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 MArray ( castSTUArray,
36 newFloatArray, writeFloatArray,
37 newDoubleArray, writeDoubleArray,
38 newIntArray, writeIntArray,
39 newAddrArray, writeAddrArray )
40 import Foreign ( Word16, Ptr(..) )
41 import Addr ( Word, Addr, nullAddr )
44 import PrelBase ( Int(..) )
45 import PrelGHC ( BCO#, newBCO#, unsafeCoerce#,
46 ByteArray#, Array#, addrToHValue#, mkApUpd0# )
47 import IOExts ( fixIO )
49 import PrelArr ( Array(..) )
50 import PrelIOBase ( IO(..) )
54 %************************************************************************
56 \subsection{Top-level stuff}
58 %************************************************************************
62 linkIModules :: ItblEnv -- incoming global itbl env; returned updated
63 -> ClosureEnv -- incoming global closure env; returned updated
64 -> [([UnlinkedBCO], ItblEnv)]
65 -> IO ([HValue], ItblEnv, ClosureEnv)
66 linkIModules gie gce mods
67 = do let (bcoss, ies) = unzip mods
69 final_gie = foldr plusFM gie ies
70 (final_gce, linked_bcos) <- linkSomeBCOs True final_gie gce bcos
71 return (linked_bcos, final_gie, final_gce)
74 linkIExpr :: ItblEnv -> ClosureEnv -> UnlinkedBCOExpr
75 -> IO HValue -- IO BCO# really
76 linkIExpr ie ce (root_ul_bco, aux_ul_bcos)
77 = do (aux_ce, _) <- linkSomeBCOs False ie ce aux_ul_bcos
78 (_, [root_bco]) <- linkSomeBCOs False ie aux_ce [root_ul_bco]
81 -- Link a bunch of BCOs and return them + updated closure env.
82 linkSomeBCOs :: Bool -- False <=> add _all_ BCOs to returned closure env
83 -- True <=> add only toplevel BCOs to closure env
87 -> IO (ClosureEnv, [HValue])
88 linkSomeBCOs toplevs_only ie ce_in ul_bcos
89 = do let nms = map nameOfUnlinkedBCO ul_bcos
91 ( \ hvs -> let ce_out = addListToFM ce_in (zipLazily nms hvs)
92 in mapM (linkBCO ie ce_out) ul_bcos )
94 let ce_all_additions = zip nms hvals
95 ce_top_additions = filter (isGlobalName.fst) ce_all_additions
96 ce_additions = if toplevs_only then ce_top_additions
98 ce_out = -- make sure we're not inserting duplicate names into the
99 -- closure environment, which leads to trouble.
100 ASSERT (all (not . (`elemFM` ce_in)) (map fst ce_additions))
101 addListToFM ce_in ce_additions
102 return (ce_out, hvals)
104 -- A lazier zip, in which no demand is propagated to the second
105 -- list unless some demand is propagated to the snd of one of the
106 -- result list elems.
108 zipLazily (x:xs) ys = (x, head ys) : zipLazily xs (tail ys)
113 (SizedSeq Word16) -- insns
114 (SizedSeq Word) -- literals
115 (SizedSeq (Either Name PrimOp)) -- ptrs
116 (SizedSeq Name) -- itbl refs
118 nameOfUnlinkedBCO (UnlinkedBCO nm _ _ _ _) = nm
120 -- When translating expressions, we need to distinguish the root
121 -- BCO for the expression
122 type UnlinkedBCOExpr = (UnlinkedBCO, [UnlinkedBCO])
124 instance Outputable UnlinkedBCO where
125 ppr (UnlinkedBCO nm insns lits ptrs itbls)
126 = sep [text "BCO", ppr nm, text "with",
127 int (sizeSS insns), text "insns",
128 int (sizeSS lits), text "lits",
129 int (sizeSS ptrs), text "ptrs",
130 int (sizeSS itbls), text "itbls"]
133 -- these need a proper home
134 type ClosureEnv = FiniteMap Name HValue
135 data HValue = HValue -- dummy type, actually a pointer to some Real Code.
137 -- remove all entries for a given set of modules from the environment;
138 -- note that this removes all local names too (ie. temporary bindings from
139 -- the command line).
140 filterNameMap :: [ModuleName] -> FiniteMap Name a -> FiniteMap Name a
141 filterNameMap mods env
142 = filterFM (\n _ -> isGlobalName n &&
143 moduleName (nameModule n) `elem` mods) env
146 %************************************************************************
148 \subsection{The bytecode assembler}
150 %************************************************************************
152 The object format for bytecodes is: 16 bits for the opcode, and 16 for
153 each field -- so the code can be considered a sequence of 16-bit ints.
