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 ( 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(..) )
43 import Addr ( Word, Addr, nullAddr )
46 import PrelBase ( Int(..) )
47 import PrelGHC ( BCO#, newBCO#, unsafeCoerce#,
48 ByteArray#, Array#, addrToHValue#, mkApUpd0# )
49 import IOExts ( fixIO )
50 import PrelArr ( Array(..) )
51 import ArrayBase ( UArray(..) )
52 import PrelIOBase ( IO(..) )
56 %************************************************************************
58 \subsection{Top-level stuff}
60 %************************************************************************
64 linkIModules :: ItblEnv -- incoming global itbl env; returned updated
65 -> ClosureEnv -- incoming global closure env; returned updated
66 -> [([UnlinkedBCO], ItblEnv)]
67 -> IO ([HValue], ItblEnv, ClosureEnv)
68 linkIModules gie gce mods
69 = do let (bcoss, ies) = unzip mods
71 final_gie = foldr plusFM gie ies
72 (final_gce, linked_bcos) <- linkSomeBCOs True final_gie gce bcos
73 return (linked_bcos, final_gie, final_gce)
76 linkIExpr :: ItblEnv -> ClosureEnv -> UnlinkedBCOExpr
77 -> IO HValue -- IO BCO# really
78 linkIExpr ie ce (root_ul_bco, aux_ul_bcos)
79 = do (aux_ce, _) <- linkSomeBCOs False ie ce aux_ul_bcos
80 (_, [root_bco]) <- linkSomeBCOs False ie aux_ce [root_ul_bco]
83 -- Link a bunch of BCOs and return them + updated closure env.
84 linkSomeBCOs :: Bool -- False <=> add _all_ BCOs to returned closure env
85 -- True <=> add only toplevel BCOs to closure env
89 -> IO (ClosureEnv, [HValue])
90 linkSomeBCOs toplevs_only ie ce_in ul_bcos
91 = do let nms = map nameOfUnlinkedBCO ul_bcos
93 ( \ hvs -> let ce_out = addListToFM ce_in (zipLazily nms hvs)
94 in mapM (linkBCO ie ce_out) ul_bcos )
96 let ce_all_additions = zip nms hvals
97 ce_top_additions = filter (isGlobalName.fst) ce_all_additions
98 ce_additions = if toplevs_only then ce_top_additions
100 ce_out = -- make sure we're not inserting duplicate names into the
101 -- closure environment, which leads to trouble.
102 ASSERT (all (not . (`elemFM` ce_in)) (map fst ce_additions))
103 addListToFM ce_in ce_additions
104 return (ce_out, hvals)
106 -- A lazier zip, in which no demand is propagated to the second
107 -- list unless some demand is propagated to the snd of one of the
108 -- result list elems.
110 zipLazily (x:xs) ys = (x, head ys) : zipLazily xs (tail ys)
115 (SizedSeq Word16) -- insns
116 (SizedSeq Word) -- literals
117 (SizedSeq (Either Name PrimOp)) -- ptrs
118 (SizedSeq Name) -- itbl refs
120 nameOfUnlinkedBCO (UnlinkedBCO nm _ _ _ _) = nm
122 -- When translating expressions, we need to distinguish the root
123 -- BCO for the expression
124 type UnlinkedBCOExpr = (UnlinkedBCO, [UnlinkedBCO])
126 instance Outputable UnlinkedBCO where
127 ppr (UnlinkedBCO nm insns lits ptrs itbls)
128 = sep [text "BCO", ppr nm, text "with",
129 int (sizeSS insns), text "insns",
130 int (sizeSS lits), text "lits",
131 int (sizeSS ptrs), text "ptrs",
132 int (sizeSS itbls), text "itbls"]
135 -- these need a proper home
136 type ClosureEnv = FiniteMap Name HValue
137 data HValue = HValue -- dummy type, actually a pointer to some Real Code.
139 -- remove all entries for a given set of modules from the environment;
140 -- note that this removes all local names too (ie. temporary bindings from
141 -- the command line).
142 filterNameMap :: [ModuleName] -> FiniteMap Name a -> FiniteMap Name a
143 filterNameMap mods env
144 = filterFM (\n _ -> isGlobalName n &&
145 moduleName (nameModule n) `elem` mods) env
148 %************************************************************************
150 \subsection{The bytecode assembler}
152 %************************************************************************
154 The object format for bytecodes is: 16 bits for the opcode, and 16 for
155 each field -- so the code can be considered a sequence of 16-bit ints.
