTableForInstr config
,getJumpDestBlockId = PPC.RegInfo.getJumpDestBlockId
,canShortcut = PPC.RegInfo.canShortcut
,shortcutStatics = PPC.RegInfo.shortcutStatics
,shortcutJump = PPC.RegInfo.shortcutJump
,pprNatCmmDecl = PPC.Ppr.pprNatCmmDecl config
,maxSpillSlots = PPC.Instr.maxSpillSlots config
,allocatableRegs = PPC.Regs.allocatableRegs platform
,ncgAllocMoreStack = PPC.Instr.allocMoreStack platform
,ncgExpandTop = id
,ncgMakeFarBranches = PPC.Instr.makeFarBranches
,extractUnwindPoints = const []
,invertCondBranches = \_ _ -> id
}
where
config = initConfig dflags
platform = ncgPlatform config
sparcNcgImpl :: DynFlags -> NcgImpl RawCmmStatics SPARC.Instr.Instr SPARC.ShortcutJump.JumpDest
sparcNcgImpl dflags
= NcgImpl {
ncgConfig = config
,cmmTopCodeGen = SPARC.CodeGen.cmmTopCodeGen
,generateJumpTableForInstr = SPARC.CodeGen.generateJumpTableForInstr platform
,getJumpDestBlockId = SPARC.ShortcutJump.getJumpDestBlockId
,canShortcut = SPARC.ShortcutJump.canShortcut
,shortcutStatics = SPARC.ShortcutJump.shortcutStatics
,shortcutJump = SPARC.ShortcutJump.shortcutJump
,pprNatCmmDecl = SPARC.Ppr.pprNatCmmDecl config
,maxSpillSlots = SPARC.Instr.maxSpillSlots config
,allocatableRegs = SPARC.Regs.allocatableRegs
,ncgAllocMoreStack = noAllocMoreStack
,ncgExpandTop = map SPARC.CodeGen.Expand.expandTop
,ncgMakeFarBranches = const id
,extractUnwindPoints = const []
,invertCondBranches = \_ _ -> id
}
where
config = initConfig dflags
platform = ncgPlatform config
--
-- Allocating more stack space for spilling is currently only
-- supported for the linear register allocator on x86/x86_64, the rest
-- default to the panic below. To support allocating extra stack on
-- more platforms provide a definition of ncgAllocMoreStack.
--
noAllocMoreStack :: Int -> NatCmmDecl statics instr
-> UniqSM (NatCmmDecl statics instr, [(BlockId,BlockId)])
noAllocMoreStack amount _
= panic $ "Register allocator: out of stack slots (need " ++ show amount ++ ")\n"
++ " If you are trying to compile SHA1.hs from the crypto library then this\n"
++ " is a known limitation in the linear allocator.\n"
++ "\n"
++ " Try enabling the graph colouring allocator with -fregs-graph instead."
++ " You can still file a bug report if you like.\n"
-- | Data accumulated during code generation. Mostly about statistics,
-- but also collects debug data for DWARF generation.
data NativeGenAcc statics instr
= NGS { ngs_imports :: ![[CLabel]]
, ngs_natives :: ![[NatCmmDecl statics instr]]
-- ^ Native code generated, for statistics. This might
-- hold a lot of data, so it is important to clear this
-- field as early as possible if it isn't actually
-- required.
, ngs_colorStats :: ![[Color.RegAllocStats statics instr]]
, ngs_linearStats :: ![[Linear.RegAllocStats]]
, ngs_labels :: ![Label]
, ngs_debug :: ![DebugBlock]
, ngs_dwarfFiles :: !DwarfFiles
, ngs_unwinds :: !(LabelMap [UnwindPoint])
-- ^ see Note [Unwinding information in the NCG]
-- and Note [What is this unwinding business?] in Debug.
}
{-
Note [Unwinding information in the NCG]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Unwind information is a type of metadata which allows a debugging tool
to reconstruct the values of machine registers at the time a procedure was
entered. For the most part, the production of unwind information is handled by
the Cmm stage, where it is represented by CmmUnwind nodes.
Unfortunately, the Cmm stage doesn't know everything necessary to produce
accurate unwinding information. For instance, the x86-64 calling convention
requires that the stack pointer be aligned to 16 bytes, which in turn means that
GHC must sometimes add padding to $sp prior to performing a foreign call. When
this happens unwind information must be updated accordingly.
For this reason, we make the NCG backends responsible for producing
unwinding tables (with the extractUnwindPoints function in NcgImpl).
We accumulate the produced unwind tables over CmmGroups in the ngs_unwinds
field of NativeGenAcc. This is a label map which contains an entry for each
procedure, containing a list of unwinding points (e.g. a label and an associated
unwinding table).
See also Note [What is this unwinding business?] in Debug.
