2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section[SpecConstr]{Specialise over constructors}
11 #include "HsVersions.h"
16 import CoreUnfold ( couldBeSmallEnoughToInline )
17 import CoreLint ( showPass, endPass )
18 import CoreFVs ( exprsFreeVars )
19 import CoreTidy ( tidyRules )
20 import PprCore ( pprRules )
21 import WwLib ( mkWorkerArgs )
22 import DataCon ( dataConRepArity, dataConUnivTyVars )
23 import Type ( Type, tyConAppArgs )
24 import Coercion ( coercionKind )
25 import Id ( Id, idName, idType, isDataConWorkId_maybe,
26 mkUserLocal, mkSysLocal, idUnfolding, isLocalId )
30 import Name ( nameOccName, nameSrcLoc )
31 import Rules ( addIdSpecialisations, mkLocalRule, rulesOfBinds )
32 import OccName ( mkSpecOcc )
33 import ErrUtils ( dumpIfSet_dyn )
34 import DynFlags ( DynFlags(..), DynFlag(..) )
35 import BasicTypes ( Activation(..) )
36 import Maybes ( orElse, catMaybes )
38 import List ( nubBy, partition )
45 -----------------------------------------------------
47 -----------------------------------------------------
52 drop n (x:xs) = drop (n-1) xs
54 After the first time round, we could pass n unboxed. This happens in
55 numerical code too. Here's what it looks like in Core:
57 drop n xs = case xs of
62 _ -> drop (I# (n# -# 1#)) xs
64 Notice that the recursive call has an explicit constructor as argument.
65 Noticing this, we can make a specialised version of drop
67 RULE: drop (I# n#) xs ==> drop' n# xs
69 drop' n# xs = let n = I# n# in ...orig RHS...
71 Now the simplifier will apply the specialisation in the rhs of drop', giving
73 drop' n# xs = case xs of
77 _ -> drop (n# -# 1#) xs
81 We'd also like to catch cases where a parameter is carried along unchanged,
82 but evaluated each time round the loop:
84 f i n = if i>0 || i>n then i else f (i*2) n
86 Here f isn't strict in n, but we'd like to avoid evaluating it each iteration.
87 In Core, by the time we've w/wd (f is strict in i) we get
89 f i# n = case i# ># 0 of
91 True -> case n of n' { I# n# ->
94 True -> f (i# *# 2#) n'
96 At the call to f, we see that the argument, n is know to be (I# n#),
97 and n is evaluated elsewhere in the body of f, so we can play the same
103 We must be careful not to allocate the same constructor twice. Consider
104 f p = (...(case p of (a,b) -> e)...p...,
105 ...let t = (r,s) in ...t...(f t)...)
106 At the recursive call to f, we can see that t is a pair. But we do NOT want
107 to make a specialised copy:
108 f' a b = let p = (a,b) in (..., ...)
109 because now t is allocated by the caller, then r and s are passed to the
110 recursive call, which allocates the (r,s) pair again.
113 (a) the argument p is used in other than a case-scrutinsation way.
114 (b) the argument to the call is not a 'fresh' tuple; you have to
115 look into its unfolding to see that it's a tuple
117 Hence the "OR" part of Note [Good arguments] below.
119 ALTERNATIVE 2: pass both boxed and unboxed versions. This no longer saves
120 allocation, but does perhaps save evals. In the RULE we'd have
123 f (I# x#) = f' (I# x#) x#
125 If at the call site the (I# x) was an unfolding, then we'd have to
126 rely on CSE to eliminate the duplicate allocation.... This alternative
127 doesn't look attractive enough to pursue.
129 ALTERNATIVE 3: ignore the reboxing problem. The trouble is that
130 the conservative reboxing story prevents many useful functions from being
131 specialised. Example:
132 foo :: Maybe Int -> Int -> Int
134 foo x@(Just m) n = foo x (n-m)
135 Here the use of 'x' will clearly not require boxing in the specialised function.
137 The strictness analyser has the same problem, in fact. Example:
139 If we pass just 'a' and 'b' to the worker, it might need to rebox the
140 pair to create (a,b). A more sophisticated analysis might figure out
141 precisely the cases in which this could happen, but the strictness
142 analyser does no such analysis; it just passes 'a' and 'b', and hopes
145 So my current choice is to make SpecConstr similarly aggressive, and
146 ignore the bad potential of reboxing.
149 Note [Good arguments]
150 ~~~~~~~~~~~~~~~~~~~~~
153 * A self-recursive function. Ignore mutual recursion for now,
154 because it's less common, and the code is simpler for self-recursion.
158 a) At a recursive call, one or more parameters is an explicit
159 constructor application
161 That same parameter is scrutinised by a case somewhere in
162 the RHS of the function
166 b) At a recursive call, one or more parameters has an unfolding
167 that is an explicit constructor application
169 That same parameter is scrutinised by a case somewhere in
170 the RHS of the function
172 Those are the only uses of the parameter (see Note [Reboxing])
175 What to abstract over
176 ~~~~~~~~~~~~~~~~~~~~~
177 There's a bit of a complication with type arguments. If the call
180 f p = ...f ((:) [a] x xs)...
182 then our specialised function look like
184 f_spec x xs = let p = (:) [a] x xs in ....as before....
186 This only makes sense if either
187 a) the type variable 'a' is in scope at the top of f, or
188 b) the type variable 'a' is an argument to f (and hence fs)
190 Actually, (a) may hold for value arguments too, in which case
191 we may not want to pass them. Supose 'x' is in scope at f's
192 defn, but xs is not. Then we'd like
194 f_spec xs = let p = (:) [a] x xs in ....as before....
196 Similarly (b) may hold too. If x is already an argument at the
197 call, no need to pass it again.
199 Finally, if 'a' is not in scope at the call site, we could abstract
200 it as we do the term variables:
202 f_spec a x xs = let p = (:) [a] x xs in ...as before...
