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(..), Safety(..) )
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 )
51 import Data.Char ( ord )
53 #include "HsVersions.h"
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 ',') }
79 '!' { L _ (CmmT_SpecChar '!') }
81 '..' { L _ (CmmT_DotDot) }
82 '::' { L _ (CmmT_DoubleColon) }
83 '>>' { L _ (CmmT_Shr) }
84 '<<' { L _ (CmmT_Shl) }
85 '>=' { L _ (CmmT_Ge) }
86 '<=' { L _ (CmmT_Le) }
87 '==' { L _ (CmmT_Eq) }
88 '!=' { L _ (CmmT_Ne) }
89 '&&' { L _ (CmmT_BoolAnd) }
90 '||' { L _ (CmmT_BoolOr) }
92 'CLOSURE' { L _ (CmmT_CLOSURE) }
93 'INFO_TABLE' { L _ (CmmT_INFO_TABLE) }
94 'INFO_TABLE_RET'{ L _ (CmmT_INFO_TABLE_RET) }
95 'INFO_TABLE_FUN'{ L _ (CmmT_INFO_TABLE_FUN) }
96 'INFO_TABLE_CONSTR'{ L _ (CmmT_INFO_TABLE_CONSTR) }
97 'INFO_TABLE_SELECTOR'{ L _ (CmmT_INFO_TABLE_SELECTOR) }
98 'else' { L _ (CmmT_else) }
99 'export' { L _ (CmmT_export) }
100 'section' { L _ (CmmT_section) }
101 'align' { L _ (CmmT_align) }
102 'goto' { L _ (CmmT_goto) }
103 'if' { L _ (CmmT_if) }
104 'jump' { L _ (CmmT_jump) }
105 'foreign' { L _ (CmmT_foreign) }
106 'import' { L _ (CmmT_import) }
107 'switch' { L _ (CmmT_switch) }
108 'case' { L _ (CmmT_case) }
109 'default' { L _ (CmmT_default) }
110 'bits8' { L _ (CmmT_bits8) }
111 'bits16' { L _ (CmmT_bits16) }
112 'bits32' { L _ (CmmT_bits32) }
113 'bits64' { L _ (CmmT_bits64) }
114 'float32' { L _ (CmmT_float32) }
115 'float64' { L _ (CmmT_float64) }
117 GLOBALREG { L _ (CmmT_GlobalReg $$) }
118 NAME { L _ (CmmT_Name $$) }
119 STRING { L _ (CmmT_String $$) }
120 INT { L _ (CmmT_Int $$) }
121 FLOAT { L _ (CmmT_Float $$) }
123 %monad { P } { >>= } { return }
124 %lexer { cmmlex } { L _ CmmT_EOF }
126 %tokentype { Located CmmToken }
128 -- C-- operator precedences, taken from the C-- spec
129 %right '||' -- non-std extension, called %disjoin in C--
130 %right '&&' -- non-std extension, called %conjoin in C--
132 %nonassoc '>=' '>' '<=' '<' '!=' '=='
144 : {- empty -} { return () }
145 | cmmtop cmm { do $1; $2 }
147 cmmtop :: { ExtCode }
151 | 'CLOSURE' '(' NAME ',' NAME lits ')' ';'
152 { do lits <- sequence $6;
153 staticClosure $3 $5 (map getLit lits) }
155 -- The only static closures in the RTS are dummy closures like
156 -- stg_END_TSO_QUEUE_closure and stg_dummy_ret. We don't need
157 -- to provide the full generality of static closures here.
159 -- * CCS can always be CCS_DONT_CARE
160 -- * closure is always extern
161 -- * payload is always empty
162 -- * we can derive closure and info table labels from a single NAME
164 cmmdata :: { ExtCode }
165 : 'section' STRING '{' statics '}'
166 { do ss <- sequence $4;
167 code (emitData (section $2) (concat ss)) }
169 statics :: { [ExtFCode [CmmStatic]] }
171 | static statics { $1 : $2 }
173 -- Strings aren't used much in the RTS HC code, so it doesn't seem
174 -- worth allowing inline strings. C-- doesn't allow them anyway.