-}
nativeCodeGen' :: (Outputable statics, Outputable instr,Outputable jumpDest,
Instruction instr)
=> DynFlags
-> Module -> ModLocation
-> NcgImpl statics instr jumpDest
-> Handle
-> UniqSupply
-> Stream IO RawCmmGroup a
-> IO a
nativeCodeGen' dflags this_mod modLoc ncgImpl h us cmms
= do
-- BufHandle is a performance hack. We could hide it inside
-- Pretty if it weren't for the fact that we do lots of little
-- printDocs here (in order to do codegen in constant space).
bufh <- newBufHandle h
let ngs0 = NGS [] [] [] [] [] [] emptyUFM mapEmpty
(ngs, us', a) <- cmmNativeGenStream dflags this_mod modLoc ncgImpl bufh us
cmms ngs0
_ <- finishNativeGen dflags modLoc bufh us' ngs
return a
finishNativeGen :: Instruction instr
=> DynFlags
-> ModLocation
-> BufHandle
-> UniqSupply
-> NativeGenAcc statics instr
-> IO UniqSupply
finishNativeGen dflags modLoc bufh@(BufHandle _ _ h) us ngs
= withTimingSilent dflags (text "NCG") (`seq` ()) $ do
-- Write debug data and finish
let emitDw = debugLevel dflags > 0
us' <- if not emitDw then return us else do
(dwarf, us') <- dwarfGen dflags modLoc us (ngs_debug ngs)
emitNativeCode dflags bufh dwarf
return us'
bFlush bufh
-- dump global NCG stats for graph coloring allocator
let stats = concat (ngs_colorStats ngs)
unless (null stats) $ do
-- build the global register conflict graph
let graphGlobal
= foldl' Color.union Color.initGraph
$ [ Color.raGraph stat
| stat@Color.RegAllocStatsStart{} <- stats]
dump_stats (Color.pprStats stats graphGlobal)
let platform = targetPlatform dflags
dumpIfSet_dyn dflags
Opt_D_dump_asm_conflicts "Register conflict graph"
FormatText
$ Color.dotGraph
(targetRegDotColor platform)
(Color.trivColorable platform
(targetVirtualRegSqueeze platform)
(targetRealRegSqueeze platform))
$ graphGlobal
-- dump global NCG stats for linear allocator
let linearStats = concat (ngs_linearStats ngs)
unless (null linearStats) $
dump_stats (Linear.pprStats (concat (ngs_natives ngs)) linearStats)
-- write out the imports
printSDocLn Pretty.LeftMode dflags h (mkCodeStyle AsmStyle)
$ makeImportsDoc dflags (concat (ngs_imports ngs))
return us'
where
dump_stats = dumpAction dflags (mkDumpStyle dflags alwaysQualify)
(dumpOptionsFromFlag Opt_D_dump_asm_stats) "NCG stats"
FormatText
cmmNativeGenStream :: (Outputable statics, Outputable instr
,Outputable jumpDest, Instruction instr)
=> DynFlags
-> Module -> ModLocation
-> NcgImpl statics instr jumpDest
-> BufHandle
-> UniqSupply
-> Stream IO RawCmmGroup a
-> NativeGenAcc statics instr
-> IO (NativeGenAcc statics instr, UniqSupply, a)
cmmNativeGenStream dflags this_mod modLoc ncgImpl h us cmm_stream ngs
= do r <- Stream.runStream cmm_stream
case r of
Left a ->
return (ngs { ngs_imports = reverse $ ngs_imports ngs
, ngs_natives = reverse $ ngs_natives ngs
, ngs_colorStats = reverse $ ngs_colorStats ngs
, ngs_linearStats = reverse $ ngs_linearStats ngs
},
us,
a)
Right (cmms, cmm_stream') -> do
(us', ngs'') <-
withTimingSilent
dflags
ncglabel (\(a, b) -> a `seq` b `seq` ()) $ do
-- Generate debug information
let debugFlag = debugLevel dflags > 0
!ndbgs | debugFlag = cmmDebugGen modLoc cmms
| otherwise = []
dbgMap = debugToMap ndbgs
-- Generate native code
(ngs',us') <- cmmNativeGens dflags this_mod modLoc ncgImpl h
dbgMap us cmms ngs 0
-- Link native code information into debug blocks
-- See Note [What is this unwinding business?] in Debug.
let !ldbgs = cmmDebugLink (ngs_labels ngs') (ngs_unwinds ngs') ndbgs
unless (null ldbgs) $
dumpIfSet_dyn dflags Opt_D_dump_debug "Debug Infos" FormatText
(vcat $ map ppr ldbgs)
-- Accumulate debug information for emission in finishNativeGen.
let ngs'' = ngs' { ngs_debug = ngs_debug ngs' ++ ldbgs, ngs_labels = [] }
return (us', ngs'')
cmmNativeGenStream dflags this_mod modLoc ncgImpl h us'
cmm_stream' ngs''
where ncglabel = text "NCG"
-- | Do native code generation on all these cmms.
--
cmmNativeGens :: forall statics instr jumpDest.
