2 % (c) The GRASP/AQUA Project, Glasgow University, 1993-1996
4 \section[Specialise]{Stamping out overloading, and (optionally) polymorphism}
12 #include "HsVersions.h"
14 import MkId ( mkUserLocal )
15 import Id ( Id, DictVar, idType,
17 getIdSpecialisation, setIdSpecialisation,
19 IdSet, mkIdSet, addOneToIdSet, intersectIdSets, isEmptyIdSet,
20 emptyIdSet, unionIdSets, minusIdSet, unitIdSet, elementOfIdSet,
22 IdEnv, mkIdEnv, lookupIdEnv, addOneToIdEnv, delOneFromIdEnv
25 import Type ( Type, mkTyVarTy, splitSigmaTy, instantiateTy, isDictTy,
26 tyVarsOfType, tyVarsOfTypes, applyTys, mkForAllTys
28 import TyCon ( TyCon )
30 TyVarSet, mkTyVarSet, isEmptyTyVarSet, intersectTyVarSets,
31 elementOfTyVarSet, unionTyVarSets, emptyTyVarSet,
32 TyVarEnv, mkTyVarEnv, delFromTyVarEnv
35 import PprCore () -- Instances
36 import Name ( NamedThing(..), getSrcLoc )
37 import SpecEnv ( addToSpecEnv, lookupSpecEnv, specEnvValues )
39 import UniqSupply ( UniqSupply,
40 UniqSM, initUs, thenUs, returnUs, getUnique, mapUs
44 import Maybes ( MaybeErr(..), maybeToBool )
46 import List ( partition )
47 import Util ( zipEqual )
54 %************************************************************************
56 \subsection[notes-Specialise]{Implementation notes [SLPJ, Aug 18 1993]}
58 %************************************************************************
60 These notes describe how we implement specialisation to eliminate
61 overloading, and optionally to eliminate unboxed polymorphism, and
64 The specialisation pass is a partial evaluator which works on Core
65 syntax, complete with all the explicit dictionary application,
66 abstraction and construction as added by the type checker. The
67 existing type checker remains largely as it is.
69 One important thought: the {\em types} passed to an overloaded
70 function, and the {\em dictionaries} passed are mutually redundant.
71 If the same function is applied to the same type(s) then it is sure to
72 be applied to the same dictionary(s)---or rather to the same {\em
73 values}. (The arguments might look different but they will evaluate
76 Second important thought: we know that we can make progress by
77 treating dictionary arguments as static and worth specialising on. So
78 we can do without binding-time analysis, and instead specialise on
79 dictionary arguments and no others.
88 and suppose f is overloaded.
90 STEP 1: CALL-INSTANCE COLLECTION
92 We traverse <body>, accumulating all applications of f to types and
95 (Might there be partial applications, to just some of its types and
96 dictionaries? In principle yes, but in practice the type checker only
97 builds applications of f to all its types and dictionaries, so partial
98 applications could only arise as a result of transformation, and even
99 then I think it's unlikely. In any case, we simply don't accumulate such
100 partial applications.)
102 There's a choice of whether to collect details of all *polymorphic* functions
103 or simply all *overloaded* ones. How to sort this out?
104 Pass in a predicate on the function to say if it is "interesting"?
105 This is dependent on the user flags: SpecialiseOverloaded
111 So now we have a collection of calls to f:
115 Notice that f may take several type arguments. To avoid ambiguity, we
116 say that f is called at type t1/t2 and t3/t4.
118 We take equivalence classes using equality of the *types* (ignoring
119 the dictionary args, which as mentioned previously are redundant).
121 STEP 3: SPECIALISATION
123 For each equivalence class, choose a representative (f t1 t2 d1 d2),
124 and create a local instance of f, defined thus:
126 f@t1/t2 = <f_rhs> t1 t2 d1 d2
128 (f_rhs presumably has some big lambdas and dictionary lambdas, so lots
129 of simplification will now result.) Then we should recursively do
132 The new id has its own unique, but its print-name (if exported) has
133 an explicit representation of the instance types t1/t2.
135 Add this new id to f's IdInfo, to record that f has a specialised version.
137 Before doing any of this, check that f's IdInfo doesn't already
138 tell us about an existing instance of f at the required type/s.
139 (This might happen if specialisation was applied more than once, or
140 it might arise from user SPECIALIZE pragmas.)
144 Wait a minute! What if f is recursive? Then we can't just plug in
145 its right-hand side, can we?
147 But it's ok. The type checker *always* creates non-recursive definitions
148 for overloaded recursive functions. For example:
150 f x = f (x+x) -- Yes I know its silly
154 f a (d::Num a) = let p = +.sel a d
156 letrec fl (y::a) = fl (p y y)
160 We still have recusion for non-overloadd functions which we
161 speciailise, but the recursive call should get speciailised to the
162 same recursive version.
168 All this is crystal clear when the function is applied to *constant
169 types*; that is, types which have no type variables inside. But what if
170 it is applied to non-constant types? Suppose we find a call of f at type
171 t1/t2. There are two possibilities:
173 (a) The free type variables of t1, t2 are in scope at the definition point
174 of f. In this case there's no problem, we proceed just as before. A common
175 example is as follows. Here's the Haskell:
180 After typechecking we have
182 g a (d::Num a) (y::a) = let f b (d'::Num b) (x::b) = +.sel b d' x x
183 in +.sel a d (f a d y) (f a d y)
185 Notice that the call to f is at type type "a"; a non-constant type.
186 Both calls to f are at the same type, so we can specialise to give:
188 g a (d::Num a) (y::a) = let f@a (x::a) = +.sel a d x x
189 in +.sel a d (f@a y) (f@a y)
192 (b) The other case is when the type variables in the instance types
193 are *not* in scope at the definition point of f. The example we are
194 working with above is a good case. There are two instances of (+.sel a d),
195 but "a" is not in scope at the definition of +.sel. Can we do anything?
