1 -- -----------------------------------------------------------------------------
3 -- (c) The University of Glasgow 1993-2004
5 -- This is the top-level module in the native code generator.
7 -- -----------------------------------------------------------------------------
10 module AsmCodeGen ( nativeCodeGen ) where
12 #include "HsVersions.h"
13 #include "nativeGen/NCG.h"
22 #elif i386_TARGET_ARCH || x86_64_TARGET_ARCH
28 #elif sparc_TARGET_ARCH
30 import SPARC.CodeGen.Expand
34 import SPARC.ShortcutJump
36 #elif powerpc_TARGET_ARCH
45 #error "AsmCodeGen: unknown architecture"
49 import RegAlloc.Liveness
50 import qualified RegAlloc.Linear.Main as Linear
52 import qualified GraphColor as Color
53 import qualified RegAlloc.Graph.Main as Color
54 import qualified RegAlloc.Graph.Stats as Color
55 import qualified RegAlloc.Graph.TrivColorable as Color
65 import CgUtils ( fixStgRegisters )
67 import CmmOpt ( cmmMiniInline, cmmMachOpFold )
72 import Unique ( Unique, getUnique )
75 #if powerpc_TARGET_ARCH
76 import StaticFlags ( opt_Static, opt_PIC )
79 #if !defined(darwin_TARGET_OS)
80 import Config ( cProjectVersion )
84 import qualified Pretty
101 The native-code generator has machine-independent and
102 machine-dependent modules.
104 This module ("AsmCodeGen") is the top-level machine-independent
105 module. Before entering machine-dependent land, we do some
106 machine-independent optimisations (defined below) on the
109 We convert to the machine-specific 'Instr' datatype with
110 'cmmCodeGen', assuming an infinite supply of registers. We then use
111 a machine-independent register allocator ('regAlloc') to rejoin
112 reality. Obviously, 'regAlloc' has machine-specific helper
113 functions (see about "RegAllocInfo" below).
115 Finally, we order the basic blocks of the function so as to minimise
116 the number of jumps between blocks, by utilising fallthrough wherever
119 The machine-dependent bits break down as follows:
121 * ["MachRegs"] Everything about the target platform's machine
122 registers (and immediate operands, and addresses, which tend to
123 intermingle/interact with registers).
125 * ["MachInstrs"] Includes the 'Instr' datatype (possibly should
126 have a module of its own), plus a miscellany of other things
127 (e.g., 'targetDoubleSize', 'smStablePtrTable', ...)
129 * ["MachCodeGen"] is where 'Cmm' stuff turns into
130 machine instructions.
132 * ["PprMach"] 'pprInstr' turns an 'Instr' into text (well, really
135 * ["RegAllocInfo"] In the register allocator, we manipulate
136 'MRegsState's, which are 'BitSet's, one bit per machine register.
137 When we want to say something about a specific machine register
138 (e.g., ``it gets clobbered by this instruction''), we set/unset
139 its bit. Obviously, we do this 'BitSet' thing for efficiency
142 The 'RegAllocInfo' module collects together the machine-specific
143 info needed to do register allocation.
145 * ["RegisterAlloc"] The (machine-independent) register allocator.
148 -- -----------------------------------------------------------------------------
149 -- Top-level of the native codegen
152 nativeCodeGen :: DynFlags -> Handle -> UniqSupply -> [RawCmm] -> IO ()
153 nativeCodeGen dflags h us cmms
155 let split_cmms = concat $ map add_split cmms
157 -- BufHandle is a performance hack. We could hide it inside
158 -- Pretty if it weren't for the fact that we do lots of little
159 -- printDocs here (in order to do codegen in constant space).
