1 -----------------------------------------------------------------------------
3 -- (c) The University of Glasgow, 2004
5 -- Parser for concrete Cmm.
7 -----------------------------------------------------------------------------
10 module CmmParse ( parseCmmFile ) where
19 import CgTailCall ( pushUnboxedTuple )
20 import CgStackery ( emitPushUpdateFrame )
21 import ClosureInfo ( C_SRT(..) )
22 import CgCallConv ( smallLiveness )
23 import CgClosure ( emitBlackHoleCode )
24 import CostCentre ( dontCareCCS )
28 import CmmUtils ( mkIntCLit )
32 import SMRep ( fixedHdrSize, CgRep(..) )
35 import ForeignCall ( CCallConv(..) )
36 import Literal ( mkMachInt )
40 import DynFlags ( DynFlags, DynFlag(..) )
41 import Packages ( HomeModules )
42 import StaticFlags ( opt_SccProfilingOn )
43 import ErrUtils ( printError, dumpIfSet_dyn, showPass )
44 import StringBuffer ( hGetStringBuffer )
46 import Panic ( panic )
47 import Constants ( wORD_SIZE )
52 #include "HsVersions.h"
56 ':' { L _ (CmmT_SpecChar ':') }
57 ';' { L _ (CmmT_SpecChar ';') }
58 '{' { L _ (CmmT_SpecChar '{') }
59 '}' { L _ (CmmT_SpecChar '}') }
60 '[' { L _ (CmmT_SpecChar '[') }
61 ']' { L _ (CmmT_SpecChar ']') }
62 '(' { L _ (CmmT_SpecChar '(') }
63 ')' { L _ (CmmT_SpecChar ')') }
64 '=' { L _ (CmmT_SpecChar '=') }
65 '`' { L _ (CmmT_SpecChar '`') }
66 '~' { L _ (CmmT_SpecChar '~') }
67 '/' { L _ (CmmT_SpecChar '/') }
68 '*' { L _ (CmmT_SpecChar '*') }
69 '%' { L _ (CmmT_SpecChar '%') }
70 '-' { L _ (CmmT_SpecChar '-') }
71 '+' { L _ (CmmT_SpecChar '+') }
72 '&' { L _ (CmmT_SpecChar '&') }
73 '^' { L _ (CmmT_SpecChar '^') }
74 '|' { L _ (CmmT_SpecChar '|') }
75 '>' { L _ (CmmT_SpecChar '>') }
76 '<' { L _ (CmmT_SpecChar '<') }
77 ',' { L _ (CmmT_SpecChar ',') }
78 '!' { L _ (CmmT_SpecChar '!') }
80 '..' { L _ (CmmT_DotDot) }
81 '::' { L _ (CmmT_DoubleColon) }
82 '>>' { L _ (CmmT_Shr) }
83 '<<' { L _ (CmmT_Shl) }
84 '>=' { L _ (CmmT_Ge) }
85 '<=' { L _ (CmmT_Le) }
86 '==' { L _ (CmmT_Eq) }
87 '!=' { L _ (CmmT_Ne) }
88 '&&' { L _ (CmmT_BoolAnd) }
89 '||' { L _ (CmmT_BoolOr) }
91 'CLOSURE' { L _ (CmmT_CLOSURE) }
92 'INFO_TABLE' { L _ (CmmT_INFO_TABLE) }
93 'INFO_TABLE_RET'{ L _ (CmmT_INFO_TABLE_RET) }
94 'INFO_TABLE_FUN'{ L _ (CmmT_INFO_TABLE_FUN) }
95 'INFO_TABLE_CONSTR'{ L _ (CmmT_INFO_TABLE_CONSTR) }
96 'INFO_TABLE_SELECTOR'{ L _ (CmmT_INFO_TABLE_SELECTOR) }
97 'else' { L _ (CmmT_else) }
98 'export' { L _ (CmmT_export) }
99 'section' { L _ (CmmT_section) }
100 'align' { L _ (CmmT_align) }
101 'goto' { L _ (CmmT_goto) }
102 'if' { L _ (CmmT_if) }
103 'jump' { L _ (CmmT_jump) }
104 'foreign' { L _ (CmmT_foreign) }
105 'import' { L _ (CmmT_import) }
106 'switch' { L _ (CmmT_switch) }
107 'case' { L _ (CmmT_case) }
108 'default' { L _ (CmmT_default) }
109 'bits8' { L _ (CmmT_bits8) }
110 'bits16' { L _ (CmmT_bits16) }
111 'bits32' { L _ (CmmT_bits32) }
112 'bits64' { L _ (CmmT_bits64) }
113 'float32' { L _ (CmmT_float32) }
114 'float64' { L _ (CmmT_float64) }
116 GLOBALREG { L _ (CmmT_GlobalReg $$) }
117 NAME { L _ (CmmT_Name $$) }
118 STRING { L _ (CmmT_String $$) }
119 INT { L _ (CmmT_Int $$) }
120 FLOAT { L _ (CmmT_Float $$) }
122 %monad { P } { >>= } { return }
123 %lexer { cmmlex } { L _ CmmT_EOF }
125 %tokentype { Located CmmToken }
127 -- C-- operator precedences, taken from the C-- spec
128 %right '||' -- non-std extension, called %disjoin in C--
129 %right '&&' -- non-std extension, called %conjoin in C--
131 %nonassoc '>=' '>' '<=' '<' '!=' '=='
143 : {- empty -} { return () }
144 | cmmtop cmm { do $1; $2 }
146 cmmtop :: { ExtCode }
150 | 'CLOSURE' '(' NAME ',' NAME lits ')' ';'
151 { do lits <- sequence $6;
152 staticClosure $3 $5 (map getLit lits) }
154 -- The only static closures in the RTS are dummy closures like
155 -- stg_END_TSO_QUEUE_closure and stg_dummy_ret. We don't need
156 -- to provide the full generality of static closures here.
