2 % (c) The University of Glasgow 2000
4 \section[ByteCodeGen]{Generate bytecode from Core}
7 module ByteCodeGen ( UnlinkedBCO, UnlinkedBCOExpr, ItblEnv, ClosureEnv, HValue,
9 byteCodeGen, coreExprToBCOs
12 #include "HsVersions.h"
15 import Name ( Name, getName )
16 import Id ( Id, idType, isDataConId_maybe, isPrimOpId_maybe, isFCallId,
17 idPrimRep, mkSysLocal, idName, isFCallId_maybe, isPrimOpId )
18 import ForeignCall ( ForeignCall(..), CCallTarget(..), CCallSpec(..) )
19 import OrdList ( OrdList, consOL, snocOL, appOL, unitOL,
20 nilOL, toOL, concatOL, fromOL )
21 import FiniteMap ( FiniteMap, addListToFM, listToFM, elemFM,
22 addToFM, lookupFM, fmToList )
24 import PprCore ( pprCoreExpr )
25 import Literal ( Literal(..), literalPrimRep )
26 import PrimRep ( PrimRep(..) )
27 import PrimOp ( PrimOp(..) )
28 import CoreFVs ( freeVars )
29 import Type ( typePrimRep, isUnLiftedType, splitTyConApp_maybe, isTyVarTy )
30 import DataCon ( dataConTag, fIRST_TAG, dataConTyCon,
31 dataConWrapId, isUnboxedTupleCon )
32 import TyCon ( TyCon(..), tyConFamilySize, isDataTyCon, tyConDataCons,
33 isFunTyCon, isUnboxedTupleTyCon )
34 import Class ( Class, classTyCon )
35 import Type ( Type, repType, splitFunTys, dropForAlls )
36 import Util ( zipEqual, zipWith4Equal, naturalMergeSortLe, nOfThem,
37 isSingleton, lengthIs, notNull )
38 import DataCon ( dataConRepArity )
39 import Var ( isTyVar )
40 import VarSet ( VarSet, varSetElems )
41 import PrimRep ( isFollowableRep )
42 import CmdLineOpts ( DynFlags, DynFlag(..) )
43 import ErrUtils ( showPass, dumpIfSet_dyn )
44 import Unique ( mkPseudoUnique3 )
45 import FastString ( FastString(..), unpackFS )
46 import Panic ( GhcException(..) )
47 import PprType ( pprType )
48 import SMRep ( arrWordsHdrSize, arrPtrsHdrSize )
49 import Constants ( wORD_SIZE )
50 import ByteCodeInstr ( BCInstr(..), ProtoBCO(..), nameOfProtoBCO, bciStackUse )
51 import ByteCodeItbls ( ItblEnv, mkITbls )
52 import ByteCodeLink ( UnlinkedBCO, UnlinkedBCOExpr, assembleBCO,
53 ClosureEnv, HValue, filterNameMap, linkFail,
54 iNTERP_STACK_CHECK_THRESH )
55 import ByteCodeFFI ( taggedSizeW, untaggedSizeW, mkMarshalCode, moan64 )
56 import Linker ( lookupSymbol )
58 import List ( intersperse, sortBy, zip4 )
59 import Foreign ( Ptr, castPtr, mallocBytes, pokeByteOff, Word8 )
60 import CTypes ( CInt )
61 import Exception ( throwDyn )
63 import GlaExts ( Int(..), ByteArray# )
66 import Maybe ( isJust )
70 %************************************************************************
72 \subsection{Functions visible from outside this module.}
74 %************************************************************************
78 byteCodeGen :: DynFlags
81 -> IO ([UnlinkedBCO], ItblEnv)
82 byteCodeGen dflags binds local_tycons local_classes
83 = do showPass dflags "ByteCodeGen"
84 let tycs = local_tycons ++ map classTyCon local_classes
85 itblenv <- mkITbls tycs
87 let flatBinds = concatMap getBind binds
88 getBind (NonRec bndr rhs) = [(bndr, freeVars rhs)]
89 getBind (Rec binds) = [(bndr, freeVars rhs) | (bndr,rhs) <- binds]
91 (BcM_State proto_bcos final_ctr mallocd, ())
92 <- runBc (BcM_State [] 0 [])
93 (mapBc (schemeR True []) flatBinds `thenBc_` returnBc ())
95 -- better be no free vars in these top-level bindings
97 when (notNull mallocd)
98 (panic "ByteCodeGen.byteCodeGen: missing final emitBc?")
100 dumpIfSet_dyn dflags Opt_D_dump_BCOs
101 "Proto-bcos" (vcat (intersperse (char ' ') (map ppr proto_bcos)))
103 bcos <- mapM assembleBCO proto_bcos
105 return (bcos, itblenv)
108 -- Returns: (the root BCO for this expression,
109 -- a list of auxilary BCOs resulting from compiling closures)
110 coreExprToBCOs :: DynFlags
112 -> IO UnlinkedBCOExpr
113 coreExprToBCOs dflags expr
114 = do showPass dflags "ByteCodeGen"
116 -- create a totally bogus name for the top-level BCO; this
117 -- should be harmless, since it's never used for anything
118 let invented_id = mkSysLocal FSLIT("ExprTopLevel")
120 (panic "invented_id's type")
121 let invented_name = idName invented_id
123 annexpr = freeVars expr
124 fvs = filter (not.isTyVar) (varSetElems (fst annexpr))
126 (BcM_State all_proto_bcos final_ctr mallocd, ())
127 <- runBc (BcM_State [] 0 [])
128 (schemeR True fvs (invented_id, annexpr))
130 when (notNull mallocd)
131 (panic "ByteCodeGen.coreExprToBCOs: missing final emitBc?")
133 dumpIfSet_dyn dflags Opt_D_dump_BCOs
134 "Proto-bcos" (vcat (intersperse (char ' ') (map ppr all_proto_bcos)))
137 = case filter ((== invented_name).nameOfProtoBCO) all_proto_bcos of
138 [root_bco] -> root_bco
140 = filter ((/= invented_name).nameOfProtoBCO) all_proto_bcos
142 auxiliary_bcos <- mapM assembleBCO auxiliary_proto_bcos
143 root_bco <- assembleBCO root_proto_bco
145 return (root_bco, auxiliary_bcos)
148 %************************************************************************
150 \subsection{Compilation schema for the bytecode generator.}
152 %************************************************************************
156 type BCInstrList = OrdList BCInstr
158 type Sequel = Int -- back off to this depth before ENTER
160 -- Maps Ids to the offset from the stack _base_ so we don't have
161 -- to mess with it after each push/pop.
