2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section[SpecConstr]{Specialise over constructors}
7 -- The above warning supression flag is a temporary kludge.
8 -- While working on this module you are encouraged to remove it and fix
9 -- any warnings in the module. See
10 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
16 , SpecConstrAnnotation(..)
20 #include "HsVersions.h"
25 import CoreUnfold ( couldBeSmallEnoughToInline )
26 import CoreFVs ( exprsFreeVars )
28 import HscTypes ( ModGuts(..) )
29 import WwLib ( mkWorkerArgs )
33 import Type hiding( substTy )
35 import MkCore ( mkImpossibleExpr )
41 import DynFlags ( DynFlags(..) )
42 import StaticFlags ( opt_PprStyle_Debug )
43 import Maybes ( orElse, catMaybes, isJust, isNothing )
45 import DmdAnal ( both )
46 import Serialized ( deserializeWithData )
53 import Control.Monad ( zipWithM )
57 -- See Note [SpecConstrAnnotation]
59 type SpecConstrAnnotation = ()
61 import Literal ( literalType )
62 import TyCon ( TyCon )
63 import GHC.Exts( SpecConstrAnnotation(..) )
67 -----------------------------------------------------
69 -----------------------------------------------------
74 drop n (x:xs) = drop (n-1) xs
76 After the first time round, we could pass n unboxed. This happens in
77 numerical code too. Here's what it looks like in Core:
79 drop n xs = case xs of
84 _ -> drop (I# (n# -# 1#)) xs
86 Notice that the recursive call has an explicit constructor as argument.
87 Noticing this, we can make a specialised version of drop
89 RULE: drop (I# n#) xs ==> drop' n# xs
91 drop' n# xs = let n = I# n# in ...orig RHS...
93 Now the simplifier will apply the specialisation in the rhs of drop', giving
95 drop' n# xs = case xs of
99 _ -> drop (n# -# 1#) xs
103 We'd also like to catch cases where a parameter is carried along unchanged,
104 but evaluated each time round the loop:
106 f i n = if i>0 || i>n then i else f (i*2) n
108 Here f isn't strict in n, but we'd like to avoid evaluating it each iteration.
109 In Core, by the time we've w/wd (f is strict in i) we get
111 f i# n = case i# ># 0 of
113 True -> case n of n' { I# n# ->
116 True -> f (i# *# 2#) n'
118 At the call to f, we see that the argument, n is know to be (I# n#),
119 and n is evaluated elsewhere in the body of f, so we can play the same
125 We must be careful not to allocate the same constructor twice. Consider
126 f p = (...(case p of (a,b) -> e)...p...,
127 ...let t = (r,s) in ...t...(f t)...)
128 At the recursive call to f, we can see that t is a pair. But we do NOT want
129 to make a specialised copy:
130 f' a b = let p = (a,b) in (..., ...)
131 because now t is allocated by the caller, then r and s are passed to the
132 recursive call, which allocates the (r,s) pair again.
135 (a) the argument p is used in other than a case-scrutinsation way.
136 (b) the argument to the call is not a 'fresh' tuple; you have to
137 look into its unfolding to see that it's a tuple
139 Hence the "OR" part of Note [Good arguments] below.
141 ALTERNATIVE 2: pass both boxed and unboxed versions. This no longer saves
142 allocation, but does perhaps save evals. In the RULE we'd have
145 f (I# x#) = f' (I# x#) x#
147 If at the call site the (I# x) was an unfolding, then we'd have to
148 rely on CSE to eliminate the duplicate allocation.... This alternative
149 doesn't look attractive enough to pursue.
151 ALTERNATIVE 3: ignore the reboxing problem. The trouble is that
152 the conservative reboxing story prevents many useful functions from being
153 specialised. Example:
154 foo :: Maybe Int -> Int -> Int
156 foo x@(Just m) n = foo x (n-m)
157 Here the use of 'x' will clearly not require boxing in the specialised function.
159 The strictness analyser has the same problem, in fact. Example:
161 If we pass just 'a' and 'b' to the worker, it might need to rebox the
162 pair to create (a,b). A more sophisticated analysis might figure out
163 precisely the cases in which this could happen, but the strictness
164 analyser does no such analysis; it just passes 'a' and 'b', and hopes
167 So my current choice is to make SpecConstr similarly aggressive, and
168 ignore the bad potential of reboxing.
171 Note [Good arguments]
172 ~~~~~~~~~~~~~~~~~~~~~
175 * A self-recursive function. Ignore mutual recursion for now,
176 because it's less common, and the code is simpler for self-recursion.
180 a) At a recursive call, one or more parameters is an explicit
181 constructor application
183 That same parameter is scrutinised by a case somewhere in
184 the RHS of the function
188 b) At a recursive call, one or more parameters has an unfolding
189 that is an explicit constructor application
191 That same parameter is scrutinised by a case somewhere in
192 the RHS of the function
194 Those are the only uses of the parameter (see Note [Reboxing])
197 What to abstract over
198 ~~~~~~~~~~~~~~~~~~~~~
199 There's a bit of a complication with type arguments. If the call
202 f p = ...f ((:) [a] x xs)...
204 then our specialised function look like
206 f_spec x xs = let p = (:) [a] x xs in ....as before....
208 This only makes sense if either
209 a) the type variable 'a' is in scope at the top of f, or
210 b) the type variable 'a' is an argument to f (and hence fs)
212 Actually, (a) may hold for value arguments too, in which case
213 we may not want to pass them. Supose 'x' is in scope at f's
214 defn, but xs is not. Then we'd like
216 f_spec xs = let p = (:) [a] x xs in ....as before....
218 Similarly (b) may hold too. If x is already an argument at the
219 call, no need to pass it again.
221 Finally, if 'a' is not in scope at the call site, we could abstract
222 it as we do the term variables:
224 f_spec a x xs = let p = (:) [a] x xs in ...as before...
226 So the grand plan is:
228 * abstract the call site to a constructor-only pattern
229 e.g. C x (D (f p) (g q)) ==> C s1 (D s2 s3)
231 * Find the free variables of the abstracted pattern
233 * Pass these variables, less any that are in scope at
234 the fn defn. But see Note [Shadowing] below.
237 NOTICE that we only abstract over variables that are not in scope,
238 so we're in no danger of shadowing variables used in "higher up"
244 In this pass we gather up usage information that may mention variables
245 that are bound between the usage site and the definition site; or (more
246 seriously) may be bound to something different at the definition site.
249 f x = letrec g y v = let x = ...
252 Since 'x' is in scope at the call site, we may make a rewrite rule that
254 RULE forall a,b. g (a,b) x = ...
255 But this rule will never match, because it's really a different 'x' at
256 the call site -- and that difference will be manifest by the time the
257 simplifier gets to it. [A worry: the simplifier doesn't *guarantee*
258 no-shadowing, so perhaps it may not be distinct?]
260 Anyway, the rule isn't actually wrong, it's just not useful. One possibility
261 is to run deShadowBinds before running SpecConstr, but instead we run the
262 simplifier. That gives the simplest possible program for SpecConstr to
263 chew on; and it virtually guarantees no shadowing.
265 Note [Specialising for constant parameters]
266 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
267 This one is about specialising on a *constant* (but not necessarily
268 constructor) argument
270 foo :: Int -> (Int -> Int) -> Int
272 foo m f = foo (f m) (+1)
276 lvl_rmV :: GHC.Base.Int -> GHC.Base.Int
278 \ (ds_dlk :: GHC.Base.Int) ->
279 case ds_dlk of wild_alH { GHC.Base.I# x_alG ->
280 GHC.Base.I# (GHC.Prim.+# x_alG 1)
282 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
285 \ (ww_sme :: GHC.Prim.Int#) (w_smg :: GHC.Base.Int -> GHC.Base.Int) ->
286 case ww_sme of ds_Xlw {
288 case w_smg (GHC.Base.I# ds_Xlw) of w1_Xmo { GHC.Base.I# ww1_Xmz ->
289 T.$wfoo ww1_Xmz lvl_rmV
294 The recursive call has lvl_rmV as its argument, so we could create a specialised copy
295 with that argument baked in; that is, not passed at all. Now it can perhaps be inlined.
