1 -----------------------------------------------------------------------------
3 -- (c) The University of Glasgow, 2004-2006
5 -- Parser for concrete Cmm.
7 -----------------------------------------------------------------------------
11 -- The above warning supression flag is a temporary kludge.
12 -- While working on this module you are encouraged to remove it and fix
13 -- any warnings in the module. See
14 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
17 module CmmParse ( parseCmmFile ) where
58 import Data.Char ( ord )
61 #include "HsVersions.h"
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 '-') }
80 '+' { L _ (CmmT_SpecChar '+') }
81 '&' { L _ (CmmT_SpecChar '&') }
82 '^' { L _ (CmmT_SpecChar '^') }
83 '|' { L _ (CmmT_SpecChar '|') }
84 '>' { L _ (CmmT_SpecChar '>') }
85 '<' { L _ (CmmT_SpecChar '<') }
86 ',' { L _ (CmmT_SpecChar ',') }
87 '!' { L _ (CmmT_SpecChar '!') }
89 '..' { L _ (CmmT_DotDot) }
90 '::' { L _ (CmmT_DoubleColon) }
91 '>>' { L _ (CmmT_Shr) }
92 '<<' { L _ (CmmT_Shl) }
93 '>=' { L _ (CmmT_Ge) }
94 '<=' { L _ (CmmT_Le) }
95 '==' { L _ (CmmT_Eq) }
96 '!=' { L _ (CmmT_Ne) }
97 '&&' { L _ (CmmT_BoolAnd) }
98 '||' { L _ (CmmT_BoolOr) }
100 'CLOSURE' { L _ (CmmT_CLOSURE) }
101 'INFO_TABLE' { L _ (CmmT_INFO_TABLE) }
102 'INFO_TABLE_RET'{ L _ (CmmT_INFO_TABLE_RET) }
103 'INFO_TABLE_FUN'{ L _ (CmmT_INFO_TABLE_FUN) }
104 'INFO_TABLE_CONSTR'{ L _ (CmmT_INFO_TABLE_CONSTR) }
105 'INFO_TABLE_SELECTOR'{ L _ (CmmT_INFO_TABLE_SELECTOR) }
106 'else' { L _ (CmmT_else) }
107 'export' { L _ (CmmT_export) }
108 'section' { L _ (CmmT_section) }
109 'align' { L _ (CmmT_align) }
110 'goto' { L _ (CmmT_goto) }
111 'if' { L _ (CmmT_if) }
112 'jump' { L _ (CmmT_jump) }
113 'foreign' { L _ (CmmT_foreign) }
114 'never' { L _ (CmmT_never) }
115 'prim' { L _ (CmmT_prim) }
116 'return' { L _ (CmmT_return) }
117 'returns' { L _ (CmmT_returns) }
118 'import' { L _ (CmmT_import) }
119 'switch' { L _ (CmmT_switch) }
120 'case' { L _ (CmmT_case) }
121 'default' { L _ (CmmT_default) }
122 'bits8' { L _ (CmmT_bits8) }
123 'bits16' { L _ (CmmT_bits16) }
124 'bits32' { L _ (CmmT_bits32) }
125 'bits64' { L _ (CmmT_bits64) }
126 'float32' { L _ (CmmT_float32) }
127 'float64' { L _ (CmmT_float64) }
129 GLOBALREG { L _ (CmmT_GlobalReg $$) }
130 NAME { L _ (CmmT_Name $$) }
131 STRING { L _ (CmmT_String $$) }
132 INT { L _ (CmmT_Int $$) }
133 FLOAT { L _ (CmmT_Float $$) }
135 %monad { P } { >>= } { return }
136 %lexer { cmmlex } { L _ CmmT_EOF }
138 %tokentype { Located CmmToken }
140 -- C-- operator precedences, taken from the C-- spec
141 %right '||' -- non-std extension, called %disjoin in C--
142 %right '&&' -- non-std extension, called %conjoin in C--
144 %nonassoc '>=' '>' '<=' '<' '!=' '=='
156 : {- empty -} { return () }
157 | cmmtop cmm { do $1; $2 }
159 cmmtop :: { ExtCode }
163 | 'CLOSURE' '(' NAME ',' NAME lits ')' ';'
164 { do lits <- sequence $6;
165 staticClosure $3 $5 (map getLit lits) }
167 -- The only static closures in the RTS are dummy closures like
168 -- stg_END_TSO_QUEUE_closure and stg_dummy_ret. We don't need
169 -- to provide the full generality of static closures here.
171 -- * CCS can always be CCS_DONT_CARE
172 -- * closure is always extern
173 -- * payload is always empty
174 -- * we can derive closure and info table labels from a single NAME
176 cmmdata :: { ExtCode }
177 : 'section' STRING '{' statics '}'
178 { do ss <- sequence $4;
179 code (emitData (section $2) (concat ss)) }
181 statics :: { [ExtFCode [CmmStatic]] }
183 | static statics { $1 : $2 }
185 -- Strings aren't used much in the RTS HC code, so it doesn't seem
186 -- worth allowing inline strings. C-- doesn't allow them anyway.
