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 )
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 :: ItblEnv -> ClosureEnv -> [UnlinkedBCO]
61 -> IO (ClosureEnv, [HValue])
62 linkSomeBCOs ie ce_in ul_bcos
63 = do let nms = map nameOfUnlinkedBCO ul_bcos
65 ( \ hvs -> let ce_out = addListToFM ce_in (zipLazily nms hvs)
66 in mapM (linkBCO ie ce_out) ul_bcos )
67 let ce_out = addListToFM ce_in (zip nms hvals)
68 return (ce_out, hvals)
70 -- A lazier zip, in which no demand is propagated to the second
71 -- list unless some demand is propagated to the snd of one of the
74 zipLazily (x:xs) ys = (x, head ys) : zipLazily xs (tail ys)
79 (SizedSeq Word16) -- insns
80 (SizedSeq Word) -- literals
81 (SizedSeq (Either Name PrimOp)) -- ptrs
82 (SizedSeq Name) -- itbl refs
84 nameOfUnlinkedBCO (UnlinkedBCO nm _ _ _ _) = nm
86 -- When translating expressions, we need to distinguish the root
87 -- BCO for the expression
88 type UnlinkedBCOExpr = (UnlinkedBCO, [UnlinkedBCO])
90 instance Outputable UnlinkedBCO where
91 ppr (UnlinkedBCO nm insns lits ptrs itbls)
92 = sep [text "BCO", ppr nm, text "with",
93 int (sizeSS insns), text "insns",
94 int (sizeSS lits), text "lits",
95 int (sizeSS ptrs), text "ptrs",
96 int (sizeSS itbls), text "itbls"]
99 -- these need a proper home
100 type ClosureEnv = FiniteMap Name HValue
101 data HValue = HValue -- dummy type, actually a pointer to some Real Code.
103 -- remove all entries for a given set of modules from the environment
104 filterNameMap :: [ModuleName] -> FiniteMap Name a -> FiniteMap Name a
105 filterNameMap mods env
106 = filterFM (\n _ -> moduleName (nameModule n) `notElem` mods) env
110 %************************************************************************
112 \subsection{The bytecode assembler}
114 %************************************************************************
116 The object format for bytecodes is: 16 bits for the opcode, and 16 for
117 each field -- so the code can be considered a sequence of 16-bit ints.
118 Each field denotes either a stack offset or number of items on the
119 stack (eg SLIDE), and index into the pointer table (eg PUSH_G), an
120 index into the literal table (eg PUSH_I/D/L), or a bytecode address in
124 -- Top level assembler fn.
125 assembleBCO :: ProtoBCO Name -> IO UnlinkedBCO
127 assembleBCO (ProtoBCO nm instrs origin)
129 -- pass 1: collect up the offsets of the local labels.
130 -- Remember that the first insn starts at offset 1 since offset 0
131 -- (eventually) will hold the total # of insns.
132 label_env = mkLabelEnv emptyFM 1 instrs
134 mkLabelEnv env i_offset [] = env
135 mkLabelEnv env i_offset (i:is)
137 = case i of LABEL n -> addToFM env n i_offset ; _ -> env
138 in mkLabelEnv new_env (i_offset + instrSize16s i) is
141 = case lookupFM label_env lab of
142 Just bco_offset -> bco_offset
143 Nothing -> pprPanic "assembleBCO.findLabel" (int lab)
145 do -- pass 2: generate the instruction, ptr and nonptr bits
146 insns <- return emptySS :: IO (SizedSeq Word16)
147 lits <- return emptySS :: IO (SizedSeq Word)
148 ptrs <- return emptySS :: IO (SizedSeq (Either Name PrimOp))
149 itbls <- return emptySS :: IO (SizedSeq Name)
150 let init_asm_state = (insns,lits,ptrs,itbls)
151 (final_insns, final_lits, final_ptrs, final_itbls)
152 <- mkBits findLabel init_asm_state instrs
154 return (UnlinkedBCO nm final_insns final_lits final_ptrs final_itbls)
156 -- instrs nonptrs ptrs itbls
157 type AsmState = (SizedSeq Word16, SizedSeq Word,