297 When is this worth it? Call the constant 'lvl'
298 - If 'lvl' has an unfolding that is a constructor, see if the corresponding
299 parameter is scrutinised anywhere in the body.
301 - If 'lvl' has an unfolding that is a inlinable function, see if the corresponding
302 parameter is applied (...to enough arguments...?)
304 Also do this is if the function has RULES?
308 Note [Specialising for lambda parameters]
309 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
310 foo :: Int -> (Int -> Int) -> Int
312 foo m f = foo (f m) (\n -> n-m)
314 This is subtly different from the previous one in that we get an
315 explicit lambda as the argument:
317 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
320 \ (ww_sm8 :: GHC.Prim.Int#) (w_sma :: GHC.Base.Int -> GHC.Base.Int) ->
321 case ww_sm8 of ds_Xlr {
323 case w_sma (GHC.Base.I# ds_Xlr) of w1_Xmf { GHC.Base.I# ww1_Xmq ->
326 (\ (n_ad3 :: GHC.Base.Int) ->
327 case n_ad3 of wild_alB { GHC.Base.I# x_alA ->
328 GHC.Base.I# (GHC.Prim.-# x_alA ds_Xlr)
334 I wonder if SpecConstr couldn't be extended to handle this? After all,
335 lambda is a sort of constructor for functions and perhaps it already
336 has most of the necessary machinery?
338 Furthermore, there's an immediate win, because you don't need to allocate the lamda
339 at the call site; and if perchance it's called in the recursive call, then you
340 may avoid allocating it altogether. Just like for constructors.
342 Looks cool, but probably rare...but it might be easy to implement.
345 Note [SpecConstr for casts]
346 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
349 data instance T Int = T Int
354 go (T n) = go (T (n-1))
356 The recursive call ends up looking like
357 go (T (I# ...) `cast` g)
358 So we want to spot the construtor application inside the cast.
359 That's why we have the Cast case in argToPat
361 Note [Local recursive groups]
362 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
363 For a *local* recursive group, we can see all the calls to the
364 function, so we seed the specialisation loop from the calls in the
365 body, not from the calls in the RHS. Consider:
367 bar m n = foo n (n,n) (n,n) (n,n) (n,n)
371 | n > 3000 = case p of { (p1,p2) -> foo (n-1) (p2,p1) q r s }
372 | n > 2000 = case q of { (q1,q2) -> foo (n-1) p (q2,q1) r s }
373 | n > 1000 = case r of { (r1,r2) -> foo (n-1) p q (r2,r1) s }
374 | otherwise = case s of { (s1,s2) -> foo (n-1) p q r (s2,s1) }
376 If we start with the RHSs of 'foo', we get lots and lots of specialisations,
377 most of which are not needed. But if we start with the (single) call
378 in the rhs of 'bar' we get exactly one fully-specialised copy, and all
379 the recursive calls go to this fully-specialised copy. Indeed, the original
380 function is later collected as dead code. This is very important in
381 specialising the loops arising from stream fusion, for example in NDP where
382 we were getting literally hundreds of (mostly unused) specialisations of
385 Note [Do not specialise diverging functions]
386 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
387 Specialising a function that just diverges is a waste of code.
388 Furthermore, it broke GHC (simpl014) thus:
390 f = \x. case x of (a,b) -> f x
391 If we specialise f we get
392 f = \x. case x of (a,b) -> fspec a b
393 But fspec doesn't have decent strictnes info. As it happened,
394 (f x) :: IO t, so the state hack applied and we eta expanded fspec,
395 and hence f. But now f's strictness is less than its arity, which
398 Note [SpecConstrAnnotation]
399 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
400 SpecConstrAnnotation is defined in GHC.Exts, and is only guaranteed to
401 be available in stage 2 (well, until the bootstrap compiler can be
402 guaranteed to have it)
404 So we define it to be () in stage1 (ie when GHCI is undefined), and
405 '#ifdef' out the code that uses it.
407 See also Note [Forcing specialisation]
409 Note [Forcing specialisation]
410 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
411 With stream fusion and in other similar cases, we want to fully specialise
412 some (but not necessarily all!) loops regardless of their size and the
413 number of specialisations. We allow a library to specify this by annotating
414 a type with ForceSpecConstr and then adding a parameter of that type to the
415 loop. Here is a (simplified) example from the vector library:
417 data SPEC = SPEC | SPEC2
418 {-# ANN type SPEC ForceSpecConstr #-}
420 foldl :: (a -> b -> a) -> a -> Stream b -> a
422 foldl f z (Stream step s _) = foldl_loop SPEC z s
424 foldl_loop SPEC z s = case step s of
425 Yield x s' -> foldl_loop SPEC (f z x) s'
426 Skip -> foldl_loop SPEC z s'
429 SpecConstr will spot the SPEC parameter and always fully specialise
430 foldl_loop. Note that we can't just annotate foldl_loop since it isn't a
431 top-level function but even if we could, inlining etc. could easily drop the
432 annotation. We also have to prevent the SPEC argument from being removed by
433 w/w which is why SPEC is a sum type. This is all quite ugly; we ought to come
434 up with a better design.
436 ForceSpecConstr arguments are spotted in scExpr' and scTopBinds which then set
437 force_spec to True when calling specLoop. This flag makes specLoop and
438 specialise ignore specConstrCount and specConstrThreshold when deciding
439 whether to specialise a function.
441 -----------------------------------------------------
442 Stuff not yet handled
443 -----------------------------------------------------
445 Here are notes arising from Roman's work that I don't want to lose.
451 foo :: Int -> T Int -> Int
453 foo x t | even x = case t of { T n -> foo (x-n) t }
454 | otherwise = foo (x-1) t
456 SpecConstr does no specialisation, because the second recursive call
457 looks like a boxed use of the argument. A pity.
459 $wfoo_sFw :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
461 \ (ww_sFo [Just L] :: GHC.Prim.Int#) (w_sFq [Just L] :: T.T GHC.Base.Int) ->
462 case ww_sFo of ds_Xw6 [Just L] {
464 case GHC.Prim.remInt# ds_Xw6 2 of wild1_aEF [Dead Just A] {
465 __DEFAULT -> $wfoo_sFw (GHC.Prim.-# ds_Xw6 1) w_sFq;
467 case w_sFq of wild_Xy [Just L] { T.T n_ad5 [Just U(L)] ->
468 case n_ad5 of wild1_aET [Just A] { GHC.Base.I# y_aES [Just L] ->
469 $wfoo_sFw (GHC.Prim.-# ds_Xw6 y_aES) wild_Xy
475 data a :*: b = !a :*: !b
478 foo :: (Int :*: T Int) -> Int
480 foo (x :*: t) | even x = case t of { T n -> foo ((x-n) :*: t) }
481 | otherwise = foo ((x-1) :*: t)
483 Very similar to the previous one, except that the parameters are now in
484 a strict tuple. Before SpecConstr, we have
486 $wfoo_sG3 :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
488 \ (ww_sFU [Just L] :: GHC.Prim.Int#) (ww_sFW [Just L] :: T.T
490 case ww_sFU of ds_Xws [Just L] {
492 case GHC.Prim.remInt# ds_Xws 2 of wild1_aEZ [Dead Just A] {
494 case ww_sFW of tpl_B2 [Just L] { T.T a_sFo [Just A] ->
495 $wfoo_sG3 (GHC.Prim.-# ds_Xws 1) tpl_B2 -- $wfoo1
498 case ww_sFW of wild_XB [Just A] { T.T n_ad7 [Just S(L)] ->
499 case n_ad7 of wild1_aFd [Just L] { GHC.Base.I# y_aFc [Just L] ->
500 $wfoo_sG3 (GHC.Prim.-# ds_Xws y_aFc) wild_XB -- $wfoo2
504 We get two specialisations:
505 "SC:$wfoo1" [0] __forall {a_sFB :: GHC.Base.Int sc_sGC :: GHC.Prim.Int#}
506 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int a_sFB)
507 = Foo.$s$wfoo1 a_sFB sc_sGC ;
508 "SC:$wfoo2" [0] __forall {y_aFp :: GHC.Prim.Int# sc_sGC :: GHC.Prim.Int#}
509 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int (GHC.Base.I# y_aFp))
510 = Foo.$s$wfoo y_aFp sc_sGC ;
512 But perhaps the first one isn't good. After all, we know that tpl_B2 is
513 a T (I# x) really, because T is strict and Int has one constructor. (We can't
514 unbox the strict fields, becuase T is polymorphic!)
