2 % (c) The University of Glasgow 2002
4 \section[ByteCodeGen]{Generate bytecode from Core}
7 module ByteCodeGen ( UnlinkedBCO, byteCodeGen, coreExprToBCOs ) where
9 #include "HsVersions.h"
12 import ByteCodeFFI ( mkMarshalCode, moan64 )
13 import ByteCodeAsm ( CompiledByteCode(..), UnlinkedBCO,
14 assembleBCO, assembleBCOs, iNTERP_STACK_CHECK_THRESH )
15 import ByteCodeLink ( lookupStaticPtr )
18 import Name ( Name, getName, mkSystemName )
21 import ForeignCall ( ForeignCall(..), CCallTarget(..), CCallSpec(..) )
22 import HscTypes ( TypeEnv, typeEnvTyCons, typeEnvClasses )
23 import CoreUtils ( exprType )
25 import PprCore ( pprCoreExpr )
26 import Literal ( Literal(..), literalPrimRep )
28 import PrimOp ( PrimOp(..) )
29 import CoreFVs ( freeVars )
30 import Type ( typePrimRep, isUnLiftedType, splitTyConApp_maybe )
31 import DataCon ( DataCon, dataConTag, fIRST_TAG, dataConTyCon,
32 isUnboxedTupleCon, isNullaryDataCon, dataConWorkId,
34 import TyCon ( tyConFamilySize, isDataTyCon, tyConDataCons,
36 import Class ( Class, classTyCon )
37 import Type ( Type, repType, splitFunTys, dropForAlls, pprType )
39 import DataCon ( dataConRepArity )
40 import Var ( isTyVar )
41 import VarSet ( VarSet, varSetElems )
42 import TysPrim ( arrayPrimTyCon, mutableArrayPrimTyCon,
43 byteArrayPrimTyCon, mutableByteArrayPrimTyCon
45 import PrimRep ( isFollowableRep )
46 import CmdLineOpts ( DynFlags, DynFlag(..) )
47 import ErrUtils ( showPass, dumpIfSet_dyn )
48 import Unique ( mkPseudoUnique3 )
49 import FastString ( FastString(..), unpackFS )
50 import Panic ( GhcException(..) )
51 import SMRep ( arrWordsHdrSize, arrPtrsHdrSize, StgWord )
52 import Bitmap ( intsToReverseBitmap, mkBitmap )
54 import Constants ( wORD_SIZE )
56 import Data.List ( intersperse, sortBy, zip4, zip5, partition )
57 import Foreign ( Ptr, castPtr, mallocBytes, pokeByteOff, Word8 )
58 import Foreign.C ( CInt )
59 import Control.Exception ( throwDyn )
61 import GHC.Exts ( Int(..), ByteArray# )
63 import Control.Monad ( when )
64 import Data.Char ( ord, chr )
66 -- -----------------------------------------------------------------------------
67 -- Generating byte code for a complete module
69 byteCodeGen :: DynFlags
72 -> IO CompiledByteCode
73 byteCodeGen dflags binds type_env
74 = do showPass dflags "ByteCodeGen"
75 let local_tycons = typeEnvTyCons type_env
76 local_classes = typeEnvClasses type_env
77 tycs = local_tycons ++ map classTyCon local_classes
79 let flatBinds = [ (bndr, freeVars rhs)
80 | (bndr, rhs) <- flattenBinds binds]
82 (BcM_State final_ctr mallocd, proto_bcos)
83 <- runBc (mapM schemeTopBind flatBinds)
85 when (notNull mallocd)
86 (panic "ByteCodeGen.byteCodeGen: missing final emitBc?")
88 dumpIfSet_dyn dflags Opt_D_dump_BCOs
89 "Proto-BCOs" (vcat (intersperse (char ' ') (map ppr proto_bcos)))
91 assembleBCOs proto_bcos tycs
93 -- -----------------------------------------------------------------------------
94 -- Generating byte code for an expression
96 -- Returns: (the root BCO for this expression,
97 -- a list of auxilary BCOs resulting from compiling closures)
98 coreExprToBCOs :: DynFlags
101 coreExprToBCOs dflags expr
102 = do showPass dflags "ByteCodeGen"
104 -- create a totally bogus name for the top-level BCO; this
105 -- should be harmless, since it's never used for anything
106 let invented_name = mkSystemName (mkPseudoUnique3 0) FSLIT("ExprTopLevel")
107 invented_id = mkLocalId invented_name (panic "invented_id's type")
109 (BcM_State final_ctr mallocd, proto_bco)
110 <- runBc (schemeTopBind (invented_id, freeVars expr))
112 when (notNull mallocd)
113 (panic "ByteCodeGen.coreExprToBCOs: missing final emitBc?")
115 dumpIfSet_dyn dflags Opt_D_dump_BCOs "Proto-BCOs" (ppr proto_bco)
117 assembleBCO proto_bco
120 -- -----------------------------------------------------------------------------
121 -- Compilation schema for the bytecode generator
123 type BCInstrList = OrdList BCInstr
125 type Sequel = Int -- back off to this depth before ENTER
127 -- Maps Ids to the offset from the stack _base_ so we don't have
128 -- to mess with it after each push/pop.
129 type BCEnv = FiniteMap Id Int -- To find vars on the stack
131 ppBCEnv :: BCEnv -> SDoc
134 $$ nest 4 (vcat (map pp_one (sortBy cmp_snd (fmToList p))))
137 pp_one (var, offset) = int offset <> colon <+> ppr var <+> ppr (idPrimRep var)
138 cmp_snd x y = compare (snd x) (snd y)
140 -- Create a BCO and do a spot of peephole optimisation on the insns
145 -> Either [AnnAlt Id VarSet] (AnnExpr Id VarSet)
149 -> Bool -- True <=> is a return point, rather than a function
152 mkProtoBCO nm instrs_ordlist origin arity bitmap_size bitmap
153 is_ret mallocd_blocks
156 protoBCOInstrs = maybe_with_stack_check,
157 protoBCOBitmap = bitmap,
158 protoBCOBitmapSize = bitmap_size,
159 protoBCOArity = arity,
160 protoBCOExpr = origin,
161 protoBCOPtrs = mallocd_blocks
164 -- Overestimate the stack usage (in words) of this BCO,
165 -- and if >= iNTERP_STACK_CHECK_THRESH, add an explicit
166 -- stack check. (The interpreter always does a stack check
167 -- for iNTERP_STACK_CHECK_THRESH words at the start of each
168 -- BCO anyway, so we only need to add an explicit on in the
169 -- (hopefully rare) cases when the (overestimated) stack use
170 -- exceeds iNTERP_STACK_CHECK_THRESH.
