1 -----------------------------------------------------------------------------
3 -- (c) The University of Glasgow, 2004-2006
5 -- Parser for concrete Cmm.
6 -- This doesn't just parse the Cmm file, we also do some code generation
7 -- along the way for switches and foreign calls etc.
9 -----------------------------------------------------------------------------
11 -- TODO: Add support for interruptible/uninterruptible foreign call specification
14 {-# LANGUAGE BangPatterns #-} -- required for versions of Happy before 1.18.6
15 {-# OPTIONS -Wwarn -w #-}
16 -- The above warning supression flag is a temporary kludge.
17 -- While working on this module you are encouraged to remove it and fix
18 -- any warnings in the module. See
19 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
22 module CmmParse ( parseCmmFile ) where
24 import CgMonad hiding (getDynFlags)
63 import Bag ( emptyBag, unitBag )
68 import Data.Char ( ord )
71 #include "HsVersions.h"
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 '~') }
88 '/' { L _ (CmmT_SpecChar '/') }
89 '*' { L _ (CmmT_SpecChar '*') }
90 '%' { L _ (CmmT_SpecChar '%') }
91 '-' { L _ (CmmT_SpecChar '-') }
92 '+' { L _ (CmmT_SpecChar '+') }
93 '&' { L _ (CmmT_SpecChar '&') }
94 '^' { L _ (CmmT_SpecChar '^') }
95 '|' { L _ (CmmT_SpecChar '|') }
96 '>' { L _ (CmmT_SpecChar '>') }
97 '<' { L _ (CmmT_SpecChar '<') }
98 ',' { L _ (CmmT_SpecChar ',') }
99 '!' { L _ (CmmT_SpecChar '!') }
101 '..' { L _ (CmmT_DotDot) }
102 '::' { L _ (CmmT_DoubleColon) }
103 '>>' { L _ (CmmT_Shr) }
104 '<<' { L _ (CmmT_Shl) }
105 '>=' { L _ (CmmT_Ge) }
106 '<=' { L _ (CmmT_Le) }
107 '==' { L _ (CmmT_Eq) }
108 '!=' { L _ (CmmT_Ne) }
109 '&&' { L _ (CmmT_BoolAnd) }
110 '||' { L _ (CmmT_BoolOr) }
112 'CLOSURE' { L _ (CmmT_CLOSURE) }
113 'INFO_TABLE' { L _ (CmmT_INFO_TABLE) }
114 'INFO_TABLE_RET'{ L _ (CmmT_INFO_TABLE_RET) }
115 'INFO_TABLE_FUN'{ L _ (CmmT_INFO_TABLE_FUN) }
116 'INFO_TABLE_CONSTR'{ L _ (CmmT_INFO_TABLE_CONSTR) }
117 'INFO_TABLE_SELECTOR'{ L _ (CmmT_INFO_TABLE_SELECTOR) }
118 'else' { L _ (CmmT_else) }
119 'export' { L _ (CmmT_export) }
120 'section' { L _ (CmmT_section) }
121 'align' { L _ (CmmT_align) }
122 'goto' { L _ (CmmT_goto) }
123 'if' { L _ (CmmT_if) }
124 'jump' { L _ (CmmT_jump) }
125 'foreign' { L _ (CmmT_foreign) }
126 'never' { L _ (CmmT_never) }
127 'prim' { L _ (CmmT_prim) }
128 'return' { L _ (CmmT_return) }
129 'returns' { L _ (CmmT_returns) }
130 'import' { L _ (CmmT_import) }
131 'switch' { L _ (CmmT_switch) }
132 'case' { L _ (CmmT_case) }
133 'default' { L _ (CmmT_default) }
134 'bits8' { L _ (CmmT_bits8) }
135 'bits16' { L _ (CmmT_bits16) }
136 'bits32' { L _ (CmmT_bits32) }
137 'bits64' { L _ (CmmT_bits64) }
138 'float32' { L _ (CmmT_float32) }
139 'float64' { L _ (CmmT_float64) }
140 'gcptr' { L _ (CmmT_gcptr) }
142 GLOBALREG { L _ (CmmT_GlobalReg $$) }
143 NAME { L _ (CmmT_Name $$) }
144 STRING { L _ (CmmT_String $$) }
145 INT { L _ (CmmT_Int $$) }
146 FLOAT { L _ (CmmT_Float $$) }
148 %monad { P } { >>= } { return }
149 %lexer { cmmlex } { L _ CmmT_EOF }
151 %tokentype { Located CmmToken }
153 -- C-- operator precedences, taken from the C-- spec
154 %right '||' -- non-std extension, called %disjoin in C--
155 %right '&&' -- non-std extension, called %conjoin in C--
157 %nonassoc '>=' '>' '<=' '<' '!=' '=='
169 : {- empty -} { return () }
170 | cmmtop cmm { do $1; $2 }
172 cmmtop :: { ExtCode }
176 | 'CLOSURE' '(' NAME ',' NAME lits ')' ';'
177 {% withThisPackage $ \pkg ->
178 do lits <- sequence $6;
179 staticClosure pkg $3 $5 (map getLit lits) }
181 -- The only static closures in the RTS are dummy closures like
182 -- stg_END_TSO_QUEUE_closure and stg_dummy_ret. We don't need
183 -- to provide the full generality of static closures here.
