1 -- -----------------------------------------------------------------------------
3 -- (c) The University of Glasgow 1993-2004
5 -- This is the top-level module in the native code generator.
7 -- -----------------------------------------------------------------------------
11 -- The above warning supression flag is a temporary kludge.
12 -- While working on this module you are encouraged to remove it and fix
13 -- any warnings in the module. See
14 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
17 module AsmCodeGen ( nativeCodeGen ) where
19 #include "HsVersions.h"
20 #include "nativeGen/NCG.h"
29 #elif i386_TARGET_ARCH || x86_64_TARGET_ARCH
36 #elif sparc_TARGET_ARCH
41 import SPARC.ShortcutJump
43 #elif powerpc_TARGET_ARCH
52 #error "AsmCodeGen: unknown architecture"
56 import RegAlloc.Liveness
57 import qualified RegAlloc.Linear.Main as Linear
59 import qualified GraphColor as Color
60 import qualified RegAlloc.Graph.Main as Color
61 import qualified RegAlloc.Graph.Stats as Color
62 import qualified RegAlloc.Graph.Coalesce as Color
63 import qualified RegAlloc.Graph.TrivColorable as Color
65 import qualified SPARC.CodeGen.Expand as SPARC
76 import CgUtils ( fixStgRegisters )
78 import CmmOpt ( cmmMiniInline, cmmMachOpFold )
84 import Unique ( Unique, getUnique )
87 #if powerpc_TARGET_ARCH
88 import StaticFlags ( opt_Static, opt_PIC )
91 import Config ( cProjectVersion )
95 import qualified Pretty
115 The native-code generator has machine-independent and
116 machine-dependent modules.
118 This module ("AsmCodeGen") is the top-level machine-independent
119 module. Before entering machine-dependent land, we do some
120 machine-independent optimisations (defined below) on the
123 We convert to the machine-specific 'Instr' datatype with
124 'cmmCodeGen', assuming an infinite supply of registers. We then use
125 a machine-independent register allocator ('regAlloc') to rejoin
126 reality. Obviously, 'regAlloc' has machine-specific helper
127 functions (see about "RegAllocInfo" below).
129 Finally, we order the basic blocks of the function so as to minimise
130 the number of jumps between blocks, by utilising fallthrough wherever
133 The machine-dependent bits break down as follows:
135 * ["MachRegs"] Everything about the target platform's machine
136 registers (and immediate operands, and addresses, which tend to
137 intermingle/interact with registers).
139 * ["MachInstrs"] Includes the 'Instr' datatype (possibly should
140 have a module of its own), plus a miscellany of other things
141 (e.g., 'targetDoubleSize', 'smStablePtrTable', ...)
143 * ["MachCodeGen"] is where 'Cmm' stuff turns into
144 machine instructions.
146 * ["PprMach"] 'pprInstr' turns an 'Instr' into text (well, really
149 * ["RegAllocInfo"] In the register allocator, we manipulate
150 'MRegsState's, which are 'BitSet's, one bit per machine register.
151 When we want to say something about a specific machine register
152 (e.g., ``it gets clobbered by this instruction''), we set/unset
153 its bit. Obviously, we do this 'BitSet' thing for efficiency
156 The 'RegAllocInfo' module collects together the machine-specific
157 info needed to do register allocation.
159 * ["RegisterAlloc"] The (machine-independent) register allocator.
162 -- -----------------------------------------------------------------------------
163 -- Top-level of the native codegen
166 nativeCodeGen :: DynFlags -> Handle -> UniqSupply -> [RawCmm] -> IO ()
167 nativeCodeGen dflags h us cmms
169 let split_cmms = concat $ map add_split cmms
171 -- BufHandle is a performance hack. We could hide it inside
172 -- Pretty if it weren't for the fact that we do lots of little
173 -- printDocs here (in order to do codegen in constant space).