516 %************************************************************************
518 \subsection{Top level wrapper stuff}
520 %************************************************************************
523 specConstrProgram :: ModGuts -> CoreM ModGuts
524 specConstrProgram guts
526 dflags <- getDynFlags
527 us <- getUniqueSupplyM
528 annos <- getFirstAnnotations deserializeWithData guts
529 let binds' = fst $ initUs us (go (initScEnv dflags annos) (mg_binds guts))
530 return (guts { mg_binds = binds' })
533 go env (bind:binds) = do (env', bind') <- scTopBind env bind
534 binds' <- go env' binds
535 return (bind' : binds')
539 %************************************************************************
541 \subsection{Environment: goes downwards}
543 %************************************************************************
546 data ScEnv = SCE { sc_size :: Maybe Int, -- Size threshold
547 sc_count :: Maybe Int, -- Max # of specialisations for any one fn
548 -- See Note [Avoiding exponential blowup]
550 sc_subst :: Subst, -- Current substitution
551 -- Maps InIds to OutExprs
553 sc_how_bound :: HowBoundEnv,
554 -- Binds interesting non-top-level variables
555 -- Domain is OutVars (*after* applying the substitution)
558 -- Domain is OutIds (*after* applying the substitution)
559 -- Used even for top-level bindings (but not imported ones)
561 sc_annotations :: UniqFM SpecConstrAnnotation
564 ---------------------
565 -- As we go, we apply a substitution (sc_subst) to the current term
566 type InExpr = CoreExpr -- _Before_ applying the subst
569 type OutExpr = CoreExpr -- _After_ applying the subst
573 ---------------------
574 type HowBoundEnv = VarEnv HowBound -- Domain is OutVars
576 ---------------------
577 type ValueEnv = IdEnv Value -- Domain is OutIds
578 data Value = ConVal AltCon [CoreArg] -- _Saturated_ constructors
579 -- The AltCon is never DEFAULT
580 | LambdaVal -- Inlinable lambdas or PAPs
582 instance Outputable Value where
583 ppr (ConVal con args) = ppr con <+> interpp'SP args
584 ppr LambdaVal = ptext (sLit "<Lambda>")
586 ---------------------
587 initScEnv :: DynFlags -> UniqFM SpecConstrAnnotation -> ScEnv
588 initScEnv dflags anns
589 = SCE { sc_size = specConstrThreshold dflags,
590 sc_count = specConstrCount dflags,
591 sc_subst = emptySubst,
592 sc_how_bound = emptyVarEnv,
593 sc_vals = emptyVarEnv,
594 sc_annotations = anns }
596 data HowBound = RecFun -- These are the recursive functions for which
597 -- we seek interesting call patterns
599 | RecArg -- These are those functions' arguments, or their sub-components;
600 -- we gather occurrence information for these
602 instance Outputable HowBound where
603 ppr RecFun = text "RecFun"
604 ppr RecArg = text "RecArg"
606 lookupHowBound :: ScEnv -> Id -> Maybe HowBound
607 lookupHowBound env id = lookupVarEnv (sc_how_bound env) id
609 scSubstId :: ScEnv -> Id -> CoreExpr
610 scSubstId env v = lookupIdSubst (text "scSubstId") (sc_subst env) v
612 scSubstTy :: ScEnv -> Type -> Type
613 scSubstTy env ty = substTy (sc_subst env) ty
615 zapScSubst :: ScEnv -> ScEnv
616 zapScSubst env = env { sc_subst = zapSubstEnv (sc_subst env) }
618 extendScInScope :: ScEnv -> [Var] -> ScEnv
619 -- Bring the quantified variables into scope
620 extendScInScope env qvars = env { sc_subst = extendInScopeList (sc_subst env) qvars }
622 -- Extend the substitution
623 extendScSubst :: ScEnv -> Var -> OutExpr -> ScEnv
624 extendScSubst env var expr = env { sc_subst = extendSubst (sc_subst env) var expr }
626 extendScSubstList :: ScEnv -> [(Var,OutExpr)] -> ScEnv
627 extendScSubstList env prs = env { sc_subst = extendSubstList (sc_subst env) prs }
629 extendHowBound :: ScEnv -> [Var] -> HowBound -> ScEnv
630 extendHowBound env bndrs how_bound
631 = env { sc_how_bound = extendVarEnvList (sc_how_bound env)
632 [(bndr,how_bound) | bndr <- bndrs] }
634 extendBndrsWith :: HowBound -> ScEnv -> [Var] -> (ScEnv, [Var])
635 extendBndrsWith how_bound env bndrs
636 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndrs')
638 (subst', bndrs') = substBndrs (sc_subst env) bndrs
639 hb_env' = sc_how_bound env `extendVarEnvList`
640 [(bndr,how_bound) | bndr <- bndrs']
642 extendBndrWith :: HowBound -> ScEnv -> Var -> (ScEnv, Var)
643 extendBndrWith how_bound env bndr
644 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndr')
646 (subst', bndr') = substBndr (sc_subst env) bndr
647 hb_env' = extendVarEnv (sc_how_bound env) bndr' how_bound
649 extendRecBndrs :: ScEnv -> [Var] -> (ScEnv, [Var])
650 extendRecBndrs env bndrs = (env { sc_subst = subst' }, bndrs')
652 (subst', bndrs') = substRecBndrs (sc_subst env) bndrs
654 extendBndr :: ScEnv -> Var -> (ScEnv, Var)
655 extendBndr env bndr = (env { sc_subst = subst' }, bndr')
657 (subst', bndr') = substBndr (sc_subst env) bndr
659 extendValEnv :: ScEnv -> Id -> Maybe Value -> ScEnv
660 extendValEnv env _ Nothing = env
661 extendValEnv env id (Just cv) = env { sc_vals = extendVarEnv (sc_vals env) id cv }
663 extendCaseBndrs :: ScEnv -> Id -> AltCon -> [Var] -> (ScEnv, [Var])
667 -- we want to bind b, to (C x y)
668 -- NB1: Extends only the sc_vals part of the envt
669 -- NB2: Kill the dead-ness info on the pattern binders x,y, since
670 -- they are potentially made alive by the [b -> C x y] binding
671 extendCaseBndrs env case_bndr con alt_bndrs
672 | isDeadBinder case_bndr
675 = (env1, map zap alt_bndrs)
676 -- NB: We used to bind v too, if scrut = (Var v); but
677 -- the simplifer has already done this so it seems
678 -- redundant to do so here
680 -- Var v -> extendValEnv env1 v cval
683 zap v | isTyCoVar v = v -- See NB2 above
684 | otherwise = zapIdOccInfo v
685 env1 = extendValEnv env case_bndr cval
688 LitAlt {} -> Just (ConVal con [])
689 DataAlt {} -> Just (ConVal con vanilla_args)
691 vanilla_args = map Type (tyConAppArgs (idType case_bndr)) ++
692 varsToCoreExprs alt_bndrs
695 decreaseSpecCount :: ScEnv -> Int -> ScEnv
696 -- See Note [Avoiding exponential blowup]
697 decreaseSpecCount env n_specs
698 = env { sc_count = case sc_count env of
700 Just n -> Just (n `div` (n_specs + 1)) }
701 -- The "+1" takes account of the original function;
702 -- See Note [Avoiding exponential blowup]
704 ---------------------------------------------------
705 -- See Note [SpecConstrAnnotation]
706 ignoreType :: ScEnv -> Type -> Bool
707 ignoreAltCon :: ScEnv -> AltCon -> Bool
708 forceSpecBndr :: ScEnv -> Var -> Bool
710 ignoreType _ _ = False
711 ignoreAltCon _ _ = False
712 forceSpecBndr _ _ = False
716 ignoreAltCon env (DataAlt dc) = ignoreTyCon env (dataConTyCon dc)
717 ignoreAltCon env (LitAlt lit) = ignoreType env (literalType lit)
718 ignoreAltCon _ DEFAULT = panic "ignoreAltCon" -- DEFAULT cannot be in a ConVal
721 = case splitTyConApp_maybe ty of
722 Just (tycon, _) -> ignoreTyCon env tycon
725 ignoreTyCon :: ScEnv -> TyCon -> Bool
726 ignoreTyCon env tycon
727 = lookupUFM (sc_annotations env) tycon == Just NoSpecConstr
729 forceSpecBndr env var = forceSpecFunTy env . snd . splitForAllTys . varType $ var
731 forceSpecFunTy :: ScEnv -> Type -> Bool
732 forceSpecFunTy env = any (forceSpecArgTy env) . fst . splitFunTys
734 forceSpecArgTy :: ScEnv -> Type -> Bool
735 forceSpecArgTy env ty
736 | Just ty' <- coreView ty = forceSpecArgTy env ty'
738 forceSpecArgTy env ty
739 | Just (tycon, tys) <- splitTyConApp_maybe ty
741 = lookupUFM (sc_annotations env) tycon == Just ForceSpecConstr
742 || any (forceSpecArgTy env) tys
744 forceSpecArgTy _ _ = False
748 Note [Avoiding exponential blowup]
749 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
750 The sc_count field of the ScEnv says how many times we are prepared to
751 duplicate a single function. But we must take care with recursive
752 specialiations. Consider
754 let $j1 = let $j2 = let $j3 = ...