204 So the grand plan is:
206 * abstract the call site to a constructor-only pattern
207 e.g. C x (D (f p) (g q)) ==> C s1 (D s2 s3)
209 * Find the free variables of the abstracted pattern
211 * Pass these variables, less any that are in scope at
212 the fn defn. But see Note [Shadowing] below.
215 NOTICE that we only abstract over variables that are not in scope,
216 so we're in no danger of shadowing variables used in "higher up"
222 In this pass we gather up usage information that may mention variables
223 that are bound between the usage site and the definition site; or (more
224 seriously) may be bound to something different at the definition site.
227 f x = letrec g y v = let x = ...
230 Since 'x' is in scope at the call site, we may make a rewrite rule that
232 RULE forall a,b. g (a,b) x = ...
233 But this rule will never match, because it's really a different 'x' at
234 the call site -- and that difference will be manifest by the time the
235 simplifier gets to it. [A worry: the simplifier doesn't *guarantee*
236 no-shadowing, so perhaps it may not be distinct?]
238 Anyway, the rule isn't actually wrong, it's just not useful. One possibility
239 is to run deShadowBinds before running SpecConstr, but instead we run the
240 simplifier. That gives the simplest possible program for SpecConstr to
241 chew on; and it virtually guarantees no shadowing.
243 Note [Specialising for constant parameters]
244 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
245 This one is about specialising on a *constant* (but not necessarily
246 constructor) argument
248 foo :: Int -> (Int -> Int) -> Int
250 foo m f = foo (f m) (+1)
254 lvl_rmV :: GHC.Base.Int -> GHC.Base.Int
256 \ (ds_dlk :: GHC.Base.Int) ->
257 case ds_dlk of wild_alH { GHC.Base.I# x_alG ->
258 GHC.Base.I# (GHC.Prim.+# x_alG 1)
260 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
263 \ (ww_sme :: GHC.Prim.Int#) (w_smg :: GHC.Base.Int -> GHC.Base.Int) ->
264 case ww_sme of ds_Xlw {
266 case w_smg (GHC.Base.I# ds_Xlw) of w1_Xmo { GHC.Base.I# ww1_Xmz ->
267 T.$wfoo ww1_Xmz lvl_rmV
272 The recursive call has lvl_rmV as its argument, so we could create a specialised copy
273 with that argument baked in; that is, not passed at all. Now it can perhaps be inlined.
275 When is this worth it? Call the constant 'lvl'
276 - If 'lvl' has an unfolding that is a constructor, see if the corresponding
277 parameter is scrutinised anywhere in the body.
279 - If 'lvl' has an unfolding that is a inlinable function, see if the corresponding
280 parameter is applied (...to enough arguments...?)
282 Also do this is if the function has RULES?
286 Note [Specialising for lambda parameters]
287 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
288 foo :: Int -> (Int -> Int) -> Int
290 foo m f = foo (f m) (\n -> n-m)
292 This is subtly different from the previous one in that we get an
293 explicit lambda as the argument:
295 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
298 \ (ww_sm8 :: GHC.Prim.Int#) (w_sma :: GHC.Base.Int -> GHC.Base.Int) ->
299 case ww_sm8 of ds_Xlr {
301 case w_sma (GHC.Base.I# ds_Xlr) of w1_Xmf { GHC.Base.I# ww1_Xmq ->
304 (\ (n_ad3 :: GHC.Base.Int) ->
305 case n_ad3 of wild_alB { GHC.Base.I# x_alA ->
306 GHC.Base.I# (GHC.Prim.-# x_alA ds_Xlr)
312 I wonder if SpecConstr couldn't be extended to handle this? After all,
313 lambda is a sort of constructor for functions and perhaps it already
314 has most of the necessary machinery?
316 Furthermore, there's an immediate win, because you don't need to allocate the lamda
317 at the call site; and if perchance it's called in the recursive call, then you
318 may avoid allocating it altogether. Just like for constructors.
320 Looks cool, but probably rare...but it might be easy to implement.
323 Note [SpecConstr for casts]
324 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
327 data instance T Int = T Int
332 go (T n) = go (T (n-1))
334 The recursive call ends up looking like
335 go (T (I# ...) `cast` g)
336 So we want to spot the construtor application inside the cast.
337 That's why we have the Cast case in argToPat
340 -----------------------------------------------------
341 Stuff not yet handled
342 -----------------------------------------------------
344 Here are notes arising from Roman's work that I don't want to lose.
350 foo :: Int -> T Int -> Int
352 foo x t | even x = case t of { T n -> foo (x-n) t }
353 | otherwise = foo (x-1) t
355 SpecConstr does no specialisation, because the second recursive call
356 looks like a boxed use of the argument. A pity.
358 $wfoo_sFw :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
360 \ (ww_sFo [Just L] :: GHC.Prim.Int#) (w_sFq [Just L] :: T.T GHC.Base.Int) ->
361 case ww_sFo of ds_Xw6 [Just L] {
363 case GHC.Prim.remInt# ds_Xw6 2 of wild1_aEF [Dead Just A] {
364 __DEFAULT -> $wfoo_sFw (GHC.Prim.-# ds_Xw6 1) w_sFq;
366 case w_sFq of wild_Xy [Just L] { T.T n_ad5 [Just U(L)] ->
367 case n_ad5 of wild1_aET [Just A] { GHC.Base.I# y_aES [Just L] ->
368 $wfoo_sFw (GHC.Prim.-# ds_Xw6 y_aES) wild_Xy
374 data a :*: b = !a :*: !b
377 foo :: (Int :*: T Int) -> Int
379 foo (x :*: t) | even x = case t of { T n -> foo ((x-n) :*: t) }
380 | otherwise = foo ((x-1) :*: t)
382 Very similar to the previous one, except that the parameters are now in
383 a strict tuple. Before SpecConstr, we have
385 $wfoo_sG3 :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
387 \ (ww_sFU [Just L] :: GHC.Prim.Int#) (ww_sFW [Just L] :: T.T
389 case ww_sFU of ds_Xws [Just L] {
391 case GHC.Prim.remInt# ds_Xws 2 of wild1_aEZ [Dead Just A] {
393 case ww_sFW of tpl_B2 [Just L] { T.T a_sFo [Just A] ->
394 $wfoo_sG3 (GHC.Prim.-# ds_Xws 1) tpl_B2 -- $wfoo1
397 case ww_sFW of wild_XB [Just A] { T.T n_ad7 [Just S(L)] ->
398 case n_ad7 of wild1_aFd [Just L] { GHC.Base.I# y_aFc [Just L] ->
399 $wfoo_sG3 (GHC.Prim.-# ds_Xws y_aFc) wild_XB -- $wfoo2
403 We get two specialisations:
404 "SC:$wfoo1" [0] __forall {a_sFB :: GHC.Base.Int sc_sGC :: GHC.Prim.Int#}
405 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int a_sFB)
406 = Foo.$s$wfoo1 a_sFB sc_sGC ;
407 "SC:$wfoo2" [0] __forall {y_aFp :: GHC.Prim.Int# sc_sGC :: GHC.Prim.Int#}
408 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int (GHC.Base.I# y_aFp))
409 = Foo.$s$wfoo y_aFp sc_sGC ;
411 But perhaps the first one isn't good. After all, we know that tpl_B2 is
412 a T (I# x) really, because T is strict and Int has one constructor. (We can't
413 unbox the strict fields, becuase T is polymorphic!)
