2 % (c) The GRASP/AQUA Project, Glasgow University, 1993-1998
4 \section[Specialise]{Stamping out overloading, and (optionally) polymorphism}
7 module Specialise ( specProgram ) where
9 #include "HsVersions.h"
20 import CoreUtils ( exprIsTrivial, applyTypeToArgs, mkPiTypes )
21 import CoreFVs ( exprFreeVars, exprsFreeVars, idFreeVars )
22 import UniqSupply ( UniqSM, initUs_, MonadUnique(..) )
24 import MkId ( voidArgId, realWorldPrimId )
25 import Maybes ( catMaybes, isJust )
34 import qualified Data.Map as Map
35 import qualified FiniteMap as Map
38 %************************************************************************
40 \subsection[notes-Specialise]{Implementation notes [SLPJ, Aug 18 1993]}
42 %************************************************************************
44 These notes describe how we implement specialisation to eliminate
47 The specialisation pass works on Core
48 syntax, complete with all the explicit dictionary application,
49 abstraction and construction as added by the type checker. The
50 existing type checker remains largely as it is.
52 One important thought: the {\em types} passed to an overloaded
53 function, and the {\em dictionaries} passed are mutually redundant.
54 If the same function is applied to the same type(s) then it is sure to
55 be applied to the same dictionary(s)---or rather to the same {\em
56 values}. (The arguments might look different but they will evaluate
59 Second important thought: we know that we can make progress by
60 treating dictionary arguments as static and worth specialising on. So
61 we can do without binding-time analysis, and instead specialise on
62 dictionary arguments and no others.
71 and suppose f is overloaded.
73 STEP 1: CALL-INSTANCE COLLECTION
75 We traverse <body>, accumulating all applications of f to types and
78 (Might there be partial applications, to just some of its types and
79 dictionaries? In principle yes, but in practice the type checker only
80 builds applications of f to all its types and dictionaries, so partial
81 applications could only arise as a result of transformation, and even
82 then I think it's unlikely. In any case, we simply don't accumulate such
83 partial applications.)
88 So now we have a collection of calls to f:
92 Notice that f may take several type arguments. To avoid ambiguity, we
93 say that f is called at type t1/t2 and t3/t4.
95 We take equivalence classes using equality of the *types* (ignoring
96 the dictionary args, which as mentioned previously are redundant).
98 STEP 3: SPECIALISATION
100 For each equivalence class, choose a representative (f t1 t2 d1 d2),
101 and create a local instance of f, defined thus:
103 f@t1/t2 = <f_rhs> t1 t2 d1 d2
105 f_rhs presumably has some big lambdas and dictionary lambdas, so lots
106 of simplification will now result. However we don't actually *do* that
107 simplification. Rather, we leave it for the simplifier to do. If we
108 *did* do it, though, we'd get more call instances from the specialised
109 RHS. We can work out what they are by instantiating the call-instance
110 set from f's RHS with the types t1, t2.
112 Add this new id to f's IdInfo, to record that f has a specialised version.
114 Before doing any of this, check that f's IdInfo doesn't already
115 tell us about an existing instance of f at the required type/s.
116 (This might happen if specialisation was applied more than once, or
117 it might arise from user SPECIALIZE pragmas.)
121 Wait a minute! What if f is recursive? Then we can't just plug in
122 its right-hand side, can we?
124 But it's ok. The type checker *always* creates non-recursive definitions
125 for overloaded recursive functions. For example:
127 f x = f (x+x) -- Yes I know its silly
131 f a (d::Num a) = let p = +.sel a d
133 letrec fl (y::a) = fl (p y y)
137 We still have recusion for non-overloaded functions which we
138 speciailise, but the recursive call should get specialised to the
139 same recursive version.
145 All this is crystal clear when the function is applied to *constant
146 types*; that is, types which have no type variables inside. But what if
147 it is applied to non-constant types? Suppose we find a call of f at type
148 t1/t2. There are two possibilities:
150 (a) The free type variables of t1, t2 are in scope at the definition point
151 of f. In this case there's no problem, we proceed just as before. A common
152 example is as follows. Here's the Haskell:
157 After typechecking we have
159 g a (d::Num a) (y::a) = let f b (d'::Num b) (x::b) = +.sel b d' x x
160 in +.sel a d (f a d y) (f a d y)
162 Notice that the call to f is at type type "a"; a non-constant type.
163 Both calls to f are at the same type, so we can specialise to give:
165 g a (d::Num a) (y::a) = let f@a (x::a) = +.sel a d x x
166 in +.sel a d (f@a y) (f@a y)
169 (b) The other case is when the type variables in the instance types
170 are *not* in scope at the definition point of f. The example we are
171 working with above is a good case. There are two instances of (+.sel a d),
172 but "a" is not in scope at the definition of +.sel. Can we do anything?
173 Yes, we can "common them up", a sort of limited common sub-expression deal.
176 g a (d::Num a) (y::a) = let +.sel@a = +.sel a d
177 f@a (x::a) = +.sel@a x x
178 in +.sel@a (f@a y) (f@a y)
180 This can save work, and can't be spotted by the type checker, because
181 the two instances of +.sel weren't originally at the same type.
185 * There are quite a few variations here. For example, the defn of
186 +.sel could be floated ouside the \y, to attempt to gain laziness.
187 It certainly mustn't be floated outside the \d because the d has to
190 * We don't want to inline f_rhs in this case, because
191 that will duplicate code. Just commoning up the call is the point.
193 * Nothing gets added to +.sel's IdInfo.
195 * Don't bother unless the equivalence class has more than one item!
197 Not clear whether this is all worth it. It is of course OK to
198 simply discard call-instances when passing a big lambda.
200 Polymorphism 2 -- Overloading
202 Consider a function whose most general type is
204 f :: forall a b. Ord a => [a] -> b -> b
206 There is really no point in making a version of g at Int/Int and another
207 at Int/Bool, because it's only instancing the type variable "a" which
208 buys us any efficiency. Since g is completely polymorphic in b there
209 ain't much point in making separate versions of g for the different
212 That suggests that we should identify which of g's type variables
213 are constrained (like "a") and which are unconstrained (like "b").
214 Then when taking equivalence classes in STEP 2, we ignore the type args
215 corresponding to unconstrained type variable. In STEP 3 we make
216 polymorphic versions. Thus:
218 f@t1/ = /\b -> <f_rhs> t1 b d1 d2
227 f a (d::Num a) = let g = ...
229 ...(let d1::Ord a = Num.Ord.sel a d in g a d1)...
231 Here, g is only called at one type, but the dictionary isn't in scope at the
232 definition point for g. Usually the type checker would build a
233 definition for d1 which enclosed g, but the transformation system
234 might have moved d1's defn inward. Solution: float dictionary bindings
235 outwards along with call instances.
239 f x = let g p q = p==q
245 Before specialisation, leaving out type abstractions we have
247 f df x = let g :: Eq a => a -> a -> Bool
249 h :: Num a => a -> a -> (a, Bool)
250 h dh r s = let deq = eqFromNum dh
251 in (+ dh r s, g deq r s)
255 After specialising h we get a specialised version of h, like this:
257 h' r s = let deq = eqFromNum df
258 in (+ df r s, g deq r s)
260 But we can't naively make an instance for g from this, because deq is not in scope
261 at the defn of g. Instead, we have to float out the (new) defn of deq
262 to widen its scope. Notice that this floating can't be done in advance -- it only
263 shows up when specialisation is done.
265 User SPECIALIZE pragmas
266 ~~~~~~~~~~~~~~~~~~~~~~~
267 Specialisation pragmas can be digested by the type checker, and implemented
268 by adding extra definitions along with that of f, in the same way as before
270 f@t1/t2 = <f_rhs> t1 t2 d1 d2
272 Indeed the pragmas *have* to be dealt with by the type checker, because
273 only it knows how to build the dictionaries d1 and d2! For example
275 g :: Ord a => [a] -> [a]
276 {-# SPECIALIZE f :: [Tree Int] -> [Tree Int] #-}
278 Here, the specialised version of g is an application of g's rhs to the
279 Ord dictionary for (Tree Int), which only the type checker can conjure
280 up. There might not even *be* one, if (Tree Int) is not an instance of
281 Ord! (All the other specialision has suitable dictionaries to hand
284 Problem. The type checker doesn't have to hand a convenient <f_rhs>, because
285 it is buried in a complex (as-yet-un-desugared) binding group.
288 f@t1/t2 = f* t1 t2 d1 d2
290 where f* is the Id f with an IdInfo which says "inline me regardless!".
291 Indeed all the specialisation could be done in this way.
292 That in turn means that the simplifier has to be prepared to inline absolutely
293 any in-scope let-bound thing.
296 Again, the pragma should permit polymorphism in unconstrained variables:
298 h :: Ord a => [a] -> b -> b
299 {-# SPECIALIZE h :: [Int] -> b -> b #-}
301 We *insist* that all overloaded type variables are specialised to ground types,
302 (and hence there can be no context inside a SPECIALIZE pragma).