762 If we specialise $j1 then in each specialisation (as well as the original)
763 we can specialise $j2, and similarly $j3. Even if we make just *one*
764 specialisation of each, becuase we also have the original we'll get 2^n
765 copies of $j3, which is not good.
767 So when recursively specialising we divide the sc_count by the number of
768 copies we are making at this level, including the original.
771 %************************************************************************
773 \subsection{Usage information: flows upwards}
775 %************************************************************************
780 scu_calls :: CallEnv, -- Calls
781 -- The functions are a subset of the
782 -- RecFuns in the ScEnv
784 scu_occs :: !(IdEnv ArgOcc) -- Information on argument occurrences
785 } -- The domain is OutIds
787 type CallEnv = IdEnv [Call]
788 type Call = (ValueEnv, [CoreArg])
789 -- The arguments of the call, together with the
790 -- env giving the constructor bindings at the call site
793 nullUsage = SCU { scu_calls = emptyVarEnv, scu_occs = emptyVarEnv }
795 combineCalls :: CallEnv -> CallEnv -> CallEnv
796 combineCalls = plusVarEnv_C (++)
798 combineUsage :: ScUsage -> ScUsage -> ScUsage
799 combineUsage u1 u2 = SCU { scu_calls = combineCalls (scu_calls u1) (scu_calls u2),
800 scu_occs = plusVarEnv_C combineOcc (scu_occs u1) (scu_occs u2) }
802 combineUsages :: [ScUsage] -> ScUsage
803 combineUsages [] = nullUsage
804 combineUsages us = foldr1 combineUsage us
806 lookupOcc :: ScUsage -> OutVar -> (ScUsage, ArgOcc)
807 lookupOcc (SCU { scu_calls = sc_calls, scu_occs = sc_occs }) bndr
808 = (SCU {scu_calls = sc_calls, scu_occs = delVarEnv sc_occs bndr},
809 lookupVarEnv sc_occs bndr `orElse` NoOcc)
811 lookupOccs :: ScUsage -> [OutVar] -> (ScUsage, [ArgOcc])
812 lookupOccs (SCU { scu_calls = sc_calls, scu_occs = sc_occs }) bndrs
813 = (SCU {scu_calls = sc_calls, scu_occs = delVarEnvList sc_occs bndrs},
814 [lookupVarEnv sc_occs b `orElse` NoOcc | b <- bndrs])
816 data ArgOcc = NoOcc -- Doesn't occur at all; or a type argument
817 | UnkOcc -- Used in some unknown way
819 | ScrutOcc (UniqFM [ArgOcc]) -- See Note [ScrutOcc]
821 | BothOcc -- Definitely taken apart, *and* perhaps used in some other way
825 An occurrence of ScrutOcc indicates that the thing, or a `cast` version of the thing,
826 is *only* taken apart or applied.
828 Functions, literal: ScrutOcc emptyUFM
829 Data constructors: ScrutOcc subs,
831 where (subs :: UniqFM [ArgOcc]) gives usage of the *pattern-bound* components,
832 The domain of the UniqFM is the Unique of the data constructor
834 The [ArgOcc] is the occurrences of the *pattern-bound* components
835 of the data structure. E.g.
836 data T a = forall b. MkT a b (b->a)
837 A pattern binds b, x::a, y::b, z::b->a, but not 'a'!
841 instance Outputable ArgOcc where
842 ppr (ScrutOcc xs) = ptext (sLit "scrut-occ") <> ppr xs
843 ppr UnkOcc = ptext (sLit "unk-occ")
844 ppr BothOcc = ptext (sLit "both-occ")
845 ppr NoOcc = ptext (sLit "no-occ")
847 -- Experimentally, this vesion of combineOcc makes ScrutOcc "win", so
848 -- that if the thing is scrutinised anywhere then we get to see that
849 -- in the overall result, even if it's also used in a boxed way
850 -- This might be too agressive; see Note [Reboxing] Alternative 3
851 combineOcc :: ArgOcc -> ArgOcc -> ArgOcc
852 combineOcc NoOcc occ = occ
853 combineOcc occ NoOcc = occ
854 combineOcc (ScrutOcc xs) (ScrutOcc ys) = ScrutOcc (plusUFM_C combineOccs xs ys)
855 combineOcc _occ (ScrutOcc ys) = ScrutOcc ys
856 combineOcc (ScrutOcc xs) _occ = ScrutOcc xs
857 combineOcc UnkOcc UnkOcc = UnkOcc
858 combineOcc _ _ = BothOcc
860 combineOccs :: [ArgOcc] -> [ArgOcc] -> [ArgOcc]
861 combineOccs xs ys = zipWithEqual "combineOccs" combineOcc xs ys
863 setScrutOcc :: ScEnv -> ScUsage -> OutExpr -> ArgOcc -> ScUsage
864 -- _Overwrite_ the occurrence info for the scrutinee, if the scrutinee
865 -- is a variable, and an interesting variable
866 setScrutOcc env usg (Cast e _) occ = setScrutOcc env usg e occ
867 setScrutOcc env usg (Note _ e) occ = setScrutOcc env usg e occ
868 setScrutOcc env usg (Var v) occ
869 | Just RecArg <- lookupHowBound env v = usg { scu_occs = extendVarEnv (scu_occs usg) v occ }
871 setScrutOcc _env usg _other _occ -- Catch-all
874 conArgOccs :: ArgOcc -> AltCon -> [ArgOcc]
875 -- Find usage of components of data con; returns [UnkOcc...] if unknown
876 -- See Note [ScrutOcc] for the extra UnkOccs in the vanilla datacon case
878 conArgOccs (ScrutOcc fm) (DataAlt dc)
879 | Just pat_arg_occs <- lookupUFM fm dc
880 = [UnkOcc | _ <- dataConUnivTyVars dc] ++ pat_arg_occs
882 conArgOccs _other _con = repeat UnkOcc
885 %************************************************************************
887 \subsection{The main recursive function}
889 %************************************************************************
891 The main recursive function gathers up usage information, and
892 creates specialised versions of functions.