187 static :: { ExtFCode [CmmStatic] }
188 : NAME ':' { return [CmmDataLabel (mkRtsDataLabelFS $1)] }
189 | type expr ';' { do e <- $2;
190 return [CmmStaticLit (getLit e)] }
191 | type ';' { return [CmmUninitialised
192 (machRepByteWidth $1)] }
193 | 'bits8' '[' ']' STRING ';' { return [mkString $4] }
194 | 'bits8' '[' INT ']' ';' { return [CmmUninitialised
196 | typenot8 '[' INT ']' ';' { return [CmmUninitialised
197 (machRepByteWidth $1 *
199 | 'align' INT ';' { return [CmmAlign (fromIntegral $2)] }
200 | 'CLOSURE' '(' NAME lits ')'
201 { do lits <- sequence $4;
202 return $ map CmmStaticLit $
203 mkStaticClosure (mkRtsInfoLabelFS $3)
204 dontCareCCS (map getLit lits) [] [] [] }
205 -- arrays of closures required for the CHARLIKE & INTLIKE arrays
207 lits :: { [ExtFCode CmmExpr] }
209 | ',' expr lits { $2 : $3 }
211 cmmproc :: { ExtCode }
212 -- TODO: add real SRT/info tables to parsed Cmm
213 : info maybe_formals_without_kinds maybe_gc_block maybe_frame '{' body '}'
214 { do ((entry_ret_label, info, live, formals, gc_block, frame), stmts) <-
215 getCgStmtsEC' $ loopDecls $ do {
216 (entry_ret_label, info, live) <- $1;
217 formals <- sequence $2;
221 return (entry_ret_label, info, live, formals, gc_block, frame) }
222 blks <- code (cgStmtsToBlocks stmts)
223 code (emitInfoTableAndCode entry_ret_label (CmmInfo gc_block frame info) formals blks) }
225 | info maybe_formals_without_kinds ';'
226 { do (entry_ret_label, info, live) <- $1;
227 formals <- sequence $2;
228 code (emitInfoTableAndCode entry_ret_label (CmmInfo Nothing Nothing info) formals []) }
230 | NAME maybe_formals_without_kinds maybe_gc_block maybe_frame '{' body '}'
231 { do ((formals, gc_block, frame), stmts) <-
232 getCgStmtsEC' $ loopDecls $ do {
233 formals <- sequence $2;
237 return (formals, gc_block, frame) }
238 blks <- code (cgStmtsToBlocks stmts)
239 code (emitProc (CmmInfo gc_block frame CmmNonInfoTable) (mkRtsCodeLabelFS $1) formals blks) }
241 info :: { ExtFCode (CLabel, CmmInfoTable, [Maybe LocalReg]) }
242 : 'INFO_TABLE' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
243 -- ptrs, nptrs, closure type, description, type
244 { do prof <- profilingInfo $11 $13
245 return (mkRtsEntryLabelFS $3,
246 CmmInfoTable prof (fromIntegral $9)
247 (ThunkInfo (fromIntegral $5, fromIntegral $7) NoC_SRT),
250 | 'INFO_TABLE_FUN' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ',' INT ')'
251 -- ptrs, nptrs, closure type, description, type, fun type
252 { do prof <- profilingInfo $11 $13
253 return (mkRtsEntryLabelFS $3,
254 CmmInfoTable prof (fromIntegral $9)
255 (FunInfo (fromIntegral $5, fromIntegral $7) NoC_SRT (fromIntegral $15) 0
259 -- we leave most of the fields zero here. This is only used
260 -- to generate the BCO info table in the RTS at the moment.
262 -- A variant with a non-zero arity (needed to write Main_main in Cmm)
263 | 'INFO_TABLE_FUN' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ',' INT ',' INT ')'
264 -- ptrs, nptrs, closure type, description, type, fun type, arity
265 { do prof <- profilingInfo $11 $13
266 return (mkRtsEntryLabelFS $3,
267 CmmInfoTable prof (fromIntegral $9)
268 (FunInfo (fromIntegral $5, fromIntegral $7) NoC_SRT (fromIntegral $15) (fromIntegral $17)
272 -- we leave most of the fields zero here. This is only used
273 -- to generate the BCO info table in the RTS at the moment.
275 | 'INFO_TABLE_CONSTR' '(' NAME ',' INT ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
276 -- ptrs, nptrs, tag, closure type, description, type
277 { do prof <- profilingInfo $13 $15
278 -- If profiling is on, this string gets duplicated,
279 -- but that's the way the old code did it we can fix it some other time.
280 desc_lit <- code $ mkStringCLit $13
281 return (mkRtsEntryLabelFS $3,
282 CmmInfoTable prof (fromIntegral $11)
283 (ConstrInfo (fromIntegral $5, fromIntegral $7) (fromIntegral $9) desc_lit),
286 | 'INFO_TABLE_SELECTOR' '(' NAME ',' INT ',' INT ',' STRING ',' STRING ')'
287 -- selector, closure type, description, type
288 { do prof <- profilingInfo $9 $11
289 return (mkRtsEntryLabelFS $3,
290 CmmInfoTable prof (fromIntegral $7)
291 (ThunkSelectorInfo (fromIntegral $5) NoC_SRT),
294 | 'INFO_TABLE_RET' '(' NAME ',' INT ')'
295 -- closure type (no live regs)
296 { do let infoLabel = mkRtsInfoLabelFS $3
297 return (mkRtsRetLabelFS $3,
298 CmmInfoTable (ProfilingInfo zeroCLit zeroCLit) (fromIntegral $5)
299 (ContInfo [] NoC_SRT),
302 | 'INFO_TABLE_RET' '(' NAME ',' INT ',' formals_without_kinds0 ')'
303 -- closure type, live regs
304 { do live <- sequence (map (liftM Just) $7)
305 return (mkRtsRetLabelFS $3,
306 CmmInfoTable (ProfilingInfo zeroCLit zeroCLit) (fromIntegral $5)
307 (ContInfo live NoC_SRT),
311 : {- empty -} { return () }
312 | decl body { do $1; $2 }
313 | stmt body { do $1; $2 }
316 : type names ';' { mapM_ (newLocal defaultKind $1) $2 }
317 | STRING type names ';' {% do k <- parseGCKind $1;
318 return $ mapM_ (newLocal k $2) $3 }
320 | 'import' names ';' { mapM_ newImport $2 }
321 | 'export' names ';' { return () } -- ignore exports
323 names :: { [FastString] }
325 | NAME ',' names { $1 : $3 }
331 { do l <- newLabel $1; code (labelC l) }
334 { do reg <- $1; e <- $3; stmtEC (CmmAssign reg e) }
335 | type '[' expr ']' '=' expr ';'
338 -- Gah! We really want to say "maybe_results" but that causes
339 -- a shift/reduce conflict with assignment. We either
340 -- we expand out the no-result and single result cases or
341 -- we tweak the syntax to avoid the conflict. The later
342 -- option is taken here because the other way would require
343 -- multiple levels of expanding and get unwieldy.