162 type BCEnv = FiniteMap Id Int -- To find vars on the stack
164 ppBCEnv :: BCEnv -> SDoc
167 $$ nest 4 (vcat (map pp_one (sortBy cmp_snd (fmToList p))))
170 pp_one (var, offset) = int offset <> colon <+> ppr var
171 cmp_snd x y = compare (snd x) (snd y)
173 -- Create a BCO and do a spot of peephole optimisation on the insns
175 mkProtoBCO nm instrs_ordlist origin mallocd_blocks
176 = ProtoBCO nm maybe_with_stack_check origin mallocd_blocks
178 -- Overestimate the stack usage (in words) of this BCO,
179 -- and if >= iNTERP_STACK_CHECK_THRESH, add an explicit
180 -- stack check. (The interpreter always does a stack check
181 -- for iNTERP_STACK_CHECK_THRESH words at the start of each
182 -- BCO anyway, so we only need to add an explicit on in the
183 -- (hopefully rare) cases when the (overestimated) stack use
184 -- exceeds iNTERP_STACK_CHECK_THRESH.
185 maybe_with_stack_check
186 | stack_overest >= 65535
187 = pprPanic "mkProtoBCO: stack use won't fit in 16 bits"
189 | stack_overest >= iNTERP_STACK_CHECK_THRESH
190 = (STKCHECK stack_overest) : peep_d
192 = peep_d -- the supposedly common case
194 stack_overest = sum (map bciStackUse peep_d)
195 + 10 {- just to be really really sure -}
198 -- Merge local pushes
199 peep_d = peep (fromOL instrs_ordlist)
201 peep (PUSH_L off1 : PUSH_L off2 : PUSH_L off3 : rest)
202 = PUSH_LLL off1 (off2-1) (off3-2) : peep rest
203 peep (PUSH_L off1 : PUSH_L off2 : rest)
204 = PUSH_LL off1 (off2-1) : peep rest
211 -- Compile code for the right hand side of a let binding.
212 -- Park the resulting BCO in the monad. Also requires the
213 -- variable to which this value was bound, so as to give the
214 -- resulting BCO a name. Bool indicates top-levelness.
216 schemeR :: Bool -> [Id] -> (Id, AnnExpr Id VarSet) -> BcM ()
217 schemeR is_top fvs (nm, rhs)
221 $$ (ppr.filter (not.isTyVar).varSetElems.fst) rhs
222 $$ pprCoreExpr (deAnnotate rhs)
228 = schemeR_wrk is_top fvs rhs nm (collect [] rhs)
231 collect xs (_, AnnNote note e)
233 collect xs (_, AnnLam x e)
234 = collect (if isTyVar x then xs else (x:xs)) e
235 collect xs not_lambda
236 = (reverse xs, not_lambda)
238 schemeR_wrk is_top fvs original_body nm (args, body)
239 | Just dcon <- maybe_toplevel_null_con_rhs
240 = --trace ("nullary constructor! " ++ showSDocDebug (ppr nm)) (
241 emitBc (mkProtoBCO (getName nm) (toOL [PACK dcon 0, ENTER])
242 (Right original_body))
246 = let all_args = reverse args ++ fvs
247 szsw_args = map taggedIdSizeW all_args
248 szw_args = sum szsw_args
249 p_init = listToFM (zip all_args (mkStackOffsets 0 szsw_args))
250 argcheck = unitOL (ARGCHECK szw_args)
252 schemeE szw_args 0 p_init body `thenBc` \ body_code ->
253 emitBc (mkProtoBCO (getName nm) (appOL argcheck body_code)
254 (Right original_body))
257 maybe_toplevel_null_con_rhs
258 | is_top && null args
259 = case nukeTyArgs (snd body) of
261 -> case isDataConId_maybe v_wrk of
263 Just dc_wrk | nm == dataConWrapId dc_wrk
271 nukeTyArgs (AnnApp f (_, AnnType _)) = nukeTyArgs (snd f)
272 nukeTyArgs other = other
275 -- Let szsw be the sizes in words of some items pushed onto the stack,
276 -- which has initial depth d'. Return the values which the stack environment
277 -- should map these items to.
278 mkStackOffsets :: Int -> [Int] -> [Int]
279 mkStackOffsets original_depth szsw
280 = map (subtract 1) (tail (scanl (+) original_depth szsw))
282 -- Compile code to apply the given expression to the remaining args
283 -- on the stack, returning a HNF.
284 schemeE :: Int -> Sequel -> BCEnv -> AnnExpr Id VarSet -> BcM BCInstrList
286 -- Delegate tail-calls to schemeT.
287 schemeE d s p e@(fvs, AnnApp f a)
288 = schemeT d s p (fvs, AnnApp f a)
290 schemeE d s p e@(fvs, AnnVar v)
291 | not (isUnLiftedType v_type)
292 = -- Lifted-type thing; push it in the normal way
293 schemeT d s p (fvs, AnnVar v)
296 = -- Returning an unlifted value.
297 -- Heave it on the stack, SLIDE, and RETURN.
298 pushAtom True d p (AnnVar v) `thenBc` \ (push, szw) ->
299 returnBc (push -- value onto stack
300 `appOL` mkSLIDE szw (d-s) -- clear to sequel
301 `snocOL` RETURN v_rep) -- go
304 v_rep = typePrimRep v_type
306 schemeE d s p (fvs, AnnLit literal)
307 = pushAtom True d p (AnnLit literal) `thenBc` \ (push, szw) ->
308 let l_rep = literalPrimRep literal
309 in returnBc (push -- value onto stack
310 `appOL` mkSLIDE szw (d-s) -- clear to sequel
311 `snocOL` RETURN l_rep) -- go
315 Deal specially with the cases
316 let x = fn atom1 .. atomn in B
318 let x = Con atom1 .. atomn in B
319 (Con must be saturated)
321 In these cases, generate code to allocate in-line.
323 This is optimisation of the general case for let, which follows
324 this one; this case can safely be omitted. The reduction in
325 interpreter execution time seems to be around 5% for some programs,
326 with a similar drop in allocations.