175 static :: { ExtFCode [CmmStatic] }
176 : NAME ':' { return [CmmDataLabel (mkRtsDataLabelFS $1)] }
177 | type expr ';' { do e <- $2;
178 return [CmmStaticLit (getLit e)] }
179 | type ';' { return [CmmUninitialised
180 (machRepByteWidth $1)] }
181 | 'bits8' '[' ']' STRING ';' { return [mkString $4] }
182 | 'bits8' '[' INT ']' ';' { return [CmmUninitialised
184 | typenot8 '[' INT ']' ';' { return [CmmUninitialised
185 (machRepByteWidth $1 *
187 | 'align' INT ';' { return [CmmAlign (fromIntegral $2)] }
188 | 'CLOSURE' '(' NAME lits ')'
189 { do lits <- sequence $4;
190 return $ map CmmStaticLit $
191 mkStaticClosure (mkRtsInfoLabelFS $3)
192 dontCareCCS (map getLit lits) [] [] [] }
193 -- arrays of closures required for the CHARLIKE & INTLIKE arrays
195 lits :: { [ExtFCode CmmExpr] }
197 | ',' expr lits { $2 : $3 }
199 cmmproc :: { ExtCode }
201 { do (info_lbl, info1, info2) <- $1;
202 stmts <- getCgStmtsEC (loopDecls $3)
203 blks <- code (cgStmtsToBlocks stmts)
204 code (emitInfoTableAndCode info_lbl info1 info2 [] blks) }
207 { do (info_lbl, info1, info2) <- $1;
208 code (emitInfoTableAndCode info_lbl info1 info2 [] []) }
211 { do stmts <- getCgStmtsEC (loopDecls $3);
212 blks <- code (cgStmtsToBlocks stmts)
213 code (emitProc [] (mkRtsCodeLabelFS $1) [] blks) }
215 info :: { ExtFCode (CLabel, [CmmLit],[CmmLit]) }
216 : 'INFO_TABLE' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
217 -- ptrs, nptrs, closure type, description, type
218 { stdInfo $3 $5 $7 0 $9 $11 $13 }
220 | 'INFO_TABLE_FUN' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ',' INT ')'
221 -- ptrs, nptrs, closure type, description, type, fun type
222 { funInfo $3 $5 $7 $9 $11 $13 $15 }
224 | 'INFO_TABLE_CONSTR' '(' NAME ',' INT ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
225 -- ptrs, nptrs, tag, closure type, description, type
226 { stdInfo $3 $5 $7 $9 $11 $13 $15 }
228 | 'INFO_TABLE_SELECTOR' '(' NAME ',' INT ',' INT ',' STRING ',' STRING ')'
229 -- selector, closure type, description, type
230 { basicInfo $3 (mkIntCLit (fromIntegral $5)) 0 $7 $9 $11 }
232 | 'INFO_TABLE_RET' '(' NAME ',' INT ',' INT ',' INT maybe_vec ')'
233 { retInfo $3 $5 $7 $9 $10 }
235 maybe_vec :: { [CmmLit] }
237 | ',' NAME maybe_vec { CmmLabel (mkRtsCodeLabelFS $2) : $3 }
240 : {- empty -} { return () }
241 | decl body { do $1; $2 }
242 | stmt body { do $1; $2 }
245 : type names ';' { mapM_ (newLocal $1) $2 }
246 | 'import' names ';' { return () } -- ignore imports
247 | 'export' names ';' { return () } -- ignore exports
249 names :: { [FastString] }
251 | NAME ',' names { $1 : $3 }
257 { do l <- newLabel $1; code (labelC l) }
260 { do reg <- $1; e <- $3; stmtEC (CmmAssign reg e) }
261 | type '[' expr ']' '=' expr ';'
263 | 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
264 {% foreignCall $2 [] $3 $5 $7 }
265 | lreg '=' 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
266 {% let result = do r <- $1; return (r,NoHint) in
267 foreignCall $4 [result] $5 $7 $9 }
268 | STRING lreg '=' 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
269 {% do h <- parseHint $1;
270 let result = do r <- $2; return (r,h) in
271 foreignCall $5 [result] $6 $8 $10 }
272 -- stmt-level macros, stealing syntax from ordinary C-- function calls.
273 -- Perhaps we ought to use the %%-form?