344 | maybe_results 'foreign' STRING expr '(' cmm_kind_exprs0 ')' safety vols opt_never_returns ';'
345 {% foreignCall $3 $1 $4 $6 $9 $8 $10 }
346 | maybe_results 'prim' '%' NAME '(' cmm_kind_exprs0 ')' safety vols ';'
347 {% primCall $1 $4 $6 $9 $8 }
348 -- stmt-level macros, stealing syntax from ordinary C-- function calls.
349 -- Perhaps we ought to use the %%-form?
350 | NAME '(' exprs0 ')' ';'
352 | 'switch' maybe_range expr '{' arms default '}'
353 { doSwitch $2 $3 $5 $6 }
355 { do l <- lookupLabel $2; stmtEC (CmmBranch l) }
356 | 'jump' expr maybe_actuals ';'
357 { do e1 <- $2; e2 <- sequence $3; stmtEC (CmmJump e1 e2) }
358 | 'return' maybe_actuals ';'
359 { do e <- sequence $2; stmtEC (CmmReturn e) }
360 | 'if' bool_expr '{' body '}' else
361 { ifThenElse $2 $4 $6 }
363 opt_never_returns :: { CmmReturnInfo }
365 | 'never' 'returns' { CmmNeverReturns }
367 bool_expr :: { ExtFCode BoolExpr }
369 | expr { do e <- $1; return (BoolTest e) }
371 bool_op :: { ExtFCode BoolExpr }
372 : bool_expr '&&' bool_expr { do e1 <- $1; e2 <- $3;
373 return (BoolAnd e1 e2) }
374 | bool_expr '||' bool_expr { do e1 <- $1; e2 <- $3;
375 return (BoolOr e1 e2) }
376 | '!' bool_expr { do e <- $2; return (BoolNot e) }
377 | '(' bool_op ')' { $2 }
379 -- This is not C-- syntax. What to do?
380 safety :: { CmmSafety }
381 : {- empty -} { CmmUnsafe } -- Default may change soon
382 | STRING {% parseSafety $1 }
384 -- This is not C-- syntax. What to do?
385 vols :: { Maybe [GlobalReg] }
386 : {- empty -} { Nothing }
387 | '[' ']' { Just [] }
388 | '[' globals ']' { Just $2 }
390 globals :: { [GlobalReg] }
392 | GLOBALREG ',' globals { $1 : $3 }
394 maybe_range :: { Maybe (Int,Int) }
395 : '[' INT '..' INT ']' { Just (fromIntegral $2, fromIntegral $4) }
396 | {- empty -} { Nothing }
398 arms :: { [([Int],ExtCode)] }
400 | arm arms { $1 : $2 }
402 arm :: { ([Int],ExtCode) }
403 : 'case' ints ':' '{' body '}' { ($2, $5) }
406 : INT { [ fromIntegral $1 ] }
407 | INT ',' ints { fromIntegral $1 : $3 }
409 default :: { Maybe ExtCode }
410 : 'default' ':' '{' body '}' { Just $4 }
411 -- taking a few liberties with the C-- syntax here; C-- doesn't have
412 -- 'default' branches
413 | {- empty -} { Nothing }
416 : {- empty -} { nopEC }
417 | 'else' '{' body '}' { $3 }
419 -- we have to write this out longhand so that Happy's precedence rules
421 expr :: { ExtFCode CmmExpr }
422 : expr '/' expr { mkMachOp MO_U_Quot [$1,$3] }
423 | expr '*' expr { mkMachOp MO_Mul [$1,$3] }
424 | expr '%' expr { mkMachOp MO_U_Rem [$1,$3] }
425 | expr '-' expr { mkMachOp MO_Sub [$1,$3] }
426 | expr '+' expr { mkMachOp MO_Add [$1,$3] }
427 | expr '>>' expr { mkMachOp MO_U_Shr [$1,$3] }
428 | expr '<<' expr { mkMachOp MO_Shl [$1,$3] }
429 | expr '&' expr { mkMachOp MO_And [$1,$3] }
430 | expr '^' expr { mkMachOp MO_Xor [$1,$3] }
431 | expr '|' expr { mkMachOp MO_Or [$1,$3] }
432 | expr '>=' expr { mkMachOp MO_U_Ge [$1,$3] }
433 | expr '>' expr { mkMachOp MO_U_Gt [$1,$3] }
434 | expr '<=' expr { mkMachOp MO_U_Le [$1,$3] }
435 | expr '<' expr { mkMachOp MO_U_Lt [$1,$3] }
436 | expr '!=' expr { mkMachOp MO_Ne [$1,$3] }
437 | expr '==' expr { mkMachOp MO_Eq [$1,$3] }
438 | '~' expr { mkMachOp MO_Not [$2] }
439 | '-' expr { mkMachOp MO_S_Neg [$2] }
440 | expr0 '`' NAME '`' expr0 {% do { mo <- nameToMachOp $3 ;
441 return (mkMachOp mo [$1,$5]) } }
444 expr0 :: { ExtFCode CmmExpr }
445 : INT maybe_ty { return (CmmLit (CmmInt $1 $2)) }
446 | FLOAT maybe_ty { return (CmmLit (CmmFloat $1 $2)) }
447 | STRING { do s <- code (mkStringCLit $1);
450 | type '[' expr ']' { do e <- $3; return (CmmLoad e $1) }
451 | '%' NAME '(' exprs0 ')' {% exprOp $2 $4 }
452 | '(' expr ')' { $2 }
455 -- leaving out the type of a literal gives you the native word size in C--
456 maybe_ty :: { MachRep }
457 : {- empty -} { wordRep }
460 maybe_actuals :: { [ExtFCode CmmActual] }