196 Yes, we can "common them up", a sort of limited common sub-expression deal.
199 g a (d::Num a) (y::a) = let +.sel@a = +.sel a d
200 f@a (x::a) = +.sel@a x x
201 in +.sel@a (f@a y) (f@a y)
203 This can save work, and can't be spotted by the type checker, because
204 the two instances of +.sel weren't originally at the same type.
208 * There are quite a few variations here. For example, the defn of
209 +.sel could be floated ouside the \y, to attempt to gain laziness.
210 It certainly mustn't be floated outside the \d because the d has to
213 * We don't want to inline f_rhs in this case, because
214 that will duplicate code. Just commoning up the call is the point.
216 * Nothing gets added to +.sel's IdInfo.
218 * Don't bother unless the equivalence class has more than one item!
220 Not clear whether this is all worth it. It is of course OK to
221 simply discard call-instances when passing a big lambda.
223 Polymorphism 2 -- Overloading
225 Consider a function whose most general type is
227 f :: forall a b. Ord a => [a] -> b -> b
229 There is really no point in making a version of g at Int/Int and another
230 at Int/Bool, because it's only instancing the type variable "a" which
231 buys us any efficiency. Since g is completely polymorphic in b there
232 ain't much point in making separate versions of g for the different
235 That suggests that we should identify which of g's type variables
236 are constrained (like "a") and which are unconstrained (like "b").
237 Then when taking equivalence classes in STEP 2, we ignore the type args
238 corresponding to unconstrained type variable. In STEP 3 we make
239 polymorphic versions. Thus:
241 f@t1/ = /\b -> <f_rhs> t1 b d1 d2
243 This seems pretty simple, and a Good Thing.
245 Polymorphism 3 -- Unboxed
248 If we are speciailising at unboxed types we must speciailise
249 regardless of the overloading constraint. In the exaple above it is
250 worth speciailising at types Int/Int#, Int/Bool# and a/Int#, Int#/Int#
253 Note that specialising an overloaded type at an uboxed type requires
254 an unboxed instance -- we cannot default to an unspecialised version!
261 f x = let g p q = p==q
267 Before specialisation, leaving out type abstractions we have
269 f df x = let g :: Eq a => a -> a -> Bool
271 h :: Num a => a -> a -> (a, Bool)
272 h dh r s = let deq = eqFromNum dh
273 in (+ dh r s, g deq r s)
277 After specialising h we get a specialised version of h, like this:
279 h' r s = let deq = eqFromNum df
280 in (+ df r s, g deq r s)
282 But we can't naively make an instance for g from this, because deq is not in scope
283 at the defn of g. Instead, we have to float out the (new) defn of deq
284 to widen its scope. Notice that this floating can't be done in advance -- it only
285 shows up when specialisation is done.
287 DELICATE MATTER: the way we tell a dictionary binding is by looking to
288 see if it has a Dict type. If the type has been "undictify'd", so that
289 it looks like a tuple, then the dictionary binding won't be floated, and
290 an opportunity to specialise might be lost.
292 User SPECIALIZE pragmas
293 ~~~~~~~~~~~~~~~~~~~~~~~
294 Specialisation pragmas can be digested by the type checker, and implemented
295 by adding extra definitions along with that of f, in the same way as before
297 f@t1/t2 = <f_rhs> t1 t2 d1 d2
299 Indeed the pragmas *have* to be dealt with by the type checker, because
300 only it knows how to build the dictionaries d1 and d2! For example
302 g :: Ord a => [a] -> [a]
303 {-# SPECIALIZE f :: [Tree Int] -> [Tree Int] #-}
305 Here, the specialised version of g is an application of g's rhs to the
306 Ord dictionary for (Tree Int), which only the type checker can conjure
307 up. There might not even *be* one, if (Tree Int) is not an instance of
308 Ord! (All the other specialision has suitable dictionaries to hand
311 Problem. The type checker doesn't have to hand a convenient <f_rhs>, because
312 it is buried in a complex (as-yet-un-desugared) binding group.
315 f@t1/t2 = f* t1 t2 d1 d2
317 where f* is the Id f with an IdInfo which says "inline me regardless!".
318 Indeed all the specialisation could be done in this way.
319 That in turn means that the simplifier has to be prepared to inline absolutely
320 any in-scope let-bound thing.
323 Again, the pragma should permit polymorphism in unconstrained variables:
325 h :: Ord a => [a] -> b -> b
326 {-# SPECIALIZE h :: [Int] -> b -> b #-}
328 We *insist* that all overloaded type variables are specialised to ground types,
329 (and hence there can be no context inside a SPECIALIZE pragma).
330 We *permit* unconstrained type variables to be specialised to
332 - or left as a polymorphic type variable
333 but nothing in between. So
335 {-# SPECIALIZE h :: [Int] -> [c] -> [c] #-}
337 is *illegal*. (It can be handled, but it adds complication, and gains the
341 SPECIALISING INSTANCE DECLARATIONS
342 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
345 instance Foo a => Foo [a] where
347 {-# SPECIALIZE instance Foo [Int] #-}
349 The original instance decl creates a dictionary-function
352 dfun.Foo.List :: forall a. Foo a -> Foo [a]
354 The SPECIALIZE pragma just makes a specialised copy, just as for
355 ordinary function definitions:
357 dfun.Foo.List@Int :: Foo [Int]
358 dfun.Foo.List@Int = dfun.Foo.List Int dFooInt
360 The information about what instance of the dfun exist gets added to
361 the dfun's IdInfo in the same way as a user-defined function too.