171 maybe_with_stack_check
173 -- don't do stack checks at return points;
174 -- everything is aggregated up to the top BCO
175 -- (which must be a function)
176 | stack_overest >= 65535
177 = pprPanic "mkProtoBCO: stack use won't fit in 16 bits"
179 | stack_overest >= iNTERP_STACK_CHECK_THRESH
180 = STKCHECK stack_overest : peep_d
182 = peep_d -- the supposedly common case
184 stack_overest = sum (map bciStackUse peep_d)
186 -- Merge local pushes
187 peep_d = peep (fromOL instrs_ordlist)
189 peep (PUSH_L off1 : PUSH_L off2 : PUSH_L off3 : rest)
190 = PUSH_LLL off1 (off2-1) (off3-2) : peep rest
191 peep (PUSH_L off1 : PUSH_L off2 : rest)
192 = PUSH_LL off1 (off2-1) : peep rest
198 argBits :: [PrimRep] -> [Bool]
201 | isFollowableRep rep = False : argBits args
202 | otherwise = take (getPrimRepSize rep) (repeat True) ++ argBits args
204 -- -----------------------------------------------------------------------------
207 -- Compile code for the right-hand side of a top-level binding
209 schemeTopBind :: (Id, AnnExpr Id VarSet) -> BcM (ProtoBCO Name)
212 schemeTopBind (id, rhs)
213 | Just data_con <- isDataConWorkId_maybe id,
214 isNullaryDataCon data_con
215 = -- Special case for the worker of a nullary data con.
216 -- It'll look like this: Nil = /\a -> Nil a
217 -- If we feed it into schemeR, we'll get
219 -- because mkConAppCode treats nullary constructor applications
220 -- by just re-using the single top-level definition. So
221 -- for the worker itself, we must allocate it directly.
222 emitBc (mkProtoBCO (getName id) (toOL [PACK data_con 0, ENTER])
223 (Right rhs) 0 0 [{-no bitmap-}] False{-not alts-})
226 = schemeR [{- No free variables -}] (id, rhs)
228 -- -----------------------------------------------------------------------------
231 -- Compile code for a right-hand side, to give a BCO that,
232 -- when executed with the free variables and arguments on top of the stack,
233 -- will return with a pointer to the result on top of the stack, after
234 -- removing the free variables and arguments.
236 -- Park the resulting BCO in the monad. Also requires the
237 -- variable to which this value was bound, so as to give the
238 -- resulting BCO a name.
240 schemeR :: [Id] -- Free vars of the RHS, ordered as they
241 -- will appear in the thunk. Empty for
242 -- top-level things, which have no free vars.
243 -> (Id, AnnExpr Id VarSet)
244 -> BcM (ProtoBCO Name)
245 schemeR fvs (nm, rhs)
249 $$ (ppr.filter (not.isTyVar).varSetElems.fst) rhs
250 $$ pprCoreExpr (deAnnotate rhs)
256 = schemeR_wrk fvs nm rhs (collect [] rhs)
258 collect xs (_, AnnNote note e) = collect xs e
259 collect xs (_, AnnLam x e) = collect (if isTyVar x then xs else (x:xs)) e
260 collect xs (_, not_lambda) = (reverse xs, not_lambda)
262 schemeR_wrk fvs nm original_body (args, body)
264 all_args = reverse args ++ fvs
265 arity = length all_args
266 -- all_args are the args in reverse order. We're compiling a function
267 -- \fv1..fvn x1..xn -> e
268 -- i.e. the fvs come first
270 szsw_args = map idSizeW all_args
271 szw_args = sum szsw_args
272 p_init = listToFM (zip all_args (mkStackOffsets 0 szsw_args))
274 -- make the arg bitmap
275 bits = argBits (reverse (map idPrimRep all_args))
276 bitmap_size = length bits
277 bitmap = mkBitmap bits
279 schemeE szw_args 0 p_init body `thenBc` \ body_code ->
280 emitBc (mkProtoBCO (getName nm) body_code (Right original_body)
281 arity bitmap_size bitmap False{-not alts-})
284 fvsToEnv :: BCEnv -> VarSet -> [Id]
285 -- Takes the free variables of a right-hand side, and
286 -- delivers an ordered list of the local variables that will
287 -- be captured in the thunk for the RHS
288 -- The BCEnv argument tells which variables are in the local
289 -- environment: these are the ones that should be captured
291 -- The code that constructs the thunk, and the code that executes
292 -- it, have to agree about this layout
293 fvsToEnv p fvs = [v | v <- varSetElems fvs,
294 isId v, -- Could be a type variable
297 -- -----------------------------------------------------------------------------
300 -- Compile code to apply the given expression to the remaining args
301 -- on the stack, returning a HNF.
302 schemeE :: Int -> Sequel -> BCEnv -> AnnExpr' Id VarSet -> BcM BCInstrList
304 -- Delegate tail-calls to schemeT.
305 schemeE d s p e@(AnnApp f a)
308 schemeE d s p e@(AnnVar v)
309 | not (isUnLiftedType v_type)
310 = -- Lifted-type thing; push it in the normal way
314 = -- Returning an unlifted value.
315 -- Heave it on the stack, SLIDE, and RETURN.
316 pushAtom d p (AnnVar v) `thenBc` \ (push, szw) ->
317 returnBc (push -- value onto stack
318 `appOL` mkSLIDE szw (d-s) -- clear to sequel
319 `snocOL` RETURN_UBX v_rep) -- go
322 v_rep = typePrimRep v_type
324 schemeE d s p (AnnLit literal)
325 = pushAtom d p (AnnLit literal) `thenBc` \ (push, szw) ->
326 let l_rep = literalPrimRep literal
327 in returnBc (push -- value onto stack
328 `appOL` mkSLIDE szw (d-s) -- clear to sequel
329 `snocOL` RETURN_UBX l_rep) -- go
332 schemeE d s p (AnnLet (AnnNonRec x (_,rhs)) (_,body))
333 | (AnnVar v, args_r_to_l) <- splitApp rhs,
334 Just data_con <- isDataConWorkId_maybe v,
335 dataConRepArity data_con == length args_r_to_l
336 = -- Special case for a non-recursive let whose RHS is a
337 -- saturatred constructor application.
338 -- Just allocate the constructor and carry on
339 mkConAppCode d s p data_con args_r_to_l `thenBc` \ alloc_code ->
340 schemeE (d+1) s (addToFM p x d) body `thenBc` \ body_code ->
341 returnBc (alloc_code `appOL` body_code)
343 -- General case for let. Generates correct, if inefficient, code in
345 schemeE d s p (AnnLet binds (_,body))
346 = let (xs,rhss) = case binds of AnnNonRec x rhs -> ([x],[rhs])
347 AnnRec xs_n_rhss -> unzip xs_n_rhss
350 fvss = map (fvsToEnv p' . fst) rhss
352 -- Sizes of free vars
353 sizes = map (\rhs_fvs -> sum (map idSizeW rhs_fvs)) fvss
355 -- the arity of each rhs
356 arities = map (length . fst . collect []) rhss
358 -- This p', d' defn is safe because all the items being pushed
359 -- are ptrs, so all have size 1. d' and p' reflect the stack
360 -- after the closures have been allocated in the heap (but not
361 -- filled in), and pointers to them parked on the stack.
