2 % (c) The University of Glasgow 2000
4 \section[ByteCodeLink]{Bytecode assembler and linker}
7 module ByteCodeLink ( UnlinkedBCO, UnlinkedBCOExpr, assembleBCO,
8 ClosureEnv, HValue, linkSomeBCOs, filterNameMap
11 #include "HsVersions.h"
14 import Name ( Name, getName, nameModule, toRdrName, isGlobalName )
15 import RdrName ( rdrNameOcc, rdrNameModule )
16 import OccName ( occNameString )
17 import FiniteMap ( FiniteMap, addListToFM, filterFM,
18 addToFM, lookupFM, emptyFM )
20 import Literal ( Literal(..) )
21 import PrimOp ( PrimOp, primOpOcc )
22 import PrimRep ( PrimRep(..) )
23 import Constants ( wORD_SIZE )
24 import Module ( ModuleName, moduleName, moduleNameFS )
25 import Linker ( lookupSymbol )
26 import FastString ( FastString(..) )
27 import ByteCodeInstr ( BCInstr(..), ProtoBCO(..) )
28 import ByteCodeItbls ( ItblEnv, ItblPtr )
31 import Monad ( foldM )
33 import MArray ( castSTUArray,
34 newFloatArray, writeFloatArray,
35 newDoubleArray, writeDoubleArray,
36 newIntArray, writeIntArray,
37 newAddrArray, writeAddrArray )
38 import Foreign ( Word16, Ptr(..) )
39 import Addr ( Word, Addr )
41 import PrelBase ( Int(..) )
42 import PrelGHC ( BCO#, newBCO#, unsafeCoerce#,
43 ByteArray#, Array#, addrToHValue#, mkApUpd0# )
44 import IOExts ( fixIO )
46 import PrelArr ( Array(..) )
47 import PrelIOBase ( IO(..) )
51 %************************************************************************
53 \subsection{Top-level stuff}
55 %************************************************************************
59 -- Link a bunch of BCOs and return them + updated closure env.
60 linkSomeBCOs :: Bool -- False <=> add _all_ BCOs to returned closure env
61 -- True <=> add only toplevel BCOs to closure env
65 -> IO (ClosureEnv, [HValue])
66 linkSomeBCOs toplevs_only ie ce_in ul_bcos
67 = do let nms = map nameOfUnlinkedBCO ul_bcos
69 ( \ hvs -> let ce_out = addListToFM ce_in (zipLazily nms hvs)
70 in mapM (linkBCO ie ce_out) ul_bcos )
72 let ce_all_additions = zip nms hvals
73 ce_top_additions = filter (isGlobalName.fst) ce_all_additions
74 ce_additions = if toplevs_only then ce_top_additions
76 ce_out = addListToFM ce_in ce_additions
77 return (ce_out, hvals)
79 -- A lazier zip, in which no demand is propagated to the second
80 -- list unless some demand is propagated to the snd of one of the
83 zipLazily (x:xs) ys = (x, head ys) : zipLazily xs (tail ys)
88 (SizedSeq Word16) -- insns
89 (SizedSeq Word) -- literals
90 (SizedSeq (Either Name PrimOp)) -- ptrs
91 (SizedSeq Name) -- itbl refs
93 nameOfUnlinkedBCO (UnlinkedBCO nm _ _ _ _) = nm
95 -- When translating expressions, we need to distinguish the root
96 -- BCO for the expression
97 type UnlinkedBCOExpr = (UnlinkedBCO, [UnlinkedBCO])
99 instance Outputable UnlinkedBCO where
100 ppr (UnlinkedBCO nm insns lits ptrs itbls)
101 = sep [text "BCO", ppr nm, text "with",
102 int (sizeSS insns), text "insns",
103 int (sizeSS lits), text "lits",
104 int (sizeSS ptrs), text "ptrs",
105 int (sizeSS itbls), text "itbls"]
108 -- these need a proper home
109 type ClosureEnv = FiniteMap Name HValue
110 data HValue = HValue -- dummy type, actually a pointer to some Real Code.
112 -- remove all entries for a given set of modules from the environment
113 filterNameMap :: [ModuleName] -> FiniteMap Name a -> FiniteMap Name a
114 filterNameMap mods env
115 = filterFM (\n _ -> moduleName (nameModule n) `notElem` mods) env
119 %************************************************************************
121 \subsection{The bytecode assembler}
123 %************************************************************************
125 The object format for bytecodes is: 16 bits for the opcode, and 16 for
126 each field -- so the code can be considered a sequence of 16-bit ints.