363 In fact, matters are a little bit more complicated than this.
364 When we make one of these specialised instances, we are defining
365 a constant dictionary, and so we want immediate access to its constant
366 methods and superclasses. Indeed, these constant methods and superclasses
367 must be in the IdInfo for the class selectors! We need help from the
368 typechecker to sort this out, perhaps by generating a separate IdInfo
371 Automatic instance decl specialisation?
372 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
373 Can instance decls be specialised automatically? It's tricky.
374 We could collect call-instance information for each dfun, but
375 then when we specialised their bodies we'd get new call-instances
376 for ordinary functions; and when we specialised their bodies, we might get
377 new call-instances of the dfuns, and so on. This all arises because of
378 the unrestricted mutual recursion between instance decls and value decls.
380 Furthermore, instance decls are usually exported and used non-locally,
381 so we'll want to compile enough to get those specialisations done.
383 Lastly, there's no such thing as a local instance decl, so we can
384 survive solely by spitting out *usage* information, and then reading that
385 back in as a pragma when next compiling the file. So for now,
386 we only specialise instance decls in response to pragmas.
388 That means that even if an instance decl ain't otherwise exported it
389 needs to be spat out as with a SPECIALIZE pragma. Furthermore, it needs
390 something to say which module defined the instance, so the usage info
391 can be fed into the right reqts info file. Blegh.
394 SPECIAILISING DATA DECLARATIONS
395 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
397 With unboxed specialisation (or full specialisation) we also require
398 data types (and their constructors) to be speciailised on unboxed
401 In addition to normal call instances we gather TyCon call instances at
402 unboxed types, determine equivalence classes for the locally defined
403 TyCons and build speciailised data constructor Ids for each TyCon and
404 substitute these in the Con calls.
406 We need the list of local TyCons to partition the TyCon instance info.
407 We pass out a FiniteMap from local TyCons to Specialised Instances to
408 give to the interface and code genertors.
410 N.B. The specialised data constructors reference the original data
411 constructor and type constructor which do not have the updated
412 specialisation info attached. Any specialisation info must be
413 extracted from the TyCon map returned.
416 SPITTING OUT USAGE INFORMATION
417 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
419 To spit out usage information we need to traverse the code collecting
420 call-instance information for all imported (non-prelude?) functions
421 and data types. Then we equivalence-class it and spit it out.
423 This is done at the top-level when all the call instances which escape
424 must be for imported functions and data types.
427 Partial specialisation by pragmas
428 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
429 What about partial specialisation:
431 k :: (Ord a, Eq b) => [a] -> b -> b -> [a]
432 {-# SPECIALIZE k :: Eq b => [Int] -> b -> b -> [a] #-}
436 {-# SPECIALIZE k :: Eq b => [Int] -> [b] -> [b] -> [a] #-}
438 Seems quite reasonable. Similar things could be done with instance decls:
440 instance (Foo a, Foo b) => Foo (a,b) where
442 {-# SPECIALIZE instance Foo a => Foo (a,Int) #-}
443 {-# SPECIALIZE instance Foo b => Foo (Int,b) #-}
445 Ho hum. Things are complex enough without this. I pass.
448 Requirements for the simplifer
449 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
450 The simplifier has to be able to take advantage of the specialisation.
452 * When the simplifier finds an application of a polymorphic f, it looks in
453 f's IdInfo in case there is a suitable instance to call instead. This converts
455 f t1 t2 d1 d2 ===> f_t1_t2
457 Note that the dictionaries get eaten up too!
459 * Dictionary selection operations on constant dictionaries must be
462 +.sel Int d ===> +Int
464 The obvious way to do this is in the same way as other specialised
465 calls: +.sel has inside it some IdInfo which tells that if it's applied
466 to the type Int then it should eat a dictionary and transform to +Int.
468 In short, dictionary selectors need IdInfo inside them for constant
471 * Exactly the same applies if a superclass dictionary is being
474 Eq.sel Int d ===> dEqInt
476 * Something similar applies to dictionary construction too. Suppose
477 dfun.Eq.List is the function taking a dictionary for (Eq a) to
478 one for (Eq [a]). Then we want
480 dfun.Eq.List Int d ===> dEq.List_Int
482 Where does the Eq [Int] dictionary come from? It is built in
483 response to a SPECIALIZE pragma on the Eq [a] instance decl.
485 In short, dfun Ids need IdInfo with a specialisation for each
486 constant instance of their instance declaration.
489 What does the specialisation IdInfo look like?
490 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
493 [Maybe Type] -- Instance types
494 Int -- No of dicts to eat
495 Id -- Specialised version
497 For example, if f has this SpecInfo:
499 SpecInfo [Just t1, Nothing, Just t3] 2 f'
503 f t1 t2 t3 d1 d2 ===> f t2
505 The "Nothings" identify type arguments in which the specialised
506 version is polymorphic.