328 This optimisation should be done more cleanly. As-is, it is
329 inapplicable to RHSs in letrecs, and needlessly duplicates code in
330 schemeR and schemeT. Some refactoring of the machinery would cure
333 schemeE d s p ee@(fvs, AnnLet (AnnNonRec x rhs) b)
335 = let d_init = if is_con then d else d'
337 mkPushes d_init args_r_to_l_reordered `thenBc` \ (d_final, push_code) ->
338 schemeE d' s p' b `thenBc` \ body_code ->
339 let size = d_final - d_init
340 alloc = if is_con then nilOL else unitOL (ALLOC size)
341 pack = unitOL (if is_con then PACK the_dcon size else MKAP size size)
343 returnBc (alloc `appOL` push_code `appOL` pack
346 -- Decide whether we can do this or not
347 (ok_to_go, is_con, the_dcon, the_fn)
349 Nothing -> (False, bomb 1, bomb 2, bomb 3)
350 Just (Left fn) -> (True, False, bomb 5, fn)
352 | dataConRepArity dcon <= length args_r_to_l
353 -> (True, True, dcon, bomb 6)
355 -> (False, bomb 7, bomb 8, bomb 9)
356 bomb n = panic ("schemeE.is_con(hacky hack hack) " ++ show n)
358 -- Extract the args (R -> L) and fn
359 args_r_to_l_reordered
363 = filter (not.isPtr.snd) args_r_to_l ++ filter (isPtr.snd) args_r_to_l
364 where isPtr = isFollowableRep . atomRep
366 args_r_to_l = filter (not.isTypeAtom.snd) args_r_to_l_raw
367 isTypeAtom (AnnType _) = True
370 (args_r_to_l_raw, maybe_fn) = chomp rhs
374 | isFCallId v || isPrimOpId v
377 -> case isDataConId_maybe v of
378 Just dcon -> ([], Just (Right dcon))
379 Nothing -> ([], Just (Left v))
380 AnnApp f a -> case chomp f of (az, f) -> (a:az, f)
381 AnnNote n e -> chomp e
382 other -> ([], Nothing)
384 -- This is the env in which to translate the body
388 -- Shove the args on the stack, including the fn in the non-dcon case
389 tag_when_push = not is_con
391 mkPushes :: Int{-curr depth-} -> [AnnExpr Id VarSet]
392 -> BcM (Int{-final depth-}, BCInstrList)
395 = returnBc (dd, nilOL)
397 = pushAtom False dd p' (AnnVar the_fn) `thenBc` \ (fn_push_code, fn_szw) ->
398 returnBc (dd+fn_szw, fn_push_code)
399 mkPushes dd (atom:atoms)
400 = pushAtom tag_when_push dd p' (snd atom)
401 `thenBc` \ (push1_code, push1_szw) ->
402 mkPushes (dd+push1_szw) atoms `thenBc` \ (dd_final, push_rest) ->
403 returnBc (dd_final, push1_code `appOL` push_rest)
406 -- General case for let. Generates correct, if inefficient, code in
408 schemeE d s p (fvs, AnnLet binds b)
409 = let (xs,rhss) = case binds of AnnNonRec x rhs -> ([x],[rhs])
410 AnnRec xs_n_rhss -> unzip xs_n_rhss
413 is_local id = not (isTyVar id) && elemFM id p'
414 fvss = map (filter is_local . varSetElems . fst) rhss
416 -- Sizes of tagged free vars, + 1 for the fn
417 sizes = map (\rhs_fvs -> 1 + sum (map taggedIdSizeW rhs_fvs)) fvss
419 -- This p', d' defn is safe because all the items being pushed
420 -- are ptrs, so all have size 1. d' and p' reflect the stack
421 -- after the closures have been allocated in the heap (but not
422 -- filled in), and pointers to them parked on the stack.
423 p' = addListToFM p (zipE xs (mkStackOffsets d (nOfThem n 1)))
426 infos = zipE4 fvss sizes xs [n, n-1 .. 1]
427 zipE = zipEqual "schemeE"
428 zipE4 = zipWith4Equal "schemeE" (\a b c d -> (a,b,c,d))
430 -- ToDo: don't build thunks for things with no free variables
431 buildThunk dd ([], size, id, off)
432 = returnBc (PUSH_G (Left (getName id))
433 `consOL` unitOL (MKAP (off+size-1) size))
434 buildThunk dd ((fv:fvs), size, id, off)
435 = pushAtom True dd p' (AnnVar fv)
436 `thenBc` \ (push_code, pushed_szw) ->
437 buildThunk (dd+pushed_szw) (fvs, size, id, off)
438 `thenBc` \ more_push_code ->
439 returnBc (push_code `appOL` more_push_code)
441 genThunkCode = mapBc (buildThunk d') infos `thenBc` \ tcodes ->
442 returnBc (concatOL tcodes)
444 allocCode = toOL (map ALLOC sizes)
446 schemeRs [] _ _ = returnBc ()
447 schemeRs (fvs:fvss) (x:xs) (rhs:rhss) =
448 schemeR False fvs (x,rhs) `thenBc_` schemeRs fvss xs rhss
450 schemeE d' s p' b `thenBc` \ bodyCode ->
451 schemeRs fvss xs rhss `thenBc_`
452 genThunkCode `thenBc` \ thunkCode ->
453 returnBc (allocCode `appOL` thunkCode `appOL` bodyCode)
459 schemeE d s p (fvs_case, AnnCase (fvs_scrut, scrut) bndr
460 [(DEFAULT, [], (fvs_rhs, rhs))])
462 | let isFunType var_type
463 = case splitTyConApp_maybe var_type of
464 Just (tycon,_) | isFunTyCon tycon -> True
466 ty_bndr = repType (idType bndr)
467 in isFunType ty_bndr || isTyVarTy ty_bndr
470 -- case scrut::suspect of bndr { DEFAULT -> rhs }
472 -- let bndr = scrut in rhs
473 -- when suspect is polymorphic or arrowtyped
474 -- So the required strictness properties are not observed.
475 -- At some point, must fix this properly.
479 (AnnNonRec bndr (fvs_scrut, scrut)) (fvs_rhs, rhs)
482 in trace ("WARNING: ignoring polymorphic case in interpreted mode.\n" ++
483 " Possibly due to strict polymorphic/functional constructor args.\n" ++
484 " Your program may leak space unexpectedly.\n")
485 (schemeE d s p new_expr)
489 schemeE d s p (fvs, AnnCase scrut bndr [(DataAlt dc, [bind1, bind2], rhs)])
490 | isUnboxedTupleCon dc && VoidRep == typePrimRep (idType bind1)
492 -- case .... of x { (# VoidRep'd-thing, a #) -> ... }
494 -- case .... of a { DEFAULT -> ... }
495 -- becuse the return convention for both are identical.
497 -- Note that it does not matter losing the void-rep thing from the
498 -- envt (it won't be bound now) because we never look such things up.
500 = --trace "automagic mashing of case alts (# VoidRep, a #)" (
501 schemeE d s p (fvs, AnnCase scrut bind2 [(DEFAULT, [], rhs)])
505 schemeE d s p (fvs, AnnCase scrut bndr [(DataAlt dc, [bind1], rhs)])
506 | isUnboxedTupleCon dc
507 -- Similarly, convert
508 -- case .... of x { (# a #) -> ... }
510 -- case .... of a { DEFAULT -> ... }
511 = --trace "automagic mashing of case alts (# a #)" (
512 schemeE d s p (fvs, AnnCase scrut bind1 [(DEFAULT, [], rhs)])
515 schemeE d s p (fvs, AnnCase scrut bndr alts)
517 -- Top of stack is the return itbl, as usual.
518 -- underneath it is the pointer to the alt_code BCO.
519 -- When an alt is entered, it assumes the returned value is
520 -- on top of the itbl.
523 -- Env and depth in which to compile the alts, not including
524 -- any vars bound by the alts themselves
525 d' = d + ret_frame_sizeW + taggedIdSizeW bndr
526 p' = addToFM p bndr (d' - 1)
528 scrut_primrep = typePrimRep (idType bndr)
530 | scrut_primrep == PtrRep
533 = WARN( scrut_primrep `elem` bad_reps,
534 text "Dire warning: strange rep in primitive case:" <+> ppr bndr )
535 -- We don't expect to see any of these
538 bad_reps = [CodePtrRep, DataPtrRep, RetRep, CostCentreRep]
540 -- given an alt, return a discr and code for it.