127 Each field denotes either a stack offset or number of items on the
128 stack (eg SLIDE), and index into the pointer table (eg PUSH_G), an
129 index into the literal table (eg PUSH_I/D/L), or a bytecode address in
133 -- Top level assembler fn.
134 assembleBCO :: ProtoBCO Name -> IO UnlinkedBCO
136 assembleBCO (ProtoBCO nm instrs origin)
138 -- pass 1: collect up the offsets of the local labels.
139 -- Remember that the first insn starts at offset 1 since offset 0
140 -- (eventually) will hold the total # of insns.
141 label_env = mkLabelEnv emptyFM 1 instrs
143 mkLabelEnv env i_offset [] = env
144 mkLabelEnv env i_offset (i:is)
146 = case i of LABEL n -> addToFM env n i_offset ; _ -> env
147 in mkLabelEnv new_env (i_offset + instrSize16s i) is
150 = case lookupFM label_env lab of
151 Just bco_offset -> bco_offset
152 Nothing -> pprPanic "assembleBCO.findLabel" (int lab)
154 do -- pass 2: generate the instruction, ptr and nonptr bits
155 insns <- return emptySS :: IO (SizedSeq Word16)
156 lits <- return emptySS :: IO (SizedSeq Word)
157 ptrs <- return emptySS :: IO (SizedSeq (Either Name PrimOp))
158 itbls <- return emptySS :: IO (SizedSeq Name)
159 let init_asm_state = (insns,lits,ptrs,itbls)
160 (final_insns, final_lits, final_ptrs, final_itbls)
161 <- mkBits findLabel init_asm_state instrs
163 return (UnlinkedBCO nm final_insns final_lits final_ptrs final_itbls)
165 -- instrs nonptrs ptrs itbls
166 type AsmState = (SizedSeq Word16, SizedSeq Word,
167 SizedSeq (Either Name PrimOp), SizedSeq Name)
169 data SizedSeq a = SizedSeq !Int [a]
170 emptySS = SizedSeq 0 []
171 addToSS (SizedSeq n r_xs) x = return (SizedSeq (n+1) (x:r_xs))
172 addListToSS (SizedSeq n r_xs) xs
173 = return (SizedSeq (n + length xs) (reverse xs ++ r_xs))
174 sizeSS (SizedSeq n r_xs) = n
175 listFromSS (SizedSeq n r_xs) = return (reverse r_xs)
178 -- This is where all the action is (pass 2 of the assembler)
179 mkBits :: (Int -> Int) -- label finder
181 -> [BCInstr] -- instructions (in)
184 mkBits findLabel st proto_insns
185 = foldM doInstr st proto_insns
187 doInstr :: AsmState -> BCInstr -> IO AsmState
190 ARGCHECK n -> instr2 st i_ARGCHECK n
191 PUSH_L o1 -> instr2 st i_PUSH_L o1
192 PUSH_LL o1 o2 -> instr3 st i_PUSH_LL o1 o2
193 PUSH_LLL o1 o2 o3 -> instr4 st i_PUSH_LLL o1 o2 o3
194 PUSH_G nm -> do (p, st2) <- ptr st nm
195 instr2 st2 i_PUSH_G p
196 PUSH_AS nm pk -> do (p, st2) <- ptr st (Left nm)
197 (np, st3) <- ctoi_itbl st2 pk
198 instr3 st3 i_PUSH_AS p np
199 PUSH_UBX lit nws -> do (np, st2) <- literal st lit
200 instr3 st2 i_PUSH_UBX np nws
201 PUSH_TAG tag -> instr2 st i_PUSH_TAG tag
202 SLIDE n by -> instr3 st i_SLIDE n by
203 ALLOC n -> instr2 st i_ALLOC n
204 MKAP off sz -> instr3 st i_MKAP off sz
205 UNPACK n -> instr2 st i_UNPACK n
206 UPK_TAG n m k -> instr4 st i_UPK_TAG n m k
207 PACK dcon sz -> do (itbl_no,st2) <- itbl st dcon
208 instr3 st2 i_PACK itbl_no sz
209 LABEL lab -> return st
210 TESTLT_I i l -> do (np, st2) <- int st i
211 instr3 st2 i_TESTLT_I np (findLabel l)
212 TESTEQ_I i l -> do (np, st2) <- int st i
213 instr3 st2 i_TESTEQ_I np (findLabel l)
214 TESTLT_F f l -> do (np, st2) <- float st f
215 instr3 st2 i_TESTLT_F np (findLabel l)
216 TESTEQ_F f l -> do (np, st2) <- float st f
217 instr3 st2 i_TESTEQ_F np (findLabel l)
218 TESTLT_D d l -> do (np, st2) <- double st d