160 bufh <- newBufHandle h
161 (imports, prof) <- cmmNativeGens dflags bufh us split_cmms [] [] 0
164 let (native, colorStats, linearStats)
169 Opt_D_dump_asm "Asm code"
170 (vcat $ map (docToSDoc . pprNatCmmTop) $ concat native)
172 -- dump global NCG stats for graph coloring allocator
173 (case concat $ catMaybes colorStats of
176 -- build the global register conflict graph
178 = foldl Color.union Color.initGraph
179 $ [ Color.raGraph stat
180 | stat@Color.RegAllocStatsStart{} <- stats]
182 dumpSDoc dflags Opt_D_dump_asm_stats "NCG stats"
183 $ Color.pprStats stats graphGlobal
186 Opt_D_dump_asm_conflicts "Register conflict graph"
190 targetVirtualRegSqueeze
191 targetRealRegSqueeze)
195 -- dump global NCG stats for linear allocator
196 (case concat $ catMaybes linearStats of
198 stats -> dumpSDoc dflags Opt_D_dump_asm_stats "NCG stats"
199 $ Linear.pprStats (concat native) stats)
201 -- write out the imports
202 Pretty.printDoc Pretty.LeftMode h
203 $ makeImportsDoc dflags (concat imports)
207 where add_split (Cmm tops)
208 | dopt Opt_SplitObjs dflags = split_marker : tops
211 split_marker = CmmProc [] mkSplitMarkerLabel [] (ListGraph [])
214 -- | Do native code generation on all these cmms.
216 cmmNativeGens :: DynFlags
221 -> [ ([NatCmmTop Instr],
222 Maybe [Color.RegAllocStats Instr],
223 Maybe [Linear.RegAllocStats]) ]
227 Maybe [Color.RegAllocStats Instr],
228 Maybe [Linear.RegAllocStats])] )
230 cmmNativeGens _ _ _ [] impAcc profAcc _
231 = return (reverse impAcc, reverse profAcc)
233 cmmNativeGens dflags h us (cmm : cmms) impAcc profAcc count
235 (us', native, imports, colorStats, linearStats)
236 <- cmmNativeGen dflags us cmm count
238 Pretty.bufLeftRender h
239 $ {-# SCC "pprNativeCode" #-} Pretty.vcat $ map pprNatCmmTop native
241 -- carefully evaluate this strictly. Binding it with 'let'
242 -- and then using 'seq' doesn't work, because the let
243 -- apparently gets inlined first.
244 lsPprNative <- return $!
245 if dopt Opt_D_dump_asm dflags
246 || dopt Opt_D_dump_asm_stats dflags
250 count' <- return $! count + 1;
252 -- force evaulation all this stuff to avoid space leaks
253 seqString (showSDoc $ vcat $ map ppr imports) `seq` return ()
255 cmmNativeGens dflags h us' cmms
257 ((lsPprNative, colorStats, linearStats) : profAcc)
260 where seqString [] = ()
261 seqString (x:xs) = x `seq` seqString xs `seq` ()
264 -- | Complete native code generation phase for a single top-level chunk of Cmm.
265 -- Dumping the output of each stage along the way.
266 -- Global conflict graph and NGC stats
270 -> RawCmmTop -- ^ the cmm to generate code for
271 -> Int -- ^ sequence number of this top thing
273 , [NatCmmTop Instr] -- native code
274 , [CLabel] -- things imported by this cmm
275 , Maybe [Color.RegAllocStats Instr] -- stats for the coloring register allocator
276 , Maybe [Linear.RegAllocStats]) -- stats for the linear register allocators
278 cmmNativeGen dflags us cmm count
281 -- rewrite assignments to global regs
283 {-# SCC "fixStgRegisters" #-}
286 -- cmm to cmm optimisations
287 let (opt_cmm, imports) =
288 {-# SCC "cmmToCmm" #-}
289 cmmToCmm dflags fixed_cmm
292 Opt_D_dump_opt_cmm "Optimised Cmm"