303 We *permit* unconstrained type variables to be specialised to
305 - or left as a polymorphic type variable
306 but nothing in between. So
308 {-# SPECIALIZE h :: [Int] -> [c] -> [c] #-}
310 is *illegal*. (It can be handled, but it adds complication, and gains the
314 SPECIALISING INSTANCE DECLARATIONS
315 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
318 instance Foo a => Foo [a] where
320 {-# SPECIALIZE instance Foo [Int] #-}
322 The original instance decl creates a dictionary-function
325 dfun.Foo.List :: forall a. Foo a -> Foo [a]
327 The SPECIALIZE pragma just makes a specialised copy, just as for
328 ordinary function definitions:
330 dfun.Foo.List@Int :: Foo [Int]
331 dfun.Foo.List@Int = dfun.Foo.List Int dFooInt
333 The information about what instance of the dfun exist gets added to
334 the dfun's IdInfo in the same way as a user-defined function too.
337 Automatic instance decl specialisation?
338 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
339 Can instance decls be specialised automatically? It's tricky.
340 We could collect call-instance information for each dfun, but
341 then when we specialised their bodies we'd get new call-instances
342 for ordinary functions; and when we specialised their bodies, we might get
343 new call-instances of the dfuns, and so on. This all arises because of
344 the unrestricted mutual recursion between instance decls and value decls.
346 Still, there's no actual problem; it just means that we may not do all
347 the specialisation we could theoretically do.
349 Furthermore, instance decls are usually exported and used non-locally,
350 so we'll want to compile enough to get those specialisations done.
352 Lastly, there's no such thing as a local instance decl, so we can
353 survive solely by spitting out *usage* information, and then reading that
354 back in as a pragma when next compiling the file. So for now,
355 we only specialise instance decls in response to pragmas.
358 SPITTING OUT USAGE INFORMATION
359 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
361 To spit out usage information we need to traverse the code collecting
362 call-instance information for all imported (non-prelude?) functions
363 and data types. Then we equivalence-class it and spit it out.
365 This is done at the top-level when all the call instances which escape
366 must be for imported functions and data types.
368 *** Not currently done ***
371 Partial specialisation by pragmas
372 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
373 What about partial specialisation:
375 k :: (Ord a, Eq b) => [a] -> b -> b -> [a]
376 {-# SPECIALIZE k :: Eq b => [Int] -> b -> b -> [a] #-}
380 {-# SPECIALIZE k :: Eq b => [Int] -> [b] -> [b] -> [a] #-}
382 Seems quite reasonable. Similar things could be done with instance decls:
384 instance (Foo a, Foo b) => Foo (a,b) where
386 {-# SPECIALIZE instance Foo a => Foo (a,Int) #-}
387 {-# SPECIALIZE instance Foo b => Foo (Int,b) #-}
389 Ho hum. Things are complex enough without this. I pass.
392 Requirements for the simplifer
393 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
394 The simplifier has to be able to take advantage of the specialisation.
396 * When the simplifier finds an application of a polymorphic f, it looks in
397 f's IdInfo in case there is a suitable instance to call instead. This converts
399 f t1 t2 d1 d2 ===> f_t1_t2
401 Note that the dictionaries get eaten up too!
403 * Dictionary selection operations on constant dictionaries must be
406 +.sel Int d ===> +Int
408 The obvious way to do this is in the same way as other specialised
409 calls: +.sel has inside it some IdInfo which tells that if it's applied
410 to the type Int then it should eat a dictionary and transform to +Int.
412 In short, dictionary selectors need IdInfo inside them for constant
415 * Exactly the same applies if a superclass dictionary is being
418 Eq.sel Int d ===> dEqInt
420 * Something similar applies to dictionary construction too. Suppose
421 dfun.Eq.List is the function taking a dictionary for (Eq a) to
422 one for (Eq [a]). Then we want
424 dfun.Eq.List Int d ===> dEq.List_Int
426 Where does the Eq [Int] dictionary come from? It is built in
427 response to a SPECIALIZE pragma on the Eq [a] instance decl.
429 In short, dfun Ids need IdInfo with a specialisation for each
430 constant instance of their instance declaration.
432 All this uses a single mechanism: the SpecEnv inside an Id
435 What does the specialisation IdInfo look like?
436 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
438 The SpecEnv of an Id maps a list of types (the template) to an expression
442 For example, if f has this SpecInfo:
444 [Int, a] -> \d:Ord Int. f' a
446 it means that we can replace the call
448 f Int t ===> (\d. f' t)
450 This chucks one dictionary away and proceeds with the
451 specialised version of f, namely f'.
454 What can't be done this way?
455 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
456 There is no way, post-typechecker, to get a dictionary for (say)
457 Eq a from a dictionary for Eq [a]. So if we find
461 we can't transform to
466 eqList :: (a->a->Bool) -> [a] -> [a] -> Bool
468 Of course, we currently have no way to automatically derive
469 eqList, nor to connect it to the Eq [a] instance decl, but you
470 can imagine that it might somehow be possible. Taking advantage
471 of this is permanently ruled out.
473 Still, this is no great hardship, because we intend to eliminate
474 overloading altogether anyway!
476 A note about non-tyvar dictionaries
477 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
478 Some Ids have types like
480 forall a,b,c. Eq a -> Ord [a] -> tau
482 This seems curious at first, because we usually only have dictionary
483 args whose types are of the form (C a) where a is a type variable.
484 But this doesn't hold for the functions arising from instance decls,
485 which sometimes get arguements with types of form (C (T a)) for some
488 Should we specialise wrt this compound-type dictionary? We used to say
490 "This is a heuristic judgement, as indeed is the fact that we
491 specialise wrt only dictionaries. We choose *not* to specialise
492 wrt compound dictionaries because at the moment the only place
493 they show up is in instance decls, where they are simply plugged
494 into a returned dictionary. So nothing is gained by specialising
497 But it is simpler and more uniform to specialise wrt these dicts too;
498 and in future GHC is likely to support full fledged type signatures
500 f :: Eq [(a,b)] => ...
503 %************************************************************************
505 \subsubsection{The new specialiser}
507 %************************************************************************
509 Our basic game plan is this. For let(rec) bound function
510 f :: (C a, D c) => (a,b,c,d) -> Bool
512 * Find any specialised calls of f, (f ts ds), where
513 ts are the type arguments t1 .. t4, and
514 ds are the dictionary arguments d1 .. d2.
516 * Add a new definition for f1 (say):
518 f1 = /\ b d -> (..body of f..) t1 b t3 d d1 d2
520 Note that we abstract over the unconstrained type arguments.
524 [t1,b,t3,d] |-> \d1 d2 -> f1 b d
526 to the specialisations of f. This will be used by the
527 simplifier to replace calls
528 (f t1 t2 t3 t4) da db
530 (\d1 d1 -> f1 t2 t4) da db
532 All the stuff about how many dictionaries to discard, and what types
533 to apply the specialised function to, are handled by the fact that the
534 SpecEnv contains a template for the result of the specialisation.
536 We don't build *partial* specialisations for f. For example:
538 f :: Eq a => a -> a -> Bool
539 {-# SPECIALISE f :: (Eq b, Eq c) => (b,c) -> (b,c) -> Bool #-}
541 Here, little is gained by making a specialised copy of f.
542 There's a distinct danger that the specialised version would
543 first build a dictionary for (Eq b, Eq c), and then select the (==)
544 method from it! Even if it didn't, not a great deal is saved.
