2 % (c) The University of Glasgow 2000
4 \section[ByteCodeLink]{Bytecode assembler and linker}
7 module ByteCodeLink ( UnlinkedBCO, UnlinkedBCOExpr, assembleBCO,
8 ClosureEnv, HValue, filterNameMap,
9 linkIModules, linkIExpr,
10 iNTERP_STACK_CHECK_THRESH
13 #include "HsVersions.h"
16 import Name ( Name, getName, nameModule, toRdrName, isGlobalName )
17 import RdrName ( rdrNameOcc, rdrNameModule )
18 import OccName ( occNameString )
19 import FiniteMap ( FiniteMap, addListToFM, filterFM,
20 addToFM, lookupFM, emptyFM )
22 import Literal ( Literal(..) )
23 import PrimOp ( PrimOp, primOpOcc )
24 import PrimRep ( PrimRep(..) )
25 import Constants ( wORD_SIZE )
26 import Module ( ModuleName, moduleName, moduleNameFS )
27 import Linker ( lookupSymbol )
28 import FastString ( FastString(..) )
29 import ByteCodeInstr ( BCInstr(..), ProtoBCO(..) )
30 import ByteCodeItbls ( ItblEnv, ItblPtr )
33 import Monad ( when, foldM )
35 import IArray ( array )
36 import MArray ( castSTUArray,
37 newFloatArray, writeFloatArray,
38 newDoubleArray, writeDoubleArray,
39 newIntArray, writeIntArray,
40 newAddrArray, writeAddrArray,
42 import Foreign ( Word16, Ptr(..), free )
43 import Addr ( Word, Addr(..), nullAddr )
44 import Weak ( addFinalizer )
47 import PrelBase ( Int(..) )
48 import PrelGHC ( BCO#, newBCO#, unsafeCoerce#,
49 ByteArray#, Array#, addrToHValue#, mkApUpd0# )
50 import IOExts ( fixIO )
51 import PrelArr ( Array(..) )
52 import ArrayBase ( UArray(..) )
53 import PrelIOBase ( IO(..) )
57 %************************************************************************
59 \subsection{Top-level stuff}
61 %************************************************************************
65 linkIModules :: ItblEnv -- incoming global itbl env; returned updated
66 -> ClosureEnv -- incoming global closure env; returned updated
67 -> [([UnlinkedBCO], ItblEnv)]
68 -> IO ([HValue], ItblEnv, ClosureEnv)
69 linkIModules gie gce mods
70 = do let (bcoss, ies) = unzip mods
72 final_gie = foldr plusFM gie ies
73 (final_gce, linked_bcos) <- linkSomeBCOs True final_gie gce bcos
74 return (linked_bcos, final_gie, final_gce)
77 linkIExpr :: ItblEnv -> ClosureEnv -> UnlinkedBCOExpr
78 -> IO HValue -- IO BCO# really
79 linkIExpr ie ce (root_ul_bco, aux_ul_bcos)
80 = do (aux_ce, _) <- linkSomeBCOs False ie ce aux_ul_bcos
81 (_, [root_bco]) <- linkSomeBCOs False ie aux_ce [root_ul_bco]
84 -- Link a bunch of BCOs and return them + updated closure env.
85 linkSomeBCOs :: Bool -- False <=> add _all_ BCOs to returned closure env
86 -- True <=> add only toplevel BCOs to closure env
90 -> IO (ClosureEnv, [HValue])
91 linkSomeBCOs toplevs_only ie ce_in ul_bcos
92 = do let nms = map nameOfUnlinkedBCO ul_bcos
94 ( \ hvs -> let ce_out = addListToFM ce_in (zipLazily nms hvs)
95 in mapM (linkBCO ie ce_out) ul_bcos )
97 let ce_all_additions = zip nms hvals
98 ce_top_additions = filter (isGlobalName.fst) ce_all_additions
99 ce_additions = if toplevs_only then ce_top_additions
100 else ce_all_additions
101 ce_out = -- make sure we're not inserting duplicate names into the
102 -- closure environment, which leads to trouble.