462 | '(' cmm_kind_exprs0 ')' { $2 }
464 cmm_kind_exprs0 :: { [ExtFCode CmmActual] }
466 | cmm_kind_exprs { $1 }
468 cmm_kind_exprs :: { [ExtFCode CmmActual] }
469 : cmm_kind_expr { [$1] }
470 | cmm_kind_expr ',' cmm_kind_exprs { $1 : $3 }
472 cmm_kind_expr :: { ExtFCode CmmActual }
473 : expr { do e <- $1; return (CmmHinted e (inferCmmKind e)) }
474 | expr STRING {% do h <- parseCmmKind $2;
476 e <- $1; return (CmmHinted e h) }
478 exprs0 :: { [ExtFCode CmmExpr] }
482 exprs :: { [ExtFCode CmmExpr] }
484 | expr ',' exprs { $1 : $3 }
486 reg :: { ExtFCode CmmExpr }
487 : NAME { lookupName $1 }
488 | GLOBALREG { return (CmmReg (CmmGlobal $1)) }
490 maybe_results :: { [ExtFCode CmmFormal] }
492 | '(' cmm_formals ')' '=' { $2 }
494 cmm_formals :: { [ExtFCode CmmFormal] }
495 : cmm_formal { [$1] }
496 | cmm_formal ',' { [$1] }
497 | cmm_formal ',' cmm_formals { $1 : $3 }
499 cmm_formal :: { ExtFCode CmmFormal }
500 : local_lreg { do e <- $1; return (CmmHinted e (inferCmmKind (CmmReg (CmmLocal e)))) }
501 | STRING local_lreg {% do h <- parseCmmKind $1;
503 e <- $2; return (CmmHinted e h) }
505 local_lreg :: { ExtFCode LocalReg }
506 : NAME { do e <- lookupName $1;
509 CmmReg (CmmLocal r) -> r
510 other -> pprPanic "CmmParse:" (ftext $1 <> text " not a local register") }
512 lreg :: { ExtFCode CmmReg }
513 : NAME { do e <- lookupName $1;
517 other -> pprPanic "CmmParse:" (ftext $1 <> text " not a register") }
518 | GLOBALREG { return (CmmGlobal $1) }
520 maybe_formals_without_kinds :: { [ExtFCode LocalReg] }
522 | '(' formals_without_kinds0 ')' { $2 }
524 formals_without_kinds0 :: { [ExtFCode LocalReg] }
526 | formals_without_kinds { $1 }
528 formals_without_kinds :: { [ExtFCode LocalReg] }
529 : formal_without_kind ',' { [$1] }
530 | formal_without_kind { [$1] }
531 | formal_without_kind ',' formals_without_kinds { $1 : $3 }
533 formal_without_kind :: { ExtFCode LocalReg }
534 : type NAME { newLocal defaultKind $1 $2 }
535 | STRING type NAME {% do k <- parseGCKind $1;
536 return $ newLocal k $2 $3 }
538 maybe_frame :: { ExtFCode (Maybe UpdateFrame) }
539 : {- empty -} { return Nothing }
540 | 'jump' expr '(' exprs0 ')' { do { target <- $2;
542 return $ Just (UpdateFrame target args) } }
544 maybe_gc_block :: { ExtFCode (Maybe BlockId) }
545 : {- empty -} { return Nothing }
547 { do l <- lookupLabel $2; return (Just l) }
553 typenot8 :: { MachRep }
560 section :: String -> Section
561 section "text" = Text
562 section "data" = Data
563 section "rodata" = ReadOnlyData
564 section "relrodata" = RelocatableReadOnlyData
565 section "bss" = UninitialisedData
566 section s = OtherSection s
568 mkString :: String -> CmmStatic
569 mkString s = CmmString (map (fromIntegral.ord) s)
571 -- mkMachOp infers the type of the MachOp from the type of its first
572 -- argument. We assume that this is correct: for MachOps that don't have
573 -- symmetrical args (e.g. shift ops), the first arg determines the type of
575 mkMachOp :: (MachRep -> MachOp) -> [ExtFCode CmmExpr] -> ExtFCode CmmExpr
576 mkMachOp fn args = do
577 arg_exprs <- sequence args
578 return (CmmMachOp (fn (cmmExprRep (head arg_exprs))) arg_exprs)
580 getLit :: CmmExpr -> CmmLit
581 getLit (CmmLit l) = l
582 getLit (CmmMachOp (MO_S_Neg _) [CmmLit (CmmInt i r)]) = CmmInt (negate i) r
583 getLit _ = panic "invalid literal" -- TODO messy failure
585 nameToMachOp :: FastString -> P (MachRep -> MachOp)
587 case lookupUFM machOps name of
588 Nothing -> fail ("unknown primitive " ++ unpackFS name)
591 exprOp :: FastString -> [ExtFCode CmmExpr] -> P (ExtFCode CmmExpr)
592 exprOp name args_code =
593 case lookupUFM exprMacros name of
594 Just f -> return $ do
595 args <- sequence args_code
598 mo <- nameToMachOp name
599 return $ mkMachOp mo args_code
601 exprMacros :: UniqFM ([CmmExpr] -> CmmExpr)
602 exprMacros = listToUFM [
603 ( FSLIT("ENTRY_CODE"), \ [x] -> entryCode x ),