(Outputable statics, Outputable instr
,Outputable jumpDest, Instruction instr)
=> DynFlags
-> Module -> ModLocation
-> NcgImpl statics instr jumpDest
-> BufHandle
-> LabelMap DebugBlock
-> UniqSupply
-> [RawCmmDecl]
-> NativeGenAcc statics instr
-> Int
-> IO (NativeGenAcc statics instr, UniqSupply)
cmmNativeGens dflags this_mod modLoc ncgImpl h dbgMap = go
where
go :: UniqSupply -> [RawCmmDecl]
-> NativeGenAcc statics instr -> Int
-> IO (NativeGenAcc statics instr, UniqSupply)
go us [] ngs !_ =
return (ngs, us)
go us (cmm : cmms) ngs count = do
let fileIds = ngs_dwarfFiles ngs
(us', fileIds', native, imports, colorStats, linearStats, unwinds)
<- {-# SCC "cmmNativeGen" #-}
cmmNativeGen dflags this_mod modLoc ncgImpl us fileIds dbgMap
cmm count
-- Generate .file directives for every new file that has been
-- used. Note that it is important that we generate these in
-- ascending order, as Clang's 3.6 assembler complains.
let newFileIds = sortBy (comparing snd) $
nonDetEltsUFM $ fileIds' `minusUFM` fileIds
-- See Note [Unique Determinism and code generation]
pprDecl (f,n) = text "\t.file " <> ppr n <+>
pprFilePathString (unpackFS f)
emitNativeCode dflags h $ vcat $
map pprDecl newFileIds ++
map (pprNatCmmDecl ncgImpl) native
-- force evaluation all this stuff to avoid space leaks
{-# SCC "seqString" #-} evaluate $ seqList (showSDoc dflags $ vcat $ map ppr imports) ()
let !labels' = if debugLevel dflags > 0
then cmmDebugLabels isMetaInstr native else []
!natives' = if dopt Opt_D_dump_asm_stats dflags
then native : ngs_natives ngs else []
mCon = maybe id (:)
ngs' = ngs{ ngs_imports = imports : ngs_imports ngs
, ngs_natives = natives'
, ngs_colorStats = colorStats `mCon` ngs_colorStats ngs
, ngs_linearStats = linearStats `mCon` ngs_linearStats ngs
, ngs_labels = ngs_labels ngs ++ labels'
, ngs_dwarfFiles = fileIds'
, ngs_unwinds = ngs_unwinds ngs `mapUnion` unwinds
}
go us' cmms ngs' (count + 1)
emitNativeCode :: DynFlags -> BufHandle -> SDoc -> IO ()
emitNativeCode dflags h sdoc = do
{-# SCC "pprNativeCode" #-} bufLeftRenderSDoc dflags h
(mkCodeStyle AsmStyle) sdoc
-- dump native code
dumpIfSet_dyn dflags
Opt_D_dump_asm "Asm code" FormatASM
sdoc
-- | Complete native code generation phase for a single top-level chunk of Cmm.
-- Dumping the output of each stage along the way.
-- Global conflict graph and NGC stats
cmmNativeGen
:: forall statics instr jumpDest. (Instruction instr,
Outputable statics, Outputable instr, Outputable jumpDest)
=> DynFlags
-> Module -> ModLocation
-> NcgImpl statics instr jumpDest
-> UniqSupply
-> DwarfFiles
-> LabelMap DebugBlock
-> RawCmmDecl -- ^ the cmm to generate code for
-> Int -- ^ sequence number of this top thing
-> IO ( UniqSupply
, DwarfFiles
, [NatCmmDecl statics instr] -- native code
, [CLabel] -- things imported by this cmm
, Maybe [Color.RegAllocStats statics instr] -- stats for the coloring register allocator
, Maybe [Linear.RegAllocStats] -- stats for the linear register allocators
, LabelMap [UnwindPoint] -- unwinding information for blocks
)
cmmNativeGen dflags this_mod modLoc ncgImpl us fileIds dbgMap cmm count
= do
let config = ncgConfig ncgImpl
let platform = ncgPlatform config
let proc_name = case cmm of
(CmmProc _ entry_label _ _) -> ppr entry_label
_ -> text "DataChunk"
-- rewrite assignments to global regs
let fixed_cmm =
{-# SCC "fixStgRegisters" #-}
fixStgRegisters dflags cmm
-- cmm to cmm optimisations
let (opt_cmm, imports) =
{-# SCC "cmmToCmm" #-}
cmmToCmm dflags this_mod fixed_cmm
dumpIfSet_dyn dflags
Opt_D_dump_opt_cmm "Optimised Cmm" FormatCMM
(pprCmmGroup [opt_cmm])
let cmmCfg = {-# SCC "getCFG" #-}
getCfgProc (cfgWeightInfo dflags) opt_cmm
-- generate native code from cmm
let ((native, lastMinuteImports, fileIds', nativeCfgWeights), usGen) =
{-# SCC "genMachCode" #-}
initUs us $ genMachCode dflags this_mod modLoc
(cmmTopCodeGen ncgImpl)
fileIds dbgMap opt_cmm cmmCfg
dumpIfSet_dyn dflags
Opt_D_dump_asm_native "Native code" FormatASM
(vcat $ map (pprNatCmmDecl ncgImpl) native)
maybeDumpCfg dflags (Just nativeCfgWeights) "CFG Weights - Native" proc_name
-- tag instructions with register liveness information
-- also drops dead code. We don't keep the cfg in sync on
-- some backends, so don't use it there.
