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 (Var v) = return (specVar subst v, emptyUDs)
713 specExpr _ (Lit lit) = return (Lit lit, emptyUDs)
714 specExpr subst (Cast e co) = do
715 (e', uds) <- specExpr subst e
716 return ((Cast e' (CoreSubst.substTy subst co)), uds)
717 specExpr subst (Note note body) = do
718 (body', uds) <- specExpr subst body
719 return (Note (specNote subst note) body', uds)
722 ---------------- Applications might generate a call instance --------------------
723 specExpr subst expr@(App {})
726 go (App fun arg) args = do (arg', uds_arg) <- specExpr subst arg
727 (fun', uds_app) <- go fun (arg':args)
728 return (App fun' arg', uds_arg `plusUDs` uds_app)
730 go (Var f) args = case specVar subst f of
731 Var f' -> return (Var f', mkCallUDs f' args)
732 e' -> return (e', emptyUDs) -- I don't expect this!
733 go other _ = specExpr subst other
735 ---------------- Lambda/case require dumping of usage details --------------------
736 specExpr subst e@(Lam _ _) = do
737 (body', uds) <- specExpr subst' body
738 let (free_uds, dumped_dbs) = dumpUDs bndrs' uds
739 return (mkLams bndrs' (wrapDictBindsE dumped_dbs body'), free_uds)
741 (bndrs, body) = collectBinders e
742 (subst', bndrs') = substBndrs subst bndrs
743 -- More efficient to collect a group of binders together all at once
744 -- and we don't want to split a lambda group with dumped bindings
746 specExpr subst (Case scrut case_bndr ty alts)
747 = do { (scrut', scrut_uds) <- specExpr subst scrut
748 ; (scrut'', case_bndr', alts', alts_uds)
749 <- specCase subst scrut' case_bndr alts
750 ; return (Case scrut'' case_bndr' (CoreSubst.substTy subst ty) alts'
751 , scrut_uds `plusUDs` alts_uds) }
753 ---------------- Finally, let is the interesting case --------------------
754 specExpr subst (Let bind body) = do
756 (rhs_subst, body_subst, bind') <- cloneBindSM subst bind
758 -- Deal with the body
759 (body', body_uds) <- specExpr body_subst body
761 -- Deal with the bindings
762 (binds', uds) <- specBind rhs_subst bind' body_uds
765 return (foldr Let body' binds', uds)
767 -- Must apply the type substitution to coerceions
768 specNote :: Subst -> Note -> Note
769 specNote _ note = note
773 -> CoreExpr -- Scrutinee, already done
775 -> SpecM ( CoreExpr -- New scrutinee
779 specCase subst scrut' case_bndr [(con, args, rhs)]
780 | isDictId case_bndr -- See Note [Floating dictionaries out of cases]
781 , interestingDict scrut'
782 , not (isDeadBinder case_bndr && null sc_args')
783 = do { (case_bndr_flt : sc_args_flt) <- mapM clone_me (case_bndr' : sc_args')
785 ; let sc_rhss = [ Case (Var case_bndr_flt) case_bndr' (idType sc_arg')
786 [(con, args', Var sc_arg')]
787 | sc_arg' <- sc_args' ]
789 -- Extend the substitution for RHS to map the *original* binders
790 -- to their floated verions. Attach an unfolding to these floated
791 -- binders so they look interesting to interestingDict
792 mb_sc_flts :: [Maybe DictId]
793 mb_sc_flts = map (lookupVarEnv clone_env) args'
794 clone_env = zipVarEnv sc_args' (zipWith add_unf sc_args_flt sc_rhss)
795 subst_prs = (case_bndr, Var (add_unf case_bndr_flt scrut'))
796 : [ (arg, Var sc_flt)
797 | (arg, Just sc_flt) <- args `zip` mb_sc_flts ]
798 subst_rhs' = extendIdSubstList subst_rhs subst_prs
800 ; (rhs', rhs_uds) <- specExpr subst_rhs' rhs
801 ; let scrut_bind = mkDB (NonRec case_bndr_flt scrut')
802 case_bndr_set = unitVarSet case_bndr_flt
803 sc_binds = [(NonRec sc_arg_flt sc_rhs, case_bndr_set)
804 | (sc_arg_flt, sc_rhs) <- sc_args_flt `zip` sc_rhss ]
805 flt_binds = scrut_bind : sc_binds
806 (free_uds, dumped_dbs) = dumpUDs (case_bndr':args') rhs_uds
807 all_uds = flt_binds `addDictBinds` free_uds
808 alt' = (con, args', wrapDictBindsE dumped_dbs rhs')
809 ; return (Var case_bndr_flt, case_bndr', [alt'], all_uds) }
811 (subst_rhs, (case_bndr':args')) = substBndrs subst (case_bndr:args)
812 sc_args' = filter is_flt_sc_arg args'
814 clone_me bndr = do { uniq <- getUniqueM
815 ; return (mkUserLocal occ uniq ty loc) }
819 occ = nameOccName name
820 loc = getSrcSpan name
822 add_unf sc_flt sc_rhs -- Sole purpose: make sc_flt respond True to interestingDictId
823 = setIdUnfolding sc_flt (mkSimpleUnfolding sc_rhs)
825 arg_set = mkVarSet args'
826 is_flt_sc_arg var = isId var
827 && not (isDeadBinder var)
829 && not (tyVarsOfType var_ty `intersectsVarSet` arg_set)
834 specCase subst scrut case_bndr alts
835 = do { (alts', uds_alts) <- mapAndCombineSM spec_alt alts
836 ; return (scrut, case_bndr', alts', uds_alts) }
838 (subst_alt, case_bndr') = substBndr subst case_bndr
839 spec_alt (con, args, rhs) = do
840 (rhs', uds) <- specExpr subst_rhs rhs
841 let (free_uds, dumped_dbs) = dumpUDs (case_bndr' : args') uds
842 return ((con, args', wrapDictBindsE dumped_dbs rhs'), free_uds)
844 (subst_rhs, args') = substBndrs subst_alt args
847 Note [Floating dictionaries out of cases]
848 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
850 g = \d. case d of { MkD sc ... -> ...(f sc)... }
851 Naively we can't float d2's binding out of the case expression,
852 because 'sc' is bound by the case, and that in turn means we can't
853 specialise f, which seems a pity.
