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, mkLblExpr )
32 import SMRep ( tablesNextToCode, fixedHdrSize, CgRep(..) )
35 import ForeignCall ( CCallConv(..) )
36 import Literal ( mkMachInt )
40 import CmdLineOpts ( DynFlags, DynFlag(..), opt_SccProfilingOn )
41 import ErrUtils ( printError, dumpIfSet_dyn, showPass )
42 import StringBuffer ( hGetStringBuffer )
44 import Panic ( panic )
45 import Constants ( wORD_SIZE )
50 #include "HsVersions.h"
54 ':' { L _ (CmmT_SpecChar ':') }
55 ';' { L _ (CmmT_SpecChar ';') }
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 '!') }
78 '..' { L _ (CmmT_DotDot) }
79 '::' { L _ (CmmT_DoubleColon) }
80 '>>' { L _ (CmmT_Shr) }
81 '<<' { L _ (CmmT_Shl) }
82 '>=' { L _ (CmmT_Ge) }
83 '<=' { L _ (CmmT_Le) }
84 '==' { L _ (CmmT_Eq) }
85 '!=' { L _ (CmmT_Ne) }
86 '&&' { L _ (CmmT_BoolAnd) }
87 '||' { L _ (CmmT_BoolOr) }
89 'CLOSURE' { L _ (CmmT_CLOSURE) }
90 'INFO_TABLE' { L _ (CmmT_INFO_TABLE) }
91 'INFO_TABLE_RET'{ L _ (CmmT_INFO_TABLE_RET) }
92 'INFO_TABLE_FUN'{ L _ (CmmT_INFO_TABLE_FUN) }
93 'INFO_TABLE_CONSTR'{ L _ (CmmT_INFO_TABLE_CONSTR) }
94 'INFO_TABLE_SELECTOR'{ L _ (CmmT_INFO_TABLE_SELECTOR) }
95 'else' { L _ (CmmT_else) }
96 'export' { L _ (CmmT_export) }
97 'section' { L _ (CmmT_section) }
98 'align' { L _ (CmmT_align) }
99 'goto' { L _ (CmmT_goto) }
100 'if' { L _ (CmmT_if) }
101 'jump' { L _ (CmmT_jump) }
102 'foreign' { L _ (CmmT_foreign) }
103 'import' { L _ (CmmT_import) }
104 'switch' { L _ (CmmT_switch) }
105 'case' { L _ (CmmT_case) }
106 'default' { L _ (CmmT_default) }
107 'bits8' { L _ (CmmT_bits8) }
108 'bits16' { L _ (CmmT_bits16) }
109 'bits32' { L _ (CmmT_bits32) }
110 'bits64' { L _ (CmmT_bits64) }
111 'float32' { L _ (CmmT_float32) }
112 'float64' { L _ (CmmT_float64) }
114 GLOBALREG { L _ (CmmT_GlobalReg $$) }
115 NAME { L _ (CmmT_Name $$) }
116 STRING { L _ (CmmT_String $$) }
117 INT { L _ (CmmT_Int $$) }
118 FLOAT { L _ (CmmT_Float $$) }
120 %monad { P } { >>= } { return }
121 %lexer { cmmlex } { L _ CmmT_EOF }
123 %tokentype { Located CmmToken }
125 -- C-- operator precedences, taken from the C-- spec
126 %right '||' -- non-std extension, called %disjoin in C--
127 %right '&&' -- non-std extension, called %conjoin in C--
129 %nonassoc '>=' '>' '<=' '<' '!=' '=='
141 : {- empty -} { return () }
142 | cmmtop cmm { do $1; $2 }
144 cmmtop :: { ExtCode }
148 | 'CLOSURE' '(' NAME ',' NAME lits ')' ';'
149 { do lits <- sequence $6;
150 staticClosure $3 $5 (map getLit lits) }
152 -- The only static closures in the RTS are dummy closures like
153 -- stg_END_TSO_QUEUE_closure and stg_dummy_ret. We don't need
154 -- to provide the full generality of static closures here.
156 -- * CCS can always be CCS_DONT_CARE
157 -- * closure is always extern
158 -- * payload is always empty
159 -- * we can derive closure and info table labels from a single NAME
161 cmmdata :: { ExtCode }
162 : 'section' STRING '{' statics '}'
163 { do ss <- sequence $4;
164 code (emitData (section $2) (concat ss)) }
166 statics :: { [ExtFCode [CmmStatic]] }
168 | static statics { $1 : $2 }
170 -- Strings aren't used much in the RTS HC code, so it doesn't seem
171 -- worth allowing inline strings. C-- doesn't allow them anyway.
