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, mkTemplateLocals,
17 getIdSpecialisation, setIdSpecialisation, isSpecPragmaId,
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 )
29 import TyVar ( TyVar, mkTyVar,
30 TyVarSet, mkTyVarSet, isEmptyTyVarSet, intersectTyVarSets,
31 elementOfTyVarSet, unionTyVarSets, emptyTyVarSet,
33 TyVarEnv, mkTyVarEnv, delFromTyVarEnv
35 import Kind ( mkBoxedTypeKind )
37 import PprCore () -- Instances
38 import Name ( NamedThing(..), getSrcLoc, mkSysLocalName )
39 import SrcLoc ( noSrcLoc )
40 import SpecEnv ( addToSpecEnv, lookupSpecEnv, specEnvValues )
42 import UniqSupply ( UniqSupply,
43 UniqSM, initUs, thenUs, returnUs, getUnique, mapUs
45 import Unique ( mkAlphaTyVarUnique )
47 import Maybes ( MaybeErr(..), maybeToBool )
49 import List ( partition )
50 import Util ( zipEqual )
57 %************************************************************************
59 \subsection[notes-Specialise]{Implementation notes [SLPJ, Aug 18 1993]}
61 %************************************************************************
63 These notes describe how we implement specialisation to eliminate
66 The specialisation pass works on Core
67 syntax, complete with all the explicit dictionary application,
68 abstraction and construction as added by the type checker. The
69 existing type checker remains largely as it is.
71 One important thought: the {\em types} passed to an overloaded
72 function, and the {\em dictionaries} passed are mutually redundant.
73 If the same function is applied to the same type(s) then it is sure to
74 be applied to the same dictionary(s)---or rather to the same {\em
75 values}. (The arguments might look different but they will evaluate
78 Second important thought: we know that we can make progress by
79 treating dictionary arguments as static and worth specialising on. So
80 we can do without binding-time analysis, and instead specialise on
81 dictionary arguments and no others.
90 and suppose f is overloaded.
92 STEP 1: CALL-INSTANCE COLLECTION
94 We traverse <body>, accumulating all applications of f to types and
97 (Might there be partial applications, to just some of its types and
98 dictionaries? In principle yes, but in practice the type checker only
99 builds applications of f to all its types and dictionaries, so partial
100 applications could only arise as a result of transformation, and even
101 then I think it's unlikely. In any case, we simply don't accumulate such
102 partial applications.)
104 There's a choice of whether to collect details of all *polymorphic* functions
105 or simply all *overloaded* ones. How to sort this out?
106 Pass in a predicate on the function to say if it is "interesting"?
107 This is dependent on the user flags: SpecialiseOverloaded
113 So now we have a collection of calls to f:
117 Notice that f may take several type arguments. To avoid ambiguity, we
118 say that f is called at type t1/t2 and t3/t4.
120 We take equivalence classes using equality of the *types* (ignoring
121 the dictionary args, which as mentioned previously are redundant).
123 STEP 3: SPECIALISATION
125 For each equivalence class, choose a representative (f t1 t2 d1 d2),
126 and create a local instance of f, defined thus:
128 f@t1/t2 = <f_rhs> t1 t2 d1 d2
130 f_rhs presumably has some big lambdas and dictionary lambdas, so lots
131 of simplification will now result. However we don't actually *do* that
132 simplification. Rather, we leave it for the simplifier to do. If we
133 *did* do it, though, we'd get more call instances from the specialised
134 RHS. We can work out what they are by instantiating the call-instance
135 set from f's RHS with the types t1, t2.
137 Add this new id to f's IdInfo, to record that f has a specialised version.
139 Before doing any of this, check that f's IdInfo doesn't already
140 tell us about an existing instance of f at the required type/s.
141 (This might happen if specialisation was applied more than once, or
142 it might arise from user SPECIALIZE pragmas.)
146 Wait a minute! What if f is recursive? Then we can't just plug in
147 its right-hand side, can we?
149 But it's ok. The type checker *always* creates non-recursive definitions
150 for overloaded recursive functions. For example:
152 f x = f (x+x) -- Yes I know its silly
156 f a (d::Num a) = let p = +.sel a d
158 letrec fl (y::a) = fl (p y y)
162 We still have recusion for non-overloaded functions which we
163 speciailise, but the recursive call should get specialised to the
164 same recursive version.
170 All this is crystal clear when the function is applied to *constant
171 types*; that is, types which have no type variables inside. But what if
172 it is applied to non-constant types? Suppose we find a call of f at type
173 t1/t2. There are two possibilities:
175 (a) The free type variables of t1, t2 are in scope at the definition point
176 of f. In this case there's no problem, we proceed just as before. A common
177 example is as follows. Here's the Haskell:
182 After typechecking we have
184 g a (d::Num a) (y::a) = let f b (d'::Num b) (x::b) = +.sel b d' x x
185 in +.sel a d (f a d y) (f a d y)
187 Notice that the call to f is at type type "a"; a non-constant type.
188 Both calls to f are at the same type, so we can specialise to give:
190 g a (d::Num a) (y::a) = let f@a (x::a) = +.sel a d x x
191 in +.sel a d (f@a y) (f@a y)
194 (b) The other case is when the type variables in the instance types
195 are *not* in scope at the definition point of f. The example we are
196 working with above is a good case. There are two instances of (+.sel a d),
197 but "a" is not in scope at the definition of +.sel. Can we do anything?