546 We do, however, generate polymorphic, but not overloaded, specialisations:
548 f :: Eq a => [a] -> b -> b -> b
549 {#- SPECIALISE f :: [Int] -> b -> b -> b #-}
551 Hence, the invariant is this:
553 *** no specialised version is overloaded ***
556 %************************************************************************
558 \subsubsection{The exported function}
560 %************************************************************************
563 specProgram :: ModGuts -> CoreM ModGuts
565 = do { hpt_rules <- getRuleBase
566 ; let local_rules = mg_rules guts
567 rule_base = extendRuleBaseList hpt_rules (mg_rules guts)
569 -- Specialise the bindings of this module
570 ; (binds', uds) <- runSpecM (go (mg_binds guts))
572 -- Specialise imported functions
573 ; (new_rules, spec_binds) <- specImports emptyVarSet rule_base uds
575 ; let final_binds | null spec_binds = binds'
576 | otherwise = Rec (flattenBinds spec_binds) : binds'
577 -- Note [Glom the bindings if imported functions are specialised]
579 ; return (guts { mg_binds = final_binds
580 , mg_rules = new_rules ++ local_rules }) }
582 -- We need to start with a Subst that knows all the things
583 -- that are in scope, so that the substitution engine doesn't
584 -- accidentally re-use a unique that's already in use
585 -- Easiest thing is to do it all at once, as if all the top-level
586 -- decls were mutually recursive
587 top_subst = mkEmptySubst $ mkInScopeSet $ mkVarSet $
588 bindersOfBinds $ mg_binds guts
590 go [] = return ([], emptyUDs)
591 go (bind:binds) = do (binds', uds) <- go binds
592 (bind', uds') <- specBind top_subst bind uds
593 return (bind' ++ binds', uds')
595 specImports :: VarSet -- Don't specialise these ones
596 -- See Note [Avoiding recursive specialisation]
597 -> RuleBase -- Rules from this module and the home package
598 -- (but not external packages, which can change)
599 -> UsageDetails -- Calls for imported things, and floating bindings
600 -> CoreM ( [CoreRule] -- New rules
601 , [CoreBind] ) -- Specialised bindings and floating bindings
602 -- See Note [Specialise imported INLINABLE things]
603 specImports done rb uds
604 = do { let import_calls = varEnvElts (ud_calls uds)
605 ; (rules, spec_binds) <- go rb import_calls
606 ; return (rules, wrapDictBinds (ud_binds uds) spec_binds) }
608 go _ [] = return ([], [])
609 go rb (CIS fn calls_for_fn : other_calls)
610 = do { (rules1, spec_binds1) <- specImport done rb fn (Map.toList calls_for_fn)
611 ; (rules2, spec_binds2) <- go (extendRuleBaseList rb rules1) other_calls
612 ; return (rules1 ++ rules2, spec_binds1 ++ spec_binds2) }
614 specImport :: VarSet -- Don't specialise these
615 -- See Note [Avoiding recursive specialisation]
616 -> RuleBase -- Rules from this module
617 -> Id -> [CallInfo] -- Imported function and calls for it
618 -> CoreM ( [CoreRule] -- New rules
619 , [CoreBind] ) -- Specialised bindings
620 specImport done rb fn calls_for_fn
621 | fn `elemVarSet` done
622 = return ([], []) -- No warning. This actually happens all the time
623 -- when specialising a recursive function, becuase
624 -- the RHS of the specialised function contains a recursive
625 -- call to the original function
627 | isInlinablePragma (idInlinePragma fn)
628 , Just rhs <- maybeUnfoldingTemplate (realIdUnfolding fn)
629 = do { -- Get rules from the external package state
630 -- We keep doing this in case we "page-fault in"
631 -- more rules as we go along
632 ; hsc_env <- getHscEnv
633 ; eps <- liftIO $ hscEPS hsc_env
634 ; let full_rb = unionRuleBase rb (eps_rule_base eps)
635 rules_for_fn = getRules full_rb fn
637 ; (rules1, spec_pairs, uds) <- runSpecM $
638 specCalls emptySubst rules_for_fn calls_for_fn fn rhs
639 ; let spec_binds1 = [NonRec b r | (b,r) <- spec_pairs]
640 -- After the rules kick in we may get recursion, but
641 -- we rely on a global GlomBinds to sort that out later
642 -- See Note [Glom the bindings if imported functions are specialised]
644 -- Now specialise any cascaded calls
645 ; (rules2, spec_binds2) <- specImports (extendVarSet done fn)
646 (extendRuleBaseList rb rules1)
649 ; return (rules2 ++ rules1, spec_binds2 ++ spec_binds1) }
652 = WARN( True, ptext (sLit "specImport discard") <+> ppr fn <+> ppr calls_for_fn )
656 Note [Specialise imported INLINABLE things]
657 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
658 We specialise INLINABLE things but not INLINE things. The latter
659 should be inlined bodily, so not much point in specialising them.
660 Moreover, we risk lots of orphan modules from vigorous specialisation.
662 Note [Glom the bindings if imported functions are specialised]
663 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
664 Suppose we have an imported, *recursive*, INLINABLE function
666 f = /\a \d x. ...(f a d)...
667 In the module being compiled we have
669 Now we'll make a specialised function
671 f_spec = \x -> ...(f Int dInt)...
672 {-# RULE f Int _ = f_spec #-}
674 Note that f_spec doesn't look recursive
675 After rewriting with the RULE, we get
676 f_spec = \x -> ...(f_spec)...
677 BUT since f_spec was non-recursive before it'll *stay* non-recursive.
678 The occurrence analyser never turns a NonRec into a Rec. So we must
679 make sure that f_spec is recursive. Easiest thing is to make all
680 the specialisations for imported bindings recursive.
683 Note [Avoiding recursive specialisation]
684 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
685 When we specialise 'f' we may find new overloaded calls to 'g', 'h' in
686 'f's RHS. So we want to specialise g,h. But we don't want to
687 specialise f any more! It's possible that f's RHS might have a
688 recursive yet-more-specialised call, so we'd diverge in that case.
689 And if the call is to the same type, one specialisation is enough.
690 Avoiding this recursive specialisation loop is the reason for the
691 'done' VarSet passed to specImports and specImport.
693 %************************************************************************
695 \subsubsection{@specExpr@: the main function}
697 %************************************************************************
700 specVar :: Subst -> Id -> CoreExpr
701 specVar subst v = lookupIdSubst (text "specVar") subst v
703 specExpr :: Subst -> CoreExpr -> SpecM (CoreExpr, UsageDetails)
704 -- We carry a substitution down:
705 -- a) we must clone any binding that might float outwards,
706 -- to avoid name clashes
707 -- b) we carry a type substitution to use when analysing
708 -- the RHS of specialised bindings (no type-let!)
710 ---------------- First the easy cases --------------------
711 specExpr subst (Type ty) = return (Type (CoreSubst.substTy subst ty), emptyUDs)
712 specExpr subst (Coercion co) = return (Coercion (CoreSubst.substCo subst co), emptyUDs)
713 specExpr subst (Var v) = return (specVar subst v, emptyUDs)
714 specExpr _ (Lit lit) = return (Lit lit, emptyUDs)
715 specExpr subst (Cast e co) = do
716 (e', uds) <- specExpr subst e
717 return ((Cast e' (CoreSubst.substCo subst co)), uds)
718 specExpr subst (Note note body) = do
719 (body', uds) <- specExpr subst body
720 return (Note (specNote subst note) body', uds)
723 ---------------- Applications might generate a call instance --------------------
724 specExpr subst expr@(App {})
727 go (App fun arg) args = do (arg', uds_arg) <- specExpr subst arg
728 (fun', uds_app) <- go fun (arg':args)
729 return (App fun' arg', uds_arg `plusUDs` uds_app)
731 go (Var f) args = case specVar subst f of
732 Var f' -> return (Var f', mkCallUDs f' args)
733 e' -> return (e', emptyUDs) -- I don't expect this!
734 go other _ = specExpr subst other
736 ---------------- Lambda/case require dumping of usage details --------------------
737 specExpr subst e@(Lam _ _) = do
738 (body', uds) <- specExpr subst' body
739 let (free_uds, dumped_dbs) = dumpUDs bndrs' uds
740 return (mkLams bndrs' (wrapDictBindsE dumped_dbs body'), free_uds)
742 (bndrs, body) = collectBinders e
743 (subst', bndrs') = substBndrs subst bndrs
744 -- More efficient to collect a group of binders together all at once
745 -- and we don't want to split a lambda group with dumped bindings
747 specExpr subst (Case scrut case_bndr ty alts)
748 = do { (scrut', scrut_uds) <- specExpr subst scrut
749 ; (scrut'', case_bndr', alts', alts_uds)
750 <- specCase subst scrut' case_bndr alts
751 ; return (Case scrut'' case_bndr' (CoreSubst.substTy subst ty) alts'
752 , scrut_uds `plusUDs` alts_uds) }
754 ---------------- Finally, let is the interesting case --------------------
755 specExpr subst (Let bind body) = do
757 (rhs_subst, body_subst, bind') <- cloneBindSM subst bind
759 -- Deal with the body
760 (body', body_uds) <- specExpr body_subst body
762 -- Deal with the bindings
763 (binds', uds) <- specBind rhs_subst bind' body_uds
766 return (foldr Let body' binds', uds)
768 -- Must apply the type substitution to coerceions
769 specNote :: Subst -> Note -> Note
770 specNote _ note = note
774 -> CoreExpr -- Scrutinee, already done
776 -> SpecM ( CoreExpr -- New scrutinee
780 specCase subst scrut' case_bndr [(con, args, rhs)]
781 | isDictId case_bndr -- See Note [Floating dictionaries out of cases]
782 , interestingDict scrut'
783 , not (isDeadBinder case_bndr && null sc_args')
784 = do { (case_bndr_flt : sc_args_flt) <- mapM clone_me (case_bndr' : sc_args')
786 ; let sc_rhss = [ Case (Var case_bndr_flt) case_bndr' (idType sc_arg')
787 [(con, args', Var sc_arg')]
788 | sc_arg' <- sc_args' ]
790 -- Extend the substitution for RHS to map the *original* binders
791 -- to their floated verions. Attach an unfolding to these floated
792 -- binders so they look interesting to interestingDict
793 mb_sc_flts :: [Maybe DictId]
794 mb_sc_flts = map (lookupVarEnv clone_env) args'
795 clone_env = zipVarEnv sc_args' (zipWith add_unf sc_args_flt sc_rhss)
796 subst_prs = (case_bndr, Var (add_unf case_bndr_flt scrut'))
797 : [ (arg, Var sc_flt)
798 | (arg, Just sc_flt) <- args `zip` mb_sc_flts ]
799 subst_rhs' = extendIdSubstList subst_rhs subst_prs
801 ; (rhs', rhs_uds) <- specExpr subst_rhs' rhs
802 ; let scrut_bind = mkDB (NonRec case_bndr_flt scrut')
803 case_bndr_set = unitVarSet case_bndr_flt
804 sc_binds = [(NonRec sc_arg_flt sc_rhs, case_bndr_set)
805 | (sc_arg_flt, sc_rhs) <- sc_args_flt `zip` sc_rhss ]
806 flt_binds = scrut_bind : sc_binds
807 (free_uds, dumped_dbs) = dumpUDs (case_bndr':args') rhs_uds
808 all_uds = flt_binds `addDictBinds` free_uds
809 alt' = (con, args', wrapDictBindsE dumped_dbs rhs')
810 ; return (Var case_bndr_flt, case_bndr', [alt'], all_uds) }
812 (subst_rhs, (case_bndr':args')) = substBndrs subst (case_bndr:args)
813 sc_args' = filter is_flt_sc_arg args'
815 clone_me bndr = do { uniq <- getUniqueM
816 ; return (mkUserLocal occ uniq ty loc) }
820 occ = nameOccName name
821 loc = getSrcSpan name
823 add_unf sc_flt sc_rhs -- Sole purpose: make sc_flt respond True to interestingDictId
824 = setIdUnfolding sc_flt (mkSimpleUnfolding sc_rhs)
826 arg_set = mkVarSet args'
827 is_flt_sc_arg var = isId var
828 && not (isDeadBinder var)
830 && not (tyVarsOfType var_ty `intersectsVarSet` arg_set)
835 specCase subst scrut case_bndr alts
836 = do { (alts', uds_alts) <- mapAndCombineSM spec_alt alts
837 ; return (scrut, case_bndr', alts', uds_alts) }
839 (subst_alt, case_bndr') = substBndr subst case_bndr
840 spec_alt (con, args, rhs) = do
841 (rhs', uds) <- specExpr subst_rhs rhs
842 let (free_uds, dumped_dbs) = dumpUDs (case_bndr' : args') uds
843 return ((con, args', wrapDictBindsE dumped_dbs rhs'), free_uds)
845 (subst_rhs, args') = substBndrs subst_alt args
848 Note [Floating dictionaries out of cases]
849 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
851 g = \d. case d of { MkD sc ... -> ...(f sc)... }
852 Naively we can't float d2's binding out of the case expression,
853 because 'sc' is bound by the case, and that in turn means we can't
854 specialise f, which seems a pity.