185 -- * CCS can always be CCS_DONT_CARE
186 -- * closure is always extern
187 -- * payload is always empty
188 -- * we can derive closure and info table labels from a single NAME
190 cmmdata :: { ExtCode }
191 : 'section' STRING '{' statics '}'
192 { do ss <- sequence $4;
193 code (emitData (section $2) (concat ss)) }
195 statics :: { [ExtFCode [CmmStatic]] }
197 | static statics { $1 : $2 }
199 -- Strings aren't used much in the RTS HC code, so it doesn't seem
200 -- worth allowing inline strings. C-- doesn't allow them anyway.
201 static :: { ExtFCode [CmmStatic] }
203 {% withThisPackage $ \pkg ->
204 return [CmmDataLabel (mkCmmDataLabel pkg $1)] }
206 | type expr ';' { do e <- $2;
207 return [CmmStaticLit (getLit e)] }
208 | type ';' { return [CmmUninitialised
209 (widthInBytes (typeWidth $1))] }
210 | 'bits8' '[' ']' STRING ';' { return [mkString $4] }
211 | 'bits8' '[' INT ']' ';' { return [CmmUninitialised
213 | typenot8 '[' INT ']' ';' { return [CmmUninitialised
214 (widthInBytes (typeWidth $1) *
216 | 'align' INT ';' { return [CmmAlign (fromIntegral $2)] }
217 | 'CLOSURE' '(' NAME lits ')'
218 { do lits <- sequence $4;
219 return $ map CmmStaticLit $
220 mkStaticClosure (mkForeignLabel $3 Nothing ForeignLabelInExternalPackage IsData)
221 -- mkForeignLabel because these are only used
222 -- for CHARLIKE and INTLIKE closures in the RTS.
223 dontCareCCS (map getLit lits) [] [] [] }
224 -- arrays of closures required for the CHARLIKE & INTLIKE arrays
226 lits :: { [ExtFCode CmmExpr] }
228 | ',' expr lits { $2 : $3 }
230 cmmproc :: { ExtCode }
231 -- TODO: add real SRT/info tables to parsed Cmm
232 : info maybe_formals_without_hints maybe_gc_block maybe_frame '{' body '}'
233 { do ((entry_ret_label, info, live, formals, gc_block, frame), stmts) <-
234 getCgStmtsEC' $ loopDecls $ do {
235 (entry_ret_label, info, live) <- $1;
236 formals <- sequence $2;
240 return (entry_ret_label, info, live, formals, gc_block, frame) }
241 blks <- code (cgStmtsToBlocks stmts)
242 code (emitInfoTableAndCode entry_ret_label (CmmInfo gc_block frame info) formals blks) }
244 | info maybe_formals_without_hints ';'
245 { do (entry_ret_label, info, live) <- $1;
246 formals <- sequence $2;
247 code (emitInfoTableAndCode entry_ret_label (CmmInfo Nothing Nothing info) formals []) }
249 | NAME maybe_formals_without_hints maybe_gc_block maybe_frame '{' body '}'
250 {% withThisPackage $ \pkg ->
251 do newFunctionName $1 pkg
252 ((formals, gc_block, frame), stmts) <-
253 getCgStmtsEC' $ loopDecls $ do {
254 formals <- sequence $2;
258 return (formals, gc_block, frame) }
259 blks <- code (cgStmtsToBlocks stmts)
260 code (emitProc (CmmInfo gc_block frame CmmNonInfoTable) (mkCmmCodeLabel pkg $1) formals blks) }
262 info :: { ExtFCode (CLabel, CmmInfoTable, [Maybe LocalReg]) }
263 : 'INFO_TABLE' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
264 -- ptrs, nptrs, closure type, description, type
265 {% withThisPackage $ \pkg ->
266 do prof <- profilingInfo $11 $13
267 return (mkCmmEntryLabel pkg $3,
268 CmmInfoTable False prof (fromIntegral $9)
269 (ThunkInfo (fromIntegral $5, fromIntegral $7) NoC_SRT),
272 | 'INFO_TABLE_FUN' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ',' INT ')'
273 -- ptrs, nptrs, closure type, description, type, fun type
274 {% withThisPackage $ \pkg ->
275 do prof <- profilingInfo $11 $13
276 return (mkCmmEntryLabel pkg $3,
277 CmmInfoTable False prof (fromIntegral $9)
278 (FunInfo (fromIntegral $5, fromIntegral $7) NoC_SRT
280 (ArgSpec (fromIntegral $15))
283 -- we leave most of the fields zero here. This is only used
284 -- to generate the BCO info table in the RTS at the moment.
286 -- A variant with a non-zero arity (needed to write Main_main in Cmm)
287 | 'INFO_TABLE_FUN' '(' NAME ',' INT ',' INT ',' INT ',' STRING ',' STRING ',' INT ',' INT ')'
288 -- ptrs, nptrs, closure type, description, type, fun type, arity
289 {% withThisPackage $ \pkg ->
290 do prof <- profilingInfo $11 $13
291 return (mkCmmEntryLabel pkg $3,
292 CmmInfoTable False prof (fromIntegral $9)
293 (FunInfo (fromIntegral $5, fromIntegral $7) NoC_SRT (fromIntegral $17)
294 (ArgSpec (fromIntegral $15))
297 -- we leave most of the fields zero here. This is only used
298 -- to generate the BCO info table in the RTS at the moment.
300 | 'INFO_TABLE_CONSTR' '(' NAME ',' INT ',' INT ',' INT ',' INT ',' STRING ',' STRING ')'
301 -- ptrs, nptrs, tag, closure type, description, type
302 {% withThisPackage $ \pkg ->
303 do prof <- profilingInfo $13 $15
304 -- If profiling is on, this string gets duplicated,
305 -- but that's the way the old code did it we can fix it some other time.