103 ASSERT (all (not . (`elemFM` ce_in)) (map fst ce_additions))
104 addListToFM ce_in ce_additions
105 return (ce_out, hvals)
107 -- A lazier zip, in which no demand is propagated to the second
108 -- list unless some demand is propagated to the snd of one of the
109 -- result list elems.
111 zipLazily (x:xs) ys = (x, head ys) : zipLazily xs (tail ys)
116 (SizedSeq Word16) -- insns
117 (SizedSeq Word) -- literals
118 (SizedSeq (Either Name PrimOp)) -- ptrs
119 (SizedSeq Name) -- itbl refs
121 nameOfUnlinkedBCO (UnlinkedBCO nm _ _ _ _) = nm
123 -- When translating expressions, we need to distinguish the root
124 -- BCO for the expression
125 type UnlinkedBCOExpr = (UnlinkedBCO, [UnlinkedBCO])
127 instance Outputable UnlinkedBCO where
128 ppr (UnlinkedBCO nm insns lits ptrs itbls)
129 = sep [text "BCO", ppr nm, text "with",
130 int (sizeSS insns), text "insns",
131 int (sizeSS lits), text "lits",
132 int (sizeSS ptrs), text "ptrs",
133 int (sizeSS itbls), text "itbls"]
136 -- these need a proper home
137 type ClosureEnv = FiniteMap Name HValue
138 data HValue = HValue -- dummy type, actually a pointer to some Real Code.
140 -- remove all entries for a given set of modules from the environment;
141 -- note that this removes all local names too (ie. temporary bindings from
142 -- the command line).
143 filterNameMap :: [ModuleName] -> FiniteMap Name a -> FiniteMap Name a
144 filterNameMap mods env
145 = filterFM (\n _ -> isGlobalName n &&
146 moduleName (nameModule n) `elem` mods) env
149 %************************************************************************
151 \subsection{The bytecode assembler}
153 %************************************************************************
155 The object format for bytecodes is: 16 bits for the opcode, and 16 for
156 each field -- so the code can be considered a sequence of 16-bit ints.
157 Each field denotes either a stack offset or number of items on the
158 stack (eg SLIDE), and index into the pointer table (eg PUSH_G), an
159 index into the literal table (eg PUSH_I/D/L), or a bytecode address in
163 -- Top level assembler fn.
164 assembleBCO :: ProtoBCO Name -> IO UnlinkedBCO
166 assembleBCO (ProtoBCO nm instrs origin malloced)
168 -- pass 1: collect up the offsets of the local labels.
169 -- Remember that the first insn starts at offset 1 since offset 0
170 -- (eventually) will hold the total # of insns.
171 label_env = mkLabelEnv emptyFM 1 instrs
173 mkLabelEnv env i_offset [] = env
174 mkLabelEnv env i_offset (i:is)
176 = case i of LABEL n -> addToFM env n i_offset ; _ -> env
177 in mkLabelEnv new_env (i_offset + instrSize16s i) is
180 = case lookupFM label_env lab of
181 Just bco_offset -> bco_offset
182 Nothing -> pprPanic "assembleBCO.findLabel" (int lab)
184 do -- pass 2: generate the instruction, ptr and nonptr bits
185 insns <- return emptySS :: IO (SizedSeq Word16)
186 lits <- return emptySS :: IO (SizedSeq Word)
187 ptrs <- return emptySS :: IO (SizedSeq (Either Name PrimOp))
188 itbls <- return emptySS :: IO (SizedSeq Name)
189 let init_asm_state = (insns,lits,ptrs,itbls)