158 -- * CCS can always be CCS_DONT_CARE
159 -- * closure is always extern
160 -- * payload is always empty
161 -- * we can derive closure and info table labels from a single NAME
163 cmmdata :: { ExtCode }
164 : 'section' STRING '{' statics '}'
165 { do ss <- sequence $4;
166 code (emitData (section $2) (concat ss)) }
168 statics :: { [ExtFCode [CmmStatic]] }
170 | static statics { $1 : $2 }
172 -- Strings aren't used much in the RTS HC code, so it doesn't seem
173 -- worth allowing inline strings. C-- doesn't allow them anyway.
174 static :: { ExtFCode [CmmStatic] }
175 : NAME ':' { return [CmmDataLabel (mkRtsDataLabelFS $1)] }
176 | type expr ';' { do e <- $2;
177 return [CmmStaticLit (getLit e)] }
178 | type ';' { return [CmmUninitialised
179 (machRepByteWidth $1)] }
180 | 'bits8' '[' ']' STRING ';' { return [CmmString $4] }
181 | 'bits8' '[' INT ']' ';' { return [CmmUninitialised
183 | typenot8 '[' INT ']' ';' { return [CmmUninitialised
184 (machRepByteWidth $1 *
186 | 'align' INT ';' { return [CmmAlign (fromIntegral $2)] }
187 | 'CLOSURE' '(' NAME lits ')'
188 { do lits <- sequence $4;
189 return $ map CmmStaticLit $
190 mkStaticClosure (mkRtsInfoLabelFS $3)
191 dontCareCCS (map getLit lits) [] [] [] }
192 -- arrays of closures required for the CHARLIKE & INTLIKE arrays
194 lits :: { [ExtFCode CmmExpr] }
196 | ',' expr lits { $2 : $3 }
198 cmmproc :: { ExtCode }
200 { do (info_lbl, info1, info2) <- $1;
201 stmts <- getCgStmtsEC (loopDecls $3)
202 blks <- code (cgStmtsToBlocks stmts)
203 code (emitInfoTableAndCode info_lbl info1 info2 [] blks) }
206 { do (info_lbl, info1, info2) <- $1;
207 code (emitInfoTableAndCode info_lbl info1 info2 [] []) }
210 { do stmts <- getCgStmtsEC (loopDecls $3);
211 blks <- code (cgStmtsToBlocks stmts)
212 code (emitProc [] (mkRtsCodeLabelFS $1) [] blks) }
214 info :: { ExtFCode (CLabel, [CmmLit],[CmmLit]) }
215 : 'INFO_TABLE' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
216 -- ptrs, nptrs, closure type, description, type
217 { stdInfo $3 $5 $7 0 $9 $11 $13 }
219 | 'INFO_TABLE_FUN' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ',' INT ')'
220 -- ptrs, nptrs, closure type, description, type, fun type
221 { funInfo $3 $5 $7 $9 $11 $13 $15 }
223 | 'INFO_TABLE_CONSTR' '(' NAME ',' INT ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
224 -- ptrs, nptrs, tag, closure type, description, type
225 { stdInfo $3 $5 $7 $9 $11 $13 $15 }
227 | 'INFO_TABLE_SELECTOR' '(' NAME ',' INT ',' INT ',' STRING ',' STRING ')'
228 -- selector, closure type, description, type
229 { basicInfo $3 (mkIntCLit (fromIntegral $5)) 0 $7 $9 $11 }
231 | 'INFO_TABLE_RET' '(' NAME ',' INT ',' INT ',' INT maybe_vec ')'
232 { retInfo $3 $5 $7 $9 $10 }
234 maybe_vec :: { [CmmLit] }
236 | ',' NAME maybe_vec { CmmLabel (mkRtsCodeLabelFS $2) : $3 }
239 : {- empty -} { return () }
240 | decl body { do $1; $2 }
241 | stmt body { do $1; $2 }
244 : type names ';' { mapM_ (newLocal $1) $2 }
245 | 'import' names ';' { return () } -- ignore imports
246 | 'export' names ';' { return () } -- ignore exports
248 names :: { [FastString] }
250 | NAME ',' names { $1 : $3 }
255 | block_id ':' { code (labelC $1) }