293 (pprCmm $ Cmm [opt_cmm])
295 -- generate native code from cmm
296 let ((native, lastMinuteImports), usGen) =
297 {-# SCC "genMachCode" #-}
298 initUs us $ genMachCode dflags opt_cmm
301 Opt_D_dump_asm_native "Native code"
302 (vcat $ map (docToSDoc . pprNatCmmTop) native)
304 -- tag instructions with register liveness information
305 let (withLiveness, usLive) =
306 {-# SCC "regLiveness" #-}
309 $ map natCmmTopToLive native
312 Opt_D_dump_asm_liveness "Liveness annotations added"
313 (vcat $ map ppr withLiveness)
315 -- allocate registers
316 (alloced, usAlloc, ppr_raStatsColor, ppr_raStatsLinear) <-
317 if ( dopt Opt_RegsGraph dflags
318 || dopt Opt_RegsIterative dflags)
320 -- the regs usable for allocation
321 let (alloc_regs :: UniqFM (UniqSet RealReg))
322 = foldr (\r -> plusUFM_C unionUniqSets
323 $ unitUFM (targetClassOfRealReg r) (unitUniqSet r))
327 -- do the graph coloring register allocation
328 let ((alloced, regAllocStats), usAlloc)
329 = {-# SCC "RegAlloc" #-}
334 (mkUniqSet [0..maxSpillSlots])
337 -- dump out what happened during register allocation
339 Opt_D_dump_asm_regalloc "Registers allocated"
340 (vcat $ map (docToSDoc . pprNatCmmTop) alloced)
343 Opt_D_dump_asm_regalloc_stages "Build/spill stages"
344 (vcat $ map (\(stage, stats)
345 -> text "# --------------------------"
346 $$ text "# cmm " <> int count <> text " Stage " <> int stage
348 $ zip [0..] regAllocStats)
351 if dopt Opt_D_dump_asm_stats dflags
352 then Just regAllocStats else Nothing
354 -- force evaluation of the Maybe to avoid space leak
355 mPprStats `seq` return ()
357 return ( alloced, usAlloc
362 -- do linear register allocation
363 let ((alloced, regAllocStats), usAlloc)
364 = {-# SCC "RegAlloc" #-}
367 $ mapUs Linear.regAlloc withLiveness
370 Opt_D_dump_asm_regalloc "Registers allocated"
371 (vcat $ map (docToSDoc . pprNatCmmTop) alloced)
374 if dopt Opt_D_dump_asm_stats dflags
375 then Just (catMaybes regAllocStats) else Nothing
377 -- force evaluation of the Maybe to avoid space leak
378 mPprStats `seq` return ()
380 return ( alloced, usAlloc
384 ---- shortcut branches
386 {-# SCC "shortcutBranches" #-}
387 shortcutBranches dflags alloced
391 {-# SCC "sequenceBlocks" #-}
392 map sequenceTop shorted
397 {-# SCC "x86fp_kludge" #-}
398 map x86fp_kludge sequenced
403 ---- expansion of SPARC synthetic instrs
404 #if sparc_TARGET_ARCH
406 {-# SCC "sparc_expand" #-}
407 map expandTop kludged
410 Opt_D_dump_asm_expanded "Synthetic instructions expanded"
411 (vcat $ map (docToSDoc . pprNatCmmTop) expanded)
419 , lastMinuteImports ++ imports
425 x86fp_kludge :: NatCmmTop Instr -> NatCmmTop Instr
426 x86fp_kludge top@(CmmData _ _) = top
427 x86fp_kludge (CmmProc info lbl params (ListGraph code)) =
428 CmmProc info lbl params (ListGraph $ i386_insert_ffrees code)
432 -- | Build a doc for all the imports.
434 makeImportsDoc :: DynFlags -> [CLabel] -> Pretty.Doc
435 makeImportsDoc dflags imports
438 #if HAVE_SUBSECTIONS_VIA_SYMBOLS
439 -- On recent versions of Darwin, the linker supports
440 -- dead-stripping of code and data on a per-symbol basis.
441 -- There's a hack to make this work in PprMach.pprNatCmmTop.