let livenessCfg = if backendMaintainsCfg platform
then Just nativeCfgWeights
else Nothing
let (withLiveness, usLive) =
{-# SCC "regLiveness" #-}
initUs usGen
$ mapM (cmmTopLiveness livenessCfg platform) native
dumpIfSet_dyn dflags
Opt_D_dump_asm_liveness "Liveness annotations added"
FormatCMM
(vcat $ map ppr withLiveness)
-- allocate registers
(alloced, usAlloc, ppr_raStatsColor, ppr_raStatsLinear, raStats, stack_updt_blks) <-
if ( gopt Opt_RegsGraph dflags
|| gopt Opt_RegsIterative dflags )
then do
-- the regs usable for allocation
let (alloc_regs :: UniqFM (UniqSet RealReg))
= foldr (\r -> plusUFM_C unionUniqSets
$ unitUFM (targetClassOfRealReg platform r) (unitUniqSet r))
emptyUFM
$ allocatableRegs ncgImpl
-- do the graph coloring register allocation
let ((alloced, maybe_more_stack, regAllocStats), usAlloc)
= {-# SCC "RegAlloc-color" #-}
initUs usLive
$ Color.regAlloc
config
alloc_regs
(mkUniqSet [0 .. maxSpillSlots ncgImpl])
(maxSpillSlots ncgImpl)
withLiveness
livenessCfg
let ((alloced', stack_updt_blks), usAlloc')
= initUs usAlloc $
case maybe_more_stack of
Nothing -> return (alloced, [])
Just amount -> do
(alloced',stack_updt_blks) <- unzip <$>
(mapM ((ncgAllocMoreStack ncgImpl) amount) alloced)
return (alloced', concat stack_updt_blks )
-- dump out what happened during register allocation
dumpIfSet_dyn dflags
Opt_D_dump_asm_regalloc "Registers allocated"
FormatCMM
(vcat $ map (pprNatCmmDecl ncgImpl) alloced)
dumpIfSet_dyn dflags
Opt_D_dump_asm_regalloc_stages "Build/spill stages"
FormatText
(vcat $ map (\(stage, stats)
-> text "# --------------------------"
$$ text "# cmm " <> int count <> text " Stage " <> int stage
$$ ppr stats)
$ zip [0..] regAllocStats)
let mPprStats =
if dopt Opt_D_dump_asm_stats dflags
then Just regAllocStats else Nothing
-- force evaluation of the Maybe to avoid space leak
mPprStats `seq` return ()
return ( alloced', usAlloc'
, mPprStats
, Nothing
, [], stack_updt_blks)
else do
-- do linear register allocation
let reg_alloc proc = do
(alloced, maybe_more_stack, ra_stats) <-
Linear.regAlloc config proc
case maybe_more_stack of
Nothing -> return ( alloced, ra_stats, [] )
Just amount -> do
(alloced',stack_updt_blks) <-
ncgAllocMoreStack ncgImpl amount alloced
return (alloced', ra_stats, stack_updt_blks )
let ((alloced, regAllocStats, stack_updt_blks), usAlloc)
= {-# SCC "RegAlloc-linear" #-}
initUs usLive
$ liftM unzip3
$ mapM reg_alloc withLiveness
dumpIfSet_dyn dflags
Opt_D_dump_asm_regalloc "Registers allocated"
FormatCMM
(vcat $ map (pprNatCmmDecl ncgImpl) alloced)
let mPprStats =
if dopt Opt_D_dump_asm_stats dflags
then Just (catMaybes regAllocStats) else Nothing
-- force evaluation of the Maybe to avoid space leak
mPprStats `seq` return ()
return ( alloced, usAlloc
, Nothing
, mPprStats, (catMaybes regAllocStats)
, concat stack_updt_blks )
-- Fixupblocks the register allocator inserted (from, regMoves, to)
let cfgRegAllocUpdates :: [(BlockId,BlockId,BlockId)]
cfgRegAllocUpdates = (concatMap Linear.ra_fixupList raStats)
let cfgWithFixupBlks =
(\cfg -> addNodesBetween dflags cfg cfgRegAllocUpdates) <$> livenessCfg
-- Insert stack update blocks
let postRegCFG =
pure (foldl' (\m (from,to) -> addImmediateSuccessor dflags from to m ))
<*> cfgWithFixupBlks
<*> pure stack_updt_blks
---- generate jump tables
let tabled =
{-# SCC "generateJumpTables" #-}
generateJumpTables ncgImpl alloced
when (not $ null nativeCfgWeights) $ dumpIfSet_dyn dflags
Opt_D_dump_cfg_weights "CFG Update information"
FormatText
( text "stack:" <+> ppr stack_updt_blks $$
text "linearAlloc:" <+> ppr cfgRegAllocUpdates )
---- shortcut branches
let (shorted, postShortCFG) =
{-# SCC "shortcutBranches" #-}
shortcutBranches dflags ncgImpl tabled postRegCFG
let optimizedCFG :: Maybe CFG
optimizedCFG =
optimizeCFG (cfgWeightInfo dflags) cmm <$!> postShortCFG
maybeDumpCfg dflags optimizedCFG "CFG Weights - Final" proc_name
--TODO: Partially check validity of the cfg.
