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 )
15 import RdrName ( rdrNameOcc, rdrNameModule )
16 import OccName ( occNameString, occNameUserString )
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 Util ( global )
24 import Constants ( wORD_SIZE )
25 import Module ( ModuleName, moduleName, moduleNameFS )
26 import Linker ( lookupSymbol )
27 import FastString ( FastString(..) )
28 import ByteCodeInstr ( BCInstr(..), ProtoBCO(..) )
29 import ByteCodeItbls ( ItblEnv, ItblPtr )
32 import Monad ( foldM )
34 import MArray ( castSTUArray,
35 newFloatArray, writeFloatArray,
36 newDoubleArray, writeDoubleArray,
37 newIntArray, writeIntArray,
38 newAddrArray, writeAddrArray )
39 import Foreign ( Word16, Ptr(..) )
40 import Addr ( Word, Addr )
42 import PrelBase ( Int(..) )
43 import PrelGHC ( BCO#, newBCO#, unsafeCoerce#,
44 ByteArray#, Array#, addrToHValue#, mkApUpd0# )
45 import IOExts ( IORef, fixIO, readIORef, writeIORef )
47 import PrelArr ( Array(..) )
48 import PrelIOBase ( IO(..) )
52 %************************************************************************
54 \subsection{Top-level stuff}
56 %************************************************************************
60 -- Link a bunch of BCOs and return them + updated closure env.
61 linkSomeBCOs :: ItblEnv -> ClosureEnv -> [UnlinkedBCO]
62 -> IO (ClosureEnv, [HValue])
63 linkSomeBCOs ie ce_in ul_bcos
64 = do let nms = map nameOfUnlinkedBCO ul_bcos
66 ( \ hvs -> let ce_out = addListToFM ce_in (zipLazily nms hvs)
67 in mapM (linkBCO ie ce_out) ul_bcos )
68 let ce_out = addListToFM ce_in (zip nms hvals)
69 return (ce_out, hvals)
71 -- A lazier zip, in which no demand is propagated to the second
72 -- list unless some demand is propagated to the snd of one of the
75 zipLazily (x:xs) ys = (x, head ys) : zipLazily xs (tail ys)
80 (SizedSeq Word16) -- insns
81 (SizedSeq Word) -- literals
82 (SizedSeq (Either Name PrimOp)) -- ptrs
83 (SizedSeq Name) -- itbl refs
85 nameOfUnlinkedBCO (UnlinkedBCO nm _ _ _ _) = nm
87 -- When translating expressions, we need to distinguish the root
88 -- BCO for the expression
89 type UnlinkedBCOExpr = (UnlinkedBCO, [UnlinkedBCO])
91 instance Outputable UnlinkedBCO where
92 ppr (UnlinkedBCO nm insns lits ptrs itbls)
93 = sep [text "BCO", ppr nm, text "with",
94 int (sizeSS insns), text "insns",
95 int (sizeSS lits), text "lits",
96 int (sizeSS ptrs), text "ptrs",
97 int (sizeSS itbls), text "itbls"]
100 -- these need a proper home
101 type ClosureEnv = FiniteMap Name HValue
102 data HValue = HValue -- dummy type, actually a pointer to some Real Code.
104 -- remove all entries for a given set of modules from the environment
105 filterNameMap :: [ModuleName] -> FiniteMap Name a -> FiniteMap Name a
106 filterNameMap mods env
107 = filterFM (\n _ -> moduleName (nameModule n) `notElem` mods) env
111 %************************************************************************
113 \subsection{The bytecode assembler}
115 %************************************************************************
117 The object format for bytecodes is: 16 bits for the opcode, and 16 for
118 each field -- so the code can be considered a sequence of 16-bit ints.
119 Each field denotes either a stack offset or number of items on the
120 stack (eg SLIDE), and index into the pointer table (eg PUSH_G), an
121 index into the literal table (eg PUSH_I/D/L), or a bytecode address in
125 -- Top level assembler fn.
126 assembleBCO :: ProtoBCO Name -> IO UnlinkedBCO
128 assembleBCO (ProtoBCO nm instrs origin)
130 -- pass 1: collect up the offsets of the local labels.
131 -- Remember that the first insn starts at offset 1 since offset 0
132 -- (eventually) will hold the total # of insns.