174 bufh <- newBufHandle h
175 (imports, prof) <- cmmNativeGens dflags bufh us split_cmms [] [] 0
178 let (native, colorStats, linearStats)
183 Opt_D_dump_asm "Asm code"
184 (vcat $ map (docToSDoc . pprNatCmmTop) $ concat native)
186 -- dump global NCG stats for graph coloring allocator
187 (case concat $ catMaybes colorStats of
190 -- build the global register conflict graph
192 = foldl Color.union Color.initGraph
193 $ [ Color.raGraph stat
194 | stat@Color.RegAllocStatsStart{} <- stats]
196 dumpSDoc dflags Opt_D_dump_asm_stats "NCG stats"
197 $ Color.pprStats stats graphGlobal
200 Opt_D_dump_asm_conflicts "Register conflict graph"
204 targetVirtualRegSqueeze
205 targetRealRegSqueeze)
209 -- dump global NCG stats for linear allocator
210 (case concat $ catMaybes linearStats of
212 stats -> dumpSDoc dflags Opt_D_dump_asm_stats "NCG stats"
213 $ Linear.pprStats (concat native) stats)
215 -- write out the imports
216 Pretty.printDoc Pretty.LeftMode h
217 $ makeImportsDoc dflags (concat imports)
221 where add_split (Cmm tops)
222 | dopt Opt_SplitObjs dflags = split_marker : tops
225 split_marker = CmmProc [] mkSplitMarkerLabel [] (ListGraph [])
228 -- | Do native code generation on all these cmms.
230 cmmNativeGens dflags h us [] impAcc profAcc count
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" #-}
307 initUs usGen $ mapUs regLiveness native
310 Opt_D_dump_asm_liveness "Liveness annotations added"
311 (vcat $ map ppr withLiveness)
313 -- allocate registers
314 (alloced, usAlloc, ppr_raStatsColor, ppr_raStatsLinear) <-
315 if ( dopt Opt_RegsGraph dflags
316 || dopt Opt_RegsIterative dflags)
318 -- the regs usable for allocation
319 let (alloc_regs :: UniqFM (UniqSet RealReg))
320 = foldr (\r -> plusUFM_C unionUniqSets
321 $ unitUFM (targetClassOfRealReg r) (unitUniqSet r))
325 -- do the graph coloring register allocation
326 let ((alloced, regAllocStats), usAlloc)
327 = {-# SCC "RegAlloc" #-}
332 (mkUniqSet [0..maxSpillSlots])
335 -- dump out what happened during register allocation
337 Opt_D_dump_asm_regalloc "Registers allocated"
338 (vcat $ map (docToSDoc . pprNatCmmTop) alloced)
341 Opt_D_dump_asm_regalloc_stages "Build/spill stages"
342 (vcat $ map (\(stage, stats)
343 -> text "# --------------------------"
344 $$ text "# cmm " <> int count <> text " Stage " <> int stage
346 $ zip [0..] regAllocStats)
349 if dopt Opt_D_dump_asm_stats dflags
350 then Just regAllocStats else Nothing
352 -- force evaluation of the Maybe to avoid space leak
353 mPprStats `seq` return ()
355 return ( alloced, usAlloc
360 -- do linear register allocation
361 let ((alloced, regAllocStats), usAlloc)
362 = {-# SCC "RegAlloc" #-}
365 $ mapUs Linear.regAlloc withLiveness
368 Opt_D_dump_asm_regalloc "Registers allocated"
369 (vcat $ map (docToSDoc . pprNatCmmTop) alloced)
372 if dopt Opt_D_dump_asm_stats dflags
373 then Just (catMaybes regAllocStats) else Nothing
375 -- force evaluation of the Maybe to avoid space leak
376 mPprStats `seq` return ()
378 return ( alloced, usAlloc
382 ---- shortcut branches
384 {-# SCC "shortcutBranches" #-}
385 shortcutBranches dflags alloced
389 {-# SCC "sequenceBlocks" #-}
390 map sequenceTop shorted
395 {-# SCC "x86fp_kludge" #-}
396 map x86fp_kludge sequenced
401 ---- expansion of SPARC synthetic instrs
402 #if sparc_TARGET_ARCH
404 {-# SCC "sparc_expand" #-}
405 map SPARC.expandTop kludged
408 Opt_D_dump_asm_expanded "Synthetic instructions expanded"
409 (vcat $ map (docToSDoc . pprNatCmmTop) expanded)
417 , lastMinuteImports ++ imports
423 x86fp_kludge :: NatCmmTop Instr -> NatCmmTop Instr
424 x86fp_kludge top@(CmmData _ _) = top
425 x86fp_kludge top@(CmmProc info lbl params (ListGraph code)) =
426 CmmProc info lbl params (ListGraph $ i386_insert_ffrees code)
430 -- | Build a doc for all the imports.