508 What can't be done this way?
509 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
510 There is no way, post-typechecker, to get a dictionary for (say)
511 Eq a from a dictionary for Eq [a]. So if we find
515 we can't transform to
520 eqList :: (a->a->Bool) -> [a] -> [a] -> Bool
522 Of course, we currently have no way to automatically derive
523 eqList, nor to connect it to the Eq [a] instance decl, but you
524 can imagine that it might somehow be possible. Taking advantage
525 of this is permanently ruled out.
527 Still, this is no great hardship, because we intend to eliminate
528 overloading altogether anyway!
533 What about types/classes mentioned in SPECIALIZE pragmas spat out,
534 but not otherwise exported. Even if they are exported, what about
535 their original names.
537 Suggestion: use qualified names in pragmas, omitting module for
538 prelude and "this module".
545 f a (d::Num a) = let g = ...
547 ...(let d1::Ord a = Num.Ord.sel a d in g a d1)...
549 Here, g is only called at one type, but the dictionary isn't in scope at the
550 definition point for g. Usually the type checker would build a
551 definition for d1 which enclosed g, but the transformation system
552 might have moved d1's defn inward.
558 What should we do when a value is specialised to a *strict* unboxed value?
560 map_*_* f (x:xs) = let h = f x
564 Could convert let to case:
566 map_*_Int# f (x:xs) = case f x of h# ->
570 This may be undesirable since it forces evaluation here, but the value
571 may not be used in all branches of the body. In the general case this
572 transformation is impossible since the mutual recursion in a letrec
573 cannot be expressed as a case.
575 There is also a problem with top-level unboxed values, since our
576 implementation cannot handle unboxed values at the top level.
578 Solution: Lift the binding of the unboxed value and extract it when it
581 map_*_Int# f (x:xs) = let h = case (f x) of h# -> _Lift h#
586 Now give it to the simplifier and the _Lifting will be optimised away.
588 The benfit is that we have given the specialised "unboxed" values a
589 very simplep lifted semantics and then leave it up to the simplifier to
590 optimise it --- knowing that the overheads will be removed in nearly
593 In particular, the value will only be evaluted in the branches of the
594 program which use it, rather than being forced at the point where the
595 value is bound. For example:
597 filtermap_*_* p f (x:xs)
604 filtermap_*_Int# p f (x:xs)
605 = let h = case (f x) of h# -> _Lift h#
608 True -> case h of _Lift h#
612 The binding for h can still be inlined in the one branch and the
616 Question: When won't the _Lifting be eliminated?
618 Answer: When they at the top-level (where it is necessary) or when
619 inlining would duplicate work (or possibly code depending on
620 options). However, the _Lifting will still be eliminated if the
621 strictness analyser deems the lifted binding strict.
624 A note about non-tyvar dictionaries
625 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
626 Some Ids have types like
628 forall a,b,c. Eq a -> Ord [a] -> tau
630 This seems curious at first, because we usually only have dictionary
631 args whose types are of the form (C a) where a is a type variable.
632 But this doesn't hold for the functions arising from instance decls,
633 which sometimes get arguements with types of form (C (T a)) for some
636 Should we specialise wrt this compound-type dictionary? We used to say
638 "This is a heuristic judgement, as indeed is the fact that we
639 specialise wrt only dictionaries. We choose *not* to specialise
640 wrt compound dictionaries because at the moment the only place
641 they show up is in instance decls, where they are simply plugged
642 into a returned dictionary. So nothing is gained by specialising
645 But it is simpler and more uniform to specialise wrt these dicts too;
646 and in future GHC is likely to support full fledged type signatures
648 f ;: Eq [(a,b)] => ...
651 %************************************************************************
653 \subsubsection{The new specialiser}
655 %************************************************************************
657 Our basic game plan is this. For let(rec) bound function
658 f :: (C a, D c) => (a,b,c,d) -> Bool
660 * Find any specialised calls of f, (f ts ds), where
661 ts are the type arguments t1 .. t4, and
662 ds are the dictionary arguments d1 .. d2.
664 * Add a new definition for f1 (say):
666 f1 = /\ b d -> (..body of f..) t1 b t3 d d1 d2
668 Note that we abstract over the unconstrained type arguments.
672 [t1,b,t3,d] |-> \d1 d2 -> f1 b d
674 to the specialisations of f. This will be used by the
675 simplifier to replace calls
676 (f t1 t2 t3 t4) da db
678 (\d1 d1 -> f1 t2 t4) da db
680 All the stuff about how many dictionaries to discard, and what types
681 to apply the specialised function to, are handled by the fact that the
682 SpecEnv contains a template for the result of the specialisation.
684 We don't build *partial* specialisations for f. For example:
686 f :: Eq a => a -> a -> Bool
687 {-# SPECIALISE f :: (Eq b, Eq c) => (b,c) -> (b,c) -> Bool #-}
689 Here, little is gained by making a specialised copy of f.
690 There's a distinct danger that the specialised version would
691 first build a dictionary for (Eq b, Eq c), and then select the (==)
692 method from it! Even if it didn't, not a great deal is saved.