362 p' = addListToFM p (zipE xs (mkStackOffsets d (nOfThem n_binds 1)))
364 zipE = zipEqual "schemeE"
366 -- ToDo: don't build thunks for things with no free variables
367 build_thunk dd [] size bco off
368 = returnBc (PUSH_BCO bco
369 `consOL` unitOL (MKAP (off+size) size))
370 build_thunk dd (fv:fvs) size bco off = do
371 (push_code, pushed_szw) <- pushAtom dd p' (AnnVar fv)
372 more_push_code <- build_thunk (dd+pushed_szw) fvs size bco off
373 returnBc (push_code `appOL` more_push_code)
375 alloc_code = toOL (zipWith mkAlloc sizes arities)
376 where mkAlloc sz 0 = ALLOC_AP sz
377 mkAlloc sz arity = ALLOC_PAP arity sz
379 compile_bind d' fvs x rhs size off = do
380 bco <- schemeR fvs (x,rhs)
381 build_thunk d' fvs size bco off
384 [ compile_bind d' fvs x rhs size n
385 | (fvs, x, rhs, size, n) <-
386 zip5 fvss xs rhss sizes [n_binds, n_binds-1 .. 1]
389 body_code <- schemeE d' s p' body
390 thunk_codes <- sequence compile_binds
391 returnBc (alloc_code `appOL` concatOL thunk_codes `appOL` body_code)
395 schemeE d s p (AnnCase scrut bndr [(DataAlt dc, [bind1, bind2], rhs)])
396 | isUnboxedTupleCon dc && VoidRep == typePrimRep (idType bind1)
398 -- case .... of x { (# VoidRep'd-thing, a #) -> ... }
400 -- case .... of a { DEFAULT -> ... }
401 -- becuse the return convention for both are identical.
403 -- Note that it does not matter losing the void-rep thing from the
404 -- envt (it won't be bound now) because we never look such things up.
406 = --trace "automagic mashing of case alts (# VoidRep, a #)" $
407 doCase d s p scrut bind2 [(DEFAULT, [], rhs)] True{-unboxed tuple-}
409 | isUnboxedTupleCon dc && VoidRep == typePrimRep (idType bind2)
410 = --trace "automagic mashing of case alts (# a, VoidRep #)" $
411 doCase d s p scrut bind1 [(DEFAULT, [], rhs)] True{-unboxed tuple-}
413 schemeE d s p (AnnCase scrut bndr [(DataAlt dc, [bind1], rhs)])
414 | isUnboxedTupleCon dc
415 -- Similarly, convert
416 -- case .... of x { (# a #) -> ... }
418 -- case .... of a { DEFAULT -> ... }
419 = --trace "automagic mashing of case alts (# a #)" $
420 doCase d s p scrut bind1 [(DEFAULT, [], rhs)] True{-unboxed tuple-}
422 schemeE d s p (AnnCase scrut bndr alts)
423 = doCase d s p scrut bndr alts False{-not an unboxed tuple-}
425 schemeE d s p (AnnNote note (_, body))
429 = pprPanic "ByteCodeGen.schemeE: unhandled case"
430 (pprCoreExpr (deAnnotate' other))
433 -- Compile code to do a tail call. Specifically, push the fn,
434 -- slide the on-stack app back down to the sequel depth,
435 -- and enter. Four cases:
438 -- An application "GHC.Prim.tagToEnum# <type> unboxed-int".
439 -- The int will be on the stack. Generate a code sequence
440 -- to convert it to the relevant constructor, SLIDE and ENTER.
442 -- 1. The fn denotes a ccall. Defer to generateCCall.
444 -- 2. (Another nasty hack). Spot (# a::VoidRep, b #) and treat
445 -- it simply as b -- since the representations are identical
446 -- (the VoidRep takes up zero stack space). Also, spot
447 -- (# b #) and treat it as b.
449 -- 3. Application of a constructor, by defn saturated.
450 -- Split the args into ptrs and non-ptrs, and push the nonptrs,
451 -- then the ptrs, and then do PACK and RETURN.
453 -- 4. Otherwise, it must be a function call. Push the args
454 -- right to left, SLIDE and ENTER.
456 schemeT :: Int -- Stack depth
457 -> Sequel -- Sequel depth
458 -> BCEnv -- stack env
459 -> AnnExpr' Id VarSet
464 -- | trace ("schemeT: env in = \n" ++ showSDocDebug (ppBCEnv p)) False
465 -- = panic "schemeT ?!?!"
467 -- | trace ("\nschemeT\n" ++ showSDoc (pprCoreExpr (deAnnotate' app)) ++ "\n") False
471 | Just (arg, constr_names) <- maybe_is_tagToEnum_call
472 = pushAtom d p arg `thenBc` \ (push, arg_words) ->
473 implement_tagToId constr_names `thenBc` \ tagToId_sequence ->
474 returnBc (push `appOL` tagToId_sequence
475 `appOL` mkSLIDE 1 (d+arg_words-s)
479 | Just (CCall ccall_spec) <- isFCallId_maybe fn
480 = generateCCall d s p ccall_spec fn args_r_to_l
482 -- Case 2: Constructor application
483 | Just con <- maybe_saturated_dcon,
484 isUnboxedTupleCon con
485 = case args_r_to_l of
486 [arg1,arg2] | isVoidRepAtom arg1 ->
487 unboxedTupleReturn d s p arg2
488 [arg1,arg2] | isVoidRepAtom arg2 ->
489 unboxedTupleReturn d s p arg1
490 _other -> unboxedTupleException
492 -- Case 3: Ordinary data constructor
493 | Just con <- maybe_saturated_dcon
494 = mkConAppCode d s p con args_r_to_l `thenBc` \ alloc_con ->
495 returnBc (alloc_con `appOL`
496 mkSLIDE 1 (d - s) `snocOL`
499 -- Case 4: Tail call of function
501 = doTailCall d s p fn args_r_to_l
504 -- Detect and extract relevant info for the tagToEnum kludge.
505 maybe_is_tagToEnum_call
506 = let extract_constr_Names ty
507 | Just (tyc, []) <- splitTyConApp_maybe (repType ty),
509 = map (getName . dataConWorkId) (tyConDataCons tyc)
510 -- NOTE: use the worker name, not the source name of
511 -- the DataCon. See DataCon.lhs for details.