219 instr3 st2 i_TESTLT_D np (findLabel l)
220 TESTEQ_D d l -> do (np, st2) <- double st d
221 instr3 st2 i_TESTEQ_D np (findLabel l)
222 TESTLT_P i l -> instr3 st i_TESTLT_P i (findLabel l)
223 TESTEQ_P i l -> instr3 st i_TESTEQ_P i (findLabel l)
224 CASEFAIL -> instr1 st i_CASEFAIL
225 ENTER -> instr1 st i_ENTER
226 RETURN rep -> do (itbl_no,st2) <- itoc_itbl st rep
227 instr2 st2 i_RETURN itbl_no
232 instr1 (st_i0,st_l0,st_p0,st_I0) i1
233 = do st_i1 <- addToSS st_i0 (i2s i1)
234 return (st_i1,st_l0,st_p0,st_I0)
236 instr2 (st_i0,st_l0,st_p0,st_I0) i1 i2
237 = do st_i1 <- addToSS st_i0 (i2s i1)
238 st_i2 <- addToSS st_i1 (i2s i2)
239 return (st_i2,st_l0,st_p0,st_I0)
241 instr3 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3
242 = do st_i1 <- addToSS st_i0 (i2s i1)
243 st_i2 <- addToSS st_i1 (i2s i2)
244 st_i3 <- addToSS st_i2 (i2s i3)
245 return (st_i3,st_l0,st_p0,st_I0)
247 instr4 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3 i4
248 = do st_i1 <- addToSS st_i0 (i2s i1)
249 st_i2 <- addToSS st_i1 (i2s i2)
250 st_i3 <- addToSS st_i2 (i2s i3)
251 st_i4 <- addToSS st_i3 (i2s i4)
252 return (st_i4,st_l0,st_p0,st_I0)
254 float (st_i0,st_l0,st_p0,st_I0) f
255 = do let ws = mkLitF f
256 st_l1 <- addListToSS st_l0 ws
257 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
259 double (st_i0,st_l0,st_p0,st_I0) d
260 = do let ws = mkLitD d
261 st_l1 <- addListToSS st_l0 ws
262 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
264 int (st_i0,st_l0,st_p0,st_I0) i
265 = do let ws = mkLitI i
266 st_l1 <- addListToSS st_l0 ws
267 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
269 addr (st_i0,st_l0,st_p0,st_I0) a
270 = do let ws = mkLitA a
271 st_l1 <- addListToSS st_l0 ws
272 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
274 ptr (st_i0,st_l0,st_p0,st_I0) p
275 = do st_p1 <- addToSS st_p0 p
276 return (sizeSS st_p0, (st_i0,st_l0,st_p1,st_I0))
278 itbl (st_i0,st_l0,st_p0,st_I0) dcon
279 = do st_I1 <- addToSS st_I0 (getName dcon)
280 return (sizeSS st_I0, (st_i0,st_l0,st_p0,st_I1))
282 literal st (MachWord w) = int st (fromIntegral w)
283 literal st (MachInt j) = int st (fromIntegral j)
284 literal st (MachFloat r) = float st (fromRational r)
285 literal st (MachDouble r) = double st (fromRational r)
286 literal st (MachChar c) = int st c
289 = addr st ret_itbl_addr
291 ret_itbl_addr = case pk of
292 PtrRep -> stg_ctoi_ret_R1p_info
293 WordRep -> stg_ctoi_ret_R1n_info
294 IntRep -> stg_ctoi_ret_R1n_info
295 AddrRep -> stg_ctoi_ret_R1n_info
296 CharRep -> stg_ctoi_ret_R1n_info
297 FloatRep -> stg_ctoi_ret_F1_info
298 DoubleRep -> stg_ctoi_ret_D1_info
299 _ -> pprPanic "mkBits.ctoi_itbl" (ppr pk)
302 = addr st ret_itbl_addr
304 ret_itbl_addr = case pk of
305 CharRep -> stg_gc_unbx_r1_info
306 IntRep -> stg_gc_unbx_r1_info
307 FloatRep -> stg_gc_f1_info
308 DoubleRep -> stg_gc_d1_info
310 foreign label "stg_ctoi_ret_R1p_info" stg_ctoi_ret_R1p_info :: Addr
311 foreign label "stg_ctoi_ret_R1n_info" stg_ctoi_ret_R1n_info :: Addr
312 foreign label "stg_ctoi_ret_F1_info" stg_ctoi_ret_F1_info :: Addr
313 foreign label "stg_ctoi_ret_D1_info" stg_ctoi_ret_D1_info :: Addr
315 foreign label "stg_gc_unbx_r1_info" stg_gc_unbx_r1_info :: Addr
316 foreign label "stg_gc_f1_info" stg_gc_f1_info :: Addr
317 foreign label "stg_gc_d1_info" stg_gc_d1_info :: Addr
319 -- The size in 16-bit entities of an instruction.