604 ( FSLIT("INFO_PTR"), \ [x] -> closureInfoPtr x ),
605 ( FSLIT("STD_INFO"), \ [x] -> infoTable x ),
606 ( FSLIT("FUN_INFO"), \ [x] -> funInfoTable x ),
607 ( FSLIT("GET_ENTRY"), \ [x] -> entryCode (closureInfoPtr x) ),
608 ( FSLIT("GET_STD_INFO"), \ [x] -> infoTable (closureInfoPtr x) ),
609 ( FSLIT("GET_FUN_INFO"), \ [x] -> funInfoTable (closureInfoPtr x) ),
610 ( FSLIT("INFO_TYPE"), \ [x] -> infoTableClosureType x ),
611 ( FSLIT("INFO_PTRS"), \ [x] -> infoTablePtrs x ),
612 ( FSLIT("INFO_NPTRS"), \ [x] -> infoTableNonPtrs x )
615 -- we understand a subset of C-- primitives:
616 machOps = listToUFM $
617 map (\(x, y) -> (mkFastString x, y)) [
624 ( "quot", MO_S_Quot ),
626 ( "divu", MO_U_Quot ),
627 ( "modu", MO_U_Rem ),
645 ( "fneg", MO_S_Neg ),
652 ( "shrl", MO_U_Shr ),
653 ( "shra", MO_S_Shr ),
655 ( "lobits8", flip MO_U_Conv I8 ),
656 ( "lobits16", flip MO_U_Conv I16 ),
657 ( "lobits32", flip MO_U_Conv I32 ),
658 ( "lobits64", flip MO_U_Conv I64 ),
659 ( "sx16", flip MO_S_Conv I16 ),
660 ( "sx32", flip MO_S_Conv I32 ),
661 ( "sx64", flip MO_S_Conv I64 ),
662 ( "zx16", flip MO_U_Conv I16 ),
663 ( "zx32", flip MO_U_Conv I32 ),
664 ( "zx64", flip MO_U_Conv I64 ),
665 ( "f2f32", flip MO_S_Conv F32 ), -- TODO; rounding mode
666 ( "f2f64", flip MO_S_Conv F64 ), -- TODO; rounding mode
667 ( "f2i8", flip MO_S_Conv I8 ),
668 ( "f2i16", flip MO_S_Conv I16 ),
669 ( "f2i32", flip MO_S_Conv I32 ),
670 ( "f2i64", flip MO_S_Conv I64 ),
671 ( "i2f32", flip MO_S_Conv F32 ),
672 ( "i2f64", flip MO_S_Conv F64 )
675 callishMachOps = listToUFM $
676 map (\(x, y) -> (mkFastString x, y)) [
677 ( "write_barrier", MO_WriteBarrier )
678 -- ToDo: the rest, maybe
681 parseSafety :: String -> P CmmSafety
682 parseSafety "safe" = return (CmmSafe NoC_SRT)
683 parseSafety "unsafe" = return CmmUnsafe
684 parseSafety str = fail ("unrecognised safety: " ++ str)
686 parseCmmKind :: String -> P CmmKind
687 parseCmmKind "ptr" = return PtrHint
688 parseCmmKind "signed" = return SignedHint
689 parseCmmKind "float" = return FloatHint
690 parseCmmKind str = fail ("unrecognised hint: " ++ str)
692 parseGCKind :: String -> P GCKind
693 parseGCKind "ptr" = return GCKindPtr
694 parseGCKind str = fail ("unrecognized kin: " ++ str)
696 defaultKind :: GCKind
697 defaultKind = GCKindNonPtr
699 -- labels are always pointers, so we might as well infer the hint
700 inferCmmKind :: CmmExpr -> CmmKind
701 inferCmmKind (CmmLit (CmmLabel _)) = PtrHint
702 inferCmmKind (CmmReg (CmmGlobal g)) | isPtrGlobalReg g = PtrHint
703 inferCmmKind _ = NoHint
705 isPtrGlobalReg Sp = True
706 isPtrGlobalReg SpLim = True
707 isPtrGlobalReg Hp = True
708 isPtrGlobalReg HpLim = True
709 isPtrGlobalReg CurrentTSO = True
710 isPtrGlobalReg CurrentNursery = True
711 isPtrGlobalReg _ = False
714 happyError = srcParseFail
716 -- -----------------------------------------------------------------------------
717 -- Statement-level macros
719 stmtMacro :: FastString -> [ExtFCode CmmExpr] -> P ExtCode
720 stmtMacro fun args_code = do
721 case lookupUFM stmtMacros fun of
722 Nothing -> fail ("unknown macro: " ++ unpackFS fun)
723 Just fcode -> return $ do
724 args <- sequence args_code
727 stmtMacros :: UniqFM ([CmmExpr] -> Code)
728 stmtMacros = listToUFM [
729 ( FSLIT("CCS_ALLOC"), \[words,ccs] -> profAlloc words ccs ),
730 ( FSLIT("CLOSE_NURSERY"), \[] -> emitCloseNursery ),
731 ( FSLIT("ENTER_CCS_PAP_CL"), \[e] -> enterCostCentrePAP e ),
732 ( FSLIT("ENTER_CCS_THUNK"), \[e] -> enterCostCentreThunk e ),
733 ( FSLIT("HP_CHK_GEN"), \[words,liveness,reentry] ->
734 hpChkGen words liveness reentry ),
735 ( FSLIT("HP_CHK_NP_ASSIGN_SP0"), \[e,f] -> hpChkNodePointsAssignSp0 e f ),
736 ( FSLIT("LOAD_THREAD_STATE"), \[] -> emitLoadThreadState ),
737 ( FSLIT("LDV_ENTER"), \[e] -> ldvEnter e ),
738 ( FSLIT("LDV_RECORD_CREATE"), \[e] -> ldvRecordCreate e ),