274 | NAME '(' exprs0 ')' ';'
276 | 'switch' maybe_range expr '{' arms default '}'
277 { doSwitch $2 $3 $5 $6 }
279 { do l <- lookupLabel $2; stmtEC (CmmBranch l) }
280 | 'jump' expr {-maybe_actuals-} ';'
281 { do e <- $2; stmtEC (CmmJump e []) }
282 | 'if' bool_expr '{' body '}' else
283 { ifThenElse $2 $4 $6 }
285 bool_expr :: { ExtFCode BoolExpr }
287 | expr { do e <- $1; return (BoolTest e) }
289 bool_op :: { ExtFCode BoolExpr }
290 : bool_expr '&&' bool_expr { do e1 <- $1; e2 <- $3;
291 return (BoolAnd e1 e2) }
292 | bool_expr '||' bool_expr { do e1 <- $1; e2 <- $3;
293 return (BoolOr e1 e2) }
294 | '!' bool_expr { do e <- $2; return (BoolNot e) }
295 | '(' bool_op ')' { $2 }
297 -- This is not C-- syntax. What to do?
298 vols :: { Maybe [GlobalReg] }
299 : {- empty -} { Nothing }
300 | '[' ']' { Just [] }
301 | '[' globals ']' { Just $2 }
303 globals :: { [GlobalReg] }
305 | GLOBALREG ',' globals { $1 : $3 }
307 maybe_range :: { Maybe (Int,Int) }
308 : '[' INT '..' INT ']' { Just (fromIntegral $2, fromIntegral $4) }
309 | {- empty -} { Nothing }
311 arms :: { [([Int],ExtCode)] }
313 | arm arms { $1 : $2 }
315 arm :: { ([Int],ExtCode) }
316 : 'case' ints ':' '{' body '}' { ($2, $5) }
319 : INT { [ fromIntegral $1 ] }
320 | INT ',' ints { fromIntegral $1 : $3 }
322 default :: { Maybe ExtCode }
323 : 'default' ':' '{' body '}' { Just $4 }
324 -- taking a few liberties with the C-- syntax here; C-- doesn't have
325 -- 'default' branches
326 | {- empty -} { Nothing }
329 : {- empty -} { nopEC }
330 | 'else' '{' body '}' { $3 }
332 -- we have to write this out longhand so that Happy's precedence rules
334 expr :: { ExtFCode CmmExpr }
335 : expr '/' expr { mkMachOp MO_U_Quot [$1,$3] }
336 | expr '*' expr { mkMachOp MO_Mul [$1,$3] }
337 | expr '%' expr { mkMachOp MO_U_Rem [$1,$3] }
338 | expr '-' expr { mkMachOp MO_Sub [$1,$3] }
339 | expr '+' expr { mkMachOp MO_Add [$1,$3] }
340 | expr '>>' expr { mkMachOp MO_U_Shr [$1,$3] }
341 | expr '<<' expr { mkMachOp MO_Shl [$1,$3] }
342 | expr '&' expr { mkMachOp MO_And [$1,$3] }
343 | expr '^' expr { mkMachOp MO_Xor [$1,$3] }
344 | expr '|' expr { mkMachOp MO_Or [$1,$3] }
345 | expr '>=' expr { mkMachOp MO_U_Ge [$1,$3] }
346 | expr '>' expr { mkMachOp MO_U_Gt [$1,$3] }
347 | expr '<=' expr { mkMachOp MO_U_Le [$1,$3] }
348 | expr '<' expr { mkMachOp MO_U_Lt [$1,$3] }
349 | expr '!=' expr { mkMachOp MO_Ne [$1,$3] }
350 | expr '==' expr { mkMachOp MO_Eq [$1,$3] }
351 | '~' expr { mkMachOp MO_Not [$2] }
352 | '-' expr { mkMachOp MO_S_Neg [$2] }
353 | expr0 '`' NAME '`' expr0 {% do { mo <- nameToMachOp $3 ;
354 return (mkMachOp mo [$1,$5]) } }
357 expr0 :: { ExtFCode CmmExpr }
358 : INT maybe_ty { return (CmmLit (CmmInt $1 $2)) }
359 | FLOAT maybe_ty { return (CmmLit (CmmFloat $1 $2)) }
360 | STRING { do s <- code (mkStringCLit $1);
363 | type '[' expr ']' { do e <- $3; return (CmmLoad e $1) }
364 | '%' NAME '(' exprs0 ')' {% exprOp $2 $4 }
365 | '(' expr ')' { $2 }
368 -- leaving out the type of a literal gives you the native word size in C--
369 maybe_ty :: { MachRep }
370 : {- empty -} { wordRep }