198 Yes, we can "common them up", a sort of limited common sub-expression deal.
201 g a (d::Num a) (y::a) = let +.sel@a = +.sel a d
202 f@a (x::a) = +.sel@a x x
203 in +.sel@a (f@a y) (f@a y)
205 This can save work, and can't be spotted by the type checker, because
206 the two instances of +.sel weren't originally at the same type.
210 * There are quite a few variations here. For example, the defn of
211 +.sel could be floated ouside the \y, to attempt to gain laziness.
212 It certainly mustn't be floated outside the \d because the d has to
215 * We don't want to inline f_rhs in this case, because
216 that will duplicate code. Just commoning up the call is the point.
218 * Nothing gets added to +.sel's IdInfo.
220 * Don't bother unless the equivalence class has more than one item!
222 Not clear whether this is all worth it. It is of course OK to
223 simply discard call-instances when passing a big lambda.
225 Polymorphism 2 -- Overloading
227 Consider a function whose most general type is
229 f :: forall a b. Ord a => [a] -> b -> b
231 There is really no point in making a version of g at Int/Int and another
232 at Int/Bool, because it's only instancing the type variable "a" which
233 buys us any efficiency. Since g is completely polymorphic in b there
234 ain't much point in making separate versions of g for the different
237 That suggests that we should identify which of g's type variables
238 are constrained (like "a") and which are unconstrained (like "b").
239 Then when taking equivalence classes in STEP 2, we ignore the type args
240 corresponding to unconstrained type variable. In STEP 3 we make
241 polymorphic versions. Thus:
243 f@t1/ = /\b -> <f_rhs> t1 b d1 d2
252 f a (d::Num a) = let g = ...
254 ...(let d1::Ord a = Num.Ord.sel a d in g a d1)...
256 Here, g is only called at one type, but the dictionary isn't in scope at the
257 definition point for g. Usually the type checker would build a
258 definition for d1 which enclosed g, but the transformation system
259 might have moved d1's defn inward. Solution: float dictionary bindings
260 outwards along with call instances.
264 f x = let g p q = p==q
270 Before specialisation, leaving out type abstractions we have
272 f df x = let g :: Eq a => a -> a -> Bool
274 h :: Num a => a -> a -> (a, Bool)
275 h dh r s = let deq = eqFromNum dh
276 in (+ dh r s, g deq r s)
280 After specialising h we get a specialised version of h, like this:
282 h' r s = let deq = eqFromNum df
283 in (+ df r s, g deq r s)
285 But we can't naively make an instance for g from this, because deq is not in scope
286 at the defn of g. Instead, we have to float out the (new) defn of deq
287 to widen its scope. Notice that this floating can't be done in advance -- it only
288 shows up when specialisation is done.
290 User SPECIALIZE pragmas
291 ~~~~~~~~~~~~~~~~~~~~~~~
292 Specialisation pragmas can be digested by the type checker, and implemented
293 by adding extra definitions along with that of f, in the same way as before
295 f@t1/t2 = <f_rhs> t1 t2 d1 d2
297 Indeed the pragmas *have* to be dealt with by the type checker, because
298 only it knows how to build the dictionaries d1 and d2! For example
300 g :: Ord a => [a] -> [a]
301 {-# SPECIALIZE f :: [Tree Int] -> [Tree Int] #-}
303 Here, the specialised version of g is an application of g's rhs to the
304 Ord dictionary for (Tree Int), which only the type checker can conjure
305 up. There might not even *be* one, if (Tree Int) is not an instance of
306 Ord! (All the other specialision has suitable dictionaries to hand
309 Problem. The type checker doesn't have to hand a convenient <f_rhs>, because
310 it is buried in a complex (as-yet-un-desugared) binding group.
313 f@t1/t2 = f* t1 t2 d1 d2
315 where f* is the Id f with an IdInfo which says "inline me regardless!".
316 Indeed all the specialisation could be done in this way.
317 That in turn means that the simplifier has to be prepared to inline absolutely
318 any in-scope let-bound thing.
321 Again, the pragma should permit polymorphism in unconstrained variables:
323 h :: Ord a => [a] -> b -> b
324 {-# SPECIALIZE h :: [Int] -> b -> b #-}
326 We *insist* that all overloaded type variables are specialised to ground types,
327 (and hence there can be no context inside a SPECIALIZE pragma).
328 We *permit* unconstrained type variables to be specialised to
330 - or left as a polymorphic type variable
331 but nothing in between. So
333 {-# SPECIALIZE h :: [Int] -> [c] -> [c] #-}
335 is *illegal*. (It can be handled, but it adds complication, and gains the
339 SPECIALISING INSTANCE DECLARATIONS
340 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
343 instance Foo a => Foo [a] where
345 {-# SPECIALIZE instance Foo [Int] #-}
347 The original instance decl creates a dictionary-function
350 dfun.Foo.List :: forall a. Foo a -> Foo [a]
352 The SPECIALIZE pragma just makes a specialised copy, just as for
353 ordinary function definitions:
355 dfun.Foo.List@Int :: Foo [Int]
356 dfun.Foo.List@Int = dfun.Foo.List Int dFooInt
358 The information about what instance of the dfun exist gets added to
359 the dfun's IdInfo in the same way as a user-defined function too.