306 desc_lit <- code $ mkStringCLit $13
307 return (mkCmmEntryLabel pkg $3,
308 CmmInfoTable False prof (fromIntegral $11)
309 (ConstrInfo (fromIntegral $5, fromIntegral $7) (fromIntegral $9) desc_lit),
312 | 'INFO_TABLE_SELECTOR' '(' NAME ',' INT ',' INT ',' STRING ',' STRING ')'
313 -- selector, closure type, description, type
314 {% withThisPackage $ \pkg ->
315 do prof <- profilingInfo $9 $11
316 return (mkCmmEntryLabel pkg $3,
317 CmmInfoTable False prof (fromIntegral $7)
318 (ThunkSelectorInfo (fromIntegral $5) NoC_SRT),
321 | 'INFO_TABLE_RET' '(' NAME ',' INT ')'
322 -- closure type (no live regs)
323 {% withThisPackage $ \pkg ->
324 do let infoLabel = mkCmmInfoLabel pkg $3
325 return (mkCmmRetLabel pkg $3,
326 CmmInfoTable False (ProfilingInfo zeroCLit zeroCLit) (fromIntegral $5)
327 (ContInfo [] NoC_SRT),
330 | 'INFO_TABLE_RET' '(' NAME ',' INT ',' formals_without_hints0 ')'
331 -- closure type, live regs
332 {% withThisPackage $ \pkg ->
333 do live <- sequence (map (liftM Just) $7)
334 return (mkCmmRetLabel pkg $3,
335 CmmInfoTable False (ProfilingInfo zeroCLit zeroCLit) (fromIntegral $5)
336 (ContInfo live NoC_SRT),
340 : {- empty -} { return () }
341 | decl body { do $1; $2 }
342 | stmt body { do $1; $2 }
345 : type names ';' { mapM_ (newLocal $1) $2 }
346 | 'import' importNames ';' { mapM_ newImport $2 }
347 | 'export' names ';' { return () } -- ignore exports
350 -- an imported function name, with optional packageId
352 :: { [(FastString, CLabel)] }
353 : importName { [$1] }
354 | importName ',' importNames { $1 : $3 }
357 :: { (FastString, CLabel) }
359 -- A label imported without an explicit packageId.
360 -- These are taken to come frome some foreign, unnamed package.
362 { ($1, mkForeignLabel $1 Nothing ForeignLabelInExternalPackage IsFunction) }
364 -- A label imported with an explicit packageId.
366 { ($2, mkCmmCodeLabel (fsToPackageId (mkFastString $1)) $2) }
369 names :: { [FastString] }
371 | NAME ',' names { $1 : $3 }
377 { do l <- newLabel $1; code (labelC l) }
380 { do reg <- $1; e <- $3; stmtEC (CmmAssign reg e) }
381 | type '[' expr ']' '=' expr ';'
384 -- Gah! We really want to say "maybe_results" but that causes
385 -- a shift/reduce conflict with assignment. We either
386 -- we expand out the no-result and single result cases or
387 -- we tweak the syntax to avoid the conflict. The later
388 -- option is taken here because the other way would require
389 -- multiple levels of expanding and get unwieldy.
390 | maybe_results 'foreign' STRING expr '(' cmm_hint_exprs0 ')' safety vols opt_never_returns ';'
391 {% foreignCall $3 $1 $4 $6 $9 $8 $10 }
392 | maybe_results 'prim' '%' NAME '(' cmm_hint_exprs0 ')' safety vols ';'
393 {% primCall $1 $4 $6 $9 $8 }
394 -- stmt-level macros, stealing syntax from ordinary C-- function calls.
395 -- Perhaps we ought to use the %%-form?
396 | NAME '(' exprs0 ')' ';'
398 | 'switch' maybe_range expr '{' arms default '}'
399 { do as <- sequence $5; doSwitch $2 $3 as $6 }
401 { do l <- lookupLabel $2; stmtEC (CmmBranch l) }
402 | 'jump' expr maybe_actuals ';'
403 { do e1 <- $2; e2 <- sequence $3; stmtEC (CmmJump e1 e2) }
404 | 'return' maybe_actuals ';'
405 { do e <- sequence $2; stmtEC (CmmReturn e) }
406 | 'if' bool_expr 'goto' NAME
407 { do l <- lookupLabel $4; cmmRawIf $2 l }
408 | 'if' bool_expr '{' body '}' else
409 { cmmIfThenElse $2 $4 $6 }
411 opt_never_returns :: { CmmReturnInfo }
413 | 'never' 'returns' { CmmNeverReturns }
415 bool_expr :: { ExtFCode BoolExpr }
417 | expr { do e <- $1; return (BoolTest e) }
419 bool_op :: { ExtFCode BoolExpr }
420 : bool_expr '&&' bool_expr { do e1 <- $1; e2 <- $3;
421 return (BoolAnd e1 e2) }
422 | bool_expr '||' bool_expr { do e1 <- $1; e2 <- $3;
423 return (BoolOr e1 e2) }
424 | '!' bool_expr { do e <- $2; return (BoolNot e) }
425 | '(' bool_op ')' { $2 }
427 -- This is not C-- syntax. What to do?
428 safety :: { CmmSafety }
429 : {- empty -} { CmmUnsafe } -- Default may change soon
430 | STRING {% parseSafety $1 }
432 -- This is not C-- syntax. What to do?