133 label_env = mkLabelEnv emptyFM 1 instrs
135 mkLabelEnv env i_offset [] = env
136 mkLabelEnv env i_offset (i:is)
138 = case i of LABEL n -> addToFM env n i_offset ; _ -> env
139 in mkLabelEnv new_env (i_offset + instrSize16s i) is
142 = case lookupFM label_env lab of
143 Just bco_offset -> bco_offset
144 Nothing -> pprPanic "assembleBCO.findLabel" (int lab)
146 do -- pass 2: generate the instruction, ptr and nonptr bits
147 insns <- return emptySS :: IO (SizedSeq Word16)
148 lits <- return emptySS :: IO (SizedSeq Word)
149 ptrs <- return emptySS :: IO (SizedSeq (Either Name PrimOp))
150 itbls <- return emptySS :: IO (SizedSeq Name)
151 let init_asm_state = (insns,lits,ptrs,itbls)
152 (final_insns, final_lits, final_ptrs, final_itbls)
153 <- mkBits findLabel init_asm_state instrs
155 return (UnlinkedBCO nm final_insns final_lits final_ptrs final_itbls)
157 -- instrs nonptrs ptrs itbls
158 type AsmState = (SizedSeq Word16, SizedSeq Word,
159 SizedSeq (Either Name PrimOp), SizedSeq Name)
161 data SizedSeq a = SizedSeq !Int [a]
162 emptySS = SizedSeq 0 []
163 addToSS (SizedSeq n r_xs) x = return (SizedSeq (n+1) (x:r_xs))
164 addListToSS (SizedSeq n r_xs) xs
165 = return (SizedSeq (n + length xs) (reverse xs ++ r_xs))
166 sizeSS (SizedSeq n r_xs) = n
167 listFromSS (SizedSeq n r_xs) = return (reverse r_xs)
170 -- This is where all the action is (pass 2 of the assembler)
171 mkBits :: (Int -> Int) -- label finder
173 -> [BCInstr] -- instructions (in)
176 mkBits findLabel st proto_insns
177 = foldM doInstr st proto_insns
179 doInstr :: AsmState -> BCInstr -> IO AsmState
182 ARGCHECK n -> instr2 st i_ARGCHECK n
183 PUSH_L o1 -> instr2 st i_PUSH_L o1
184 PUSH_LL o1 o2 -> instr3 st i_PUSH_LL o1 o2
185 PUSH_LLL o1 o2 o3 -> instr4 st i_PUSH_LLL o1 o2 o3
186 PUSH_G nm -> do (p, st2) <- ptr st nm
187 instr2 st2 i_PUSH_G p
188 PUSH_AS nm pk -> do (p, st2) <- ptr st (Left nm)
189 (np, st3) <- ctoi_itbl st2 pk
190 instr3 st3 i_PUSH_AS p np
191 PUSH_UBX lit nws -> do (np, st2) <- literal st lit
192 instr3 st2 i_PUSH_UBX np nws
193 PUSH_TAG tag -> instr2 st i_PUSH_TAG tag
194 SLIDE n by -> instr3 st i_SLIDE n by
195 ALLOC n -> instr2 st i_ALLOC n
196 MKAP off sz -> instr3 st i_MKAP off sz
197 UNPACK n -> instr2 st i_UNPACK n
198 UPK_TAG n m k -> instr4 st i_UPK_TAG n m k
199 PACK dcon sz -> do (itbl_no,st2) <- itbl st dcon
200 instr3 st2 i_PACK itbl_no sz
201 LABEL lab -> return st
202 TESTLT_I i l -> do (np, st2) <- int st i
203 instr3 st2 i_TESTLT_I np (findLabel l)
204 TESTEQ_I i l -> do (np, st2) <- int st i
205 instr3 st2 i_TESTEQ_I np (findLabel l)
206 TESTLT_F f l -> do (np, st2) <- float st f
207 instr3 st2 i_TESTLT_F np (findLabel l)
208 TESTEQ_F f l -> do (np, st2) <- float st f
209 instr3 st2 i_TESTEQ_F np (findLabel l)
210 TESTLT_D d l -> do (np, st2) <- double st d
211 instr3 st2 i_TESTLT_D np (findLabel l)
212 TESTEQ_D d l -> do (np, st2) <- double st d
213 instr3 st2 i_TESTEQ_D np (findLabel l)
214 TESTLT_P i l -> instr3 st i_TESTLT_P i (findLabel l)
215 TESTEQ_P i l -> instr3 st i_TESTEQ_P i (findLabel l)
216 CASEFAIL -> instr1 st i_CASEFAIL
217 ENTER -> instr1 st i_ENTER
218 RETURN rep -> do (itbl_no,st2) <- itoc_itbl st rep