856 So we invert the case, by floating out a binding
858 sc_flt = case d of { MkD sc ... -> sc }
859 Now we can float the call instance for 'f'. Indeed this is just
860 what'll happen if 'sc' was originally bound with a let binding,
861 but case is more efficient, and necessary with equalities. So it's
862 good to work with both.
864 You might think that this won't make any difference, because the
865 call instance will only get nuked by the \d. BUT if 'g' itself is
866 specialised, then transitively we should be able to specialise f.
869 case e of cb { MkD sc ... -> ...(f sc)... }
872 sc_flt = case cb_flt of { MkD sc ... -> sc }
874 case cb_flt of bg { MkD sc ... -> ....(f sc_flt)... }
876 The "_flt" things are the floated binds; we use the current substitution
877 to substitute sc -> sc_flt in the RHS
879 %************************************************************************
881 Dealing with a binding
883 %************************************************************************
886 specBind :: Subst -- Use this for RHSs
888 -> UsageDetails -- Info on how the scope of the binding
889 -> SpecM ([CoreBind], -- New bindings
890 UsageDetails) -- And info to pass upstream
892 -- Returned UsageDetails:
893 -- No calls for binders of this bind
894 specBind rhs_subst (NonRec fn rhs) body_uds
895 = do { (rhs', rhs_uds) <- specExpr rhs_subst rhs
896 ; (fn', spec_defns, body_uds1) <- specDefn rhs_subst body_uds fn rhs
898 ; let pairs = spec_defns ++ [(fn', rhs')]
899 -- fn' mentions the spec_defns in its rules,
900 -- so put the latter first
902 combined_uds = body_uds1 `plusUDs` rhs_uds
903 -- This way round a call in rhs_uds of a function f
904 -- at type T will override a call of f at T in body_uds1; and
905 -- that is good because it'll tend to keep "earlier" calls
906 -- See Note [Specialisation of dictionary functions]
908 (free_uds, dump_dbs, float_all) = dumpBindUDs [fn] combined_uds
909 -- See Note [From non-recursive to recursive]
911 final_binds | isEmptyBag dump_dbs = [NonRec b r | (b,r) <- pairs]
912 | otherwise = [Rec (flattenDictBinds dump_dbs pairs)]
915 -- Rather than discard the calls mentioning the bound variables
916 -- we float this binding along with the others
917 return ([], free_uds `snocDictBinds` final_binds)
919 -- No call in final_uds mentions bound variables,
920 -- so we can just leave the binding here
921 return (final_binds, free_uds) }
924 specBind rhs_subst (Rec pairs) body_uds
925 -- Note [Specialising a recursive group]
926 = do { let (bndrs,rhss) = unzip pairs
927 ; (rhss', rhs_uds) <- mapAndCombineSM (specExpr rhs_subst) rhss
928 ; let scope_uds = body_uds `plusUDs` rhs_uds
929 -- Includes binds and calls arising from rhss
931 ; (bndrs1, spec_defns1, uds1) <- specDefns rhs_subst scope_uds pairs
933 ; (bndrs3, spec_defns3, uds3)
934 <- if null spec_defns1 -- Common case: no specialisation
935 then return (bndrs1, [], uds1)
936 else do { -- Specialisation occurred; do it again
937 (bndrs2, spec_defns2, uds2)
938 <- specDefns rhs_subst uds1 (bndrs1 `zip` rhss)
939 ; return (bndrs2, spec_defns2 ++ spec_defns1, uds2) }
941 ; let (final_uds, dumped_dbs, float_all) = dumpBindUDs bndrs uds3
942 bind = Rec (flattenDictBinds dumped_dbs $
943 spec_defns3 ++ zip bndrs3 rhss')
946 return ([], final_uds `snocDictBind` bind)
948 return ([bind], final_uds) }
951 ---------------------------
953 -> UsageDetails -- Info on how it is used in its scope
954 -> [(Id,CoreExpr)] -- The things being bound and their un-processed RHS
955 -> SpecM ([Id], -- Original Ids with RULES added
956 [(Id,CoreExpr)], -- Extra, specialised bindings
957 UsageDetails) -- Stuff to fling upwards from the specialised versions
959 -- Specialise a list of bindings (the contents of a Rec), but flowing usages
960 -- upwards binding by binding. Example: { f = ...g ...; g = ...f .... }
961 -- Then if the input CallDetails has a specialised call for 'g', whose specialisation
962 -- in turn generates a specialised call for 'f', we catch that in this one sweep.
963 -- But not vice versa (it's a fixpoint problem).
965 specDefns _subst uds []
966 = return ([], [], uds)
967 specDefns subst uds ((bndr,rhs):pairs)
968 = do { (bndrs1, spec_defns1, uds1) <- specDefns subst uds pairs
969 ; (bndr1, spec_defns2, uds2) <- specDefn subst uds1 bndr rhs
970 ; return (bndr1 : bndrs1, spec_defns1 ++ spec_defns2, uds2) }
972 ---------------------------
974 -> UsageDetails -- Info on how it is used in its scope
975 -> Id -> CoreExpr -- The thing being bound and its un-processed RHS
976 -> SpecM (Id, -- Original Id with added RULES
977 [(Id,CoreExpr)], -- Extra, specialised bindings
978 UsageDetails) -- Stuff to fling upwards from the specialised versions
980 specDefn subst body_uds fn rhs
981 = do { let (body_uds_without_me, calls_for_me) = callsForMe fn body_uds
982 rules_for_me = idCoreRules fn
983 ; (rules, spec_defns, spec_uds) <- specCalls subst rules_for_me
985 ; return ( fn `addIdSpecialisations` rules
987 , body_uds_without_me `plusUDs` spec_uds) }
988 -- It's important that the `plusUDs` is this way
989 -- round, because body_uds_without_me may bind
990 -- dictionaries that are used in calls_for_me passed
991 -- to specDefn. So the dictionary bindings in
992 -- spec_uds may mention dictionaries bound in
993 -- body_uds_without_me
995 ---------------------------
997 -> [CoreRule] -- Existing RULES for the fn
1000 -> SpecM ([CoreRule], -- New RULES for the fn
1001 [(Id,CoreExpr)], -- Extra, specialised bindings
1002 UsageDetails) -- New usage details from the specialised RHSs
1004 -- This function checks existing rules, and does not create
1005 -- duplicate ones. So the caller does not need to do this filtering.