442 Pretty.$$ Pretty.text ".subsections_via_symbols"
444 #if HAVE_GNU_NONEXEC_STACK
445 -- On recent GNU ELF systems one can mark an object file
446 -- as not requiring an executable stack. If all objects
447 -- linked into a program have this note then the program
448 -- will not use an executable stack, which is good for
449 -- security. GHC generated code does not need an executable
450 -- stack so add the note in:
451 Pretty.$$ Pretty.text ".section .note.GNU-stack,\"\",@progbits"
453 #if !defined(darwin_TARGET_OS)
454 -- And just because every other compiler does, lets stick in
455 -- an identifier directive: .ident "GHC x.y.z"
456 Pretty.$$ let compilerIdent = Pretty.text "GHC" Pretty.<+>
457 Pretty.text cProjectVersion
458 in Pretty.text ".ident" Pretty.<+>
459 Pretty.doubleQuotes compilerIdent
463 -- Generate "symbol stubs" for all external symbols that might
464 -- come from a dynamic library.
465 dyld_stubs :: [CLabel] -> Pretty.Doc
466 {- dyld_stubs imps = Pretty.vcat $ map pprDyldSymbolStub $
467 map head $ group $ sort imps-}
469 arch = platformArch $ targetPlatform dflags
470 os = platformOS $ targetPlatform dflags
472 -- (Hack) sometimes two Labels pretty-print the same, but have
473 -- different uniques; so we compare their text versions...
475 | needImportedSymbols arch os
477 (pprGotDeclaration arch os :) $
478 map ( pprImportedSymbol arch os . fst . head) $
479 groupBy (\(_,a) (_,b) -> a == b) $
480 sortBy (\(_,a) (_,b) -> compare a b) $
486 doPpr lbl = (lbl, Pretty.render $ pprCLabel lbl astyle)
487 astyle = mkCodeStyle AsmStyle
490 -- -----------------------------------------------------------------------------
491 -- Sequencing the basic blocks
493 -- Cmm BasicBlocks are self-contained entities: they always end in a
494 -- jump, either non-local or to another basic block in the same proc.
495 -- In this phase, we attempt to place the basic blocks in a sequence
496 -- such that as many of the local jumps as possible turn into
503 sequenceTop top@(CmmData _ _) = top
504 sequenceTop (CmmProc info lbl params (ListGraph blocks)) =
505 CmmProc info lbl params (ListGraph $ makeFarBranches $ sequenceBlocks blocks)
507 -- The algorithm is very simple (and stupid): we make a graph out of
508 -- the blocks where there is an edge from one block to another iff the
509 -- first block ends by jumping to the second. Then we topologically
510 -- sort this graph. Then traverse the list: for each block, we first
511 -- output the block, then if it has an out edge, we move the
512 -- destination of the out edge to the front of the list, and continue.
514 -- FYI, the classic layout for basic blocks uses postorder DFS; this
515 -- algorithm is implemented in cmm/ZipCfg.hs (NR 6 Sep 2007).
519 => [NatBasicBlock instr]
520 -> [NatBasicBlock instr]
522 sequenceBlocks [] = []
523 sequenceBlocks (entry:blocks) =
524 seqBlocks (mkNode entry : reverse (flattenSCCs (sccBlocks blocks)))
525 -- the first block is the entry point ==> it must remain at the start.
530 => [NatBasicBlock instr]
531 -> [SCC ( NatBasicBlock instr
535 sccBlocks blocks = stronglyConnCompFromEdgedVerticesR (map mkNode blocks)
537 -- we're only interested in the last instruction of
538 -- the block, and only if it has a single destination.
541 => [instr] -> [Unique]
544 = case jumpDestsOfInstr (last instrs) of
545 [one] -> [getUnique one]
548 mkNode :: (Instruction t)
550 -> (GenBasicBlock t, Unique, [Unique])
551 mkNode block@(BasicBlock id instrs) = (block, getUnique id, getOutEdges instrs)
553 seqBlocks :: (Eq t) => [(GenBasicBlock t1, t, [t])] -> [GenBasicBlock t1]
555 seqBlocks ((block,_,[]) : rest)
556 = block : seqBlocks rest
557 seqBlocks ((block@(BasicBlock id instrs),_,[next]) : rest)
558 | can_fallthrough = BasicBlock id (init instrs) : seqBlocks rest'
559 | otherwise = block : seqBlocks rest'
561 (can_fallthrough, rest') = reorder next [] rest
562 -- TODO: we should do a better job for cycles; try to maximise the
563 -- fallthroughs within a loop.
