3 1. Use a library type rather than an annotation for ForceSpecConstr
7 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
9 \section[SpecConstr]{Specialise over constructors}
12 -- The above warning supression flag is a temporary kludge.
13 -- While working on this module you are encouraged to remove it and fix
14 -- any warnings in the module. See
15 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
21 , SpecConstrAnnotation(..)
25 #include "HsVersions.h"
30 import CoreUnfold ( couldBeSmallEnoughToInline )
31 import CoreFVs ( exprsFreeVars )
33 import HscTypes ( ModGuts(..) )
34 import WwLib ( mkWorkerArgs )
38 import Type hiding( substTy )
40 import MkCore ( mkImpossibleExpr )
46 import DynFlags ( DynFlags(..) )
47 import StaticFlags ( opt_PprStyle_Debug )
48 import Maybes ( orElse, catMaybes, isJust, isNothing )
50 import DmdAnal ( both )
51 import Serialized ( deserializeWithData )
58 import Control.Monad ( zipWithM )
62 -- See Note [SpecConstrAnnotation]
64 type SpecConstrAnnotation = ()
66 import Literal ( literalType )
67 import TyCon ( TyCon )
68 import GHC.Exts( SpecConstrAnnotation(..) )
72 -----------------------------------------------------
74 -----------------------------------------------------
79 drop n (x:xs) = drop (n-1) xs
81 After the first time round, we could pass n unboxed. This happens in
82 numerical code too. Here's what it looks like in Core:
84 drop n xs = case xs of
89 _ -> drop (I# (n# -# 1#)) xs
91 Notice that the recursive call has an explicit constructor as argument.
92 Noticing this, we can make a specialised version of drop
94 RULE: drop (I# n#) xs ==> drop' n# xs
96 drop' n# xs = let n = I# n# in ...orig RHS...
98 Now the simplifier will apply the specialisation in the rhs of drop', giving
100 drop' n# xs = case xs of
104 _ -> drop (n# -# 1#) xs
108 We'd also like to catch cases where a parameter is carried along unchanged,
109 but evaluated each time round the loop:
111 f i n = if i>0 || i>n then i else f (i*2) n
113 Here f isn't strict in n, but we'd like to avoid evaluating it each iteration.
114 In Core, by the time we've w/wd (f is strict in i) we get
116 f i# n = case i# ># 0 of
118 True -> case n of n' { I# n# ->
121 True -> f (i# *# 2#) n'
123 At the call to f, we see that the argument, n is know to be (I# n#),
124 and n is evaluated elsewhere in the body of f, so we can play the same
130 We must be careful not to allocate the same constructor twice. Consider
131 f p = (...(case p of (a,b) -> e)...p...,
132 ...let t = (r,s) in ...t...(f t)...)
133 At the recursive call to f, we can see that t is a pair. But we do NOT want
134 to make a specialised copy:
135 f' a b = let p = (a,b) in (..., ...)
136 because now t is allocated by the caller, then r and s are passed to the
137 recursive call, which allocates the (r,s) pair again.
140 (a) the argument p is used in other than a case-scrutinsation way.
141 (b) the argument to the call is not a 'fresh' tuple; you have to
142 look into its unfolding to see that it's a tuple
144 Hence the "OR" part of Note [Good arguments] below.
146 ALTERNATIVE 2: pass both boxed and unboxed versions. This no longer saves
147 allocation, but does perhaps save evals. In the RULE we'd have
150 f (I# x#) = f' (I# x#) x#
152 If at the call site the (I# x) was an unfolding, then we'd have to
153 rely on CSE to eliminate the duplicate allocation.... This alternative
154 doesn't look attractive enough to pursue.
156 ALTERNATIVE 3: ignore the reboxing problem. The trouble is that
157 the conservative reboxing story prevents many useful functions from being
158 specialised. Example:
159 foo :: Maybe Int -> Int -> Int
161 foo x@(Just m) n = foo x (n-m)
162 Here the use of 'x' will clearly not require boxing in the specialised function.
164 The strictness analyser has the same problem, in fact. Example:
166 If we pass just 'a' and 'b' to the worker, it might need to rebox the
167 pair to create (a,b). A more sophisticated analysis might figure out
168 precisely the cases in which this could happen, but the strictness
169 analyser does no such analysis; it just passes 'a' and 'b', and hopes
172 So my current choice is to make SpecConstr similarly aggressive, and
173 ignore the bad potential of reboxing.
176 Note [Good arguments]
177 ~~~~~~~~~~~~~~~~~~~~~
180 * A self-recursive function. Ignore mutual recursion for now,
181 because it's less common, and the code is simpler for self-recursion.
185 a) At a recursive call, one or more parameters is an explicit
186 constructor application
188 That same parameter is scrutinised by a case somewhere in
189 the RHS of the function
193 b) At a recursive call, one or more parameters has an unfolding
194 that is an explicit constructor application
196 That same parameter is scrutinised by a case somewhere in
197 the RHS of the function
199 Those are the only uses of the parameter (see Note [Reboxing])
202 What to abstract over
203 ~~~~~~~~~~~~~~~~~~~~~
204 There's a bit of a complication with type arguments. If the call
207 f p = ...f ((:) [a] x xs)...