513 = panic "maybe_is_tagToEnum_call.extract_constr_Ids"
516 (AnnApp (_, AnnApp (_, AnnVar v) (_, AnnType t)) arg)
517 -> case isPrimOpId_maybe v of
518 Just TagToEnumOp -> Just (snd arg, extract_constr_Names t)
522 -- Extract the args (R->L) and fn
523 -- The function will necessarily be a variable,
524 -- because we are compiling a tail call
525 (AnnVar fn, args_r_to_l) = splitApp app
527 -- Only consider this to be a constructor application iff it is
528 -- saturated. Otherwise, we'll call the constructor wrapper.
529 n_args = length args_r_to_l
531 = case isDataConWorkId_maybe fn of
532 Just con | dataConRepArity con == n_args -> Just con
535 -- -----------------------------------------------------------------------------
536 -- Generate code to build a constructor application,
537 -- leaving it on top of the stack
539 mkConAppCode :: Int -> Sequel -> BCEnv
540 -> DataCon -- The data constructor
541 -> [AnnExpr' Id VarSet] -- Args, in *reverse* order
544 mkConAppCode orig_d s p con [] -- Nullary constructor
545 = ASSERT( isNullaryDataCon con )
546 returnBc (unitOL (PUSH_G (getName (dataConWorkId con))))
547 -- Instead of doing a PACK, which would allocate a fresh
548 -- copy of this constructor, use the single shared version.
550 mkConAppCode orig_d s p con args_r_to_l
551 = ASSERT( dataConRepArity con == length args_r_to_l )
552 do_pushery orig_d (non_ptr_args ++ ptr_args)
554 -- The args are already in reverse order, which is the way PACK
555 -- expects them to be. We must push the non-ptrs after the ptrs.
556 (ptr_args, non_ptr_args) = partition isPtrAtom args_r_to_l
558 do_pushery d (arg:args)
559 = pushAtom d p arg `thenBc` \ (push, arg_words) ->
560 do_pushery (d+arg_words) args `thenBc` \ more_push_code ->
561 returnBc (push `appOL` more_push_code)
563 = returnBc (unitOL (PACK con n_arg_words))
565 n_arg_words = d - orig_d
568 -- -----------------------------------------------------------------------------
569 -- Returning an unboxed tuple with one non-void component (the only
570 -- case we can handle).
572 -- Remember, we don't want to *evaluate* the component that is being
573 -- returned, even if it is a pointed type. We always just return.
576 :: Int -> Sequel -> BCEnv
577 -> AnnExpr' Id VarSet -> BcM BCInstrList
578 unboxedTupleReturn d s p arg = do
579 (push, sz) <- pushAtom d p arg
580 returnBc (push `appOL`
581 mkSLIDE sz (d-s) `snocOL`
582 RETURN_UBX (atomRep arg))
584 -- -----------------------------------------------------------------------------
585 -- Generate code for a tail-call
588 :: Int -> Sequel -> BCEnv
589 -> Id -> [AnnExpr' Id VarSet]
591 doTailCall init_d s p fn args
592 = do_pushes init_d args (map (primRepToArgRep.atomRep) args)
594 do_pushes d [] reps = do
596 (push_fn, sz) <- pushAtom d p (AnnVar fn)
598 returnBc (push_fn `appOL` (
599 mkSLIDE ((d-init_d) + 1) (init_d - s) `appOL`
601 do_pushes d args reps = do
602 let (push_apply, n, rest_of_reps) = findPushSeq reps
603 (these_args, rest_of_args) = splitAt n args
604 (next_d, push_code) <- push_seq d these_args
605 instrs <- do_pushes (next_d + 1) rest_of_args rest_of_reps
606 -- ^^^ for the PUSH_APPLY_ instruction
607 returnBc (push_code `appOL` (push_apply `consOL` instrs))
609 push_seq d [] = return (d, nilOL)
610 push_seq d (arg:args) = do
611 (push_code, sz) <- pushAtom d p arg
612 (final_d, more_push_code) <- push_seq (d+sz) args
613 return (final_d, push_code `appOL` more_push_code)
615 -- v. similar to CgStackery.findMatch, ToDo: merge
616 findPushSeq (RepP: RepP: RepP: RepP: RepP: RepP: RepP: rest)
617 = (PUSH_APPLY_PPPPPPP, 7, rest)
618 findPushSeq (RepP: RepP: RepP: RepP: RepP: RepP: rest)
619 = (PUSH_APPLY_PPPPPP, 6, rest)
620 findPushSeq (RepP: RepP: RepP: RepP: RepP: rest)
621 = (PUSH_APPLY_PPPPP, 5, rest)
622 findPushSeq (RepP: RepP: RepP: RepP: rest)
623 = (PUSH_APPLY_PPPP, 4, rest)
624 findPushSeq (RepP: RepP: RepP: rest)
625 = (PUSH_APPLY_PPP, 3, rest)
626 findPushSeq (RepP: RepP: rest)
627 = (PUSH_APPLY_PP, 2, rest)
628 findPushSeq (RepP: rest)
629 = (PUSH_APPLY_P, 1, rest)
630 findPushSeq (RepV: rest)
631 = (PUSH_APPLY_V, 1, rest)
632 findPushSeq (RepN: rest)
633 = (PUSH_APPLY_N, 1, rest)
634 findPushSeq (RepF: rest)
635 = (PUSH_APPLY_F, 1, rest)
636 findPushSeq (RepD: rest)
637 = (PUSH_APPLY_D, 1, rest)
638 findPushSeq (RepL: rest)
639 = (PUSH_APPLY_L, 1, rest)
641 = panic "ByteCodeGen.findPushSeq"
643 -- -----------------------------------------------------------------------------
646 doCase :: Int -> Sequel -> BCEnv
647 -> AnnExpr Id VarSet -> Id -> [AnnAlt Id VarSet]
648 -> Bool -- True <=> is an unboxed tuple case, don't enter the result
650 doCase d s p (_,scrut)
651 bndr alts is_unboxed_tuple
653 -- Top of stack is the return itbl, as usual.
654 -- underneath it is the pointer to the alt_code BCO.
655 -- When an alt is entered, it assumes the returned value is
656 -- on top of the itbl.
659 -- An unlifted value gets an extra info table pushed on top
660 -- when it is returned.
661 unlifted_itbl_sizeW | isAlgCase = 0
664 -- depth of stack after the return value has been pushed
665 d_bndr = d + ret_frame_sizeW + idSizeW bndr
667 -- depth of stack after the extra info table for an unboxed return
668 -- has been pushed, if any. This is the stack depth at the
670 d_alts = d_bndr + unlifted_itbl_sizeW
672 -- Env in which to compile the alts, not including
673 -- any vars bound by the alts themselves
674 p_alts = addToFM p bndr (d_bndr - 1)
676 bndr_ty = idType bndr
677 isAlgCase = not (isUnLiftedType bndr_ty) && not is_unboxed_tuple
679 -- given an alt, return a discr and code for it.