373 hint_exprs0 :: { [ExtFCode (CmmExpr, MachHint)] }
377 hint_exprs :: { [ExtFCode (CmmExpr, MachHint)] }
379 | hint_expr ',' hint_exprs { $1 : $3 }
381 hint_expr :: { ExtFCode (CmmExpr, MachHint) }
382 : expr { do e <- $1; return (e, inferHint e) }
383 | expr STRING {% do h <- parseHint $2;
385 e <- $1; return (e,h) }
387 exprs0 :: { [ExtFCode CmmExpr] }
391 exprs :: { [ExtFCode CmmExpr] }
393 | expr ',' exprs { $1 : $3 }
395 reg :: { ExtFCode CmmExpr }
396 : NAME { lookupName $1 }
397 | GLOBALREG { return (CmmReg (CmmGlobal $1)) }
399 lreg :: { ExtFCode CmmReg }
400 : NAME { do e <- lookupName $1;
404 other -> pprPanic "CmmParse:" (ftext $1 <> text " not a register") }
405 | GLOBALREG { return (CmmGlobal $1) }
411 typenot8 :: { MachRep }
418 section :: String -> Section
419 section "text" = Text
420 section "data" = Data
421 section "rodata" = ReadOnlyData
422 section "bss" = UninitialisedData
423 section s = OtherSection s
425 mkString :: String -> CmmStatic
426 mkString s = CmmString (map (fromIntegral.ord) s)
428 -- mkMachOp infers the type of the MachOp from the type of its first
429 -- argument. We assume that this is correct: for MachOps that don't have
430 -- symmetrical args (e.g. shift ops), the first arg determines the type of
432 mkMachOp :: (MachRep -> MachOp) -> [ExtFCode CmmExpr] -> ExtFCode CmmExpr
433 mkMachOp fn args = do
434 arg_exprs <- sequence args
435 return (CmmMachOp (fn (cmmExprRep (head arg_exprs))) arg_exprs)
437 getLit :: CmmExpr -> CmmLit
438 getLit (CmmLit l) = l
439 getLit (CmmMachOp (MO_S_Neg _) [CmmLit (CmmInt i r)]) = CmmInt (negate i) r
440 getLit _ = panic "invalid literal" -- TODO messy failure
442 nameToMachOp :: FastString -> P (MachRep -> MachOp)
444 case lookupUFM machOps name of
445 Nothing -> fail ("unknown primitive " ++ unpackFS name)
448 exprOp :: FastString -> [ExtFCode CmmExpr] -> P (ExtFCode CmmExpr)
449 exprOp name args_code =
450 case lookupUFM exprMacros name of
451 Just f -> return $ do
452 args <- sequence args_code
455 mo <- nameToMachOp name
456 return $ mkMachOp mo args_code
458 exprMacros :: UniqFM ([CmmExpr] -> CmmExpr)
459 exprMacros = listToUFM [
460 ( FSLIT("ENTRY_CODE"), \ [x] -> entryCode x ),
461 ( FSLIT("INFO_PTR"), \ [x] -> closureInfoPtr x ),
462 ( FSLIT("STD_INFO"), \ [x] -> infoTable x ),
463 ( FSLIT("FUN_INFO"), \ [x] -> funInfoTable x ),
464 ( FSLIT("GET_ENTRY"), \ [x] -> entryCode (closureInfoPtr x) ),
465 ( FSLIT("GET_STD_INFO"), \ [x] -> infoTable (closureInfoPtr x) ),
466 ( FSLIT("GET_FUN_INFO"), \ [x] -> funInfoTable (closureInfoPtr x) ),
467 ( FSLIT("INFO_TYPE"), \ [x] -> infoTableClosureType x ),
468 ( FSLIT("INFO_PTRS"), \ [x] -> infoTablePtrs x ),
469 ( FSLIT("INFO_NPTRS"), \ [x] -> infoTableNonPtrs x ),
470 ( FSLIT("RET_VEC"), \ [info, conZ] -> retVec info conZ )
473 -- we understand a subset of C-- primitives:
474 machOps = listToUFM $
475 map (\(x, y) -> (mkFastString x, y)) [
482 ( "quot", MO_S_Quot ),
484 ( "divu", MO_U_Quot ),
485 ( "modu", MO_U_Rem ),
503 ( "fneg", MO_S_Neg ),
510 ( "shrl", MO_U_Shr ),
511 ( "shra", MO_S_Shr ),
513 ( "lobits8", flip MO_U_Conv I8 ),
514 ( "lobits16", flip MO_U_Conv I16 ),