190 (final_insns, final_lits, final_ptrs, final_itbls)
191 <- mkBits findLabel init_asm_state instrs
193 let ul_bco = UnlinkedBCO nm final_insns final_lits final_ptrs final_itbls
194 when (not (null malloced)) (addFinalizer ul_bco (mapM_ zonk malloced))
197 zonk (A# a#) = do -- putStrLn ("freeing malloc'd block at " ++ show (A# a#))
200 -- instrs nonptrs ptrs itbls
201 type AsmState = (SizedSeq Word16, SizedSeq Word,
202 SizedSeq (Either Name PrimOp), SizedSeq Name)
204 data SizedSeq a = SizedSeq !Int [a]
205 emptySS = SizedSeq 0 []
206 addToSS (SizedSeq n r_xs) x = return (SizedSeq (n+1) (x:r_xs))
207 addListToSS (SizedSeq n r_xs) xs
208 = return (SizedSeq (n + length xs) (reverse xs ++ r_xs))
209 sizeSS (SizedSeq n r_xs) = n
210 listFromSS (SizedSeq n r_xs) = return (reverse r_xs)
213 -- This is where all the action is (pass 2 of the assembler)
214 mkBits :: (Int -> Int) -- label finder
216 -> [BCInstr] -- instructions (in)
219 mkBits findLabel st proto_insns
220 = foldM doInstr st proto_insns
222 doInstr :: AsmState -> BCInstr -> IO AsmState
225 ARGCHECK n -> instr2 st i_ARGCHECK n
226 STKCHECK n -> instr2 st i_STKCHECK n
227 PUSH_L o1 -> instr2 st i_PUSH_L o1
228 PUSH_LL o1 o2 -> instr3 st i_PUSH_LL o1 o2
229 PUSH_LLL o1 o2 o3 -> instr4 st i_PUSH_LLL o1 o2 o3
230 PUSH_G nm -> do (p, st2) <- ptr st nm
231 instr2 st2 i_PUSH_G p
232 PUSH_AS nm pk -> do (p, st2) <- ptr st (Left nm)
233 (np, st3) <- ctoi_itbl st2 pk
234 instr3 st3 i_PUSH_AS p np
235 PUSH_UBX (Left lit) nws
236 -> do (np, st2) <- literal st lit
237 instr3 st2 i_PUSH_UBX np nws
238 PUSH_UBX (Right aa) nws
239 -> do (np, st2) <- addr st aa
240 instr3 st2 i_PUSH_UBX np nws
242 PUSH_TAG tag -> instr2 st i_PUSH_TAG tag
243 SLIDE n by -> instr3 st i_SLIDE n by
244 ALLOC n -> instr2 st i_ALLOC n
245 MKAP off sz -> instr3 st i_MKAP off sz
246 UNPACK n -> instr2 st i_UNPACK n
247 UPK_TAG n m k -> instr4 st i_UPK_TAG n m k
248 PACK dcon sz -> do (itbl_no,st2) <- itbl st dcon
249 instr3 st2 i_PACK itbl_no sz
250 LABEL lab -> return st
251 TESTLT_I i l -> do (np, st2) <- int st i
252 instr3 st2 i_TESTLT_I np (findLabel l)
253 TESTEQ_I i l -> do (np, st2) <- int st i
254 instr3 st2 i_TESTEQ_I np (findLabel l)
255 TESTLT_F f l -> do (np, st2) <- float st f
256 instr3 st2 i_TESTLT_F np (findLabel l)
257 TESTEQ_F f l -> do (np, st2) <- float st f
258 instr3 st2 i_TESTEQ_F np (findLabel l)
259 TESTLT_D d l -> do (np, st2) <- double st d
260 instr3 st2 i_TESTLT_D np (findLabel l)
261 TESTEQ_D d l -> do (np, st2) <- double st d
262 instr3 st2 i_TESTEQ_D np (findLabel l)
263 TESTLT_P i l -> instr3 st i_TESTLT_P i (findLabel l)
264 TESTEQ_P i l -> instr3 st i_TESTEQ_P i (findLabel l)
265 CASEFAIL -> instr1 st i_CASEFAIL
266 JMP l -> instr2 st i_JMP (findLabel l)
267 ENTER -> instr1 st i_ENTER
268 RETURN rep -> do (itbl_no,st2) <- itoc_itbl st rep
269 instr2 st2 i_RETURN itbl_no
270 CCALL m_addr -> do (np, st2) <- addr st m_addr
271 instr2 st2 i_CCALL np
276 instr1 (st_i0,st_l0,st_p0,st_I0) i1
277 = do st_i1 <- addToSS st_i0 (i2s i1)
278 return (st_i1,st_l0,st_p0,st_I0)
280 instr2 (st_i0,st_l0,st_p0,st_I0) i1 i2