1006 -- See 'already_covered'
1008 specCalls subst rules_for_me calls_for_me fn rhs
1009 -- The first case is the interesting one
1010 | rhs_tyvars `lengthIs` n_tyvars -- Rhs of fn's defn has right number of big lambdas
1011 && rhs_ids `lengthAtLeast` n_dicts -- and enough dict args
1012 && notNull calls_for_me -- And there are some calls to specialise
1013 && not (isNeverActive (idInlineActivation fn))
1014 -- Don't specialise NOINLINE things
1015 -- See Note [Auto-specialisation and RULES]
1017 -- && not (certainlyWillInline (idUnfolding fn)) -- And it's not small
1018 -- See Note [Inline specialisation] for why we do not
1019 -- switch off specialisation for inline functions
1021 = -- pprTrace "specDefn: some" (ppr fn $$ ppr calls_for_me $$ ppr rules_for_me) $
1022 do { stuff <- mapM spec_call calls_for_me
1023 ; let (spec_defns, spec_uds, spec_rules) = unzip3 (catMaybes stuff)
1024 ; return (spec_rules, spec_defns, plusUDList spec_uds) }
1026 | otherwise -- No calls or RHS doesn't fit our preconceptions
1027 = WARN( notNull calls_for_me, ptext (sLit "Missed specialisation opportunity for")
1028 <+> ppr fn $$ _trace_doc )
1029 -- Note [Specialisation shape]
1030 -- pprTrace "specDefn: none" (ppr fn $$ ppr calls_for_me) $
1031 return ([], [], emptyUDs)
1033 _trace_doc = vcat [ ppr rhs_tyvars, ppr n_tyvars
1034 , ppr rhs_ids, ppr n_dicts
1035 , ppr (idInlineActivation fn) ]
1038 fn_arity = idArity fn
1039 fn_unf = realIdUnfolding fn -- Ignore loop-breaker-ness here
1040 (tyvars, theta, _) = tcSplitSigmaTy fn_type
1041 n_tyvars = length tyvars
1042 n_dicts = length theta
1043 inl_prag = idInlinePragma fn
1044 inl_act = inlinePragmaActivation inl_prag
1045 is_local = isLocalId fn
1047 -- Figure out whether the function has an INLINE pragma
1048 -- See Note [Inline specialisations]
1050 spec_arity = unfoldingArity fn_unf - n_dicts -- Arity of the *specialised* inline rule
1052 (rhs_tyvars, rhs_ids, rhs_body) = collectTyAndValBinders rhs
1054 rhs_dict_ids = take n_dicts rhs_ids
1055 body = mkLams (drop n_dicts rhs_ids) rhs_body
1056 -- Glue back on the non-dict lambdas
1058 already_covered :: [CoreExpr] -> Bool
1059 already_covered args -- Note [Specialisations already covered]
1060 = isJust (lookupRule (const True) realIdUnfolding
1061 (substInScope subst)
1062 fn args rules_for_me)
1064 mk_ty_args :: [Maybe Type] -> [CoreExpr]
1065 mk_ty_args call_ts = zipWithEqual "spec_call" mk_ty_arg rhs_tyvars call_ts
1067 mk_ty_arg rhs_tyvar Nothing = Type (mkTyVarTy rhs_tyvar)
1068 mk_ty_arg _ (Just ty) = Type ty
1070 ----------------------------------------------------------
1071 -- Specialise to one particular call pattern
1072 spec_call :: CallInfo -- Call instance
1073 -> SpecM (Maybe ((Id,CoreExpr), -- Specialised definition
1074 UsageDetails, -- Usage details from specialised body
1075 CoreRule)) -- Info for the Id's SpecEnv
1076 spec_call (CallKey call_ts, (call_ds, _))
1077 = ASSERT( call_ts `lengthIs` n_tyvars && call_ds `lengthIs` n_dicts )
1079 -- Suppose f's defn is f = /\ a b c -> \ d1 d2 -> rhs
1080 -- Supppose the call is for f [Just t1, Nothing, Just t3] [dx1, dx2]
1082 -- Construct the new binding
1083 -- f1 = SUBST[a->t1,c->t3, d1->d1', d2->d2'] (/\ b -> rhs)
1084 -- PLUS the usage-details
1085 -- { d1' = dx1; d2' = dx2 }
1086 -- where d1', d2' are cloned versions of d1,d2, with the type substitution
1087 -- applied. These auxiliary bindings just avoid duplication of dx1, dx2
1089 -- Note that the substitution is applied to the whole thing.
1090 -- This is convenient, but just slightly fragile. Notably:
1091 -- * There had better be no name clashes in a/b/c
1093 -- poly_tyvars = [b] in the example above
1094 -- spec_tyvars = [a,c]
1095 -- ty_args = [t1,b,t3]
1096 poly_tyvars = [tv | (tv, Nothing) <- rhs_tyvars `zip` call_ts]
1097 spec_tv_binds = [(tv,ty) | (tv, Just ty) <- rhs_tyvars `zip` call_ts]
1098 spec_ty_args = map snd spec_tv_binds
1099 ty_args = mk_ty_args call_ts
1100 rhs_subst = CoreSubst.extendTvSubstList subst spec_tv_binds
1102 ; (rhs_subst1, inst_dict_ids) <- newDictBndrs rhs_subst rhs_dict_ids
1103 -- Clone rhs_dicts, including instantiating their types
1105 ; let (rhs_subst2, dx_binds) = bindAuxiliaryDicts rhs_subst1 $
1106 (my_zipEqual rhs_dict_ids inst_dict_ids call_ds)
1107 inst_args = ty_args ++ map Var inst_dict_ids
1109 ; if already_covered inst_args then
1112 { -- Figure out the type of the specialised function
1113 let body_ty = applyTypeToArgs rhs fn_type inst_args
1114 (lam_args, app_args) -- Add a dummy argument if body_ty is unlifted
1115 | isUnLiftedType body_ty -- C.f. WwLib.mkWorkerArgs
1116 = (poly_tyvars ++ [voidArgId], poly_tyvars ++ [realWorldPrimId])
1117 | otherwise = (poly_tyvars, poly_tyvars)
1118 spec_id_ty = mkPiTypes lam_args body_ty
1120 ; spec_f <- newSpecIdSM fn spec_id_ty
1121 ; (spec_rhs, rhs_uds) <- specExpr rhs_subst2 (mkLams lam_args body)
1123 -- The rule to put in the function's specialisation is:
1124 -- forall b, d1',d2'. f t1 b t3 d1' d2' = f1 b
1125 rule_name = mkFastString ("SPEC " ++ showSDoc (ppr fn <+> ppr spec_ty_args))
1126 spec_env_rule = mkRule True {- Auto generated -} is_local
1128 inl_act -- Note [Auto-specialisation and RULES]
1130 (poly_tyvars ++ inst_dict_ids)
1132 (mkVarApps (Var spec_f) app_args)
1134 -- Add the { d1' = dx1; d2' = dx2 } usage stuff
1135 final_uds = foldr consDictBind rhs_uds dx_binds
1137 --------------------------------------
1138 -- Add a suitable unfolding if the spec_inl_prag says so
1139 -- See Note [Inline specialisations]
1142 InlinePragma { inl_inline = Inlinable }
1143 -> inl_prag { inl_inline = EmptyInlineSpec }
1147 = case inlinePragmaSpec spec_inl_prag of
1148 Inline -> mkInlineUnfolding (Just spec_arity) spec_rhs
1149 Inlinable -> mkInlinableUnfolding spec_rhs
1152 --------------------------------------
1153 -- Adding arity information just propagates it a bit faster
1154 -- See Note [Arity decrease] in Simplify
1155 -- Copy InlinePragma information from the parent Id.
1156 -- So if f has INLINE[1] so does spec_f
1157 spec_f_w_arity = spec_f `setIdArity` max 0 (fn_arity - n_dicts)
1158 `setInlinePragma` spec_inl_prag
1159 `setIdUnfolding` spec_unf
1161 ; return (Just ((spec_f_w_arity, spec_rhs), final_uds, spec_env_rule)) } }
1163 my_zipEqual xs ys zs
1164 | debugIsOn && not (equalLength xs ys && equalLength ys zs)
1165 = pprPanic "my_zipEqual" (vcat [ ppr xs, ppr ys
1166 , ppr fn <+> ppr call_ts
1167 , ppr (idType fn), ppr theta
1168 , ppr n_dicts, ppr rhs_dict_ids
1170 | otherwise = zip3 xs ys zs
1174 -> [(DictId,DictId,CoreExpr)] -- (orig_dict, inst_dict, dx)
1175 -> (Subst, -- Substitute for all orig_dicts
1176 [CoreBind]) -- Auxiliary bindings
1177 -- Bind any dictionary arguments to fresh names, to preserve sharing
1178 -- Substitution already substitutes orig_dict -> inst_dict
1179 bindAuxiliaryDicts subst triples = go subst [] triples
1181 go subst binds [] = (subst, binds)
1182 go subst binds ((d, dx_id, dx) : pairs)
1183 | exprIsTrivial dx = go (extendIdSubst subst d dx) binds pairs
1184 -- No auxiliary binding necessary
1185 -- Note that we bind the *original* dict in the substitution,
1186 -- overriding any d->dx_id binding put there by substBndrs
1188 | otherwise = go subst_w_unf (NonRec dx_id dx : binds) pairs
1190 dx_id1 = dx_id `setIdUnfolding` mkSimpleUnfolding dx
1191 subst_w_unf = extendIdSubst subst d (Var dx_id1)
1192 -- Important! We're going to substitute dx_id1 for d
1193 -- and we want it to look "interesting", else we won't gather *any*
1194 -- consequential calls. E.g.