515 ( "lobits32", flip MO_U_Conv I32 ),
516 ( "lobits64", flip MO_U_Conv I64 ),
517 ( "sx16", flip MO_S_Conv I16 ),
518 ( "sx32", flip MO_S_Conv I32 ),
519 ( "sx64", flip MO_S_Conv I64 ),
520 ( "zx16", flip MO_U_Conv I16 ),
521 ( "zx32", flip MO_U_Conv I32 ),
522 ( "zx64", flip MO_U_Conv I64 ),
523 ( "f2f32", flip MO_S_Conv F32 ), -- TODO; rounding mode
524 ( "f2f64", flip MO_S_Conv F64 ), -- TODO; rounding mode
525 ( "f2i8", flip MO_S_Conv I8 ),
526 ( "f2i16", flip MO_S_Conv I8 ),
527 ( "f2i32", flip MO_S_Conv I8 ),
528 ( "f2i64", flip MO_S_Conv I8 ),
529 ( "i2f32", flip MO_S_Conv F32 ),
530 ( "i2f64", flip MO_S_Conv F64 )
533 parseHint :: String -> P MachHint
534 parseHint "ptr" = return PtrHint
535 parseHint "signed" = return SignedHint
536 parseHint "float" = return FloatHint
537 parseHint str = fail ("unrecognised hint: " ++ str)
539 -- labels are always pointers, so we might as well infer the hint
540 inferHint :: CmmExpr -> MachHint
541 inferHint (CmmLit (CmmLabel _)) = PtrHint
542 inferHint (CmmReg (CmmGlobal g)) | isPtrGlobalReg g = PtrHint
545 isPtrGlobalReg Sp = True
546 isPtrGlobalReg SpLim = True
547 isPtrGlobalReg Hp = True
548 isPtrGlobalReg HpLim = True
549 isPtrGlobalReg CurrentTSO = True
550 isPtrGlobalReg CurrentNursery = True
551 isPtrGlobalReg _ = False
554 happyError = srcParseFail
556 -- -----------------------------------------------------------------------------
557 -- Statement-level macros
559 stmtMacro :: FastString -> [ExtFCode CmmExpr] -> P ExtCode
560 stmtMacro fun args_code = do
561 case lookupUFM stmtMacros fun of
562 Nothing -> fail ("unknown macro: " ++ unpackFS fun)
563 Just fcode -> return $ do
564 args <- sequence args_code
567 stmtMacros :: UniqFM ([CmmExpr] -> Code)
568 stmtMacros = listToUFM [
569 ( FSLIT("CCS_ALLOC"), \[words,ccs] -> profAlloc words ccs ),
570 ( FSLIT("CLOSE_NURSERY"), \[] -> emitCloseNursery ),
571 ( FSLIT("ENTER_CCS_PAP_CL"), \[e] -> enterCostCentrePAP e ),
572 ( FSLIT("ENTER_CCS_THUNK"), \[e] -> enterCostCentreThunk e ),
573 ( FSLIT("HP_CHK_GEN"), \[words,liveness,reentry] ->
574 hpChkGen words liveness reentry ),
575 ( FSLIT("HP_CHK_NP_ASSIGN_SP0"), \[e,f] -> hpChkNodePointsAssignSp0 e f ),
576 ( FSLIT("LOAD_THREAD_STATE"), \[] -> emitLoadThreadState ),
577 ( FSLIT("LDV_ENTER"), \[e] -> ldvEnter e ),
578 ( FSLIT("LDV_RECORD_CREATE"), \[e] -> ldvRecordCreate e ),
579 ( FSLIT("OPEN_NURSERY"), \[] -> emitOpenNursery ),
580 ( FSLIT("PUSH_UPD_FRAME"), \[sp,e] -> emitPushUpdateFrame sp e ),
581 ( FSLIT("SAVE_THREAD_STATE"), \[] -> emitSaveThreadState ),
582 ( FSLIT("SET_HDR"), \[ptr,info,ccs] ->
583 emitSetDynHdr ptr info ccs ),
584 ( FSLIT("STK_CHK_GEN"), \[words,liveness,reentry] ->
585 stkChkGen words liveness reentry ),
586 ( FSLIT("STK_CHK_NP"), \[e] -> stkChkNodePoints e ),
587 ( FSLIT("TICK_ALLOC_PRIM"), \[hdr,goods,slop] ->
588 tickyAllocPrim hdr goods slop ),
589 ( FSLIT("TICK_ALLOC_PAP"), \[goods,slop] ->
590 tickyAllocPAP goods slop ),
591 ( FSLIT("TICK_ALLOC_UP_THK"), \[goods,slop] ->
592 tickyAllocThunk goods slop ),
593 ( FSLIT("UPD_BH_UPDATABLE"), \[] -> emitBlackHoleCode False ),