154 Each field denotes either a stack offset or number of items on the
155 stack (eg SLIDE), and index into the pointer table (eg PUSH_G), an
156 index into the literal table (eg PUSH_I/D/L), or a bytecode address in
160 -- Top level assembler fn.
161 assembleBCO :: ProtoBCO Name -> IO UnlinkedBCO
163 assembleBCO (ProtoBCO nm instrs origin)
165 -- pass 1: collect up the offsets of the local labels.
166 -- Remember that the first insn starts at offset 1 since offset 0
167 -- (eventually) will hold the total # of insns.
168 label_env = mkLabelEnv emptyFM 1 instrs
170 mkLabelEnv env i_offset [] = env
171 mkLabelEnv env i_offset (i:is)
173 = case i of LABEL n -> addToFM env n i_offset ; _ -> env
174 in mkLabelEnv new_env (i_offset + instrSize16s i) is
177 = case lookupFM label_env lab of
178 Just bco_offset -> bco_offset
179 Nothing -> pprPanic "assembleBCO.findLabel" (int lab)
181 do -- pass 2: generate the instruction, ptr and nonptr bits
182 insns <- return emptySS :: IO (SizedSeq Word16)
183 lits <- return emptySS :: IO (SizedSeq Word)
184 ptrs <- return emptySS :: IO (SizedSeq (Either Name PrimOp))
185 itbls <- return emptySS :: IO (SizedSeq Name)
186 let init_asm_state = (insns,lits,ptrs,itbls)
187 (final_insns, final_lits, final_ptrs, final_itbls)
188 <- mkBits findLabel init_asm_state instrs
190 return (UnlinkedBCO nm final_insns final_lits final_ptrs final_itbls)
192 -- instrs nonptrs ptrs itbls
193 type AsmState = (SizedSeq Word16, SizedSeq Word,
194 SizedSeq (Either Name PrimOp), SizedSeq Name)
196 data SizedSeq a = SizedSeq !Int [a]
197 emptySS = SizedSeq 0 []
198 addToSS (SizedSeq n r_xs) x = return (SizedSeq (n+1) (x:r_xs))
199 addListToSS (SizedSeq n r_xs) xs
200 = return (SizedSeq (n + length xs) (reverse xs ++ r_xs))
201 sizeSS (SizedSeq n r_xs) = n
202 listFromSS (SizedSeq n r_xs) = return (reverse r_xs)
205 -- This is where all the action is (pass 2 of the assembler)
206 mkBits :: (Int -> Int) -- label finder
208 -> [BCInstr] -- instructions (in)
211 mkBits findLabel st proto_insns
212 = foldM doInstr st proto_insns
214 doInstr :: AsmState -> BCInstr -> IO AsmState
217 ARGCHECK n -> instr2 st i_ARGCHECK n
218 STKCHECK n -> instr2 st i_STKCHECK n
219 PUSH_L o1 -> instr2 st i_PUSH_L o1
220 PUSH_LL o1 o2 -> instr3 st i_PUSH_LL o1 o2
221 PUSH_LLL o1 o2 o3 -> instr4 st i_PUSH_LLL o1 o2 o3
222 PUSH_G nm -> do (p, st2) <- ptr st nm
223 instr2 st2 i_PUSH_G p
224 PUSH_AS nm pk -> do (p, st2) <- ptr st (Left nm)
225 (np, st3) <- ctoi_itbl st2 pk
226 instr3 st3 i_PUSH_AS p np
227 PUSH_UBX lit nws -> do (np, st2) <- literal st lit
228 instr3 st2 i_PUSH_UBX np nws
229 PUSH_TAG tag -> instr2 st i_PUSH_TAG tag
230 SLIDE n by -> instr3 st i_SLIDE n by
231 ALLOC n -> instr2 st i_ALLOC n
232 MKAP off sz -> instr3 st i_MKAP off sz
233 UNPACK n -> instr2 st i_UNPACK n
234 UPK_TAG n m k -> instr4 st i_UPK_TAG n m k
235 PACK dcon sz -> do (itbl_no,st2) <- itbl st dcon
236 instr3 st2 i_PACK itbl_no sz
237 LABEL lab -> return st
238 TESTLT_I i l -> do (np, st2) <- int st i
239 instr3 st2 i_TESTLT_I np (findLabel l)
240 TESTEQ_I i l -> do (np, st2) <- int st i
241 instr3 st2 i_TESTEQ_I np (findLabel l)
242 TESTLT_F f l -> do (np, st2) <- float st f
243 instr3 st2 i_TESTLT_F np (findLabel l)
244 TESTEQ_F f l -> do (np, st2) <- float st f
245 instr3 st2 i_TESTEQ_F np (findLabel l)
246 TESTLT_D d l -> do (np, st2) <- double st d
247 instr3 st2 i_TESTLT_D np (findLabel l)
248 TESTEQ_D d l -> do (np, st2) <- double st d