362 Automatic instance decl specialisation?
363 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
364 Can instance decls be specialised automatically? It's tricky.
365 We could collect call-instance information for each dfun, but
366 then when we specialised their bodies we'd get new call-instances
367 for ordinary functions; and when we specialised their bodies, we might get
368 new call-instances of the dfuns, and so on. This all arises because of
369 the unrestricted mutual recursion between instance decls and value decls.
371 Still, there's no actual problem; it just means that we may not do all
372 the specialisation we could theoretically do.
374 Furthermore, instance decls are usually exported and used non-locally,
375 so we'll want to compile enough to get those specialisations done.
377 Lastly, there's no such thing as a local instance decl, so we can
378 survive solely by spitting out *usage* information, and then reading that
379 back in as a pragma when next compiling the file. So for now,
380 we only specialise instance decls in response to pragmas.
383 SPITTING OUT USAGE INFORMATION
384 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
386 To spit out usage information we need to traverse the code collecting
387 call-instance information for all imported (non-prelude?) functions
388 and data types. Then we equivalence-class it and spit it out.
390 This is done at the top-level when all the call instances which escape
391 must be for imported functions and data types.
393 *** Not currently done ***
396 Partial specialisation by pragmas
397 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
398 What about partial specialisation:
400 k :: (Ord a, Eq b) => [a] -> b -> b -> [a]
401 {-# SPECIALIZE k :: Eq b => [Int] -> b -> b -> [a] #-}
405 {-# SPECIALIZE k :: Eq b => [Int] -> [b] -> [b] -> [a] #-}
407 Seems quite reasonable. Similar things could be done with instance decls:
409 instance (Foo a, Foo b) => Foo (a,b) where
411 {-# SPECIALIZE instance Foo a => Foo (a,Int) #-}
412 {-# SPECIALIZE instance Foo b => Foo (Int,b) #-}
414 Ho hum. Things are complex enough without this. I pass.
417 Requirements for the simplifer
418 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
419 The simplifier has to be able to take advantage of the specialisation.
421 * When the simplifier finds an application of a polymorphic f, it looks in
422 f's IdInfo in case there is a suitable instance to call instead. This converts
424 f t1 t2 d1 d2 ===> f_t1_t2
426 Note that the dictionaries get eaten up too!
428 * Dictionary selection operations on constant dictionaries must be
431 +.sel Int d ===> +Int
433 The obvious way to do this is in the same way as other specialised
434 calls: +.sel has inside it some IdInfo which tells that if it's applied
435 to the type Int then it should eat a dictionary and transform to +Int.
437 In short, dictionary selectors need IdInfo inside them for constant
440 * Exactly the same applies if a superclass dictionary is being
443 Eq.sel Int d ===> dEqInt
445 * Something similar applies to dictionary construction too. Suppose
446 dfun.Eq.List is the function taking a dictionary for (Eq a) to
447 one for (Eq [a]). Then we want
449 dfun.Eq.List Int d ===> dEq.List_Int
451 Where does the Eq [Int] dictionary come from? It is built in
452 response to a SPECIALIZE pragma on the Eq [a] instance decl.
454 In short, dfun Ids need IdInfo with a specialisation for each
455 constant instance of their instance declaration.
457 All this uses a single mechanism: the SpecEnv inside an Id
460 What does the specialisation IdInfo look like?
461 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
463 The SpecEnv of an Id maps a list of types (the template) to an expression
467 For example, if f has this SpecInfo:
469 [Int, a] -> \d:Ord Int. f' a
471 it means that we can replace the call
473 f Int t ===> (\d. f' t)
475 This chucks one dictionary away and proceeds with the
476 specialised version of f, namely f'.
479 What can't be done this way?
480 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
481 There is no way, post-typechecker, to get a dictionary for (say)
482 Eq a from a dictionary for Eq [a]. So if we find
486 we can't transform to
491 eqList :: (a->a->Bool) -> [a] -> [a] -> Bool
493 Of course, we currently have no way to automatically derive
494 eqList, nor to connect it to the Eq [a] instance decl, but you
495 can imagine that it might somehow be possible. Taking advantage
496 of this is permanently ruled out.
498 Still, this is no great hardship, because we intend to eliminate
499 overloading altogether anyway!
503 A note about non-tyvar dictionaries
504 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
505 Some Ids have types like
507 forall a,b,c. Eq a -> Ord [a] -> tau
509 This seems curious at first, because we usually only have dictionary
510 args whose types are of the form (C a) where a is a type variable.
511 But this doesn't hold for the functions arising from instance decls,
512 which sometimes get arguements with types of form (C (T a)) for some
515 Should we specialise wrt this compound-type dictionary? We used to say
517 "This is a heuristic judgement, as indeed is the fact that we
518 specialise wrt only dictionaries. We choose *not* to specialise
519 wrt compound dictionaries because at the moment the only place
520 they show up is in instance decls, where they are simply plugged
521 into a returned dictionary. So nothing is gained by specialising
524 But it is simpler and more uniform to specialise wrt these dicts too;
525 and in future GHC is likely to support full fledged type signatures
527 f ;: Eq [(a,b)] => ...
530 %************************************************************************
532 \subsubsection{The new specialiser}
534 %************************************************************************
536 Our basic game plan is this. For let(rec) bound function
537 f :: (C a, D c) => (a,b,c,d) -> Bool
539 * Find any specialised calls of f, (f ts ds), where
540 ts are the type arguments t1 .. t4, and
541 ds are the dictionary arguments d1 .. d2.