564 seqBlocks _ = panic "AsmCodegen:seqBlocks"
566 reorder :: (Eq a) => a -> [(t, a, t1)] -> [(t, a, t1)] -> (Bool, [(t, a, t1)])
567 reorder _ accum [] = (False, reverse accum)
568 reorder id accum (b@(block,id',out) : rest)
569 | id == id' = (True, (block,id,out) : reverse accum ++ rest)
570 | otherwise = reorder id (b:accum) rest
573 -- -----------------------------------------------------------------------------
574 -- Making far branches
576 -- Conditional branches on PowerPC are limited to +-32KB; if our Procs get too
577 -- big, we have to work around this limitation.
580 :: [NatBasicBlock Instr]
581 -> [NatBasicBlock Instr]
583 #if powerpc_TARGET_ARCH
584 makeFarBranches blocks
585 | last blockAddresses < nearLimit = blocks
586 | otherwise = zipWith handleBlock blockAddresses blocks
588 blockAddresses = scanl (+) 0 $ map blockLen blocks
589 blockLen (BasicBlock _ instrs) = length instrs
591 handleBlock addr (BasicBlock id instrs)
592 = BasicBlock id (zipWith makeFar [addr..] instrs)
594 makeFar addr (BCC ALWAYS tgt) = BCC ALWAYS tgt
595 makeFar addr (BCC cond tgt)
596 | abs (addr - targetAddr) >= nearLimit
600 where Just targetAddr = lookupUFM blockAddressMap tgt
601 makeFar addr other = other
603 nearLimit = 7000 -- 8192 instructions are allowed; let's keep some
604 -- distance, as we have a few pseudo-insns that are
605 -- pretty-printed as multiple instructions,
606 -- and it's just not worth the effort to calculate
609 blockAddressMap = listToUFM $ zip (map blockId blocks) blockAddresses
614 -- -----------------------------------------------------------------------------
622 shortcutBranches dflags tops
623 | optLevel dflags < 1 = tops -- only with -O or higher
624 | otherwise = map (apply_mapping mapping) tops'
626 (tops', mappings) = mapAndUnzip build_mapping tops
627 mapping = foldr plusUFM emptyUFM mappings
629 build_mapping :: GenCmmTop d t (ListGraph Instr)
630 -> (GenCmmTop d t (ListGraph Instr), UniqFM JumpDest)
631 build_mapping top@(CmmData _ _) = (top, emptyUFM)
632 build_mapping (CmmProc info lbl params (ListGraph []))
633 = (CmmProc info lbl params (ListGraph []), emptyUFM)
634 build_mapping (CmmProc info lbl params (ListGraph (head:blocks)))
635 = (CmmProc info lbl params (ListGraph (head:others)), mapping)
636 -- drop the shorted blocks, but don't ever drop the first one,
637 -- because it is pointed to by a global label.
639 -- find all the blocks that just consist of a jump that can be
641 -- Don't completely eliminate loops here -- that can leave a dangling jump!