258 { do reg <- $1; e <- $3; stmtEC (CmmAssign reg e) }
259 | type '[' expr ']' '=' expr ';'
261 | 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
262 {% foreignCall $2 [] $3 $5 $7 }
263 | lreg '=' 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
264 {% let result = do r <- $1; return (r,NoHint) in
265 foreignCall $4 [result] $5 $7 $9 }
266 | STRING lreg '=' 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
267 {% do h <- parseHint $1;
268 let result = do r <- $2; return (r,h) in
269 foreignCall $5 [result] $6 $8 $10 }
270 -- stmt-level macros, stealing syntax from ordinary C-- function calls.
271 -- Perhaps we ought to use the %%-form?
272 | NAME '(' exprs0 ')' ';'
274 | 'switch' maybe_range expr '{' arms default '}'
275 { doSwitch $2 $3 $5 $6 }
276 | 'goto' block_id ';'
277 { stmtEC (CmmBranch $2) }
278 | 'jump' expr {-maybe_actuals-} ';'
279 { do e <- $2; stmtEC (CmmJump e []) }
280 | 'if' bool_expr '{' body '}' else
281 { ifThenElse $2 $4 $6 }
283 bool_expr :: { ExtFCode BoolExpr }
285 | expr { do e <- $1; return (BoolTest e) }
287 bool_op :: { ExtFCode BoolExpr }
288 : bool_expr '&&' bool_expr { do e1 <- $1; e2 <- $3;
289 return (BoolAnd e1 e2) }
290 | bool_expr '||' bool_expr { do e1 <- $1; e2 <- $3;
291 return (BoolOr e1 e2) }
292 | '!' bool_expr { do e <- $2; return (BoolNot e) }
293 | '(' bool_op ')' { $2 }
295 -- This is not C-- syntax. What to do?
296 vols :: { Maybe [GlobalReg] }
297 : {- empty -} { Nothing }
298 | '[' globals ']' { Just $2 }
300 globals :: { [GlobalReg] }
302 | GLOBALREG ',' globals { $1 : $3 }
304 maybe_range :: { Maybe (Int,Int) }
305 : '[' INT '..' INT ']' { Just (fromIntegral $2, fromIntegral $4) }
306 | {- empty -} { Nothing }
308 arms :: { [([Int],ExtCode)] }
310 | arm arms { $1 : $2 }
312 arm :: { ([Int],ExtCode) }
313 : 'case' ints ':' '{' body '}' { ($2, $5) }
316 : INT { [ fromIntegral $1 ] }
317 | INT ',' ints { fromIntegral $1 : $3 }
319 default :: { Maybe ExtCode }
320 : 'default' ':' '{' body '}' { Just $4 }
321 -- taking a few liberties with the C-- syntax here; C-- doesn't have
322 -- 'default' branches
323 | {- empty -} { Nothing }
326 : {- empty -} { nopEC }
327 | 'else' '{' body '}' { $3 }
329 -- we have to write this out longhand so that Happy's precedence rules
331 expr :: { ExtFCode CmmExpr }
332 : expr '/' expr { mkMachOp MO_U_Quot [$1,$3] }
333 | expr '*' expr { mkMachOp MO_Mul [$1,$3] }
334 | expr '%' expr { mkMachOp MO_U_Rem [$1,$3] }
335 | expr '-' expr { mkMachOp MO_Sub [$1,$3] }
336 | expr '+' expr { mkMachOp MO_Add [$1,$3] }
337 | expr '>>' expr { mkMachOp MO_U_Shr [$1,$3] }
338 | expr '<<' expr { mkMachOp MO_Shl [$1,$3] }
339 | expr '&' expr { mkMachOp MO_And [$1,$3] }
340 | expr '^' expr { mkMachOp MO_Xor [$1,$3] }
341 | expr '|' expr { mkMachOp MO_Or [$1,$3] }
342 | expr '>=' expr { mkMachOp MO_U_Ge [$1,$3] }
343 | expr '>' expr { mkMachOp MO_U_Gt [$1,$3] }
344 | expr '<=' expr { mkMachOp MO_U_Le [$1,$3] }
345 | expr '<' expr { mkMachOp MO_U_Lt [$1,$3] }
346 | expr '!=' expr { mkMachOp MO_Ne [$1,$3] }
347 | expr '==' expr { mkMachOp MO_Eq [$1,$3] }
348 | '~' expr { mkMachOp MO_Not [$2] }
349 | '-' expr { mkMachOp MO_S_Neg [$2] }