543 * Add a new definition for f1 (say):
545 f1 = /\ b d -> (..body of f..) t1 b t3 d d1 d2
547 Note that we abstract over the unconstrained type arguments.
551 [t1,b,t3,d] |-> \d1 d2 -> f1 b d
553 to the specialisations of f. This will be used by the
554 simplifier to replace calls
555 (f t1 t2 t3 t4) da db
557 (\d1 d1 -> f1 t2 t4) da db
559 All the stuff about how many dictionaries to discard, and what types
560 to apply the specialised function to, are handled by the fact that the
561 SpecEnv contains a template for the result of the specialisation.
563 We don't build *partial* specialisations for f. For example:
565 f :: Eq a => a -> a -> Bool
566 {-# SPECIALISE f :: (Eq b, Eq c) => (b,c) -> (b,c) -> Bool #-}
568 Here, little is gained by making a specialised copy of f.
569 There's a distinct danger that the specialised version would
570 first build a dictionary for (Eq b, Eq c), and then select the (==)
571 method from it! Even if it didn't, not a great deal is saved.
573 We do, however, generate polymorphic, but not overloaded, specialisations:
575 f :: Eq a => [a] -> b -> b -> b
576 {#- SPECIALISE f :: [Int] -> b -> b -> b #-}
578 Hence, the invariant is this:
580 *** no specialised version is overloaded ***
583 %************************************************************************
585 \subsubsection{The exported function}
587 %************************************************************************
590 specProgram :: UniqSupply -> [CoreBinding] -> [CoreBinding]
592 = initSM us (go binds `thenSM` \ (binds', uds') ->
593 returnSM (dumpAllDictBinds uds' binds')
596 go [] = returnSM ([], emptyUDs)
597 go (bind:binds) = go binds `thenSM` \ (binds', uds) ->
598 specBind bind uds `thenSM` \ (bind', uds') ->
599 returnSM (bind' ++ binds', uds')
602 %************************************************************************
604 \subsubsection{@specExpr@: the main function}
606 %************************************************************************
609 specExpr :: CoreExpr -> SpecM (CoreExpr, UsageDetails)
611 ---------------- First the easy cases --------------------
612 specExpr e@(Var _) = returnSM (e, emptyUDs)
613 specExpr e@(Lit _) = returnSM (e, emptyUDs)
614 specExpr e@(Con _ _) = returnSM (e, emptyUDs)
615 specExpr e@(Prim _ _) = returnSM (e, emptyUDs)
617 specExpr (Note note body)
618 = specExpr body `thenSM` \ (body', uds) ->
619 returnSM (Note note body', uds)
622 ---------------- Applications might generate a call instance --------------------
623 specExpr e@(App fun arg)
626 go (App fun arg) args = go fun (arg:args)
627 go (Var f) args = returnSM (e, mkCallUDs f args)
628 go other args = specExpr other `thenSM` \ (e', uds) ->
629 returnSM (foldl App e' args, uds)
631 ---------------- Lambda/case require dumping of usage details --------------------
633 = specExpr body `thenSM` \ (body', uds) ->
635 (filtered_uds, body'') = dumpUDs bndrs uds body'
637 returnSM (foldr Lam body'' bndrs, filtered_uds)
639 (bndrs, body) = go [] e
641 -- More efficient to collect a group of binders together all at once
642 go bndrs (Lam bndr e) = go (bndr:bndrs) e
643 go bndrs e = (reverse bndrs, e)
646 specExpr (Case scrut alts)
647 = specExpr scrut `thenSM` \ (scrut', uds_scrut) ->
648 spec_alts alts `thenSM` \ (alts', uds_alts) ->
649 returnSM (Case scrut' alts', uds_scrut `plusUDs` uds_alts)
651 spec_alts (AlgAlts alts deflt)
652 = mapAndCombineSM spec_alg_alt alts `thenSM` \ (alts', uds1) ->
653 spec_deflt deflt `thenSM` \ (deflt', uds2) ->
654 returnSM (AlgAlts alts' deflt', uds1 `plusUDs` uds2)
656 spec_alts (PrimAlts alts deflt)
657 = mapAndCombineSM spec_prim_alt alts `thenSM` \ (alts', uds1) ->
658 spec_deflt deflt `thenSM` \ (deflt', uds2) ->
659 returnSM (PrimAlts alts' deflt', uds1 `plusUDs` uds2)
661 spec_alg_alt (con, args, rhs)
662 = specExpr rhs `thenSM` \ (rhs', uds) ->
664 (uds', rhs'') = dumpUDs (map ValBinder args) uds rhs'
666 returnSM ((con, args, rhs''), uds')
668 spec_prim_alt (lit, rhs)
669 = specExpr rhs `thenSM` \ (rhs', uds) ->
670 returnSM ((lit, rhs'), uds)
672 spec_deflt NoDefault = returnSM (NoDefault, emptyUDs)
673 spec_deflt (BindDefault arg rhs)
674 = specExpr rhs `thenSM` \ (rhs', uds) ->
676 (uds', rhs'') = dumpUDs [ValBinder arg] uds rhs'
678 returnSM (BindDefault arg rhs'', uds')
680 ---------------- Finally, let is the interesting case --------------------
681 specExpr (Let bind body)
682 = -- Deal with the body
683 specExpr body `thenSM` \ (body', body_uds) ->
685 -- Deal with the bindings
686 specBind bind body_uds `thenSM` \ (binds', uds) ->
689 returnSM (foldr Let body' binds', uds)
692 %************************************************************************
694 \subsubsection{Dealing with a binding}
696 %************************************************************************
699 specBind :: CoreBinding
700 -> UsageDetails -- Info on how the scope of the binding
701 -> SpecM ([CoreBinding], -- New bindings
702 UsageDetails) -- And info to pass upstream
704 specBind (NonRec bndr rhs) body_uds
705 | isDictTy (idType bndr)
706 = -- It's a dictionary binding
707 -- Pick it up and float it outwards.