432 makeImportsDoc :: DynFlags -> [CLabel] -> Pretty.Doc
433 makeImportsDoc dflags imports
436 #if HAVE_SUBSECTIONS_VIA_SYMBOLS
437 -- On recent versions of Darwin, the linker supports
438 -- dead-stripping of code and data on a per-symbol basis.
439 -- There's a hack to make this work in PprMach.pprNatCmmTop.
440 Pretty.$$ Pretty.text ".subsections_via_symbols"
442 #if HAVE_GNU_NONEXEC_STACK
443 -- On recent GNU ELF systems one can mark an object file
444 -- as not requiring an executable stack. If all objects
445 -- linked into a program have this note then the program
446 -- will not use an executable stack, which is good for
447 -- security. GHC generated code does not need an executable
448 -- stack so add the note in:
449 Pretty.$$ Pretty.text ".section .note.GNU-stack,\"\",@progbits"
451 #if !defined(darwin_TARGET_OS)
452 -- And just because every other compiler does, lets stick in
453 -- an identifier directive: .ident "GHC x.y.z"
454 Pretty.$$ let compilerIdent = Pretty.text "GHC" Pretty.<+>
455 Pretty.text cProjectVersion
456 in Pretty.text ".ident" Pretty.<+>
457 Pretty.doubleQuotes compilerIdent
461 -- Generate "symbol stubs" for all external symbols that might
462 -- come from a dynamic library.
463 dyld_stubs :: [CLabel] -> Pretty.Doc
464 {- dyld_stubs imps = Pretty.vcat $ map pprDyldSymbolStub $
465 map head $ group $ sort imps-}
467 arch = platformArch $ targetPlatform dflags
468 os = platformOS $ targetPlatform dflags
470 -- (Hack) sometimes two Labels pretty-print the same, but have
471 -- different uniques; so we compare their text versions...
473 | needImportedSymbols arch os
475 (pprGotDeclaration arch os :) $
476 map ( pprImportedSymbol arch os . fst . head) $
477 groupBy (\(_,a) (_,b) -> a == b) $
478 sortBy (\(_,a) (_,b) -> compare a b) $
484 doPpr lbl = (lbl, Pretty.render $ pprCLabel lbl astyle)
485 astyle = mkCodeStyle AsmStyle
488 -- -----------------------------------------------------------------------------
489 -- Sequencing the basic blocks
491 -- Cmm BasicBlocks are self-contained entities: they always end in a
492 -- jump, either non-local or to another basic block in the same proc.
493 -- In this phase, we attempt to place the basic blocks in a sequence
494 -- such that as many of the local jumps as possible turn into
501 sequenceTop top@(CmmData _ _) = top
502 sequenceTop (CmmProc info lbl params (ListGraph blocks)) =
503 CmmProc info lbl params (ListGraph $ makeFarBranches $ sequenceBlocks blocks)
505 -- The algorithm is very simple (and stupid): we make a graph out of
506 -- the blocks where there is an edge from one block to another iff the
507 -- first block ends by jumping to the second. Then we topologically
508 -- sort this graph. Then traverse the list: for each block, we first
509 -- output the block, then if it has an out edge, we move the
510 -- destination of the out edge to the front of the list, and continue.
512 -- FYI, the classic layout for basic blocks uses postorder DFS; this
513 -- algorithm is implemented in cmm/ZipCfg.hs (NR 6 Sep 2007).
517 => [NatBasicBlock instr]
518 -> [NatBasicBlock instr]
520 sequenceBlocks [] = []
521 sequenceBlocks (entry:blocks) =
522 seqBlocks (mkNode entry : reverse (flattenSCCs (sccBlocks blocks)))
523 -- the first block is the entry point ==> it must remain at the start.