541 codeAlt alt@(discr, binds_f, rhs)
543 = let (unpack_code, d_after_unpack, p_after_unpack)
544 = mkUnpackCode (filter (not.isTyVar) binds_f) d' p'
545 in schemeE d_after_unpack s p_after_unpack rhs
546 `thenBc` \ rhs_code ->
547 returnBc (my_discr alt, unpack_code `appOL` rhs_code)
549 = ASSERT(null binds_f)
550 schemeE d' s p' rhs `thenBc` \ rhs_code ->
551 returnBc (my_discr alt, rhs_code)
553 my_discr (DEFAULT, binds, rhs) = NoDiscr
554 my_discr (DataAlt dc, binds, rhs)
555 | isUnboxedTupleCon dc
556 = unboxedTupleException
558 = DiscrP (dataConTag dc - fIRST_TAG)
559 my_discr (LitAlt l, binds, rhs)
560 = case l of MachInt i -> DiscrI (fromInteger i)
561 MachFloat r -> DiscrF (fromRational r)
562 MachDouble r -> DiscrD (fromRational r)
563 MachChar i -> DiscrI i
564 _ -> pprPanic "schemeE(AnnCase).my_discr" (ppr l)
567 | not isAlgCase = Nothing
569 = case [dc | (DataAlt dc, _, _) <- alts] of
571 (dc:_) -> Just (tyConFamilySize (dataConTyCon dc))
574 mapBc codeAlt alts `thenBc` \ alt_stuff ->
575 mkMultiBranch maybe_ncons alt_stuff `thenBc` \ alt_final ->
577 alt_final_ac = ARGCHECK (taggedIdSizeW bndr) `consOL` alt_final
578 alt_bco_name = getName bndr
579 alt_bco = mkProtoBCO alt_bco_name alt_final_ac (Left alts)
581 schemeE (d + ret_frame_sizeW)
582 (d + ret_frame_sizeW) p scrut `thenBc` \ scrut_code ->
584 emitBc alt_bco `thenBc_`
585 returnBc (PUSH_AS alt_bco_name scrut_primrep `consOL` scrut_code)
588 schemeE d s p (fvs, AnnNote note body)
592 = pprPanic "ByteCodeGen.schemeE: unhandled case"
593 (pprCoreExpr (deAnnotate other))
596 -- Compile code to do a tail call. Specifically, push the fn,
597 -- slide the on-stack app back down to the sequel depth,
598 -- and enter. Four cases:
601 -- An application "GHC.Prim.tagToEnum# <type> unboxed-int".
602 -- The int will be on the stack. Generate a code sequence
603 -- to convert it to the relevant constructor, SLIDE and ENTER.
605 -- 1. A nullary constructor. Push its closure on the stack
606 -- and SLIDE and RETURN.
608 -- 2. (Another nasty hack). Spot (# a::VoidRep, b #) and treat
609 -- it simply as b -- since the representations are identical
610 -- (the VoidRep takes up zero stack space). Also, spot
611 -- (# b #) and treat it as b.
613 -- 3. The fn denotes a ccall. Defer to generateCCall.
615 -- 4. Application of a non-nullary constructor, by defn saturated.
616 -- Split the args into ptrs and non-ptrs, and push the nonptrs,
617 -- then the ptrs, and then do PACK and RETURN.
619 -- 5. Otherwise, it must be a function call. Push the args
620 -- right to left, SLIDE and ENTER.
622 schemeT :: Int -- Stack depth
623 -> Sequel -- Sequel depth
624 -> BCEnv -- stack env
630 -- | trace ("schemeT: env in = \n" ++ showSDocDebug (ppBCEnv p)) False
631 -- = panic "schemeT ?!?!"
633 -- | trace ("\nschemeT\n" ++ showSDoc (pprCoreExpr (deAnnotate app)) ++ "\n") False
637 | Just (arg, constr_names) <- maybe_is_tagToEnum_call
638 = pushAtom True d p arg `thenBc` \ (push, arg_words) ->
639 implement_tagToId constr_names `thenBc` \ tagToId_sequence ->
640 returnBc (push `appOL` tagToId_sequence
641 `appOL` mkSLIDE 1 (d+arg_words-s)
645 | is_con_call && null args_r_to_l
647 (PUSH_G (Left (getName con)) `consOL` mkSLIDE 1 (d-s))
652 | [arg1,arg2] <- args_r_to_l,
654 isVoidRepAtom (_, AnnVar v) = typePrimRep (idType v) == VoidRep
655 isVoidRepAtom (_, AnnNote n e) = isVoidRepAtom e
656 isVoidRepAtom _ = False
659 = --trace (if isSingleton args_r_to_l
660 -- then "schemeT: unboxed singleton"
661 -- else "schemeT: unboxed pair with Void first component") (
666 | Just (CCall ccall_spec) <- isFCallId_maybe fn
667 = generateCCall d s p ccall_spec fn args_r_to_l
671 = if is_con_call && isUnboxedTupleCon con
672 then unboxedTupleException
673 else do_pushery d (map snd args_final_r_to_l)
676 -- Detect and extract relevant info for the tagToEnum kludge.
677 maybe_is_tagToEnum_call
678 = let extract_constr_Names ty
679 = case splitTyConApp_maybe (repType ty) of
680 (Just (tyc, [])) | isDataTyCon tyc
681 -> map getName (tyConDataCons tyc)
682 other -> panic "maybe_is_tagToEnum_call.extract_constr_Ids"
685 (_, AnnApp (_, AnnApp (_, AnnVar v) (_, AnnType t)) arg)
686 -> case isPrimOpId_maybe v of
687 Just TagToEnumOp -> Just (snd arg, extract_constr_Names t)
691 -- Extract the args (R->L) and fn
692 (args_r_to_l, fn) = chomp app
697 | isTypeAtom (snd a) -> chomp f
698 | otherwise -> case chomp f of (az, f) -> (a:az, f)
699 AnnNote n e -> chomp e
700 other -> pprPanic "schemeT"
701 (ppr (deAnnotate (panic "schemeT.chomp", other)))
703 n_args = length args_r_to_l
705 isTypeAtom (AnnType _) = True
708 -- decide if this is a constructor application, because we need
709 -- to rearrange the arguments on the stack if so. For building
710 -- a constructor, we put pointers before non-pointers and omit
713 -- Also if the constructor is not saturated, we just arrange to
714 -- call the curried worker instead.