433 vols :: { Maybe [GlobalReg] }
434 : {- empty -} { Nothing }
435 | '[' ']' { Just [] }
436 | '[' globals ']' { Just $2 }
438 globals :: { [GlobalReg] }
440 | GLOBALREG ',' globals { $1 : $3 }
442 maybe_range :: { Maybe (Int,Int) }
443 : '[' INT '..' INT ']' { Just (fromIntegral $2, fromIntegral $4) }
444 | {- empty -} { Nothing }
446 arms :: { [ExtFCode ([Int],Either BlockId ExtCode)] }
448 | arm arms { $1 : $2 }
450 arm :: { ExtFCode ([Int],Either BlockId ExtCode) }
451 : 'case' ints ':' arm_body { do b <- $4; return ($2, b) }
453 arm_body :: { ExtFCode (Either BlockId ExtCode) }
454 : '{' body '}' { return (Right $2) }
455 | 'goto' NAME ';' { do l <- lookupLabel $2; return (Left l) }
458 : INT { [ fromIntegral $1 ] }
459 | INT ',' ints { fromIntegral $1 : $3 }
461 default :: { Maybe ExtCode }
462 : 'default' ':' '{' body '}' { Just $4 }
463 -- taking a few liberties with the C-- syntax here; C-- doesn't have
464 -- 'default' branches
465 | {- empty -} { Nothing }
467 -- Note: OldCmm doesn't support a first class 'else' statement, though
470 : {- empty -} { nopEC }
471 | 'else' '{' body '}' { $3 }
473 -- we have to write this out longhand so that Happy's precedence rules
475 expr :: { ExtFCode CmmExpr }
476 : expr '/' expr { mkMachOp MO_U_Quot [$1,$3] }
477 | expr '*' expr { mkMachOp MO_Mul [$1,$3] }
478 | expr '%' expr { mkMachOp MO_U_Rem [$1,$3] }
479 | expr '-' expr { mkMachOp MO_Sub [$1,$3] }
480 | expr '+' expr { mkMachOp MO_Add [$1,$3] }
481 | expr '>>' expr { mkMachOp MO_U_Shr [$1,$3] }
482 | expr '<<' expr { mkMachOp MO_Shl [$1,$3] }
483 | expr '&' expr { mkMachOp MO_And [$1,$3] }
484 | expr '^' expr { mkMachOp MO_Xor [$1,$3] }
485 | expr '|' expr { mkMachOp MO_Or [$1,$3] }
486 | expr '>=' expr { mkMachOp MO_U_Ge [$1,$3] }
487 | expr '>' expr { mkMachOp MO_U_Gt [$1,$3] }
488 | expr '<=' expr { mkMachOp MO_U_Le [$1,$3] }
489 | expr '<' expr { mkMachOp MO_U_Lt [$1,$3] }
490 | expr '!=' expr { mkMachOp MO_Ne [$1,$3] }
491 | expr '==' expr { mkMachOp MO_Eq [$1,$3] }
492 | '~' expr { mkMachOp MO_Not [$2] }
493 | '-' expr { mkMachOp MO_S_Neg [$2] }
494 | expr0 '`' NAME '`' expr0 {% do { mo <- nameToMachOp $3 ;
495 return (mkMachOp mo [$1,$5]) } }
498 expr0 :: { ExtFCode CmmExpr }
499 : INT maybe_ty { return (CmmLit (CmmInt $1 (typeWidth $2))) }
500 | FLOAT maybe_ty { return (CmmLit (CmmFloat $1 (typeWidth $2))) }
501 | STRING { do s <- code (mkStringCLit $1);
504 | type '[' expr ']' { do e <- $3; return (CmmLoad e $1) }
505 | '%' NAME '(' exprs0 ')' {% exprOp $2 $4 }
506 | '(' expr ')' { $2 }
509 -- leaving out the type of a literal gives you the native word size in C--
510 maybe_ty :: { CmmType }
511 : {- empty -} { bWord }
514 maybe_actuals :: { [ExtFCode HintedCmmActual] }
516 | '(' cmm_hint_exprs0 ')' { $2 }
518 cmm_hint_exprs0 :: { [ExtFCode HintedCmmActual] }
520 | cmm_hint_exprs { $1 }
522 cmm_hint_exprs :: { [ExtFCode HintedCmmActual] }
523 : cmm_hint_expr { [$1] }
524 | cmm_hint_expr ',' cmm_hint_exprs { $1 : $3 }
526 cmm_hint_expr :: { ExtFCode HintedCmmActual }
527 : expr { do e <- $1; return (CmmHinted e (inferCmmHint e)) }
528 | expr STRING {% do h <- parseCmmHint $2;
530 e <- $1; return (CmmHinted e h) }
532 exprs0 :: { [ExtFCode CmmExpr] }
536 exprs :: { [ExtFCode CmmExpr] }
538 | expr ',' exprs { $1 : $3 }
540 reg :: { ExtFCode CmmExpr }
541 : NAME { lookupName $1 }
542 | GLOBALREG { return (CmmReg (CmmGlobal $1)) }
544 maybe_results :: { [ExtFCode HintedCmmFormal] }
546 | '(' cmm_formals ')' '=' { $2 }
548 cmm_formals :: { [ExtFCode HintedCmmFormal] }
549 : cmm_formal { [$1] }
550 | cmm_formal ',' { [$1] }
551 | cmm_formal ',' cmm_formals { $1 : $3 }
553 cmm_formal :: { ExtFCode HintedCmmFormal }
554 : local_lreg { do e <- $1; return (CmmHinted e (inferCmmHint (CmmReg (CmmLocal e)))) }
555 | STRING local_lreg {% do h <- parseCmmHint $1;
557 e <- $2; return (CmmHinted e h) }
559 local_lreg :: { ExtFCode LocalReg }
