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
|
-- -----------------------------------------------------------------------------
--
-- (c) The University of Glasgow 1993-2004
--
-- This is the top-level module in the native code generator.
--
-- -----------------------------------------------------------------------------
\begin{code}
module AsmCodeGen ( nativeCodeGen ) where
#include "HsVersions.h"
#include "NCG.h"
import MachInstrs
import MachRegs
import MachCodeGen
import PprMach
import RegisterAlloc
import RegAllocInfo ( jumpDests )
import NCGMonad
import PositionIndependentCode
import Cmm
import CmmOpt ( cmmMiniInline, cmmMachOpFold )
import PprCmm ( pprStmt, pprCmms )
import MachOp
import CLabel ( CLabel, mkSplitMarkerLabel, mkAsmTempLabel )
#if powerpc_TARGET_ARCH
import CLabel ( mkRtsCodeLabel )
#endif
import UniqFM
import Unique ( Unique, getUnique )
import UniqSupply
import FastTypes
import List ( groupBy, sortBy )
import CLabel ( pprCLabel )
import ErrUtils ( dumpIfSet_dyn )
import DynFlags ( DynFlags, DynFlag(..), dopt )
import StaticFlags ( opt_Static, opt_PIC )
import Digraph
import qualified Pretty
import Outputable
import FastString
-- DEBUGGING ONLY
--import OrdList
#ifdef NCG_DEBUG
import List ( intersperse )
#endif
import DATA_INT
import DATA_WORD
import DATA_BITS
import GLAEXTS
{-
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 'Doc').
* ["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.
-}
-- -----------------------------------------------------------------------------
-- Top-level of the native codegen
-- NB. We *lazilly* compile each block of code for space reasons.
nativeCodeGen :: DynFlags -> [Cmm] -> UniqSupply -> IO Pretty.Doc
nativeCodeGen dflags cmms us
= let (res, _) = initUs us $
cgCmm (concat (map add_split cmms))
cgCmm :: [CmmTop] -> UniqSM (Cmm, Pretty.Doc, [CLabel])
cgCmm tops =
lazyMapUs (cmmNativeGen dflags) tops `thenUs` \ results ->
case unzip3 results of { (cmms,docs,imps) ->
returnUs (Cmm cmms, my_vcat docs, concat imps)
}
in
case res of { (ppr_cmms, insn_sdoc, imports) -> do
dumpIfSet_dyn dflags Opt_D_dump_opt_cmm "Optimised Cmm" (pprCmms [ppr_cmms])
return (insn_sdoc Pretty.$$ dyld_stubs imports
#if HAVE_SUBSECTIONS_VIA_SYMBOLS
-- On recent versions of Darwin, the linker supports
-- dead-stripping of code and data on a per-symbol basis.
-- There's a hack to make this work in PprMach.pprNatCmmTop.
Pretty.$$ Pretty.text ".subsections_via_symbols"
#endif
)
}
where
add_split (Cmm tops)
| dopt Opt_SplitObjs dflags = split_marker : tops
| otherwise = tops
split_marker = CmmProc [] mkSplitMarkerLabel [] []
-- Generate "symbol stubs" for all external symbols that might
-- come from a dynamic library.
{- dyld_stubs imps = Pretty.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
= Pretty.vcat $
(pprGotDeclaration :) $
map (pprImportedSymbol . fst . head) $
groupBy (\(_,a) (_,b) -> a == b) $
sortBy (\(_,a) (_,b) -> compare a b) $
map doPpr $
imps
| otherwise
= Pretty.empty
where doPpr lbl = (lbl, Pretty.render $ pprCLabel lbl astyle)
astyle = mkCodeStyle AsmStyle
#ifndef NCG_DEBUG
my_vcat sds = Pretty.vcat sds
#else
my_vcat sds = Pretty.vcat (
intersperse (
Pretty.char ' '
Pretty.$$ Pretty.ptext SLIT("# ___ncg_debug_marker")
Pretty.$$ Pretty.char ' '
)
sds
)
#endif
-- Complete native code generation phase for a single top-level chunk
-- of Cmm.