350 | expr0 '`' NAME '`' expr0 {% do { mo <- nameToMachOp $3 ;
351 return (mkMachOp mo [$1,$5]) } }
354 expr0 :: { ExtFCode CmmExpr }
355 : INT maybe_ty { return (CmmLit (CmmInt $1 $2)) }
356 | FLOAT maybe_ty { return (CmmLit (CmmFloat $1 $2)) }
357 | STRING { do s <- code (mkStringCLit $1);
360 | type '[' expr ']' { do e <- $3; return (CmmLoad e $1) }
361 | '%' NAME '(' exprs0 ')' {% exprOp $2 $4 }
362 | '(' expr ')' { $2 }
365 -- leaving out the type of a literal gives you the native word size in C--
366 maybe_ty :: { MachRep }
367 : {- empty -} { wordRep }
370 hint_exprs0 :: { [ExtFCode (CmmExpr, MachHint)] }
374 hint_exprs :: { [ExtFCode (CmmExpr, MachHint)] }
376 | hint_expr ',' hint_exprs { $1 : $3 }
378 hint_expr :: { ExtFCode (CmmExpr, MachHint) }
379 : expr { do e <- $1; return (e, inferHint e) }
380 | expr STRING {% do h <- parseHint $2;
382 e <- $1; return (e,h) }
384 exprs0 :: { [ExtFCode CmmExpr] }
388 exprs :: { [ExtFCode CmmExpr] }
390 | expr ',' exprs { $1 : $3 }
392 reg :: { ExtFCode CmmExpr }
393 : NAME { lookupName $1 }
394 | GLOBALREG { return (CmmReg (CmmGlobal $1)) }
396 lreg :: { ExtFCode CmmReg }
397 : NAME { do e <- lookupName $1;
401 other -> pprPanic "CmmParse:" (ftext $1 <> text " not a register") }
402 | GLOBALREG { return (CmmGlobal $1) }
404 block_id :: { BlockId }
405 : NAME { BlockId (newTagUnique (getUnique $1) 'L') }
406 -- TODO: ugh. The unique of a FastString has a null
407 -- tag, so we have to put our own tag on. We should
408 -- really make a new unique for every label, and keep
409 -- them in an environment.
415 typenot8 :: { MachRep }
422 section :: String -> Section
423 section "text" = Text
424 section "data" = Data
425 section "rodata" = ReadOnlyData
426 section "bss" = UninitialisedData
427 section s = OtherSection s
429 -- mkMachOp infers the type of the MachOp from the type of its first
430 -- argument. We assume that this is correct: for MachOps that don't have
431 -- symmetrical args (e.g. shift ops), the first arg determines the type of
433 mkMachOp :: (MachRep -> MachOp) -> [ExtFCode CmmExpr] -> ExtFCode CmmExpr
434 mkMachOp fn args = do
435 arg_exprs <- sequence args
436 return (CmmMachOp (fn (cmmExprRep (head arg_exprs))) arg_exprs)
438 getLit :: CmmExpr -> CmmLit
439 getLit (CmmLit l) = l
440 getLit (CmmMachOp (MO_S_Neg _) [CmmLit (CmmInt i r)]) = CmmInt (negate i) r
441 getLit _ = panic "invalid literal" -- TODO messy failure
443 nameToMachOp :: FastString -> P (MachRep -> MachOp)
445 case lookupUFM machOps name of
446 Nothing -> fail ("unknown primitive " ++ unpackFS name)
449 exprOp :: FastString -> [ExtFCode CmmExpr] -> P (ExtFCode CmmExpr)
450 exprOp name args_code =
451 case lookupUFM exprMacros name of
452 Just f -> return $ do
453 args <- sequence args_code
456 mo <- nameToMachOp name
457 return $ mkMachOp mo args_code
459 exprMacros :: UniqFM ([CmmExpr] -> CmmExpr)
460 exprMacros = listToUFM [
461 ( FSLIT("ENTRY_CODE"), \ [x] -> entryCode x ),
462 ( FSLIT("GET_ENTRY"), \ [x] -> entryCode (closureInfoPtr x) ),
463 ( FSLIT("STD_INFO"), \ [x] -> infoTable x ),
464 ( FSLIT("GET_STD_INFO"), \ [x] -> infoTable (closureInfoPtr x) ),