594 ( FSLIT("UPD_BH_SINGLE_ENTRY"), \[] -> emitBlackHoleCode True ),
596 ( FSLIT("RET_P"), \[a] -> emitRetUT [(PtrArg,a)]),
597 ( FSLIT("RET_N"), \[a] -> emitRetUT [(NonPtrArg,a)]),
598 ( FSLIT("RET_PP"), \[a,b] -> emitRetUT [(PtrArg,a),(PtrArg,b)]),
599 ( FSLIT("RET_NN"), \[a,b] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b)]),
600 ( FSLIT("RET_NP"), \[a,b] -> emitRetUT [(NonPtrArg,a),(PtrArg,b)]),
601 ( FSLIT("RET_PPP"), \[a,b,c] -> emitRetUT [(PtrArg,a),(PtrArg,b),(PtrArg,c)]),
602 ( FSLIT("RET_NNP"), \[a,b,c] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(PtrArg,c)]),
603 ( FSLIT("RET_NNNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(NonPtrArg,c),(PtrArg,d)]),
604 ( FSLIT("RET_NPNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(PtrArg,b),(NonPtrArg,c),(PtrArg,d)])
608 -- -----------------------------------------------------------------------------
609 -- Our extended FCode monad.
611 -- We add a mapping from names to CmmExpr, to support local variable names in
612 -- the concrete C-- code. The unique supply of the underlying FCode monad
613 -- is used to grab a new unique for each local variable.
615 -- In C--, a local variable can be declared anywhere within a proc,
616 -- and it scopes from the beginning of the proc to the end. Hence, we have
617 -- to collect declarations as we parse the proc, and feed the environment
618 -- back in circularly (to avoid a two-pass algorithm).
620 data Named = Var CmmExpr | Label BlockId
621 type Decls = [(FastString,Named)]
622 type Env = UniqFM Named
624 newtype ExtFCode a = EC { unEC :: Env -> Decls -> FCode (Decls, a) }
626 type ExtCode = ExtFCode ()
628 returnExtFC a = EC $ \e s -> return (s, a)
629 thenExtFC (EC m) k = EC $ \e s -> do (s',r) <- m e s; unEC (k r) e s'
631 instance Monad ExtFCode where
635 -- This function takes the variable decarations and imports and makes
636 -- an environment, which is looped back into the computation. In this
637 -- way, we can have embedded declarations that scope over the whole
638 -- procedure, and imports that scope over the entire module.
639 loopDecls :: ExtFCode a -> ExtFCode a
640 loopDecls (EC fcode) =
641 EC $ \e s -> fixC (\ ~(decls,a) -> fcode (addListToUFM e decls) [])
643 getEnv :: ExtFCode Env
644 getEnv = EC $ \e s -> return (s, e)
646 addVarDecl :: FastString -> CmmExpr -> ExtCode
647 addVarDecl var expr = EC $ \e s -> return ((var, Var expr):s, ())
649 addLabel :: FastString -> BlockId -> ExtCode
650 addLabel name block_id = EC $ \e s -> return ((name, Label block_id):s, ())
652 newLocal :: MachRep -> FastString -> ExtCode
653 newLocal ty name = do
655 addVarDecl name (CmmReg (CmmLocal (LocalReg u ty)))
657 newLabel :: FastString -> ExtFCode BlockId
660 addLabel name (BlockId u)
663 lookupLabel :: FastString -> ExtFCode BlockId
664 lookupLabel name = do
667 case lookupUFM env name of
669 _other -> BlockId (newTagUnique (getUnique name) 'L')
671 -- Unknown names are treated as if they had been 'import'ed.
672 -- This saves us a lot of bother in the RTS sources, at the expense of
673 -- deferring some errors to link time.