739 ( FSLIT("OPEN_NURSERY"), \[] -> emitOpenNursery ),
740 ( FSLIT("PUSH_UPD_FRAME"), \[sp,e] -> emitPushUpdateFrame sp e ),
741 ( FSLIT("SAVE_THREAD_STATE"), \[] -> emitSaveThreadState ),
742 ( FSLIT("SET_HDR"), \[ptr,info,ccs] ->
743 emitSetDynHdr ptr info ccs ),
744 ( FSLIT("STK_CHK_GEN"), \[words,liveness,reentry] ->
745 stkChkGen words liveness reentry ),
746 ( FSLIT("STK_CHK_NP"), \[e] -> stkChkNodePoints e ),
747 ( FSLIT("TICK_ALLOC_PRIM"), \[hdr,goods,slop] ->
748 tickyAllocPrim hdr goods slop ),
749 ( FSLIT("TICK_ALLOC_PAP"), \[goods,slop] ->
750 tickyAllocPAP goods slop ),
751 ( FSLIT("TICK_ALLOC_UP_THK"), \[goods,slop] ->
752 tickyAllocThunk goods slop ),
753 ( FSLIT("UPD_BH_UPDATABLE"), \[] -> emitBlackHoleCode False ),
754 ( FSLIT("UPD_BH_SINGLE_ENTRY"), \[] -> emitBlackHoleCode True ),
756 ( FSLIT("RET_P"), \[a] -> emitRetUT [(PtrArg,a)]),
757 ( FSLIT("RET_N"), \[a] -> emitRetUT [(NonPtrArg,a)]),
758 ( FSLIT("RET_PP"), \[a,b] -> emitRetUT [(PtrArg,a),(PtrArg,b)]),
759 ( FSLIT("RET_NN"), \[a,b] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b)]),
760 ( FSLIT("RET_NP"), \[a,b] -> emitRetUT [(NonPtrArg,a),(PtrArg,b)]),
761 ( FSLIT("RET_PPP"), \[a,b,c] -> emitRetUT [(PtrArg,a),(PtrArg,b),(PtrArg,c)]),
762 ( FSLIT("RET_NPP"), \[a,b,c] -> emitRetUT [(NonPtrArg,a),(PtrArg,b),(PtrArg,c)]),
763 ( FSLIT("RET_NNP"), \[a,b,c] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(PtrArg,c)]),
764 ( FSLIT("RET_NNNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(NonPtrArg,c),(PtrArg,d)]),
765 ( FSLIT("RET_NPNP"), \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(PtrArg,b),(NonPtrArg,c),(PtrArg,d)])
769 -- -----------------------------------------------------------------------------
770 -- Our extended FCode monad.
772 -- We add a mapping from names to CmmExpr, to support local variable names in
773 -- the concrete C-- code. The unique supply of the underlying FCode monad
774 -- is used to grab a new unique for each local variable.
776 -- In C--, a local variable can be declared anywhere within a proc,
777 -- and it scopes from the beginning of the proc to the end. Hence, we have
778 -- to collect declarations as we parse the proc, and feed the environment
779 -- back in circularly (to avoid a two-pass algorithm).
781 data Named = Var CmmExpr | Label BlockId
782 type Decls = [(FastString,Named)]
783 type Env = UniqFM Named
785 newtype ExtFCode a = EC { unEC :: Env -> Decls -> FCode (Decls, a) }
787 type ExtCode = ExtFCode ()
789 returnExtFC a = EC $ \e s -> return (s, a)
790 thenExtFC (EC m) k = EC $ \e s -> do (s',r) <- m e s; unEC (k r) e s'
792 instance Monad ExtFCode where
796 -- This function takes the variable decarations and imports and makes
797 -- an environment, which is looped back into the computation. In this
798 -- way, we can have embedded declarations that scope over the whole
799 -- procedure, and imports that scope over the entire module.
800 -- Discards the local declaration contained within decl'
801 loopDecls :: ExtFCode a -> ExtFCode a
802 loopDecls (EC fcode) =
803 EC $ \e globalDecls -> do
804 (decls', a) <- fixC (\ ~(decls,a) -> fcode (addListToUFM e (decls ++ globalDecls)) globalDecls)
805 return (globalDecls, a)
807 getEnv :: ExtFCode Env
808 getEnv = EC $ \e s -> return (s, e)
810 addVarDecl :: FastString -> CmmExpr -> ExtCode
811 addVarDecl var expr = EC $ \e s -> return ((var, Var expr):s, ())
813 addLabel :: FastString -> BlockId -> ExtCode
814 addLabel name block_id = EC $ \e s -> return ((name, Label block_id):s, ())
816 newLocal :: GCKind -> MachRep -> FastString -> ExtFCode LocalReg
817 newLocal kind ty name = do
819 let reg = LocalReg u ty kind
820 addVarDecl name (CmmReg (CmmLocal reg))
823 -- Creates a foreign label in the import. CLabel's labelDynamic
824 -- classifies these labels as dynamic, hence the code generator emits the
825 -- PIC code for them.