465 ( FSLIT("GET_FUN_INFO"), \ [x] -> funInfoTable (closureInfoPtr x) ),
466 ( FSLIT("INFO_TYPE"), \ [x] -> infoTableClosureType x ),
467 ( FSLIT("INFO_PTRS"), \ [x] -> infoTablePtrs x ),
468 ( FSLIT("INFO_NPTRS"), \ [x] -> infoTableNonPtrs x ),
469 ( FSLIT("RET_VEC"), \ [info, conZ] -> retVec info conZ )
472 -- we understand a subset of C-- primitives:
473 machOps = listToUFM $
474 map (\(x, y) -> (mkFastString x, y)) [
481 ( "quot", MO_S_Quot ),
483 ( "divu", MO_U_Quot ),
484 ( "modu", MO_U_Rem ),
502 ( "fneg", MO_S_Neg ),
509 ( "shrl", MO_U_Shr ),
510 ( "shra", MO_S_Shr ),
512 ( "lobits8", flip MO_U_Conv I8 ),
513 ( "lobits16", flip MO_U_Conv I16 ),
514 ( "lobits32", flip MO_U_Conv I32 ),
515 ( "lobits64", flip MO_U_Conv I64 ),
516 ( "sx16", flip MO_S_Conv I16 ),
517 ( "sx32", flip MO_S_Conv I32 ),
518 ( "sx64", flip MO_S_Conv I64 ),
519 ( "zx16", flip MO_U_Conv I16 ),
520 ( "zx32", flip MO_U_Conv I32 ),
521 ( "zx64", flip MO_U_Conv I64 ),
522 ( "f2f32", flip MO_S_Conv F32 ), -- TODO; rounding mode
523 ( "f2f64", flip MO_S_Conv F64 ), -- TODO; rounding mode
524 ( "f2i8", flip MO_S_Conv I8 ),
525 ( "f2i16", flip MO_S_Conv I8 ),
526 ( "f2i32", flip MO_S_Conv I8 ),
527 ( "f2i64", flip MO_S_Conv I8 ),
528 ( "i2f32", flip MO_S_Conv F32 ),
529 ( "i2f64", flip MO_S_Conv F64 )
532 parseHint :: String -> P MachHint
533 parseHint "ptr" = return PtrHint
534 parseHint "signed" = return SignedHint
535 parseHint "float" = return FloatHint
536 parseHint str = fail ("unrecognised hint: " ++ str)
538 -- labels are always pointers, so we might as well infer the hint
539 inferHint :: CmmExpr -> MachHint
540 inferHint (CmmLit (CmmLabel _)) = PtrHint
541 inferHint (CmmReg (CmmGlobal g)) | isPtrGlobalReg g = PtrHint
544 isPtrGlobalReg Sp = True
545 isPtrGlobalReg SpLim = True
546 isPtrGlobalReg Hp = True
547 isPtrGlobalReg HpLim = True
548 isPtrGlobalReg CurrentTSO = True
549 isPtrGlobalReg CurrentNursery = True
550 isPtrGlobalReg _ = False
553 happyError = srcParseFail
555 -- -----------------------------------------------------------------------------
556 -- Statement-level macros
558 stmtMacro :: FastString -> [ExtFCode CmmExpr] -> P ExtCode
559 stmtMacro fun args_code = do
560 case lookupUFM stmtMacros fun of
561 Nothing -> fail ("unknown macro: " ++ unpackFS fun)
562 Just fcode -> return $ do
563 args <- sequence args_code
566 stmtMacros :: UniqFM ([CmmExpr] -> Code)
567 stmtMacros = listToUFM [
568 ( FSLIT("CCS_ALLOC"), \[words,ccs] -> profAlloc words ccs ),
569 ( FSLIT("CLOSE_NURSERY"), \[] -> emitCloseNursery ),
570 ( FSLIT("ENTER_CCS_PAP_CL"), \[e] -> enterCostCentrePAP e ),
571 ( FSLIT("ENTER_CCS_THUNK"), \[e] -> enterCostCentreThunk e ),
572 ( FSLIT("HP_CHK_GEN"), \[words,liveness,reentry] ->
573 hpChkGen words liveness reentry ),
574 ( FSLIT("HP_CHK_NP_ASSIGN_SP0"), \[e,f] -> hpChkNodePointsAssignSp0 e f ),
575 ( FSLIT("LOAD_THREAD_STATE"), \[] -> emitLoadThreadState ),
576 ( FSLIT("LDV_ENTER"), \[e] -> ldvEnter e ),
577 ( FSLIT("LDV_RECORD_CREATE"), \[e] -> ldvRecordCreate e ),
578 ( FSLIT("OPEN_NURSERY"), \[] -> emitOpenNursery ),
579 ( FSLIT("PUSH_UPD_FRAME"), \[sp,e] -> emitPushUpdateFrame sp e ),