172 static :: { ExtFCode [CmmStatic] }
173 : NAME ':' { return [CmmDataLabel (mkRtsDataLabelFS $1)] }
174 | type expr ';' { do e <- $2;
175 return [CmmStaticLit (getLit e)] }
176 | type ';' { return [CmmUninitialised
177 (machRepByteWidth $1)] }
178 | 'bits8' '[' ']' STRING ';' { return [CmmString $4] }
179 | 'bits8' '[' INT ']' ';' { return [CmmUninitialised
181 | typenot8 '[' INT ']' ';' { return [CmmUninitialised
182 (machRepByteWidth $1 *
184 | 'align' INT ';' { return [CmmAlign (fromIntegral $2)] }
185 | 'CLOSURE' '(' NAME lits ')'
186 { do lits <- sequence $4;
187 return $ map CmmStaticLit $
188 mkStaticClosure (mkRtsInfoLabelFS $3)
189 dontCareCCS (map getLit lits) [] [] }
190 -- arrays of closures required for the CHARLIKE & INTLIKE arrays
192 lits :: { [ExtFCode CmmExpr] }
194 | ',' expr lits { $2 : $3 }
196 cmmproc :: { ExtCode }
198 { do (info_lbl, info1, info2) <- $1;
199 stmts <- getCgStmtsEC (loopDecls $3)
200 blks <- code (cgStmtsToBlocks stmts)
201 code (emitInfoTableAndCode info_lbl info1 info2 [] blks) }
204 { do (info_lbl, info1, info2) <- $1;
205 code (emitInfoTableAndCode info_lbl info1 info2 [] []) }
208 { do stmts <- getCgStmtsEC (loopDecls $3);
209 blks <- code (cgStmtsToBlocks stmts)
210 code (emitProc [] (mkRtsCodeLabelFS $1) [] blks) }
212 info :: { ExtFCode (CLabel, [CmmLit],[CmmLit]) }
213 : 'INFO_TABLE' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
214 -- ptrs, nptrs, closure type, description, type
215 { stdInfo $3 $5 $7 0 $9 $11 $13 }
217 | 'INFO_TABLE_FUN' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ',' INT ')'
218 -- ptrs, nptrs, closure type, description, type, fun type
219 { funInfo $3 $5 $7 $9 $11 $13 $15 }
221 | 'INFO_TABLE_CONSTR' '(' NAME ',' INT ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
222 -- ptrs, nptrs, tag, closure type, description, type
223 { stdInfo $3 $5 $7 $9 $11 $13 $15 }
225 | 'INFO_TABLE_SELECTOR' '(' NAME ',' INT ',' INT ',' STRING ',' STRING ')'
226 -- selector, closure type, description, type
227 { basicInfo $3 (mkIntCLit (fromIntegral $5)) 0 $7 $9 $11 }
229 | 'INFO_TABLE_RET' '(' NAME ',' INT ',' INT ',' INT maybe_vec ')'
230 { retInfo $3 $5 $7 $9 $10 }
232 maybe_vec :: { [CLabel] }
234 | ',' NAME maybe_vec { mkRtsCodeLabelFS $2 : $3 }
237 : {- empty -} { return () }
238 | decl body { do $1; $2 }
239 | stmt body { do $1; $2 }
242 : type names ';' { mapM_ (newLocal $1) $2 }
243 | 'import' names ';' { return () } -- ignore imports
244 | 'export' names ';' { return () } -- ignore exports
246 names :: { [FastString] }
248 | NAME ',' names { $1 : $3 }
253 | block_id ':' { code (labelC $1) }
256 { do reg <- $1; e <- $3; stmtEC (CmmAssign reg e) }
257 | type '[' expr ']' '=' expr ';'
259 | 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
260 {% foreignCall $2 [] $3 $5 $7 }
261 | lreg '=' 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
262 {% let result = do r <- $1; return (r,NoHint) in
263 foreignCall $4 [result] $5 $7 $9 }
264 | STRING lreg '=' 'foreign' STRING expr '(' hint_exprs0 ')' vols ';'
265 {% do h <- parseHint $1;
266 let result = do r <- $2; return (r,h) in
267 foreignCall $5 [result] $6 $8 $10 }
268 -- stmt-level macros, stealing syntax from ordinary C-- function calls.