249 instr3 st2 i_TESTEQ_D np (findLabel l)
250 TESTLT_P i l -> instr3 st i_TESTLT_P i (findLabel l)
251 TESTEQ_P i l -> instr3 st i_TESTEQ_P i (findLabel l)
252 CASEFAIL -> instr1 st i_CASEFAIL
253 JMP l -> instr2 st i_JMP (findLabel l)
254 ENTER -> instr1 st i_ENTER
255 RETURN rep -> do (itbl_no,st2) <- itoc_itbl st rep
256 instr2 st2 i_RETURN itbl_no
261 instr1 (st_i0,st_l0,st_p0,st_I0) i1
262 = do st_i1 <- addToSS st_i0 (i2s i1)
263 return (st_i1,st_l0,st_p0,st_I0)
265 instr2 (st_i0,st_l0,st_p0,st_I0) i1 i2
266 = do st_i1 <- addToSS st_i0 (i2s i1)
267 st_i2 <- addToSS st_i1 (i2s i2)
268 return (st_i2,st_l0,st_p0,st_I0)
270 instr3 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3
271 = do st_i1 <- addToSS st_i0 (i2s i1)
272 st_i2 <- addToSS st_i1 (i2s i2)
273 st_i3 <- addToSS st_i2 (i2s i3)
274 return (st_i3,st_l0,st_p0,st_I0)
276 instr4 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3 i4
277 = do st_i1 <- addToSS st_i0 (i2s i1)
278 st_i2 <- addToSS st_i1 (i2s i2)
279 st_i3 <- addToSS st_i2 (i2s i3)
280 st_i4 <- addToSS st_i3 (i2s i4)
281 return (st_i4,st_l0,st_p0,st_I0)
283 float (st_i0,st_l0,st_p0,st_I0) f
284 = do let ws = mkLitF f
285 st_l1 <- addListToSS st_l0 ws
286 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
288 double (st_i0,st_l0,st_p0,st_I0) d
289 = do let ws = mkLitD d
290 st_l1 <- addListToSS st_l0 ws
291 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
293 int (st_i0,st_l0,st_p0,st_I0) i
294 = do let ws = mkLitI i
295 st_l1 <- addListToSS st_l0 ws
296 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
298 addr (st_i0,st_l0,st_p0,st_I0) a
299 = do let ws = mkLitA a
300 st_l1 <- addListToSS st_l0 ws
301 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
303 ptr (st_i0,st_l0,st_p0,st_I0) p
304 = do st_p1 <- addToSS st_p0 p
305 return (sizeSS st_p0, (st_i0,st_l0,st_p1,st_I0))
307 itbl (st_i0,st_l0,st_p0,st_I0) dcon
308 = do st_I1 <- addToSS st_I0 (getName dcon)
309 return (sizeSS st_I0, (st_i0,st_l0,st_p0,st_I1))
311 literal st (MachWord w) = int st (fromIntegral w)
312 literal st (MachInt j) = int st (fromIntegral j)
313 literal st (MachFloat r) = float st (fromRational r)
314 literal st (MachDouble r) = double st (fromRational r)
315 literal st (MachChar c) = int st c
318 = addr st ret_itbl_addr
320 ret_itbl_addr = case pk of
321 PtrRep -> stg_ctoi_ret_R1p_info
322 WordRep -> stg_ctoi_ret_R1n_info
323 IntRep -> stg_ctoi_ret_R1n_info
324 AddrRep -> stg_ctoi_ret_R1n_info
325 CharRep -> stg_ctoi_ret_R1n_info
326 FloatRep -> stg_ctoi_ret_F1_info
327 DoubleRep -> stg_ctoi_ret_D1_info
328 VoidRep -> stg_ctoi_ret_V_info
329 _ -> pprPanic "mkBits.ctoi_itbl" (ppr pk)
332 = addr st ret_itbl_addr
334 ret_itbl_addr = case pk of
335 CharRep -> stg_gc_unbx_r1_info
336 IntRep -> stg_gc_unbx_r1_info
337 FloatRep -> stg_gc_f1_info
338 DoubleRep -> stg_gc_d1_info
340 -- Interpreter.c spots this special case
342 foreign label "stg_ctoi_ret_R1p_info" stg_ctoi_ret_R1p_info :: Addr
343 foreign label "stg_ctoi_ret_R1n_info" stg_ctoi_ret_R1n_info :: Addr
344 foreign label "stg_ctoi_ret_F1_info" stg_ctoi_ret_F1_info :: Addr
345 foreign label "stg_ctoi_ret_D1_info" stg_ctoi_ret_D1_info :: Addr
346 foreign label "stg_ctoi_ret_V_info" stg_ctoi_ret_V_info :: Addr
348 foreign label "stg_gc_unbx_r1_info" stg_gc_unbx_r1_info :: Addr
349 foreign label "stg_gc_f1_info" stg_gc_f1_info :: Addr
350 foreign label "stg_gc_d1_info" stg_gc_d1_info :: Addr
352 -- The size in 16-bit entities of an instruction.