219 instr2 st2 i_RETURN itbl_no
224 instr1 (st_i0,st_l0,st_p0,st_I0) i1
225 = do st_i1 <- addToSS st_i0 (i2s i1)
226 return (st_i1,st_l0,st_p0,st_I0)
228 instr2 (st_i0,st_l0,st_p0,st_I0) i1 i2
229 = do st_i1 <- addToSS st_i0 (i2s i1)
230 st_i2 <- addToSS st_i1 (i2s i2)
231 return (st_i2,st_l0,st_p0,st_I0)
233 instr3 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3
234 = do st_i1 <- addToSS st_i0 (i2s i1)
235 st_i2 <- addToSS st_i1 (i2s i2)
236 st_i3 <- addToSS st_i2 (i2s i3)
237 return (st_i3,st_l0,st_p0,st_I0)
239 instr4 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3 i4
240 = do st_i1 <- addToSS st_i0 (i2s i1)
241 st_i2 <- addToSS st_i1 (i2s i2)
242 st_i3 <- addToSS st_i2 (i2s i3)
243 st_i4 <- addToSS st_i3 (i2s i4)
244 return (st_i4,st_l0,st_p0,st_I0)
246 float (st_i0,st_l0,st_p0,st_I0) f
247 = do let ws = mkLitF f
248 st_l1 <- addListToSS st_l0 ws
249 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
251 double (st_i0,st_l0,st_p0,st_I0) d
252 = do let ws = mkLitD d
253 st_l1 <- addListToSS st_l0 ws
254 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
256 int (st_i0,st_l0,st_p0,st_I0) i
257 = do let ws = mkLitI i
258 st_l1 <- addListToSS st_l0 ws
259 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
261 addr (st_i0,st_l0,st_p0,st_I0) a
262 = do let ws = mkLitA a
263 st_l1 <- addListToSS st_l0 ws
264 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
266 ptr (st_i0,st_l0,st_p0,st_I0) p
267 = do st_p1 <- addToSS st_p0 p
268 return (sizeSS st_p0, (st_i0,st_l0,st_p1,st_I0))
270 itbl (st_i0,st_l0,st_p0,st_I0) dcon
271 = do st_I1 <- addToSS st_I0 (getName dcon)
272 return (sizeSS st_I0, (st_i0,st_l0,st_p0,st_I1))
274 literal st (MachInt j) = int st (fromIntegral j)
275 literal st (MachFloat r) = float st (fromRational r)
276 literal st (MachDouble r) = double st (fromRational r)
277 literal st (MachChar c) = int st c
280 = addr st ret_itbl_addr
282 ret_itbl_addr = case pk of
283 PtrRep -> stg_ctoi_ret_R1p_info
284 IntRep -> stg_ctoi_ret_R1n_info
285 AddrRep -> stg_ctoi_ret_R1n_info
286 CharRep -> stg_ctoi_ret_R1n_info
287 FloatRep -> stg_ctoi_ret_F1_info
288 DoubleRep -> stg_ctoi_ret_D1_info
289 _ -> pprPanic "mkBits.ctoi_itbl" (ppr pk)
292 = addr st ret_itbl_addr
294 ret_itbl_addr = case pk of
295 CharRep -> stg_gc_unbx_r1_info
296 IntRep -> stg_gc_unbx_r1_info
297 FloatRep -> stg_gc_f1_info
298 DoubleRep -> stg_gc_d1_info
300 foreign label "stg_ctoi_ret_R1p_info" stg_ctoi_ret_R1p_info :: Addr
301 foreign label "stg_ctoi_ret_R1n_info" stg_ctoi_ret_R1n_info :: Addr
302 foreign label "stg_ctoi_ret_F1_info" stg_ctoi_ret_F1_info :: Addr
303 foreign label "stg_ctoi_ret_D1_info" stg_ctoi_ret_D1_info :: Addr
305 foreign label "stg_gc_unbx_r1_info" stg_gc_unbx_r1_info :: Addr
306 foreign label "stg_gc_f1_info" stg_gc_f1_info :: Addr
307 foreign label "stg_gc_d1_info" stg_gc_d1_info :: Addr
309 -- The size in 16-bit entities of an instruction.
310 instrSize16s :: BCInstr -> Int
341 -- Make lists of host-sized words for literals, so that when the
342 -- words are placed in memory at increasing addresses, the
343 -- bit pattern is correct for the host's word size and endianness.