694 We do, however, generate polymorphic, but not overloaded, specialisations:
696 f :: Eq a => [a] -> b -> b -> b
697 {#- SPECIALISE f :: [Int] -> b -> b -> b #-}
699 Hence, the invariant is this:
701 *** no specialised version is overloaded ***
704 %************************************************************************
706 \subsubsection{The exported function}
708 %************************************************************************
711 specProgram :: UniqSupply -> [CoreBinding] -> [CoreBinding]
713 = initSM us (go binds `thenSM` \ (binds', uds') ->
714 returnSM (dumpAllDictBinds uds' binds')
717 go [] = returnSM ([], emptyUDs)
718 go (bind:binds) = go binds `thenSM` \ (binds', uds) ->
719 specBind bind uds `thenSM` \ (bind', uds') ->
720 returnSM (bind' ++ binds', uds')
723 %************************************************************************
725 \subsubsection{@specExpr@: the main function}
727 %************************************************************************
730 specExpr :: CoreExpr -> SpecM (CoreExpr, UsageDetails)
732 ---------------- First the easy cases --------------------
733 specExpr e@(Var _) = returnSM (e, emptyUDs)
734 specExpr e@(Lit _) = returnSM (e, emptyUDs)
735 specExpr e@(Con _ _) = returnSM (e, emptyUDs)
736 specExpr e@(Prim _ _) = returnSM (e, emptyUDs)
738 specExpr (Note note body)
739 = specExpr body `thenSM` \ (body', uds) ->
740 returnSM (Note note body', uds)
743 ---------------- Applications might generate a call instance --------------------
744 specExpr e@(App fun arg)
747 go (App fun arg) args = go fun (arg:args)
748 go (Var f) args = returnSM (e, mkCallUDs f args)
749 go other args = specExpr other `thenSM` \ (e', uds) ->
750 returnSM (foldl App e' args, uds)
752 ---------------- Lambda/case require dumping of usage details --------------------
754 = specExpr body `thenSM` \ (body', uds) ->
756 (filtered_uds, body'') = dumpUDs bndrs uds body'
758 returnSM (foldr Lam body'' bndrs, filtered_uds)
760 (bndrs, body) = go [] e
762 -- More efficient to collect a group of binders together all at once
763 go bndrs (Lam bndr e) = go (bndr:bndrs) e
764 go bndrs e = (reverse bndrs, e)
767 specExpr (Case scrut alts)
768 = specExpr scrut `thenSM` \ (scrut', uds_scrut) ->
769 spec_alts alts `thenSM` \ (alts', uds_alts) ->
770 returnSM (Case scrut' alts', uds_scrut `plusUDs` uds_alts)
772 spec_alts (AlgAlts alts deflt)
773 = mapAndCombineSM spec_alg_alt alts `thenSM` \ (alts', uds1) ->
774 spec_deflt deflt `thenSM` \ (deflt', uds2) ->
775 returnSM (AlgAlts alts' deflt', uds1 `plusUDs` uds2)
777 spec_alts (PrimAlts alts deflt)
778 = mapAndCombineSM spec_prim_alt alts `thenSM` \ (alts', uds1) ->
779 spec_deflt deflt `thenSM` \ (deflt', uds2) ->
780 returnSM (PrimAlts alts' deflt', uds1 `plusUDs` uds2)
782 spec_alg_alt (con, args, rhs)
783 = specExpr rhs `thenSM` \ (rhs', uds) ->
785 (uds', rhs'') = dumpUDs (map ValBinder args) uds rhs'
787 returnSM ((con, args, rhs''), uds')
789 spec_prim_alt (lit, rhs)
790 = specExpr rhs `thenSM` \ (rhs', uds) ->
791 returnSM ((lit, rhs'), uds)
793 spec_deflt NoDefault = returnSM (NoDefault, emptyUDs)
794 spec_deflt (BindDefault arg rhs)
795 = specExpr rhs `thenSM` \ (rhs', uds) ->
797 (uds', rhs'') = dumpUDs [ValBinder arg] uds rhs'
799 returnSM (BindDefault arg rhs'', uds')
801 ---------------- Finally, let is the interesting case --------------------
802 specExpr (Let bind body)
803 = -- Deal with the body
804 specExpr body `thenSM` \ (body', body_uds) ->
806 -- Deal with the bindings
807 specBind bind body_uds `thenSM` \ (binds', uds) ->
810 returnSM (foldr Let body' binds', uds)
813 %************************************************************************
815 \subsubsection{Dealing with a binding}
817 %************************************************************************
820 specBind :: CoreBinding
821 -> UsageDetails -- Info on how the scope of the binding
822 -> SpecM ([CoreBinding], -- New bindings
823 UsageDetails) -- And info to pass upstream
825 specBind (NonRec bndr rhs) body_uds
826 | isDictTy (idType bndr)
827 = -- It's a dictionary binding
828 -- Pick it up and float it outwards.