cmmNativeGen :: DynFlags -> CmmTop -> UniqSM (CmmTop, Pretty.Doc, [CLabel])
cmmNativeGen dflags cmm
= {-# SCC "fixAssigns" #-}
fixAssignsTop cmm `thenUs` \ fixed_cmm ->
{-# SCC "genericOpt" #-}
cmmToCmm fixed_cmm `bind` \ (cmm, imports) ->
(if dopt Opt_D_dump_opt_cmm dflags -- space leak avoidance
then cmm
else CmmData Text []) `bind` \ ppr_cmm ->
{-# SCC "genMachCode" #-}
genMachCode cmm `thenUs` \ (pre_regalloc, lastMinuteImports) ->
{-# SCC "regAlloc" #-}
mapUs regAlloc pre_regalloc `thenUs` \ with_regs ->
{-# SCC "sequenceBlocks" #-}
map sequenceTop with_regs `bind` \ sequenced ->
{-# SCC "x86fp_kludge" #-}
map x86fp_kludge sequenced `bind` \ final_mach_code ->
{-# SCC "vcat" #-}
Pretty.vcat (map pprNatCmmTop final_mach_code) `bind` \ final_sdoc ->
returnUs (ppr_cmm, final_sdoc Pretty.$$ Pretty.text "", lastMinuteImports ++ imports)
where
x86fp_kludge :: NatCmmTop -> NatCmmTop
x86fp_kludge top@(CmmData _ _) = top
#if i386_TARGET_ARCH
x86fp_kludge top@(CmmProc info lbl params code) =
CmmProc info lbl params (map bb_i386_insert_ffrees code)
where
bb_i386_insert_ffrees (BasicBlock id instrs) =
BasicBlock id (i386_insert_ffrees instrs)
#else
x86fp_kludge top = top
#endif
-- -----------------------------------------------------------------------------
-- Sequencing the basic blocks
-- Cmm BasicBlocks are self-contained entities: they always end in a
-- jump, either non-local or to another basic block in the same proc.
-- In this phase, we attempt to place the basic blocks in a sequence
-- such that as many of the local jumps as possible turn into
-- fallthroughs.
sequenceTop :: NatCmmTop -> NatCmmTop
sequenceTop top@(CmmData _ _) = top
sequenceTop (CmmProc info lbl params blocks) =
CmmProc info lbl params (sequenceBlocks blocks)
-- The algorithm is very simple (and stupid): we make a graph out of
-- the blocks where there is an edge from one block to another iff the
-- first block ends by jumping to the second. Then we topologically
-- sort this graph. Then traverse the list: for each block, we first
-- output the block, then if it has an out edge, we move the
-- destination of the out edge to the front of the list, and continue.
sequenceBlocks :: [NatBasicBlock] -> [NatBasicBlock]
sequenceBlocks [] = []
sequenceBlocks (entry:blocks) =
seqBlocks (mkNode entry : reverse (flattenSCCs (sccBlocks blocks)))
-- the first block is the entry point ==> it must remain at the start.
sccBlocks :: [NatBasicBlock] -> [SCC (NatBasicBlock,Unique,[Unique])]
sccBlocks blocks = stronglyConnCompR (map mkNode blocks)
getOutEdges :: [Instr] -> [Unique]
getOutEdges instrs = case jumpDests (last instrs) [] of
[one] -> [getUnique one]
_many -> []
-- we're only interested in the last instruction of
-- the block, and only if it has a single destination.
mkNode block@(BasicBlock id instrs) = (block, getUnique id, getOutEdges instrs)
seqBlocks [] = []
seqBlocks ((block,_,[]) : rest)
= block : seqBlocks rest
seqBlocks ((block@(BasicBlock id instrs),_,[next]) : rest)
| can_fallthrough = BasicBlock id (init instrs) : seqBlocks rest'
| otherwise = block : seqBlocks rest'
where
(can_fallthrough, rest') = reorder next [] rest
-- TODO: we should do a better job for cycles; try to maximise the
-- fallthroughs within a loop.
seqBlocks _ = panic "AsmCodegen:seqBlocks"
reorder id accum [] = (False, reverse accum)
reorder id accum (b@(block,id',out) : rest)
| id == id' = (True, (block,id,out) : reverse accum ++ rest)
| otherwise = reorder id (b:accum) rest
-- -----------------------------------------------------------------------------
-- 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.