528 => [NatBasicBlock instr]
529 -> [SCC ( NatBasicBlock instr
533 sccBlocks blocks = stronglyConnCompFromEdgedVerticesR (map mkNode blocks)
535 -- we're only interested in the last instruction of
536 -- the block, and only if it has a single destination.
539 => [instr] -> [Unique]
542 = case jumpDestsOfInstr (last instrs) of
543 [one] -> [getUnique one]
546 mkNode block@(BasicBlock id instrs) = (block, getUnique id, getOutEdges instrs)
549 seqBlocks ((block,_,[]) : rest)
550 = block : seqBlocks rest
551 seqBlocks ((block@(BasicBlock id instrs),_,[next]) : rest)
552 | can_fallthrough = BasicBlock id (init instrs) : seqBlocks rest'
553 | otherwise = block : seqBlocks rest'
555 (can_fallthrough, rest') = reorder next [] rest
556 -- TODO: we should do a better job for cycles; try to maximise the
557 -- fallthroughs within a loop.
558 seqBlocks _ = panic "AsmCodegen:seqBlocks"
560 reorder id accum [] = (False, reverse accum)
561 reorder id accum (b@(block,id',out) : rest)
562 | id == id' = (True, (block,id,out) : reverse accum ++ rest)
563 | otherwise = reorder id (b:accum) rest
566 -- -----------------------------------------------------------------------------
567 -- Making far branches
569 -- Conditional branches on PowerPC are limited to +-32KB; if our Procs get too
570 -- big, we have to work around this limitation.
573 :: [NatBasicBlock Instr]
574 -> [NatBasicBlock Instr]
576 #if powerpc_TARGET_ARCH
577 makeFarBranches blocks
578 | last blockAddresses < nearLimit = blocks
579 | otherwise = zipWith handleBlock blockAddresses blocks
581 blockAddresses = scanl (+) 0 $ map blockLen blocks
582 blockLen (BasicBlock _ instrs) = length instrs
584 handleBlock addr (BasicBlock id instrs)
585 = BasicBlock id (zipWith makeFar [addr..] instrs)
587 makeFar addr (BCC ALWAYS tgt) = BCC ALWAYS tgt
588 makeFar addr (BCC cond tgt)
589 | abs (addr - targetAddr) >= nearLimit
593 where Just targetAddr = lookupUFM blockAddressMap tgt
594 makeFar addr other = other
596 nearLimit = 7000 -- 8192 instructions are allowed; let's keep some
597 -- distance, as we have a few pseudo-insns that are
598 -- pretty-printed as multiple instructions,
599 -- and it's just not worth the effort to calculate
602 blockAddressMap = listToUFM $ zip (map blockId blocks) blockAddresses
607 -- -----------------------------------------------------------------------------
615 shortcutBranches dflags tops
616 | optLevel dflags < 1 = tops -- only with -O or higher
617 | otherwise = map (apply_mapping mapping) tops'
619 (tops', mappings) = mapAndUnzip build_mapping tops
620 mapping = foldr plusUFM emptyUFM mappings
622 build_mapping top@(CmmData _ _) = (top, emptyUFM)
623 build_mapping (CmmProc info lbl params (ListGraph []))
624 = (CmmProc info lbl params (ListGraph []), emptyUFM)
625 build_mapping (CmmProc info lbl params (ListGraph (head:blocks)))
626 = (CmmProc info lbl params (ListGraph (head:others)), mapping)
627 -- drop the shorted blocks, but don't ever drop the first one,
628 -- because it is pointed to by a global label.
630 -- find all the blocks that just consist of a jump that can be
632 -- Don't completely eliminate loops here -- that can leave a dangling jump!