716 maybe_dcon = case isDataConId_maybe fn of
717 Just con | dataConRepArity con == n_args -> Just con
719 is_con_call = isJust maybe_dcon
720 (Just con) = maybe_dcon
726 = filter (not.isPtr.snd) args_r_to_l ++ filter (isPtr.snd) args_r_to_l
727 where isPtr = isFollowableRep . atomRep
729 -- make code to push the args and then do the SLIDE-ENTER thing
730 tag_when_push = not is_con_call
731 narg_words = sum (map (get_arg_szw . atomRep . snd) args_r_to_l)
732 get_arg_szw = if tag_when_push then taggedSizeW else untaggedSizeW
734 do_pushery d (arg:args)
735 = pushAtom tag_when_push d p arg `thenBc` \ (push, arg_words) ->
736 do_pushery (d+arg_words) args `thenBc` \ more_push_code ->
737 returnBc (push `appOL` more_push_code)
739 | Just (CCall ccall_spec) <- isFCallId_maybe fn
740 = panic "schemeT.do_pushery: unexpected ccall"
743 Just con -> returnBc (
744 (PACK con narg_words `consOL`
745 mkSLIDE 1 (d - narg_words - s)) `snocOL`
749 -> pushAtom True d p (AnnVar fn)
750 `thenBc` \ (push, arg_words) ->
751 returnBc (push `appOL` mkSLIDE (narg_words+arg_words)
756 {- Deal with a CCall. Taggedly push the args onto the stack R->L,
757 deferencing ForeignObj#s and (ToDo: adjusting addrs to point to
758 payloads in Ptr/Byte arrays). Then, generate the marshalling
759 (machine) code for the ccall, and create bytecodes to call that and
760 then return in the right way.
762 generateCCall :: Int -> Sequel -- stack and sequel depths
764 -> CCallSpec -- where to call
765 -> Id -- of target, for type info
766 -> [AnnExpr Id VarSet] -- args (atoms)
769 generateCCall d0 s p ccall_spec@(CCallSpec target cconv safety) fn args_r_to_l
772 addr_usizeW = untaggedSizeW AddrRep
773 addr_tsizeW = taggedSizeW AddrRep
775 -- Get the args on the stack, with tags and suitably
776 -- dereferenced for the CCall. For each arg, return the
777 -- depth to the first word of the bits for that arg, and the
778 -- PrimRep of what was actually pushed.
780 pargs d [] = returnBc []
782 = let rep_arg = atomRep a
784 -- Don't push the FO; instead push the Addr# it
787 -> pushAtom False{-irrelevant-} d p a
788 `thenBc` \ (push_fo, _) ->
789 let foro_szW = taggedSizeW ForeignObjRep
790 d_now = d + addr_tsizeW
791 code = push_fo `appOL` toOL [
792 UPK_TAG addr_usizeW 0 0,
793 SLIDE addr_tsizeW foro_szW
795 in pargs d_now az `thenBc` \ rest ->
796 returnBc ((code, AddrRep) : rest)
799 -> pargs (d + addr_tsizeW) az `thenBc` \ rest ->
800 parg_ArrayishRep arrPtrsHdrSize d p a
802 returnBc ((code,AddrRep):rest)
805 -> pargs (d + addr_tsizeW) az `thenBc` \ rest ->
806 parg_ArrayishRep arrWordsHdrSize d p a
808 returnBc ((code,AddrRep):rest)
810 -- Default case: push taggedly, but otherwise intact.
812 -> pushAtom True d p a `thenBc` \ (code_a, sz_a) ->
813 pargs (d+sz_a) az `thenBc` \ rest ->
814 returnBc ((code_a, rep_arg) : rest)
816 -- Do magic for Ptr/Byte arrays. Push a ptr to the array on
817 -- the stack but then advance it over the headers, so as to
818 -- point to the payload.
819 parg_ArrayishRep hdrSizeW d p a
820 = pushAtom False{-irrel-} d p a `thenBc` \ (push_fo, _) ->
821 -- The ptr points at the header. Advance it over the
822 -- header and then pretend this is an Addr# (push a tag).
823 returnBc (push_fo `snocOL`
824 SWIZZLE 0 (hdrSizeW * untaggedSizeW PtrRep
827 PUSH_TAG addr_usizeW)
830 pargs d0 args_r_to_l `thenBc` \ code_n_reps ->
832 (pushs_arg, a_reps_pushed_r_to_l) = unzip code_n_reps
834 push_args = concatOL pushs_arg
835 d_after_args = d0 + sum (map taggedSizeW a_reps_pushed_r_to_l)
837 | null a_reps_pushed_r_to_l || head a_reps_pushed_r_to_l /= VoidRep
838 = panic "ByteCodeGen.generateCCall: missing or invalid World token?"
840 = reverse (tail a_reps_pushed_r_to_l)
842 -- Now: a_reps_pushed_RAW are the reps which are actually on the stack.
843 -- push_args is the code to do that.
844 -- d_after_args is the stack depth once the args are on.
846 -- Get the result rep.
847 (returns_void, r_rep)
848 = case maybe_getCCallReturnRep (idType fn) of
849 Nothing -> (True, VoidRep)
850 Just rr -> (False, rr)
852 Because the Haskell stack grows down, the a_reps refer to
853 lowest to highest addresses in that order. The args for the call
854 are on the stack. Now push an unboxed, tagged Addr# indicating
855 the C function to call. Then push a dummy placeholder for the
856 result. Finally, emit a CCALL insn with an offset pointing to the
857 Addr# just pushed, and a literal field holding the mallocville
858 address of the piece of marshalling code we generate.
859 So, just prior to the CCALL insn, the stack looks like this
860 (growing down, as usual):
865 Addr# address_of_C_fn
866 <placeholder-for-result#> (must be an unboxed type)
868 The interpreter then calls the marshall code mentioned
869 in the CCALL insn, passing it (& <placeholder-for-result#>),
870 that is, the addr of the topmost word in the stack.
871 When this returns, the placeholder will have been
872 filled in. The placeholder is slid down to the sequel
873 depth, and we RETURN.
875 This arrangement makes it simple to do f-i-dynamic since the Addr#
876 value is the first arg anyway. It also has the virtue that the
877 stack is GC-understandable at all times.
879 The marshalling code is generated specifically for this
880 call site, and so knows exactly the (Haskell) stack
881 offsets of the args, fn address and placeholder. It
882 copies the args to the C stack, calls the stacked addr,
883 and parks the result back in the placeholder. The interpreter
884 calls it as a normal C call, assuming it has a signature
885 void marshall_code ( StgWord* ptr_to_top_of_stack )
887 -- resolve static address
891 -> returnBc (False, panic "ByteCodeGen.generateCCall(dyn)")
893 -> let sym_to_find = unpackFS target in
894 ioToBc (lookupSymbol sym_to_find) `thenBc` \res ->
896 Just aa -> returnBc (True, aa)
897 Nothing -> ioToBc (linkFail "ByteCodeGen.generateCCall"
900 -> pprPanic "ByteCodeGen.generateCCall: casm" (ppr ccall_spec)
902 get_target_info `thenBc` \ (is_static, static_target_addr) ->
905 -- Get the arg reps, zapping the leading Addr# in the dynamic case
906 a_reps -- | trace (showSDoc (ppr a_reps_pushed_RAW)) False = error "???"
907 | is_static = a_reps_pushed_RAW
908 | otherwise = if null a_reps_pushed_RAW
909 then panic "ByteCodeGen.generateCCall: dyn with no args"
910 else tail a_reps_pushed_RAW
913 (push_Addr, d_after_Addr)
915 = (toOL [PUSH_UBX (Right static_target_addr) addr_usizeW,
916 PUSH_TAG addr_usizeW],
917 d_after_args + addr_tsizeW)
918 | otherwise -- is already on the stack
919 = (nilOL, d_after_args)
921 -- Push the return placeholder. For a call returning nothing,
922 -- this is a VoidRep (tag).