580 ( FSLIT("SAVE_THREAD_STATE"), \[] -> emitSaveThreadState ),
581 ( FSLIT("SET_HDR"), \[ptr,info,ccs] ->
582 emitSetDynHdr ptr info ccs ),
583 ( FSLIT("STK_CHK_GEN"), \[words,liveness,reentry] ->
584 stkChkGen words liveness reentry ),
585 ( FSLIT("STK_CHK_NP"), \[e] -> stkChkNodePoints e ),
586 ( FSLIT("TICK_ALLOC_PRIM"), \[hdr,goods,slop] ->
587 tickyAllocPrim hdr goods slop ),
588 ( FSLIT("TICK_ALLOC_PAP"), \[goods,slop] ->
589 tickyAllocPAP goods slop ),
590 ( FSLIT("TICK_ALLOC_UP_THK"), \[goods,slop] ->
591 tickyAllocThunk goods slop ),
592 ( FSLIT("UPD_BH_UPDATABLE"), \[] -> emitBlackHoleCode False ),
593 ( FSLIT("UPD_BH_SINGLE_ENTRY"), \[] -> emitBlackHoleCode True ),
595 ( FSLIT("RET_P"), \[a] -> emitRetUT [(PtrArg,a)]),
596 ( FSLIT("RET_N"), \[a] -> emitRetUT [(NonPtrArg,a)]),
597 ( FSLIT("RET_PP"), \[a,b] -> emitRetUT [(PtrArg,a),(PtrArg,b)]),
598 ( FSLIT("RET_NN"), \[a,b] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b)]),
599 ( FSLIT("RET_NP"), \[a,b] -> emitRetUT [(NonPtrArg,a),(PtrArg,b)]),
600 ( FSLIT("RET_PPP"), \[a,b,c] -> emitRetUT [(PtrArg,a),(PtrArg,b),(PtrArg,c)]),
601 ( FSLIT("RET_NNP"), \[a,b,c] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(PtrArg,c)]),
602 ( FSLIT("RET_NNNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(NonPtrArg,c),(PtrArg,d)]),
603 ( FSLIT("RET_NPNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(PtrArg,b),(NonPtrArg,c),(PtrArg,d)])
607 -- -----------------------------------------------------------------------------
608 -- Our extended FCode monad.
610 -- We add a mapping from names to CmmExpr, to support local variable names in
611 -- the concrete C-- code. The unique supply of the underlying FCode monad
612 -- is used to grab a new unique for each local variable.
614 -- In C--, a local variable can be declared anywhere within a proc,
615 -- and it scopes from the beginning of the proc to the end. Hence, we have
616 -- to collect declarations as we parse the proc, and feed the environment
617 -- back in circularly (to avoid a two-pass algorithm).
619 type Decls = [(FastString,CmmExpr)]
620 type Env = UniqFM CmmExpr
622 newtype ExtFCode a = EC { unEC :: Env -> Decls -> FCode (Decls, a) }
624 type ExtCode = ExtFCode ()
626 returnExtFC a = EC $ \e s -> return (s, a)
627 thenExtFC (EC m) k = EC $ \e s -> do (s',r) <- m e s; unEC (k r) e s'
629 instance Monad ExtFCode where
633 -- This function takes the variable decarations and imports and makes
634 -- an environment, which is looped back into the computation. In this
635 -- way, we can have embedded declarations that scope over the whole
636 -- procedure, and imports that scope over the entire module.
637 loopDecls :: ExtFCode a -> ExtFCode a
638 loopDecls (EC fcode) =
639 EC $ \e s -> fixC (\ ~(decls,a) -> fcode (addListToUFM e decls) [])
641 getEnv :: ExtFCode Env
642 getEnv = EC $ \e s -> return (s, e)
644 addVarDecl :: FastString -> CmmExpr -> ExtCode
645 addVarDecl var expr = EC $ \e s -> return ((var,expr):s, ())
647 newLocal :: MachRep -> FastString -> ExtCode
648 newLocal ty name = do
650 addVarDecl name (CmmReg (CmmLocal (LocalReg u ty)))
652 -- Unknown names are treated as if they had been 'import'ed.