158 SizedSeq (Either Name PrimOp), SizedSeq Name)
160 data SizedSeq a = SizedSeq !Int [a]
161 emptySS = SizedSeq 0 []
162 addToSS (SizedSeq n r_xs) x = return (SizedSeq (n+1) (x:r_xs))
163 addListToSS (SizedSeq n r_xs) xs
164 = return (SizedSeq (n + length xs) (reverse xs ++ r_xs))
165 sizeSS (SizedSeq n r_xs) = n
166 listFromSS (SizedSeq n r_xs) = return (reverse r_xs)
169 -- This is where all the action is (pass 2 of the assembler)
170 mkBits :: (Int -> Int) -- label finder
172 -> [BCInstr] -- instructions (in)
175 mkBits findLabel st proto_insns
176 = foldM doInstr st proto_insns
178 doInstr :: AsmState -> BCInstr -> IO AsmState
181 ARGCHECK n -> instr2 st i_ARGCHECK n
182 PUSH_L o1 -> instr2 st i_PUSH_L o1
183 PUSH_LL o1 o2 -> instr3 st i_PUSH_LL o1 o2
184 PUSH_LLL o1 o2 o3 -> instr4 st i_PUSH_LLL o1 o2 o3
185 PUSH_G nm -> do (p, st2) <- ptr st nm
186 instr2 st2 i_PUSH_G p
187 PUSH_AS nm pk -> do (p, st2) <- ptr st (Left nm)
188 (np, st3) <- ctoi_itbl st2 pk
189 instr3 st3 i_PUSH_AS p np
190 PUSH_UBX lit nws -> do (np, st2) <- literal st lit
191 instr3 st2 i_PUSH_UBX np nws
192 PUSH_TAG tag -> instr2 st i_PUSH_TAG tag
193 SLIDE n by -> instr3 st i_SLIDE n by
194 ALLOC n -> instr2 st i_ALLOC n
195 MKAP off sz -> instr3 st i_MKAP off sz
196 UNPACK n -> instr2 st i_UNPACK n
197 UPK_TAG n m k -> instr4 st i_UPK_TAG n m k
198 PACK dcon sz -> do (itbl_no,st2) <- itbl st dcon
199 instr3 st2 i_PACK itbl_no sz
200 LABEL lab -> return st
201 TESTLT_I i l -> do (np, st2) <- int st i
202 instr3 st2 i_TESTLT_I np (findLabel l)
203 TESTEQ_I i l -> do (np, st2) <- int st i
204 instr3 st2 i_TESTEQ_I np (findLabel l)
205 TESTLT_F f l -> do (np, st2) <- float st f
206 instr3 st2 i_TESTLT_F np (findLabel l)
207 TESTEQ_F f l -> do (np, st2) <- float st f
208 instr3 st2 i_TESTEQ_F np (findLabel l)
209 TESTLT_D d l -> do (np, st2) <- double st d
210 instr3 st2 i_TESTLT_D np (findLabel l)
211 TESTEQ_D d l -> do (np, st2) <- double st d
212 instr3 st2 i_TESTEQ_D np (findLabel l)
213 TESTLT_P i l -> instr3 st i_TESTLT_P i (findLabel l)
214 TESTEQ_P i l -> instr3 st i_TESTEQ_P i (findLabel l)
215 CASEFAIL -> instr1 st i_CASEFAIL
216 ENTER -> instr1 st i_ENTER
217 RETURN rep -> do (itbl_no,st2) <- itoc_itbl st rep
218 instr2 st2 i_RETURN itbl_no
223 instr1 (st_i0,st_l0,st_p0,st_I0) i1
224 = do st_i1 <- addToSS st_i0 (i2s i1)
225 return (st_i1,st_l0,st_p0,st_I0)
227 instr2 (st_i0,st_l0,st_p0,st_I0) i1 i2
228 = do st_i1 <- addToSS st_i0 (i2s i1)
229 st_i2 <- addToSS st_i1 (i2s i2)
230 return (st_i2,st_l0,st_p0,st_I0)
232 instr3 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3
233 = do st_i1 <- addToSS st_i0 (i2s i1)
234 st_i2 <- addToSS st_i1 (i2s i2)
235 st_i3 <- addToSS st_i2 (i2s i3)
236 return (st_i3,st_l0,st_p0,st_I0)
238 instr4 (st_i0,st_l0,st_p0,st_I0) i1 i2 i3 i4
239 = do st_i1 <- addToSS st_i0 (i2s i1)
240 st_i2 <- addToSS st_i1 (i2s i2)
241 st_i3 <- addToSS st_i2 (i2s i3)
242 st_i4 <- addToSS st_i3 (i2s i4)
243 return (st_i4,st_l0,st_p0,st_I0)
245 float (st_i0,st_l0,st_p0,st_I0) f
246 = do let ws = mkLitF f
247 st_l1 <- addListToSS st_l0 ws
248 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
250 double (st_i0,st_l0,st_p0,st_I0) d
251 = do let ws = mkLitD d
252 st_l1 <- addListToSS st_l0 ws
253 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
255 int (st_i0,st_l0,st_p0,st_I0) i
256 = do let ws = mkLitI i
257 st_l1 <- addListToSS st_l0 ws
258 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