353 instrSize16s :: BCInstr -> Int
386 -- Make lists of host-sized words for literals, so that when the
387 -- words are placed in memory at increasing addresses, the
388 -- bit pattern is correct for the host's word size and endianness.
389 mkLitI :: Int -> [Word]
390 mkLitF :: Float -> [Word]
391 mkLitD :: Double -> [Word]
392 mkLitA :: Addr -> [Word]
396 arr <- newFloatArray ((0::Int),0)
397 writeFloatArray arr 0 f
398 f_arr <- castSTUArray arr
399 w0 <- readWordArray f_arr 0
406 arr <- newDoubleArray ((0::Int),1)
407 writeDoubleArray arr 0 d
408 d_arr <- castSTUArray arr
409 w0 <- readWordArray d_arr 0
410 w1 <- readWordArray d_arr 1
415 arr <- newDoubleArray ((0::Int),0)
416 writeDoubleArray arr 0 d
417 d_arr <- castSTUArray arr
418 w0 <- readWordArray d_arr 0
424 arr <- newIntArray ((0::Int),0)
425 writeIntArray arr 0 i
426 i_arr <- castSTUArray arr
427 w0 <- readWordArray i_arr 0
433 arr <- newAddrArray ((0::Int),0)
434 writeAddrArray arr 0 a
435 a_arr <- castSTUArray arr
436 w0 <- readWordArray a_arr 0
442 %************************************************************************
444 \subsection{Linking interpretables into something we can run}
446 %************************************************************************
451 data BCO# = BCO# ByteArray# -- instrs :: array Word16#
452 ByteArray# -- literals :: array Word32#
453 PtrArray# -- ptrs :: Array HValue
454 ByteArray# -- itbls :: Array Addr#
457 linkBCO ie ce (UnlinkedBCO nm insnsSS literalsSS ptrsSS itblsSS)
458 = do insns <- listFromSS insnsSS
459 literals <- listFromSS literalsSS
460 ptrs <- listFromSS ptrsSS
461 itbls <- listFromSS itblsSS
463 linked_ptrs <- mapM (lookupCE ce) ptrs
464 linked_itbls <- mapM (lookupIE ie) itbls
466 let n_insns = sizeSS insnsSS
467 n_literals = sizeSS literalsSS
468 n_ptrs = sizeSS ptrsSS
469 n_itbls = sizeSS itblsSS
471 let ptrs_arr = array (0, n_ptrs-1) (indexify linked_ptrs)
473 ptrs_parr = case ptrs_arr of Array lo hi parr -> parr
475 itbls_arr = array (0, n_itbls-1) (indexify linked_itbls)
476 :: UArray Int ItblPtr
477 itbls_barr = case itbls_arr of UArray lo hi barr -> barr
479 insns_arr | n_insns > 65535
480 = panic "linkBCO: >= 64k insns in BCO"
483 (indexify (fromIntegral n_insns:insns))
485 insns_barr = case insns_arr of UArray lo hi barr -> barr
487 literals_arr = array (0, n_literals-1) (indexify literals)
489 literals_barr = case literals_arr of UArray lo hi barr -> barr
491 indexify :: [a] -> [(Int, a)]
492 indexify xs = zip [0..] xs
494 BCO bco# <- newBCO insns_barr literals_barr ptrs_parr itbls_barr
496 -- WAS: return (unsafeCoerce# bco#)
497 case mkApUpd0# (unsafeCoerce# bco#) of
498 (# final_bco #) -> return final_bco
503 newBCO :: ByteArray# -> ByteArray# -> Array# a -> ByteArray# -> IO BCO