708 specExpr rhs `thenSM` \ (rhs', rhs_uds) ->
710 all_uds = rhs_uds `plusUDs` addDictBind body_uds bndr rhs'
712 returnSM ([], all_uds)
714 | isSpecPragmaId bndr
715 = specExpr rhs `thenSM` \ (rhs', rhs_uds) ->
716 returnSM ([], rhs_uds `plusUDs` body_uds)
719 = -- Deal with the RHS, specialising it according
720 -- to the calls found in the body
721 specDefn (calls body_uds) (bndr,rhs) `thenSM` \ ((bndr',rhs'), spec_defns, spec_uds) ->
723 (all_uds, (dict_binds, dump_calls))
724 = splitUDs [ValBinder bndr]
725 (body_uds `plusUDs` spec_uds)
726 -- It's important that the `plusUDs` is this way round,
727 -- because body_uds may bind dictionaries that are
728 -- used in the calls passed to specDefn. So the
729 -- dictionary bindings in spec_uds may mention
730 -- dictionaries bound in body_uds.
732 -- If we make specialisations then we Rec the whole lot together
733 -- If not, leave it as a NonRec
734 new_bind | null spec_defns = NonRec bndr' rhs'
735 | otherwise = Rec ((bndr',rhs'):spec_defns)
737 returnSM ( new_bind : mkDictBinds dict_binds, all_uds )
739 specBind (Rec pairs) body_uds
740 = mapSM (specDefn (calls body_uds)) pairs `thenSM` \ stuff ->
742 (pairs', spec_defns_s, spec_uds_s) = unzip3 stuff
743 spec_defns = concat spec_defns_s
744 spec_uds = plusUDList spec_uds_s
746 (all_uds, (dict_binds, dump_calls))
747 = splitUDs (map (ValBinder . fst) pairs)
748 (body_uds `plusUDs` spec_uds)
749 -- See notes for non-rec case
751 new_bind = Rec (spec_defns ++ pairs')
753 returnSM ( new_bind : mkDictBinds dict_binds, all_uds )
755 specDefn :: CallDetails -- Info on how it is used in its scope
756 -> (Id, CoreExpr) -- The thing being bound and its un-processed RHS
757 -> SpecM ((Id, CoreExpr), -- The thing and its processed RHS
758 -- the Id may now have specialisations attached
759 [(Id,CoreExpr)], -- Extra, specialised bindings
760 UsageDetails -- Stuff to fling upwards from the RHS and its
761 ) -- specialised versions
763 specDefn calls (fn, rhs)
764 -- The first case is the interesting one
765 | n_tyvars == length rhs_tyvars -- Rhs of fn's defn has right number of big lambdas
766 && n_dicts <= length rhs_bndrs -- and enough dict args
767 && not (null calls_for_me) -- And there are some calls to specialise
768 = -- Specialise the body of the function
769 specExpr body `thenSM` \ (body', body_uds) ->
771 (float_uds, bound_uds@(dict_binds,_)) = splitUDs rhs_bndrs body_uds
774 -- Make a specialised version for each call in calls_for_me
775 mapSM (spec_call bound_uds) calls_for_me `thenSM` \ stuff ->
777 (spec_defns, spec_uds, spec_env_stuff) = unzip3 stuff
779 fn' = addIdSpecialisations fn spec_env_stuff
780 rhs' = foldr Lam (mkDictLets dict_binds body') rhs_bndrs
782 returnSM ((fn',rhs'),
784 float_uds `plusUDs` plusUDList spec_uds)
786 | otherwise -- No calls or RHS doesn't fit our preconceptions
787 = specExpr rhs `thenSM` \ (rhs', rhs_uds) ->
788 returnSM ((fn, rhs'), [], rhs_uds)
792 (tyvars, theta, tau) = splitSigmaTy fn_type
793 n_tyvars = length tyvars
794 n_dicts = length theta
795 mk_spec_tys call_ts = zipWith mk_spec_ty call_ts tyVarTemplates
797 mk_spec_ty (Just ty) _ = ty
798 mk_spec_ty Nothing tyvar = mkTyVarTy tyvar
800 (rhs_tyvars, rhs_ids, rhs_body) = collectBinders rhs
801 rhs_dicts = take n_dicts rhs_ids
802 rhs_bndrs = map TyBinder rhs_tyvars ++ map ValBinder rhs_dicts
803 body = mkValLam (drop n_dicts rhs_ids) rhs_body
804 -- Glue back on the non-dict lambdas
806 calls_for_me = case lookupFM calls fn of
808 Just cs -> fmToList cs
810 ----------------------------------------------------------
811 -- Specialise to one particular call pattern
812 spec_call :: ProtoUsageDetails -- From the original body, captured by
813 -- the dictionary lambdas
814 -> ([Maybe Type], [DictVar]) -- Call instance
815 -> SpecM ((Id,CoreExpr), -- Specialised definition
816 UsageDetails, -- Usage details from specialised body
817 ([TyVar], [Type], CoreExpr)) -- Info for the Id's SpecEnv
818 spec_call bound_uds (call_ts, call_ds)
819 = ASSERT( length call_ts == n_tyvars && length call_ds == n_dicts )
820 -- Calls are only recorded for properly-saturated applications
822 -- Supppose the call is for f [Just t1, Nothing, Just t3, Nothing] [d1, d2]
824 -- Construct the new binding
825 -- f1 = /\ b d -> (..rhs of f..) t1 b t3 d d1 d2
826 -- and the type of this binder
828 spec_tyvars = [tyvar | (tyvar, Nothing) <- tyVarTemplates `zip` call_ts]
829 spec_tys = mk_spec_tys call_ts
830 spec_rhs = mkTyLam spec_tyvars $
831 mkGenApp rhs (map TyArg spec_tys ++ map VarArg call_ds)
832 spec_id_ty = mkForAllTys spec_tyvars (instantiateTy ty_env tau)
833 ty_env = mkTyVarEnv (zipEqual "spec_call" tyvars spec_tys)
836 newIdSM fn spec_id_ty `thenSM` \ spec_f ->
839 -- Construct the stuff for f's spec env
840 -- [b,d] [t1,b,t3,d] |-> \d1 d2 -> f1 b d
841 -- The only awkward bit is that d1,d2 might well be global
842 -- dictionaries, so it's tidier to make new local variables
843 -- for the lambdas in the RHS, rather than lambda-bind the
844 -- dictionaries themselves.