417 %************************************************************************
419 \subsection{Top level wrapper stuff}
421 %************************************************************************
424 specConstrProgram :: DynFlags -> UniqSupply -> [CoreBind] -> IO [CoreBind]
425 specConstrProgram dflags us binds
427 showPass dflags "SpecConstr"
429 let (binds', _) = initUs us (go (initScEnv dflags) binds)
431 endPass dflags "SpecConstr" Opt_D_dump_spec binds'
433 dumpIfSet_dyn dflags Opt_D_dump_rules "Top-level specialisations"
434 (pprRules (tidyRules emptyTidyEnv (rulesOfBinds binds')))
438 go env [] = returnUs []
439 go env (bind:binds) = scBind env bind `thenUs` \ (env', _, bind') ->
440 go env' binds `thenUs` \ binds' ->
441 returnUs (bind' : binds')
445 %************************************************************************
447 \subsection{Environment: goes downwards}
449 %************************************************************************
452 data ScEnv = SCE { sc_size :: Int, -- Size threshold
454 sc_subst :: Subst, -- Current subsitution
456 sc_how_bound :: HowBoundEnv,
457 -- Binds interesting non-top-level variables
458 -- Look up in here *after* applying the substitution
461 -- Look up in here *after* applying the substitution
464 type HowBoundEnv = VarEnv HowBound
466 type ConstrEnv = IdEnv ConValue
467 data ConValue = CV AltCon [CoreArg]
468 -- Variables known to be bound to a constructor
469 -- in a particular case alternative
472 instance Outputable ConValue where
473 ppr (CV con args) = ppr con <+> interpp'SP args
476 = SCE { sc_size = specThreshold dflags,
477 sc_subst = emptySubst,
478 sc_how_bound = emptyVarEnv,
479 sc_cons = emptyVarEnv }
481 data HowBound = RecFun -- These are the recursive functions for which
482 -- we seek interesting call patterns
484 | RecArg -- These are those functions' arguments, or their sub-components;
485 -- we gather occurrence information for these
487 instance Outputable HowBound where
488 ppr RecFun = text "RecFun"
489 ppr RecArg = text "RecArg"
491 lookupHowBound :: ScEnv -> Id -> Maybe HowBound
492 lookupHowBound env id = lookupVarEnv (sc_how_bound env) id
494 scSubstId :: ScEnv -> Id -> CoreExpr
495 scSubstId env v = lookupIdSubst (sc_subst env) v
497 scSubstTy :: ScEnv -> Type -> Type
498 scSubstTy env ty = substTy (sc_subst env) ty
500 zapScSubst :: ScEnv -> ScEnv
501 zapScSubst env = env { sc_subst = zapSubstEnv (sc_subst env) }
503 extendScInScope :: ScEnv -> [Var] -> ScEnv
504 -- Bring the quantified variables into scope
505 extendScInScope env qvars = env { sc_subst = extendInScopeList (sc_subst env) qvars }
507 extendScSubst :: ScEnv -> [(Var,CoreArg)] -> ScEnv
508 -- Extend the substitution
509 extendScSubst env prs = env { sc_subst = extendSubstList (sc_subst env) prs }
511 extendHowBound :: ScEnv -> [Var] -> HowBound -> ScEnv
512 extendHowBound env bndrs how_bound
513 = env { sc_how_bound = extendVarEnvList (sc_how_bound env)
514 [(bndr,how_bound) | bndr <- bndrs] }
516 extendBndrsWith :: HowBound -> ScEnv -> [Var] -> (ScEnv, [Var])
517 extendBndrsWith how_bound env bndrs
518 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndrs')
520 (subst', bndrs') = substBndrs (sc_subst env) bndrs
521 hb_env' = sc_how_bound env `extendVarEnvList`
522 [(bndr,how_bound) | bndr <- bndrs']
524 extendBndrWith :: HowBound -> ScEnv -> Var -> (ScEnv, Var)
525 extendBndrWith how_bound env bndr
526 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndr')
528 (subst', bndr') = substBndr (sc_subst env) bndr
529 hb_env' = extendVarEnv (sc_how_bound env) bndr' how_bound
531 extendRecBndrs :: ScEnv -> [Var] -> (ScEnv, [Var])
532 extendRecBndrs env bndrs = (env { sc_subst = subst' }, bndrs')
534 (subst', bndrs') = substRecBndrs (sc_subst env) bndrs
536 extendBndr :: ScEnv -> Var -> (ScEnv, Var)
537 extendBndr env bndr = (env { sc_subst = subst' }, bndr')
539 (subst', bndr') = substBndr (sc_subst env) bndr
541 extendConEnv :: ScEnv -> Id -> Maybe ConValue -> ScEnv
542 extendConEnv env id Nothing = env
543 extendConEnv env id (Just cv) = env { sc_cons = extendVarEnv (sc_cons env) id cv }
545 extendCaseBndrs :: ScEnv -> CoreExpr -> Id -> AltCon -> [Var] -> ScEnv
549 -- we want to bind b, and perhaps scrut too, to (C x y)
550 -- NB: Extends only the sc_cons part of the envt
551 extendCaseBndrs env scrut case_bndr con alt_bndrs
553 Var v -> extendConEnv env1 v cval
556 env1 = extendConEnv env case_bndr cval
559 LitAlt lit -> Just (CV con [])
560 DataAlt dc -> Just (CV con vanilla_args)