560 : NAME { do e <- lookupName $1;
563 CmmReg (CmmLocal r) -> r
564 other -> pprPanic "CmmParse:" (ftext $1 <> text " not a local register") }
566 lreg :: { ExtFCode CmmReg }
567 : NAME { do e <- lookupName $1;
571 other -> pprPanic "CmmParse:" (ftext $1 <> text " not a register") }
572 | GLOBALREG { return (CmmGlobal $1) }
574 maybe_formals_without_hints :: { [ExtFCode LocalReg] }
576 | '(' formals_without_hints0 ')' { $2 }
578 formals_without_hints0 :: { [ExtFCode LocalReg] }
580 | formals_without_hints { $1 }
582 formals_without_hints :: { [ExtFCode LocalReg] }
583 : formal_without_hint ',' { [$1] }
584 | formal_without_hint { [$1] }
585 | formal_without_hint ',' formals_without_hints { $1 : $3 }
587 formal_without_hint :: { ExtFCode LocalReg }
588 : type NAME { newLocal $1 $2 }
590 maybe_frame :: { ExtFCode (Maybe UpdateFrame) }
591 : {- empty -} { return Nothing }
592 | 'jump' expr '(' exprs0 ')' { do { target <- $2;
594 return $ Just (UpdateFrame target args) } }
596 maybe_gc_block :: { ExtFCode (Maybe BlockId) }
597 : {- empty -} { return Nothing }
599 { do l <- lookupLabel $2; return (Just l) }
605 typenot8 :: { CmmType }
613 section :: String -> Section
614 section "text" = Text
615 section "data" = Data
616 section "rodata" = ReadOnlyData
617 section "relrodata" = RelocatableReadOnlyData
618 section "bss" = UninitialisedData
619 section s = OtherSection s
621 mkString :: String -> CmmStatic
622 mkString s = CmmString (map (fromIntegral.ord) s)
624 -- mkMachOp infers the type of the MachOp from the type of its first
625 -- argument. We assume that this is correct: for MachOps that don't have
626 -- symmetrical args (e.g. shift ops), the first arg determines the type of
628 mkMachOp :: (Width -> MachOp) -> [ExtFCode CmmExpr] -> ExtFCode CmmExpr
629 mkMachOp fn args = do
630 arg_exprs <- sequence args
631 return (CmmMachOp (fn (typeWidth (cmmExprType (head arg_exprs)))) arg_exprs)
633 getLit :: CmmExpr -> CmmLit
634 getLit (CmmLit l) = l
635 getLit (CmmMachOp (MO_S_Neg _) [CmmLit (CmmInt i r)]) = CmmInt (negate i) r
636 getLit _ = panic "invalid literal" -- TODO messy failure
638 nameToMachOp :: FastString -> P (Width -> MachOp)
640 case lookupUFM machOps name of
641 Nothing -> fail ("unknown primitive " ++ unpackFS name)
644 exprOp :: FastString -> [ExtFCode CmmExpr] -> P (ExtFCode CmmExpr)
645 exprOp name args_code =
646 case lookupUFM exprMacros name of
647 Just f -> return $ do
648 args <- sequence args_code
651 mo <- nameToMachOp name
652 return $ mkMachOp mo args_code
654 exprMacros :: UniqFM ([CmmExpr] -> CmmExpr)
655 exprMacros = listToUFM [
656 ( fsLit "ENTRY_CODE", \ [x] -> entryCode x ),
657 ( fsLit "INFO_PTR", \ [x] -> closureInfoPtr x ),
658 ( fsLit "STD_INFO", \ [x] -> infoTable x ),
659 ( fsLit "FUN_INFO", \ [x] -> funInfoTable x ),
660 ( fsLit "GET_ENTRY", \ [x] -> entryCode (closureInfoPtr x) ),
661 ( fsLit "GET_STD_INFO", \ [x] -> infoTable (closureInfoPtr x) ),
662 ( fsLit "GET_FUN_INFO", \ [x] -> funInfoTable (closureInfoPtr x) ),
663 ( fsLit "INFO_TYPE", \ [x] -> infoTableClosureType x ),
664 ( fsLit "INFO_PTRS", \ [x] -> infoTablePtrs x ),
665 ( fsLit "INFO_NPTRS", \ [x] -> infoTableNonPtrs x )
668 -- we understand a subset of C-- primitives:
669 machOps = listToUFM $
670 map (\(x, y) -> (mkFastString x, y)) [
677 ( "quot", MO_S_Quot ),
679 ( "divu", MO_U_Quot ),
680 ( "modu", MO_U_Rem ),
698 ( "fneg", MO_S_Neg ),
705 ( "shrl", MO_U_Shr ),
706 ( "shra", MO_S_Shr ),
708 ( "lobits8", flip MO_UU_Conv W8 ),
709 ( "lobits16", flip MO_UU_Conv W16 ),
710 ( "lobits32", flip MO_UU_Conv W32 ),
711 ( "lobits64", flip MO_UU_Conv W64 ),
713 ( "zx16", flip MO_UU_Conv W16 ),
714 ( "zx32", flip MO_UU_Conv W32 ),
715 ( "zx64", flip MO_UU_Conv W64 ),
717 ( "sx16", flip MO_SS_Conv W16 ),
718 ( "sx32", flip MO_SS_Conv W32 ),
719 ( "sx64", flip MO_SS_Conv W64 ),
721 ( "f2f32", flip MO_FF_Conv W32 ), -- TODO; rounding mode