633 (_, shortcut_blocks, others) = foldl split (emptyBlockSet, [], []) blocks
634 split (s, shortcut_blocks, others) b@(BasicBlock id [insn])
635 | Just (DestBlockId dest) <- canShortcut insn,
636 (elemBlockSet dest s) || dest == id -- loop checks
637 = (s, shortcut_blocks, b : others)
638 split (s, shortcut_blocks, others) (BasicBlock id [insn])
639 | Just dest <- canShortcut insn
640 = (extendBlockSet s id, (id,dest) : shortcut_blocks, others)
641 split (s, shortcut_blocks, others) other = (s, shortcut_blocks, other : others)
644 -- build a mapping from BlockId to JumpDest for shorting branches
645 mapping = foldl add emptyUFM shortcut_blocks
646 add ufm (id,dest) = addToUFM ufm id dest
648 apply_mapping ufm (CmmData sec statics)
649 = CmmData sec (map (shortcutStatic (lookupUFM ufm)) statics)
650 -- we need to get the jump tables, so apply the mapping to the entries
652 apply_mapping ufm (CmmProc info lbl params (ListGraph blocks))
653 = CmmProc info lbl params (ListGraph $ map short_bb blocks)
655 short_bb (BasicBlock id insns) = BasicBlock id $! map short_insn insns
656 short_insn i = shortcutJump (lookupUFM ufm) i
657 -- shortcutJump should apply the mapping repeatedly,
658 -- just in case we can short multiple branches.
660 -- -----------------------------------------------------------------------------
661 -- Instruction selection
663 -- Native code instruction selection for a chunk of stix code. For
664 -- this part of the computation, we switch from the UniqSM monad to
665 -- the NatM monad. The latter carries not only a Unique, but also an
666 -- Int denoting the current C stack pointer offset in the generated
667 -- code; this is needed for creating correct spill offsets on
668 -- architectures which don't offer, or for which it would be
669 -- prohibitively expensive to employ, a frame pointer register. Viz,
672 -- The offset is measured in bytes, and indicates the difference
673 -- between the current (simulated) C stack-ptr and the value it was at
674 -- the beginning of the block. For stacks which grow down, this value
675 -- should be either zero or negative.
677 -- Switching between the two monads whilst carrying along the same
678 -- Unique supply breaks abstraction. Is that bad?
687 genMachCode dflags cmm_top
688 = do { initial_us <- getUs
689 ; let initial_st = mkNatM_State initial_us 0 dflags
690 (new_tops, final_st) = initNat initial_st (cmmTopCodeGen dflags cmm_top)
691 final_delta = natm_delta final_st
692 final_imports = natm_imports final_st
693 ; if final_delta == 0
694 then return (new_tops, final_imports)
695 else pprPanic "genMachCode: nonzero final delta" (int final_delta)
699 -- -----------------------------------------------------------------------------
700 -- Generic Cmm optimiser
706 (b) Simple inlining: a temporary which is assigned to and then
707 used, once, can be shorted.
708 (c) Position independent code and dynamic linking
709 (i) introduce the appropriate indirections
710 and position independent refs
711 (ii) compile a list of imported symbols
713 Ideas for other things we could do (ToDo):
715 - shortcut jumps-to-jumps
716 - eliminate dead code blocks
717 - simple CSE: if an expr is assigned to a temp, then replace later occs of
718 that expr with the temp, until the expr is no longer valid (can push through
719 temp assignments, and certain assigns to mem...)