680 codeALt alt@(DEFAULT, _, (_,rhs))
681 = schemeE d_alts s p_alts rhs `thenBc` \ rhs_code ->
682 returnBc (NoDiscr, rhs_code)
683 codeAlt alt@(discr, bndrs, (_,rhs))
684 -- primitive or nullary constructor alt: no need to UNPACK
685 | null real_bndrs = do
686 rhs_code <- schemeE d_alts s p_alts rhs
687 returnBc (my_discr alt, rhs_code)
688 -- algebraic alt with some binders
689 | ASSERT(isAlgCase) otherwise =
691 (ptrs,nptrs) = partition (isFollowableRep.idPrimRep) real_bndrs
692 ptr_sizes = map idSizeW ptrs
693 nptrs_sizes = map idSizeW nptrs
694 bind_sizes = ptr_sizes ++ nptrs_sizes
695 size = sum ptr_sizes + sum nptrs_sizes
696 -- the UNPACK instruction unpacks in reverse order...
697 p' = addListToFM p_alts
698 (zip (reverse (ptrs ++ nptrs))
699 (mkStackOffsets d_alts (reverse bind_sizes)))
701 rhs_code <- schemeE (d_alts+size) s p' rhs
702 return (my_discr alt, unitOL (UNPACK size) `appOL` rhs_code)
704 real_bndrs = filter (not.isTyVar) bndrs
707 my_discr (DEFAULT, binds, rhs) = NoDiscr {-shouldn't really happen-}
708 my_discr (DataAlt dc, binds, rhs)
709 | isUnboxedTupleCon dc
710 = unboxedTupleException
712 = DiscrP (dataConTag dc - fIRST_TAG)
713 my_discr (LitAlt l, binds, rhs)
714 = case l of MachInt i -> DiscrI (fromInteger i)
715 MachFloat r -> DiscrF (fromRational r)
716 MachDouble r -> DiscrD (fromRational r)
717 MachChar i -> DiscrI (ord i)
718 _ -> pprPanic "schemeE(AnnCase).my_discr" (ppr l)
721 | not isAlgCase = Nothing
723 = case [dc | (DataAlt dc, _, _) <- alts] of
725 (dc:_) -> Just (tyConFamilySize (dataConTyCon dc))
727 -- the bitmap is relative to stack depth d, i.e. before the
728 -- BCO, info table and return value are pushed on.
729 -- This bit of code is v. similar to buildLivenessMask in CgBindery,
730 -- except that here we build the bitmap from the known bindings of
731 -- things that are pointers, whereas in CgBindery the code builds the
732 -- bitmap from the free slots and unboxed bindings.
734 bitmap = intsToReverseBitmap d{-size-} (sortLt (<) rel_slots)
737 rel_slots = concat (map spread binds)
739 | isFollowableRep (idPrimRep id) = [ rel_offset ]
741 where rel_offset = d - offset - 1
744 alt_stuff <- mapM codeAlt alts
745 alt_final <- mkMultiBranch maybe_ncons alt_stuff
747 alt_bco_name = getName bndr
748 alt_bco = mkProtoBCO alt_bco_name alt_final (Left alts)
749 0{-no arity-} d{-bitmap size-} bitmap True{-is alts-}
751 -- trace ("case: bndr = " ++ showSDocDebug (ppr bndr) ++ "\ndepth = " ++ show d ++ "\nenv = \n" ++ showSDocDebug (ppBCEnv p) ++
752 -- "\n bitmap = " ++ show bitmap) $ do
753 scrut_code <- schemeE (d + ret_frame_sizeW) (d + ret_frame_sizeW) p scrut
754 alt_bco' <- emitBc alt_bco
756 | isAlgCase = PUSH_ALTS alt_bco'
757 | otherwise = PUSH_ALTS_UNLIFTED alt_bco' (typePrimRep bndr_ty)
758 returnBc (push_alts `consOL` scrut_code)
761 -- -----------------------------------------------------------------------------
762 -- Deal with a CCall.
764 -- Taggedly push the args onto the stack R->L,
765 -- deferencing ForeignObj#s and adjusting addrs to point to
766 -- payloads in Ptr/Byte arrays. Then, generate the marshalling
767 -- (machine) code for the ccall, and create bytecodes to call that and
768 -- then return in the right way.
770 generateCCall :: Int -> Sequel -- stack and sequel depths
772 -> CCallSpec -- where to call
773 -> Id -- of target, for type info
774 -> [AnnExpr' Id VarSet] -- args (atoms)
777 generateCCall d0 s p ccall_spec@(CCallSpec target cconv safety) fn args_r_to_l
780 addr_sizeW = getPrimRepSize AddrRep
782 -- Get the args on the stack, with tags and suitably
783 -- dereferenced for the CCall. For each arg, return the
784 -- depth to the first word of the bits for that arg, and the
785 -- PrimRep of what was actually pushed.
787 pargs d [] = returnBc []
789 = let arg_ty = repType (exprType (deAnnotate' a))
791 in case splitTyConApp_maybe arg_ty of
792 -- Don't push the FO; instead push the Addr# it
795 | t == arrayPrimTyCon || t == mutableArrayPrimTyCon
796 -> pargs (d + addr_sizeW) az `thenBc` \ rest ->
797 parg_ArrayishRep arrPtrsHdrSize d p a
799 returnBc ((code,AddrRep):rest)
801 | t == byteArrayPrimTyCon || t == mutableByteArrayPrimTyCon
802 -> pargs (d + addr_sizeW) az `thenBc` \ rest ->
803 parg_ArrayishRep arrWordsHdrSize d p a
805 returnBc ((code,AddrRep):rest)
807 -- Default case: push taggedly, but otherwise intact.
809 -> pushAtom d p a `thenBc` \ (code_a, sz_a) ->
810 pargs (d+sz_a) az `thenBc` \ rest ->
811 returnBc ((code_a, atomRep a) : rest)
813 -- Do magic for Ptr/Byte arrays. Push a ptr to the array on
814 -- the stack but then advance it over the headers, so as to
815 -- point to the payload.
816 parg_ArrayishRep hdrSizeW d p a
817 = pushAtom d p a `thenBc` \ (push_fo, _) ->
818 -- The ptr points at the header. Advance it over the
819 -- header and then pretend this is an Addr#.
820 returnBc (push_fo `snocOL`
821 SWIZZLE 0 (hdrSizeW * getPrimRepSize WordRep
825 pargs d0 args_r_to_l `thenBc` \ code_n_reps ->
827 (pushs_arg, a_reps_pushed_r_to_l) = unzip code_n_reps
829 push_args = concatOL pushs_arg
830 d_after_args = d0 + sum (map getPrimRepSize a_reps_pushed_r_to_l)
832 | null a_reps_pushed_r_to_l || head a_reps_pushed_r_to_l /= VoidRep
833 = panic "ByteCodeGen.generateCCall: missing or invalid World token?"