722 ( "f2f64", flip MO_FF_Conv W64 ), -- TODO; rounding mode
723 ( "f2i8", flip MO_FS_Conv W8 ),
724 ( "f2i16", flip MO_FS_Conv W16 ),
725 ( "f2i32", flip MO_FS_Conv W32 ),
726 ( "f2i64", flip MO_FS_Conv W64 ),
727 ( "i2f32", flip MO_SF_Conv W32 ),
728 ( "i2f64", flip MO_SF_Conv W64 )
731 callishMachOps = listToUFM $
732 map (\(x, y) -> (mkFastString x, y)) [
733 ( "write_barrier", MO_WriteBarrier )
734 -- ToDo: the rest, maybe
737 parseSafety :: String -> P CmmSafety
738 parseSafety "safe" = return (CmmSafe NoC_SRT)
739 parseSafety "unsafe" = return CmmUnsafe
740 parseSafety "interruptible" = return CmmInterruptible
741 parseSafety str = fail ("unrecognised safety: " ++ str)
743 parseCmmHint :: String -> P ForeignHint
744 parseCmmHint "ptr" = return AddrHint
745 parseCmmHint "signed" = return SignedHint
746 parseCmmHint str = fail ("unrecognised hint: " ++ str)
748 -- labels are always pointers, so we might as well infer the hint
749 inferCmmHint :: CmmExpr -> ForeignHint
750 inferCmmHint (CmmLit (CmmLabel _)) = AddrHint
751 inferCmmHint (CmmReg (CmmGlobal g)) | isPtrGlobalReg g = AddrHint
752 inferCmmHint _ = NoHint
754 isPtrGlobalReg Sp = True
755 isPtrGlobalReg SpLim = True
756 isPtrGlobalReg Hp = True
757 isPtrGlobalReg HpLim = True
758 isPtrGlobalReg CurrentTSO = True
759 isPtrGlobalReg CurrentNursery = True
760 isPtrGlobalReg (VanillaReg _ VGcPtr) = True
761 isPtrGlobalReg _ = False
764 happyError = srcParseFail
766 -- -----------------------------------------------------------------------------
767 -- Statement-level macros
769 stmtMacro :: FastString -> [ExtFCode CmmExpr] -> P ExtCode
770 stmtMacro fun args_code = do
771 case lookupUFM stmtMacros fun of
772 Nothing -> fail ("unknown macro: " ++ unpackFS fun)
773 Just fcode -> return $ do
774 args <- sequence args_code
777 stmtMacros :: UniqFM ([CmmExpr] -> Code)
778 stmtMacros = listToUFM [
779 ( fsLit "CCS_ALLOC", \[words,ccs] -> profAlloc words ccs ),
780 ( fsLit "CLOSE_NURSERY", \[] -> emitCloseNursery ),
781 ( fsLit "ENTER_CCS_PAP_CL", \[e] -> enterCostCentrePAP e ),
782 ( fsLit "ENTER_CCS_THUNK", \[e] -> enterCostCentreThunk e ),
783 ( fsLit "HP_CHK_GEN", \[words,liveness,reentry] ->
784 hpChkGen words liveness reentry ),
785 ( fsLit "HP_CHK_NP_ASSIGN_SP0", \[e,f] -> hpChkNodePointsAssignSp0 e f ),
786 ( fsLit "LOAD_THREAD_STATE", \[] -> emitLoadThreadState ),
787 ( fsLit "LDV_ENTER", \[e] -> ldvEnter e ),
788 ( fsLit "LDV_RECORD_CREATE", \[e] -> ldvRecordCreate e ),
789 ( fsLit "OPEN_NURSERY", \[] -> emitOpenNursery ),
790 ( fsLit "PUSH_UPD_FRAME", \[sp,e] -> emitPushUpdateFrame sp e ),
791 ( fsLit "SAVE_THREAD_STATE", \[] -> emitSaveThreadState ),
792 ( fsLit "SET_HDR", \[ptr,info,ccs] ->
793 emitSetDynHdr ptr info ccs ),
794 ( fsLit "STK_CHK_GEN", \[words,liveness,reentry] ->
795 stkChkGen words liveness reentry ),
796 ( fsLit "STK_CHK_NP", \[e] -> stkChkNodePoints e ),
797 ( fsLit "TICK_ALLOC_PRIM", \[hdr,goods,slop] ->
798 tickyAllocPrim hdr goods slop ),
799 ( fsLit "TICK_ALLOC_PAP", \[goods,slop] ->
800 tickyAllocPAP goods slop ),
801 ( fsLit "TICK_ALLOC_UP_THK", \[goods,slop] ->
802 tickyAllocThunk goods slop ),
803 ( fsLit "UPD_BH_UPDATABLE", \[] -> emitBlackHoleCode False ),
804 ( fsLit "UPD_BH_SINGLE_ENTRY", \[] -> emitBlackHoleCode True ),
806 ( fsLit "RET_P", \[a] -> emitRetUT [(PtrArg,a)]),
807 ( fsLit "RET_N", \[a] -> emitRetUT [(NonPtrArg,a)]),
808 ( fsLit "RET_PP", \[a,b] -> emitRetUT [(PtrArg,a),(PtrArg,b)]),
809 ( fsLit "RET_NN", \[a,b] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b)]),
810 ( fsLit "RET_NP", \[a,b] -> emitRetUT [(NonPtrArg,a),(PtrArg,b)]),
811 ( fsLit "RET_PPP", \[a,b,c] -> emitRetUT [(PtrArg,a),(PtrArg,b),(PtrArg,c)]),
812 ( fsLit "RET_NPP", \[a,b,c] -> emitRetUT [(NonPtrArg,a),(PtrArg,b),(PtrArg,c)]),
813 ( fsLit "RET_NNP", \[a,b,c] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(PtrArg,c)]),