722 cmmToCmm :: DynFlags -> RawCmmTop -> (RawCmmTop, [CLabel])
723 cmmToCmm _ top@(CmmData _ _) = (top, [])
724 cmmToCmm dflags (CmmProc info lbl params (ListGraph blocks)) = runCmmOpt dflags $ do
725 blocks' <- mapM cmmBlockConFold (cmmMiniInline blocks)
726 return $ CmmProc info lbl params (ListGraph blocks')
728 newtype CmmOptM a = CmmOptM (([CLabel], DynFlags) -> (# a, [CLabel] #))
730 instance Monad CmmOptM where
731 return x = CmmOptM $ \(imports, _) -> (# x,imports #)
733 CmmOptM $ \(imports, dflags) ->
734 case f (imports, dflags) of
737 CmmOptM g' -> g' (imports', dflags)
739 addImportCmmOpt :: CLabel -> CmmOptM ()
740 addImportCmmOpt lbl = CmmOptM $ \(imports, dflags) -> (# (), lbl:imports #)
742 getDynFlagsCmmOpt :: CmmOptM DynFlags
743 getDynFlagsCmmOpt = CmmOptM $ \(imports, dflags) -> (# dflags, imports #)
745 runCmmOpt :: DynFlags -> CmmOptM a -> (a, [CLabel])
746 runCmmOpt dflags (CmmOptM f) = case f ([], dflags) of
747 (# result, imports #) -> (result, imports)
749 cmmBlockConFold :: CmmBasicBlock -> CmmOptM CmmBasicBlock
750 cmmBlockConFold (BasicBlock id stmts) = do
751 stmts' <- mapM cmmStmtConFold stmts
752 return $ BasicBlock id stmts'
757 -> do src' <- cmmExprConFold DataReference src
758 return $ case src' of
759 CmmReg reg' | reg == reg' -> CmmNop
760 new_src -> CmmAssign reg new_src
763 -> do addr' <- cmmExprConFold DataReference addr
764 src' <- cmmExprConFold DataReference src
765 return $ CmmStore addr' src'
768 -> do addr' <- cmmExprConFold JumpReference addr
769 return $ CmmJump addr' regs
771 CmmCall target regs args srt returns
772 -> do target' <- case target of
773 CmmCallee e conv -> do
774 e' <- cmmExprConFold CallReference e
775 return $ CmmCallee e' conv
776 other -> return other
777 args' <- mapM (\(CmmHinted arg hint) -> do
778 arg' <- cmmExprConFold DataReference arg
779 return (CmmHinted arg' hint)) args
780 return $ CmmCall target' regs args' srt returns
782 CmmCondBranch test dest
783 -> do test' <- cmmExprConFold DataReference test
784 return $ case test' of
785 CmmLit (CmmInt 0 _) ->
786 CmmComment (mkFastString ("deleted: " ++
787 showSDoc (pprStmt stmt)))
789 CmmLit (CmmInt n _) -> CmmBranch dest
790 other -> CmmCondBranch test' dest
793 -> do expr' <- cmmExprConFold DataReference expr
794 return $ CmmSwitch expr' ids
800 cmmExprConFold referenceKind expr
803 -> do addr' <- cmmExprConFold DataReference addr
804 return $ CmmLoad addr' rep
807 -- For MachOps, we first optimize the children, and then we try
808 -- our hand at some constant-folding.
809 -> do args' <- mapM (cmmExprConFold DataReference) args
810 return $ cmmMachOpFold mop args'
812 CmmLit (CmmLabel lbl)
814 dflags <- getDynFlagsCmmOpt
815 cmmMakeDynamicReference dflags addImportCmmOpt referenceKind lbl
816 CmmLit (CmmLabelOff lbl off)
818 dflags <- getDynFlagsCmmOpt
819 dynRef <- cmmMakeDynamicReference dflags addImportCmmOpt referenceKind lbl
820 return $ cmmMachOpFold (MO_Add wordWidth) [
822 (CmmLit $ CmmInt (fromIntegral off) wordWidth)
825 #if powerpc_TARGET_ARCH
826 -- On powerpc (non-PIC), it's easier to jump directly to a label than
827 -- to use the register table, so we replace these registers
828 -- with the corresponding labels:
829 CmmReg (CmmGlobal EagerBlackholeInfo)
831 -> cmmExprConFold referenceKind $
832 CmmLit (CmmLabel (mkCmmCodeLabel rtsPackageId (fsLit "__stg_EAGER_BLACKHOLE_info")))
833 CmmReg (CmmGlobal GCEnter1)
835 -> cmmExprConFold referenceKind $
836 CmmLit (CmmLabel (mkCmmCodeLabel rtsPackageId (fsLit "__stg_gc_enter_1")))
837 CmmReg (CmmGlobal GCFun)
839 -> cmmExprConFold referenceKind $
840 CmmLit (CmmLabel (mkCmmCodeLabel rtsPackageId (fsLit "__stg_gc_fun")))