281 = do st_i1 <- addToSS st_i0 (i2s i1)
282 st_i2 <- addToSS st_i1 (i2s i2)
283 return (st_i2,st_l0,st_p0,st_I0)
285 instr3 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3
286 = do st_i1 <- addToSS st_i0 (i2s i1)
287 st_i2 <- addToSS st_i1 (i2s i2)
288 st_i3 <- addToSS st_i2 (i2s i3)
289 return (st_i3,st_l0,st_p0,st_I0)
291 instr4 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3 i4
292 = do st_i1 <- addToSS st_i0 (i2s i1)
293 st_i2 <- addToSS st_i1 (i2s i2)
294 st_i3 <- addToSS st_i2 (i2s i3)
295 st_i4 <- addToSS st_i3 (i2s i4)
296 return (st_i4,st_l0,st_p0,st_I0)
298 float (st_i0,st_l0,st_p0,st_I0) f
299 = do let ws = mkLitF f
300 st_l1 <- addListToSS st_l0 ws
301 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
303 double (st_i0,st_l0,st_p0,st_I0) d
304 = do let ws = mkLitD d
305 st_l1 <- addListToSS st_l0 ws
306 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
308 int (st_i0,st_l0,st_p0,st_I0) i
309 = do let ws = mkLitI i
310 st_l1 <- addListToSS st_l0 ws
311 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
313 addr (st_i0,st_l0,st_p0,st_I0) a
314 = do let ws = mkLitA a
315 st_l1 <- addListToSS st_l0 ws
316 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
318 ptr (st_i0,st_l0,st_p0,st_I0) p
319 = do st_p1 <- addToSS st_p0 p
320 return (sizeSS st_p0, (st_i0,st_l0,st_p1,st_I0))
322 itbl (st_i0,st_l0,st_p0,st_I0) dcon
323 = do st_I1 <- addToSS st_I0 (getName dcon)
324 return (sizeSS st_I0, (st_i0,st_l0,st_p0,st_I1))
326 literal st (MachWord w) = int st (fromIntegral w)
327 literal st (MachInt j) = int st (fromIntegral j)
328 literal st (MachFloat r) = float st (fromRational r)
329 literal st (MachDouble r) = double st (fromRational r)
330 literal st (MachChar c) = int st c
331 literal st other = pprPanic "ByteCodeLink.literal" (ppr other)
334 = addr st ret_itbl_addr
338 PtrRep -> stg_ctoi_ret_R1p_info
339 WordRep -> stg_ctoi_ret_R1n_info
340 IntRep -> stg_ctoi_ret_R1n_info
341 AddrRep -> stg_ctoi_ret_R1n_info
342 CharRep -> stg_ctoi_ret_R1n_info
343 FloatRep -> stg_ctoi_ret_F1_info
344 DoubleRep -> stg_ctoi_ret_D1_info
345 VoidRep -> stg_ctoi_ret_V_info
346 other -> pprPanic "ByteCodeLink.ctoi_itbl" (ppr pk)
349 = addr st ret_itbl_addr
353 CharRep -> stg_gc_unbx_r1_ret_info
354 IntRep -> stg_gc_unbx_r1_ret_info
355 AddrRep -> stg_gc_unbx_r1_ret_info
356 FloatRep -> stg_gc_f1_ret_info
357 DoubleRep -> stg_gc_d1_ret_info
359 -- Interpreter.c spots this special case
360 other -> pprPanic "ByteCodeLink.itoc_itbl" (ppr pk)
362 foreign label "stg_ctoi_ret_R1p_info" stg_ctoi_ret_R1p_info :: Addr
363 foreign label "stg_ctoi_ret_R1n_info" stg_ctoi_ret_R1n_info :: Addr
364 foreign label "stg_ctoi_ret_F1_info" stg_ctoi_ret_F1_info :: Addr
365 foreign label "stg_ctoi_ret_D1_info" stg_ctoi_ret_D1_info :: Addr
366 foreign label "stg_ctoi_ret_V_info" stg_ctoi_ret_V_info :: Addr
368 foreign label "stg_gc_unbx_r1_ret_info" stg_gc_unbx_r1_ret_info :: Addr
369 foreign label "stg_gc_f1_ret_info" stg_gc_f1_ret_info :: Addr
370 foreign label "stg_gc_d1_ret_info" stg_gc_d1_ret_info :: Addr
372 -- The size in 16-bit entities of an instruction.