269 -- Perhaps we ought to use the %%-form?
270 | NAME '(' exprs0 ')' ';'
272 | 'switch' maybe_range expr '{' arms default '}'
273 { doSwitch $2 $3 $5 $6 }
274 | 'goto' block_id ';'
275 { stmtEC (CmmBranch $2) }
276 | 'jump' expr {-maybe_actuals-} ';'
277 { do e <- $2; stmtEC (CmmJump e []) }
278 | 'if' bool_expr '{' body '}' else
279 { ifThenElse $2 $4 $6 }
281 bool_expr :: { ExtFCode BoolExpr }
283 | expr { do e <- $1; return (BoolTest e) }
285 bool_op :: { ExtFCode BoolExpr }
286 : bool_expr '&&' bool_expr { do e1 <- $1; e2 <- $3;
287 return (BoolAnd e1 e2) }
288 | bool_expr '||' bool_expr { do e1 <- $1; e2 <- $3;
289 return (BoolOr e1 e2) }
290 | '!' bool_expr { do e <- $2; return (BoolNot e) }
291 | '(' bool_op ')' { $2 }
293 -- This is not C-- syntax. What to do?
294 vols :: { Maybe [GlobalReg] }
295 : {- empty -} { Nothing }
296 | '[' globals ']' { Just $2 }
298 globals :: { [GlobalReg] }
300 | GLOBALREG ',' globals { $1 : $3 }
302 maybe_range :: { Maybe (Int,Int) }
303 : '[' INT '..' INT ']' { Just (fromIntegral $2, fromIntegral $4) }
304 | {- empty -} { Nothing }
306 arms :: { [([Int],ExtCode)] }
308 | arm arms { $1 : $2 }
310 arm :: { ([Int],ExtCode) }
311 : 'case' ints ':' '{' body '}' { ($2, $5) }
314 : INT { [ fromIntegral $1 ] }
315 | INT ',' ints { fromIntegral $1 : $3 }
317 default :: { Maybe ExtCode }
318 : 'default' ':' '{' body '}' { Just $4 }
319 -- taking a few liberties with the C-- syntax here; C-- doesn't have
320 -- 'default' branches
321 | {- empty -} { Nothing }
324 : {- empty -} { nopEC }
325 | 'else' '{' body '}' { $3 }
327 -- we have to write this out longhand so that Happy's precedence rules
329 expr :: { ExtFCode CmmExpr }
330 : expr '/' expr { mkMachOp MO_U_Quot [$1,$3] }
331 | expr '*' expr { mkMachOp MO_Mul [$1,$3] }
332 | expr '%' expr { mkMachOp MO_U_Rem [$1,$3] }
333 | expr '-' expr { mkMachOp MO_Sub [$1,$3] }
334 | expr '+' expr { mkMachOp MO_Add [$1,$3] }
335 | expr '>>' expr { mkMachOp MO_U_Shr [$1,$3] }
336 | expr '<<' expr { mkMachOp MO_Shl [$1,$3] }
337 | expr '&' expr { mkMachOp MO_And [$1,$3] }
338 | expr '^' expr { mkMachOp MO_Xor [$1,$3] }
339 | expr '|' expr { mkMachOp MO_Or [$1,$3] }
340 | expr '>=' expr { mkMachOp MO_U_Ge [$1,$3] }
341 | expr '>' expr { mkMachOp MO_U_Gt [$1,$3] }
342 | expr '<=' expr { mkMachOp MO_U_Le [$1,$3] }
343 | expr '<' expr { mkMachOp MO_U_Lt [$1,$3] }
344 | expr '!=' expr { mkMachOp MO_Ne [$1,$3] }
345 | expr '==' expr { mkMachOp MO_Eq [$1,$3] }
346 | '~' expr { mkMachOp MO_Not [$2] }
347 | '-' expr { mkMachOp MO_S_Neg [$2] }
348 | expr0 '`' NAME '`' expr0 {% do { mo <- nameToMachOp $3 ;
349 return (mkMachOp mo [$1,$5]) } }
352 expr0 :: { ExtFCode CmmExpr }
353 : INT maybe_ty { return (CmmLit (CmmInt $1 $2)) }
354 | FLOAT maybe_ty { return (CmmLit (CmmFloat $1 $2)) }
355 | STRING { do s <- code (mkStringCLit $1);
358 | type '[' expr ']' { do e <- $3; return (CmmLoad e $1) }
359 | '%' NAME '(' exprs0 ')' {% exprOp $2 $4 }
360 | '(' expr ')' { $2 }
363 -- leaving out the type of a literal gives you the native word size in C--