562 vanilla_args = map Type (tyConAppArgs (idType case_bndr)) ++
563 varsToCoreExprs alt_bndrs
567 %************************************************************************
569 \subsection{Usage information: flows upwards}
571 %************************************************************************
576 calls :: CallEnv, -- Calls
577 -- The functions are a subset of the
578 -- RecFuns in the ScEnv
580 occs :: !(IdEnv ArgOcc) -- Information on argument occurrences
581 } -- The variables are a subset of the
582 -- RecArg in the ScEnv
584 type CallEnv = IdEnv [Call]
585 type Call = (ConstrEnv, [CoreArg])
586 -- The arguments of the call, together with the
587 -- env giving the constructor bindings at the call site
589 nullUsage = SCU { calls = emptyVarEnv, occs = emptyVarEnv }
591 combineCalls :: CallEnv -> CallEnv -> CallEnv
592 combineCalls = plusVarEnv_C (++)
594 combineUsage u1 u2 = SCU { calls = combineCalls (calls u1) (calls u2),
595 occs = plusVarEnv_C combineOcc (occs u1) (occs u2) }
597 combineUsages [] = nullUsage
598 combineUsages us = foldr1 combineUsage us
600 lookupOcc :: ScUsage -> Var -> (ScUsage, ArgOcc)
601 lookupOcc (SCU { calls = sc_calls, occs = sc_occs }) bndr
602 = (SCU {calls = sc_calls, occs = delVarEnv sc_occs bndr},
603 lookupVarEnv sc_occs bndr `orElse` NoOcc)
605 lookupOccs :: ScUsage -> [Var] -> (ScUsage, [ArgOcc])
606 lookupOccs (SCU { calls = sc_calls, occs = sc_occs }) bndrs
607 = (SCU {calls = sc_calls, occs = delVarEnvList sc_occs bndrs},
608 [lookupVarEnv sc_occs b `orElse` NoOcc | b <- bndrs])
610 data ArgOcc = NoOcc -- Doesn't occur at all; or a type argument
611 | UnkOcc -- Used in some unknown way
613 | ScrutOcc (UniqFM [ArgOcc]) -- See Note [ScrutOcc]
615 | BothOcc -- Definitely taken apart, *and* perhaps used in some other way
619 An occurrence of ScrutOcc indicates that the thing, or a `cast` version of the thing,
620 is *only* taken apart or applied.
622 Functions, literal: ScrutOcc emptyUFM
623 Data constructors: ScrutOcc subs,
625 where (subs :: UniqFM [ArgOcc]) gives usage of the *pattern-bound* components,
626 The domain of the UniqFM is the Unique of the data constructor
628 The [ArgOcc] is the occurrences of the *pattern-bound* components
629 of the data structure. E.g.
630 data T a = forall b. MkT a b (b->a)
631 A pattern binds b, x::a, y::b, z::b->a, but not 'a'!
635 instance Outputable ArgOcc where
636 ppr (ScrutOcc xs) = ptext SLIT("scrut-occ") <> ppr xs
637 ppr UnkOcc = ptext SLIT("unk-occ")
638 ppr BothOcc = ptext SLIT("both-occ")
639 ppr NoOcc = ptext SLIT("no-occ")
641 -- Experimentally, this vesion of combineOcc makes ScrutOcc "win", so
642 -- that if the thing is scrutinised anywhere then we get to see that
643 -- in the overall result, even if it's also used in a boxed way
644 -- This might be too agressive; see Note [Reboxing] Alternative 3
645 combineOcc NoOcc occ = occ
646 combineOcc occ NoOcc = occ
647 combineOcc (ScrutOcc xs) (ScrutOcc ys) = ScrutOcc (plusUFM_C combineOccs xs ys)
648 combineOcc occ (ScrutOcc ys) = ScrutOcc ys
649 combineOcc (ScrutOcc xs) occ = ScrutOcc xs
650 combineOcc UnkOcc UnkOcc = UnkOcc
651 combineOcc _ _ = BothOcc
653 combineOccs :: [ArgOcc] -> [ArgOcc] -> [ArgOcc]
654 combineOccs xs ys = zipWithEqual "combineOccs" combineOcc xs ys
656 setScrutOcc :: ScEnv -> ScUsage -> CoreExpr -> ArgOcc -> ScUsage
657 -- *Overwrite* the occurrence info for the scrutinee, if the scrutinee
658 -- is a variable, and an interesting variable
659 setScrutOcc env usg (Cast e _) occ = setScrutOcc env usg e occ
660 setScrutOcc env usg (Note _ e) occ = setScrutOcc env usg e occ
661 setScrutOcc env usg (Var v) occ
662 | Just RecArg <- lookupHowBound env v = usg { occs = extendVarEnv (occs usg) v occ }
664 setScrutOcc env usg other occ -- Catch-all
667 conArgOccs :: ArgOcc -> AltCon -> [ArgOcc]
668 -- Find usage of components of data con; returns [UnkOcc...] if unknown
669 -- See Note [ScrutOcc] for the extra UnkOccs in the vanilla datacon case
671 conArgOccs (ScrutOcc fm) (DataAlt dc)
672 | Just pat_arg_occs <- lookupUFM fm dc
673 = [UnkOcc | tv <- dataConUnivTyVars dc] ++ pat_arg_occs
675 conArgOccs other con = repeat UnkOcc
678 %************************************************************************
680 \subsection{The main recursive function}
682 %************************************************************************
684 The main recursive function gathers up usage information, and
685 creates specialised versions of functions.