855 So we invert the case, by floating out a binding
857 sc_flt = case d of { MkD sc ... -> sc }
858 Now we can float the call instance for 'f'. Indeed this is just
859 what'll happen if 'sc' was originally bound with a let binding,
860 but case is more efficient, and necessary with equalities. So it's
861 good to work with both.
863 You might think that this won't make any difference, because the
864 call instance will only get nuked by the \d. BUT if 'g' itself is
865 specialised, then transitively we should be able to specialise f.
868 case e of cb { MkD sc ... -> ...(f sc)... }
871 sc_flt = case cb_flt of { MkD sc ... -> sc }
873 case cb_flt of bg { MkD sc ... -> ....(f sc_flt)... }
875 The "_flt" things are the floated binds; we use the current substitution
876 to substitute sc -> sc_flt in the RHS
878 %************************************************************************
880 Dealing with a binding
882 %************************************************************************
885 specBind :: Subst -- Use this for RHSs
887 -> UsageDetails -- Info on how the scope of the binding
888 -> SpecM ([CoreBind], -- New bindings
889 UsageDetails) -- And info to pass upstream
891 -- Returned UsageDetails:
892 -- No calls for binders of this bind
893 specBind rhs_subst (NonRec fn rhs) body_uds
894 = do { (rhs', rhs_uds) <- specExpr rhs_subst rhs
895 ; (fn', spec_defns, body_uds1) <- specDefn rhs_subst body_uds fn rhs
897 ; let pairs = spec_defns ++ [(fn', rhs')]
898 -- fn' mentions the spec_defns in its rules,
899 -- so put the latter first
901 combined_uds = body_uds1 `plusUDs` rhs_uds
902 -- This way round a call in rhs_uds of a function f
903 -- at type T will override a call of f at T in body_uds1; and
904 -- that is good because it'll tend to keep "earlier" calls
905 -- See Note [Specialisation of dictionary functions]
907 (free_uds, dump_dbs, float_all) = dumpBindUDs [fn] combined_uds
908 -- See Note [From non-recursive to recursive]
910 final_binds | isEmptyBag dump_dbs = [NonRec b r | (b,r) <- pairs]
911 | otherwise = [Rec (flattenDictBinds dump_dbs pairs)]
914 -- Rather than discard the calls mentioning the bound variables
915 -- we float this binding along with the others
916 return ([], free_uds `snocDictBinds` final_binds)
918 -- No call in final_uds mentions bound variables,
919 -- so we can just leave the binding here
920 return (final_binds, free_uds) }
923 specBind rhs_subst (Rec pairs) body_uds
924 -- Note [Specialising a recursive group]
925 = do { let (bndrs,rhss) = unzip pairs
926 ; (rhss', rhs_uds) <- mapAndCombineSM (specExpr rhs_subst) rhss
927 ; let scope_uds = body_uds `plusUDs` rhs_uds
928 -- Includes binds and calls arising from rhss
930 ; (bndrs1, spec_defns1, uds1) <- specDefns rhs_subst scope_uds pairs
932 ; (bndrs3, spec_defns3, uds3)
933 <- if null spec_defns1 -- Common case: no specialisation
934 then return (bndrs1, [], uds1)
935 else do { -- Specialisation occurred; do it again
936 (bndrs2, spec_defns2, uds2)
937 <- specDefns rhs_subst uds1 (bndrs1 `zip` rhss)
938 ; return (bndrs2, spec_defns2 ++ spec_defns1, uds2) }
940 ; let (final_uds, dumped_dbs, float_all) = dumpBindUDs bndrs uds3
941 bind = Rec (flattenDictBinds dumped_dbs $
942 spec_defns3 ++ zip bndrs3 rhss')
945 return ([], final_uds `snocDictBind` bind)
947 return ([bind], final_uds) }
950 ---------------------------
952 -> UsageDetails -- Info on how it is used in its scope
953 -> [(Id,CoreExpr)] -- The things being bound and their un-processed RHS
954 -> SpecM ([Id], -- Original Ids with RULES added
955 [(Id,CoreExpr)], -- Extra, specialised bindings
956 UsageDetails) -- Stuff to fling upwards from the specialised versions
958 -- Specialise a list of bindings (the contents of a Rec), but flowing usages
959 -- upwards binding by binding. Example: { f = ...g ...; g = ...f .... }
960 -- Then if the input CallDetails has a specialised call for 'g', whose specialisation
961 -- in turn generates a specialised call for 'f', we catch that in this one sweep.
962 -- But not vice versa (it's a fixpoint problem).
964 specDefns _subst uds []
965 = return ([], [], uds)
966 specDefns subst uds ((bndr,rhs):pairs)
967 = do { (bndrs1, spec_defns1, uds1) <- specDefns subst uds pairs
968 ; (bndr1, spec_defns2, uds2) <- specDefn subst uds1 bndr rhs
969 ; return (bndr1 : bndrs1, spec_defns1 ++ spec_defns2, uds2) }
971 ---------------------------
973 -> UsageDetails -- Info on how it is used in its scope
974 -> Id -> CoreExpr -- The thing being bound and its un-processed RHS
975 -> SpecM (Id, -- Original Id with added RULES
976 [(Id,CoreExpr)], -- Extra, specialised bindings
977 UsageDetails) -- Stuff to fling upwards from the specialised versions
979 specDefn subst body_uds fn rhs
980 = do { let (body_uds_without_me, calls_for_me) = callsForMe fn body_uds
981 rules_for_me = idCoreRules fn
982 ; (rules, spec_defns, spec_uds) <- specCalls subst rules_for_me
984 ; return ( fn `addIdSpecialisations` rules
986 , body_uds_without_me `plusUDs` spec_uds) }
987 -- It's important that the `plusUDs` is this way
988 -- round, because body_uds_without_me may bind
989 -- dictionaries that are used in calls_for_me passed
990 -- to specDefn. So the dictionary bindings in
991 -- spec_uds may mention dictionaries bound in
992 -- body_uds_without_me
994 ---------------------------
996 -> [CoreRule] -- Existing RULES for the fn
999 -> SpecM ([CoreRule], -- New RULES for the fn
1000 [(Id,CoreExpr)], -- Extra, specialised bindings
1001 UsageDetails) -- New usage details from the specialised RHSs
1003 -- This function checks existing rules, and does not create
1004 -- duplicate ones. So the caller does not need to do this filtering.