364 maybe_ty :: { MachRep }
365 : {- empty -} { wordRep }
368 hint_exprs0 :: { [ExtFCode (CmmExpr, MachHint)] }
372 hint_exprs :: { [ExtFCode (CmmExpr, MachHint)] }
374 | hint_expr ',' hint_exprs { $1 : $3 }
376 hint_expr :: { ExtFCode (CmmExpr, MachHint) }
377 : expr { do e <- $1; return (e, inferHint e) }
378 | expr STRING {% do h <- parseHint $2;
380 e <- $1; return (e,h) }
382 exprs0 :: { [ExtFCode CmmExpr] }
386 exprs :: { [ExtFCode CmmExpr] }
388 | expr ',' exprs { $1 : $3 }
390 reg :: { ExtFCode CmmExpr }
391 : NAME { lookupName $1 }
392 | GLOBALREG { return (CmmReg (CmmGlobal $1)) }
394 lreg :: { ExtFCode CmmReg }
395 : NAME { do e <- lookupName $1;
399 other -> pprPanic "CmmParse:" (ftext $1 <> text " not a register") }
400 | GLOBALREG { return (CmmGlobal $1) }
402 block_id :: { BlockId }
403 : NAME { BlockId (newTagUnique (getUnique $1) 'L') }
404 -- TODO: ugh. The unique of a FastString has a null
405 -- tag, so we have to put our own tag on. We should
406 -- really make a new unique for every label, and keep
407 -- them in an environment.
413 typenot8 :: { MachRep }
420 section :: String -> Section
421 section "text" = Text
422 section "data" = Data
423 section "rodata" = ReadOnlyData
424 section "bss" = UninitialisedData
425 section s = OtherSection s
427 -- mkMachOp infers the type of the MachOp from the type of its first
428 -- argument. We assume that this is correct: for MachOps that don't have
429 -- symmetrical args (e.g. shift ops), the first arg determines the type of
431 mkMachOp :: (MachRep -> MachOp) -> [ExtFCode CmmExpr] -> ExtFCode CmmExpr
432 mkMachOp fn args = do
433 arg_exprs <- sequence args
434 return (CmmMachOp (fn (cmmExprRep (head arg_exprs))) arg_exprs)
436 getLit :: CmmExpr -> CmmLit
437 getLit (CmmLit l) = l
438 getLit (CmmMachOp (MO_S_Neg _) [CmmLit (CmmInt i r)]) = CmmInt (negate i) r
439 getLit _ = panic "invalid literal" -- TODO messy failure
441 nameToMachOp :: FastString -> P (MachRep -> MachOp)
443 case lookupUFM machOps name of
444 Nothing -> fail ("unknown primitive " ++ unpackFS name)
447 exprOp :: FastString -> [ExtFCode CmmExpr] -> P (ExtFCode CmmExpr)
448 exprOp name args_code =
449 case lookupUFM exprMacros name of
450 Just f -> return $ do
451 args <- sequence args_code
454 mo <- nameToMachOp name
455 return $ mkMachOp mo args_code
457 exprMacros :: UniqFM ([CmmExpr] -> CmmExpr)
458 exprMacros = listToUFM [
459 ( FSLIT("ENTRY_CODE"), \ [x] -> entryCode x ),
460 ( FSLIT("GET_ENTRY"), \ [x] -> entryCode (closureInfoPtr x) ),
461 ( FSLIT("STD_INFO"), \ [x] -> infoTable x ),
462 ( FSLIT("GET_STD_INFO"), \ [x] -> infoTable (closureInfoPtr x) ),
463 ( FSLIT("GET_FUN_INFO"), \ [x] -> funInfoTable (closureInfoPtr x) ),
464 ( FSLIT("INFO_TYPE"), \ [x] -> infoTableClosureType x ),
465 ( FSLIT("INFO_PTRS"), \ [x] -> infoTablePtrs x ),
466 ( FSLIT("INFO_NPTRS"), \ [x] -> infoTableNonPtrs x ),
467 ( FSLIT("RET_VEC"), \ [info, conZ] -> CmmLoad (vectorSlot info conZ) wordRep )
470 -- we understand a subset of C-- primitives:
471 machOps = listToUFM $
472 map (\(x, y) -> (mkFastString x, y)) [
479 ( "quot", MO_S_Quot ),
481 ( "divu", MO_U_Quot ),
482 ( "modu", MO_U_Rem ),
500 ( "fneg", MO_S_Neg ),
507 ( "shrl", MO_U_Shr ),
508 ( "shra", MO_S_Shr ),
510 ( "lobits8", flip MO_U_Conv I8 ),
511 ( "lobits16", flip MO_U_Conv I16 ),
512 ( "lobits32", flip MO_U_Conv I32 ),
513 ( "lobits64", flip MO_U_Conv I64 ),
514 ( "sx16", flip MO_S_Conv I16 ),
515 ( "sx32", flip MO_S_Conv I32 ),
516 ( "sx64", flip MO_S_Conv I64 ),
517 ( "zx16", flip MO_U_Conv I16 ),
518 ( "zx32", flip MO_U_Conv I32 ),
519 ( "zx64", flip MO_U_Conv I64 ),
520 ( "f2f32", flip MO_S_Conv F32 ), -- TODO; rounding mode
521 ( "f2f64", flip MO_S_Conv F64 ), -- TODO; rounding mode
522 ( "f2i8", flip MO_S_Conv I8 ),
523 ( "f2i16", flip MO_S_Conv I8 ),
524 ( "f2i32", flip MO_S_Conv I8 ),
525 ( "f2i64", flip MO_S_Conv I8 ),
526 ( "i2f32", flip MO_S_Conv F32 ),
527 ( "i2f64", flip MO_S_Conv F64 )
530 parseHint :: String -> P MachHint
531 parseHint "ptr" = return PtrHint
532 parseHint "signed" = return SignedHint
533 parseHint "float" = return FloatHint
534 parseHint str = fail ("unrecognised hint: " ++ str)
536 -- labels are always pointers, so we might as well infer the hint
537 inferHint :: CmmExpr -> MachHint
538 inferHint (CmmLit (CmmLabel _)) = PtrHint
539 inferHint (CmmReg (CmmGlobal g)) | isPtrGlobalReg g = PtrHint
542 isPtrGlobalReg Sp = True
543 isPtrGlobalReg SpLim = True
544 isPtrGlobalReg Hp = True
545 isPtrGlobalReg HpLim = True
546 isPtrGlobalReg CurrentTSO = True
547 isPtrGlobalReg CurrentNursery = True
548 isPtrGlobalReg _ = False
551 happyError = srcParseFail
553 -- -----------------------------------------------------------------------------
554 -- Statement-level macros
556 stmtMacro :: FastString -> [ExtFCode CmmExpr] -> P ExtCode
557 stmtMacro fun args_code = do
558 case lookupUFM stmtMacros fun of
559 Nothing -> fail ("unknown macro: " ++ unpackFS fun)
560 Just fcode -> return $ do
561 args <- sequence args_code
564 stmtMacros :: UniqFM ([CmmExpr] -> Code)
565 stmtMacros = listToUFM [
566 ( FSLIT("CCS_ALLOC"), \[words,ccs] -> profAlloc words ccs ),
567 ( FSLIT("CLOSE_NURSERY"), \[] -> emitCloseNursery ),
568 ( FSLIT("ENTER_CCS_PAP_CL"), \[e] -> enterCostCentrePAP e ),
569 ( FSLIT("ENTER_CCS_THUNK"), \[e] -> enterCostCentreThunk e ),
570 ( FSLIT("HP_CHK_GEN"), \[words,liveness,reentry] ->
571 hpChkGen words liveness reentry ),
572 ( FSLIT("HP_CHK_NP_ASSIGN_SP0"), \[e,f] -> hpChkNodePointsAssignSp0 e f ),
573 ( FSLIT("LOAD_THREAD_STATE"), \[] -> emitLoadThreadState ),
574 ( FSLIT("LDV_ENTER"), \[e] -> ldvEnter e ),
575 ( FSLIT("LDV_RECORD_CREATE"), \[e] -> ldvRecordCreate e ),
576 ( FSLIT("OPEN_NURSERY"), \[] -> emitOpenNursery ),
577 ( FSLIT("PUSH_UPD_FRAME"), \[sp,e] -> emitPushUpdateFrame sp e ),
578 ( FSLIT("SAVE_THREAD_STATE"), \[] -> emitSaveThreadState ),
579 ( FSLIT("SET_HDR"), \[ptr,info,ccs] ->
580 emitSetDynHdr ptr info ccs ),
581 ( FSLIT("STK_CHK_GEN"), \[words,liveness,reentry] ->
582 stkChkGen words liveness reentry ),