156 Each field denotes either a stack offset or number of items on the
157 stack (eg SLIDE), and index into the pointer table (eg PUSH_G), an
158 index into the literal table (eg PUSH_I/D/L), or a bytecode address in
162 -- Top level assembler fn.
163 assembleBCO :: ProtoBCO Name -> IO UnlinkedBCO
165 assembleBCO (ProtoBCO nm instrs origin)
167 -- pass 1: collect up the offsets of the local labels.
168 -- Remember that the first insn starts at offset 1 since offset 0
169 -- (eventually) will hold the total # of insns.
170 label_env = mkLabelEnv emptyFM 1 instrs
172 mkLabelEnv env i_offset [] = env
173 mkLabelEnv env i_offset (i:is)
175 = case i of LABEL n -> addToFM env n i_offset ; _ -> env
176 in mkLabelEnv new_env (i_offset + instrSize16s i) is
179 = case lookupFM label_env lab of
180 Just bco_offset -> bco_offset
181 Nothing -> pprPanic "assembleBCO.findLabel" (int lab)
183 do -- pass 2: generate the instruction, ptr and nonptr bits
184 insns <- return emptySS :: IO (SizedSeq Word16)
185 lits <- return emptySS :: IO (SizedSeq Word)
186 ptrs <- return emptySS :: IO (SizedSeq (Either Name PrimOp))
187 itbls <- return emptySS :: IO (SizedSeq Name)
188 let init_asm_state = (insns,lits,ptrs,itbls)
189 (final_insns, final_lits, final_ptrs, final_itbls)
190 <- mkBits findLabel init_asm_state instrs
192 return (UnlinkedBCO nm final_insns final_lits final_ptrs final_itbls)
194 -- instrs nonptrs ptrs itbls
195 type AsmState = (SizedSeq Word16, SizedSeq Word,
196 SizedSeq (Either Name PrimOp), SizedSeq Name)
198 data SizedSeq a = SizedSeq !Int [a]
199 emptySS = SizedSeq 0 []
200 addToSS (SizedSeq n r_xs) x = return (SizedSeq (n+1) (x:r_xs))
201 addListToSS (SizedSeq n r_xs) xs
202 = return (SizedSeq (n + length xs) (reverse xs ++ r_xs))
203 sizeSS (SizedSeq n r_xs) = n
204 listFromSS (SizedSeq n r_xs) = return (reverse r_xs)
207 -- This is where all the action is (pass 2 of the assembler)
208 mkBits :: (Int -> Int) -- label finder
210 -> [BCInstr] -- instructions (in)
213 mkBits findLabel st proto_insns
214 = foldM doInstr st proto_insns
216 doInstr :: AsmState -> BCInstr -> IO AsmState
219 ARGCHECK n -> instr2 st i_ARGCHECK n
220 STKCHECK n -> instr2 st i_STKCHECK n
221 PUSH_L o1 -> instr2 st i_PUSH_L o1
222 PUSH_LL o1 o2 -> instr3 st i_PUSH_LL o1 o2
223 PUSH_LLL o1 o2 o3 -> instr4 st i_PUSH_LLL o1 o2 o3
224 PUSH_G nm -> do (p, st2) <- ptr st nm
225 instr2 st2 i_PUSH_G p
226 PUSH_AS nm pk -> do (p, st2) <- ptr st (Left nm)
227 (np, st3) <- ctoi_itbl st2 pk
228 instr3 st3 i_PUSH_AS p np
229 PUSH_UBX (Left lit) nws
230 -> do (np, st2) <- literal st lit
231 instr3 st2 i_PUSH_UBX np nws
232 PUSH_UBX (Right aa) nws
233 -> do (np, st2) <- addr st aa
234 instr3 st2 i_PUSH_UBX np nws
236 PUSH_TAG tag -> instr2 st i_PUSH_TAG tag
237 SLIDE n by -> instr3 st i_SLIDE n by
238 ALLOC n -> instr2 st i_ALLOC n
239 MKAP off sz -> instr3 st i_MKAP off sz
240 UNPACK n -> instr2 st i_UNPACK n
241 UPK_TAG n m k -> instr4 st i_UPK_TAG n m k
242 PACK dcon sz -> do (itbl_no,st2) <- itbl st dcon
243 instr3 st2 i_PACK itbl_no sz
244 LABEL lab -> return st
245 TESTLT_I i l -> do (np, st2) <- int st i
246 instr3 st2 i_TESTLT_I np (findLabel l)
247 TESTEQ_I i l -> do (np, st2) <- int st i
248 instr3 st2 i_TESTEQ_I np (findLabel l)
249 TESTLT_F f l -> do (np, st2) <- float st f
250 instr3 st2 i_TESTLT_F np (findLabel l)
251 TESTEQ_F f l -> do (np, st2) <- float st f
252 instr3 st2 i_TESTEQ_F np (findLabel l)
253 TESTLT_D d l -> do (np, st2) <- double st d
254 instr3 st2 i_TESTLT_D np (findLabel l)
255 TESTEQ_D d l -> do (np, st2) <- double st d
256 instr3 st2 i_TESTEQ_D np (findLabel l)