895 scExpr, scExpr' :: ScEnv -> CoreExpr -> UniqSM (ScUsage, CoreExpr)
896 -- The unique supply is needed when we invent
897 -- a new name for the specialised function and its args
899 scExpr env e = scExpr' env e
902 scExpr' env (Var v) = case scSubstId env v of
903 Var v' -> return (varUsage env v' UnkOcc, Var v')
904 e' -> scExpr (zapScSubst env) e'
906 scExpr' env (Type t) = return (nullUsage, Type (scSubstTy env t))
907 scExpr' _ e@(Lit {}) = return (nullUsage, e)
908 scExpr' env (Note n e) = do (usg,e') <- scExpr env e
909 return (usg, Note n e')
910 scExpr' env (Cast e co) = do (usg, e') <- scExpr env e
911 return (usg, Cast e' (scSubstTy env co))
912 scExpr' env e@(App _ _) = scApp env (collectArgs e)
913 scExpr' env (Lam b e) = do let (env', b') = extendBndr env b
914 (usg, e') <- scExpr env' e
915 return (usg, Lam b' e')
917 scExpr' env (Case scrut b ty alts)
918 = do { (scrut_usg, scrut') <- scExpr env scrut
919 ; case isValue (sc_vals env) scrut' of
920 Just (ConVal con args) -> sc_con_app con args scrut'
921 _other -> sc_vanilla scrut_usg scrut'
924 sc_con_app con args scrut' -- Known constructor; simplify
925 = do { let (_, bs, rhs) = findAlt con alts
926 `orElse` (DEFAULT, [], mkImpossibleExpr (coreAltsType alts))
927 alt_env' = extendScSubstList env ((b,scrut') : bs `zip` trimConArgs con args)
928 ; scExpr alt_env' rhs }
930 sc_vanilla scrut_usg scrut' -- Normal case
931 = do { let (alt_env,b') = extendBndrWith RecArg env b
932 -- Record RecArg for the components
934 ; (alt_usgs, alt_occs, alts')
935 <- mapAndUnzip3M (sc_alt alt_env scrut' b') alts
937 ; let (alt_usg, b_occ) = lookupOcc (combineUsages alt_usgs) b'
938 scrut_occ = foldr combineOcc b_occ alt_occs
939 scrut_usg' = setScrutOcc env scrut_usg scrut' scrut_occ
940 -- The combined usage of the scrutinee is given
941 -- by scrut_occ, which is passed to scScrut, which
942 -- in turn treats a bare-variable scrutinee specially
944 ; return (alt_usg `combineUsage` scrut_usg',
945 Case scrut' b' (scSubstTy env ty) alts') }
947 sc_alt env _scrut' b' (con,bs,rhs)
948 = do { let (env1, bs1) = extendBndrsWith RecArg env bs
949 (env2, bs2) = extendCaseBndrs env1 b' con bs1
950 ; (usg,rhs') <- scExpr env2 rhs
951 ; let (usg', arg_occs) = lookupOccs usg bs2
952 scrut_occ = case con of
953 DataAlt dc -> ScrutOcc (unitUFM dc arg_occs)
954 _ -> ScrutOcc emptyUFM
955 ; return (usg', scrut_occ, (con, bs2, rhs')) }
957 scExpr' env (Let (NonRec bndr rhs) body)
958 | isTyCoVar bndr -- Type-lets may be created by doBeta
959 = scExpr' (extendScSubst env bndr rhs) body
962 = do { let (body_env, bndr') = extendBndr env bndr
963 ; (rhs_usg, rhs_info) <- scRecRhs env (bndr',rhs)
965 ; let body_env2 = extendHowBound body_env [bndr'] RecFun
966 -- Note [Local let bindings]
967 RI _ rhs' _ _ _ = rhs_info
968 body_env3 = extendValEnv body_env2 bndr' (isValue (sc_vals env) rhs')
970 ; (body_usg, body') <- scExpr body_env3 body
972 -- NB: We don't use the ForceSpecConstr mechanism (see
973 -- Note [Forcing specialisation]) for non-recursive bindings
974 -- at the moment. I'm not sure if this is the right thing to do.
975 ; let force_spec = False
976 ; (spec_usg, specs) <- specialise env force_spec
979 (SI [] 0 (Just rhs_usg))
981 ; return (body_usg { scu_calls = scu_calls body_usg `delVarEnv` bndr' }
982 `combineUsage` spec_usg,
983 mkLets [NonRec b r | (b,r) <- specInfoBinds rhs_info specs] body')
987 -- A *local* recursive group: see Note [Local recursive groups]
988 scExpr' env (Let (Rec prs) body)
989 = do { let (bndrs,rhss) = unzip prs
990 (rhs_env1,bndrs') = extendRecBndrs env bndrs
991 rhs_env2 = extendHowBound rhs_env1 bndrs' RecFun
992 force_spec = any (forceSpecBndr env) bndrs'
993 -- Note [Forcing specialisation]
995 ; (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs rhs_env2) (bndrs' `zip` rhss)
996 ; (body_usg, body') <- scExpr rhs_env2 body
998 -- NB: start specLoop from body_usg
999 ; (spec_usg, specs) <- specLoop rhs_env2 force_spec
1000 (scu_calls body_usg) rhs_infos nullUsage
1001 [SI [] 0 (Just usg) | usg <- rhs_usgs]
1002 -- Do not unconditionally use rhs_usgs.
1003 -- Instead use them only if we find an unspecialised call
1004 -- See Note [Local recursive groups]
1006 ; let all_usg = spec_usg `combineUsage` body_usg
1007 bind' = Rec (concat (zipWith specInfoBinds rhs_infos specs))
1009 ; return (all_usg { scu_calls = scu_calls all_usg `delVarEnvList` bndrs' },
1013 Note [Local let bindings]
1014 ~~~~~~~~~~~~~~~~~~~~~~~~~
1015 It is not uncommon to find this
1017 let $j = \x. <blah> in ...$j True...$j True...
1019 Here $j is an arbitrary let-bound function, but it often comes up for
1020 join points. We might like to specialise $j for its call patterns.
1021 Notice the difference from a letrec, where we look for call patterns
1022 in the *RHS* of the function. Here we look for call patterns in the
1025 At one point I predicated this on the RHS mentioning the outer
1026 recursive function, but that's not essential and might even be
1027 harmful. I'm not sure.
1031 scApp :: ScEnv -> (InExpr, [InExpr]) -> UniqSM (ScUsage, CoreExpr)
1033 scApp env (Var fn, args) -- Function is a variable
1034 = ASSERT( not (null args) )
1035 do { args_w_usgs <- mapM (scExpr env) args
1036 ; let (arg_usgs, args') = unzip args_w_usgs
1037 arg_usg = combineUsages arg_usgs
1038 ; case scSubstId env fn of
1039 fn'@(Lam {}) -> scExpr (zapScSubst env) (doBeta fn' args')
1040 -- Do beta-reduction and try again
1042 Var fn' -> return (arg_usg `combineUsage` fn_usg, mkApps (Var fn') args')
1044 fn_usg = case lookupHowBound env fn' of
1045 Just RecFun -> SCU { scu_calls = unitVarEnv fn' [(sc_vals env, args')],
1046 scu_occs = emptyVarEnv }
1047 Just RecArg -> SCU { scu_calls = emptyVarEnv,
1048 scu_occs = unitVarEnv fn' (ScrutOcc emptyUFM) }
1049 Nothing -> nullUsage
1052 other_fn' -> return (arg_usg, mkApps other_fn' args') }
1053 -- NB: doing this ignores any usage info from the substituted
1054 -- function, but I don't think that matters. If it does
1057 doBeta :: OutExpr -> [OutExpr] -> OutExpr
1058 -- ToDo: adjust for System IF
1059 doBeta (Lam bndr body) (arg : args) = Let (NonRec bndr arg) (doBeta body args)
1060 doBeta fn args = mkApps fn args
1062 -- The function is almost always a variable, but not always.