1005 -- See 'already_covered'
1007 specCalls subst rules_for_me calls_for_me fn rhs
1008 -- The first case is the interesting one
1009 | rhs_tyvars `lengthIs` n_tyvars -- Rhs of fn's defn has right number of big lambdas
1010 && rhs_ids `lengthAtLeast` n_dicts -- and enough dict args
1011 && notNull calls_for_me -- And there are some calls to specialise
1012 && not (isNeverActive (idInlineActivation fn))
1013 -- Don't specialise NOINLINE things
1014 -- See Note [Auto-specialisation and RULES]
1016 -- && not (certainlyWillInline (idUnfolding fn)) -- And it's not small
1017 -- See Note [Inline specialisation] for why we do not
1018 -- switch off specialisation for inline functions
1020 = -- pprTrace "specDefn: some" (ppr fn $$ ppr calls_for_me $$ ppr rules_for_me) $
1021 do { stuff <- mapM spec_call calls_for_me
1022 ; let (spec_defns, spec_uds, spec_rules) = unzip3 (catMaybes stuff)
1023 ; return (spec_rules, spec_defns, plusUDList spec_uds) }
1025 | otherwise -- No calls or RHS doesn't fit our preconceptions
1026 = WARN( notNull calls_for_me, ptext (sLit "Missed specialisation opportunity for")
1027 <+> ppr fn $$ _trace_doc )
1028 -- Note [Specialisation shape]
1029 -- pprTrace "specDefn: none" (ppr fn $$ ppr calls_for_me) $
1030 return ([], [], emptyUDs)
1032 _trace_doc = vcat [ ppr rhs_tyvars, ppr n_tyvars
1033 , ppr rhs_ids, ppr n_dicts
1034 , ppr (idInlineActivation fn) ]
1037 fn_arity = idArity fn
1038 fn_unf = realIdUnfolding fn -- Ignore loop-breaker-ness here
1039 (tyvars, theta, _) = tcSplitSigmaTy fn_type
1040 n_tyvars = length tyvars
1041 n_dicts = length theta
1042 inl_prag = idInlinePragma fn
1043 inl_act = inlinePragmaActivation inl_prag
1044 is_local = isLocalId fn
1046 -- Figure out whether the function has an INLINE pragma
1047 -- See Note [Inline specialisations]
1049 spec_arity = unfoldingArity fn_unf - n_dicts -- Arity of the *specialised* inline rule
1051 (rhs_tyvars, rhs_ids, rhs_body) = collectTyAndValBinders rhs
1053 rhs_dict_ids = take n_dicts rhs_ids
1054 body = mkLams (drop n_dicts rhs_ids) rhs_body
1055 -- Glue back on the non-dict lambdas
1057 already_covered :: [CoreExpr] -> Bool
1058 already_covered args -- Note [Specialisations already covered]
1059 = isJust (lookupRule (const True) realIdUnfolding
1060 (substInScope subst)
1061 fn args rules_for_me)
1063 mk_ty_args :: [Maybe Type] -> [CoreExpr]
1064 mk_ty_args call_ts = zipWithEqual "spec_call" mk_ty_arg rhs_tyvars call_ts
1066 mk_ty_arg rhs_tyvar Nothing = Type (mkTyVarTy rhs_tyvar)
1067 mk_ty_arg _ (Just ty) = Type ty
1069 ----------------------------------------------------------
1070 -- Specialise to one particular call pattern
1071 spec_call :: CallInfo -- Call instance
1072 -> SpecM (Maybe ((Id,CoreExpr), -- Specialised definition
1073 UsageDetails, -- Usage details from specialised body
1074 CoreRule)) -- Info for the Id's SpecEnv
1075 spec_call (CallKey call_ts, (call_ds, _))
1076 = ASSERT( call_ts `lengthIs` n_tyvars && call_ds `lengthIs` n_dicts )
1078 -- Suppose f's defn is f = /\ a b c -> \ d1 d2 -> rhs
1079 -- Supppose the call is for f [Just t1, Nothing, Just t3] [dx1, dx2]
1081 -- Construct the new binding
1082 -- f1 = SUBST[a->t1,c->t3, d1->d1', d2->d2'] (/\ b -> rhs)
1083 -- PLUS the usage-details
1084 -- { d1' = dx1; d2' = dx2 }
1085 -- where d1', d2' are cloned versions of d1,d2, with the type substitution
1086 -- applied. These auxiliary bindings just avoid duplication of dx1, dx2
1088 -- Note that the substitution is applied to the whole thing.
1089 -- This is convenient, but just slightly fragile. Notably:
1090 -- * There had better be no name clashes in a/b/c
1092 -- poly_tyvars = [b] in the example above
1093 -- spec_tyvars = [a,c]
1094 -- ty_args = [t1,b,t3]
1095 poly_tyvars = [tv | (tv, Nothing) <- rhs_tyvars `zip` call_ts]
1096 spec_tv_binds = [(tv,ty) | (tv, Just ty) <- rhs_tyvars `zip` call_ts]
1097 spec_ty_args = map snd spec_tv_binds
1098 ty_args = mk_ty_args call_ts
1099 rhs_subst = CoreSubst.extendTvSubstList subst spec_tv_binds
1101 ; (rhs_subst1, inst_dict_ids) <- newDictBndrs rhs_subst rhs_dict_ids
1102 -- Clone rhs_dicts, including instantiating their types
1104 ; let (rhs_subst2, dx_binds) = bindAuxiliaryDicts rhs_subst1 $
1105 (my_zipEqual rhs_dict_ids inst_dict_ids call_ds)
1106 inst_args = ty_args ++ map Var inst_dict_ids
1108 ; if already_covered inst_args then
1111 { -- Figure out the type of the specialised function
1112 let body_ty = applyTypeToArgs rhs fn_type inst_args
1113 (lam_args, app_args) -- Add a dummy argument if body_ty is unlifted
1114 | isUnLiftedType body_ty -- C.f. WwLib.mkWorkerArgs
1115 = (poly_tyvars ++ [voidArgId], poly_tyvars ++ [realWorldPrimId])
1116 | otherwise = (poly_tyvars, poly_tyvars)
1117 spec_id_ty = mkPiTypes lam_args body_ty
1119 ; spec_f <- newSpecIdSM fn spec_id_ty
1120 ; (spec_rhs, rhs_uds) <- specExpr rhs_subst2 (mkLams lam_args body)
1122 -- The rule to put in the function's specialisation is:
1123 -- forall b, d1',d2'. f t1 b t3 d1' d2' = f1 b
1124 rule_name = mkFastString ("SPEC " ++ showSDoc (ppr fn <+> ppr spec_ty_args))
1125 spec_env_rule = mkRule True {- Auto generated -} is_local
1127 inl_act -- Note [Auto-specialisation and RULES]
1129 (poly_tyvars ++ inst_dict_ids)
1131 (mkVarApps (Var spec_f) app_args)
1133 -- Add the { d1' = dx1; d2' = dx2 } usage stuff
1134 final_uds = foldr consDictBind rhs_uds dx_binds
1136 --------------------------------------
1137 -- Add a suitable unfolding if the spec_inl_prag says so
1138 -- See Note [Inline specialisations]
1141 InlinePragma { inl_inline = Inlinable }
1142 -> inl_prag { inl_inline = EmptyInlineSpec }
1146 = case inlinePragmaSpec spec_inl_prag of
1147 Inline -> mkInlineUnfolding (Just spec_arity) spec_rhs
1148 Inlinable -> mkInlinableUnfolding spec_rhs
1151 --------------------------------------
1152 -- Adding arity information just propagates it a bit faster
1153 -- See Note [Arity decrease] in Simplify
1154 -- Copy InlinePragma information from the parent Id.