814 ( fsLit "RET_NNN", \[a,b,c] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(NonPtrArg,c)]),
815 ( fsLit "RET_NNNN", \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(NonPtrArg,c),(NonPtrArg,d)]),
816 ( fsLit "RET_NNNP", \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(NonPtrArg,b),(NonPtrArg,c),(PtrArg,d)]),
817 ( fsLit "RET_NPNP", \[a,b,c,d] -> emitRetUT [(NonPtrArg,a),(PtrArg,b),(NonPtrArg,c),(PtrArg,d)])
823 profilingInfo desc_str ty_str = do
824 lit1 <- if opt_SccProfilingOn
825 then code $ mkStringCLit desc_str
826 else return (mkIntCLit 0)
827 lit2 <- if opt_SccProfilingOn
828 then code $ mkStringCLit ty_str
829 else return (mkIntCLit 0)
830 return (ProfilingInfo lit1 lit2)
833 staticClosure :: PackageId -> FastString -> FastString -> [CmmLit] -> ExtCode
834 staticClosure pkg cl_label info payload
835 = code $ emitDataLits (mkCmmDataLabel pkg cl_label) lits
836 where lits = mkStaticClosure (mkCmmInfoLabel pkg info) dontCareCCS payload [] [] []
840 -> [ExtFCode HintedCmmFormal]
842 -> [ExtFCode HintedCmmActual]
847 foreignCall conv_string results_code expr_code args_code vols safety ret
848 = do convention <- case conv_string of
849 "C" -> return CCallConv
850 "stdcall" -> return StdCallConv
851 "C--" -> return CmmCallConv
852 _ -> fail ("unknown calling convention: " ++ conv_string)
854 results <- sequence results_code
856 args <- sequence args_code
857 --code (stmtC (CmmCall (CmmCallee expr convention) results args safety))
859 -- Temporary hack so at least some functions are CmmSafe
860 CmmCallConv -> code (stmtC (CmmCall (CmmCallee expr convention) results args safety ret))
862 let expr' = adjCallTarget convention expr args in
865 code (emitForeignCall' PlayRisky results
866 (CmmCallee expr' convention) args vols NoC_SRT ret)
868 code (emitForeignCall' (PlaySafe unused) results
869 (CmmCallee expr' convention) args vols NoC_SRT ret) where
870 unused = panic "not used by emitForeignCall'"
872 code (emitForeignCall' PlayInterruptible results
873 (CmmCallee expr' convention) args vols NoC_SRT ret)
875 adjCallTarget :: CCallConv -> CmmExpr -> [CmmHinted CmmExpr] -> CmmExpr
876 #ifdef mingw32_TARGET_OS
877 -- On Windows, we have to add the '@N' suffix to the label when making
878 -- a call with the stdcall calling convention.
879 adjCallTarget StdCallConv (CmmLit (CmmLabel lbl)) args
880 = CmmLit (CmmLabel (addLabelSize lbl (sum (map size args))))
881 where size (CmmHinted e _) = max wORD_SIZE (widthInBytes (typeWidth (cmmExprType e)))
882 -- c.f. CgForeignCall.emitForeignCall
884 adjCallTarget _ expr _
888 :: [ExtFCode HintedCmmFormal]
890 -> [ExtFCode HintedCmmActual]
894 primCall results_code name args_code vols safety
895 = case lookupUFM callishMachOps name of
896 Nothing -> fail ("unknown primitive " ++ unpackFS name)
897 Just p -> return $ do
898 results <- sequence results_code
899 args <- sequence args_code
902 code (emitForeignCall' PlayRisky results
903 (CmmPrim p) args vols NoC_SRT CmmMayReturn)
905 code (emitForeignCall' (PlaySafe unused) results
906 (CmmPrim p) args vols NoC_SRT CmmMayReturn) where
907 unused = panic "not used by emitForeignCall'"
909 code (emitForeignCall' PlayInterruptible results
910 (CmmPrim p) args vols NoC_SRT CmmMayReturn)
912 doStore :: CmmType -> ExtFCode CmmExpr -> ExtFCode CmmExpr -> ExtCode
913 doStore rep addr_code val_code
914 = do addr <- addr_code
916 -- if the specified store type does not match the type of the expr
917 -- on the rhs, then we insert a coercion that will cause the type
918 -- mismatch to be flagged by cmm-lint. If we don't do this, then
919 -- the store will happen at the wrong type, and the error will not
921 let val_width = typeWidth (cmmExprType val)
922 rep_width = typeWidth rep
924 | val_width /= rep_width = CmmMachOp (MO_UU_Conv val_width rep_width) [val]
926 stmtEC (CmmStore addr coerce_val)
928 -- Return an unboxed tuple.