let getBlks (CmmProc _info _lbl _live (ListGraph blocks)) = blocks
getBlks _ = []
when ( backendMaintainsCfg platform &&
(gopt Opt_DoAsmLinting dflags || debugIsOn )) $ do
let blocks = concatMap getBlks shorted
let labels = setFromList $ fmap blockId blocks :: LabelSet
let cfg = fromJust optimizedCFG
return $! seq (sanityCheckCfg cfg labels $
text "cfg not in lockstep") ()
---- sequence blocks
let sequenced :: [NatCmmDecl statics instr]
sequenced =
checkLayout shorted $
{-# SCC "sequenceBlocks" #-}
map (BlockLayout.sequenceTop
dflags
ncgImpl optimizedCFG)
shorted
let branchOpt :: [NatCmmDecl statics instr]
branchOpt =
{-# SCC "invertCondBranches" #-}
map invert sequenced
where
invertConds :: LabelMap RawCmmStatics -> [NatBasicBlock instr]
-> [NatBasicBlock instr]
invertConds = invertCondBranches ncgImpl optimizedCFG
invert top@CmmData {} = top
invert (CmmProc info lbl live (ListGraph blocks)) =
CmmProc info lbl live (ListGraph $ invertConds info blocks)
---- expansion of SPARC synthetic instrs
let expanded =
{-# SCC "sparc_expand" #-}
ncgExpandTop ncgImpl branchOpt
--ncgExpandTop ncgImpl sequenced
dumpIfSet_dyn dflags
Opt_D_dump_asm_expanded "Synthetic instructions expanded"
FormatCMM
(vcat $ map (pprNatCmmDecl ncgImpl) expanded)
-- generate unwinding information from cmm
let unwinds :: BlockMap [UnwindPoint]
unwinds =
{-# SCC "unwindingInfo" #-}
foldl' addUnwind mapEmpty expanded
where
addUnwind acc proc =
acc `mapUnion` computeUnwinding dflags ncgImpl proc
return ( usAlloc
, fileIds'
, expanded
, lastMinuteImports ++ imports
, ppr_raStatsColor
, ppr_raStatsLinear
, unwinds )
maybeDumpCfg :: DynFlags -> Maybe CFG -> String -> SDoc -> IO ()
maybeDumpCfg _dflags Nothing _ _ = return ()
maybeDumpCfg dflags (Just cfg) msg proc_name
| null cfg = return ()
| otherwise
= dumpIfSet_dyn
dflags Opt_D_dump_cfg_weights msg
FormatText
(proc_name <> char ':' $$ pprEdgeWeights cfg)
-- | Make sure all blocks we want the layout algorithm to place have been placed.
checkLayout :: [NatCmmDecl statics instr] -> [NatCmmDecl statics instr]
-> [NatCmmDecl statics instr]
checkLayout procsUnsequenced procsSequenced =
ASSERT2(setNull diff,
ppr "Block sequencing dropped blocks:" <> ppr diff)
procsSequenced
where
blocks1 = foldl' (setUnion) setEmpty $
map getBlockIds procsUnsequenced :: LabelSet
blocks2 = foldl' (setUnion) setEmpty $
map getBlockIds procsSequenced
diff = setDifference blocks1 blocks2
getBlockIds (CmmData _ _) = setEmpty
getBlockIds (CmmProc _ _ _ (ListGraph blocks)) =
setFromList $ map blockId blocks
-- | Compute unwinding tables for the blocks of a procedure
computeUnwinding :: Instruction instr
=> DynFlags -> NcgImpl statics instr jumpDest
-> NatCmmDecl statics instr
-- ^ the native code generated for the procedure
-> LabelMap [UnwindPoint]
-- ^ unwinding tables for all points of all blocks of the
-- procedure
computeUnwinding dflags _ _
| debugLevel dflags == 0 = mapEmpty
computeUnwinding _ _ (CmmData _ _) = mapEmpty
computeUnwinding _ ncgImpl (CmmProc _ _ _ (ListGraph blks)) =
-- In general we would need to push unwinding information down the
-- block-level call-graph to ensure that we fully account for all
-- relevant register writes within a procedure.
--
-- However, the only unwinding information that we care about in GHC is for
-- Sp. The fact that GHC.Cmm.LayoutStack already ensures that we have unwind
-- information at the beginning of every block means that there is no need
-- to perform this sort of push-down.
mapFromList [ (blk_lbl, extractUnwindPoints ncgImpl instrs)
| BasicBlock blk_lbl instrs <- blks ]
-- | Build a doc for all the imports.
--
makeImportsDoc :: DynFlags -> [CLabel] -> SDoc
makeImportsDoc dflags imports
= dyld_stubs imports
$$
-- On recent versions of Darwin, the linker supports
-- dead-stripping of code and data on a per-symbol basis.
-- There's a hack to make this work in PprMach.pprNatCmmDecl.