923 r_usizeW = untaggedSizeW r_rep
924 r_tsizeW = taggedSizeW r_rep
925 d_after_r = d_after_Addr + r_tsizeW
926 r_lit = mkDummyLiteral r_rep
927 push_r = (if returns_void
929 else unitOL (PUSH_UBX (Left r_lit) r_usizeW))
931 unitOL (PUSH_TAG r_usizeW)
933 -- generate the marshalling code we're going to call
936 arg1_offW = r_tsizeW + addr_tsizeW
937 args_offW = map (arg1_offW +)
938 (init (scanl (+) 0 (map taggedSizeW a_reps)))
940 ioToBc (mkMarshalCode cconv
941 (r_offW, r_rep) addr_offW
942 (zip args_offW a_reps)) `thenBc` \ addr_of_marshaller ->
943 recordMallocBc addr_of_marshaller `thenBc_`
946 do_call = unitOL (CCALL (castPtr addr_of_marshaller))
948 wrapup = mkSLIDE r_tsizeW (d_after_r - r_tsizeW - s)
949 `snocOL` RETURN r_rep
951 --trace (show (arg1_offW, args_offW , (map taggedSizeW a_reps) )) (
954 push_Addr `appOL` push_r `appOL` do_call `appOL` wrapup
959 -- Make a dummy literal, to be used as a placeholder for FFI return
960 -- values on the stack.
961 mkDummyLiteral :: PrimRep -> Literal
964 CharRep -> MachChar 0
966 WordRep -> MachWord 0
967 DoubleRep -> MachDouble 0
968 FloatRep -> MachFloat 0
969 AddrRep | taggedSizeW AddrRep == taggedSizeW WordRep -> MachWord 0
970 _ -> moan64 "mkDummyLiteral" (ppr pr)
974 -- GHC.Prim.Char# -> GHC.Prim.State# GHC.Prim.RealWorld
975 -- -> (# GHC.Prim.State# GHC.Prim.RealWorld, GHC.Prim.Int# #)
978 -- and check that an unboxed pair is returned wherein the first arg is VoidRep'd.
980 -- Alternatively, for call-targets returning nothing, convert
982 -- GHC.Prim.Char# -> GHC.Prim.State# GHC.Prim.RealWorld
983 -- -> (# GHC.Prim.State# GHC.Prim.RealWorld #)
987 maybe_getCCallReturnRep :: Type -> Maybe PrimRep
988 maybe_getCCallReturnRep fn_ty
989 = let (a_tys, r_ty) = splitFunTys (dropForAlls fn_ty)
991 = if isSingleton r_reps then Nothing else Just (r_reps !! 1)
993 = case splitTyConApp_maybe (repType r_ty) of
994 (Just (tyc, tys)) -> (tyc, map typePrimRep tys)
996 ok = ( ( r_reps `lengthIs` 2 && VoidRep == head r_reps)
997 || r_reps == [VoidRep] )
998 && isUnboxedTupleTyCon r_tycon
999 && case maybe_r_rep_to_go of
1001 Just r_rep -> r_rep /= PtrRep
1002 -- if it was, it would be impossible
1003 -- to create a valid return value
1004 -- placeholder on the stack
1005 blargh = pprPanic "maybe_getCCallReturn: can't handle:"
1008 --trace (showSDoc (ppr (a_reps, r_reps))) (
1009 if ok then maybe_r_rep_to_go else blargh
1012 atomRep (AnnVar v) = typePrimRep (idType v)
1013 atomRep (AnnLit l) = literalPrimRep l
1014 atomRep (AnnNote n b) = atomRep (snd b)
1015 atomRep (AnnApp f (_, AnnType _)) = atomRep (snd f)
1016 atomRep (AnnLam x e) | isTyVar x = atomRep (snd e)
1017 atomRep other = pprPanic "atomRep" (ppr (deAnnotate (undefined,other)))
1020 -- Compile code which expects an unboxed Int on the top of stack,
1021 -- (call it i), and pushes the i'th closure in the supplied list
1022 -- as a consequence.
1023 implement_tagToId :: [Name] -> BcM BCInstrList
1024 implement_tagToId names
1025 = ASSERT( notNull names )
1026 getLabelsBc (length names) `thenBc` \ labels ->
1027 getLabelBc `thenBc` \ label_fail ->
1028 getLabelBc `thenBc` \ label_exit ->
1029 zip4 labels (tail labels ++ [label_fail])
1030 [0 ..] names `bind` \ infos ->
1031 map (mkStep label_exit) infos `bind` \ steps ->
1032 returnBc (concatOL steps
1034 toOL [LABEL label_fail, CASEFAIL, LABEL label_exit])
1036 mkStep l_exit (my_label, next_label, n, name_for_n)
1037 = toOL [LABEL my_label,
1038 TESTEQ_I n next_label,
1039 PUSH_G (Left name_for_n),
1043 -- Make code to unpack the top-of-stack constructor onto the stack,
1044 -- adding tags for the unboxed bits. Takes the PrimReps of the
1045 -- constructor's arguments. off_h and off_s are travelling offsets
1046 -- along the constructor and the stack.
1048 -- Supposing a constructor in the heap has layout
1050 -- Itbl p_1 ... p_i np_1 ... np_j
1052 -- then we add to the stack, shown growing down, the following:
1064 -- so that in the common case (ptrs only) a single UNPACK instr can
1065 -- copy all the payload of the constr onto the stack with no further ado.
1067 mkUnpackCode :: [Id] -- constr args
1068 -> Int -- depth before unpack
1069 -> BCEnv -- env before unpack
1070 -> (BCInstrList, Int, BCEnv)
1071 mkUnpackCode vars d p
1072 = --trace ("mkUnpackCode: " ++ showSDocDebug (ppr vars)
1073 -- ++ " --> " ++ show d' ++ "\n" ++ showSDocDebug (ppBCEnv p')
1075 (code_p `appOL` code_np, d', p')
1079 vreps = [(var, typePrimRep (idType var)) | var <- vars]
1081 -- ptrs and nonptrs, forward
1082 vreps_p = filter (isFollowableRep.snd) vreps
1083 vreps_np = filter (not.isFollowableRep.snd) vreps
1085 -- the order in which we will augment the environment
1086 vreps_env = reverse vreps_p ++ reverse vreps_np
1088 -- new env and depth
1089 vreps_env_tszsw = map (taggedSizeW.snd) vreps_env
1090 p' = addListToFM p (zip (map fst vreps_env)
1091 (mkStackOffsets d vreps_env_tszsw))
1092 d' = d + sum vreps_env_tszsw
1094 -- code to unpack the ptrs
1095 ptrs_szw = sum (map (untaggedSizeW.snd) vreps_p)
1096 code_p | null vreps_p = nilOL
1097 | otherwise = unitOL (UNPACK ptrs_szw)
1099 -- code to unpack the nonptrs
1100 vreps_env_uszw = sum (map (untaggedSizeW.snd) vreps_env)
1101 code_np = do_nptrs vreps_env_uszw ptrs_szw (reverse (map snd vreps_np))
1102 do_nptrs off_h off_s [] = nilOL
1103 do_nptrs off_h off_s (npr:nprs)
1104 | npr `elem` [IntRep, WordRep, FloatRep, DoubleRep,
1105 CharRep, AddrRep, StablePtrRep]
1108 = moan64 "ByteCodeGen.mkUnpackCode" (ppr npr)
1110 approved = UPK_TAG usizeW (off_h-usizeW) off_s `consOL` theRest
1111 theRest = do_nptrs (off_h-usizeW) (off_s + tsizeW) nprs
1112 usizeW = untaggedSizeW npr
1113 tsizeW = taggedSizeW npr
1116 -- Push an atom onto the stack, returning suitable code & number of
1117 -- stack words used. Pushes it either tagged or untagged, since
1118 -- pushAtom is used to set up the stack prior to copying into the
1119 -- heap for both APs (requiring tags) and constructors (which don't).