257 TESTLT_P i l -> instr3 st i_TESTLT_P i (findLabel l)
258 TESTEQ_P i l -> instr3 st i_TESTEQ_P i (findLabel l)
259 CASEFAIL -> instr1 st i_CASEFAIL
260 JMP l -> instr2 st i_JMP (findLabel l)
261 ENTER -> instr1 st i_ENTER
262 RETURN rep -> do (itbl_no,st2) <- itoc_itbl st rep
263 instr2 st2 i_RETURN itbl_no
264 CCALL m_addr -> do (np, st2) <- addr st m_addr
265 instr2 st2 i_CCALL np
270 instr1 (st_i0,st_l0,st_p0,st_I0) i1
271 = do st_i1 <- addToSS st_i0 (i2s i1)
272 return (st_i1,st_l0,st_p0,st_I0)
274 instr2 (st_i0,st_l0,st_p0,st_I0) i1 i2
275 = do st_i1 <- addToSS st_i0 (i2s i1)
276 st_i2 <- addToSS st_i1 (i2s i2)
277 return (st_i2,st_l0,st_p0,st_I0)
279 instr3 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3
280 = do st_i1 <- addToSS st_i0 (i2s i1)
281 st_i2 <- addToSS st_i1 (i2s i2)
282 st_i3 <- addToSS st_i2 (i2s i3)
283 return (st_i3,st_l0,st_p0,st_I0)
285 instr4 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3 i4
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 st_i4 <- addToSS st_i3 (i2s i4)
290 return (st_i4,st_l0,st_p0,st_I0)
292 float (st_i0,st_l0,st_p0,st_I0) f
293 = do let ws = mkLitF f
294 st_l1 <- addListToSS st_l0 ws
295 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
297 double (st_i0,st_l0,st_p0,st_I0) d
298 = do let ws = mkLitD d
299 st_l1 <- addListToSS st_l0 ws
300 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
302 int (st_i0,st_l0,st_p0,st_I0) i
303 = do let ws = mkLitI i
304 st_l1 <- addListToSS st_l0 ws
305 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
307 addr (st_i0,st_l0,st_p0,st_I0) a
308 = do let ws = mkLitA a
309 st_l1 <- addListToSS st_l0 ws
310 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
312 ptr (st_i0,st_l0,st_p0,st_I0) p
313 = do st_p1 <- addToSS st_p0 p
314 return (sizeSS st_p0, (st_i0,st_l0,st_p1,st_I0))
316 itbl (st_i0,st_l0,st_p0,st_I0) dcon
317 = do st_I1 <- addToSS st_I0 (getName dcon)
318 return (sizeSS st_I0, (st_i0,st_l0,st_p0,st_I1))
320 literal st (MachWord w) = int st (fromIntegral w)
321 literal st (MachInt j) = int st (fromIntegral j)
322 literal st (MachFloat r) = float st (fromRational r)
323 literal st (MachDouble r) = double st (fromRational r)
324 literal st (MachChar c) = int st c
325 literal st other = pprPanic "ByteCodeLink.literal" (ppr other)
328 = addr st ret_itbl_addr
332 PtrRep -> stg_ctoi_ret_R1p_info
333 WordRep -> stg_ctoi_ret_R1n_info
334 IntRep -> stg_ctoi_ret_R1n_info
335 AddrRep -> stg_ctoi_ret_R1n_info
336 CharRep -> stg_ctoi_ret_R1n_info
337 FloatRep -> stg_ctoi_ret_F1_info
338 DoubleRep -> stg_ctoi_ret_D1_info
339 VoidRep -> stg_ctoi_ret_V_info
340 other -> pprPanic "ByteCodeLink.ctoi_itbl" (ppr pk)
343 = addr st ret_itbl_addr
347 CharRep -> stg_gc_unbx_r1_ret_info
348 IntRep -> stg_gc_unbx_r1_ret_info
349 AddrRep -> stg_gc_unbx_r1_ret_info
350 FloatRep -> stg_gc_f1_ret_info
351 DoubleRep -> stg_gc_d1_ret_info
353 -- Interpreter.c spots this special case
354 other -> pprPanic "ByteCodeLink.itoc_itbl" (ppr pk)
356 foreign label "stg_ctoi_ret_R1p_info" stg_ctoi_ret_R1p_info :: Addr
357 foreign label "stg_ctoi_ret_R1n_info" stg_ctoi_ret_R1n_info :: Addr
358 foreign label "stg_ctoi_ret_F1_info" stg_ctoi_ret_F1_info :: Addr
359 foreign label "stg_ctoi_ret_D1_info" stg_ctoi_ret_D1_info :: Addr
360 foreign label "stg_ctoi_ret_V_info" stg_ctoi_ret_V_info :: Addr
362 foreign label "stg_gc_unbx_r1_ret_info" stg_gc_unbx_r1_ret_info :: Addr
363 foreign label "stg_gc_f1_ret_info" stg_gc_f1_ret_info :: Addr
364 foreign label "stg_gc_d1_ret_info" stg_gc_d1_ret_info :: Addr
366 -- The size in 16-bit entities of an instruction.