835 = reverse (tail a_reps_pushed_r_to_l)
837 -- Now: a_reps_pushed_RAW are the reps which are actually on the stack.
838 -- push_args is the code to do that.
839 -- d_after_args is the stack depth once the args are on.
841 -- Get the result rep.
842 (returns_void, r_rep)
843 = case maybe_getCCallReturnRep (idType fn) of
844 Nothing -> (True, VoidRep)
845 Just rr -> (False, rr)
847 Because the Haskell stack grows down, the a_reps refer to
848 lowest to highest addresses in that order. The args for the call
849 are on the stack. Now push an unboxed Addr# indicating
850 the C function to call. Then push a dummy placeholder for the
851 result. Finally, emit a CCALL insn with an offset pointing to the
852 Addr# just pushed, and a literal field holding the mallocville
853 address of the piece of marshalling code we generate.
854 So, just prior to the CCALL insn, the stack looks like this
855 (growing down, as usual):
860 Addr# address_of_C_fn
861 <placeholder-for-result#> (must be an unboxed type)
863 The interpreter then calls the marshall code mentioned
864 in the CCALL insn, passing it (& <placeholder-for-result#>),
865 that is, the addr of the topmost word in the stack.
866 When this returns, the placeholder will have been
867 filled in. The placeholder is slid down to the sequel
868 depth, and we RETURN.
870 This arrangement makes it simple to do f-i-dynamic since the Addr#
871 value is the first arg anyway.
873 The marshalling code is generated specifically for this
874 call site, and so knows exactly the (Haskell) stack
875 offsets of the args, fn address and placeholder. It
876 copies the args to the C stack, calls the stacked addr,
877 and parks the result back in the placeholder. The interpreter
878 calls it as a normal C call, assuming it has a signature
879 void marshall_code ( StgWord* ptr_to_top_of_stack )
881 -- resolve static address
885 -> returnBc (False, panic "ByteCodeGen.generateCCall(dyn)")
887 -> ioToBc (lookupStaticPtr target) `thenBc` \res ->
890 get_target_info `thenBc` \ (is_static, static_target_addr) ->
893 -- Get the arg reps, zapping the leading Addr# in the dynamic case
894 a_reps -- | trace (showSDoc (ppr a_reps_pushed_RAW)) False = error "???"
895 | is_static = a_reps_pushed_RAW
896 | otherwise = if null a_reps_pushed_RAW
897 then panic "ByteCodeGen.generateCCall: dyn with no args"
898 else tail a_reps_pushed_RAW
901 (push_Addr, d_after_Addr)
903 = (toOL [PUSH_UBX (Right static_target_addr) addr_sizeW],
904 d_after_args + addr_sizeW)
905 | otherwise -- is already on the stack
906 = (nilOL, d_after_args)
908 -- Push the return placeholder. For a call returning nothing,
909 -- this is a VoidRep (tag).
910 r_sizeW = getPrimRepSize r_rep
911 d_after_r = d_after_Addr + r_sizeW
912 r_lit = mkDummyLiteral r_rep
913 push_r = (if returns_void
915 else unitOL (PUSH_UBX (Left r_lit) r_sizeW))
917 -- generate the marshalling code we're going to call
920 arg1_offW = r_sizeW + addr_sizeW
921 args_offW = map (arg1_offW +)
922 (init (scanl (+) 0 (map getPrimRepSize a_reps)))
924 ioToBc (mkMarshalCode cconv
925 (r_offW, r_rep) addr_offW
926 (zip args_offW a_reps)) `thenBc` \ addr_of_marshaller ->
927 recordMallocBc addr_of_marshaller `thenBc_`
929 -- Offset of the next stack frame down the stack. The CCALL
930 -- instruction needs to describe the chunk of stack containing
931 -- the ccall args to the GC, so it needs to know how large it
932 -- is. See comment in Interpreter.c with the CCALL instruction.
933 stk_offset = d_after_r - s
936 do_call = unitOL (CCALL stk_offset (castPtr addr_of_marshaller))
938 wrapup = mkSLIDE r_sizeW (d_after_r - r_sizeW - s)
939 `snocOL` RETURN_UBX r_rep
941 --trace (show (arg1_offW, args_offW , (map getPrimRepSize a_reps) )) $
944 push_Addr `appOL` push_r `appOL` do_call `appOL` wrapup
948 -- Make a dummy literal, to be used as a placeholder for FFI return
949 -- values on the stack.
950 mkDummyLiteral :: PrimRep -> Literal
953 CharRep -> MachChar (chr 0)
955 WordRep -> MachWord 0
956 DoubleRep -> MachDouble 0
957 FloatRep -> MachFloat 0
958 AddrRep | getPrimRepSize AddrRep == getPrimRepSize WordRep -> MachWord 0
959 _ -> moan64 "mkDummyLiteral" (ppr pr)
963 -- GHC.Prim.Char# -> GHC.Prim.State# GHC.Prim.RealWorld
964 -- -> (# GHC.Prim.State# GHC.Prim.RealWorld, GHC.Prim.Int# #)
967 -- and check that an unboxed pair is returned wherein the first arg is VoidRep'd.
969 -- Alternatively, for call-targets returning nothing, convert
971 -- GHC.Prim.Char# -> GHC.Prim.State# GHC.Prim.RealWorld
972 -- -> (# GHC.Prim.State# GHC.Prim.RealWorld #)
976 maybe_getCCallReturnRep :: Type -> Maybe PrimRep
977 maybe_getCCallReturnRep fn_ty
978 = let (a_tys, r_ty) = splitFunTys (dropForAlls fn_ty)
980 = if isSingleton r_reps then Nothing else Just (r_reps !! 1)
982 = case splitTyConApp_maybe (repType r_ty) of
983 (Just (tyc, tys)) -> (tyc, map typePrimRep tys)
985 ok = ( ( r_reps `lengthIs` 2 && VoidRep == head r_reps)
986 || r_reps == [VoidRep] )
987 && isUnboxedTupleTyCon r_tycon
988 && case maybe_r_rep_to_go of
990 Just r_rep -> r_rep /= PtrRep
991 -- if it was, it would be impossible
992 -- to create a valid return value
993 -- placeholder on the stack
994 blargh = pprPanic "maybe_getCCallReturn: can't handle:"
997 --trace (showSDoc (ppr (a_reps, r_reps))) $
998 if ok then maybe_r_rep_to_go else blargh
1000 -- Compile code which expects an unboxed Int on the top of stack,
1001 -- (call it i), and pushes the i'th closure in the supplied list
1002 -- as a consequence.