1063 -- In particular, if this pass follows float-in,
1064 -- which it may, we can get
1065 -- (let f = ...f... in f) arg1 arg2
1066 scApp env (other_fn, args)
1067 = do { (fn_usg, fn') <- scExpr env other_fn
1068 ; (arg_usgs, args') <- mapAndUnzipM (scExpr env) args
1069 ; return (combineUsages arg_usgs `combineUsage` fn_usg, mkApps fn' args') }
1071 ----------------------
1072 scTopBind :: ScEnv -> CoreBind -> UniqSM (ScEnv, CoreBind)
1073 scTopBind env (Rec prs)
1074 | Just threshold <- sc_size env
1076 , not (all (couldBeSmallEnoughToInline threshold) rhss)
1077 -- No specialisation
1078 = do { let (rhs_env,bndrs') = extendRecBndrs env bndrs
1079 ; (_, rhss') <- mapAndUnzipM (scExpr rhs_env) rhss
1080 ; return (rhs_env, Rec (bndrs' `zip` rhss')) }
1081 | otherwise -- Do specialisation
1082 = do { let (rhs_env1,bndrs') = extendRecBndrs env bndrs
1083 rhs_env2 = extendHowBound rhs_env1 bndrs' RecFun
1085 ; (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs rhs_env2) (bndrs' `zip` rhss)
1086 ; let rhs_usg = combineUsages rhs_usgs
1088 ; (_, specs) <- specLoop rhs_env2 force_spec
1089 (scu_calls rhs_usg) rhs_infos nullUsage
1090 [SI [] 0 Nothing | _ <- bndrs]
1092 ; return (rhs_env1, -- For the body of the letrec, delete the RecFun business
1093 Rec (concat (zipWith specInfoBinds rhs_infos specs))) }
1095 (bndrs,rhss) = unzip prs
1096 force_spec = any (forceSpecBndr env) bndrs
1097 -- Note [Forcing specialisation]
1099 scTopBind env (NonRec bndr rhs)
1100 = do { (_, rhs') <- scExpr env rhs
1101 ; let (env1, bndr') = extendBndr env bndr
1102 env2 = extendValEnv env1 bndr' (isValue (sc_vals env) rhs')
1103 ; return (env2, NonRec bndr' rhs') }
1105 ----------------------
1106 scRecRhs :: ScEnv -> (OutId, InExpr) -> UniqSM (ScUsage, RhsInfo)
1107 scRecRhs env (bndr,rhs)
1108 = do { let (arg_bndrs,body) = collectBinders rhs
1109 (body_env, arg_bndrs') = extendBndrsWith RecArg env arg_bndrs
1110 ; (body_usg, body') <- scExpr body_env body
1111 ; let (rhs_usg, arg_occs) = lookupOccs body_usg arg_bndrs'
1112 ; return (rhs_usg, RI bndr (mkLams arg_bndrs' body')
1113 arg_bndrs body arg_occs) }
1114 -- The arg_occs says how the visible,
1115 -- lambda-bound binders of the RHS are used
1116 -- (including the TyVar binders)
1117 -- Two pats are the same if they match both ways
1119 ----------------------
1120 specInfoBinds :: RhsInfo -> SpecInfo -> [(Id,CoreExpr)]
1121 specInfoBinds (RI fn new_rhs _ _ _) (SI specs _ _)
1122 = [(id,rhs) | OS _ _ id rhs <- specs] ++
1123 [(fn `addIdSpecialisations` rules, new_rhs)]
1125 rules = [r | OS _ r _ _ <- specs]
1127 ----------------------
1128 varUsage :: ScEnv -> OutVar -> ArgOcc -> ScUsage
1130 | Just RecArg <- lookupHowBound env v = SCU { scu_calls = emptyVarEnv
1131 , scu_occs = unitVarEnv v use }
1132 | otherwise = nullUsage
1136 %************************************************************************
1138 The specialiser itself
1140 %************************************************************************
1143 data RhsInfo = RI OutId -- The binder
1144 OutExpr -- The new RHS
1145 [InVar] InExpr -- The *original* RHS (\xs.body)
1146 -- Note [Specialise original body]
1147 [ArgOcc] -- Info on how the xs occur in body
1149 data SpecInfo = SI [OneSpec] -- The specialisations we have generated
1151 Int -- Length of specs; used for numbering them
1153 (Maybe ScUsage) -- Nothing => we have generated specialisations
1154 -- from calls in the *original* RHS
1155 -- Just cs => we haven't, and this is the usage
1156 -- of the original RHS
1157 -- See Note [Local recursive groups]
1159 -- One specialisation: Rule plus definition
1160 data OneSpec = OS CallPat -- Call pattern that generated this specialisation
1161 CoreRule -- Rule connecting original id with the specialisation
1162 OutId OutExpr -- Spec id + its rhs
1166 -> Bool -- force specialisation?
1167 -- Note [Forcing specialisation]
1170 -> ScUsage -> [SpecInfo] -- One per binder; acccumulating parameter
1171 -> UniqSM (ScUsage, [SpecInfo]) -- ...ditto...
1172 specLoop env force_spec all_calls rhs_infos usg_so_far specs_so_far
1173 = do { specs_w_usg <- zipWithM (specialise env force_spec all_calls) rhs_infos specs_so_far
1174 ; let (new_usg_s, all_specs) = unzip specs_w_usg
1175 new_usg = combineUsages new_usg_s
1176 new_calls = scu_calls new_usg
1177 all_usg = usg_so_far `combineUsage` new_usg
1178 ; if isEmptyVarEnv new_calls then
1179 return (all_usg, all_specs)
1181 specLoop env force_spec new_calls rhs_infos all_usg all_specs }
1185 -> Bool -- force specialisation?
1186 -- Note [Forcing specialisation]
1187 -> CallEnv -- Info on calls
1189 -> SpecInfo -- Original RHS plus patterns dealt with
1190 -> UniqSM (ScUsage, SpecInfo) -- New specialised versions and their usage
1192 -- Note: the rhs here is the optimised version of the original rhs
1193 -- So when we make a specialised copy of the RHS, we're starting
1194 -- from an RHS whose nested functions have been optimised already.
1196 specialise env force_spec bind_calls (RI fn _ arg_bndrs body arg_occs)
1197 spec_info@(SI specs spec_count mb_unspec)
1198 | not (isBottomingId fn) -- Note [Do not specialise diverging functions]
1199 , not (isNeverActive (idInlineActivation fn)) -- See Note [Transfer activation]
1200 , notNull arg_bndrs -- Only specialise functions
1201 , Just all_calls <- lookupVarEnv bind_calls fn
1202 = do { (boring_call, pats) <- callsToPats env specs arg_occs all_calls
1203 -- ; pprTrace "specialise" (vcat [ ppr fn <+> text "with" <+> int (length pats) <+> text "good patterns"
1204 -- , text "arg_occs" <+> ppr arg_occs
1205 -- , text "calls" <+> ppr all_calls
1206 -- , text "good pats" <+> ppr pats]) $
1209 -- Bale out if too many specialisations
1210 ; let n_pats = length pats
1211 spec_count' = n_pats + spec_count
1212 ; case sc_count env of
1213 Just max | not force_spec && spec_count' > max
1214 -> pprTrace "SpecConstr" msg $
1215 return (nullUsage, spec_info)
1217 msg = vcat [ sep [ ptext (sLit "Function") <+> quotes (ppr fn)
1218 , nest 2 (ptext (sLit "has") <+>
1219 speakNOf spec_count' (ptext (sLit "call pattern")) <> comma <+>
1220 ptext (sLit "but the limit is") <+> int max) ]
1221 , ptext (sLit "Use -fspec-constr-count=n to set the bound")
1223 extra | not opt_PprStyle_Debug = ptext (sLit "Use -dppr-debug to see specialisations")
1224 | otherwise = ptext (sLit "Specialisations:") <+> ppr (pats ++ [p | OS p _ _ _ <- specs])
1226 _normal_case -> do {
1228 let spec_env = decreaseSpecCount env n_pats
1229 ; (spec_usgs, new_specs) <- mapAndUnzipM (spec_one spec_env fn arg_bndrs body)
1230 (pats `zip` [spec_count..])
1231 -- See Note [Specialise original body]
1233 ; let spec_usg = combineUsages spec_usgs
1234 (new_usg, mb_unspec')
1236 Just rhs_usg | boring_call -> (spec_usg `combineUsage` rhs_usg, Nothing)
1237 _ -> (spec_usg, mb_unspec)
1239 ; return (new_usg, SI (new_specs ++ specs) spec_count' mb_unspec') } }
1241 = return (nullUsage, spec_info) -- The boring case
1244 ---------------------
1246 -> OutId -- Function
1247 -> [InVar] -- Lambda-binders of RHS; should match patterns
1248 -> InExpr -- Body of the original function
1250 -> UniqSM (ScUsage, OneSpec) -- Rule and binding
1252 -- spec_one creates a specialised copy of the function, together
1253 -- with a rule for using it. I'm very proud of how short this
1254 -- function is, considering what it does :-).
1260 f = /\b \y::[(a,b)] -> ....f (b,c) ((:) (a,(b,c)) (x,v) (h w))...