829 specExpr rhs `thenSM` \ (rhs', rhs_uds) ->
831 all_uds = rhs_uds `plusUDs` addDictBind body_uds bndr rhs'
833 returnSM ([], all_uds)
836 = -- Deal with the RHS, specialising it according
837 -- to the calls found in the body
838 specDefn (calls body_uds) (bndr,rhs) `thenSM` \ ((bndr',rhs'), spec_defns, spec_uds) ->
840 (all_uds, (dict_binds, dump_calls))
841 = splitUDs [ValBinder bndr] (spec_uds `plusUDs` body_uds)
843 -- If we make specialisations then we Rec the whole lot together
844 -- If not, leave it as a NonRec
845 new_bind | null spec_defns = NonRec bndr' rhs'
846 | otherwise = Rec ((bndr',rhs'):spec_defns)
848 returnSM ( new_bind : dict_binds, all_uds )
850 specBind (Rec pairs) body_uds
851 = mapSM (specDefn (calls body_uds)) pairs `thenSM` \ stuff ->
853 (pairs', spec_defns_s, spec_uds_s) = unzip3 stuff
854 spec_defns = concat spec_defns_s
855 spec_uds = plusUDList spec_uds_s
856 (all_uds, (dict_binds, dump_calls))
857 = splitUDs (map (ValBinder . fst) pairs) (spec_uds `plusUDs` body_uds)
858 new_bind = Rec (spec_defns ++ pairs')
860 returnSM ( new_bind : dict_binds, all_uds )
862 specDefn :: CallDetails -- Info on how it is used in its scope
863 -> (Id, CoreExpr) -- The thing being bound and its un-processed RHS
864 -> SpecM ((Id, CoreExpr), -- The thing and its processed RHS
865 -- the Id may now have specialisations attached
866 [(Id,CoreExpr)], -- Extra, specialised bindings
867 UsageDetails -- Stuff to fling upwards from the RHS and its
868 ) -- specialised versions
870 specDefn calls (fn, rhs)
871 -- The first case is the interesting one
872 | n_tyvars == length rhs_tyvars -- Rhs of fn's defn has right number of big lambdas
873 && n_dicts <= length rhs_bndrs -- and enough dict args
874 && not (null calls_for_me) -- And there are some calls to specialise
875 = -- Specialise the body of the function
876 specExpr body `thenSM` \ (body', body_uds) ->
878 (float_uds, bound_uds@(dict_binds,_)) = splitUDs rhs_bndrs body_uds
881 -- Make a specialised version for each call in calls_for_me
882 mapSM (spec_call bound_uds) calls_for_me `thenSM` \ stuff ->
884 (spec_defns, spec_uds, spec_env_stuff) = unzip3 stuff
886 fn' = addIdSpecialisations fn spec_env_stuff
887 rhs' = foldr Lam (foldr Let body' dict_binds) rhs_bndrs
889 returnSM ((fn',rhs'),
891 float_uds `plusUDs` plusUDList spec_uds)
893 | otherwise -- No calls or RHS doesn't fit our preconceptions
894 = specExpr rhs `thenSM` \ (rhs', rhs_uds) ->
895 returnSM ((fn, rhs'), [], rhs_uds)
899 (tyvars, theta, tau) = splitSigmaTy fn_type
900 n_tyvars = length tyvars
901 n_dicts = length theta
902 mk_spec_tys call_ts = zipWith mk_spec_ty call_ts tyvars
904 mk_spec_ty (Just ty) _ = ty
905 mk_spec_ty Nothing tyvar = mkTyVarTy tyvar
907 (rhs_tyvars, rhs_ids, rhs_body) = collectBinders rhs
908 rhs_dicts = take n_dicts rhs_ids
909 rhs_bndrs = map TyBinder rhs_tyvars ++ map ValBinder rhs_dicts
910 body = mkValLam (drop n_dicts rhs_ids) rhs_body
911 -- Glue back on the non-dict lambdas
913 calls_for_me = case lookupFM calls fn of
915 Just cs -> fmToList cs
917 -- Filter out calls for which we already have a specialisation
918 calls_to_spec = filter spec_me calls_for_me
919 spec_me (call_ts, _) = not (maybeToBool (lookupSpecEnv id_spec_env (mk_spec_tys call_ts)))
920 id_spec_env = getIdSpecialisation fn
922 ----------------------------------------------------------
923 -- Specialise to one particular call pattern
924 spec_call :: ProtoUsageDetails -- From the original body, captured by
925 -- the dictionary lambdas
926 -> ([Maybe Type], [DictVar]) -- Call instance
927 -> SpecM ((Id,CoreExpr), -- Specialised definition
928 UsageDetails, -- Usage details from specialised body
929 ([TyVar], [Type], CoreExpr)) -- Info for the Id's SpecEnv
930 spec_call bound_uds (call_ts, call_ds)
931 = ASSERT( length call_ts == n_tyvars && length call_ds == n_dicts )
932 -- Calls are only recorded for properly-saturated applications
934 -- Supppose the call is for f [Just t1, Nothing, Just t3, Nothing] [d1, d2]
936 -- Construct the new binding
937 -- f1 = /\ b d -> (..rhs of f..) t1 b t3 d d1 d2
938 -- and the type of this binder
940 spec_tyvars = [tyvar | (tyvar, Nothing) <- tyvars `zip` call_ts]
941 spec_tys = mk_spec_tys call_ts
942 spec_rhs = mkTyLam spec_tyvars $
943 mkGenApp rhs (map TyArg spec_tys ++ map VarArg call_ds)
944 spec_id_ty = mkForAllTys spec_tyvars (instantiateTy ty_env tau)
945 ty_env = mkTyVarEnv (zipEqual "spec_call" tyvars spec_tys)
947 newIdSM fn spec_id_ty `thenSM` \ spec_f ->
950 -- Construct the stuff for f's spec env
951 -- [b,d] [t1,b,t3,d] |-> \d1 d2 -> f1 b d
953 spec_env_rhs = mkValLam call_ds $
954 mkTyApp (Var spec_f) $
955 map mkTyVarTy spec_tyvars
956 spec_env_info = (spec_tyvars, spec_tys, spec_env_rhs)
959 -- Specialise the UDs from f's RHS
961 -- Only the overloaded tyvars should be free in the uds
962 ty_env = [ (rhs_tyvar,ty)
963 | (rhs_tyvar, Just ty) <- zipEqual "specUDs1" rhs_tyvars call_ts
965 dict_env = zipEqual "specUDs2" rhs_dicts call_ds
967 specUDs ty_env dict_env bound_uds `thenSM` \ spec_uds ->
969 returnSM ((spec_f, spec_rhs),
975 %************************************************************************
977 \subsubsection{UsageDetails and suchlike}
979 %************************************************************************
982 type FreeDicts = IdSet
986 dict_binds :: !(Bag (DictVar, CoreExpr, TyVarSet, FreeDicts)),
987 -- Floated dictionary bindings
988 -- The order is important;
989 -- in ds1 `union` ds2, bindings in ds2 can depend on those in ds1
990 -- (Remember, Bags preserve order in GHC.)