1121 -- NB this means NO GC between pushing atoms for a constructor and
1122 -- copying them into the heap. It probably also means that
1123 -- tail calls MUST be of the form atom{atom ... atom} since if the
1124 -- expression head was allowed to be arbitrary, there could be GC
1125 -- in between pushing the arg atoms and completing the head.
1126 -- (not sure; perhaps the allocate/doYouWantToGC interface means this
1127 -- isn't a problem; but only if arbitrary graph construction for the
1128 -- head doesn't leave this BCO, since GC might happen at the start of
1129 -- each BCO (we consult doYouWantToGC there).
1131 -- Blargh. JRS 001206
1133 -- NB (further) that the env p must map each variable to the highest-
1134 -- numbered stack slot for it. For example, if the stack has depth 4
1135 -- and we tagged-ly push (v :: Int#) on it, the value will be in stack[4],
1136 -- the tag in stack[5], the stack will have depth 6, and p must map v to
1137 -- 5 and not to 4. Stack locations are numbered from zero, so a depth
1138 -- 6 stack has valid words 0 .. 5.
1140 pushAtom :: Bool -> Int -> BCEnv -> AnnExpr' Id VarSet -> BcM (BCInstrList, Int)
1141 pushAtom tagged d p (AnnVar v)
1143 | idPrimRep v == VoidRep
1144 = if tagged then returnBc (unitOL (PUSH_TAG 0), 1)
1145 else panic "ByteCodeGen.pushAtom(VoidRep,untaggedly)"
1148 = pprPanic "pushAtom: shouldn't get an FCallId here" (ppr v)
1150 | Just primop <- isPrimOpId_maybe v
1151 = returnBc (unitOL (PUSH_G (Right primop)), 1)
1155 str = "\npushAtom " ++ showSDocDebug (ppr v)
1156 ++ " :: " ++ showSDocDebug (pprType (idType v))
1157 ++ ", depth = " ++ show d
1158 ++ ", tagged = " ++ show tagged ++ ", env =\n" ++
1159 showSDocDebug (ppBCEnv p)
1160 ++ " --> words: " ++ show (snd result) ++ "\n" ++
1161 showSDoc (nest 4 (vcat (map ppr (fromOL (fst result)))))
1162 ++ "\nendPushAtom " ++ showSDocDebug (ppr v)
1166 = case lookupBCEnv_maybe p v of
1167 Just d_v -> (toOL (nOfThem nwords (PUSH_L (d-d_v+sz_t-2))), nwords)
1168 Nothing -> ASSERT(sz_t == 1) (unitOL (PUSH_G (Left nm)), nwords)
1170 nm = case isDataConId_maybe v of
1172 Nothing -> getName v
1174 sz_t = taggedIdSizeW v
1175 sz_u = untaggedIdSizeW v
1176 nwords = if tagged then sz_t else sz_u
1180 pushAtom True d p (AnnLit lit)
1181 = pushAtom False d p (AnnLit lit) `thenBc` \ (ubx_code, ubx_size) ->
1182 returnBc (ubx_code `snocOL` PUSH_TAG ubx_size, 1 + ubx_size)
1184 pushAtom False d p (AnnLit lit)
1186 MachLabel fs -> code CodePtrRep
1187 MachWord w -> code WordRep
1188 MachInt i -> code IntRep
1189 MachFloat r -> code FloatRep
1190 MachDouble r -> code DoubleRep
1191 MachChar c -> code CharRep
1192 MachStr s -> pushStr s
1195 = let size_host_words = untaggedSizeW rep
1196 in returnBc (unitOL (PUSH_UBX (Left lit) size_host_words),
1200 = let getMallocvilleAddr
1202 FastString _ l ba ->
1203 -- sigh, a string in the heap is no good to us.
1204 -- We need a static C pointer, since the type of
1205 -- a string literal is Addr#. So, copy the string
1206 -- into C land and introduce a memory leak
1207 -- at the same time.
1209 -- CAREFUL! Chars are 32 bits in ghc 4.09+
1210 in ioToBc (mallocBytes (n+1)) `thenBc` \ ptr ->
1211 recordMallocBc ptr `thenBc_`
1213 do memcpy ptr ba (fromIntegral n)
1214 pokeByteOff ptr n (fromIntegral (ord '\0') :: Word8)
1217 other -> panic "ByteCodeGen.pushAtom.pushStr"
1219 getMallocvilleAddr `thenBc` \ addr ->
1220 -- Get the addr on the stack, untaggedly
1221 returnBc (unitOL (PUSH_UBX (Right addr) 1), 1)
1227 pushAtom tagged d p (AnnApp f (_, AnnType _))
1228 = pushAtom tagged d p (snd f)
1230 pushAtom tagged d p (AnnNote note e)
1231 = pushAtom tagged d p (snd e)
1233 pushAtom tagged d p (AnnLam x e)
1235 = pushAtom tagged d p (snd e)
1237 pushAtom tagged d p other
1238 = pprPanic "ByteCodeGen.pushAtom"
1239 (pprCoreExpr (deAnnotate (undefined, other)))
1241 foreign import "memcpy" memcpy :: Ptr a -> ByteArray# -> CInt -> IO ()
1244 -- Given a bunch of alts code and their discrs, do the donkey work
1245 -- of making a multiway branch using a switch tree.
1246 -- What a load of hassle!