826 newImport :: FastString -> ExtFCode ()
828 = addVarDecl name (CmmLit (CmmLabel (mkForeignLabel name Nothing True)))
830 newLabel :: FastString -> ExtFCode BlockId
833 addLabel name (BlockId u)
836 lookupLabel :: FastString -> ExtFCode BlockId
837 lookupLabel name = do
840 case lookupUFM env name of
842 _other -> BlockId (newTagUnique (getUnique name) 'L')
844 -- Unknown names are treated as if they had been 'import'ed.
845 -- This saves us a lot of bother in the RTS sources, at the expense of
846 -- deferring some errors to link time.
847 lookupName :: FastString -> ExtFCode CmmExpr
851 case lookupUFM env name of
853 _other -> CmmLit (CmmLabel (mkRtsCodeLabelFS name))
855 -- Lifting FCode computations into the ExtFCode monad:
856 code :: FCode a -> ExtFCode a
857 code fc = EC $ \e s -> do r <- fc; return (s, r)
859 code2 :: (FCode (Decls,b) -> FCode ((Decls,b),c))
860 -> ExtFCode b -> ExtFCode c
861 code2 f (EC ec) = EC $ \e s -> do ((s',b),c) <- f (ec e s); return (s',c)
864 stmtEC stmt = code (stmtC stmt)
865 stmtsEC stmts = code (stmtsC stmts)
866 getCgStmtsEC = code2 getCgStmts'
867 getCgStmtsEC' = code2 (\m -> getCgStmts' m >>= f)
868 where f ((decl, b), c) = return ((decl, b), (b, c))
870 forkLabelledCodeEC ec = do
871 stmts <- getCgStmtsEC ec
872 code (forkCgStmts stmts)
875 profilingInfo desc_str ty_str = do
876 lit1 <- if opt_SccProfilingOn
877 then code $ mkStringCLit desc_str
878 else return (mkIntCLit 0)
879 lit2 <- if opt_SccProfilingOn
880 then code $ mkStringCLit ty_str
881 else return (mkIntCLit 0)
882 return (ProfilingInfo lit1 lit2)
885 staticClosure :: FastString -> FastString -> [CmmLit] -> ExtCode
886 staticClosure cl_label info payload
887 = code $ emitDataLits (mkRtsDataLabelFS cl_label) lits
888 where lits = mkStaticClosure (mkRtsInfoLabelFS info) dontCareCCS payload [] [] []
892 -> [ExtFCode CmmFormal]
894 -> [ExtFCode CmmActual]
899 foreignCall conv_string results_code expr_code args_code vols safety ret
900 = do convention <- case conv_string of
901 "C" -> return CCallConv
902 "stdcall" -> return StdCallConv
903 "C--" -> return CmmCallConv
904 _ -> fail ("unknown calling convention: " ++ conv_string)
906 results <- sequence results_code
908 args <- sequence args_code
909 --code (stmtC (CmmCall (CmmCallee expr convention) results args safety))
911 -- Temporary hack so at least some functions are CmmSafe
912 CmmCallConv -> code (stmtC (CmmCall (CmmCallee expr convention) results args safety ret))
914 let expr' = adjCallTarget convention expr args in
917 code (emitForeignCall' PlayRisky results
918 (CmmCallee expr' convention) args vols NoC_SRT ret)
920 code (emitForeignCall' (PlaySafe unused) results
921 (CmmCallee expr' convention) args vols NoC_SRT ret) where
922 unused = panic "not used by emitForeignCall'"
924 adjCallTarget :: CCallConv -> CmmExpr -> [CmmHinted CmmExpr] -> CmmExpr
925 #ifdef mingw32_TARGET_OS
926 -- On Windows, we have to add the '@N' suffix to the label when making
927 -- a call with the stdcall calling convention.
928 adjCallTarget StdCallConv (CmmLit (CmmLabel lbl)) args
929 = CmmLit (CmmLabel (addLabelSize lbl (sum (map size args))))
930 where size (CmmHinted e _) = max wORD_SIZE (machRepByteWidth (cmmExprRep e))
931 -- c.f. CgForeignCall.emitForeignCall
933 adjCallTarget _ expr _
937 :: [ExtFCode CmmFormal]
939 -> [ExtFCode CmmActual]
943 primCall results_code name args_code vols safety
944 = case lookupUFM callishMachOps name of
945 Nothing -> fail ("unknown primitive " ++ unpackFS name)
946 Just p -> return $ do
947 results <- sequence results_code
948 args <- sequence args_code
951 code (emitForeignCall' PlayRisky results
952 (CmmPrim p) args vols NoC_SRT CmmMayReturn)
954 code (emitForeignCall' (PlaySafe unused) results
955 (CmmPrim p) args vols NoC_SRT CmmMayReturn) where
956 unused = panic "not used by emitForeignCall'"
958 doStore :: MachRep -> ExtFCode CmmExpr -> ExtFCode CmmExpr -> ExtCode
959 doStore rep addr_code val_code
960 = do addr <- addr_code
962 -- if the specified store type does not match the type of the expr
963 -- on the rhs, then we insert a coercion that will cause the type
964 -- mismatch to be flagged by cmm-lint. If we don't do this, then
965 -- the store will happen at the wrong type, and the error will not
968 | cmmExprRep val /= rep = CmmMachOp (MO_U_Conv rep rep) [val]
970 stmtEC (CmmStore addr coerce_val)
972 -- Return an unboxed tuple.