991 -- The FreeDicts is the free vars of the RHS
993 calls :: !CallDetails
996 emptyUDs = MkUD { dict_binds = emptyBag, calls = emptyFM }
998 type ProtoUsageDetails = ([CoreBinding], -- Dict bindings
999 [(Id, [Maybe Type], [DictVar])]
1002 ------------------------------------------------------------
1003 type CallDetails = FiniteMap Id CallInfo
1004 type CallInfo = FiniteMap [Maybe Type] -- Nothing => unconstrained type argument
1005 [DictVar] -- Dict args
1006 -- The finite maps eliminate duplicates
1007 -- The list of types and dictionaries is guaranteed to
1008 -- match the type of f
1010 callDetailsToList calls = [ (id,tys,dicts)
1011 | (id,fm) <- fmToList calls,
1012 (tys,dicts) <- fmToList fm
1015 listToCallDetails calls = foldr (unionCalls . singleCall) emptyFM calls
1017 unionCalls :: CallDetails -> CallDetails -> CallDetails
1018 unionCalls c1 c2 = plusFM_C plusFM c1 c2
1020 singleCall (id, tys, dicts) = unitFM id (unitFM tys dicts)
1024 || length spec_tys /= n_tyvars
1025 || length dicts /= n_dicts
1026 = emptyUDs -- Not overloaded
1029 = MkUD {dict_binds = emptyBag,
1030 calls = singleCall (f, spec_tys, dicts)
1033 (tyvars, theta, tau) = splitSigmaTy (idType f)
1034 constrained_tyvars = foldr (unionTyVarSets . tyVarsOfTypes . snd) emptyTyVarSet theta
1035 n_tyvars = length tyvars
1036 n_dicts = length theta
1038 spec_tys = [mk_spec_ty tv ty | (tv, TyArg ty) <- tyvars `zip` args]
1039 dicts = [d | (_, VarArg d) <- theta `zip` (drop n_tyvars args)]
1041 mk_spec_ty tyvar ty | tyvar `elementOfTyVarSet` constrained_tyvars
1046 ------------------------------------------------------------
1047 plusUDs :: UsageDetails -> UsageDetails -> UsageDetails
1048 plusUDs (MkUD {dict_binds = db1, calls = calls1})
1049 (MkUD {dict_binds = db2, calls = calls2})
1050 = MkUD {dict_binds, calls}
1052 dict_binds = db1 `unionBags` db2
1053 calls = calls1 `unionCalls` calls2
1055 plusUDList = foldr plusUDs emptyUDs
1057 mkDB dict rhs = (dict, rhs, db_ftvs, db_fvs)
1059 db_ftvs = tyVarsOfType (idType dict) -- Superset of RHS fvs
1060 db_fvs = dictRhsFVs rhs
1062 addDictBind uds dict rhs = uds { dict_binds = mkDB dict rhs `consBag` dict_binds uds }
1064 dumpAllDictBinds (MkUD {dict_binds = dbs}) binds
1065 = foldrBag add binds dbs
1067 add (dict,rhs,_,_) binds = NonRec dict rhs : binds
1069 dumpUDs :: [CoreBinder]
1070 -> UsageDetails -> CoreExpr
1071 -> (UsageDetails, CoreExpr)
1072 dumpUDs bndrs uds body
1073 = (free_uds, foldr Let body dict_binds)
1075 (free_uds, (dict_binds, _)) = splitUDs bndrs uds
1077 splitUDs :: [CoreBinder]
1079 -> (UsageDetails, -- These don't mention the binders
1080 ProtoUsageDetails) -- These do
1082 splitUDs bndrs uds@(MkUD {dict_binds = orig_dbs,
1083 calls = orig_calls})
1085 = if isEmptyBag dump_dbs && null dump_calls then
1086 -- Common case: binder doesn't affect floats
1090 -- Binders bind some of the fvs of the floats
1091 (MkUD {dict_binds = free_dbs,
1092 calls = listToCallDetails free_calls},
1093 (bagToList dump_dbs, dump_calls)
1097 tyvar_set = mkTyVarSet [tv | TyBinder tv <- bndrs]
1098 id_set = mkIdSet [id | ValBinder id <- bndrs]
1100 (free_dbs, dump_dbs, dump_idset)
1101 = foldlBag dump_db (emptyBag, emptyBag, id_set) orig_dbs
1102 -- Important that it's foldl not foldr;
1103 -- we're accumulating the set of dumped ids in dump_set
1105 -- Filter out any calls that mention things that are being dumped
1106 -- Don't need to worry about the tyvars because the dicts will
1107 -- spot the captured ones; any fully polymorphic arguments will
1108 -- be Nothings in the call details
1109 orig_call_list = callDetailsToList orig_calls
1110 (dump_calls, free_calls) = partition captured orig_call_list
1111 captured (id,tys,dicts) = any (`elementOfIdSet` dump_idset) (id:dicts)
1113 dump_db (free_dbs, dump_dbs, dump_idset) db@(dict, rhs, ftvs, fvs)
1114 | isEmptyIdSet (dump_idset `intersectIdSets` fvs)
1115 && isEmptyTyVarSet (tyvar_set `intersectTyVarSets` ftvs)
1116 = (free_dbs `snocBag` db, dump_dbs, dump_idset)
1118 | otherwise -- Dump it
1119 = (free_dbs, dump_dbs `snocBag` NonRec dict rhs,