1247 mkMultiBranch :: Maybe Int -- # datacons in tycon, if alg alt
1248 -- a hint; generates better code
1249 -- Nothing is always safe
1250 -> [(Discr, BCInstrList)]
1252 mkMultiBranch maybe_ncons raw_ways
1253 = let d_way = filter (isNoDiscr.fst) raw_ways
1254 notd_ways = naturalMergeSortLe
1255 (\w1 w2 -> leAlt (fst w1) (fst w2))
1256 (filter (not.isNoDiscr.fst) raw_ways)
1258 mkTree :: [(Discr, BCInstrList)] -> Discr -> Discr -> BcM BCInstrList
1259 mkTree [] range_lo range_hi = returnBc the_default
1261 mkTree [val] range_lo range_hi
1262 | range_lo `eqAlt` range_hi
1263 = returnBc (snd val)
1265 = getLabelBc `thenBc` \ label_neq ->
1266 returnBc (mkTestEQ (fst val) label_neq
1268 `appOL` unitOL (LABEL label_neq)
1269 `appOL` the_default))
1271 mkTree vals range_lo range_hi
1272 = let n = length vals `div` 2
1273 vals_lo = take n vals
1274 vals_hi = drop n vals
1275 v_mid = fst (head vals_hi)
1277 getLabelBc `thenBc` \ label_geq ->
1278 mkTree vals_lo range_lo (dec v_mid) `thenBc` \ code_lo ->
1279 mkTree vals_hi v_mid range_hi `thenBc` \ code_hi ->
1280 returnBc (mkTestLT v_mid label_geq
1282 `appOL` unitOL (LABEL label_geq)
1286 = case d_way of [] -> unitOL CASEFAIL
1289 -- None of these will be needed if there are no non-default alts
1290 (mkTestLT, mkTestEQ, init_lo, init_hi)
1292 = panic "mkMultiBranch: awesome foursome"
1294 = case fst (head notd_ways) of {
1295 DiscrI _ -> ( \(DiscrI i) fail_label -> TESTLT_I i fail_label,
1296 \(DiscrI i) fail_label -> TESTEQ_I i fail_label,
1299 DiscrF _ -> ( \(DiscrF f) fail_label -> TESTLT_F f fail_label,
1300 \(DiscrF f) fail_label -> TESTEQ_F f fail_label,
1303 DiscrD _ -> ( \(DiscrD d) fail_label -> TESTLT_D d fail_label,
1304 \(DiscrD d) fail_label -> TESTEQ_D d fail_label,
1307 DiscrP _ -> ( \(DiscrP i) fail_label -> TESTLT_P i fail_label,
1308 \(DiscrP i) fail_label -> TESTEQ_P i fail_label,
1310 DiscrP algMaxBound )
1313 (algMinBound, algMaxBound)
1314 = case maybe_ncons of
1315 Just n -> (0, n - 1)
1316 Nothing -> (minBound, maxBound)
1318 (DiscrI i1) `eqAlt` (DiscrI i2) = i1 == i2
1319 (DiscrF f1) `eqAlt` (DiscrF f2) = f1 == f2
1320 (DiscrD d1) `eqAlt` (DiscrD d2) = d1 == d2
1321 (DiscrP i1) `eqAlt` (DiscrP i2) = i1 == i2
1322 NoDiscr `eqAlt` NoDiscr = True
1325 (DiscrI i1) `leAlt` (DiscrI i2) = i1 <= i2
1326 (DiscrF f1) `leAlt` (DiscrF f2) = f1 <= f2
1327 (DiscrD d1) `leAlt` (DiscrD d2) = d1 <= d2
1328 (DiscrP i1) `leAlt` (DiscrP i2) = i1 <= i2
1329 NoDiscr `leAlt` NoDiscr = True
1332 isNoDiscr NoDiscr = True
1335 dec (DiscrI i) = DiscrI (i-1)
1336 dec (DiscrP i) = DiscrP (i-1)
1337 dec other = other -- not really right, but if you
1338 -- do cases on floating values, you'll get what you deserve
1340 -- same snotty comment applies to the following
1342 minD, maxD :: Double
1348 mkTree notd_ways init_lo init_hi
1352 %************************************************************************
1354 \subsection{Supporting junk for the compilation schemes}
1356 %************************************************************************
1360 -- Describes case alts
1368 instance Outputable Discr where
1369 ppr (DiscrI i) = int i
1370 ppr (DiscrF f) = text (show f)
1371 ppr (DiscrD d) = text (show d)
1372 ppr (DiscrP i) = int i
1373 ppr NoDiscr = text "DEF"
1376 -- Find things in the BCEnv (the what's-on-the-stack-env)
1377 -- See comment preceding pushAtom for precise meaning of env contents
1378 --lookupBCEnv :: BCEnv -> Id -> Int
1379 --lookupBCEnv env nm
1380 -- = case lookupFM env nm of
1381 -- Nothing -> pprPanic "lookupBCEnv"
1382 -- (ppr nm $$ char ' ' $$ vcat (map ppr (fmToList env)))
1385 lookupBCEnv_maybe :: BCEnv -> Id -> Maybe Int
1386 lookupBCEnv_maybe = lookupFM
1389 taggedIdSizeW, untaggedIdSizeW :: Id -> Int
1390 taggedIdSizeW = taggedSizeW . typePrimRep . idType
1391 untaggedIdSizeW = untaggedSizeW . typePrimRep . idType
1393 unboxedTupleException :: a
1394 unboxedTupleException
1397 ("Bytecode generator can't handle unboxed tuples. Possibly due\n" ++
1398 "\tto foreign import/export decls in source. Workaround:\n" ++
1399 "\tcompile this module to a .o file, then restart session."))
1402 mkSLIDE n d = if d == 0 then nilOL else unitOL (SLIDE n d)
1407 %************************************************************************
1409 \subsection{The bytecode generator's monad}
1411 %************************************************************************
1415 = BcM_State { bcos :: [ProtoBCO Name], -- accumulates completed BCOs
1416 nextlabel :: Int, -- for generating local labels
1417 malloced :: [Ptr ()] } -- ptrs malloced for current BCO
1418 -- Should be free()d when it is GCd
1419 type BcM r = BcM_State -> IO (BcM_State, r)
1421 ioToBc :: IO a -> BcM a
1422 ioToBc io st = do x <- io
1425 runBc :: BcM_State -> BcM r -> IO (BcM_State, r)
1426 runBc st0 m = do (st1, res) <- m st0
1429 thenBc :: BcM a -> (a -> BcM b) -> BcM b
1430 thenBc expr cont st0
1431 = do (st1, q) <- expr st0
1432 (st2, r) <- cont q st1
1435 thenBc_ :: BcM a -> BcM b -> BcM b
1436 thenBc_ expr cont st0
1437 = do (st1, q) <- expr st0
1438 (st2, r) <- cont st1
1441 returnBc :: a -> BcM a
1442 returnBc result st = return (st, result)
1445 mapBc :: (a -> BcM b) -> [a] -> BcM [b]
1446 mapBc f [] = returnBc []
1448 = f x `thenBc` \ r ->
1449 mapBc f xs `thenBc` \ rs ->
1452 emitBc :: ([Ptr ()] -> ProtoBCO Name) -> BcM ()
1454 = return (st{bcos = bco (malloced st) : bcos st, malloced=[]}, ())
1458 | notNull (malloced st)
1459 = panic "ByteCodeGen.newbcoBc: missed prior emitBc?"
1463 recordMallocBc :: Ptr a -> BcM ()
1465 = return (st{malloced = castPtr a : malloced st}, ())
1467 getLabelBc :: BcM Int
1469 = return (st{nextlabel = 1 + nextlabel st}, nextlabel st)
1471 getLabelsBc :: Int -> BcM [Int]
1473 = let ctr = nextlabel st
1474 in return (st{nextlabel = ctr+n}, [ctr .. ctr+n-1])