1155 -- So if f has INLINE[1] so does spec_f
1156 spec_f_w_arity = spec_f `setIdArity` max 0 (fn_arity - n_dicts)
1157 `setInlinePragma` spec_inl_prag
1158 `setIdUnfolding` spec_unf
1160 ; return (Just ((spec_f_w_arity, spec_rhs), final_uds, spec_env_rule)) } }
1162 my_zipEqual xs ys zs
1163 | debugIsOn && not (equalLength xs ys && equalLength ys zs)
1164 = pprPanic "my_zipEqual" (vcat [ ppr xs, ppr ys
1165 , ppr fn <+> ppr call_ts
1166 , ppr (idType fn), ppr theta
1167 , ppr n_dicts, ppr rhs_dict_ids
1169 | otherwise = zip3 xs ys zs
1173 -> [(DictId,DictId,CoreExpr)] -- (orig_dict, inst_dict, dx)
1174 -> (Subst, -- Substitute for all orig_dicts
1175 [CoreBind]) -- Auxiliary bindings
1176 -- Bind any dictionary arguments to fresh names, to preserve sharing
1177 -- Substitution already substitutes orig_dict -> inst_dict
1178 bindAuxiliaryDicts subst triples = go subst [] triples
1180 go subst binds [] = (subst, binds)
1181 go subst binds ((d, dx_id, dx) : pairs)
1182 | exprIsTrivial dx = go (extendIdSubst subst d dx) binds pairs
1183 -- No auxiliary binding necessary
1184 -- Note that we bind the *original* dict in the substitution,
1185 -- overriding any d->dx_id binding put there by substBndrs
1187 | otherwise = go subst_w_unf (NonRec dx_id dx : binds) pairs
1189 dx_id1 = dx_id `setIdUnfolding` mkSimpleUnfolding dx
1190 subst_w_unf = extendIdSubst subst d (Var dx_id1)
1191 -- Important! We're going to substitute dx_id1 for d
1192 -- and we want it to look "interesting", else we won't gather *any*
1193 -- consequential calls. E.g.
1195 -- If we specialise f for a call (f (dfun dNumInt)), we'll get
1196 -- a consequent call (g d') with an auxiliary definition
1198 -- We want that consequent call to look interesting
1200 -- Again, note that we bind the *original* dict in the substitution,
1201 -- overriding any d->dx_id binding put there by substBndrs
1204 Note [From non-recursive to recursive]
1205 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1206 Even in the non-recursive case, if any dict-binds depend on 'fn' we might
1207 have built a recursive knot
1210 MkUD { ud_binds = d7 = MkD ..f..
1211 , ud_calls = ...(f T d7)... }
1215 Rec { fs x = <blah>[T/a, d7/d]
1220 Here the recursion is only through the RULE.
1223 Note [Specialisation of dictionary functions]
1224 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1225 Here is a nasty example that bit us badly: see Trac #3591
1228 instance Eq [a] => C [a]
1231 dfun :: Eq [a] -> C [a]
1232 dfun a d = MkD a d (meth d)
1234 d4 :: Eq [T] = <blah>
1235 d2 :: C [T] = dfun T d4
1236 d1 :: Eq [T] = $p1 d2
1237 d3 :: C [T] = dfun T d1
1239 None of these definitions is recursive. What happened was that we
1240 generated a specialisation:
1242 RULE forall d. dfun T d = dT :: C [T]
1243 dT = (MkD a d (meth d)) [T/a, d1/d]
1244 = MkD T d1 (meth d1)
1246 But now we use the RULE on the RHS of d2, to get
1248 d2 = dT = MkD d1 (meth d1)
1251 and now d1 is bottom! The problem is that when specialising 'dfun' we
1252 should first dump "below" the binding all floated dictionary bindings
1253 that mention 'dfun' itself. So d2 and d3 (and hence d1) must be
1254 placed below 'dfun', and thus unavailable to it when specialising
1255 'dfun'. That in turn means that the call (dfun T d1) must be
1256 discarded. On the other hand, the call (dfun T d4) is fine, assuming
1257 d4 doesn't mention dfun.
1261 class C a where { foo,bar :: [a] -> [a] }
1263 instance C Int where
1267 r_bar :: C a => [a] -> [a]
1268 r_bar xs = bar (xs ++ xs)
1272 r_bar a (c::C a) (xs::[a]) = bar a d (xs ++ xs)
1274 Rec { $fCInt :: C Int = MkC foo_help reverse
1275 foo_help (xs::[Int]) = r_bar Int $fCInt xs }
1277 The call (r_bar $fCInt) mentions $fCInt,
1278 which mentions foo_help,
1279 which mentions r_bar
1280 But we DO want to specialise r_bar at Int:
1282 Rec { $fCInt :: C Int = MkC foo_help reverse
1283 foo_help (xs::[Int]) = r_bar Int $fCInt xs
1285 r_bar a (c::C a) (xs::[a]) = bar a d (xs ++ xs)
1286 RULE r_bar Int _ = r_bar_Int
1288 r_bar_Int xs = bar Int $fCInt (xs ++ xs)
1291 Note that, because of its RULE, r_bar joins the recursive
1292 group. (In this case it'll unravel a short moment later.)