929 emitRetUT :: [(CgRep,CmmExpr)] -> Code
931 tickyUnboxedTupleReturn (length args) -- TICK
932 (sp, stmts) <- pushUnboxedTuple 0 args
933 emitSimultaneously stmts -- NB. the args might overlap with the stack slots
934 -- or regs that we assign to, so better use
935 -- simultaneous assignments here (#3546)
936 when (sp /= 0) $ stmtC (CmmAssign spReg (cmmRegOffW spReg (-sp)))
937 stmtC (CmmJump (entryCode (CmmLoad (cmmRegOffW spReg sp) bWord)) [])
938 -- TODO (when using CPS): emitStmt (CmmReturn (map snd args))
940 -- -----------------------------------------------------------------------------
941 -- If-then-else and boolean expressions
944 = BoolExpr `BoolAnd` BoolExpr
945 | BoolExpr `BoolOr` BoolExpr
949 -- ToDo: smart constructors which simplify the boolean expression.
951 cmmIfThenElse cond then_part else_part = do
952 then_id <- code newLabelC
953 join_id <- code newLabelC
957 stmtEC (CmmBranch join_id)
958 code (labelC then_id)
960 -- fall through to join
961 code (labelC join_id)
963 cmmRawIf cond then_id = do
967 -- 'emitCond cond true_id' emits code to test whether the cond is true,
968 -- branching to true_id if so, and falling through otherwise.
969 emitCond (BoolTest e) then_id = do
970 stmtEC (CmmCondBranch e then_id)
971 emitCond (BoolNot (BoolTest (CmmMachOp op args))) then_id
972 | Just op' <- maybeInvertComparison op
973 = emitCond (BoolTest (CmmMachOp op' args)) then_id
974 emitCond (BoolNot e) then_id = do
975 else_id <- code newLabelC
977 stmtEC (CmmBranch then_id)
978 code (labelC else_id)
979 emitCond (e1 `BoolOr` e2) then_id = do
982 emitCond (e1 `BoolAnd` e2) then_id = do
983 -- we'd like to invert one of the conditionals here to avoid an
984 -- extra branch instruction, but we can't use maybeInvertComparison
985 -- here because we can't look too closely at the expression since
987 and_id <- code newLabelC
988 else_id <- code newLabelC
990 stmtEC (CmmBranch else_id)
993 code (labelC else_id)
996 -- -----------------------------------------------------------------------------
999 -- We use a simplified form of C-- switch statements for now. A
1000 -- switch statement always compiles to a table jump. Each arm can
1001 -- specify a list of values (not ranges), and there can be a single
1002 -- default branch. The range of the table is given either by the
1003 -- optional range on the switch (eg. switch [0..7] {...}), or by
1004 -- the minimum/maximum values from the branches.
1006 doSwitch :: Maybe (Int,Int) -> ExtFCode CmmExpr -> [([Int],Either BlockId ExtCode)]
1007 -> Maybe ExtCode -> ExtCode
1008 doSwitch mb_range scrut arms deflt
1010 -- Compile code for the default branch
1013 Nothing -> return Nothing
1014 Just e -> do b <- forkLabelledCodeEC e; return (Just b)
1016 -- Compile each case branch
1017 table_entries <- mapM emitArm arms
1019 -- Construct the table
1021 all_entries = concat table_entries
1022 ixs = map fst all_entries
1024 | Just (l,u) <- mb_range = (l,u)
1025 | otherwise = (minimum ixs, maximum ixs)
1027 entries = elems (accumArray (\_ a -> Just a) dflt_entry (min,max)
1030 -- ToDo: check for out of range and jump to default if necessary
1031 stmtEC (CmmSwitch expr entries)
1033 emitArm :: ([Int],Either BlockId ExtCode) -> ExtFCode [(Int,BlockId)]
1034 emitArm (ints,Left blockid) = return [ (i,blockid) | i <- ints ]
1035 emitArm (ints,Right code) = do
1036 blockid <- forkLabelledCodeEC code
1037 return [ (i,blockid) | i <- ints ]
1039 -- -----------------------------------------------------------------------------
1040 -- Putting it all together
1042 -- The initial environment: we define some constants that the compiler
1043 -- knows about here.
1045 initEnv = listToUFM [
1046 ( fsLit "SIZEOF_StgHeader",
1047 Var (CmmLit (CmmInt (fromIntegral (fixedHdrSize * wORD_SIZE)) wordWidth) )),
1048 ( fsLit "SIZEOF_StgInfoTable",
1049 Var (CmmLit (CmmInt (fromIntegral stdInfoTableSizeB) wordWidth) ))
1052 parseCmmFile :: DynFlags -> FilePath -> IO (Messages, Maybe Cmm)
1053 parseCmmFile dflags filename = do
1054 showPass dflags "ParseCmm"
1055 buf <- hGetStringBuffer filename
1057 init_loc = mkSrcLoc (mkFastString filename) 1 1
1058 init_state = (mkPState dflags buf init_loc) { lex_state = [0] }
1059 -- reset the lex_state: the Lexer monad leaves some stuff
1060 -- in there we don't want.
1061 case unP cmmParse init_state of
1062 PFailed span err -> do
1063 let msg = mkPlainErrMsg span err
1064 return ((emptyBag, unitBag msg), Nothing)
1066 cmm <- initC dflags no_module (getCmm (unEC code initEnv [] >> return ()))
1067 let ms = getMessages pst
1068 if (errorsFound dflags ms)
1069 then return (ms, Nothing)
1071 dumpIfSet_dyn dflags Opt_D_dump_cmm "Cmm" (ppr cmm)
1072 return (ms, Just cmm)
1074 no_module = panic "parseCmmFile: no module"