688 scExpr :: ScEnv -> CoreExpr -> UniqSM (ScUsage, CoreExpr)
689 -- The unique supply is needed when we invent
690 -- a new name for the specialised function and its args
692 scExpr env e = scExpr' env e
695 scExpr' env (Var v) = case scSubstId env v of
696 Var v' -> returnUs (varUsage env v UnkOcc, Var v')
697 e' -> scExpr (zapScSubst env) e'
699 scExpr' env e@(Type t) = returnUs (nullUsage, Type (scSubstTy env t))
700 scExpr' env e@(Lit l) = returnUs (nullUsage, e)
701 scExpr' env (Note n e) = do { (usg,e') <- scExpr env e
702 ; return (usg, Note n e') }
703 scExpr' env (Cast e co) = do { (usg, e') <- scExpr env e
704 ; return (usg, Cast e' (scSubstTy env co)) }
705 scExpr' env (Lam b e) = do { let (env', b') = extendBndr env b
706 ; (usg, e') <- scExpr env' e
707 ; return (usg, Lam b' e') }
709 scExpr' env (Case scrut b ty alts)
710 = do { (scrut_usg, scrut') <- scExpr env scrut
711 ; case isConApp (sc_cons env) scrut' of
712 Nothing -> sc_vanilla scrut_usg scrut'
713 Just cval -> sc_con_app cval scrut'
716 sc_con_app cval@(CV con args) scrut' -- Known constructor; simplify
717 = do { let (_, bs, rhs) = findAlt con alts
718 alt_env' = extendScSubst env ((b,scrut') : bs `zip` trimConArgs con args)
719 ; scExpr alt_env' rhs }
722 sc_vanilla scrut_usg scrut' -- Normal case
723 = do { let (alt_env,b') = extendBndrWith RecArg env b
724 -- Record RecArg for the components
726 ; (alt_usgs, alt_occs, alts')
727 <- mapAndUnzip3Us (sc_alt alt_env scrut' b') alts
729 ; let (alt_usg, b_occ) = lookupOcc (combineUsages alt_usgs) b
730 scrut_occ = foldr combineOcc b_occ alt_occs
731 scrut_usg' = setScrutOcc env scrut_usg scrut' scrut_occ
732 -- The combined usage of the scrutinee is given
733 -- by scrut_occ, which is passed to scScrut, which
734 -- in turn treats a bare-variable scrutinee specially
736 ; return (alt_usg `combineUsage` scrut_usg',
737 Case scrut' b' (scSubstTy env ty) alts') }
739 sc_alt env scrut' b' (con,bs,rhs)
740 = do { let (env1, bs') = extendBndrsWith RecArg env bs
741 env2 = extendCaseBndrs env1 scrut' b' con bs'
742 ; (usg,rhs') <- scExpr env2 rhs
743 ; let (usg', arg_occs) = lookupOccs usg bs
744 scrut_occ = case con of
745 DataAlt dc -> ScrutOcc (unitUFM dc arg_occs)
746 other -> ScrutOcc emptyUFM
747 ; return (usg', scrut_occ, (con,bs',rhs')) }
749 scExpr' env (Let (NonRec bndr rhs) body)
750 = do { (rhs_usg, rhs_info@(_, args', rhs_body', _)) <- scRecRhs env (bndr,rhs)
751 ; if null args' || isEmptyVarEnv (calls rhs_usg) then do
753 let rhs' = mkLams args' rhs_body'
754 (body_env, bndr') = extendBndr env bndr
755 body_env2 = extendConEnv body_env bndr' (isConApp (sc_cons env) rhs')
756 -- Record if the RHS is a constructor
757 ; (body_usg, body') <- scExpr body_env2 body
758 ; return (body_usg `combineUsage` rhs_usg, Let (NonRec bndr' rhs') body') }
760 do { -- Join-point case
761 let (body_env, bndr') = extendBndrWith RecFun env bndr
762 -- If the RHS of this 'let' contains calls
763 -- to recursive functions that we're trying
764 -- to specialise, then treat this let too
765 -- as one to specialise
766 ; (body_usg, body') <- scExpr body_env body
768 ; (spec_usg, _, specs) <- specialise env (calls body_usg) ([], rhs_info)
770 ; return (body_usg { calls = calls body_usg `delVarEnv` bndr' }
771 `combineUsage` rhs_usg `combineUsage` spec_usg,
772 mkLets [NonRec b r | (b,r) <- addRules rhs_info specs] body')
775 scExpr' env (Let (Rec prs) body)
776 = do { (env', bind_usg, bind') <- scBind env (Rec prs)
777 ; (body_usg, body') <- scExpr env' body
778 ; return (bind_usg `combineUsage` body_usg, Let bind' body') }
780 scExpr' env e@(App _ _)
781 = do { let (fn, args) = collectArgs e
782 ; (fn_usg, fn') <- scExpr env fn
783 -- Process the function too. It's almost always a variable,
784 -- but not always. In particular, if this pass follows float-in,
785 -- which it may, we can get
786 -- (let f = ...f... in f) arg1 arg2
787 -- Also the substitution may replace a variable by a non-variable
789 ; let fn_usg' = setScrutOcc env fn_usg fn' (ScrutOcc emptyUFM)
790 -- We use setScrutOcc to record the fact that the function is called
791 -- Perhaps we should check that it has at least one value arg,
792 -- but currently we don't bother
794 ; (arg_usgs, args') <- mapAndUnzipUs (scExpr env) args
795 ; let call_usg = case fn' of
796 Var f | Just RecFun <- lookupHowBound env f
797 , not (null args) -- Not a proper call!