1196 -- If we specialise f for a call (f (dfun dNumInt)), we'll get
1197 -- a consequent call (g d') with an auxiliary definition
1199 -- We want that consequent call to look interesting
1201 -- Again, note that we bind the *original* dict in the substitution,
1202 -- overriding any d->dx_id binding put there by substBndrs
1205 Note [From non-recursive to recursive]
1206 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1207 Even in the non-recursive case, if any dict-binds depend on 'fn' we might
1208 have built a recursive knot
1211 MkUD { ud_binds = d7 = MkD ..f..
1212 , ud_calls = ...(f T d7)... }
1216 Rec { fs x = <blah>[T/a, d7/d]
1221 Here the recursion is only through the RULE.
1224 Note [Specialisation of dictionary functions]
1225 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1226 Here is a nasty example that bit us badly: see Trac #3591
1229 instance Eq [a] => C [a]
1232 dfun :: Eq [a] -> C [a]
1233 dfun a d = MkD a d (meth d)
1235 d4 :: Eq [T] = <blah>
1236 d2 :: C [T] = dfun T d4
1237 d1 :: Eq [T] = $p1 d2
1238 d3 :: C [T] = dfun T d1
1240 None of these definitions is recursive. What happened was that we
1241 generated a specialisation:
1243 RULE forall d. dfun T d = dT :: C [T]
1244 dT = (MkD a d (meth d)) [T/a, d1/d]
1245 = MkD T d1 (meth d1)
1247 But now we use the RULE on the RHS of d2, to get
1249 d2 = dT = MkD d1 (meth d1)
1252 and now d1 is bottom! The problem is that when specialising 'dfun' we
1253 should first dump "below" the binding all floated dictionary bindings
1254 that mention 'dfun' itself. So d2 and d3 (and hence d1) must be
1255 placed below 'dfun', and thus unavailable to it when specialising
1256 'dfun'. That in turn means that the call (dfun T d1) must be
1257 discarded. On the other hand, the call (dfun T d4) is fine, assuming
1258 d4 doesn't mention dfun.
1262 class C a where { foo,bar :: [a] -> [a] }
1264 instance C Int where
1268 r_bar :: C a => [a] -> [a]
1269 r_bar xs = bar (xs ++ xs)
1273 r_bar a (c::C a) (xs::[a]) = bar a d (xs ++ xs)
1275 Rec { $fCInt :: C Int = MkC foo_help reverse
1276 foo_help (xs::[Int]) = r_bar Int $fCInt xs }
1278 The call (r_bar $fCInt) mentions $fCInt,
1279 which mentions foo_help,
1280 which mentions r_bar
1281 But we DO want to specialise r_bar at Int:
1283 Rec { $fCInt :: C Int = MkC foo_help reverse
1284 foo_help (xs::[Int]) = r_bar Int $fCInt xs
1286 r_bar a (c::C a) (xs::[a]) = bar a d (xs ++ xs)
1287 RULE r_bar Int _ = r_bar_Int
1289 r_bar_Int xs = bar Int $fCInt (xs ++ xs)
1292 Note that, because of its RULE, r_bar joins the recursive
1293 group. (In this case it'll unravel a short moment later.)
1296 Conclusion: we catch the nasty case using filter_dfuns in
1297 callsForMe. To be honest I'm not 100% certain that this is 100%
1298 right, but it works. Sigh.
1301 Note [Specialising a recursive group]
1302 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1304 let rec { f x = ...g x'...
1305 ; g y = ...f y'.... }
1307 Here we specialise 'f' at Char; but that is very likely to lead to
1308 a specialisation of 'g' at Char. We must do the latter, else the
1309 whole point of specialisation is lost.
1311 But we do not want to keep iterating to a fixpoint, because in the
1312 presence of polymorphic recursion we might generate an infinite number
1315 So we use the following heuristic:
1316 * Arrange the rec block in dependency order, so far as possible
1317 (the occurrence analyser already does this)
1319 * Specialise it much like a sequence of lets
1321 * Then go through the block a second time, feeding call-info from
1322 the RHSs back in the bottom, as it were
1324 In effect, the ordering maxmimises the effectiveness of each sweep,
1325 and we do just two sweeps. This should catch almost every case of
1326 monomorphic recursion -- the exception could be a very knotted-up
1327 recursion with multiple cycles tied up together.
1329 This plan is implemented in the Rec case of specBindItself.
1331 Note [Specialisations already covered]
1332 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1333 We obviously don't want to generate two specialisations for the same
1334 argument pattern. There are two wrinkles
1336 1. We do the already-covered test in specDefn, not when we generate
1337 the CallInfo in mkCallUDs. We used to test in the latter place, but
1338 we now iterate the specialiser somewhat, and the Id at the call site
1339 might therefore not have all the RULES that we can see in specDefn
1341 2. What about two specialisations where the second is an *instance*
1342 of the first? If the more specific one shows up first, we'll generate
1343 specialisations for both. If the *less* specific one shows up first,
1344 we *don't* currently generate a specialisation for the more specific
1345 one. (See the call to lookupRule in already_covered.) Reasons:
1346 (a) lookupRule doesn't say which matches are exact (bad reason)
1347 (b) if the earlier specialisation is user-provided, it's
1348 far from clear that we should auto-specialise further
1350 Note [Auto-specialisation and RULES]
1351 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1353 g :: Num a => a -> a
1356 f :: (Int -> Int) -> Int
1358 {-# RULE f g = 0 #-}
1360 Suppose that auto-specialisation makes a specialised version of
1361 g::Int->Int That version won't appear in the LHS of the RULE for f.
1362 So if the specialisation rule fires too early, the rule for f may
1365 It might be possible to add new rules, to "complete" the rewrite system.
1367 RULE forall d. g Int d = g_spec
1371 But that's a bit complicated. For now we ask the programmer's help,
1372 by *copying the INLINE activation pragma* to the auto-specialised
1373 rule. So if g says {-# NOINLINE[2] g #-}, then the auto-spec rule
1374 will also not be active until phase 2. And that's what programmers
1375 should jolly well do anyway, even aside from specialisation, to ensure
1376 that g doesn't inline too early.
1378 This in turn means that the RULE would never fire for a NOINLINE
1379 thing so not much point in generating a specialisation at all.
1381 Note [Specialisation shape]
1382 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
1383 We only specialise a function if it has visible top-level lambdas
1384 corresponding to its overloading. E.g. if
1385 f :: forall a. Eq a => ....
1386 then its body must look like
1389 Reason: when specialising the body for a call (f ty dexp), we want to
1390 substitute dexp for d, and pick up specialised calls in the body of f.
1392 This doesn't always work. One example I came across was this:
1393 newtype Gen a = MkGen{ unGen :: Int -> a }
1395 choose :: Eq a => a -> Gen a
1396 choose n = MkGen (\r -> n)
1398 oneof = choose (1::Int)
1400 It's a silly exapmle, but we get
1401 choose = /\a. g `cast` co
1402 where choose doesn't have any dict arguments. Thus far I have not
1403 tried to fix this (wait till there's a real example).
1405 Note [Inline specialisations]
1406 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1407 Here is what we do with the InlinePragma of the original function
1408 * Activation/RuleMatchInfo: both transferred to the
1409 specialised function
1411 (a) An INLINE pragma is transferred
1412 (b) An INLINABLE pragma is *not* transferred
1414 Why (a)? Previously the idea is that the point of INLINE was
1415 precisely to specialise the function at its call site, and that's not
1416 so important for the specialised copies. But *pragma-directed*
1417 specialisation now takes place in the typechecker/desugarer, with
1418 manually specified INLINEs. The specialiation here is automatic.
1419 It'd be very odd if a function marked INLINE was specialised (because
1420 of some local use), and then forever after (including importing
1421 modules) the specialised version wasn't INLINEd. After all, the
1422 programmer said INLINE!
1424 You might wonder why we don't just not-specialise INLINE functions.
1425 It's because even INLINE functions are sometimes not inlined, when
1426 they aren't applied to interesting arguments. But perhaps the type
1427 arguments alone are enough to specialise (even though the args are too
1428 boring to trigger inlining), and it's certainly better to call the
1429 specialised version.
1431 Why (b)? See Trac #4874 for persuasive examples. Suppose we have
1433 f :: Ord a => [a] -> Int
1434 f xs = letrec f' = ...f'... in f'
1435 Then, when f is specialised and optimised we might get
1436 wgo :: [Int] -> Int#
1438 f_spec :: [Int] -> Int
1439 f_spec xs = case wgo xs of { r -> I# r }
1440 and we clearly want to inline f_spec at call sites. But if we still
1441 have the big, un-optimised of f (albeit specialised) captured in an
1442 INLINABLE pragma for f_spec, we won't get that optimisation.
1444 So we simply drop INLINABLE pragmas when specialising. It's not really
1445 a complete solution; ignoring specalisation for now, INLINABLE functions
1446 don't get properly strictness analysed, for example. But it works well
1447 for examples involving specialisation, which is the dominant use of
1448 INLINABLE. See Trac #4874.
1451 %************************************************************************
1453 \subsubsection{UsageDetails and suchlike}
1455 %************************************************************************
1460 ud_binds :: !(Bag DictBind),
1461 -- Floated dictionary bindings
1462 -- The order is important;
1463 -- in ds1 `union` ds2, bindings in ds2 can depend on those in ds1
1464 -- (Remember, Bags preserve order in GHC.)