674 lookupName :: FastString -> ExtFCode CmmExpr
678 case lookupUFM env name of
680 _other -> CmmLit (CmmLabel (mkRtsCodeLabelFS name))
682 -- Lifting FCode computations into the ExtFCode monad:
683 code :: FCode a -> ExtFCode a
684 code fc = EC $ \e s -> do r <- fc; return (s, r)
686 code2 :: (FCode (Decls,b) -> FCode ((Decls,b),c))
687 -> ExtFCode b -> ExtFCode c
688 code2 f (EC ec) = EC $ \e s -> do ((s',b),c) <- f (ec e s); return (s',c)
691 stmtEC stmt = code (stmtC stmt)
692 stmtsEC stmts = code (stmtsC stmts)
693 getCgStmtsEC = code2 getCgStmts'
695 forkLabelledCodeEC ec = do
696 stmts <- getCgStmtsEC ec
697 code (forkCgStmts stmts)
699 retInfo name size live_bits cl_type vector = do
700 let liveness = smallLiveness (fromIntegral size) (fromIntegral live_bits)
701 info_lbl = mkRtsRetInfoLabelFS name
702 (info1,info2) = mkRetInfoTable info_lbl liveness NoC_SRT
703 (fromIntegral cl_type) vector
704 return (info_lbl, info1, info2)
706 stdInfo name ptrs nptrs srt_bitmap cl_type desc_str ty_str =
707 basicInfo name (packHalfWordsCLit ptrs nptrs)
708 srt_bitmap cl_type desc_str ty_str
710 basicInfo name layout srt_bitmap cl_type desc_str ty_str = do
711 lit1 <- if opt_SccProfilingOn
712 then code $ mkStringCLit desc_str
713 else return (mkIntCLit 0)
714 lit2 <- if opt_SccProfilingOn
715 then code $ mkStringCLit ty_str
716 else return (mkIntCLit 0)
717 let info1 = mkStdInfoTable lit1 lit2 (fromIntegral cl_type)
718 (fromIntegral srt_bitmap)
720 return (mkRtsInfoLabelFS name, info1, [])
722 funInfo name ptrs nptrs cl_type desc_str ty_str fun_type = do
723 (label,info1,_) <- stdInfo name ptrs nptrs 0{-srt_bitmap-}
724 cl_type desc_str ty_str
725 let info2 = mkFunGenInfoExtraBits (fromIntegral fun_type) 0 zero zero zero
726 -- we leave most of the fields zero here. This is only used
727 -- to generate the BCO info table in the RTS at the moment.
728 return (label,info1,info2)
733 staticClosure :: FastString -> FastString -> [CmmLit] -> ExtCode
734 staticClosure cl_label info payload
735 = code $ emitDataLits (mkRtsDataLabelFS cl_label) lits
736 where lits = mkStaticClosure (mkRtsInfoLabelFS info) dontCareCCS payload [] [] []
740 -> [ExtFCode (CmmReg,MachHint)]
742 -> [ExtFCode (CmmExpr,MachHint)]
743 -> Maybe [GlobalReg] -> P ExtCode
744 foreignCall "C" results_code expr_code args_code vols
746 results <- sequence results_code
748 args <- sequence args_code
749 code (emitForeignCall' PlayRisky results
750 (CmmForeignCall expr CCallConv) args vols)
751 foreignCall conv _ _ _ _
752 = fail ("unknown calling convention: " ++ conv)
754 doStore :: MachRep -> ExtFCode CmmExpr -> ExtFCode CmmExpr -> ExtCode
755 doStore rep addr_code val_code
756 = do addr <- addr_code
758 -- if the specified store type does not match the type of the expr
759 -- on the rhs, then we insert a coercion that will cause the type
760 -- mismatch to be flagged by cmm-lint. If we don't do this, then
761 -- the store will happen at the wrong type, and the error will not
764 | cmmExprRep val /= rep = CmmMachOp (MO_U_Conv rep rep) [val]
766 stmtEC (CmmStore addr coerce_val)
768 -- Return an unboxed tuple.
769 emitRetUT :: [(CgRep,CmmExpr)] -> Code
771 tickyUnboxedTupleReturn (length args) -- TICK
772 (sp, stmts) <- pushUnboxedTuple 0 args
774 when (sp /= 0) $ stmtC (CmmAssign spReg (cmmRegOffW spReg (-sp)))
775 stmtC (CmmJump (entryCode (CmmLoad (cmmRegOffW spReg sp) wordRep)) [])
777 -- -----------------------------------------------------------------------------
778 -- If-then-else and boolean expressions
781 = BoolExpr `BoolAnd` BoolExpr
782 | BoolExpr `BoolOr` BoolExpr
786 -- ToDo: smart constructors which simplify the boolean expression.