1261 [c::*, v::(b,c) are presumably bound by the (...) part]
1263 f_spec = /\ b c \ v::(b,c) hw::[(a,(b,c))] ->
1264 (...entire body of f...) [b -> (b,c),
1265 y -> ((:) (a,(b,c)) (x,v) hw)]
1267 RULE: forall b::* c::*, -- Note, *not* forall a, x
1271 f (b,c) ((:) (a,(b,c)) (x,v) hw) = f_spec b c v hw
1274 spec_one env fn arg_bndrs body (call_pat@(qvars, pats), rule_number)
1275 = do { spec_uniq <- getUniqueUs
1276 ; let spec_env = extendScSubstList (extendScInScope env qvars)
1277 (arg_bndrs `zip` pats)
1279 fn_loc = nameSrcSpan fn_name
1280 spec_occ = mkSpecOcc (nameOccName fn_name)
1281 rule_name = mkFastString ("SC:" ++ showSDoc (ppr fn <> int rule_number))
1282 spec_name = mkInternalName spec_uniq spec_occ fn_loc
1283 -- ; pprTrace "{spec_one" (ppr (sc_count env) <+> ppr fn <+> ppr pats <+> text "-->" <+> ppr spec_name) $
1286 -- Specialise the body
1287 ; (spec_usg, spec_body) <- scExpr spec_env body
1289 -- ; pprTrace "done spec_one}" (ppr fn) $
1292 -- And build the results
1293 ; let spec_id = mkLocalId spec_name (mkPiTypes spec_lam_args body_ty)
1294 `setIdStrictness` spec_str -- See Note [Transfer strictness]
1295 `setIdArity` count isId spec_lam_args
1296 spec_str = calcSpecStrictness fn spec_lam_args pats
1297 (spec_lam_args, spec_call_args) = mkWorkerArgs qvars body_ty
1298 -- Usual w/w hack to avoid generating
1299 -- a spec_rhs of unlifted type and no args
1301 spec_rhs = mkLams spec_lam_args spec_body
1302 body_ty = exprType spec_body
1303 rule_rhs = mkVarApps (Var spec_id) spec_call_args
1304 inline_act = idInlineActivation fn
1305 rule = mkRule True {- Auto -} True {- Local -}
1306 rule_name inline_act fn_name qvars pats rule_rhs
1307 -- See Note [Transfer activation]
1308 ; return (spec_usg, OS call_pat rule spec_id spec_rhs) }
1310 calcSpecStrictness :: Id -- The original function
1311 -> [Var] -> [CoreExpr] -- Call pattern
1312 -> StrictSig -- Strictness of specialised thing
1313 -- See Note [Transfer strictness]
1314 calcSpecStrictness fn qvars pats
1315 = StrictSig (mkTopDmdType spec_dmds TopRes)
1317 spec_dmds = [ lookupVarEnv dmd_env qv `orElse` lazyDmd | qv <- qvars, isId qv ]
1318 StrictSig (DmdType _ dmds _) = idStrictness fn
1320 dmd_env = go emptyVarEnv dmds pats
1322 go env ds (Type {} : pats) = go env ds pats
1323 go env (d:ds) (pat : pats) = go (go_one env d pat) ds pats
1326 go_one env d (Var v) = extendVarEnv_C both env v d
1327 go_one env (Box d) e = go_one env d e
1328 go_one env (Eval (Prod ds)) e
1329 | (Var _, args) <- collectArgs e = go env ds args
1330 go_one env _ _ = env
1334 Note [Specialise original body]
1335 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1336 The RhsInfo for a binding keeps the *original* body of the binding. We
1337 must specialise that, *not* the result of applying specExpr to the RHS
1338 (which is also kept in RhsInfo). Otherwise we end up specialising a
1339 specialised RHS, and that can lead directly to exponential behaviour.
1341 Note [Transfer activation]
1342 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1343 This note is for SpecConstr, but exactly the same thing
1344 happens in the overloading specialiser; see
1345 Note [Auto-specialisation and RULES] in Specialise.
1347 In which phase should the specialise-constructor rules be active?
1348 Originally I made them always-active, but Manuel found that this
1349 defeated some clever user-written rules. Then I made them active only
1350 in Phase 0; after all, currently, the specConstr transformation is
1351 only run after the simplifier has reached Phase 0, but that meant
1352 that specialisations didn't fire inside wrappers; see test
1353 simplCore/should_compile/spec-inline.
1355 So now I just use the inline-activation of the parent Id, as the
1356 activation for the specialiation RULE, just like the main specialiser;
1358 This in turn means there is no point in specialising NOINLINE things,
1359 so we test for that.
1361 Note [Transfer strictness]
1362 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1363 We must transfer strictness information from the original function to
1364 the specialised one. Suppose, for example
1367 and a RULE f (a:as) b = f_spec a as b
1369 Now we want f_spec to have strictess LLS, otherwise we'll use call-by-need
1370 when calling f_spec instead of call-by-value. And that can result in
1371 unbounded worsening in space (cf the classic foldl vs foldl')
1373 See Trac #3437 for a good example.
1375 The function calcSpecStrictness performs the calculation.
1378 %************************************************************************
1380 \subsection{Argument analysis}
1382 %************************************************************************
1384 This code deals with analysing call-site arguments to see whether
1385 they are constructor applications.
1389 type CallPat = ([Var], [CoreExpr]) -- Quantified variables and arguments
1392 callsToPats :: ScEnv -> [OneSpec] -> [ArgOcc] -> [Call] -> UniqSM (Bool, [CallPat])
1393 -- Result has no duplicate patterns,
1394 -- nor ones mentioned in done_pats
1395 -- Bool indicates that there was at least one boring pattern
1396 callsToPats env done_specs bndr_occs calls
1397 = do { mb_pats <- mapM (callToPats env bndr_occs) calls
1399 ; let good_pats :: [([Var], [CoreArg])]
1400 good_pats = catMaybes mb_pats
1401 done_pats = [p | OS p _ _ _ <- done_specs]
1402 is_done p = any (samePat p) done_pats
1404 ; return (any isNothing mb_pats,
1405 filterOut is_done (nubBy samePat good_pats)) }
1407 callToPats :: ScEnv -> [ArgOcc] -> Call -> UniqSM (Maybe CallPat)
1408 -- The [Var] is the variables to quantify over in the rule
1409 -- Type variables come first, since they may scope
1410 -- over the following term variables
1411 -- The [CoreExpr] are the argument patterns for the rule
1412 callToPats env bndr_occs (con_env, args)
1413 | length args < length bndr_occs -- Check saturated
1416 = do { let in_scope = substInScope (sc_subst env)
1417 ; prs <- argsToPats env in_scope con_env (args `zip` bndr_occs)
1418 ; let (interesting_s, pats) = unzip prs
1419 pat_fvs = varSetElems (exprsFreeVars pats)
1420 qvars = filterOut (`elemInScopeSet` in_scope) pat_fvs
1421 -- Quantify over variables that are not in sccpe
1423 -- See Note [Shadowing] at the top
1425 (tvs, ids) = partition isTyCoVar qvars
1427 -- Put the type variables first; the type of a term
1428 -- variable may mention a type variable
1430 ; -- pprTrace "callToPats" (ppr args $$ ppr prs $$ ppr bndr_occs) $
1432 then return (Just (qvars', pats))
1433 else return Nothing }
1435 -- argToPat takes an actual argument, and returns an abstracted
1436 -- version, consisting of just the "constructor skeleton" of the
1437 -- argument, with non-constructor sub-expression replaced by new
1438 -- placeholder variables. For example:
1439 -- C a (D (f x) (g y)) ==> C p1 (D p2 p3)
1442 -> InScopeSet -- What's in scope at the fn defn site
1443 -> ValueEnv -- ValueEnv at the call site
1444 -> CoreArg -- A call arg (or component thereof)
1446 -> UniqSM (Bool, CoreArg)
1447 -- Returns (interesting, pat),
1448 -- where pat is the pattern derived from the argument
1449 -- intersting=True if the pattern is non-trivial (not a variable or type)
1450 -- E.g. x:xs --> (True, x:xs)
1451 -- f xs --> (False, w) where w is a fresh wildcard
1452 -- (f xs, 'c') --> (True, (w, 'c')) where w is a fresh wildcard
1453 -- \x. x+y --> (True, \x. x+y)
1454 -- lvl7 --> (True, lvl7) if lvl7 is bound
1455 -- somewhere further out
1457 argToPat _env _in_scope _val_env arg@(Type {}) _arg_occ
1458 = return (False, arg)
1460 argToPat env in_scope val_env (Note _ arg) arg_occ
1461 = argToPat env in_scope val_env arg arg_occ
1462 -- Note [Notes in call patterns]
1463 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1464 -- Ignore Notes. In particular, we want to ignore any InlineMe notes
1465 -- Perhaps we should not ignore profiling notes, but I'm going to
1466 -- ride roughshod over them all for now.