973 emitRetUT :: [(CgRep,CmmExpr)] -> Code
975 tickyUnboxedTupleReturn (length args) -- TICK
976 (sp, stmts) <- pushUnboxedTuple 0 args
978 when (sp /= 0) $ stmtC (CmmAssign spReg (cmmRegOffW spReg (-sp)))
979 stmtC (CmmJump (entryCode (CmmLoad (cmmRegOffW spReg sp) wordRep)) [])
980 -- TODO (when using CPS): emitStmt (CmmReturn (map snd args))
982 -- -----------------------------------------------------------------------------
983 -- If-then-else and boolean expressions
986 = BoolExpr `BoolAnd` BoolExpr
987 | BoolExpr `BoolOr` BoolExpr
991 -- ToDo: smart constructors which simplify the boolean expression.
993 ifThenElse cond then_part else_part = do
994 then_id <- code newLabelC
995 join_id <- code newLabelC
999 stmtEC (CmmBranch join_id)
1000 code (labelC then_id)
1002 -- fall through to join
1003 code (labelC join_id)
1005 -- 'emitCond cond true_id' emits code to test whether the cond is true,
1006 -- branching to true_id if so, and falling through otherwise.
1007 emitCond (BoolTest e) then_id = do
1008 stmtEC (CmmCondBranch e then_id)
1009 emitCond (BoolNot (BoolTest (CmmMachOp op args))) then_id
1010 | Just op' <- maybeInvertComparison op
1011 = emitCond (BoolTest (CmmMachOp op' args)) then_id
1012 emitCond (BoolNot e) then_id = do
1013 else_id <- code newLabelC
1015 stmtEC (CmmBranch then_id)
1016 code (labelC else_id)
1017 emitCond (e1 `BoolOr` e2) then_id = do
1020 emitCond (e1 `BoolAnd` e2) then_id = do
1021 -- we'd like to invert one of the conditionals here to avoid an
1022 -- extra branch instruction, but we can't use maybeInvertComparison
1023 -- here because we can't look too closely at the expression since
1025 and_id <- code newLabelC
1026 else_id <- code newLabelC
1028 stmtEC (CmmBranch else_id)
1029 code (labelC and_id)
1031 code (labelC else_id)
1034 -- -----------------------------------------------------------------------------
1037 -- We use a simplified form of C-- switch statements for now. A
1038 -- switch statement always compiles to a table jump. Each arm can
1039 -- specify a list of values (not ranges), and there can be a single
1040 -- default branch. The range of the table is given either by the
1041 -- optional range on the switch (eg. switch [0..7] {...}), or by
1042 -- the minimum/maximum values from the branches.
1044 doSwitch :: Maybe (Int,Int) -> ExtFCode CmmExpr -> [([Int],ExtCode)]
1045 -> Maybe ExtCode -> ExtCode
1046 doSwitch mb_range scrut arms deflt
1048 -- Compile code for the default branch
1051 Nothing -> return Nothing
1052 Just e -> do b <- forkLabelledCodeEC e; return (Just b)
1054 -- Compile each case branch
1055 table_entries <- mapM emitArm arms
1057 -- Construct the table
1059 all_entries = concat table_entries
1060 ixs = map fst all_entries
1062 | Just (l,u) <- mb_range = (l,u)
1063 | otherwise = (minimum ixs, maximum ixs)
1065 entries = elems (accumArray (\_ a -> Just a) dflt_entry (min,max)
1068 -- ToDo: check for out of range and jump to default if necessary
1069 stmtEC (CmmSwitch expr entries)
1071 emitArm :: ([Int],ExtCode) -> ExtFCode [(Int,BlockId)]
1072 emitArm (ints,code) = do
1073 blockid <- forkLabelledCodeEC code
1074 return [ (i,blockid) | i <- ints ]
1077 -- -----------------------------------------------------------------------------
1078 -- Putting it all together
1080 -- The initial environment: we define some constants that the compiler
1081 -- knows about here.
1083 initEnv = listToUFM [
1084 ( FSLIT("SIZEOF_StgHeader"),
1085 Var (CmmLit (CmmInt (fromIntegral (fixedHdrSize * wORD_SIZE)) wordRep) )),
1086 ( FSLIT("SIZEOF_StgInfoTable"),
1087 Var (CmmLit (CmmInt (fromIntegral stdInfoTableSizeB) wordRep) ))
1090 parseCmmFile :: DynFlags -> FilePath -> IO (Maybe Cmm)
1091 parseCmmFile dflags filename = do
1092 showPass dflags "ParseCmm"
1093 buf <- hGetStringBuffer filename
1095 init_loc = mkSrcLoc (mkFastString filename) 1 0
1096 init_state = (mkPState buf init_loc dflags) { lex_state = [0] }
1097 -- reset the lex_state: the Lexer monad leaves some stuff
1098 -- in there we don't want.
1099 case unP cmmParse init_state of
1100 PFailed span err -> do printError span err; return Nothing
1102 cmm <- initC dflags no_module (getCmm (unEC code initEnv [] >> return ()))
1103 let ms = getMessages pst
1104 printErrorsAndWarnings dflags ms
1105 when (errorsFound dflags ms) $ exitWith (ExitFailure 1)
1106 dumpIfSet_dyn dflags Opt_D_dump_cmm "Cmm" (ppr cmm)
1109 no_module = panic "parseCmmFile: no module"