1466 ud_calls :: !CallDetails
1468 -- INVARIANT: suppose bs = bindersOf ud_binds
1469 -- Then 'calls' may *mention* 'bs',
1470 -- but there should be no calls *for* bs
1473 instance Outputable UsageDetails where
1474 ppr (MkUD { ud_binds = dbs, ud_calls = calls })
1475 = ptext (sLit "MkUD") <+> braces (sep (punctuate comma
1476 [ptext (sLit "binds") <+> equals <+> ppr dbs,
1477 ptext (sLit "calls") <+> equals <+> ppr calls]))
1479 type DictBind = (CoreBind, VarSet)
1480 -- The set is the free vars of the binding
1481 -- both tyvars and dicts
1483 type DictExpr = CoreExpr
1485 emptyUDs :: UsageDetails
1486 emptyUDs = MkUD { ud_binds = emptyBag, ud_calls = emptyVarEnv }
1488 ------------------------------------------------------------
1489 type CallDetails = IdEnv CallInfoSet
1490 newtype CallKey = CallKey [Maybe Type] -- Nothing => unconstrained type argument
1492 -- CallInfo uses a Map, thereby ensuring that
1493 -- we record only one call instance for any key
1495 -- The list of types and dictionaries is guaranteed to
1496 -- match the type of f
1497 data CallInfoSet = CIS Id (Map CallKey ([DictExpr], VarSet))
1498 -- Range is dict args and the vars of the whole
1499 -- call (including tyvars)
1500 -- [*not* include the main id itself, of course]
1502 type CallInfo = (CallKey, ([DictExpr], VarSet))
1504 instance Outputable CallInfoSet where
1505 ppr (CIS fn map) = hang (ptext (sLit "CIS") <+> ppr fn)
1508 instance Outputable CallKey where
1509 ppr (CallKey ts) = ppr ts
1511 -- Type isn't an instance of Ord, so that we can control which
1512 -- instance we use. That's tiresome here. Oh well
1513 instance Eq CallKey where
1514 k1 == k2 = case k1 `compare` k2 of { EQ -> True; _ -> False }
1516 instance Ord CallKey where
1517 compare (CallKey k1) (CallKey k2) = cmpList cmp k1 k2
1519 cmp Nothing Nothing = EQ
1520 cmp Nothing (Just _) = LT
1521 cmp (Just _) Nothing = GT
1522 cmp (Just t1) (Just t2) = cmpType t1 t2
1524 unionCalls :: CallDetails -> CallDetails -> CallDetails
1525 unionCalls c1 c2 = plusVarEnv_C unionCallInfoSet c1 c2
1527 unionCallInfoSet :: CallInfoSet -> CallInfoSet -> CallInfoSet
1528 unionCallInfoSet (CIS f calls1) (CIS _ calls2) = CIS f (calls1 `Map.union` calls2)
1530 callDetailsFVs :: CallDetails -> VarSet
1531 callDetailsFVs calls = foldVarEnv (unionVarSet . callInfoFVs) emptyVarSet calls
1533 callInfoFVs :: CallInfoSet -> VarSet
1534 callInfoFVs (CIS _ call_info) = Map.foldRight (\(_,fv) vs -> unionVarSet fv vs) emptyVarSet call_info
1536 ------------------------------------------------------------
1537 singleCall :: Id -> [Maybe Type] -> [DictExpr] -> UsageDetails
1538 singleCall id tys dicts
1539 = MkUD {ud_binds = emptyBag,
1540 ud_calls = unitVarEnv id $ CIS id $
1541 Map.singleton (CallKey tys) (dicts, call_fvs) }
1543 call_fvs = exprsFreeVars dicts `unionVarSet` tys_fvs
1544 tys_fvs = tyVarsOfTypes (catMaybes tys)
1545 -- The type args (tys) are guaranteed to be part of the dictionary
1546 -- types, because they are just the constrained types,
1547 -- and the dictionary is therefore sure to be bound
1548 -- inside the binding for any type variables free in the type;
1549 -- hence it's safe to neglect tyvars free in tys when making
1550 -- the free-var set for this call
1551 -- BUT I don't trust this reasoning; play safe and include tys_fvs
1553 -- We don't include the 'id' itself.
1555 mkCallUDs :: Id -> [CoreExpr] -> UsageDetails
1557 | not (want_calls_for f) -- Imported from elsewhere
1558 || null theta -- Not overloaded
1559 || not (all isClassPred theta)
1560 -- Only specialise if all overloading is on class params.
1561 -- In ptic, with implicit params, the type args
1562 -- *don't* say what the value of the implicit param is!
1563 || not (spec_tys `lengthIs` n_tyvars)
1564 || not ( dicts `lengthIs` n_dicts)
1565 || not (any interestingDict dicts) -- Note [Interesting dictionary arguments]
1566 -- See also Note [Specialisations already covered]
1567 = -- pprTrace "mkCallUDs: discarding" _trace_doc
1568 emptyUDs -- Not overloaded, or no specialisation wanted
1571 = -- pprTrace "mkCallUDs: keeping" _trace_doc
1572 singleCall f spec_tys dicts
1574 _trace_doc = vcat [ppr f, ppr args, ppr n_tyvars, ppr n_dicts
1575 , ppr (map interestingDict dicts)]
1576 (tyvars, theta, _) = tcSplitSigmaTy (idType f)
1577 constrained_tyvars = tyVarsOfTheta theta
1578 n_tyvars = length tyvars
1579 n_dicts = length theta
1581 spec_tys = [mk_spec_ty tv ty | (tv, Type ty) <- tyvars `zip` args]
1582 dicts = [dict_expr | (_, dict_expr) <- theta `zip` (drop n_tyvars args)]
1585 | tyvar `elemVarSet` constrained_tyvars = Just ty
1586 | otherwise = Nothing
1588 want_calls_for f = isLocalId f || isInlinablePragma (idInlinePragma f)
1591 Note [Interesting dictionary arguments]
1592 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1594 \a.\d:Eq a. let f = ... in ...(f d)...
1595 There really is not much point in specialising f wrt the dictionary d,
1596 because the code for the specialised f is not improved at all, because
1597 d is lambda-bound. We simply get junk specialisations.
1599 What is "interesting"? Just that it has *some* structure.
1602 interestingDict :: CoreExpr -> Bool
1603 -- A dictionary argument is interesting if it has *some* structure
1604 interestingDict (Var v) = hasSomeUnfolding (idUnfolding v)
1605 || isDataConWorkId v
1606 interestingDict (Type _) = False
1607 interestingDict (Coercion _) = False
1608 interestingDict (App fn (Type _)) = interestingDict fn
1609 interestingDict (App fn (Coercion _)) = interestingDict fn
1610 interestingDict (Note _ a) = interestingDict a
1611 interestingDict (Cast e _) = interestingDict e
1612 interestingDict _ = True
1616 plusUDs :: UsageDetails -> UsageDetails -> UsageDetails
1617 plusUDs (MkUD {ud_binds = db1, ud_calls = calls1})
1618 (MkUD {ud_binds = db2, ud_calls = calls2})
1619 = MkUD { ud_binds = db1 `unionBags` db2
1620 , ud_calls = calls1 `unionCalls` calls2 }
1622 plusUDList :: [UsageDetails] -> UsageDetails
1623 plusUDList = foldr plusUDs emptyUDs
1625 -----------------------------
1626 _dictBindBndrs :: Bag DictBind -> [Id]
1627 _dictBindBndrs dbs = foldrBag ((++) . bindersOf . fst) [] dbs
1629 mkDB :: CoreBind -> DictBind
1630 mkDB bind = (bind, bind_fvs bind)
1632 bind_fvs :: CoreBind -> VarSet
1633 bind_fvs (NonRec bndr rhs) = pair_fvs (bndr,rhs)
1634 bind_fvs (Rec prs) = foldl delVarSet rhs_fvs bndrs
1637 rhs_fvs = unionVarSets (map pair_fvs prs)
1639 pair_fvs :: (Id, CoreExpr) -> VarSet
1640 pair_fvs (bndr, rhs) = exprFreeVars rhs `unionVarSet` idFreeVars bndr
1641 -- Don't forget variables mentioned in the
1642 -- rules of the bndr. C.f. OccAnal.addRuleUsage
1643 -- Also tyvars mentioned in its type; they may not appear in the RHS
1647 flattenDictBinds :: Bag DictBind -> [(Id,CoreExpr)] -> [(Id,CoreExpr)]
1648 flattenDictBinds dbs pairs
1649 = foldrBag add pairs dbs
1651 add (NonRec b r,_) pairs = (b,r) : pairs
1652 add (Rec prs1, _) pairs = prs1 ++ pairs
1654 snocDictBinds :: UsageDetails -> [CoreBind] -> UsageDetails
1655 -- Add ud_binds to the tail end of the bindings in uds
1656 snocDictBinds uds dbs
1657 = uds { ud_binds = ud_binds uds `unionBags`
1658 foldr (consBag . mkDB) emptyBag dbs }
1660 consDictBind :: CoreBind -> UsageDetails -> UsageDetails
1661 consDictBind bind uds = uds { ud_binds = mkDB bind `consBag` ud_binds uds }
1663 addDictBinds :: [DictBind] -> UsageDetails -> UsageDetails
1664 addDictBinds binds uds = uds { ud_binds = listToBag binds `unionBags` ud_binds uds }
1666 snocDictBind :: UsageDetails -> CoreBind -> UsageDetails
1667 snocDictBind uds bind = uds { ud_binds = ud_binds uds `snocBag` mkDB bind }
1669 wrapDictBinds :: Bag DictBind -> [CoreBind] -> [CoreBind]
1670 wrapDictBinds dbs binds
1671 = foldrBag add binds dbs
1673 add (bind,_) binds = bind : binds