642 (_, shortcut_blocks, others) = foldl split (emptyBlockSet, [], []) blocks
643 split (s, shortcut_blocks, others) b@(BasicBlock id [insn])
644 | Just (DestBlockId dest) <- canShortcut insn,
645 (elemBlockSet dest s) || dest == id -- loop checks
646 = (s, shortcut_blocks, b : others)
647 split (s, shortcut_blocks, others) (BasicBlock id [insn])
648 | Just dest <- canShortcut insn
649 = (extendBlockSet s id, (id,dest) : shortcut_blocks, others)
650 split (s, shortcut_blocks, others) other = (s, shortcut_blocks, other : others)
653 -- build a mapping from BlockId to JumpDest for shorting branches
654 mapping = foldl add emptyUFM shortcut_blocks
655 add ufm (id,dest) = addToUFM ufm id dest
657 apply_mapping :: UniqFM JumpDest
658 -> GenCmmTop CmmStatic h (ListGraph Instr)
659 -> GenCmmTop CmmStatic h (ListGraph Instr)
660 apply_mapping ufm (CmmData sec statics)
661 = CmmData sec (map (shortcutStatic (lookupUFM ufm)) statics)
662 -- we need to get the jump tables, so apply the mapping to the entries
664 apply_mapping ufm (CmmProc info lbl params (ListGraph blocks))
665 = CmmProc info lbl params (ListGraph $ map short_bb blocks)
667 short_bb (BasicBlock id insns) = BasicBlock id $! map short_insn insns
668 short_insn i = shortcutJump (lookupUFM ufm) i
669 -- shortcutJump should apply the mapping repeatedly,
670 -- just in case we can short multiple branches.
672 -- -----------------------------------------------------------------------------
673 -- Instruction selection
675 -- Native code instruction selection for a chunk of stix code. For
676 -- this part of the computation, we switch from the UniqSM monad to
677 -- the NatM monad. The latter carries not only a Unique, but also an
678 -- Int denoting the current C stack pointer offset in the generated
679 -- code; this is needed for creating correct spill offsets on
680 -- architectures which don't offer, or for which it would be
681 -- prohibitively expensive to employ, a frame pointer register. Viz,
684 -- The offset is measured in bytes, and indicates the difference
685 -- between the current (simulated) C stack-ptr and the value it was at
686 -- the beginning of the block. For stacks which grow down, this value
687 -- should be either zero or negative.
689 -- Switching between the two monads whilst carrying along the same
690 -- Unique supply breaks abstraction. Is that bad?
699 genMachCode dflags cmm_top
700 = do { initial_us <- getUs
701 ; let initial_st = mkNatM_State initial_us 0 dflags
702 (new_tops, final_st) = initNat initial_st (cmmTopCodeGen dflags cmm_top)
703 final_delta = natm_delta final_st
704 final_imports = natm_imports final_st
705 ; if final_delta == 0
706 then return (new_tops, final_imports)
707 else pprPanic "genMachCode: nonzero final delta" (int final_delta)
711 -- -----------------------------------------------------------------------------
712 -- Generic Cmm optimiser
718 (b) Simple inlining: a temporary which is assigned to and then
719 used, once, can be shorted.
720 (c) Position independent code and dynamic linking
721 (i) introduce the appropriate indirections
722 and position independent refs
723 (ii) compile a list of imported symbols
725 Ideas for other things we could do (ToDo):
727 - shortcut jumps-to-jumps
728 - eliminate dead code blocks
729 - simple CSE: if an expr is assigned to a temp, then replace later occs of
730 that expr with the temp, until the expr is no longer valid (can push through
731 temp assignments, and certain assigns to mem...)