(if platformHasSubsectionsViaSymbols platform
then text ".subsections_via_symbols"
else Outputable.empty)
$$
-- On recent GNU ELF systems one can mark an object file
-- as not requiring an executable stack. If all objects
-- linked into a program have this note then the program
-- will not use an executable stack, which is good for
-- security. GHC generated code does not need an executable
-- stack so add the note in:
(if platformHasGnuNonexecStack platform
then text ".section .note.GNU-stack,\"\"," <> sectionType platform "progbits"
else Outputable.empty)
$$
-- And just because every other compiler does, let's stick in
-- an identifier directive: .ident "GHC x.y.z"
(if platformHasIdentDirective platform
then let compilerIdent = text "GHC" <+> text cProjectVersion
in text ".ident" <+> doubleQuotes compilerIdent
else Outputable.empty)
where
config = initConfig dflags
platform = ncgPlatform config
-- Generate "symbol stubs" for all external symbols that might
-- come from a dynamic library.
dyld_stubs :: [CLabel] -> SDoc
{- dyld_stubs imps = vcat $ map pprDyldSymbolStub $
map head $ group $ sort imps-}
-- (Hack) sometimes two Labels pretty-print the same, but have
-- different uniques; so we compare their text versions...
dyld_stubs imps
| needImportedSymbols config
= vcat $
(pprGotDeclaration config :) $
map ( pprImportedSymbol dflags config . fst . head) $
groupBy (\(_,a) (_,b) -> a == b) $
sortBy (\(_,a) (_,b) -> compare a b) $
map doPpr $
imps
| otherwise
= Outputable.empty
doPpr lbl = (lbl, renderWithStyle
(initSDocContext dflags astyle)
(pprCLabel dflags lbl))
astyle = mkCodeStyle AsmStyle
-- -----------------------------------------------------------------------------
-- Generate jump tables
-- Analyzes all native code and generates data sections for all jump
-- table instructions.
generateJumpTables
:: NcgImpl statics instr jumpDest
-> [NatCmmDecl statics instr] -> [NatCmmDecl statics instr]
generateJumpTables ncgImpl xs = concatMap f xs
where f p@(CmmProc _ _ _ (ListGraph xs)) = p : concatMap g xs
f p = [p]
g (BasicBlock _ xs) = catMaybes (map (generateJumpTableForInstr ncgImpl) xs)
-- -----------------------------------------------------------------------------
-- Shortcut branches
shortcutBranches
:: forall statics instr jumpDest. (Outputable jumpDest) => DynFlags
-> NcgImpl statics instr jumpDest
-> [NatCmmDecl statics instr]
-> Maybe CFG
-> ([NatCmmDecl statics instr],Maybe CFG)
shortcutBranches dflags ncgImpl tops weights
| gopt Opt_AsmShortcutting dflags
= ( map (apply_mapping ncgImpl mapping) tops'
, shortcutWeightMap mappingBid <$!> weights )
| otherwise
= (tops, weights)
where
(tops', mappings) = mapAndUnzip (build_mapping ncgImpl) tops
mapping = mapUnions mappings :: LabelMap jumpDest
mappingBid = fmap (getJumpDestBlockId ncgImpl) mapping
build_mapping :: forall instr t d statics jumpDest.
NcgImpl statics instr jumpDest
-> GenCmmDecl d (LabelMap t) (ListGraph instr)
-> (GenCmmDecl d (LabelMap t) (ListGraph instr)
,LabelMap jumpDest)
build_mapping _ top@(CmmData _ _) = (top, mapEmpty)
build_mapping _ (CmmProc info lbl live (ListGraph []))
= (CmmProc info lbl live (ListGraph []), mapEmpty)
build_mapping ncgImpl (CmmProc info lbl live (ListGraph (head:blocks)))
= (CmmProc info lbl live (ListGraph (head:others)), mapping)
-- drop the shorted blocks, but don't ever drop the first one,
-- because it is pointed to by a global label.
where
-- find all the blocks that just consist of a jump that can be
-- shorted.
-- Don't completely eliminate loops here -- that can leave a dangling jump!
shortcut_blocks :: [(BlockId, jumpDest)]
(_, shortcut_blocks, others) =
foldl' split (setEmpty :: LabelSet, [], []) blocks
split (s, shortcut_blocks, others) b@(BasicBlock id [insn])
| Just jd <- canShortcut ncgImpl insn
, Just dest <- getJumpDestBlockId ncgImpl jd
, not (has_info id)
, (setMember dest s) || dest == id -- loop checks
= (s, shortcut_blocks, b : others)
split (s, shortcut_blocks, others) (BasicBlock id [insn])
| Just dest <- canShortcut ncgImpl insn
, not (has_info id)
= (setInsert id s, (id,dest) : shortcut_blocks, others)
split (s, shortcut_blocks, others) other = (s, shortcut_blocks, other : others)
-- do not eliminate blocks that have an info table
has_info l = mapMember l info
-- build a mapping from BlockId to JumpDest for shorting branches
mapping = mapFromList shortcut_blocks
apply_mapping :: NcgImpl statics instr jumpDest
-> LabelMap jumpDest
-> GenCmmDecl statics h (ListGraph instr)
-> GenCmmDecl statics h (ListGraph instr)
apply_mapping ncgImpl ufm (CmmData sec statics)
= CmmData sec (shortcutStatics ncgImpl (\bid -> mapLookup bid ufm) statics)
apply_mapping ncgImpl ufm (CmmProc info lbl live (ListGraph blocks))
= CmmProc info lbl live (ListGraph $ map short_bb blocks)
where
short_bb (BasicBlock id insns) = BasicBlock id $! map short_insn insns
short_insn i = shortcutJump ncgImpl (\bid -> mapLookup bid ufm) i
-- shortcutJump should apply the mapping repeatedly,
-- just in case we can short multiple branches.