846 -- In fact we use the standard template locals, so that the
847 -- they don't need to be "tidied" before putting in interface files
849 arg_ds = mkTemplateLocals (map idType call_ds)
850 spec_env_rhs = mkValLam arg_ds $
851 mkTyApp (Var spec_f) $
852 map mkTyVarTy spec_tyvars
853 spec_env_info = (spec_tyvars, spec_tys, spec_env_rhs)
856 -- Specialise the UDs from f's RHS
858 -- Only the overloaded tyvars should be free in the uds
859 ty_env = [ (rhs_tyvar,ty)
860 | (rhs_tyvar, Just ty) <- zipEqual "specUDs1" rhs_tyvars call_ts
862 dict_env = zipEqual "specUDs2" rhs_dicts call_ds
864 specUDs ty_env dict_env bound_uds `thenSM` \ spec_uds ->
866 returnSM ((spec_f, spec_rhs),
872 %************************************************************************
874 \subsubsection{UsageDetails and suchlike}
876 %************************************************************************
879 type FreeDicts = IdSet
883 dict_binds :: !(Bag DictBind),
884 -- Floated dictionary bindings
885 -- The order is important;
886 -- in ds1 `union` ds2, bindings in ds2 can depend on those in ds1
887 -- (Remember, Bags preserve order in GHC.)
888 -- The FreeDicts is the free vars of the RHS
890 calls :: !CallDetails
893 type DictBind = (DictVar, CoreExpr, TyVarSet, FreeDicts)
894 -- The FreeDicts are the free dictionaries (only)
895 -- of the RHS of the dictionary bindings
896 -- Similarly the TyVarSet
898 emptyUDs = MkUD { dict_binds = emptyBag, calls = emptyFM }
900 type ProtoUsageDetails = ([DictBind],
901 [(Id, [Maybe Type], [DictVar])]
904 ------------------------------------------------------------
905 type CallDetails = FiniteMap Id CallInfo
906 type CallInfo = FiniteMap [Maybe Type] -- Nothing => unconstrained type argument
907 [DictVar] -- Dict args
908 -- The finite maps eliminate duplicates
909 -- The list of types and dictionaries is guaranteed to
910 -- match the type of f
912 callDetailsToList calls = [ (id,tys,dicts)
913 | (id,fm) <- fmToList calls,
914 (tys,dicts) <- fmToList fm
917 listToCallDetails calls = foldr (unionCalls . singleCall) emptyFM calls
919 unionCalls :: CallDetails -> CallDetails -> CallDetails
920 unionCalls c1 c2 = plusFM_C plusFM c1 c2
922 singleCall (id, tys, dicts) = unitFM id (unitFM tys dicts)
926 || length spec_tys /= n_tyvars
927 || length dicts /= n_dicts
928 = emptyUDs -- Not overloaded
931 = MkUD {dict_binds = emptyBag,
932 calls = singleCall (f, spec_tys, dicts)
935 (tyvars, theta, tau) = splitSigmaTy (idType f)
936 constrained_tyvars = foldr (unionTyVarSets . tyVarsOfTypes . snd) emptyTyVarSet theta
937 n_tyvars = length tyvars
938 n_dicts = length theta
940 spec_tys = [mk_spec_ty tv ty | (tv, TyArg ty) <- tyvars `zip` args]
941 dicts = [d | (_, VarArg d) <- theta `zip` (drop n_tyvars args)]
943 mk_spec_ty tyvar ty | tyvar `elementOfTyVarSet` constrained_tyvars
948 ------------------------------------------------------------
949 plusUDs :: UsageDetails -> UsageDetails -> UsageDetails
950 plusUDs (MkUD {dict_binds = db1, calls = calls1})
951 (MkUD {dict_binds = db2, calls = calls2})
952 = MkUD {dict_binds, calls}
954 dict_binds = db1 `unionBags` db2
955 calls = calls1 `unionCalls` calls2
957 plusUDList = foldr plusUDs emptyUDs
959 mkDB dict rhs = (dict, rhs, db_ftvs, db_fvs)
961 db_ftvs = tyVarsOfType (idType dict) -- Superset of RHS fvs
962 db_fvs = dictRhsFVs rhs
964 addDictBind uds dict rhs = uds { dict_binds = mkDB dict rhs `consBag` dict_binds uds }
966 dumpAllDictBinds (MkUD {dict_binds = dbs}) binds
967 = foldrBag add binds dbs
969 add (dict,rhs,_,_) binds = NonRec dict rhs : binds
971 mkDictBinds :: [DictBind] -> [CoreBinding]
972 mkDictBinds = map (\(d,r,_,_) -> NonRec d r)
974 mkDictLets :: [DictBind] -> CoreExpr -> CoreExpr