260 addr (st_i0,st_l0,st_p0,st_I0) a
261 = do let ws = mkLitA a
262 st_l1 <- addListToSS st_l0 ws
263 return (sizeSS st_l0, (st_i0,st_l1,st_p0,st_I0))
265 ptr (st_i0,st_l0,st_p0,st_I0) p
266 = do st_p1 <- addToSS st_p0 p
267 return (sizeSS st_p0, (st_i0,st_l0,st_p1,st_I0))
269 itbl (st_i0,st_l0,st_p0,st_I0) dcon
270 = do st_I1 <- addToSS st_I0 (getName dcon)
271 return (sizeSS st_I0, (st_i0,st_l0,st_p0,st_I1))
273 literal st (MachWord w) = int st (fromIntegral w)
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 WordRep -> stg_ctoi_ret_R1n_info
285 IntRep -> stg_ctoi_ret_R1n_info
286 AddrRep -> stg_ctoi_ret_R1n_info
287 CharRep -> stg_ctoi_ret_R1n_info
288 FloatRep -> stg_ctoi_ret_F1_info
289 DoubleRep -> stg_ctoi_ret_D1_info
290 _ -> pprPanic "mkBits.ctoi_itbl" (ppr pk)
293 = addr st ret_itbl_addr
295 ret_itbl_addr = case pk of
296 CharRep -> stg_gc_unbx_r1_info
297 IntRep -> stg_gc_unbx_r1_info
298 FloatRep -> stg_gc_f1_info
299 DoubleRep -> stg_gc_d1_info
301 foreign label "stg_ctoi_ret_R1p_info" stg_ctoi_ret_R1p_info :: Addr
302 foreign label "stg_ctoi_ret_R1n_info" stg_ctoi_ret_R1n_info :: Addr
303 foreign label "stg_ctoi_ret_F1_info" stg_ctoi_ret_F1_info :: Addr
304 foreign label "stg_ctoi_ret_D1_info" stg_ctoi_ret_D1_info :: Addr
306 foreign label "stg_gc_unbx_r1_info" stg_gc_unbx_r1_info :: Addr
307 foreign label "stg_gc_f1_info" stg_gc_f1_info :: Addr
308 foreign label "stg_gc_d1_info" stg_gc_d1_info :: Addr
310 -- The size in 16-bit entities of an instruction.
311 instrSize16s :: BCInstr -> Int
342 -- Make lists of host-sized words for literals, so that when the
343 -- words are placed in memory at increasing addresses, the
344 -- bit pattern is correct for the host's word size and endianness.
345 mkLitI :: Int -> [Word]
346 mkLitF :: Float -> [Word]
347 mkLitD :: Double -> [Word]
348 mkLitA :: Addr -> [Word]
352 arr <- newFloatArray ((0::Int),0)
353 writeFloatArray arr 0 f
354 f_arr <- castSTUArray arr
355 w0 <- readWordArray f_arr 0
362 arr <- newDoubleArray ((0::Int),1)
363 writeDoubleArray arr 0 d
364 d_arr <- castSTUArray arr
365 w0 <- readWordArray d_arr 0
366 w1 <- readWordArray d_arr 1
371 arr <- newDoubleArray ((0::Int),0)
372 writeDoubleArray arr 0 d
373 d_arr <- castSTUArray arr
374 w0 <- readWordArray d_arr 0
380 arr <- newIntArray ((0::Int),0)
381 writeIntArray arr 0 i
382 i_arr <- castSTUArray arr
383 w0 <- readWordArray i_arr 0
389 arr <- newAddrArray ((0::Int),0)
390 writeAddrArray arr 0 a
391 a_arr <- castSTUArray arr
392 w0 <- readWordArray a_arr 0
398 %************************************************************************
400 \subsection{Linking interpretables into something we can run}
402 %************************************************************************
407 data BCO# = BCO# ByteArray# -- instrs :: array Word16#
408 ByteArray# -- literals :: array Word32#
409 PtrArray# -- ptrs :: Array HValue
410 ByteArray# -- itbls :: Array Addr#
413 linkBCO ie ce (UnlinkedBCO nm insnsSS literalsSS ptrsSS itblsSS)
414 = do insns <- listFromSS insnsSS
415 literals <- listFromSS literalsSS
416 ptrs <- listFromSS ptrsSS
417 itbls <- listFromSS itblsSS
419 linked_ptrs <- mapM (lookupCE ce) ptrs
420 linked_itbls <- mapM (lookupIE ie) itbls
422 let n_insns = sizeSS insnsSS
423 n_literals = sizeSS literalsSS
424 n_ptrs = sizeSS ptrsSS
425 n_itbls = sizeSS itblsSS
427 let ptrs_arr = array (0, n_ptrs-1) (indexify linked_ptrs)
429 ptrs_parr = case ptrs_arr of Array lo hi parr -> parr