1675 wrapDictBindsE :: Bag DictBind -> CoreExpr -> CoreExpr
1676 wrapDictBindsE dbs expr
1677 = foldrBag add expr dbs
1679 add (bind,_) expr = Let bind expr
1681 ----------------------
1682 dumpUDs :: [CoreBndr] -> UsageDetails -> (UsageDetails, Bag DictBind)
1683 -- Used at a lambda or case binder; just dump anything mentioning the binder
1684 dumpUDs bndrs uds@(MkUD { ud_binds = orig_dbs, ud_calls = orig_calls })
1685 | null bndrs = (uds, emptyBag) -- Common in case alternatives
1686 | otherwise = -- pprTrace "dumpUDs" (ppr bndrs $$ ppr free_uds $$ ppr dump_dbs) $
1687 (free_uds, dump_dbs)
1689 free_uds = MkUD { ud_binds = free_dbs, ud_calls = free_calls }
1690 bndr_set = mkVarSet bndrs
1691 (free_dbs, dump_dbs, dump_set) = splitDictBinds orig_dbs bndr_set
1692 free_calls = deleteCallsMentioning dump_set $ -- Drop calls mentioning bndr_set on the floor
1693 deleteCallsFor bndrs orig_calls -- Discard calls for bndr_set; there should be
1694 -- no calls for any of the dicts in dump_dbs
1696 dumpBindUDs :: [CoreBndr] -> UsageDetails -> (UsageDetails, Bag DictBind, Bool)
1697 -- Used at a lambda or case binder; just dump anything mentioning the binder
1698 dumpBindUDs bndrs (MkUD { ud_binds = orig_dbs, ud_calls = orig_calls })
1699 = -- pprTrace "dumpBindUDs" (ppr bndrs $$ ppr free_uds $$ ppr dump_dbs) $
1700 (free_uds, dump_dbs, float_all)
1702 free_uds = MkUD { ud_binds = free_dbs, ud_calls = free_calls }
1703 bndr_set = mkVarSet bndrs
1704 (free_dbs, dump_dbs, dump_set) = splitDictBinds orig_dbs bndr_set
1705 free_calls = deleteCallsFor bndrs orig_calls
1706 float_all = dump_set `intersectsVarSet` callDetailsFVs free_calls
1708 callsForMe :: Id -> UsageDetails -> (UsageDetails, [CallInfo])
1709 callsForMe fn (MkUD { ud_binds = orig_dbs, ud_calls = orig_calls })
1710 = -- pprTrace ("callsForMe")
1712 -- text "Orig dbs =" <+> ppr (_dictBindBndrs orig_dbs),
1713 -- text "Orig calls =" <+> ppr orig_calls,
1714 -- text "Dep set =" <+> ppr dep_set,
1715 -- text "Calls for me =" <+> ppr calls_for_me]) $
1716 (uds_without_me, calls_for_me)
1718 uds_without_me = MkUD { ud_binds = orig_dbs, ud_calls = delVarEnv orig_calls fn }
1719 calls_for_me = case lookupVarEnv orig_calls fn of
1721 Just (CIS _ calls) -> filter_dfuns (Map.toList calls)
1723 dep_set = foldlBag go (unitVarSet fn) orig_dbs
1724 go dep_set (db,fvs) | fvs `intersectsVarSet` dep_set
1725 = extendVarSetList dep_set (bindersOf db)
1726 | otherwise = dep_set
1728 -- Note [Specialisation of dictionary functions]
1729 filter_dfuns | isDFunId fn = filter ok_call
1730 | otherwise = \cs -> cs
1732 ok_call (_, (_,fvs)) = not (fvs `intersectsVarSet` dep_set)
1734 ----------------------
1735 splitDictBinds :: Bag DictBind -> IdSet -> (Bag DictBind, Bag DictBind, IdSet)
1736 -- Returns (free_dbs, dump_dbs, dump_set)
1737 splitDictBinds dbs bndr_set
1738 = foldlBag split_db (emptyBag, emptyBag, bndr_set) dbs
1739 -- Important that it's foldl not foldr;
1740 -- we're accumulating the set of dumped ids in dump_set
1742 split_db (free_dbs, dump_dbs, dump_idset) db@(bind, fvs)
1743 | dump_idset `intersectsVarSet` fvs -- Dump it
1744 = (free_dbs, dump_dbs `snocBag` db,
1745 extendVarSetList dump_idset (bindersOf bind))
1747 | otherwise -- Don't dump it
1748 = (free_dbs `snocBag` db, dump_dbs, dump_idset)
1751 ----------------------
1752 deleteCallsMentioning :: VarSet -> CallDetails -> CallDetails
1753 -- Remove calls *mentioning* bs
1754 deleteCallsMentioning bs calls
1755 = mapVarEnv filter_calls calls
1757 filter_calls :: CallInfoSet -> CallInfoSet
1758 filter_calls (CIS f calls) = CIS f (Map.filter keep_call calls)
1759 keep_call (_, fvs) = not (fvs `intersectsVarSet` bs)
1761 deleteCallsFor :: [Id] -> CallDetails -> CallDetails
1762 -- Remove calls *for* bs
1763 deleteCallsFor bs calls = delVarEnvList calls bs
1767 %************************************************************************
1769 \subsubsection{Boring helper functions}
1771 %************************************************************************
1774 type SpecM a = UniqSM a
1776 runSpecM:: SpecM a -> CoreM a
1777 runSpecM spec = do { us <- getUniqueSupplyM
1778 ; return (initUs_ us spec) }
1780 mapAndCombineSM :: (a -> SpecM (b, UsageDetails)) -> [a] -> SpecM ([b], UsageDetails)
1781 mapAndCombineSM _ [] = return ([], emptyUDs)
1782 mapAndCombineSM f (x:xs) = do (y, uds1) <- f x
1783 (ys, uds2) <- mapAndCombineSM f xs
1784 return (y:ys, uds1 `plusUDs` uds2)
1786 cloneBindSM :: Subst -> CoreBind -> SpecM (Subst, Subst, CoreBind)
1787 -- Clone the binders of the bind; return new bind with the cloned binders
1788 -- Return the substitution to use for RHSs, and the one to use for the body
1789 cloneBindSM subst (NonRec bndr rhs) = do
1790 us <- getUniqueSupplyM
1791 let (subst', bndr') = cloneIdBndr subst us bndr
1792 return (subst, subst', NonRec bndr' rhs)
1794 cloneBindSM subst (Rec pairs) = do
1795 us <- getUniqueSupplyM
1796 let (subst', bndrs') = cloneRecIdBndrs subst us (map fst pairs)
1797 return (subst', subst', Rec (bndrs' `zip` map snd pairs))
1799 newDictBndrs :: Subst -> [CoreBndr] -> SpecM (Subst, [CoreBndr])
1800 -- Make up completely fresh binders for the dictionaries
1801 -- Their bindings are going to float outwards
1802 newDictBndrs subst bndrs
1803 = do { bndrs' <- mapM new bndrs
1804 ; let subst' = extendIdSubstList subst
1805 [(d, Var d') | (d,d') <- bndrs `zip` bndrs']
1806 ; return (subst', bndrs' ) }
1808 new b = do { uniq <- getUniqueM
1810 ty' = CoreSubst.substTy subst (idType b)
1811 ; return (mkUserLocal (nameOccName n) uniq ty' (getSrcSpan n)) }
1813 newSpecIdSM :: Id -> Type -> SpecM Id
1814 -- Give the new Id a similar occurrence name to the old one
1815 newSpecIdSM old_id new_ty
1816 = do { uniq <- getUniqueM
1817 ; let name = idName old_id
1818 new_occ = mkSpecOcc (nameOccName name)
1819 new_id = mkUserLocal new_occ uniq new_ty (getSrcSpan name)
1824 Old (but interesting) stuff about unboxed bindings
1825 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1827 What should we do when a value is specialised to a *strict* unboxed value?
1829 map_*_* f (x:xs) = let h = f x
1833 Could convert let to case:
1835 map_*_Int# f (x:xs) = case f x of h# ->
1839 This may be undesirable since it forces evaluation here, but the value
1840 may not be used in all branches of the body. In the general case this
1841 transformation is impossible since the mutual recursion in a letrec
1842 cannot be expressed as a case.
1844 There is also a problem with top-level unboxed values, since our
1845 implementation cannot handle unboxed values at the top level.
1847 Solution: Lift the binding of the unboxed value and extract it when it
1850 map_*_Int# f (x:xs) = let h = case (f x) of h# -> _Lift h#
1855 Now give it to the simplifier and the _Lifting will be optimised away.
1857 The benfit is that we have given the specialised "unboxed" values a
1858 very simplep lifted semantics and then leave it up to the simplifier to
1859 optimise it --- knowing that the overheads will be removed in nearly
1862 In particular, the value will only be evaluted in the branches of the
1863 program which use it, rather than being forced at the point where the
1864 value is bound. For example:
1866 filtermap_*_* p f (x:xs)
1873 filtermap_*_Int# p f (x:xs)
1874 = let h = case (f x) of h# -> _Lift h#
1877 True -> case h of _Lift h#
1881 The binding for h can still be inlined in the one branch and the
1882 _Lifting eliminated.
1885 Question: When won't the _Lifting be eliminated?
1887 Answer: When they at the top-level (where it is necessary) or when
1888 inlining would duplicate work (or possibly code depending on
1889 options). However, the _Lifting will still be eliminated if the
1890 strictness analyser deems the lifted binding strict.