1295 Conclusion: we catch the nasty case using filter_dfuns in
1296 callsForMe. To be honest I'm not 100% certain that this is 100%
1297 right, but it works. Sigh.
1300 Note [Specialising a recursive group]
1301 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1303 let rec { f x = ...g x'...
1304 ; g y = ...f y'.... }
1306 Here we specialise 'f' at Char; but that is very likely to lead to
1307 a specialisation of 'g' at Char. We must do the latter, else the
1308 whole point of specialisation is lost.
1310 But we do not want to keep iterating to a fixpoint, because in the
1311 presence of polymorphic recursion we might generate an infinite number
1314 So we use the following heuristic:
1315 * Arrange the rec block in dependency order, so far as possible
1316 (the occurrence analyser already does this)
1318 * Specialise it much like a sequence of lets
1320 * Then go through the block a second time, feeding call-info from
1321 the RHSs back in the bottom, as it were
1323 In effect, the ordering maxmimises the effectiveness of each sweep,
1324 and we do just two sweeps. This should catch almost every case of
1325 monomorphic recursion -- the exception could be a very knotted-up
1326 recursion with multiple cycles tied up together.
1328 This plan is implemented in the Rec case of specBindItself.
1330 Note [Specialisations already covered]
1331 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1332 We obviously don't want to generate two specialisations for the same
1333 argument pattern. There are two wrinkles
1335 1. We do the already-covered test in specDefn, not when we generate
1336 the CallInfo in mkCallUDs. We used to test in the latter place, but
1337 we now iterate the specialiser somewhat, and the Id at the call site
1338 might therefore not have all the RULES that we can see in specDefn
1340 2. What about two specialisations where the second is an *instance*
1341 of the first? If the more specific one shows up first, we'll generate
1342 specialisations for both. If the *less* specific one shows up first,
1343 we *don't* currently generate a specialisation for the more specific
1344 one. (See the call to lookupRule in already_covered.) Reasons:
1345 (a) lookupRule doesn't say which matches are exact (bad reason)
1346 (b) if the earlier specialisation is user-provided, it's
1347 far from clear that we should auto-specialise further
1349 Note [Auto-specialisation and RULES]
1350 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1352 g :: Num a => a -> a
1355 f :: (Int -> Int) -> Int
1357 {-# RULE f g = 0 #-}
1359 Suppose that auto-specialisation makes a specialised version of
1360 g::Int->Int That version won't appear in the LHS of the RULE for f.
1361 So if the specialisation rule fires too early, the rule for f may
1364 It might be possible to add new rules, to "complete" the rewrite system.
1366 RULE forall d. g Int d = g_spec
1370 But that's a bit complicated. For now we ask the programmer's help,
1371 by *copying the INLINE activation pragma* to the auto-specialised
1372 rule. So if g says {-# NOINLINE[2] g #-}, then the auto-spec rule
1373 will also not be active until phase 2. And that's what programmers
1374 should jolly well do anyway, even aside from specialisation, to ensure
1375 that g doesn't inline too early.
1377 This in turn means that the RULE would never fire for a NOINLINE
1378 thing so not much point in generating a specialisation at all.
1380 Note [Specialisation shape]
1381 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
1382 We only specialise a function if it has visible top-level lambdas
1383 corresponding to its overloading. E.g. if
1384 f :: forall a. Eq a => ....
1385 then its body must look like
1388 Reason: when specialising the body for a call (f ty dexp), we want to
1389 substitute dexp for d, and pick up specialised calls in the body of f.
1391 This doesn't always work. One example I came across was this:
1392 newtype Gen a = MkGen{ unGen :: Int -> a }
1394 choose :: Eq a => a -> Gen a
1395 choose n = MkGen (\r -> n)
1397 oneof = choose (1::Int)
1399 It's a silly exapmle, but we get
1400 choose = /\a. g `cast` co
1401 where choose doesn't have any dict arguments. Thus far I have not
1402 tried to fix this (wait till there's a real example).
1404 Note [Inline specialisations]
1405 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1406 Here is what we do with the InlinePragma of the original function
1407 * Activation/RuleMatchInfo: both transferred to the
1408 specialised function
1410 (a) An INLINE pragma is transferred
1411 (b) An INLINABLE pragma is *not* transferred
1413 Why (a)? Previously the idea is that the point of INLINE was
1414 precisely to specialise the function at its call site, and that's not
1415 so important for the specialised copies. But *pragma-directed*
1416 specialisation now takes place in the typechecker/desugarer, with
1417 manually specified INLINEs. The specialiation here is automatic.
1418 It'd be very odd if a function marked INLINE was specialised (because
1419 of some local use), and then forever after (including importing
1420 modules) the specialised version wasn't INLINEd. After all, the
1421 programmer said INLINE!
1423 You might wonder why we don't just not-specialise INLINE functions.
1424 It's because even INLINE functions are sometimes not inlined, when
1425 they aren't applied to interesting arguments. But perhaps the type
1426 arguments alone are enough to specialise (even though the args are too
1427 boring to trigger inlining), and it's certainly better to call the
1428 specialised version.
1430 Why (b)? See Trac #4874 for persuasive examples. Suppose we have
1432 f :: Ord a => [a] -> Int
1433 f xs = letrec f' = ...f'... in f'
1434 Then, when f is specialised and optimised we might get
1435 wgo :: [Int] -> Int#
1437 f_spec :: [Int] -> Int
1438 f_spec xs = case wgo xs of { r -> I# r }
1439 and we clearly want to inline f_spec at call sites. But if we still
1440 have the big, un-optimised of f (albeit specialised) captured in an
1441 INLINABLE pragma for f_spec, we won't get that optimisation.
1443 So we simply drop INLINABLE pragmas when specialising. It's not really
1444 a complete solution; ignoring specalisation for now, INLINABLE functions
1445 don't get properly strictness analysed, for example. But it works well
1446 for examples involving specialisation, which is the dominant use of
1447 INLINABLE. See Trac #4874.
1450 %************************************************************************
1452 \subsubsection{UsageDetails and suchlike}
1454 %************************************************************************
1459 ud_binds :: !(Bag DictBind),
1460 -- Floated dictionary bindings
1461 -- The order is important;
1462 -- in ds1 `union` ds2, bindings in ds2 can depend on those in ds1
1463 -- (Remember, Bags preserve order in GHC.)