975 mkDictLets dbs body = foldr mk body dbs
977 mk (d,r,_,_) e = Let (NonRec d r) e
979 dumpUDs :: [CoreBinder]
980 -> UsageDetails -> CoreExpr
981 -> (UsageDetails, CoreExpr)
982 dumpUDs bndrs uds body
983 = (free_uds, mkDictLets dict_binds body)
985 (free_uds, (dict_binds, _)) = splitUDs bndrs uds
987 splitUDs :: [CoreBinder]
989 -> (UsageDetails, -- These don't mention the binders
990 ProtoUsageDetails) -- These do
992 splitUDs bndrs uds@(MkUD {dict_binds = orig_dbs,
995 = if isEmptyBag dump_dbs && null dump_calls then
996 -- Common case: binder doesn't affect floats
1000 -- Binders bind some of the fvs of the floats
1001 (MkUD {dict_binds = free_dbs,
1002 calls = listToCallDetails free_calls},
1003 (bagToList dump_dbs, dump_calls)
1007 tyvar_set = mkTyVarSet [tv | TyBinder tv <- bndrs]
1008 id_set = mkIdSet [id | ValBinder id <- bndrs]
1010 (free_dbs, dump_dbs, dump_idset)
1011 = foldlBag dump_db (emptyBag, emptyBag, id_set) orig_dbs
1012 -- Important that it's foldl not foldr;
1013 -- we're accumulating the set of dumped ids in dump_set
1015 -- Filter out any calls that mention things that are being dumped
1016 -- Don't need to worry about the tyvars because the dicts will
1017 -- spot the captured ones; any fully polymorphic arguments will
1018 -- be Nothings in the call details
1019 orig_call_list = callDetailsToList orig_calls
1020 (dump_calls, free_calls) = partition captured orig_call_list
1021 captured (id,tys,dicts) = any (`elementOfIdSet` dump_idset) (id:dicts)
1023 dump_db (free_dbs, dump_dbs, dump_idset) db@(dict, rhs, ftvs, fvs)
1024 | isEmptyIdSet (dump_idset `intersectIdSets` fvs)
1025 && isEmptyTyVarSet (tyvar_set `intersectTyVarSets` ftvs)
1026 = (free_dbs `snocBag` db, dump_dbs, dump_idset)
1028 | otherwise -- Dump it
1029 = (free_dbs, dump_dbs `snocBag` db,
1030 dump_idset `addOneToIdSet` dict)
1033 Given a type and value substitution, specUDs creates a specialised copy of
1037 specUDs :: [(TyVar,Type)] -> [(DictVar,DictVar)] -> ProtoUsageDetails -> SpecM UsageDetails
1038 specUDs tv_env_list dict_env_list (dbs, calls)
1039 = specDBs dict_env_list dbs `thenSM` \ (dict_env_list', dbs') ->
1041 dict_env = mkIdEnv dict_env_list'
1043 returnSM (MkUD { dict_binds = dbs',
1044 calls = listToCallDetails (map (inst_call dict_env) calls)
1047 bound_tyvars = mkTyVarSet (map fst tv_env_list)
1048 tv_env = mkTyVarEnv tv_env_list -- Doesn't change
1050 inst_call dict_env (id, tys, dicts) = (id, map inst_maybe_ty tys,
1051 map (lookupId dict_env) dicts)
1053 inst_maybe_ty Nothing = Nothing
1054 inst_maybe_ty (Just ty) = Just (instantiateTy tv_env ty)
1057 = returnSM (dict_env, emptyBag)
1058 specDBs dict_env ((dict, rhs, ftvs, fvs) : dbs)
1059 = newIdSM dict (instantiateTy tv_env (idType dict)) `thenSM` \ dict' ->
1061 rhs' = foldl App (foldr Lam rhs (t_bndrs ++ d_bndrs)) (t_args ++ d_args)
1062 (t_bndrs, t_args) = unzip [(TyBinder tv, TyArg ty) | (tv,ty) <- tv_env_list,
1063 tv `elementOfTyVarSet` ftvs]
1064 (d_bndrs, d_args) = unzip [(ValBinder d, VarArg d') | (d,d') <- dict_env,
1065 d `elementOfIdSet` fvs]
1066 dict_env' = (dict,dict') : dict_env
1067 ftvs' = tyVarsOfTypes [ty | TyArg ty <- t_args] `unionTyVarSets`
1068 (ftvs `minusTyVarSet` bound_tyvars)
1069 fvs' = mkIdSet [d | VarArg d <- d_args] `unionIdSets`
1070 (fvs `minusIdSet` mkIdSet [d | ValBinder d <- d_bndrs])
1072 specDBs dict_env' dbs `thenSM` \ (dict_env'', dbs') ->
1073 returnSM ( dict_env'', (dict', rhs', ftvs', fvs') `consBag` dbs' )
1076 %************************************************************************
1078 \subsubsection{Boring helper functions}
1080 %************************************************************************
1083 tyVarTemplates :: [TyVar]
1084 tyVarTemplates = map mk [1..]