320 instrSize16s :: BCInstr -> Int
351 -- Make lists of host-sized words for literals, so that when the
352 -- words are placed in memory at increasing addresses, the
353 -- bit pattern is correct for the host's word size and endianness.
354 mkLitI :: Int -> [Word]
355 mkLitF :: Float -> [Word]
356 mkLitD :: Double -> [Word]
357 mkLitA :: Addr -> [Word]
361 arr <- newFloatArray ((0::Int),0)
362 writeFloatArray arr 0 f
363 f_arr <- castSTUArray arr
364 w0 <- readWordArray f_arr 0
371 arr <- newDoubleArray ((0::Int),1)
372 writeDoubleArray arr 0 d
373 d_arr <- castSTUArray arr
374 w0 <- readWordArray d_arr 0
375 w1 <- readWordArray d_arr 1
380 arr <- newDoubleArray ((0::Int),0)
381 writeDoubleArray arr 0 d
382 d_arr <- castSTUArray arr
383 w0 <- readWordArray d_arr 0
389 arr <- newIntArray ((0::Int),0)
390 writeIntArray arr 0 i
391 i_arr <- castSTUArray arr
392 w0 <- readWordArray i_arr 0
398 arr <- newAddrArray ((0::Int),0)
399 writeAddrArray arr 0 a
400 a_arr <- castSTUArray arr
401 w0 <- readWordArray a_arr 0
407 %************************************************************************
409 \subsection{Linking interpretables into something we can run}
411 %************************************************************************
416 data BCO# = BCO# ByteArray# -- instrs :: array Word16#
417 ByteArray# -- literals :: array Word32#
418 PtrArray# -- ptrs :: Array HValue
419 ByteArray# -- itbls :: Array Addr#
422 linkBCO ie ce (UnlinkedBCO nm insnsSS literalsSS ptrsSS itblsSS)
423 = do insns <- listFromSS insnsSS
424 literals <- listFromSS literalsSS
425 ptrs <- listFromSS ptrsSS
426 itbls <- listFromSS itblsSS
428 linked_ptrs <- mapM (lookupCE ce) ptrs
429 linked_itbls <- mapM (lookupIE ie) itbls
431 let n_insns = sizeSS insnsSS
432 n_literals = sizeSS literalsSS
433 n_ptrs = sizeSS ptrsSS
434 n_itbls = sizeSS itblsSS
436 let ptrs_arr = array (0, n_ptrs-1) (indexify linked_ptrs)
438 ptrs_parr = case ptrs_arr of Array lo hi parr -> parr
440 itbls_arr = array (0, n_itbls-1) (indexify linked_itbls)
441 :: UArray Int ItblPtr
442 itbls_barr = case itbls_arr of UArray lo hi barr -> barr
444 insns_arr | n_insns > 65535
445 = panic "linkBCO: >= 64k insns in BCO"
448 (indexify (fromIntegral n_insns:insns))
450 insns_barr = case insns_arr of UArray lo hi barr -> barr
452 literals_arr = array (0, n_literals-1) (indexify literals)
454 literals_barr = case literals_arr of UArray lo hi barr -> barr
456 indexify :: [a] -> [(Int, a)]
457 indexify xs = zip [0..] xs
459 BCO bco# <- newBCO insns_barr literals_barr ptrs_parr itbls_barr
461 -- WAS: return (unsafeCoerce# bco#)
462 case mkApUpd0# (unsafeCoerce# bco#) of
463 (# final_bco #) -> return final_bco
468 newBCO :: ByteArray# -> ByteArray# -> Array# a -> ByteArray# -> IO BCO