1465 ud_calls :: !CallDetails
1467 -- INVARIANT: suppose bs = bindersOf ud_binds
1468 -- Then 'calls' may *mention* 'bs',
1469 -- but there should be no calls *for* bs
1472 instance Outputable UsageDetails where
1473 ppr (MkUD { ud_binds = dbs, ud_calls = calls })
1474 = ptext (sLit "MkUD") <+> braces (sep (punctuate comma
1475 [ptext (sLit "binds") <+> equals <+> ppr dbs,
1476 ptext (sLit "calls") <+> equals <+> ppr calls]))
1478 type DictBind = (CoreBind, VarSet)
1479 -- The set is the free vars of the binding
1480 -- both tyvars and dicts
1482 type DictExpr = CoreExpr
1484 emptyUDs :: UsageDetails
1485 emptyUDs = MkUD { ud_binds = emptyBag, ud_calls = emptyVarEnv }
1487 ------------------------------------------------------------
1488 type CallDetails = IdEnv CallInfoSet
1489 newtype CallKey = CallKey [Maybe Type] -- Nothing => unconstrained type argument
1491 -- CallInfo uses a Map, thereby ensuring that
1492 -- we record only one call instance for any key
1494 -- The list of types and dictionaries is guaranteed to
1495 -- match the type of f
1496 data CallInfoSet = CIS Id (Map CallKey ([DictExpr], VarSet))
1497 -- Range is dict args and the vars of the whole
1498 -- call (including tyvars)
1499 -- [*not* include the main id itself, of course]
1501 type CallInfo = (CallKey, ([DictExpr], VarSet))
1503 instance Outputable CallInfoSet where
1504 ppr (CIS fn map) = hang (ptext (sLit "CIS") <+> ppr fn)
1507 instance Outputable CallKey where
1508 ppr (CallKey ts) = ppr ts
1510 -- Type isn't an instance of Ord, so that we can control which
1511 -- instance we use. That's tiresome here. Oh well
1512 instance Eq CallKey where
1513 k1 == k2 = case k1 `compare` k2 of { EQ -> True; _ -> False }
1515 instance Ord CallKey where
1516 compare (CallKey k1) (CallKey k2) = cmpList cmp k1 k2
1518 cmp Nothing Nothing = EQ
1519 cmp Nothing (Just _) = LT
1520 cmp (Just _) Nothing = GT
1521 cmp (Just t1) (Just t2) = tcCmpType t1 t2
1523 unionCalls :: CallDetails -> CallDetails -> CallDetails
1524 unionCalls c1 c2 = plusVarEnv_C unionCallInfoSet c1 c2
1526 unionCallInfoSet :: CallInfoSet -> CallInfoSet -> CallInfoSet
1527 unionCallInfoSet (CIS f calls1) (CIS _ calls2) = CIS f (calls1 `Map.union` calls2)
1529 callDetailsFVs :: CallDetails -> VarSet
1530 callDetailsFVs calls = foldVarEnv (unionVarSet . callInfoFVs) emptyVarSet calls
1532 callInfoFVs :: CallInfoSet -> VarSet
1533 callInfoFVs (CIS _ call_info) = Map.foldRight (\(_,fv) vs -> unionVarSet fv vs) emptyVarSet call_info
1535 ------------------------------------------------------------
1536 singleCall :: Id -> [Maybe Type] -> [DictExpr] -> UsageDetails
1537 singleCall id tys dicts
1538 = MkUD {ud_binds = emptyBag,
1539 ud_calls = unitVarEnv id $ CIS id $
1540 Map.singleton (CallKey tys) (dicts, call_fvs) }
1542 call_fvs = exprsFreeVars dicts `unionVarSet` tys_fvs
1543 tys_fvs = tyVarsOfTypes (catMaybes tys)
1544 -- The type args (tys) are guaranteed to be part of the dictionary
1545 -- types, because they are just the constrained types,
1546 -- and the dictionary is therefore sure to be bound
1547 -- inside the binding for any type variables free in the type;
1548 -- hence it's safe to neglect tyvars free in tys when making
1549 -- the free-var set for this call
1550 -- BUT I don't trust this reasoning; play safe and include tys_fvs
1552 -- We don't include the 'id' itself.
1554 mkCallUDs :: Id -> [CoreExpr] -> UsageDetails
1556 | not (want_calls_for f) -- Imported from elsewhere
1557 || null theta -- Not overloaded
1558 || not (all isClassPred theta)
1559 -- Only specialise if all overloading is on class params.
1560 -- In ptic, with implicit params, the type args
1561 -- *don't* say what the value of the implicit param is!
1562 || not (spec_tys `lengthIs` n_tyvars)
1563 || not ( dicts `lengthIs` n_dicts)
1564 || not (any interestingDict dicts) -- Note [Interesting dictionary arguments]
1565 -- See also Note [Specialisations already covered]
1566 = -- pprTrace "mkCallUDs: discarding" _trace_doc
1567 emptyUDs -- Not overloaded, or no specialisation wanted
1570 = -- pprTrace "mkCallUDs: keeping" _trace_doc
1571 singleCall f spec_tys dicts
1573 _trace_doc = vcat [ppr f, ppr args, ppr n_tyvars, ppr n_dicts
1574 , ppr (map interestingDict dicts)]
1575 (tyvars, theta, _) = tcSplitSigmaTy (idType f)
1576 constrained_tyvars = tyVarsOfTheta theta
1577 n_tyvars = length tyvars
1578 n_dicts = length theta
1580 spec_tys = [mk_spec_ty tv ty | (tv, Type ty) <- tyvars `zip` args]
1581 dicts = [dict_expr | (_, dict_expr) <- theta `zip` (drop n_tyvars args)]
1584 | tyvar `elemVarSet` constrained_tyvars = Just ty
1585 | otherwise = Nothing
1587 want_calls_for f = isLocalId f || isInlinablePragma (idInlinePragma f)
1590 Note [Interesting dictionary arguments]
1591 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1593 \a.\d:Eq a. let f = ... in ...(f d)...
1594 There really is not much point in specialising f wrt the dictionary d,
1595 because the code for the specialised f is not improved at all, because
1596 d is lambda-bound. We simply get junk specialisations.