788 ifThenElse cond then_part else_part = do
789 then_id <- code newLabelC
790 join_id <- code newLabelC
794 stmtEC (CmmBranch join_id)
795 code (labelC then_id)
797 -- fall through to join
798 code (labelC join_id)
800 -- 'emitCond cond true_id' emits code to test whether the cond is true,
801 -- branching to true_id if so, and falling through otherwise.
802 emitCond (BoolTest e) then_id = do
803 stmtEC (CmmCondBranch e then_id)
804 emitCond (BoolNot (BoolTest (CmmMachOp op args))) then_id
805 | Just op' <- maybeInvertComparison op
806 = emitCond (BoolTest (CmmMachOp op' args)) then_id
807 emitCond (BoolNot e) then_id = do
808 else_id <- code newLabelC
810 stmtEC (CmmBranch then_id)
811 code (labelC else_id)
812 emitCond (e1 `BoolOr` e2) then_id = do
815 emitCond (e1 `BoolAnd` e2) then_id = do
816 -- we'd like to invert one of the conditionals here to avoid an
817 -- extra branch instruction, but we can't use maybeInvertComparison
818 -- here because we can't look too closely at the expression since
820 and_id <- code newLabelC
821 else_id <- code newLabelC
823 stmtEC (CmmBranch else_id)
826 code (labelC else_id)
829 -- -----------------------------------------------------------------------------
832 -- We use a simplified form of C-- switch statements for now. A
833 -- switch statement always compiles to a table jump. Each arm can
834 -- specify a list of values (not ranges), and there can be a single
835 -- default branch. The range of the table is given either by the
836 -- optional range on the switch (eg. switch [0..7] {...}), or by
837 -- the minimum/maximum values from the branches.
839 doSwitch :: Maybe (Int,Int) -> ExtFCode CmmExpr -> [([Int],ExtCode)]
840 -> Maybe ExtCode -> ExtCode
841 doSwitch mb_range scrut arms deflt
843 -- Compile code for the default branch
846 Nothing -> return Nothing
847 Just e -> do b <- forkLabelledCodeEC e; return (Just b)
849 -- Compile each case branch
850 table_entries <- mapM emitArm arms
852 -- Construct the table
854 all_entries = concat table_entries
855 ixs = map fst all_entries
857 | Just (l,u) <- mb_range = (l,u)
858 | otherwise = (minimum ixs, maximum ixs)
860 entries = elems (accumArray (\_ a -> Just a) dflt_entry (min,max)
863 -- ToDo: check for out of range and jump to default if necessary
864 stmtEC (CmmSwitch expr entries)
866 emitArm :: ([Int],ExtCode) -> ExtFCode [(Int,BlockId)]
867 emitArm (ints,code) = do
868 blockid <- forkLabelledCodeEC code
869 return [ (i,blockid) | i <- ints ]
872 -- -----------------------------------------------------------------------------
873 -- Putting it all together
875 -- The initial environment: we define some constants that the compiler
878 initEnv = listToUFM [
879 ( FSLIT("SIZEOF_StgHeader"),
880 Var (CmmLit (CmmInt (fromIntegral (fixedHdrSize * wORD_SIZE)) wordRep) )),
881 ( FSLIT("SIZEOF_StgInfoTable"),
882 Var (CmmLit (CmmInt (fromIntegral stdInfoTableSizeB) wordRep) ))
885 parseCmmFile :: DynFlags -> HomeModules -> FilePath -> IO (Maybe Cmm)
886 parseCmmFile dflags hmods filename = do
887 showPass dflags "ParseCmm"
888 buf <- hGetStringBuffer filename
890 init_loc = mkSrcLoc (mkFastString filename) 1 0
891 init_state = (mkPState buf init_loc dflags) { lex_state = [0] }
892 -- reset the lex_state: the Lexer monad leaves some stuff
893 -- in there we don't want.
894 case unP cmmParse init_state of
895 PFailed span err -> do printError span err; return Nothing
897 cmm <- initC dflags hmods no_module (getCmm (unEC code initEnv [] >> return ()))
898 dumpIfSet_dyn dflags Opt_D_dump_cmm "Cmm" (pprCmms [cmm])
901 no_module = panic "parseCmmFile: no module"