-- -----------------------------------------------------------------------------
-- Instruction selection
-- Native code instruction selection for a chunk of stix code. For
-- this part of the computation, we switch from the UniqSM monad to
-- the NatM monad. The latter carries not only a Unique, but also an
-- Int denoting the current C stack pointer offset in the generated
-- code; this is needed for creating correct spill offsets on
-- architectures which don't offer, or for which it would be
-- prohibitively expensive to employ, a frame pointer register. Viz,
-- x86.
-- The offset is measured in bytes, and indicates the difference
-- between the current (simulated) C stack-ptr and the value it was at
-- the beginning of the block. For stacks which grow down, this value
-- should be either zero or negative.
-- Along with the stack pointer offset, we also carry along a LabelMap of
-- DebugBlocks, which we read to generate .location directives.
--
-- Switching between the two monads whilst carrying along the same
-- Unique supply breaks abstraction. Is that bad?
genMachCode
:: DynFlags
-> Module -> ModLocation
-> (RawCmmDecl -> NatM [NatCmmDecl statics instr])
-> DwarfFiles
-> LabelMap DebugBlock
-> RawCmmDecl
-> CFG
-> UniqSM
( [NatCmmDecl statics instr]
, [CLabel]
, DwarfFiles
, CFG
)
genMachCode dflags this_mod modLoc cmmTopCodeGen fileIds dbgMap cmm_top cmm_cfg
= do { initial_us <- getUniqueSupplyM
; let initial_st = mkNatM_State initial_us 0 dflags this_mod
modLoc fileIds dbgMap cmm_cfg
(new_tops, final_st) = initNat initial_st (cmmTopCodeGen cmm_top)
final_delta = natm_delta final_st
final_imports = natm_imports final_st
final_cfg = natm_cfg final_st
; if final_delta == 0
then return (new_tops, final_imports
, natm_fileid final_st, final_cfg)
else pprPanic "genMachCode: nonzero final delta" (int final_delta)
}
-- -----------------------------------------------------------------------------
-- Generic Cmm optimiser
{-
Here we do:
(a) Constant folding
(c) Position independent code and dynamic linking
(i) introduce the appropriate indirections
and position independent refs
(ii) compile a list of imported symbols
(d) Some arch-specific optimizations
(a) will be moving to the new Hoopl pipeline, however, (c) and
(d) are only needed by the native backend and will continue to live
here.
Ideas for other things we could do (put these in Hoopl please!):
- shortcut jumps-to-jumps
- simple CSE: if an expr is assigned to a temp, then replace later occs of
that expr with the temp, until the expr is no longer valid (can push through
temp assignments, and certain assigns to mem...)
-}
cmmToCmm :: DynFlags -> Module -> RawCmmDecl -> (RawCmmDecl, [CLabel])
cmmToCmm _ _ top@(CmmData _ _) = (top, [])
cmmToCmm dflags this_mod (CmmProc info lbl live graph)
= runCmmOpt dflags this_mod $
do blocks' <- mapM cmmBlockConFold (toBlockList graph)
return $ CmmProc info lbl live (ofBlockList (g_entry graph) blocks')
-- Avoids using unboxed tuples when loading into GHCi
#if !defined(GHC_LOADED_INTO_GHCI)
type OptMResult a = (# a, [CLabel] #)
pattern OptMResult :: a -> b -> (# a, b #)
pattern OptMResult x y = (# x, y #)
{-# COMPLETE OptMResult #-}
#else
data OptMResult a = OptMResult !a ![CLabel] deriving (Functor)
#endif
newtype CmmOptM a = CmmOptM (DynFlags -> Module -> [CLabel] -> OptMResult a)
deriving (Functor)
instance Applicative CmmOptM where
pure x = CmmOptM $ \_ _ imports -> OptMResult x imports
(<*>) = ap
instance Monad CmmOptM where
(CmmOptM f) >>= g =
CmmOptM $ \dflags this_mod imports0 ->
case f dflags this_mod imports0 of
OptMResult x imports1 ->
case g x of
CmmOptM g' -> g' dflags this_mod imports1
instance CmmMakeDynamicReferenceM CmmOptM where
addImport = addImportCmmOpt
getThisModule = CmmOptM $ \_ this_mod imports -> OptMResult this_mod imports
addImportCmmOpt :: CLabel -> CmmOptM ()
addImportCmmOpt lbl = CmmOptM $ \_ _ imports -> OptMResult () (lbl:imports)
instance HasDynFlags CmmOptM where
getDynFlags = CmmOptM $ \dflags _ imports -> OptMResult dflags imports
runCmmOpt :: DynFlags -> Module -> CmmOptM a -> (a, [CLabel])
runCmmOpt dflags this_mod (CmmOptM f) =
case f dflags this_mod [] of
OptMResult result imports -> (result, imports)
cmmBlockConFold :: CmmBlock -> CmmOptM CmmBlock
cmmBlockConFold block = do
let (entry, middle, last) = blockSplit block
stmts = blockToList middle
stmts' <- mapM cmmStmtConFold stmts
last' <- cmmStmtConFold last
return $ blockJoin entry (blockFromList stmts') last'
-- This does three optimizations, but they're very quick to check, so we don't
-- bother turning them off even when the Hoopl code is active. Since
-- this is on the old Cmm representation, we can't reuse the code either:
-- * reg = reg --> nop
-- * if 0 then jump --> nop
-- * if 1 then jump --> jump
-- We might be tempted to skip this step entirely of not Opt_PIC, but
-- there is some PowerPC code for the non-PIC case, which would also
-- have to be separated.