1598 What is "interesting"? Just that it has *some* structure.
1601 interestingDict :: CoreExpr -> Bool
1602 -- A dictionary argument is interesting if it has *some* structure
1603 interestingDict (Var v) = hasSomeUnfolding (idUnfolding v)
1604 || isDataConWorkId v
1605 interestingDict (Type _) = False
1606 interestingDict (App fn (Type _)) = interestingDict fn
1607 interestingDict (Note _ a) = interestingDict a
1608 interestingDict (Cast e _) = interestingDict e
1609 interestingDict _ = True
1613 plusUDs :: UsageDetails -> UsageDetails -> UsageDetails
1614 plusUDs (MkUD {ud_binds = db1, ud_calls = calls1})
1615 (MkUD {ud_binds = db2, ud_calls = calls2})
1616 = MkUD { ud_binds = db1 `unionBags` db2
1617 , ud_calls = calls1 `unionCalls` calls2 }
1619 plusUDList :: [UsageDetails] -> UsageDetails
1620 plusUDList = foldr plusUDs emptyUDs
1622 -----------------------------
1623 _dictBindBndrs :: Bag DictBind -> [Id]
1624 _dictBindBndrs dbs = foldrBag ((++) . bindersOf . fst) [] dbs
1626 mkDB :: CoreBind -> DictBind
1627 mkDB bind = (bind, bind_fvs bind)
1629 bind_fvs :: CoreBind -> VarSet
1630 bind_fvs (NonRec bndr rhs) = pair_fvs (bndr,rhs)
1631 bind_fvs (Rec prs) = foldl delVarSet rhs_fvs bndrs
1634 rhs_fvs = unionVarSets (map pair_fvs prs)
1636 pair_fvs :: (Id, CoreExpr) -> VarSet
1637 pair_fvs (bndr, rhs) = exprFreeVars rhs `unionVarSet` idFreeVars bndr
1638 -- Don't forget variables mentioned in the
1639 -- rules of the bndr. C.f. OccAnal.addRuleUsage
1640 -- Also tyvars mentioned in its type; they may not appear in the RHS
1644 flattenDictBinds :: Bag DictBind -> [(Id,CoreExpr)] -> [(Id,CoreExpr)]
1645 flattenDictBinds dbs pairs
1646 = foldrBag add pairs dbs
1648 add (NonRec b r,_) pairs = (b,r) : pairs
1649 add (Rec prs1, _) pairs = prs1 ++ pairs
1651 snocDictBinds :: UsageDetails -> [CoreBind] -> UsageDetails
1652 -- Add ud_binds to the tail end of the bindings in uds
1653 snocDictBinds uds dbs
1654 = uds { ud_binds = ud_binds uds `unionBags`
1655 foldr (consBag . mkDB) emptyBag dbs }
1657 consDictBind :: CoreBind -> UsageDetails -> UsageDetails
1658 consDictBind bind uds = uds { ud_binds = mkDB bind `consBag` ud_binds uds }
1660 addDictBinds :: [DictBind] -> UsageDetails -> UsageDetails
1661 addDictBinds binds uds = uds { ud_binds = listToBag binds `unionBags` ud_binds uds }
1663 snocDictBind :: UsageDetails -> CoreBind -> UsageDetails
1664 snocDictBind uds bind = uds { ud_binds = ud_binds uds `snocBag` mkDB bind }
1666 wrapDictBinds :: Bag DictBind -> [CoreBind] -> [CoreBind]
1667 wrapDictBinds dbs binds
1668 = foldrBag add binds dbs
1670 add (bind,_) binds = bind : binds
1672 wrapDictBindsE :: Bag DictBind -> CoreExpr -> CoreExpr
1673 wrapDictBindsE dbs expr
1674 = foldrBag add expr dbs
1676 add (bind,_) expr = Let bind expr
1678 ----------------------
1679 dumpUDs :: [CoreBndr] -> UsageDetails -> (UsageDetails, Bag DictBind)
1680 -- Used at a lambda or case binder; just dump anything mentioning the binder
1681 dumpUDs bndrs uds@(MkUD { ud_binds = orig_dbs, ud_calls = orig_calls })
1682 | null bndrs = (uds, emptyBag) -- Common in case alternatives
1683 | otherwise = -- pprTrace "dumpUDs" (ppr bndrs $$ ppr free_uds $$ ppr dump_dbs) $
1684 (free_uds, dump_dbs)
1686 free_uds = MkUD { ud_binds = free_dbs, ud_calls = free_calls }
1687 bndr_set = mkVarSet bndrs
1688 (free_dbs, dump_dbs, dump_set) = splitDictBinds orig_dbs bndr_set
1689 free_calls = deleteCallsMentioning dump_set $ -- Drop calls mentioning bndr_set on the floor
1690 deleteCallsFor bndrs orig_calls -- Discard calls for bndr_set; there should be
1691 -- no calls for any of the dicts in dump_dbs
1693 dumpBindUDs :: [CoreBndr] -> UsageDetails -> (UsageDetails, Bag DictBind, Bool)
1694 -- Used at a lambda or case binder; just dump anything mentioning the binder
1695 dumpBindUDs bndrs (MkUD { ud_binds = orig_dbs, ud_calls = orig_calls })
1696 = -- pprTrace "dumpBindUDs" (ppr bndrs $$ ppr free_uds $$ ppr dump_dbs) $
1697 (free_uds, dump_dbs, float_all)
1699 free_uds = MkUD { ud_binds = free_dbs, ud_calls = free_calls }
1700 bndr_set = mkVarSet bndrs
1701 (free_dbs, dump_dbs, dump_set) = splitDictBinds orig_dbs bndr_set
1702 free_calls = deleteCallsFor bndrs orig_calls
1703 float_all = dump_set `intersectsVarSet` callDetailsFVs free_calls
1705 callsForMe :: Id -> UsageDetails -> (UsageDetails, [CallInfo])
1706 callsForMe fn (MkUD { ud_binds = orig_dbs, ud_calls = orig_calls })
1707 = -- pprTrace ("callsForMe")
1709 -- text "Orig dbs =" <+> ppr (_dictBindBndrs orig_dbs),
1710 -- text "Orig calls =" <+> ppr orig_calls,
1711 -- text "Dep set =" <+> ppr dep_set,
1712 -- text "Calls for me =" <+> ppr calls_for_me]) $
1713 (uds_without_me, calls_for_me)
1715 uds_without_me = MkUD { ud_binds = orig_dbs, ud_calls = delVarEnv orig_calls fn }
1716 calls_for_me = case lookupVarEnv orig_calls fn of
1718 Just (CIS _ calls) -> filter_dfuns (Map.toList calls)
1720 dep_set = foldlBag go (unitVarSet fn) orig_dbs
1721 go dep_set (db,fvs) | fvs `intersectsVarSet` dep_set
1722 = extendVarSetList dep_set (bindersOf db)
1723 | otherwise = dep_set
1725 -- Note [Specialisation of dictionary functions]
1726 filter_dfuns | isDFunId fn = filter ok_call
1727 | otherwise = \cs -> cs