583 ( FSLIT("STK_CHK_NP"), \[e] -> stkChkNodePoints e ),
584 ( FSLIT("TICK_ALLOC_PRIM"), \[hdr,goods,slop] ->
585 tickyAllocPrim hdr goods slop ),
586 ( FSLIT("TICK_ALLOC_PAP"), \[goods,slop] ->
587 tickyAllocPAP goods slop ),
588 ( FSLIT("TICK_ALLOC_UP_THK"), \[goods,slop] ->
589 tickyAllocThunk goods slop ),
590 ( FSLIT("UPD_BH_UPDATABLE"), \[] -> emitBlackHoleCode False ),
591 ( FSLIT("UPD_BH_SINGLE_ENTRY"), \[] -> emitBlackHoleCode True ),
593 ( FSLIT("RET_P"), \[a] -> emitRetUT [(PtrArg,a)]),
594 ( FSLIT("RET_N"), \[a] -> emitRetUT [(NonPtrArg,a)]),
595 ( FSLIT("RET_PP"), \[a,b] -> emitRetUT [(PtrArg,a),(PtrArg,b)]),
596 ( FSLIT("RET_NN"), \[a,b] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b)]),
597 ( FSLIT("RET_NP"), \[a,b] -> emitRetUT [(NonPtrArg,a),(PtrArg,b)]),
598 ( FSLIT("RET_PPP"), \[a,b,c] -> emitRetUT [(PtrArg,a),(PtrArg,b),(PtrArg,c)]),
599 ( FSLIT("RET_NNP"), \[a,b,c] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(PtrArg,c)]),
600 ( FSLIT("RET_NNNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(NonPtrArg,c),(PtrArg,d)]),
601 ( FSLIT("RET_NPNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(PtrArg,b),(NonPtrArg,c),(PtrArg,d)])
605 -- -----------------------------------------------------------------------------
606 -- Our extended FCode monad.
608 -- We add a mapping from names to CmmExpr, to support local variable names in
609 -- the concrete C-- code. The unique supply of the underlying FCode monad
610 -- is used to grab a new unique for each local variable.
612 -- In C--, a local variable can be declared anywhere within a proc,
613 -- and it scopes from the beginning of the proc to the end. Hence, we have
614 -- to collect declarations as we parse the proc, and feed the environment
615 -- back in circularly (to avoid a two-pass algorithm).
617 type Decls = [(FastString,CmmExpr)]
618 type Env = UniqFM CmmExpr
620 newtype ExtFCode a = EC { unEC :: Env -> Decls -> FCode (Decls, a) }
622 type ExtCode = ExtFCode ()
624 returnExtFC a = EC $ \e s -> return (s, a)
625 thenExtFC (EC m) k = EC $ \e s -> do (s',r) <- m e s; unEC (k r) e s'
627 instance Monad ExtFCode where
631 -- This function takes the variable decarations and imports and makes
632 -- an environment, which is looped back into the computation. In this
633 -- way, we can have embedded declarations that scope over the whole
634 -- procedure, and imports that scope over the entire module.
635 loopDecls :: ExtFCode a -> ExtFCode a
636 loopDecls (EC fcode) =
637 EC $ \e s -> fixC (\ ~(decls,a) -> fcode (addListToUFM e decls) [])
639 getEnv :: ExtFCode Env
640 getEnv = EC $ \e s -> return (s, e)
642 addVarDecl :: FastString -> CmmExpr -> ExtCode
643 addVarDecl var expr = EC $ \e s -> return ((var,expr):s, ())
645 newLocal :: MachRep -> FastString -> ExtCode
646 newLocal ty name = do
648 addVarDecl name (CmmReg (CmmLocal (LocalReg u ty)))
650 -- Unknown names are treated as if they had been 'import'ed.
651 -- This saves us a lot of bother in the RTS sources, at the expense of
652 -- deferring some errors to link time.