734 cmmToCmm :: DynFlags -> RawCmmTop -> (RawCmmTop, [CLabel])
735 cmmToCmm _ top@(CmmData _ _) = (top, [])
736 cmmToCmm dflags (CmmProc info lbl params (ListGraph blocks)) = runCmmOpt dflags $ do
737 blocks' <- mapM cmmBlockConFold (cmmMiniInline blocks)
738 return $ CmmProc info lbl params (ListGraph blocks')
740 newtype CmmOptM a = CmmOptM (([CLabel], DynFlags) -> (# a, [CLabel] #))
742 instance Monad CmmOptM where
743 return x = CmmOptM $ \(imports, _) -> (# x,imports #)
745 CmmOptM $ \(imports, dflags) ->
746 case f (imports, dflags) of
749 CmmOptM g' -> g' (imports', dflags)
751 addImportCmmOpt :: CLabel -> CmmOptM ()
752 addImportCmmOpt lbl = CmmOptM $ \(imports, _dflags) -> (# (), lbl:imports #)
754 getDynFlagsCmmOpt :: CmmOptM DynFlags
755 getDynFlagsCmmOpt = CmmOptM $ \(imports, dflags) -> (# dflags, imports #)
757 runCmmOpt :: DynFlags -> CmmOptM a -> (a, [CLabel])
758 runCmmOpt dflags (CmmOptM f) = case f ([], dflags) of
759 (# result, imports #) -> (result, imports)
761 cmmBlockConFold :: CmmBasicBlock -> CmmOptM CmmBasicBlock
762 cmmBlockConFold (BasicBlock id stmts) = do
763 stmts' <- mapM cmmStmtConFold stmts
764 return $ BasicBlock id stmts'
766 cmmStmtConFold :: CmmStmt -> CmmOptM CmmStmt
770 -> do src' <- cmmExprConFold DataReference src
771 return $ case src' of
772 CmmReg reg' | reg == reg' -> CmmNop
773 new_src -> CmmAssign reg new_src
776 -> do addr' <- cmmExprConFold DataReference addr
777 src' <- cmmExprConFold DataReference src
778 return $ CmmStore addr' src'
781 -> do addr' <- cmmExprConFold JumpReference addr
782 return $ CmmJump addr' regs
784 CmmCall target regs args srt returns
785 -> do target' <- case target of
786 CmmCallee e conv -> do
787 e' <- cmmExprConFold CallReference e
788 return $ CmmCallee e' conv
789 other -> return other
790 args' <- mapM (\(CmmHinted arg hint) -> do
791 arg' <- cmmExprConFold DataReference arg
792 return (CmmHinted arg' hint)) args
793 return $ CmmCall target' regs args' srt returns
795 CmmCondBranch test dest
796 -> do test' <- cmmExprConFold DataReference test
797 return $ case test' of
798 CmmLit (CmmInt 0 _) ->
799 CmmComment (mkFastString ("deleted: " ++
800 showSDoc (pprStmt stmt)))
802 CmmLit (CmmInt _ _) -> CmmBranch dest
803 _other -> CmmCondBranch test' dest
806 -> do expr' <- cmmExprConFold DataReference expr
807 return $ CmmSwitch expr' ids
813 cmmExprConFold :: ReferenceKind -> CmmExpr -> CmmOptM CmmExpr
814 cmmExprConFold referenceKind expr
817 -> do addr' <- cmmExprConFold DataReference addr
818 return $ CmmLoad addr' rep
821 -- For MachOps, we first optimize the children, and then we try
822 -- our hand at some constant-folding.
823 -> do args' <- mapM (cmmExprConFold DataReference) args
824 return $ cmmMachOpFold mop args'
826 CmmLit (CmmLabel lbl)
828 dflags <- getDynFlagsCmmOpt
829 cmmMakeDynamicReference dflags addImportCmmOpt referenceKind lbl
830 CmmLit (CmmLabelOff lbl off)
832 dflags <- getDynFlagsCmmOpt
833 dynRef <- cmmMakeDynamicReference dflags addImportCmmOpt referenceKind lbl
834 return $ cmmMachOpFold (MO_Add wordWidth) [
836 (CmmLit $ CmmInt (fromIntegral off) wordWidth)
839 #if powerpc_TARGET_ARCH
840 -- On powerpc (non-PIC), it's easier to jump directly to a label than
841 -- to use the register table, so we replace these registers
842 -- with the corresponding labels:
843 CmmReg (CmmGlobal EagerBlackholeInfo)
845 -> cmmExprConFold referenceKind $
846 CmmLit (CmmLabel (mkCmmCodeLabel rtsPackageId (fsLit "__stg_EAGER_BLACKHOLE_info")))
847 CmmReg (CmmGlobal GCEnter1)
849 -> cmmExprConFold referenceKind $
850 CmmLit (CmmLabel (mkCmmCodeLabel rtsPackageId (fsLit "__stg_gc_enter_1")))
851 CmmReg (CmmGlobal GCFun)
853 -> cmmExprConFold referenceKind $
854 CmmLit (CmmLabel (mkCmmCodeLabel rtsPackageId (fsLit "__stg_gc_fun")))