470 = IO (\s -> case newBCO# a b c d s of (# s1, bco #) -> (# s1, BCO bco #))
473 lookupCE :: ClosureEnv -> Either Name PrimOp -> IO HValue
474 lookupCE ce (Right primop)
475 = do m <- lookupSymbol (primopToCLabel primop "closure")
477 Just (Ptr addr) -> case addrToHValue# addr of
478 (# hval #) -> return hval
479 Nothing -> pprPanic "ByteCodeGen.lookupCE(primop)" (ppr primop)
480 lookupCE ce (Left nm)
481 = case lookupFM ce nm of
484 -> do m <- lookupSymbol (nameToCLabel nm "closure")
486 Just (Ptr addr) -> case addrToHValue# addr of
487 (# hval #) -> return hval
488 Nothing -> pprPanic "ByteCodeGen.lookupCE" (ppr nm)
490 lookupIE :: ItblEnv -> Name -> IO (Ptr a)
492 = case lookupFM ie con_nm of
493 Just (Ptr a) -> return (Ptr a)
495 -> do -- try looking up in the object files.
496 m <- lookupSymbol (nameToCLabel con_nm "con_info")
498 Just addr -> return addr
500 -> do -- perhaps a nullary constructor?
501 n <- lookupSymbol (nameToCLabel con_nm "static_info")
503 Just addr -> return addr
504 Nothing -> pprPanic "ByteCodeGen.lookupIE" (ppr con_nm)
506 -- HACKS!!! ToDo: cleaner
507 nameToCLabel :: Name -> String{-suffix-} -> String
508 nameToCLabel n suffix
509 = _UNPK_(moduleNameFS (rdrNameModule rn))
510 ++ '_':occNameString(rdrNameOcc rn) ++ '_':suffix
511 where rn = toRdrName n
513 primopToCLabel :: PrimOp -> String{-suffix-} -> String
514 primopToCLabel primop suffix
515 = let str = "PrelPrimopWrappers_" ++ occNameString (primOpOcc primop) ++ '_':suffix
516 in --trace ("primopToCLabel: " ++ str)
521 %************************************************************************
523 \subsection{Connect to actual values for bytecode opcodes}
525 %************************************************************************
529 #include "Bytecodes.h"
531 i_ARGCHECK = (bci_ARGCHECK :: Int)
532 i_PUSH_L = (bci_PUSH_L :: Int)
533 i_PUSH_LL = (bci_PUSH_LL :: Int)
534 i_PUSH_LLL = (bci_PUSH_LLL :: Int)
535 i_PUSH_G = (bci_PUSH_G :: Int)
536 i_PUSH_AS = (bci_PUSH_AS :: Int)
537 i_PUSH_UBX = (bci_PUSH_UBX :: Int)
538 i_PUSH_TAG = (bci_PUSH_TAG :: Int)
539 i_SLIDE = (bci_SLIDE :: Int)
540 i_ALLOC = (bci_ALLOC :: Int)
541 i_MKAP = (bci_MKAP :: Int)
542 i_UNPACK = (bci_UNPACK :: Int)
543 i_UPK_TAG = (bci_UPK_TAG :: Int)
544 i_PACK = (bci_PACK :: Int)
545 i_TESTLT_I = (bci_TESTLT_I :: Int)
546 i_TESTEQ_I = (bci_TESTEQ_I :: Int)
547 i_TESTLT_F = (bci_TESTLT_F :: Int)
548 i_TESTEQ_F = (bci_TESTEQ_F :: Int)
549 i_TESTLT_D = (bci_TESTLT_D :: Int)
550 i_TESTEQ_D = (bci_TESTEQ_D :: Int)
551 i_TESTLT_P = (bci_TESTLT_P :: Int)
552 i_TESTEQ_P = (bci_TESTEQ_P :: Int)
553 i_CASEFAIL = (bci_CASEFAIL :: Int)
554 i_ENTER = (bci_ENTER :: Int)
555 i_RETURN = (bci_RETURN :: Int)