798 -> SCU { calls = unitVarEnv f [(sc_cons env, args')],
801 ; return (combineUsages arg_usgs `combineUsage` fn_usg'
802 `combineUsage` call_usg,
806 ----------------------
807 scBind :: ScEnv -> CoreBind -> UniqSM (ScEnv, ScUsage, CoreBind)
809 | not (all (couldBeSmallEnoughToInline (sc_size env)) rhss)
811 = do { let (rhs_env,bndrs') = extendRecBndrs env bndrs
812 ; (rhs_usgs, rhss') <- mapAndUnzipUs (scExpr rhs_env) rhss
813 ; return (rhs_env, combineUsages rhs_usgs, Rec (bndrs' `zip` rhss')) }
814 | otherwise -- Do specialisation
815 = do { let (rhs_env1,bndrs') = extendRecBndrs env bndrs
816 rhs_env2 = extendHowBound rhs_env1 bndrs RecFun
818 ; (rhs_usgs, rhs_infos) <- mapAndUnzipUs (scRecRhs rhs_env2) (bndrs' `zip` rhss)
819 ; let rhs_usg = combineUsages rhs_usgs
821 ; (spec_usg, specs) <- spec_loop rhs_env2 (calls rhs_usg)
822 (repeat [] `zip` rhs_infos)
824 ; let all_usg = rhs_usg `combineUsage` spec_usg
826 ; return (rhs_env1, -- For the body of the letrec, delete the RecFun business
827 all_usg { calls = calls rhs_usg `delVarEnvList` bndrs' },
828 Rec (concat (zipWith addRules rhs_infos specs))) }
830 (bndrs,rhss) = unzip prs
834 -> [([CallPat], RhsInfo)] -- One per binder
835 -> UniqSM (ScUsage, [[SpecInfo]]) -- One list per binder
836 spec_loop env all_calls rhs_stuff
837 = do { (spec_usg_s, new_pats_s, specs) <- mapAndUnzip3Us (specialise env all_calls) rhs_stuff
838 ; let spec_usg = combineUsages spec_usg_s
839 ; if all null new_pats_s then
840 return (spec_usg, specs) else do
841 { (spec_usg1, specs1) <- spec_loop env (calls spec_usg)
842 (zipWith add_pats new_pats_s rhs_stuff)
843 ; return (spec_usg `combineUsage` spec_usg1, zipWith (++) specs specs1) } }
845 add_pats :: [CallPat] -> ([CallPat], RhsInfo) -> ([CallPat], RhsInfo)
846 add_pats new_pats (done_pats, rhs_info) = (done_pats ++ new_pats, rhs_info)
848 scBind env (NonRec bndr rhs)
849 = do { (usg, rhs') <- scExpr env rhs
850 ; let (env', bndr') = extendBndr env bndr
851 ; return (env', usg, NonRec bndr' rhs') }
853 ----------------------
854 scRecRhs :: ScEnv -> (Id,CoreExpr) -> UniqSM (ScUsage, RhsInfo)
855 scRecRhs env (bndr,rhs)
856 = do { let (arg_bndrs,body) = collectBinders rhs
857 (body_env, arg_bndrs') = extendBndrsWith RecArg env arg_bndrs
858 ; (body_usg, body') <- scExpr body_env body
859 ; let (rhs_usg, arg_occs) = lookupOccs body_usg arg_bndrs'
860 ; return (rhs_usg, (bndr, arg_bndrs', body', arg_occs)) }
862 -- The arg_occs says how the visible,
863 -- lambda-bound binders of the RHS are used
864 -- (including the TyVar binders)
865 -- Two pats are the same if they match both ways
867 ----------------------
868 addRules :: RhsInfo -> [SpecInfo] -> [(Id,CoreExpr)]
869 addRules (fn, args, body, _) specs
870 = [(id,rhs) | (_,id,rhs) <- specs] ++
871 [(fn `addIdSpecialisations` rules, mkLams args body)]
873 rules = [r | (r,_,_) <- specs]
875 ----------------------
877 | Just RecArg <- lookupHowBound env v = SCU { calls = emptyVarEnv,
878 occs = unitVarEnv v use }
879 | otherwise = nullUsage
883 %************************************************************************
885 The specialiser itself
887 %************************************************************************
890 type RhsInfo = (Id, [Var], CoreExpr, [ArgOcc])
891 -- Info about the *original* RHS of a binding we are specialising
892 -- Original binding f = \xs.body
893 -- Plus info about usage of arguments
895 type SpecInfo = (CoreRule, Var, CoreExpr)
896 -- One specialisation: Rule plus definition
901 -> CallEnv -- Info on calls
902 -> ([CallPat], RhsInfo) -- Original RHS plus patterns dealt with
903 -> UniqSM (ScUsage, [CallPat], [SpecInfo]) -- Specialised calls
905 -- Note: the rhs here is the optimised version of the original rhs
906 -- So when we make a specialised copy of the RHS, we're starting
907 -- from an RHS whose nested functions have been optimised already.
909 specialise env bind_calls (done_pats, (fn, arg_bndrs, body, arg_occs))
910 | notNull arg_bndrs, -- Only specialise functions
911 Just all_calls <- lookupVarEnv bind_calls fn
912 = do { pats <- callsToPats env done_pats arg_occs all_calls
913 -- ; pprTrace "specialise" (vcat [ppr fn <+> ppr arg_occs,
914 -- text "calls" <+> ppr all_calls,
915 -- text "good pats" <+> ppr pats]) $
918 ; (spec_usgs, specs) <- mapAndUnzipUs (spec_one env fn arg_bndrs body)
919 (pats `zip` [length done_pats..])
921 ; return (combineUsages spec_usgs, pats, specs) }
923 = return (nullUsage, [], []) -- The boring case
926 ---------------------
929 -> [Var] -- Lambda-binders of RHS; should match patterns
930 -> CoreExpr -- Body of the original function
931 -> (([Var], [CoreArg]), Int)
932 -> UniqSM (ScUsage, SpecInfo) -- Rule and binding
934 -- spec_one creates a specialised copy of the function, together
935 -- with a rule for using it. I'm very proud of how short this
936 -- function is, considering what it does :-).
942 f = /\b \y::[(a,b)] -> ....f (b,c) ((:) (a,(b,c)) (x,v) (h w))...