367 instrSize16s :: BCInstr -> Int
400 -- Make lists of host-sized words for literals, so that when the
401 -- words are placed in memory at increasing addresses, the
402 -- bit pattern is correct for the host's word size and endianness.
403 mkLitI :: Int -> [Word]
404 mkLitF :: Float -> [Word]
405 mkLitD :: Double -> [Word]
406 mkLitA :: Addr -> [Word]
410 arr <- newFloatArray ((0::Int),0)
411 writeFloatArray arr 0 f
412 f_arr <- castSTUArray arr
413 w0 <- readWordArray f_arr 0
420 arr <- newDoubleArray ((0::Int),1)
421 writeDoubleArray arr 0 d
422 d_arr <- castSTUArray arr
423 w0 <- readWordArray d_arr 0
424 w1 <- readWordArray d_arr 1
429 arr <- newDoubleArray ((0::Int),0)
430 writeDoubleArray arr 0 d
431 d_arr <- castSTUArray arr
432 w0 <- readWordArray d_arr 0
438 arr <- newIntArray ((0::Int),0)
439 writeIntArray arr 0 i
440 i_arr <- castSTUArray arr
441 w0 <- readWordArray i_arr 0
447 arr <- newAddrArray ((0::Int),0)
448 writeAddrArray arr 0 a
449 a_arr <- castSTUArray arr
450 w0 <- readWordArray a_arr 0
456 %************************************************************************
458 \subsection{Linking interpretables into something we can run}
460 %************************************************************************
465 data BCO# = BCO# ByteArray# -- instrs :: Array Word16#
466 ByteArray# -- literals :: Array Word32#
467 PtrArray# -- ptrs :: Array HValue
468 ByteArray# -- itbls :: Array Addr#
471 linkBCO ie ce (UnlinkedBCO nm insnsSS literalsSS ptrsSS itblsSS)
472 = do insns <- listFromSS insnsSS
473 literals <- listFromSS literalsSS
474 ptrs <- listFromSS ptrsSS
475 itbls <- listFromSS itblsSS
477 linked_ptrs <- mapM (lookupCE ce) ptrs
478 linked_itbls <- mapM (lookupIE ie) itbls
480 let n_insns = sizeSS insnsSS
481 n_literals = sizeSS literalsSS
482 n_ptrs = sizeSS ptrsSS
483 n_itbls = sizeSS itblsSS
485 let ptrs_arr = array (0, n_ptrs-1) (indexify linked_ptrs)
487 ptrs_parr = case ptrs_arr of Array lo hi parr -> parr
489 itbls_arr = array (0, n_itbls-1) (indexify linked_itbls)
490 :: UArray Int ItblPtr
491 itbls_barr = case itbls_arr of UArray lo hi barr -> barr
493 insns_arr | n_insns > 65535
494 = panic "linkBCO: >= 64k insns in BCO"
497 (indexify (fromIntegral n_insns:insns))
499 insns_barr = case insns_arr of UArray lo hi barr -> barr
501 literals_arr = array (0, n_literals-1) (indexify literals)
503 literals_barr = case literals_arr of UArray lo hi barr -> barr
505 indexify :: [a] -> [(Int, a)]
506 indexify xs = zip [0..] xs
508 BCO bco# <- newBCO insns_barr literals_barr ptrs_parr itbls_barr
510 -- WAS: return (unsafeCoerce# bco#)
511 case mkApUpd0# (unsafeCoerce# bco#) of
512 (# final_bco #) -> return final_bco
517 newBCO :: ByteArray# -> ByteArray# -> Array# a -> ByteArray# -> IO BCO
519 = IO (\s -> case newBCO# a b c d s of (# s1, bco #) -> (# s1, BCO bco #))
522 lookupCE :: ClosureEnv -> Either Name PrimOp -> IO HValue
523 lookupCE ce (Right primop)
524 = do m <- lookupSymbol (primopToCLabel primop "closure")
526 Just (Ptr addr) -> case addrToHValue# addr of
527 (# hval #) -> return hval
528 Nothing -> pprPanic "ByteCodeLink.lookupCE(primop)" (ppr primop)
529 lookupCE ce (Left nm)
530 = case lookupFM ce nm of
533 -> do m <- lookupSymbol (nameToCLabel nm "closure")
535 Just (Ptr addr) -> case addrToHValue# addr of
536 (# hval #) -> return hval
537 Nothing -> pprPanic "ByteCodeLink.lookupCE" (ppr nm)
539 lookupIE :: ItblEnv -> Name -> IO (Ptr a)
541 = case lookupFM ie con_nm of
542 Just (Ptr a) -> return (Ptr a)
544 -> do -- try looking up in the object files.