344 mkLitI :: Int -> [Word]
345 mkLitF :: Float -> [Word]
346 mkLitD :: Double -> [Word]
347 mkLitA :: Addr -> [Word]
351 arr <- newFloatArray ((0::Int),0)
352 writeFloatArray arr 0 f
353 f_arr <- castSTUArray arr
354 w0 <- readWordArray f_arr 0
361 arr <- newDoubleArray ((0::Int),1)
362 writeDoubleArray arr 0 d
363 d_arr <- castSTUArray arr
364 w0 <- readWordArray d_arr 0
365 w1 <- readWordArray d_arr 1
370 arr <- newDoubleArray ((0::Int),0)
371 writeDoubleArray arr 0 d
372 d_arr <- castSTUArray arr
373 w0 <- readWordArray d_arr 0
379 arr <- newIntArray ((0::Int),0)
380 writeIntArray arr 0 i
381 i_arr <- castSTUArray arr
382 w0 <- readWordArray i_arr 0
388 arr <- newAddrArray ((0::Int),0)
389 writeAddrArray arr 0 a
390 a_arr <- castSTUArray arr
391 w0 <- readWordArray a_arr 0
397 %************************************************************************
399 \subsection{Linking interpretables into something we can run}
401 %************************************************************************
406 data BCO# = BCO# ByteArray# -- instrs :: array Word16#
407 ByteArray# -- literals :: array Word32#
408 PtrArray# -- ptrs :: Array HValue
409 ByteArray# -- itbls :: Array Addr#
412 GLOBAL_VAR(v_cafTable, [], [HValue])
414 addCAF :: HValue -> IO ()
415 addCAF x = do xs <- readIORef v_cafTable
416 --putStrLn ("addCAF " ++ show (1 + length xs))
417 writeIORef v_cafTable (x:xs)
420 linkBCO ie ce (UnlinkedBCO nm insnsSS literalsSS ptrsSS itblsSS)
421 = do insns <- listFromSS insnsSS
422 literals <- listFromSS literalsSS
423 ptrs <- listFromSS ptrsSS
424 itbls <- listFromSS itblsSS
426 linked_ptrs <- mapM (lookupCE ce) ptrs
427 linked_itbls <- mapM (lookupIE ie) itbls
429 let n_insns = sizeSS insnsSS
430 n_literals = sizeSS literalsSS
431 n_ptrs = sizeSS ptrsSS
432 n_itbls = sizeSS itblsSS
434 let ptrs_arr = array (0, n_ptrs-1) (indexify linked_ptrs)
436 ptrs_parr = case ptrs_arr of Array lo hi parr -> parr
438 itbls_arr = array (0, n_itbls-1) (indexify linked_itbls)
439 :: UArray Int ItblPtr
440 itbls_barr = case itbls_arr of UArray lo hi barr -> barr
442 insns_arr | n_insns > 65535
443 = panic "linkBCO: >= 64k insns in BCO"
446 (indexify (fromIntegral n_insns:insns))
448 insns_barr = case insns_arr of UArray lo hi barr -> barr
450 literals_arr = array (0, n_literals-1) (indexify literals)
452 literals_barr = case literals_arr of UArray lo hi barr -> barr
454 indexify :: [a] -> [(Int, a)]
455 indexify xs = zip [0..] xs
457 BCO bco# <- newBCO insns_barr literals_barr ptrs_parr itbls_barr
459 -- WAS: return (unsafeCoerce# bco#)
460 case mkApUpd0# (unsafeCoerce# bco#) of
461 (# final_bco #) -> return final_bco
466 newBCO :: ByteArray# -> ByteArray# -> Array# a -> ByteArray# -> IO BCO
468 = IO (\s -> case newBCO# a b c d s of (# s1, bco #) -> (# s1, BCO bco #))
471 lookupCE :: ClosureEnv -> Either Name PrimOp -> IO HValue
472 lookupCE ce (Right primop)
473 = do m <- lookupSymbol (primopToCLabel primop "closure")
475 Just (Ptr addr) -> case addrToHValue# addr of
476 (# hval #) -> do addCAF hval
478 Nothing -> pprPanic "ByteCodeGen.lookupCE(primop)" (ppr primop)
479 lookupCE ce (Left nm)
480 = case lookupFM ce nm of
483 -> do m <- lookupSymbol (nameToCLabel nm "closure")
485 Just (Ptr addr) -> case addrToHValue# addr of
486 (# hval #) -> do addCAF 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)