373 instrSize16s :: BCInstr -> Int
406 -- Make lists of host-sized words for literals, so that when the
407 -- words are placed in memory at increasing addresses, the
408 -- bit pattern is correct for the host's word size and endianness.
409 mkLitI :: Int -> [Word]
410 mkLitF :: Float -> [Word]
411 mkLitD :: Double -> [Word]
412 mkLitA :: Addr -> [Word]
416 arr <- newFloatArray ((0::Int),0)
417 writeFloatArray arr 0 f
418 f_arr <- castSTUArray arr
419 w0 <- readWordArray f_arr 0
426 arr <- newDoubleArray ((0::Int),1)
427 writeDoubleArray arr 0 d
428 d_arr <- castSTUArray arr
429 w0 <- readWordArray d_arr 0
430 w1 <- readWordArray d_arr 1
435 arr <- newDoubleArray ((0::Int),0)
436 writeDoubleArray arr 0 d
437 d_arr <- castSTUArray arr
438 w0 <- readWordArray d_arr 0
444 arr <- newIntArray ((0::Int),0)
445 writeIntArray arr 0 i
446 i_arr <- castSTUArray arr
447 w0 <- readWordArray i_arr 0
453 arr <- newAddrArray ((0::Int),0)
454 writeAddrArray arr 0 a
455 a_arr <- castSTUArray arr
456 w0 <- readWordArray a_arr 0
462 %************************************************************************
464 \subsection{Linking interpretables into something we can run}
466 %************************************************************************
471 data BCO# = BCO# ByteArray# -- instrs :: Array Word16#
472 ByteArray# -- literals :: Array Word32#
473 PtrArray# -- ptrs :: Array HValue
474 ByteArray# -- itbls :: Array Addr#
477 linkBCO ie ce (UnlinkedBCO nm insnsSS literalsSS ptrsSS itblsSS)
478 = do insns <- listFromSS insnsSS
479 literals <- listFromSS literalsSS
480 ptrs <- listFromSS ptrsSS
481 itbls <- listFromSS itblsSS
483 linked_ptrs <- mapM (lookupCE ce) ptrs
484 linked_itbls <- mapM (lookupIE ie) itbls
486 let n_insns = sizeSS insnsSS
487 n_literals = sizeSS literalsSS
488 n_ptrs = sizeSS ptrsSS
489 n_itbls = sizeSS itblsSS
491 let ptrs_arr = array (0, n_ptrs-1) (indexify linked_ptrs)
493 ptrs_parr = case ptrs_arr of Array lo hi parr -> parr
495 itbls_arr = array (0, n_itbls-1) (indexify linked_itbls)
496 :: UArray Int ItblPtr
497 itbls_barr = case itbls_arr of UArray lo hi barr -> barr
499 insns_arr | n_insns > 65535
500 = panic "linkBCO: >= 64k insns in BCO"
503 (indexify (fromIntegral n_insns:insns))
505 insns_barr = case insns_arr of UArray lo hi barr -> barr
507 literals_arr = array (0, n_literals-1) (indexify literals)
509 literals_barr = case literals_arr of UArray lo hi barr -> barr
511 indexify :: [a] -> [(Int, a)]
512 indexify xs = zip [0..] xs
514 BCO bco# <- newBCO insns_barr literals_barr ptrs_parr itbls_barr
516 -- WAS: return (unsafeCoerce# bco#)
517 case mkApUpd0# (unsafeCoerce# bco#) of
518 (# final_bco #) -> return final_bco
523 newBCO :: ByteArray# -> ByteArray# -> Array# a -> ByteArray# -> IO BCO
525 = IO (\s -> case newBCO# a b c d s of (# s1, bco #) -> (# s1, BCO bco #))
528 lookupCE :: ClosureEnv -> Either Name PrimOp -> IO HValue
529 lookupCE ce (Right primop)
530 = do m <- lookupSymbol (primopToCLabel primop "closure")
532 Just (Ptr addr) -> case addrToHValue# addr of
533 (# hval #) -> return hval
534 Nothing -> pprPanic "ByteCodeLink.lookupCE(primop)" (ppr primop)
535 lookupCE ce (Left nm)
536 = case lookupFM ce nm of
539 -> do m <- lookupSymbol (nameToCLabel nm "closure")
541 Just (Ptr addr) -> case addrToHValue# addr of
542 (# hval #) -> return hval
543 Nothing -> pprPanic "ByteCodeLink.lookupCE" (ppr nm)
545 lookupIE :: ItblEnv -> Name -> IO (Ptr a)
547 = case lookupFM ie con_nm of
548 Just (Ptr a) -> return (Ptr a)
550 -> do -- try looking up in the object files.