653 -- This saves us a lot of bother in the RTS sources, at the expense of
654 -- deferring some errors to link time.
655 lookupName :: FastString -> ExtFCode CmmExpr
659 case lookupUFM env name of
660 Nothing -> CmmLit (CmmLabel (mkRtsCodeLabelFS name))
663 -- Lifting FCode computations into the ExtFCode monad:
664 code :: FCode a -> ExtFCode a
665 code fc = EC $ \e s -> do r <- fc; return (s, r)
667 code2 :: (FCode (Decls,b) -> FCode ((Decls,b),c))
668 -> ExtFCode b -> ExtFCode c
669 code2 f (EC ec) = EC $ \e s -> do ((s',b),c) <- f (ec e s); return (s',c)
672 stmtEC stmt = code (stmtC stmt)
673 stmtsEC stmts = code (stmtsC stmts)
674 getCgStmtsEC = code2 getCgStmts'
676 forkLabelledCodeEC ec = do
677 stmts <- getCgStmtsEC ec
678 code (forkCgStmts stmts)
680 retInfo name size live_bits cl_type vector = do
681 let liveness = smallLiveness (fromIntegral size) (fromIntegral live_bits)
682 info_lbl = mkRtsRetInfoLabelFS name
683 (info1,info2) = mkRetInfoTable info_lbl liveness NoC_SRT
684 (fromIntegral cl_type) vector
685 return (info_lbl, info1, info2)
687 stdInfo name ptrs nptrs srt_bitmap cl_type desc_str ty_str =
688 basicInfo name (packHalfWordsCLit ptrs nptrs)
689 srt_bitmap cl_type desc_str ty_str
691 basicInfo name layout srt_bitmap cl_type desc_str ty_str = do
692 lit1 <- if opt_SccProfilingOn
693 then code $ mkStringCLit desc_str
694 else return (mkIntCLit 0)
695 lit2 <- if opt_SccProfilingOn
696 then code $ mkStringCLit ty_str
697 else return (mkIntCLit 0)
698 let info1 = mkStdInfoTable lit1 lit2 (fromIntegral cl_type)
699 (fromIntegral srt_bitmap)
701 return (mkRtsInfoLabelFS name, info1, [])
703 funInfo name ptrs nptrs cl_type desc_str ty_str fun_type = do
704 (label,info1,_) <- stdInfo name ptrs nptrs 0{-srt_bitmap-}
705 cl_type desc_str ty_str
706 let info2 = mkFunGenInfoExtraBits (fromIntegral fun_type) 0 zero zero zero
707 -- we leave most of the fields zero here. This is only used
708 -- to generate the BCO info table in the RTS at the moment.
709 return (label,info1,info2)
714 staticClosure :: FastString -> FastString -> [CmmLit] -> ExtCode
715 staticClosure cl_label info payload
716 = code $ emitDataLits (mkRtsDataLabelFS cl_label) lits
717 where lits = mkStaticClosure (mkRtsInfoLabelFS info) dontCareCCS payload [] [] []
721 -> [ExtFCode (CmmReg,MachHint)]
723 -> [ExtFCode (CmmExpr,MachHint)]
724 -> Maybe [GlobalReg] -> P ExtCode
725 foreignCall "C" results_code expr_code args_code vols
727 results <- sequence results_code
729 args <- sequence args_code
730 stmtEC (CmmCall (CmmForeignCall expr CCallConv) results args vols)
731 foreignCall conv _ _ _ _
732 = fail ("unknown calling convention: " ++ conv)
734 doStore :: MachRep -> ExtFCode CmmExpr -> ExtFCode CmmExpr -> ExtCode
735 doStore rep addr_code val_code
736 = do addr <- addr_code
738 -- if the specified store type does not match the type of the expr
739 -- on the rhs, then we insert a coercion that will cause the type
740 -- mismatch to be flagged by cmm-lint. If we don't do this, then
741 -- the store will happen at the wrong type, and the error will not
744 | cmmExprRep val /= rep = CmmMachOp (MO_U_Conv rep rep) [val]
746 stmtEC (CmmStore addr coerce_val)
748 -- Return an unboxed tuple.