-- Switching between the two monads whilst carrying along the same
-- Unique supply breaks abstraction. Is that bad?
genMachCode :: CmmTop -> UniqSM ([NatCmmTop], [CLabel])
genMachCode cmm_top initial_us
= let initial_st = mkNatM_State initial_us 0
(new_tops, final_st) = initNat initial_st (cmmTopCodeGen cmm_top)
final_us = natm_us final_st
final_delta = natm_delta final_st
final_imports = natm_imports final_st
in
if final_delta == 0
then ((new_tops, final_imports), final_us)
else pprPanic "genMachCode: nonzero final delta"
(int final_delta)
-- -----------------------------------------------------------------------------
-- Fixup assignments to global registers so that they assign to
-- locations within the RegTable, if appropriate.
-- Note that we currently don't fixup reads here: they're done by
-- the generic optimiser below, to avoid having two separate passes
-- over the Cmm.
fixAssignsTop :: CmmTop -> UniqSM CmmTop
fixAssignsTop top@(CmmData _ _) = returnUs top
fixAssignsTop (CmmProc info lbl params blocks) =
mapUs fixAssignsBlock blocks `thenUs` \ blocks' ->
returnUs (CmmProc info lbl params blocks')
fixAssignsBlock :: CmmBasicBlock -> UniqSM CmmBasicBlock
fixAssignsBlock (BasicBlock id stmts) =
fixAssigns stmts `thenUs` \ stmts' ->
returnUs (BasicBlock id stmts')
fixAssigns :: [CmmStmt] -> UniqSM [CmmStmt]
fixAssigns stmts =
mapUs fixAssign stmts `thenUs` \ stmtss ->
returnUs (concat stmtss)
fixAssign :: CmmStmt -> UniqSM [CmmStmt]
fixAssign (CmmAssign (CmmGlobal BaseReg) src)
= panic "cmmStmtConFold: assignment to BaseReg";
fixAssign (CmmAssign (CmmGlobal reg) src)
| Left realreg <- reg_or_addr
= returnUs [CmmAssign (CmmGlobal reg) src]
| Right baseRegAddr <- reg_or_addr
= returnUs [CmmStore baseRegAddr src]
-- Replace register leaves with appropriate StixTrees for
-- the given target. GlobalRegs which map to a reg on this
-- arch are left unchanged. Assigning to BaseReg is always
-- illegal, so we check for that.
where
reg_or_addr = get_GlobalReg_reg_or_addr reg
fixAssign (CmmCall target results args vols)
= mapAndUnzipUs fixResult results `thenUs` \ (results',stores) ->
returnUs (caller_save ++
CmmCall target results' args vols :
caller_restore ++
concat stores)
where
-- we also save/restore any caller-saves STG registers here
(caller_save, caller_restore) = callerSaveVolatileRegs vols
fixResult g@(CmmGlobal reg,hint) =
case get_GlobalReg_reg_or_addr reg of
Left realreg -> returnUs (g, [])
Right baseRegAddr ->
getUniqueUs `thenUs` \ uq ->
let local = CmmLocal (LocalReg uq (globalRegRep reg)) in
returnUs ((local,hint),
[CmmStore baseRegAddr (CmmReg local)])
fixResult other =
returnUs (other,[])
fixAssign other_stmt = returnUs [other_stmt]
-- -----------------------------------------------------------------------------
-- Generic Cmm optimiser
{-
Here we do:
(a) Constant folding
(b) Simple inlining: a temporary which is assigned to and then
used, once, can be shorted.
(c) Replacement of references to GlobalRegs which do not have
machine registers by the appropriate memory load (eg.
Hp ==> *(BaseReg + 34) ).