551 m <- lookupSymbol (nameToCLabel con_nm "con_info")
553 Just addr -> return addr
555 -> do -- perhaps a nullary constructor?
556 n <- lookupSymbol (nameToCLabel con_nm "static_info")
558 Just addr -> return addr
559 Nothing -> pprPanic "ByteCodeLink.lookupIE" (ppr con_nm)
561 -- HACKS!!! ToDo: cleaner
562 nameToCLabel :: Name -> String{-suffix-} -> String
563 nameToCLabel n suffix
564 = _UNPK_(moduleNameFS (rdrNameModule rn))
565 ++ '_':occNameString(rdrNameOcc rn) ++ '_':suffix
566 where rn = toRdrName n
568 primopToCLabel :: PrimOp -> String{-suffix-} -> String
569 primopToCLabel primop suffix
570 = let str = "PrelPrimopWrappers_" ++ occNameString (primOpOcc primop) ++ '_':suffix
571 in --trace ("primopToCLabel: " ++ str)
576 %************************************************************************
578 \subsection{Connect to actual values for bytecode opcodes}
580 %************************************************************************
584 #include "Bytecodes.h"
586 i_ARGCHECK = (bci_ARGCHECK :: Int)
587 i_PUSH_L = (bci_PUSH_L :: Int)
588 i_PUSH_LL = (bci_PUSH_LL :: Int)
589 i_PUSH_LLL = (bci_PUSH_LLL :: Int)
590 i_PUSH_G = (bci_PUSH_G :: Int)
591 i_PUSH_AS = (bci_PUSH_AS :: Int)
592 i_PUSH_UBX = (bci_PUSH_UBX :: Int)
593 i_PUSH_TAG = (bci_PUSH_TAG :: Int)
594 i_SLIDE = (bci_SLIDE :: Int)
595 i_ALLOC = (bci_ALLOC :: Int)
596 i_MKAP = (bci_MKAP :: Int)
597 i_UNPACK = (bci_UNPACK :: Int)
598 i_UPK_TAG = (bci_UPK_TAG :: Int)
599 i_PACK = (bci_PACK :: Int)
600 i_TESTLT_I = (bci_TESTLT_I :: Int)
601 i_TESTEQ_I = (bci_TESTEQ_I :: Int)
602 i_TESTLT_F = (bci_TESTLT_F :: Int)
603 i_TESTEQ_F = (bci_TESTEQ_F :: Int)
604 i_TESTLT_D = (bci_TESTLT_D :: Int)
605 i_TESTEQ_D = (bci_TESTEQ_D :: Int)
606 i_TESTLT_P = (bci_TESTLT_P :: Int)
607 i_TESTEQ_P = (bci_TESTEQ_P :: Int)
608 i_CASEFAIL = (bci_CASEFAIL :: Int)
609 i_ENTER = (bci_ENTER :: Int)
610 i_RETURN = (bci_RETURN :: Int)
611 i_STKCHECK = (bci_STKCHECK :: Int)
612 i_JMP = (bci_JMP :: Int)
614 i_CCALL = (bci_CCALL :: Int)
616 i_CCALL = error "Sorry pal, you need to bootstrap to use i_CCALL."
619 iNTERP_STACK_CHECK_THRESH = (INTERP_STACK_CHECK_THRESH :: Int)