1086 mk i = mkTyVar (mkSysLocalName uniq occ noSrcLoc) mkBoxedTypeKind
1088 uniq = mkAlphaTyVarUnique i
1089 occ = _PK_ ("$t" ++ show i)
1093 lookupId:: IdEnv Id -> Id -> Id
1094 lookupId env id = case lookupIdEnv env id of
1098 dictRhsFVs :: CoreExpr -> IdSet
1099 -- Cheapo function for simple RHSs
1103 go (App e1 (VarArg a)) = go e1 `addOneToIdSet` a
1104 go (App e1 (LitArg l)) = go e1
1105 go (App e1 (TyArg t)) = go e1
1106 go (Var v) = unitIdSet v
1107 go (Lit l) = emptyIdSet
1108 go (Con _ args) = mkIdSet [id | VarArg id <- args]
1109 go (Note _ e) = go e
1111 go (Case e _) = go e -- Claim: no free dictionaries in the alternatives
1112 -- These case expressions are of the form
1113 -- case d of { D a b c -> b }
1115 go (Lam _ _) = emptyIdSet -- This can happen for a Functor "dict",
1116 -- which is represented by the function
1117 -- itself; but it won't have any further
1118 -- dicts inside it. I hope.
1120 go other = pprPanic "dictRhsFVs" (ppr e)
1123 addIdSpecialisations id spec_stuff
1124 = (if not (null errs) then
1125 pprTrace "Duplicate specialisations" (vcat (map ppr errs))
1128 setIdSpecialisation id new_spec_env
1130 (new_spec_env, errs) = foldr add (getIdSpecialisation id, []) spec_stuff
1132 add (tyvars, tys, template) (spec_env, errs)
1133 = case addToSpecEnv True spec_env tyvars tys template of
1134 Succeeded spec_env' -> (spec_env', errs)
1135 Failed err -> (spec_env, err:errs)
1137 -- Given an Id, isSpecVars returns all its specialisations.
1138 -- We extract these from its SpecEnv.
1139 -- This is used by the occurrence analyser and free-var finder;
1140 -- we regard an Id's specialisations as free in the Id's definition.
1142 idSpecVars :: Id -> [Id]
1144 = map get_spec (specEnvValues (getIdSpecialisation id))
1146 -- get_spec is another cheapo function like dictRhsFVs
1147 -- It knows what these specialisation temlates look like,
1148 -- and just goes for the jugular
1149 get_spec (App f _) = get_spec f
1150 get_spec (Lam _ b) = get_spec b
1151 get_spec (Var v) = v
1153 ----------------------------------------
1154 type SpecM a = UniqSM a
1158 getUniqSM = getUnique
1162 mapAndCombineSM f [] = returnSM ([], emptyUDs)
1163 mapAndCombineSM f (x:xs) = f x `thenSM` \ (y, uds1) ->
1164 mapAndCombineSM f xs `thenSM` \ (ys, uds2) ->
1165 returnSM (y:ys, uds1 `plusUDs` uds2)
1167 newIdSM old_id new_ty
1168 = getUnique `thenSM` \ uniq ->
1169 returnSM (mkUserLocal (getOccName old_id)
1177 Old (but interesting) stuff about unboxed bindings
1178 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1180 What should we do when a value is specialised to a *strict* unboxed value?
1182 map_*_* f (x:xs) = let h = f x
1186 Could convert let to case:
1188 map_*_Int# f (x:xs) = case f x of h# ->
1192 This may be undesirable since it forces evaluation here, but the value
1193 may not be used in all branches of the body. In the general case this
1194 transformation is impossible since the mutual recursion in a letrec
1195 cannot be expressed as a case.
1197 There is also a problem with top-level unboxed values, since our
1198 implementation cannot handle unboxed values at the top level.
1200 Solution: Lift the binding of the unboxed value and extract it when it
1203 map_*_Int# f (x:xs) = let h = case (f x) of h# -> _Lift h#
1208 Now give it to the simplifier and the _Lifting will be optimised away.
1210 The benfit is that we have given the specialised "unboxed" values a
1211 very simplep lifted semantics and then leave it up to the simplifier to
1212 optimise it --- knowing that the overheads will be removed in nearly
1215 In particular, the value will only be evaluted in the branches of the
1216 program which use it, rather than being forced at the point where the
1217 value is bound. For example:
1219 filtermap_*_* p f (x:xs)
1226 filtermap_*_Int# p f (x:xs)
1227 = let h = case (f x) of h# -> _Lift h#
1230 True -> case h of _Lift h#
1234 The binding for h can still be inlined in the one branch and the
1235 _Lifting eliminated.
1238 Question: When won't the _Lifting be eliminated?
1240 Answer: When they at the top-level (where it is necessary) or when
1241 inlining would duplicate work (or possibly code depending on
1242 options). However, the _Lifting will still be eliminated if the
1243 strictness analyser deems the lifted binding strict.