653 lookupName :: FastString -> ExtFCode CmmExpr
657 case lookupUFM env name of
658 Nothing -> CmmLit (CmmLabel (mkRtsCodeLabelFS name))
661 -- Lifting FCode computations into the ExtFCode monad:
662 code :: FCode a -> ExtFCode a
663 code fc = EC $ \e s -> do r <- fc; return (s, r)
665 code2 :: (FCode (Decls,b) -> FCode ((Decls,b),c))
666 -> ExtFCode b -> ExtFCode c
667 code2 f (EC ec) = EC $ \e s -> do ((s',b),c) <- f (ec e s); return (s',c)
670 stmtEC stmt = code (stmtC stmt)
671 stmtsEC stmts = code (stmtsC stmts)
672 getCgStmtsEC = code2 getCgStmts'
674 forkLabelledCodeEC ec = do
675 stmts <- getCgStmtsEC ec
676 code (forkCgStmts stmts)
678 retInfo name size live_bits cl_type vector = do
679 let liveness = smallLiveness (fromIntegral size) (fromIntegral live_bits)
680 (info1,info2) = mkRetInfoTable liveness NoC_SRT
681 (fromIntegral cl_type) vector
682 return (mkRtsRetInfoLabelFS name, info1, info2)
684 stdInfo name ptrs nptrs srt_bitmap cl_type desc_str ty_str =
685 basicInfo name (packHalfWordsCLit ptrs nptrs)
686 srt_bitmap cl_type desc_str ty_str
688 basicInfo name layout srt_bitmap cl_type desc_str ty_str = do
689 lit1 <- if opt_SccProfilingOn
690 then code $ mkStringCLit desc_str
691 else return (mkIntCLit 0)
692 lit2 <- if opt_SccProfilingOn
693 then code $ mkStringCLit ty_str
694 else return (mkIntCLit 0)
695 let info1 = mkStdInfoTable lit1 lit2 (fromIntegral cl_type)
696 (fromIntegral srt_bitmap)
698 return (mkRtsInfoLabelFS name, info1, [])
700 funInfo name ptrs nptrs cl_type desc_str ty_str fun_type = do
701 (label,info1,_) <- stdInfo name ptrs nptrs 0{-srt_bitmap-}
702 cl_type desc_str ty_str
703 let info2 = mkFunGenInfoExtraBits (fromIntegral fun_type) 0 zero zero zero
704 -- we leave most of the fields zero here. This is only used
705 -- to generate the BCO info table in the RTS at the moment.
706 return (label,info1,info2)
711 staticClosure :: FastString -> FastString -> [CmmLit] -> ExtCode
712 staticClosure cl_label info payload
713 = code $ emitDataLits (mkRtsDataLabelFS cl_label) lits
714 where lits = mkStaticClosure (mkRtsInfoLabelFS info) dontCareCCS payload [] []
718 -> [ExtFCode (CmmReg,MachHint)]
720 -> [ExtFCode (CmmExpr,MachHint)]
721 -> Maybe [GlobalReg] -> P ExtCode
722 foreignCall "C" results_code expr_code args_code vols
724 results <- sequence results_code
726 args <- sequence args_code
727 stmtEC (CmmCall (CmmForeignCall expr CCallConv) results args vols)
728 foreignCall conv _ _ _ _
729 = fail ("unknown calling convention: " ++ conv)
731 doStore :: MachRep -> ExtFCode CmmExpr -> ExtFCode CmmExpr -> ExtCode
732 doStore rep addr_code val_code
733 = do addr <- addr_code
735 -- if the specified store type does not match the type of the expr
736 -- on the rhs, then we insert a coercion that will cause the type
737 -- mismatch to be flagged by cmm-lint. If we don't do this, then
738 -- the store will happen at the wrong type, and the error will not
741 | cmmExprRep val /= rep = CmmMachOp (MO_U_Conv rep rep) [val]
743 stmtEC (CmmStore addr coerce_val)
745 -- Return an unboxed tuple.