431 itbls_arr = array (0, n_itbls-1) (indexify linked_itbls)
432 :: UArray Int ItblPtr
433 itbls_barr = case itbls_arr of UArray lo hi barr -> barr
435 insns_arr | n_insns > 65535
436 = panic "linkBCO: >= 64k insns in BCO"
439 (indexify (fromIntegral n_insns:insns))
441 insns_barr = case insns_arr of UArray lo hi barr -> barr
443 literals_arr = array (0, n_literals-1) (indexify literals)
445 literals_barr = case literals_arr of UArray lo hi barr -> barr
447 indexify :: [a] -> [(Int, a)]
448 indexify xs = zip [0..] xs
450 BCO bco# <- newBCO insns_barr literals_barr ptrs_parr itbls_barr
452 -- WAS: return (unsafeCoerce# bco#)
453 case mkApUpd0# (unsafeCoerce# bco#) of
454 (# final_bco #) -> return final_bco
459 newBCO :: ByteArray# -> ByteArray# -> Array# a -> ByteArray# -> IO BCO
461 = IO (\s -> case newBCO# a b c d s of (# s1, bco #) -> (# s1, BCO bco #))
464 lookupCE :: ClosureEnv -> Either Name PrimOp -> IO HValue
465 lookupCE ce (Right primop)
466 = do m <- lookupSymbol (primopToCLabel primop "closure")
468 Just (Ptr addr) -> case addrToHValue# addr of
469 (# hval #) -> return hval
470 Nothing -> pprPanic "ByteCodeGen.lookupCE(primop)" (ppr primop)
471 lookupCE ce (Left nm)
472 = case lookupFM ce nm of
475 -> do m <- lookupSymbol (nameToCLabel nm "closure")
477 Just (Ptr addr) -> case addrToHValue# addr of
478 (# hval #) -> return hval
479 Nothing -> pprPanic "ByteCodeGen.lookupCE" (ppr nm)
481 lookupIE :: ItblEnv -> Name -> IO (Ptr a)
483 = case lookupFM ie con_nm of
484 Just (Ptr a) -> return (Ptr a)
486 -> do -- try looking up in the object files.
487 m <- lookupSymbol (nameToCLabel con_nm "con_info")
489 Just addr -> return addr
491 -> do -- perhaps a nullary constructor?
492 n <- lookupSymbol (nameToCLabel con_nm "static_info")
494 Just addr -> return addr
495 Nothing -> pprPanic "ByteCodeGen.lookupIE" (ppr con_nm)
497 -- HACKS!!! ToDo: cleaner
498 nameToCLabel :: Name -> String{-suffix-} -> String
499 nameToCLabel n suffix
500 = _UNPK_(moduleNameFS (rdrNameModule rn))
501 ++ '_':occNameString(rdrNameOcc rn) ++ '_':suffix
502 where rn = toRdrName n
504 primopToCLabel :: PrimOp -> String{-suffix-} -> String
505 primopToCLabel primop suffix
506 = let str = "PrelPrimopWrappers_" ++ occNameString (primOpOcc primop) ++ '_':suffix
507 in --trace ("primopToCLabel: " ++ str)
512 %************************************************************************
514 \subsection{Connect to actual values for bytecode opcodes}
516 %************************************************************************
520 #include "Bytecodes.h"
522 i_ARGCHECK = (bci_ARGCHECK :: Int)
523 i_PUSH_L = (bci_PUSH_L :: Int)
524 i_PUSH_LL = (bci_PUSH_LL :: Int)
525 i_PUSH_LLL = (bci_PUSH_LLL :: Int)
526 i_PUSH_G = (bci_PUSH_G :: Int)
527 i_PUSH_AS = (bci_PUSH_AS :: Int)
528 i_PUSH_UBX = (bci_PUSH_UBX :: Int)
529 i_PUSH_TAG = (bci_PUSH_TAG :: Int)
530 i_SLIDE = (bci_SLIDE :: Int)
531 i_ALLOC = (bci_ALLOC :: Int)
532 i_MKAP = (bci_MKAP :: Int)
533 i_UNPACK = (bci_UNPACK :: Int)
534 i_UPK_TAG = (bci_UPK_TAG :: Int)
535 i_PACK = (bci_PACK :: Int)
536 i_TESTLT_I = (bci_TESTLT_I :: Int)
537 i_TESTEQ_I = (bci_TESTEQ_I :: Int)
538 i_TESTLT_F = (bci_TESTLT_F :: Int)
539 i_TESTEQ_F = (bci_TESTEQ_F :: Int)
540 i_TESTLT_D = (bci_TESTLT_D :: Int)
541 i_TESTEQ_D = (bci_TESTEQ_D :: Int)
542 i_TESTLT_P = (bci_TESTLT_P :: Int)
543 i_TESTEQ_P = (bci_TESTEQ_P :: Int)
544 i_CASEFAIL = (bci_CASEFAIL :: Int)
545 i_ENTER = (bci_ENTER :: Int)
546 i_RETURN = (bci_RETURN :: Int)