943 [c::*, v::(b,c) are presumably bound by the (...) part]
945 f_spec = /\ b c \ v::(b,c) hw::[(a,(b,c))] ->
946 (...entire body of f...) [b -> (b,c),
947 y -> ((:) (a,(b,c)) (x,v) hw)]
949 RULE: forall b::* c::*, -- Note, *not* forall a, x
953 f (b,c) ((:) (a,(b,c)) (x,v) hw) = f_spec b c v hw
956 spec_one env fn arg_bndrs body ((qvars, pats), rule_number)
957 = do { -- Specialise the body
958 let spec_env = extendScSubst (extendScInScope env qvars)
959 (arg_bndrs `zip` pats)
960 ; (spec_usg, spec_body) <- scExpr spec_env body
962 -- ; pprTrace "spec_one" (ppr fn <+> vcat [text "pats" <+> ppr pats,
963 -- text "calls" <+> (ppr (calls spec_usg))])
966 -- And build the results
967 ; spec_uniq <- getUniqueUs
968 ; let (spec_lam_args, spec_call_args) = mkWorkerArgs qvars body_ty
969 -- Usual w/w hack to avoid generating
970 -- a spec_rhs of unlifted type and no args
973 fn_loc = nameSrcLoc fn_name
974 spec_occ = mkSpecOcc (nameOccName fn_name)
975 rule_name = mkFastString ("SC:" ++ showSDoc (ppr fn <> int rule_number))
976 spec_rhs = mkLams spec_lam_args spec_body
977 spec_id = mkUserLocal spec_occ spec_uniq (mkPiTypes spec_lam_args body_ty) fn_loc
978 body_ty = exprType spec_body
979 rule_rhs = mkVarApps (Var spec_id) spec_call_args
980 rule = mkLocalRule rule_name specConstrActivation fn_name qvars pats rule_rhs
981 ; return (spec_usg, (rule, spec_id, spec_rhs)) }
983 -- In which phase should the specialise-constructor rules be active?
984 -- Originally I made them always-active, but Manuel found that
985 -- this defeated some clever user-written rules. So Plan B
986 -- is to make them active only in Phase 0; after all, currently,
987 -- the specConstr transformation is only run after the simplifier
988 -- has reached Phase 0. In general one would want it to be
989 -- flag-controllable, but for now I'm leaving it baked in
991 specConstrActivation :: Activation
992 specConstrActivation = ActiveAfter 0 -- Baked in; see comments above
995 %************************************************************************
997 \subsection{Argument analysis}
999 %************************************************************************
1001 This code deals with analysing call-site arguments to see whether
1002 they are constructor applications.
1006 type CallPat = ([Var], [CoreExpr]) -- Quantified variables and arguments
1009 callsToPats :: ScEnv -> [CallPat] -> [ArgOcc] -> [Call] -> UniqSM [CallPat]
1010 -- Result has no duplicate patterns,
1011 -- nor ones mentioned in done_pats
1012 callsToPats env done_pats bndr_occs calls
1013 = do { mb_pats <- mapM (callToPats env bndr_occs) calls
1015 ; let good_pats :: [([Var], [CoreArg])]
1016 good_pats = catMaybes mb_pats
1017 is_done p = any (samePat p) done_pats
1019 ; return (filterOut is_done (nubBy samePat good_pats)) }
1021 callToPats :: ScEnv -> [ArgOcc] -> Call -> UniqSM (Maybe CallPat)
1022 -- The [Var] is the variables to quantify over in the rule
1023 -- Type variables come first, since they may scope
1024 -- over the following term variables
1025 -- The [CoreExpr] are the argument patterns for the rule
1026 callToPats env bndr_occs (con_env, args)
1027 | length args < length bndr_occs -- Check saturated
1030 = do { let in_scope = substInScope (sc_subst env)
1031 ; prs <- argsToPats in_scope con_env (args `zip` bndr_occs)
1032 ; let (good_pats, pats) = unzip prs
1033 pat_fvs = varSetElems (exprsFreeVars pats)
1034 qvars = filterOut (`elemInScopeSet` in_scope) pat_fvs
1035 -- Quantify over variables that are not in sccpe
1037 -- See Note [Shadowing] at the top
1039 (tvs, ids) = partition isTyVar qvars
1041 -- Put the type variables first; the type of a term
1042 -- variable may mention a type variable
1044 ; -- pprTrace "callToPats" (ppr args $$ ppr prs $$ ppr bndr_occs) $
1046 then return (Just (qvars', pats))
1047 else return Nothing }
1049 -- argToPat takes an actual argument, and returns an abstracted
1050 -- version, consisting of just the "constructor skeleton" of the
1051 -- argument, with non-constructor sub-expression replaced by new
1052 -- placeholder variables. For example:
1053 -- C a (D (f x) (g y)) ==> C p1 (D p2 p3)
1055 argToPat :: InScopeSet -- What's in scope at the fn defn site
1056 -> ConstrEnv -- ConstrEnv at the call site
1057 -> CoreArg -- A call arg (or component thereof)
1059 -> UniqSM (Bool, CoreArg)
1060 -- Returns (interesting, pat),
1061 -- where pat is the pattern derived from the argument
1062 -- intersting=True if the pattern is non-trivial (not a variable or type)
1063 -- E.g. x:xs --> (True, x:xs)
1064 -- f xs --> (False, w) where w is a fresh wildcard
1065 -- (f xs, 'c') --> (True, (w, 'c')) where w is a fresh wildcard
1066 -- \x. x+y --> (True, \x. x+y)
1067 -- lvl7 --> (True, lvl7) if lvl7 is bound
1068 -- somewhere further out
1070 argToPat in_scope con_env arg@(Type ty) arg_occ
1071 = return (False, arg)
1073 argToPat in_scope con_env (Note n arg) arg_occ
1074 = argToPat in_scope con_env arg arg_occ
1075 -- Note [Notes in call patterns]
1076 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1077 -- Ignore Notes. In particular, we want to ignore any InlineMe notes
1078 -- Perhaps we should not ignore profiling notes, but I'm going to
1079 -- ride roughshod over them all for now.