1003 implement_tagToId :: [Name] -> BcM BCInstrList
1004 implement_tagToId names
1005 = ASSERT( notNull names )
1006 getLabelsBc (length names) `thenBc` \ labels ->
1007 getLabelBc `thenBc` \ label_fail ->
1008 getLabelBc `thenBc` \ label_exit ->
1009 zip4 labels (tail labels ++ [label_fail])
1010 [0 ..] names `bind` \ infos ->
1011 map (mkStep label_exit) infos `bind` \ steps ->
1012 returnBc (concatOL steps
1014 toOL [LABEL label_fail, CASEFAIL, LABEL label_exit])
1016 mkStep l_exit (my_label, next_label, n, name_for_n)
1017 = toOL [LABEL my_label,
1018 TESTEQ_I n next_label,
1023 -- -----------------------------------------------------------------------------
1026 -- Push an atom onto the stack, returning suitable code & number of
1027 -- stack words used.
1029 -- The env p must map each variable to the highest- numbered stack
1030 -- slot for it. For example, if the stack has depth 4 and we
1031 -- tagged-ly push (v :: Int#) on it, the value will be in stack[4],
1032 -- the tag in stack[5], the stack will have depth 6, and p must map v
1033 -- to 5 and not to 4. Stack locations are numbered from zero, so a
1034 -- depth 6 stack has valid words 0 .. 5.
1036 pushAtom :: Int -> BCEnv -> AnnExpr' Id VarSet -> BcM (BCInstrList, Int)
1038 pushAtom d p (AnnApp f (_, AnnType _))
1039 = pushAtom d p (snd f)
1041 pushAtom d p (AnnNote note e)
1042 = pushAtom d p (snd e)
1044 pushAtom d p (AnnLam x e)
1046 = pushAtom d p (snd e)
1048 pushAtom d p (AnnVar v)
1050 | idPrimRep v == VoidRep
1051 = returnBc (nilOL, 0)
1054 = pprPanic "pushAtom: shouldn't get an FCallId here" (ppr v)
1056 | Just primop <- isPrimOpId_maybe v
1057 = returnBc (unitOL (PUSH_PRIMOP primop), 1)
1059 | Just d_v <- lookupBCEnv_maybe p v -- v is a local variable
1060 = returnBc (toOL (nOfThem sz (PUSH_L (d-d_v+sz-2))), sz)
1061 -- d - d_v the number of words between the TOS
1062 -- and the 1st slot of the object
1064 -- d - d_v - 1 the offset from the TOS of the 1st slot
1066 -- d - d_v - 1 + sz - 1 the offset from the TOS of the last slot
1069 -- Having found the last slot, we proceed to copy the right number of
1070 -- slots on to the top of the stack.
1072 | otherwise -- v must be a global variable
1074 returnBc (unitOL (PUSH_G (getName v)), sz)
1080 pushAtom d p (AnnLit lit)
1082 MachLabel fs _ -> code CodePtrRep
1083 MachWord w -> code WordRep
1084 MachInt i -> code IntRep
1085 MachFloat r -> code FloatRep
1086 MachDouble r -> code DoubleRep
1087 MachChar c -> code CharRep
1088 MachStr s -> pushStr s
1091 = let size_host_words = getPrimRepSize rep
1092 in returnBc (unitOL (PUSH_UBX (Left lit) size_host_words),
1096 = let getMallocvilleAddr
1098 FastString _ l ba ->
1099 -- sigh, a string in the heap is no good to us.
1100 -- We need a static C pointer, since the type of
1101 -- a string literal is Addr#. So, copy the string
1102 -- into C land and remember the pointer so we can
1105 -- CAREFUL! Chars are 32 bits in ghc 4.09+
1106 in ioToBc (mallocBytes (n+1)) `thenBc` \ ptr ->
1107 recordMallocBc ptr `thenBc_`
1109 do memcpy ptr ba (fromIntegral n)
1110 pokeByteOff ptr n (fromIntegral (ord '\0') :: Word8)
1113 other -> panic "ByteCodeGen.pushAtom.pushStr"
1115 getMallocvilleAddr `thenBc` \ addr ->
1116 -- Get the addr on the stack, untaggedly
1117 returnBc (unitOL (PUSH_UBX (Right addr) 1), 1)
1120 = pprPanic "ByteCodeGen.pushAtom"
1121 (pprCoreExpr (deAnnotate (undefined, other)))
1123 foreign import ccall unsafe "memcpy"
1124 memcpy :: Ptr a -> ByteArray# -> CInt -> IO ()
1127 -- -----------------------------------------------------------------------------
1128 -- Given a bunch of alts code and their discrs, do the donkey work
1129 -- of making a multiway branch using a switch tree.
1130 -- What a load of hassle!
1132 mkMultiBranch :: Maybe Int -- # datacons in tycon, if alg alt
1133 -- a hint; generates better code
1134 -- Nothing is always safe
1135 -> [(Discr, BCInstrList)]
1137 mkMultiBranch maybe_ncons raw_ways
1138 = let d_way = filter (isNoDiscr.fst) raw_ways
1139 notd_ways = naturalMergeSortLe
1140 (\w1 w2 -> leAlt (fst w1) (fst w2))
1141 (filter (not.isNoDiscr.fst) raw_ways)
1143 mkTree :: [(Discr, BCInstrList)] -> Discr -> Discr -> BcM BCInstrList
1144 mkTree [] range_lo range_hi = returnBc the_default
1146 mkTree [val] range_lo range_hi
1147 | range_lo `eqAlt` range_hi
1148 = returnBc (snd val)
1150 = getLabelBc `thenBc` \ label_neq ->
1151 returnBc (mkTestEQ (fst val) label_neq
1153 `appOL` unitOL (LABEL label_neq)
1154 `appOL` the_default))
1156 mkTree vals range_lo range_hi
1157 = let n = length vals `div` 2
1158 vals_lo = take n vals
1159 vals_hi = drop n vals
1160 v_mid = fst (head vals_hi)
1162 getLabelBc `thenBc` \ label_geq ->
1163 mkTree vals_lo range_lo (dec v_mid) `thenBc` \ code_lo ->
1164 mkTree vals_hi v_mid range_hi `thenBc` \ code_hi ->
1165 returnBc (mkTestLT v_mid label_geq
1167 `appOL` unitOL (LABEL label_geq)
1171 = case d_way of [] -> unitOL CASEFAIL
1174 -- None of these will be needed if there are no non-default alts
1175 (mkTestLT, mkTestEQ, init_lo, init_hi)
1177 = panic "mkMultiBranch: awesome foursome"