749 emitRetUT :: [(CgRep,CmmExpr)] -> Code
751 tickyUnboxedTupleReturn (length args) -- TICK
752 (sp, stmts) <- pushUnboxedTuple 0 args
754 when (sp /= 0) $ stmtC (CmmAssign spReg (cmmRegOffW spReg (-sp)))
755 stmtC (CmmJump (entryCode (CmmLoad (cmmRegOffW spReg sp) wordRep)) [])
757 -- -----------------------------------------------------------------------------
758 -- If-then-else and boolean expressions
761 = BoolExpr `BoolAnd` BoolExpr
762 | BoolExpr `BoolOr` BoolExpr
766 -- ToDo: smart constructors which simplify the boolean expression.
768 ifThenElse cond then_part else_part = do
769 then_id <- code newLabelC
770 join_id <- code newLabelC
774 stmtEC (CmmBranch join_id)
775 code (labelC then_id)
777 -- fall through to join
778 code (labelC join_id)
780 -- 'emitCond cond true_id' emits code to test whether the cond is true,
781 -- branching to true_id if so, and falling through otherwise.
782 emitCond (BoolTest e) then_id = do
783 stmtEC (CmmCondBranch e then_id)
784 emitCond (BoolNot (BoolTest (CmmMachOp op args))) then_id
785 | Just op' <- maybeInvertComparison op
786 = emitCond (BoolTest (CmmMachOp op' args)) then_id
787 emitCond (BoolNot e) then_id = do
788 else_id <- code newLabelC
790 stmtEC (CmmBranch then_id)
791 code (labelC else_id)
792 emitCond (e1 `BoolOr` e2) then_id = do
795 emitCond (e1 `BoolAnd` e2) then_id = do
796 -- we'd like to invert one of the conditionals here to avoid an
797 -- extra branch instruction, but we can't use maybeInvertComparison
798 -- here because we can't look too closely at the expression since
800 and_id <- code newLabelC
801 else_id <- code newLabelC
803 stmtEC (CmmBranch else_id)
806 code (labelC else_id)
809 -- -----------------------------------------------------------------------------
812 -- We use a simplified form of C-- switch statements for now. A
813 -- switch statement always compiles to a table jump. Each arm can
814 -- specify a list of values (not ranges), and there can be a single
815 -- default branch. The range of the table is given either by the
816 -- optional range on the switch (eg. switch [0..7] {...}), or by
817 -- the minimum/maximum values from the branches.
819 doSwitch :: Maybe (Int,Int) -> ExtFCode CmmExpr -> [([Int],ExtCode)]
820 -> Maybe ExtCode -> ExtCode
821 doSwitch mb_range scrut arms deflt
823 -- Compile code for the default branch
826 Nothing -> return Nothing
827 Just e -> do b <- forkLabelledCodeEC e; return (Just b)
829 -- Compile each case branch
830 table_entries <- mapM emitArm arms
832 -- Construct the table
834 all_entries = concat table_entries
835 ixs = map fst all_entries
837 | Just (l,u) <- mb_range = (l,u)
838 | otherwise = (minimum ixs, maximum ixs)
840 entries = elems (accumArray (\_ a -> Just a) dflt_entry (min,max)
843 -- ToDo: check for out of range and jump to default if necessary
844 stmtEC (CmmSwitch expr entries)
846 emitArm :: ([Int],ExtCode) -> ExtFCode [(Int,BlockId)]
847 emitArm (ints,code) = do
848 blockid <- forkLabelledCodeEC code
849 return [ (i,blockid) | i <- ints ]
852 -- -----------------------------------------------------------------------------
853 -- Putting it all together
855 -- The initial environment: we define some constants that the compiler
858 initEnv = listToUFM [
859 ( FSLIT("SIZEOF_StgHeader"),
860 CmmLit (CmmInt (fromIntegral (fixedHdrSize * wORD_SIZE)) wordRep) ),
861 ( FSLIT("SIZEOF_StgInfoTable"),
862 CmmLit (CmmInt (fromIntegral stdInfoTableSizeB) wordRep) )
865 parseCmmFile :: DynFlags -> HomeModules -> FilePath -> IO (Maybe Cmm)
866 parseCmmFile dflags hmods filename = do
867 showPass dflags "ParseCmm"
868 buf <- hGetStringBuffer filename
870 init_loc = mkSrcLoc (mkFastString filename) 1 0
871 init_state = (mkPState buf init_loc dflags) { lex_state = [0] }
872 -- reset the lex_state: the Lexer monad leaves some stuff
873 -- in there we don't want.
874 case unP cmmParse init_state of
875 PFailed span err -> do printError span err; return Nothing
877 cmm <- initC dflags hmods no_module (getCmm (unEC code initEnv [] >> return ()))
878 dumpIfSet_dyn dflags Opt_D_dump_cmm "Cmm" (pprCmms [cmm])
881 no_module = panic "parseCmmFile: no module"