746 emitRetUT :: [(CgRep,CmmExpr)] -> Code
748 tickyUnboxedTupleReturn (length args) -- TICK
749 (sp, stmts) <- pushUnboxedTuple 0 args
751 when (sp /= 0) $ stmtC (CmmAssign spReg (cmmRegOffW spReg (-sp)))
752 stmtC (CmmJump (entryCode (CmmLoad (cmmRegOffW spReg sp) wordRep)) [])
754 -- -----------------------------------------------------------------------------
755 -- If-then-else and boolean expressions
758 = BoolExpr `BoolAnd` BoolExpr
759 | BoolExpr `BoolOr` BoolExpr
763 -- ToDo: smart constructors which simplify the boolean expression.
765 ifThenElse cond then_part else_part = do
766 then_id <- code newLabelC
767 join_id <- code newLabelC
771 stmtEC (CmmBranch join_id)
772 code (labelC then_id)
774 -- fall through to join
775 code (labelC join_id)
777 -- 'emitCond cond true_id' emits code to test whether the cond is true,
778 -- branching to true_id if so, and falling through otherwise.
779 emitCond (BoolTest e) then_id = do
780 stmtEC (CmmCondBranch e then_id)
781 emitCond (BoolNot (BoolTest (CmmMachOp op args))) then_id
782 | Just op' <- maybeInvertComparison op
783 = emitCond (BoolTest (CmmMachOp op' args)) then_id
784 emitCond (BoolNot e) then_id = do
785 else_id <- code newLabelC
787 stmtEC (CmmBranch then_id)
788 code (labelC else_id)
789 emitCond (e1 `BoolOr` e2) then_id = do
792 emitCond (e1 `BoolAnd` e2) then_id = do
793 -- we'd like to invert one of the conditionals here to avoid an
794 -- extra branch instruction, but we can't use maybeInvertComparison
795 -- here because we can't look too closely at the expression since
797 and_id <- code newLabelC
798 else_id <- code newLabelC
800 stmtEC (CmmBranch else_id)
803 code (labelC else_id)
806 -- -----------------------------------------------------------------------------
809 -- We use a simplified form of C-- switch statements for now. A
810 -- switch statement always compiles to a table jump. Each arm can
811 -- specify a list of values (not ranges), and there can be a single
812 -- default branch. The range of the table is given either by the
813 -- optional range on the switch (eg. switch [0..7] {...}), or by
814 -- the minimum/maximum values from the branches.
816 doSwitch :: Maybe (Int,Int) -> ExtFCode CmmExpr -> [([Int],ExtCode)]
817 -> Maybe ExtCode -> ExtCode
818 doSwitch mb_range scrut arms deflt
820 -- Compile code for the default branch
823 Nothing -> return Nothing
824 Just e -> do b <- forkLabelledCodeEC e; return (Just b)
826 -- Compile each case branch
827 table_entries <- mapM emitArm arms
829 -- Construct the table
831 all_entries = concat table_entries
832 ixs = map fst all_entries
834 | Just (l,u) <- mb_range = (l,u)
835 | otherwise = (minimum ixs, maximum ixs)
837 entries = elems (accumArray (\_ a -> Just a) dflt_entry (min,max)
840 -- ToDo: check for out of range and jump to default if necessary
841 stmtEC (CmmSwitch expr entries)
843 emitArm :: ([Int],ExtCode) -> ExtFCode [(Int,BlockId)]
844 emitArm (ints,code) = do
845 blockid <- forkLabelledCodeEC code
846 return [ (i,blockid) | i <- ints ]
849 -- -----------------------------------------------------------------------------
850 -- Putting it all together
852 -- The initial environment: we define some constants that the compiler
855 initEnv = listToUFM [
856 ( FSLIT("SIZEOF_StgHeader"),
857 CmmLit (CmmInt (fromIntegral (fixedHdrSize * wORD_SIZE)) wordRep) )
860 parseCmmFile :: DynFlags -> FilePath -> IO (Maybe Cmm)
861 parseCmmFile dflags filename = do
862 showPass dflags "ParseCmm"
863 buf <- hGetStringBuffer filename
865 init_loc = mkSrcLoc (mkFastString filename) 1 0
866 init_state = (mkPState buf init_loc dflags) { lex_state = [0] }
867 -- reset the lex_state: the Lexer monad leaves some stuff
868 -- in there we don't want.
869 case unP cmmParse init_state of
870 PFailed span err -> do printError span err; return Nothing
872 cmm <- initC no_module (getCmm (unEC code initEnv [] >> return ()))
873 dumpIfSet_dyn dflags Opt_D_dump_cmm "Cmm" (pprCmms [cmm])
876 no_module = panic "parseCmmFile: no module"