1179 = case fst (head notd_ways) of {
1180 DiscrI _ -> ( \(DiscrI i) fail_label -> TESTLT_I i fail_label,
1181 \(DiscrI i) fail_label -> TESTEQ_I i fail_label,
1184 DiscrF _ -> ( \(DiscrF f) fail_label -> TESTLT_F f fail_label,
1185 \(DiscrF f) fail_label -> TESTEQ_F f fail_label,
1188 DiscrD _ -> ( \(DiscrD d) fail_label -> TESTLT_D d fail_label,
1189 \(DiscrD d) fail_label -> TESTEQ_D d fail_label,
1192 DiscrP _ -> ( \(DiscrP i) fail_label -> TESTLT_P i fail_label,
1193 \(DiscrP i) fail_label -> TESTEQ_P i fail_label,
1195 DiscrP algMaxBound )
1198 (algMinBound, algMaxBound)
1199 = case maybe_ncons of
1200 Just n -> (0, n - 1)
1201 Nothing -> (minBound, maxBound)
1203 (DiscrI i1) `eqAlt` (DiscrI i2) = i1 == i2
1204 (DiscrF f1) `eqAlt` (DiscrF f2) = f1 == f2
1205 (DiscrD d1) `eqAlt` (DiscrD d2) = d1 == d2
1206 (DiscrP i1) `eqAlt` (DiscrP i2) = i1 == i2
1207 NoDiscr `eqAlt` NoDiscr = True
1210 (DiscrI i1) `leAlt` (DiscrI i2) = i1 <= i2
1211 (DiscrF f1) `leAlt` (DiscrF f2) = f1 <= f2
1212 (DiscrD d1) `leAlt` (DiscrD d2) = d1 <= d2
1213 (DiscrP i1) `leAlt` (DiscrP i2) = i1 <= i2
1214 NoDiscr `leAlt` NoDiscr = True
1217 isNoDiscr NoDiscr = True
1220 dec (DiscrI i) = DiscrI (i-1)
1221 dec (DiscrP i) = DiscrP (i-1)
1222 dec other = other -- not really right, but if you
1223 -- do cases on floating values, you'll get what you deserve
1225 -- same snotty comment applies to the following
1227 minD, maxD :: Double
1233 mkTree notd_ways init_lo init_hi
1236 -- -----------------------------------------------------------------------------
1237 -- Supporting junk for the compilation schemes
1239 -- Describes case alts
1247 instance Outputable Discr where
1248 ppr (DiscrI i) = int i
1249 ppr (DiscrF f) = text (show f)
1250 ppr (DiscrD d) = text (show d)
1251 ppr (DiscrP i) = int i
1252 ppr NoDiscr = text "DEF"
1255 lookupBCEnv_maybe :: BCEnv -> Id -> Maybe Int
1256 lookupBCEnv_maybe = lookupFM
1258 idSizeW :: Id -> Int
1259 idSizeW id = getPrimRepSize (typePrimRep (idType id))
1261 unboxedTupleException :: a
1262 unboxedTupleException
1265 ("Bytecode generator can't handle unboxed tuples. Possibly due\n" ++
1266 "\tto foreign import/export decls in source. Workaround:\n" ++
1267 "\tcompile this module to a .o file, then restart session."))
1270 mkSLIDE n d = if d == 0 then nilOL else unitOL (SLIDE n d)
1273 splitApp :: AnnExpr' id ann -> (AnnExpr' id ann, [AnnExpr' id ann])
1274 -- The arguments are returned in *right-to-left* order
1275 splitApp (AnnApp (_,f) (_,a))
1276 | isTypeAtom a = splitApp f
1277 | otherwise = case splitApp f of
1278 (f', as) -> (f', a:as)
1279 splitApp (AnnNote n (_,e)) = splitApp e
1280 splitApp e = (e, [])
1283 isTypeAtom :: AnnExpr' id ann -> Bool
1284 isTypeAtom (AnnType _) = True
1285 isTypeAtom _ = False
1287 isVoidRepAtom :: AnnExpr' id ann -> Bool
1288 isVoidRepAtom (AnnVar v) = typePrimRep (idType v) == VoidRep
1289 isVoidRepAtom (AnnNote n (_,e)) = isVoidRepAtom e
1290 isVoidRepAtom _ = False
1292 atomRep :: AnnExpr' Id ann -> PrimRep
1293 atomRep (AnnVar v) = typePrimRep (idType v)
1294 atomRep (AnnLit l) = literalPrimRep l
1295 atomRep (AnnNote n b) = atomRep (snd b)
1296 atomRep (AnnApp f (_, AnnType _)) = atomRep (snd f)
1297 atomRep (AnnLam x e) | isTyVar x = atomRep (snd e)
1298 atomRep other = pprPanic "atomRep" (ppr (deAnnotate (undefined,other)))
1300 isPtrAtom :: AnnExpr' Id ann -> Bool
1301 isPtrAtom e = isFollowableRep (atomRep e)
1303 -- Let szsw be the sizes in words of some items pushed onto the stack,
1304 -- which has initial depth d'. Return the values which the stack environment
1305 -- should map these items to.
1306 mkStackOffsets :: Int -> [Int] -> [Int]
1307 mkStackOffsets original_depth szsw
1308 = map (subtract 1) (tail (scanl (+) original_depth szsw))
1310 -- -----------------------------------------------------------------------------
1311 -- The bytecode generator's monad
1315 nextlabel :: Int, -- for generating local labels
1316 malloced :: [Ptr ()] } -- ptrs malloced for current BCO
1317 -- Should be free()d when it is GCd
1319 newtype BcM r = BcM (BcM_State -> IO (BcM_State, r))
1321 ioToBc :: IO a -> BcM a
1322 ioToBc io = BcM $ \st -> do
1326 runBc :: BcM r -> IO (BcM_State, r)
1327 runBc (BcM m) = m (BcM_State 0 [])
1329 thenBc :: BcM a -> (a -> BcM b) -> BcM b
1330 thenBc (BcM expr) cont = BcM $ \st0 -> do
1331 (st1, q) <- expr st0
1336 thenBc_ :: BcM a -> BcM b -> BcM b
1337 thenBc_ (BcM expr) (BcM cont) = BcM $ \st0 -> do
1338 (st1, q) <- expr st0
1339 (st2, r) <- cont st1
1342 returnBc :: a -> BcM a
1343 returnBc result = BcM $ \st -> (return (st, result))
1345 instance Monad BcM where
1350 emitBc :: ([Ptr ()] -> ProtoBCO Name) -> BcM (ProtoBCO Name)
1352 = BcM $ \st -> return (st{malloced=[]}, bco (malloced st))
1354 recordMallocBc :: Ptr a -> BcM ()
1356 = BcM $ \st -> return (st{malloced = castPtr a : malloced st}, ())
1358 getLabelBc :: BcM Int
1360 = BcM $ \st -> return (st{nextlabel = 1 + nextlabel st}, nextlabel st)
1362 getLabelsBc :: Int -> BcM [Int]
1364 = BcM $ \st -> let ctr = nextlabel st
1365 in return (st{nextlabel = ctr+n}, [ctr .. ctr+n-1])