545 m <- lookupSymbol (nameToCLabel con_nm "con_info")
547 Just addr -> return addr
549 -> do -- perhaps a nullary constructor?
550 n <- lookupSymbol (nameToCLabel con_nm "static_info")
552 Just addr -> return addr
553 Nothing -> pprPanic "ByteCodeLink.lookupIE" (ppr con_nm)
555 -- HACKS!!! ToDo: cleaner
556 nameToCLabel :: Name -> String{-suffix-} -> String
557 nameToCLabel n suffix
558 = _UNPK_(moduleNameFS (rdrNameModule rn))
559 ++ '_':occNameString(rdrNameOcc rn) ++ '_':suffix
560 where rn = toRdrName n
562 primopToCLabel :: PrimOp -> String{-suffix-} -> String
563 primopToCLabel primop suffix
564 = let str = "PrelPrimopWrappers_" ++ occNameString (primOpOcc primop) ++ '_':suffix
565 in --trace ("primopToCLabel: " ++ str)
570 %************************************************************************
572 \subsection{Connect to actual values for bytecode opcodes}
574 %************************************************************************
578 #include "Bytecodes.h"
580 i_ARGCHECK = (bci_ARGCHECK :: Int)
581 i_PUSH_L = (bci_PUSH_L :: Int)
582 i_PUSH_LL = (bci_PUSH_LL :: Int)
583 i_PUSH_LLL = (bci_PUSH_LLL :: Int)
584 i_PUSH_G = (bci_PUSH_G :: Int)
585 i_PUSH_AS = (bci_PUSH_AS :: Int)
586 i_PUSH_UBX = (bci_PUSH_UBX :: Int)
587 i_PUSH_TAG = (bci_PUSH_TAG :: Int)
588 i_SLIDE = (bci_SLIDE :: Int)
589 i_ALLOC = (bci_ALLOC :: Int)
590 i_MKAP = (bci_MKAP :: Int)
591 i_UNPACK = (bci_UNPACK :: Int)
592 i_UPK_TAG = (bci_UPK_TAG :: Int)
593 i_PACK = (bci_PACK :: Int)
594 i_TESTLT_I = (bci_TESTLT_I :: Int)
595 i_TESTEQ_I = (bci_TESTEQ_I :: Int)
596 i_TESTLT_F = (bci_TESTLT_F :: Int)
597 i_TESTEQ_F = (bci_TESTEQ_F :: Int)
598 i_TESTLT_D = (bci_TESTLT_D :: Int)
599 i_TESTEQ_D = (bci_TESTEQ_D :: Int)
600 i_TESTLT_P = (bci_TESTLT_P :: Int)
601 i_TESTEQ_P = (bci_TESTEQ_P :: Int)
602 i_CASEFAIL = (bci_CASEFAIL :: Int)
603 i_ENTER = (bci_ENTER :: Int)
604 i_RETURN = (bci_RETURN :: Int)
605 i_STKCHECK = (bci_STKCHECK :: Int)
606 i_JMP = (bci_JMP :: Int)
608 i_CCALL = (bci_CCALL :: Int)
610 i_CCALL = error "Sorry pal, you need to bootstrap to use i_CCALL."
613 iNTERP_STACK_CHECK_THRESH = (INTERP_STACK_CHECK_THRESH :: Int)