1729 ok_call (_, (_,fvs)) = not (fvs `intersectsVarSet` dep_set)
1731 ----------------------
1732 splitDictBinds :: Bag DictBind -> IdSet -> (Bag DictBind, Bag DictBind, IdSet)
1733 -- Returns (free_dbs, dump_dbs, dump_set)
1734 splitDictBinds dbs bndr_set
1735 = foldlBag split_db (emptyBag, emptyBag, bndr_set) dbs
1736 -- Important that it's foldl not foldr;
1737 -- we're accumulating the set of dumped ids in dump_set
1739 split_db (free_dbs, dump_dbs, dump_idset) db@(bind, fvs)
1740 | dump_idset `intersectsVarSet` fvs -- Dump it
1741 = (free_dbs, dump_dbs `snocBag` db,
1742 extendVarSetList dump_idset (bindersOf bind))
1744 | otherwise -- Don't dump it
1745 = (free_dbs `snocBag` db, dump_dbs, dump_idset)
1748 ----------------------
1749 deleteCallsMentioning :: VarSet -> CallDetails -> CallDetails
1750 -- Remove calls *mentioning* bs
1751 deleteCallsMentioning bs calls
1752 = mapVarEnv filter_calls calls
1754 filter_calls :: CallInfoSet -> CallInfoSet
1755 filter_calls (CIS f calls) = CIS f (Map.filter keep_call calls)
1756 keep_call (_, fvs) = not (fvs `intersectsVarSet` bs)
1758 deleteCallsFor :: [Id] -> CallDetails -> CallDetails
1759 -- Remove calls *for* bs
1760 deleteCallsFor bs calls = delVarEnvList calls bs
1764 %************************************************************************
1766 \subsubsection{Boring helper functions}
1768 %************************************************************************
1771 type SpecM a = UniqSM a
1773 runSpecM:: SpecM a -> CoreM a
1774 runSpecM spec = do { us <- getUniqueSupplyM
1775 ; return (initUs_ us spec) }
1777 mapAndCombineSM :: (a -> SpecM (b, UsageDetails)) -> [a] -> SpecM ([b], UsageDetails)
1778 mapAndCombineSM _ [] = return ([], emptyUDs)
1779 mapAndCombineSM f (x:xs) = do (y, uds1) <- f x
1780 (ys, uds2) <- mapAndCombineSM f xs
1781 return (y:ys, uds1 `plusUDs` uds2)
1783 cloneBindSM :: Subst -> CoreBind -> SpecM (Subst, Subst, CoreBind)
1784 -- Clone the binders of the bind; return new bind with the cloned binders
1785 -- Return the substitution to use for RHSs, and the one to use for the body
1786 cloneBindSM subst (NonRec bndr rhs) = do
1787 us <- getUniqueSupplyM
1788 let (subst', bndr') = cloneIdBndr subst us bndr
1789 return (subst, subst', NonRec bndr' rhs)
1791 cloneBindSM subst (Rec pairs) = do
1792 us <- getUniqueSupplyM
1793 let (subst', bndrs') = cloneRecIdBndrs subst us (map fst pairs)
1794 return (subst', subst', Rec (bndrs' `zip` map snd pairs))
1796 newDictBndrs :: Subst -> [CoreBndr] -> SpecM (Subst, [CoreBndr])
1797 -- Make up completely fresh binders for the dictionaries
1798 -- Their bindings are going to float outwards
1799 newDictBndrs subst bndrs
1800 = do { bndrs' <- mapM new bndrs
1801 ; let subst' = extendIdSubstList subst
1802 [(d, Var d') | (d,d') <- bndrs `zip` bndrs']
1803 ; return (subst', bndrs' ) }
1805 new b = do { uniq <- getUniqueM
1807 ty' = CoreSubst.substTy subst (idType b)
1808 ; return (mkUserLocal (nameOccName n) uniq ty' (getSrcSpan n)) }
1810 newSpecIdSM :: Id -> Type -> SpecM Id
1811 -- Give the new Id a similar occurrence name to the old one
1812 newSpecIdSM old_id new_ty
1813 = do { uniq <- getUniqueM
1814 ; let name = idName old_id
1815 new_occ = mkSpecOcc (nameOccName name)
1816 new_id = mkUserLocal new_occ uniq new_ty (getSrcSpan name)
1821 Old (but interesting) stuff about unboxed bindings
1822 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1824 What should we do when a value is specialised to a *strict* unboxed value?
1826 map_*_* f (x:xs) = let h = f x
1830 Could convert let to case:
1832 map_*_Int# f (x:xs) = case f x of h# ->
1836 This may be undesirable since it forces evaluation here, but the value
1837 may not be used in all branches of the body. In the general case this
1838 transformation is impossible since the mutual recursion in a letrec
1839 cannot be expressed as a case.
1841 There is also a problem with top-level unboxed values, since our
1842 implementation cannot handle unboxed values at the top level.
1844 Solution: Lift the binding of the unboxed value and extract it when it
1847 map_*_Int# f (x:xs) = let h = case (f x) of h# -> _Lift h#
1852 Now give it to the simplifier and the _Lifting will be optimised away.
1854 The benfit is that we have given the specialised "unboxed" values a
1855 very simplep lifted semantics and then leave it up to the simplifier to
1856 optimise it --- knowing that the overheads will be removed in nearly
1859 In particular, the value will only be evaluted in the branches of the
1860 program which use it, rather than being forced at the point where the
1861 value is bound. For example:
1863 filtermap_*_* p f (x:xs)
1870 filtermap_*_Int# p f (x:xs)
1871 = let h = case (f x) of h# -> _Lift h#
1874 True -> case h of _Lift h#
1878 The binding for h can still be inlined in the one branch and the
1879 _Lifting eliminated.
1882 Question: When won't the _Lifting be eliminated?
1884 Answer: When they at the top-level (where it is necessary) or when
1885 inlining would duplicate work (or possibly code depending on
1886 options). However, the _Lifting will still be eliminated if the
1887 strictness analyser deems the lifted binding strict.