1120 dump_idset `addOneToIdSet` dict)
1123 Given a type and value substitution, specUDs creates a specialised copy of
1127 specUDs :: [(TyVar,Type)] -> [(DictVar,DictVar)] -> ProtoUsageDetails -> SpecM UsageDetails
1128 specUDs tv_env_list dict_env_list (dbs, calls)
1129 = specDBs dict_env dbs `thenSM` \ (dict_env', dbs') ->
1130 returnSM (MkUD { dict_binds = dbs',
1131 calls = listToCallDetails (map (inst_call dict_env') calls)
1134 tv_env = mkTyVarEnv tv_env_list
1135 dict_env = mkIdEnv dict_env_list
1137 inst_call dict_env (id, tys, dicts) = (id, map inst_maybe_ty tys,
1138 map (lookupId dict_env) dicts)
1140 inst_maybe_ty Nothing = Nothing
1141 inst_maybe_ty (Just ty) = Just (instantiateTy tv_env ty)
1144 = returnSM (dict_env, emptyBag)
1145 specDBs dict_env (NonRec dict rhs : dbs)
1146 = newIdSM dict (instantiateTy tv_env (idType dict)) `thenSM` \ dict' ->
1148 dict_env' = addOneToIdEnv dict_env dict dict'
1149 rhs' = instantiateDictRhs tv_env dict_env rhs
1151 specDBs dict_env' dbs `thenSM` \ (dict_env'', dbs') ->
1152 returnSM ( dict_env'', mkDB dict' rhs' `consBag` dbs' )
1155 %************************************************************************
1157 \subsubsection{Boring helper functions}
1159 %************************************************************************
1162 lookupId:: IdEnv Id -> Id -> Id
1163 lookupId env id = case lookupIdEnv env id of
1167 instantiateDictRhs :: TyVarEnv Type -> IdEnv Id -> CoreExpr -> CoreExpr
1168 -- Cheapo function for simple RHSs
1169 instantiateDictRhs ty_env id_env rhs
1172 go_arg (VarArg a) = VarArg (lookupId id_env a)
1173 go_arg (TyArg t) = TyArg (instantiateTy ty_env t)
1175 go (App e1 arg) = App (go e1) (go_arg arg)
1176 go (Var v) = Var (lookupId id_env v)
1178 go (Con con args) = Con con (map go_arg args)
1179 go (Note n e) = Note (go_note n) (go e)
1180 go (Case e alts) = Case (go e) alts -- See comment below re alts
1181 go other = pprPanic "instantiateDictRhs" (ppr rhs)
1183 go_note (Coerce t1 t2) = Coerce (instantiateTy ty_env t1) (instantiateTy ty_env t2)
1186 dictRhsFVs :: CoreExpr -> IdSet
1187 -- Cheapo function for simple RHSs
1191 go (App e1 (VarArg a)) = go e1 `addOneToIdSet` a
1192 go (App e1 (TyArg t)) = go e1
1193 go (Var v) = unitIdSet v
1194 go (Lit l) = emptyIdSet
1195 go (Con _ args) = mkIdSet [id | VarArg id <- args]
1196 go (Note _ e) = go e
1198 go (Case e _) = go e -- Claim: no free dictionaries in the alternatives
1199 -- These case expressions are of the form
1200 -- case d of { D a b c -> b }
1202 go other = pprPanic "dictRhsFVs" (ppr e)
1205 addIdSpecialisations id spec_stuff
1206 = (if not (null errs) then
1207 pprTrace "Duplicate specialisations" (vcat (map ppr errs))
1210 setIdSpecialisation id new_spec_env
1212 (new_spec_env, errs) = foldr add (getIdSpecialisation id, []) spec_stuff
1214 add (tyvars, tys, template) (spec_env, errs)
1215 = case addToSpecEnv True spec_env tyvars tys template of
1216 Succeeded spec_env' -> (spec_env', errs)
1217 Failed err -> (spec_env, err:errs)
1219 -- Given an Id, isSpecVars returns all its specialisations.
1220 -- We extract these from its SpecEnv.
1221 -- This is used by the occurrence analyser and free-var finder;
1222 -- we regard an Id's specialisations as free in the Id's definition.
1224 idSpecVars :: Id -> [Id]
1226 = map get_spec (specEnvValues (getIdSpecialisation id))
1228 -- get_spec is another cheapo function like dictRhsFVs
1229 -- It knows what these specialisation temlates look like,
1230 -- and just goes for the jugular
1231 get_spec (App f _) = get_spec f
1232 get_spec (Lam _ b) = get_spec b
1233 get_spec (Var v) = v
1235 ----------------------------------------
1236 type SpecM a = UniqSM a
1240 getUniqSM = getUnique
1244 mapAndCombineSM f [] = returnSM ([], emptyUDs)
1245 mapAndCombineSM f (x:xs) = f x `thenSM` \ (y, uds1) ->
1246 mapAndCombineSM f xs `thenSM` \ (ys, uds2) ->
1247 returnSM (y:ys, uds1 `plusUDs` uds2)
1249 newIdSM old_id new_ty
1250 = getUnique `thenSM` \ uniq ->
1251 returnSM (mkUserLocal (getOccName old_id)