1080 --- See Note [Notes in RULE matching] in Rules
1082 argToPat in_scope con_env (Let _ arg) arg_occ
1083 = argToPat in_scope con_env arg arg_occ
1084 -- Look through let expressions
1085 -- e.g. f (let v = rhs in \y -> ...v...)
1086 -- Here we can specialise for f (\y -> ...)
1087 -- because the rule-matcher will look through the let.
1089 argToPat in_scope con_env (Cast arg co) arg_occ
1090 = do { (interesting, arg') <- argToPat in_scope con_env arg arg_occ
1091 ; if interesting then
1092 return (interesting, Cast arg' co)
1094 wildCardPat (snd (coercionKind co)) }
1096 {- Disabling lambda specialisation for now
1097 It's fragile, and the spec_loop can be infinite
1098 argToPat in_scope con_env arg arg_occ
1100 = return (True, arg)
1102 is_value_lam (Lam v e) -- Spot a value lambda, even if
1103 | isId v = True -- it is inside a type lambda
1104 | otherwise = is_value_lam e
1105 is_value_lam other = False
1108 -- Check for a constructor application
1109 -- NB: this *precedes* the Var case, so that we catch nullary constrs
1110 argToPat in_scope con_env arg arg_occ
1111 | Just (CV dc args) <- isConApp con_env arg
1113 ScrutOcc _ -> True -- Used only by case scrutinee
1114 BothOcc -> case arg of -- Used elsewhere
1115 App {} -> True -- see Note [Reboxing]
1117 other -> False -- No point; the arg is not decomposed
1118 = do { args' <- argsToPats in_scope con_env (args `zip` conArgOccs arg_occ dc)
1119 ; return (True, mk_con_app dc (map snd args')) }
1121 -- Check if the argument is a variable that
1122 -- is in scope at the function definition site
1123 -- It's worth specialising on this if
1124 -- (a) it's used in an interesting way in the body
1125 -- (b) we know what its value is
1126 argToPat in_scope con_env (Var v) arg_occ
1127 | not (isLocalId v) || v `elemInScopeSet` in_scope,
1128 case arg_occ of { UnkOcc -> False; other -> True }, -- (a)
1129 isValueUnfolding (idUnfolding v) -- (b)
1130 = return (True, Var v)
1132 -- I'm really not sure what this comment means
1133 -- And by not wild-carding we tend to get forall'd
1134 -- variables that are in soope, which in turn can
1135 -- expose the weakness in let-matching
1136 -- See Note [Matching lets] in Rules
1137 -- Check for a variable bound inside the function.
1138 -- Don't make a wild-card, because we may usefully share
1139 -- e.g. f a = let x = ... in f (x,x)
1140 -- NB: this case follows the lambda and con-app cases!!
1141 argToPat in_scope con_env (Var v) arg_occ
1142 = return (False, Var v)
1144 -- The default case: make a wild-card
1145 argToPat in_scope con_env arg arg_occ
1146 = wildCardPat (exprType arg)
1148 wildCardPat :: Type -> UniqSM (Bool, CoreArg)
1149 wildCardPat ty = do { uniq <- getUniqueUs
1150 ; let id = mkSysLocal FSLIT("sc") uniq ty
1151 ; return (False, Var id) }
1153 argsToPats :: InScopeSet -> ConstrEnv
1154 -> [(CoreArg, ArgOcc)]
1155 -> UniqSM [(Bool, CoreArg)]
1156 argsToPats in_scope con_env args
1159 do_one (arg,occ) = argToPat in_scope con_env arg occ
1164 isConApp :: ConstrEnv -> CoreExpr -> Maybe ConValue
1165 isConApp env (Lit lit)
1166 = Just (CV (LitAlt lit) [])
1168 isConApp env expr -- Maybe it's a constructor application
1169 | (Var fun, args) <- collectArgs expr,
1170 Just con <- isDataConWorkId_maybe fun,
1171 args `lengthAtLeast` dataConRepArity con
1172 -- Might be > because the arity excludes type args
1173 = Just (CV (DataAlt con) args)
1175 isConApp env (Var v)
1176 | Just stuff <- lookupVarEnv env v
1177 = Just stuff -- You might think we could look in the idUnfolding here
1178 -- but that doesn't take account of which branch of a
1179 -- case we are in, which is the whole point
1181 | not (isLocalId v) && isCheapUnfolding unf
1182 = isConApp env (unfoldingTemplate unf)
1185 -- However we do want to consult the unfolding
1186 -- as well, for let-bound constructors!
1188 isConApp env expr = Nothing
1190 mk_con_app :: AltCon -> [CoreArg] -> CoreExpr
1191 mk_con_app (LitAlt lit) [] = Lit lit
1192 mk_con_app (DataAlt con) args = mkConApp con args
1193 mk_con_app other args = panic "SpecConstr.mk_con_app"
1195 samePat :: CallPat -> CallPat -> Bool
1196 samePat (vs1, as1) (vs2, as2)
1199 same (Var v1) (Var v2)
1200 | v1 `elem` vs1 = v2 `elem` vs2
1201 | v2 `elem` vs2 = False
1202 | otherwise = v1 == v2
1204 same (Lit l1) (Lit l2) = l1==l2
1205 same (App f1 a1) (App f2 a2) = same f1 f2 && same a1 a2
1207 same (Type t1) (Type t2) = True -- Note [Ignore type differences]
1208 same (Note _ e1) e2 = same e1 e2 -- Ignore casts and notes
1209 same (Cast e1 _) e2 = same e1 e2
1210 same e1 (Note _ e2) = same e1 e2
1211 same e1 (Cast e2 _) = same e1 e2
1213 same e1 e2 = WARN( bad e1 || bad e2, ppr e1 $$ ppr e2)
1214 False -- Let, lambda, case should not occur
1216 bad (Case {}) = True
1223 Note [Ignore type differences]
1224 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1225 We do not want to generate specialisations where the call patterns
1226 differ only in their type arguments! Not only is it utterly useless,
1227 but it also means that (with polymorphic recursion) we can generate
1228 an infinite number of specialisations. Example is Data.Sequence.adjustTree,