(d) Position independent code and dynamic linking
(i) introduce the appropriate indirections
and position independent refs
(ii) compile a list of imported symbols
Ideas for other things we could do (ToDo):
- shortcut jumps-to-jumps
- eliminate dead code blocks
- 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 :: CmmTop -> (CmmTop, [CLabel])
cmmToCmm top@(CmmData _ _) = (top, [])
cmmToCmm (CmmProc info lbl params blocks) = runCmmOpt $ do
blocks' <- mapM cmmBlockConFold (cmmMiniInline blocks)
return $ CmmProc info lbl params blocks'
newtype CmmOptM a = CmmOptM ([CLabel] -> (# a, [CLabel] #))
instance Monad CmmOptM where
return x = CmmOptM $ \imports -> (# x,imports #)
(CmmOptM f) >>= g =
CmmOptM $ \imports ->
case f imports of
(# x, imports' #) ->
case g x of
CmmOptM g' -> g' imports'
addImportCmmOpt :: CLabel -> CmmOptM ()
addImportCmmOpt lbl = CmmOptM $ \imports -> (# (), lbl:imports #)
runCmmOpt :: CmmOptM a -> (a, [CLabel])
runCmmOpt (CmmOptM f) = case f [] of
(# result, imports #) -> (result, imports)
cmmBlockConFold :: CmmBasicBlock -> CmmOptM CmmBasicBlock
cmmBlockConFold (BasicBlock id stmts) = do
stmts' <- mapM cmmStmtConFold stmts
return $ BasicBlock id stmts'
cmmStmtConFold stmt
= case stmt of
CmmAssign reg src
-> do src' <- cmmExprConFold False src
return $ case src' of
CmmReg reg' | reg == reg' -> CmmNop
new_src -> CmmAssign reg new_src
CmmStore addr src
-> do addr' <- cmmExprConFold False addr
src' <- cmmExprConFold False src
return $ CmmStore addr' src'
CmmJump addr regs
-> do addr' <- cmmExprConFold True addr
return $ CmmJump addr' regs
CmmCall target regs args vols
-> do target' <- case target of
CmmForeignCall e conv -> do
e' <- cmmExprConFold True e
return $ CmmForeignCall e' conv
other -> return other
args' <- mapM (\(arg, hint) -> do
arg' <- cmmExprConFold False arg
return (arg', hint)) args
return $ CmmCall target' regs args' vols
CmmCondBranch test dest
-> do test' <- cmmExprConFold False test
return $ case test' of
CmmLit (CmmInt 0 _) ->
CmmComment (mkFastString ("deleted: " ++
showSDoc (pprStmt stmt)))
CmmLit (CmmInt n _) -> CmmBranch dest
other -> CmmCondBranch test' dest
CmmSwitch expr ids
-> do expr' <- cmmExprConFold False expr
return $ CmmSwitch expr' ids
other
-> return other
cmmExprConFold isJumpTarget expr
= case expr of
CmmLoad addr rep
-> do addr' <- cmmExprConFold False addr
return $ CmmLoad addr' rep
CmmMachOp mop args
-- For MachOps, we first optimize the children, and then we try
-- our hand at some constant-folding.
-> do args' <- mapM (cmmExprConFold False) args
return $ cmmMachOpFold mop args'
CmmLit (CmmLabel lbl)
-> cmmMakeDynamicReference addImportCmmOpt isJumpTarget lbl
CmmLit (CmmLabelOff lbl off)
-> do dynRef <- cmmMakeDynamicReference addImportCmmOpt isJumpTarget lbl
return $ cmmMachOpFold (MO_Add wordRep) [
dynRef,
(CmmLit $ CmmInt (fromIntegral off) wordRep)
]
#if powerpc_TARGET_ARCH
-- 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 GCEnter1)
| not opt_PIC
-> cmmExprConFold isJumpTarget $
CmmLit (CmmLabel (mkRtsCodeLabel SLIT( "__stg_gc_enter_1")))
CmmReg (CmmGlobal GCFun)
| not opt_PIC
-> cmmExprConFold isJumpTarget $
CmmLit (CmmLabel (mkRtsCodeLabel SLIT( "__stg_gc_fun")))
#endif
CmmReg (CmmGlobal mid)
-- Replace register leaves with appropriate StixTrees for
-- the given target. MagicIds which map to a reg on this
-- arch are left unchanged. For the rest, BaseReg is taken
-- to mean the address of the reg table in MainCapability,
-- and for all others we generate an indirection to its
-- location in the register table.
-> case get_GlobalReg_reg_or_addr mid of
Left realreg -> return expr
Right baseRegAddr
-> case mid of
BaseReg -> cmmExprConFold False baseRegAddr
other -> cmmExprConFold False (CmmLoad baseRegAddr
(globalRegRep mid))
-- eliminate zero offsets
CmmRegOff reg 0
-> cmmExprConFold False (CmmReg reg)
CmmRegOff (CmmGlobal mid) offset
-- RegOf leaves are just a shorthand form. If the reg maps
-- to a real reg, we keep the shorthand, otherwise, we just
-- expand it and defer to the above code.
-> case get_GlobalReg_reg_or_addr mid of
Left realreg -> return expr
Right baseRegAddr
-> cmmExprConFold False (CmmMachOp (MO_Add wordRep) [
CmmReg (CmmGlobal mid),
CmmLit (CmmInt (fromIntegral offset)
wordRep)])
other
-> return other
-- -----------------------------------------------------------------------------
-- Utils
bind f x = x $! f
\end{code}
|