cmmStmtConFold :: CmmNode e x -> CmmOptM (CmmNode e x)
cmmStmtConFold stmt
= case stmt of
CmmAssign reg src
-> do src' <- cmmExprConFold DataReference src
return $ case src' of
CmmReg reg' | reg == reg' -> CmmComment (fsLit "nop")
new_src -> CmmAssign reg new_src
CmmStore addr src
-> do addr' <- cmmExprConFold DataReference addr
src' <- cmmExprConFold DataReference src
return $ CmmStore addr' src'
CmmCall { cml_target = addr }
-> do addr' <- cmmExprConFold JumpReference addr
return $ stmt { cml_target = addr' }
CmmUnsafeForeignCall target regs args
-> do target' <- case target of
ForeignTarget e conv -> do
e' <- cmmExprConFold CallReference e
return $ ForeignTarget e' conv
PrimTarget _ ->
return target
args' <- mapM (cmmExprConFold DataReference) args
return $ CmmUnsafeForeignCall target' regs args'
CmmCondBranch test true false likely
-> do test' <- cmmExprConFold DataReference test
return $ case test' of
CmmLit (CmmInt 0 _) -> CmmBranch false
CmmLit (CmmInt _ _) -> CmmBranch true
_other -> CmmCondBranch test' true false likely
CmmSwitch expr ids
-> do expr' <- cmmExprConFold DataReference expr
return $ CmmSwitch expr' ids
other
-> return other
cmmExprConFold :: ReferenceKind -> CmmExpr -> CmmOptM CmmExpr
cmmExprConFold referenceKind expr = do
dflags <- getDynFlags
-- With -O1 and greater, the cmmSink pass does constant-folding, so
-- we don't need to do it again here.
let expr' = if optLevel dflags >= 1
then expr
else cmmExprCon dflags expr
cmmExprNative referenceKind expr'
cmmExprCon :: DynFlags -> CmmExpr -> CmmExpr
cmmExprCon dflags (CmmLoad addr rep) = CmmLoad (cmmExprCon dflags addr) rep
cmmExprCon dflags (CmmMachOp mop args)
= cmmMachOpFold platform mop (map (cmmExprCon dflags) args)
where platform = targetPlatform dflags
cmmExprCon _ other = other
-- handles both PIC and non-PIC cases... a very strange mixture
-- of things to do.
cmmExprNative :: ReferenceKind -> CmmExpr -> CmmOptM CmmExpr
cmmExprNative referenceKind expr = do
dflags <- getDynFlags
let platform = targetPlatform dflags
arch = platformArch platform
case expr of
CmmLoad addr rep
-> do addr' <- cmmExprNative DataReference addr
return $ CmmLoad addr' rep
CmmMachOp mop args
-> do args' <- mapM (cmmExprNative DataReference) args
return $ CmmMachOp mop args'
CmmLit (CmmBlock id)
-> cmmExprNative referenceKind (CmmLit (CmmLabel (infoTblLbl id)))
-- we must convert block Ids to CLabels here, because we
-- might have to do the PIC transformation. Hence we must
-- not modify BlockIds beyond this point.
CmmLit (CmmLabel lbl)
-> do
cmmMakeDynamicReference dflags referenceKind lbl
CmmLit (CmmLabelOff lbl off)
-> do
dynRef <- cmmMakeDynamicReference dflags referenceKind lbl
-- need to optimize here, since it's late
return $ cmmMachOpFold platform (MO_Add (wordWidth platform)) [
dynRef,
(CmmLit $ CmmInt (fromIntegral off) (wordWidth platform))
]
-- On powerpc (non-PIC), it's easier to jump directly to a label than
-- to use the register table, so we replace these registers
-- with the corresponding labels:
CmmReg (CmmGlobal EagerBlackholeInfo)
| arch == ArchPPC && not (positionIndependent dflags)
-> cmmExprNative referenceKind $
CmmLit (CmmLabel (mkCmmCodeLabel rtsUnitId (fsLit "__stg_EAGER_BLACKHOLE_info")))
CmmReg (CmmGlobal GCEnter1)
| arch == ArchPPC && not (positionIndependent dflags)
-> cmmExprNative referenceKind $
CmmLit (CmmLabel (mkCmmCodeLabel rtsUnitId (fsLit "__stg_gc_enter_1")))
CmmReg (CmmGlobal GCFun)
| arch == ArchPPC && not (positionIndependent dflags)
-> cmmExprNative referenceKind $
CmmLit (CmmLabel (mkCmmCodeLabel rtsUnitId (fsLit "__stg_gc_fun")))
other
-> return other
|