209 then our specialised function look like
211 f_spec x xs = let p = (:) [a] x xs in ....as before....
213 This only makes sense if either
214 a) the type variable 'a' is in scope at the top of f, or
215 b) the type variable 'a' is an argument to f (and hence fs)
217 Actually, (a) may hold for value arguments too, in which case
218 we may not want to pass them. Supose 'x' is in scope at f's
219 defn, but xs is not. Then we'd like
221 f_spec xs = let p = (:) [a] x xs in ....as before....
223 Similarly (b) may hold too. If x is already an argument at the
224 call, no need to pass it again.
226 Finally, if 'a' is not in scope at the call site, we could abstract
227 it as we do the term variables:
229 f_spec a x xs = let p = (:) [a] x xs in ...as before...
231 So the grand plan is:
233 * abstract the call site to a constructor-only pattern
234 e.g. C x (D (f p) (g q)) ==> C s1 (D s2 s3)
236 * Find the free variables of the abstracted pattern
238 * Pass these variables, less any that are in scope at
239 the fn defn. But see Note [Shadowing] below.
242 NOTICE that we only abstract over variables that are not in scope,
243 so we're in no danger of shadowing variables used in "higher up"
249 In this pass we gather up usage information that may mention variables
250 that are bound between the usage site and the definition site; or (more
251 seriously) may be bound to something different at the definition site.
254 f x = letrec g y v = let x = ...
257 Since 'x' is in scope at the call site, we may make a rewrite rule that
259 RULE forall a,b. g (a,b) x = ...
260 But this rule will never match, because it's really a different 'x' at
261 the call site -- and that difference will be manifest by the time the
262 simplifier gets to it. [A worry: the simplifier doesn't *guarantee*
263 no-shadowing, so perhaps it may not be distinct?]
265 Anyway, the rule isn't actually wrong, it's just not useful. One possibility
266 is to run deShadowBinds before running SpecConstr, but instead we run the
267 simplifier. That gives the simplest possible program for SpecConstr to
268 chew on; and it virtually guarantees no shadowing.
270 Note [Specialising for constant parameters]
271 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
272 This one is about specialising on a *constant* (but not necessarily
273 constructor) argument
275 foo :: Int -> (Int -> Int) -> Int
277 foo m f = foo (f m) (+1)
281 lvl_rmV :: GHC.Base.Int -> GHC.Base.Int
283 \ (ds_dlk :: GHC.Base.Int) ->
284 case ds_dlk of wild_alH { GHC.Base.I# x_alG ->
285 GHC.Base.I# (GHC.Prim.+# x_alG 1)
287 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
290 \ (ww_sme :: GHC.Prim.Int#) (w_smg :: GHC.Base.Int -> GHC.Base.Int) ->
291 case ww_sme of ds_Xlw {
293 case w_smg (GHC.Base.I# ds_Xlw) of w1_Xmo { GHC.Base.I# ww1_Xmz ->
294 T.$wfoo ww1_Xmz lvl_rmV
299 The recursive call has lvl_rmV as its argument, so we could create a specialised copy
300 with that argument baked in; that is, not passed at all. Now it can perhaps be inlined.
302 When is this worth it? Call the constant 'lvl'
303 - If 'lvl' has an unfolding that is a constructor, see if the corresponding
304 parameter is scrutinised anywhere in the body.
306 - If 'lvl' has an unfolding that is a inlinable function, see if the corresponding
307 parameter is applied (...to enough arguments...?)
309 Also do this is if the function has RULES?
313 Note [Specialising for lambda parameters]
314 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
315 foo :: Int -> (Int -> Int) -> Int
317 foo m f = foo (f m) (\n -> n-m)
319 This is subtly different from the previous one in that we get an
320 explicit lambda as the argument:
322 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
325 \ (ww_sm8 :: GHC.Prim.Int#) (w_sma :: GHC.Base.Int -> GHC.Base.Int) ->
326 case ww_sm8 of ds_Xlr {
328 case w_sma (GHC.Base.I# ds_Xlr) of w1_Xmf { GHC.Base.I# ww1_Xmq ->
331 (\ (n_ad3 :: GHC.Base.Int) ->
332 case n_ad3 of wild_alB { GHC.Base.I# x_alA ->
333 GHC.Base.I# (GHC.Prim.-# x_alA ds_Xlr)
339 I wonder if SpecConstr couldn't be extended to handle this? After all,
340 lambda is a sort of constructor for functions and perhaps it already
341 has most of the necessary machinery?
343 Furthermore, there's an immediate win, because you don't need to allocate the lamda
344 at the call site; and if perchance it's called in the recursive call, then you
345 may avoid allocating it altogether. Just like for constructors.