505 = IO (\s -> case newBCO# a b c d s of (# s1, bco #) -> (# s1, BCO bco #))
508 lookupCE :: ClosureEnv -> Either Name PrimOp -> IO HValue
509 lookupCE ce (Right primop)
510 = do m <- lookupSymbol (primopToCLabel primop "closure")
512 Just (Ptr addr) -> case addrToHValue# addr of
513 (# hval #) -> return hval
514 Nothing -> pprPanic "ByteCodeGen.lookupCE(primop)" (ppr primop)
515 lookupCE ce (Left nm)
516 = case lookupFM ce nm of
519 -> do m <- lookupSymbol (nameToCLabel nm "closure")
521 Just (Ptr addr) -> case addrToHValue# addr of
522 (# hval #) -> return hval
523 Nothing -> pprPanic "ByteCodeGen.lookupCE" (ppr nm)
525 lookupIE :: ItblEnv -> Name -> IO (Ptr a)
527 = case lookupFM ie con_nm of
528 Just (Ptr a) -> return (Ptr a)
530 -> do -- try looking up in the object files.
531 m <- lookupSymbol (nameToCLabel con_nm "con_info")
533 Just addr -> return addr
535 -> do -- perhaps a nullary constructor?
536 n <- lookupSymbol (nameToCLabel con_nm "static_info")
538 Just addr -> return addr
539 Nothing -> pprPanic "ByteCodeGen.lookupIE" (ppr con_nm)
541 -- HACKS!!! ToDo: cleaner
542 nameToCLabel :: Name -> String{-suffix-} -> String
543 nameToCLabel n suffix
544 = _UNPK_(moduleNameFS (rdrNameModule rn))
545 ++ '_':occNameString(rdrNameOcc rn) ++ '_':suffix
546 where rn = toRdrName n
548 primopToCLabel :: PrimOp -> String{-suffix-} -> String
549 primopToCLabel primop suffix
550 = let str = "PrelPrimopWrappers_" ++ occNameString (primOpOcc primop) ++ '_':suffix
551 in --trace ("primopToCLabel: " ++ str)
556 %************************************************************************
558 \subsection{Connect to actual values for bytecode opcodes}
560 %************************************************************************
564 #include "Bytecodes.h"
566 i_ARGCHECK = (bci_ARGCHECK :: Int)
567 i_PUSH_L = (bci_PUSH_L :: Int)
568 i_PUSH_LL = (bci_PUSH_LL :: Int)
569 i_PUSH_LLL = (bci_PUSH_LLL :: Int)
570 i_PUSH_G = (bci_PUSH_G :: Int)
571 i_PUSH_AS = (bci_PUSH_AS :: Int)
572 i_PUSH_UBX = (bci_PUSH_UBX :: Int)
573 i_PUSH_TAG = (bci_PUSH_TAG :: Int)
574 i_SLIDE = (bci_SLIDE :: Int)
575 i_ALLOC = (bci_ALLOC :: Int)
576 i_MKAP = (bci_MKAP :: Int)
577 i_UNPACK = (bci_UNPACK :: Int)
578 i_UPK_TAG = (bci_UPK_TAG :: Int)
579 i_PACK = (bci_PACK :: Int)
580 i_TESTLT_I = (bci_TESTLT_I :: Int)
581 i_TESTEQ_I = (bci_TESTEQ_I :: Int)
582 i_TESTLT_F = (bci_TESTLT_F :: Int)
583 i_TESTEQ_F = (bci_TESTEQ_F :: Int)
584 i_TESTLT_D = (bci_TESTLT_D :: Int)
585 i_TESTEQ_D = (bci_TESTEQ_D :: Int)
586 i_TESTLT_P = (bci_TESTLT_P :: Int)
587 i_TESTEQ_P = (bci_TESTEQ_P :: Int)
588 i_CASEFAIL = (bci_CASEFAIL :: Int)
589 i_ENTER = (bci_ENTER :: Int)
590 i_RETURN = (bci_RETURN :: Int)
591 i_STKCHECK = (bci_STKCHECK :: Int)
592 i_JMP = (bci_JMP :: Int)
594 iNTERP_STACK_CHECK_THRESH = (INTERP_STACK_CHECK_THRESH :: Int)