1467 --- See Note [Notes in RULE matching] in Rules
1469 argToPat env in_scope val_env (Let _ arg) arg_occ
1470 = argToPat env in_scope val_env arg arg_occ
1471 -- See Note [Matching lets] in Rule.lhs
1472 -- Look through let expressions
1473 -- e.g. f (let v = rhs in (v,w))
1474 -- Here we can specialise for f (v,w)
1475 -- because the rule-matcher will look through the let.
1477 {- Disabled; see Note [Matching cases] in Rule.lhs
1478 argToPat env in_scope val_env (Case scrut _ _ [(_, _, rhs)]) arg_occ
1479 | exprOkForSpeculation scrut -- See Note [Matching cases] in Rule.hhs
1480 = argToPat env in_scope val_env rhs arg_occ
1483 argToPat env in_scope val_env (Cast arg co) arg_occ
1484 | not (ignoreType env ty2)
1485 = do { (interesting, arg') <- argToPat env in_scope val_env arg arg_occ
1486 ; if not interesting then
1489 { -- Make a wild-card pattern for the coercion
1491 ; let co_name = mkSysTvName uniq (fsLit "sg")
1492 co_var = mkCoVar co_name (mkCoKind ty1 ty2)
1493 ; return (interesting, Cast arg' (mkTyVarTy co_var)) } }
1495 (ty1, ty2) = coercionKind co
1499 {- Disabling lambda specialisation for now
1500 It's fragile, and the spec_loop can be infinite
1501 argToPat in_scope val_env arg arg_occ
1503 = return (True, arg)
1505 is_value_lam (Lam v e) -- Spot a value lambda, even if
1506 | isId v = True -- it is inside a type lambda
1507 | otherwise = is_value_lam e
1508 is_value_lam other = False
1511 -- Check for a constructor application
1512 -- NB: this *precedes* the Var case, so that we catch nullary constrs
1513 argToPat env in_scope val_env arg arg_occ
1514 | Just (ConVal dc args) <- isValue val_env arg
1515 , not (ignoreAltCon env dc)
1517 ScrutOcc _ -> True -- Used only by case scrutinee
1518 BothOcc -> case arg of -- Used elsewhere
1519 App {} -> True -- see Note [Reboxing]
1521 _other -> False -- No point; the arg is not decomposed
1522 = do { args' <- argsToPats env in_scope val_env (args `zip` conArgOccs arg_occ dc)
1523 ; return (True, mk_con_app dc (map snd args')) }
1525 -- Check if the argument is a variable that
1526 -- is in scope at the function definition site
1527 -- It's worth specialising on this if
1528 -- (a) it's used in an interesting way in the body
1529 -- (b) we know what its value is
1530 argToPat env in_scope val_env (Var v) arg_occ
1531 | case arg_occ of { UnkOcc -> False; _other -> True }, -- (a)
1533 not (ignoreType env (varType v))
1534 = return (True, Var v)
1537 | isLocalId v = v `elemInScopeSet` in_scope
1538 && isJust (lookupVarEnv val_env v)
1539 -- Local variables have values in val_env
1540 | otherwise = isValueUnfolding (idUnfolding v)
1541 -- Imports have unfoldings
1543 -- I'm really not sure what this comment means
1544 -- And by not wild-carding we tend to get forall'd
1545 -- variables that are in soope, which in turn can
1546 -- expose the weakness in let-matching
1547 -- See Note [Matching lets] in Rules
1549 -- Check for a variable bound inside the function.
1550 -- Don't make a wild-card, because we may usefully share
1551 -- e.g. f a = let x = ... in f (x,x)
1552 -- NB: this case follows the lambda and con-app cases!!
1553 -- argToPat _in_scope _val_env (Var v) _arg_occ
1554 -- = return (False, Var v)
1555 -- SLPJ : disabling this to avoid proliferation of versions
1556 -- also works badly when thinking about seeding the loop
1557 -- from the body of the let
1558 -- f x y = letrec g z = ... in g (x,y)
1559 -- We don't want to specialise for that *particular* x,y
1561 -- The default case: make a wild-card
1562 argToPat _env _in_scope _val_env arg _arg_occ
1563 = wildCardPat (exprType arg)
1565 wildCardPat :: Type -> UniqSM (Bool, CoreArg)
1566 wildCardPat ty = do { uniq <- getUniqueUs
1567 ; let id = mkSysLocal (fsLit "sc") uniq ty
1568 ; return (False, Var id) }
1570 argsToPats :: ScEnv -> InScopeSet -> ValueEnv
1571 -> [(CoreArg, ArgOcc)]
1572 -> UniqSM [(Bool, CoreArg)]
1573 argsToPats env in_scope val_env args
1576 do_one (arg,occ) = argToPat env in_scope val_env arg occ
1581 isValue :: ValueEnv -> CoreExpr -> Maybe Value
1582 isValue _env (Lit lit)
1583 = Just (ConVal (LitAlt lit) [])
1586 | Just stuff <- lookupVarEnv env v
1587 = Just stuff -- You might think we could look in the idUnfolding here
1588 -- but that doesn't take account of which branch of a
1589 -- case we are in, which is the whole point
1591 | not (isLocalId v) && isCheapUnfolding unf
1592 = isValue env (unfoldingTemplate unf)
1595 -- However we do want to consult the unfolding
1596 -- as well, for let-bound constructors!
1598 isValue env (Lam b e)
1599 | isTyCoVar b = case isValue env e of
1600 Just _ -> Just LambdaVal
1602 | otherwise = Just LambdaVal
1604 isValue _env expr -- Maybe it's a constructor application
1605 | (Var fun, args) <- collectArgs expr
1606 = case isDataConWorkId_maybe fun of
1608 Just con | args `lengthAtLeast` dataConRepArity con
1609 -- Check saturated; might be > because the
1610 -- arity excludes type args
1611 -> Just (ConVal (DataAlt con) args)
1613 _other | valArgCount args < idArity fun
1614 -- Under-applied function
1615 -> Just LambdaVal -- Partial application
1619 isValue _env _expr = Nothing
1621 mk_con_app :: AltCon -> [CoreArg] -> CoreExpr
1622 mk_con_app (LitAlt lit) [] = Lit lit
1623 mk_con_app (DataAlt con) args = mkConApp con args
1624 mk_con_app _other _args = panic "SpecConstr.mk_con_app"
1626 samePat :: CallPat -> CallPat -> Bool
1627 samePat (vs1, as1) (vs2, as2)
1630 same (Var v1) (Var v2)
1631 | v1 `elem` vs1 = v2 `elem` vs2
1632 | v2 `elem` vs2 = False
1633 | otherwise = v1 == v2
1635 same (Lit l1) (Lit l2) = l1==l2
1636 same (App f1 a1) (App f2 a2) = same f1 f2 && same a1 a2
1638 same (Type {}) (Type {}) = True -- Note [Ignore type differences]
1639 same (Note _ e1) e2 = same e1 e2 -- Ignore casts and notes
1640 same (Cast e1 _) e2 = same e1 e2
1641 same e1 (Note _ e2) = same e1 e2
1642 same e1 (Cast e2 _) = same e1 e2
1644 same e1 e2 = WARN( bad e1 || bad e2, ppr e1 $$ ppr e2)
1645 False -- Let, lambda, case should not occur
1646 bad (Case {}) = True
1652 Note [Ignore type differences]
1653 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1654 We do not want to generate specialisations where the call patterns
1655 differ only in their type arguments! Not only is it utterly useless,
1656 but it also means that (with polymorphic recursion) we can generate
1657 an infinite number of specialisations. Example is Data.Sequence.adjustTree,