347 Looks cool, but probably rare...but it might be easy to implement.
350 Note [SpecConstr for casts]
351 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
354 data instance T Int = T Int
359 go (T n) = go (T (n-1))
361 The recursive call ends up looking like
362 go (T (I# ...) `cast` g)
363 So we want to spot the construtor application inside the cast.
364 That's why we have the Cast case in argToPat
366 Note [Local recursive groups]
367 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
368 For a *local* recursive group, we can see all the calls to the
369 function, so we seed the specialisation loop from the calls in the
370 body, not from the calls in the RHS. Consider:
372 bar m n = foo n (n,n) (n,n) (n,n) (n,n)
376 | n > 3000 = case p of { (p1,p2) -> foo (n-1) (p2,p1) q r s }
377 | n > 2000 = case q of { (q1,q2) -> foo (n-1) p (q2,q1) r s }
378 | n > 1000 = case r of { (r1,r2) -> foo (n-1) p q (r2,r1) s }
379 | otherwise = case s of { (s1,s2) -> foo (n-1) p q r (s2,s1) }
381 If we start with the RHSs of 'foo', we get lots and lots of specialisations,
382 most of which are not needed. But if we start with the (single) call
383 in the rhs of 'bar' we get exactly one fully-specialised copy, and all
384 the recursive calls go to this fully-specialised copy. Indeed, the original
385 function is later collected as dead code. This is very important in
386 specialising the loops arising from stream fusion, for example in NDP where
387 we were getting literally hundreds of (mostly unused) specialisations of
390 Note [Do not specialise diverging functions]
391 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
392 Specialising a function that just diverges is a waste of code.
393 Furthermore, it broke GHC (simpl014) thus:
395 f = \x. case x of (a,b) -> f x
396 If we specialise f we get
397 f = \x. case x of (a,b) -> fspec a b
398 But fspec doesn't have decent strictnes info. As it happened,
399 (f x) :: IO t, so the state hack applied and we eta expanded fspec,
400 and hence f. But now f's strictness is less than its arity, which
403 Note [SpecConstrAnnotation]
404 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
405 SpecConstrAnnotation is defined in GHC.Exts, and is only guaranteed to
406 be available in stage 2 (well, until the bootstrap compiler can be
407 guaranteed to have it)
409 So we define it to be () in stage1 (ie when GHCI is undefined), and
410 '#ifdef' out the code that uses it.
412 See also Note [Forcing specialisation]
414 Note [Forcing specialisation]
415 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
416 With stream fusion and in other similar cases, we want to fully specialise
417 some (but not necessarily all!) loops regardless of their size and the
418 number of specialisations. We allow a library to specify this by annotating
419 a type with ForceSpecConstr and then adding a parameter of that type to the
420 loop. Here is a (simplified) example from the vector library:
422 data SPEC = SPEC | SPEC2
423 {-# ANN type SPEC ForceSpecConstr #-}
425 foldl :: (a -> b -> a) -> a -> Stream b -> a
427 foldl f z (Stream step s _) = foldl_loop SPEC z s
429 foldl_loop !sPEC z s = case step s of
430 Yield x s' -> foldl_loop sPEC (f z x) s'
431 Skip -> foldl_loop sPEC z s'
434 SpecConstr will spot the SPEC parameter and always fully specialise
435 foldl_loop. Note that
437 * We have to prevent the SPEC argument from being removed by
438 w/w which is why (a) SPEC is a sum type, and (b) we have to seq on
441 * And lastly, the SPEC argument is ultimately eliminated by
442 SpecConstr itself so there is no runtime overhead.
444 This is all quite ugly; we ought to come up with a better design.
446 ForceSpecConstr arguments are spotted in scExpr' and scTopBinds which then set
447 sc_force to True when calling specLoop. This flag does three things:
448 * Ignore specConstrThreshold, to specialise functions of arbitrary size
450 * Ignore specConstrCount, to make arbitrary numbers of specialisations
452 * Specialise even for arguments that are not scrutinised in the loop
453 (see argToPat; Trac #4488)
455 What alternatives did I consider? Annotating the loop itself doesn't
456 work because (a) it is local and (b) it will be w/w'ed and I having
457 w/w propagating annotation somehow doesn't seem like a good idea. The
458 types of the loop arguments really seem to be the most persistent
461 Annotating the types that make up the loop state doesn't work,
462 either, because (a) it would prevent us from using types like Either
463 or tuples here, (b) we don't want to restrict the set of types that
464 can be used in Stream states and (c) some types are fixed by the user
465 (e.g., the accumulator here) but we still want to specialise as much
468 ForceSpecConstr is done by way of an annotation:
469 data SPEC = SPEC | SPEC2
470 {-# ANN type SPEC ForceSpecConstr #-}
471 But SPEC is the *only* type so annotated, so it'd be better to
472 use a particular library type.
474 Alternatives to ForceSpecConstr
475 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
476 Instead of giving the loop an extra argument of type SPEC, we
477 also considered *wrapping* arguments in SPEC, thus
478 data SPEC a = SPEC a | SPEC2
480 loop = \arg -> case arg of
482 case state of (x,y) -> ... loop (SPEC (x',y')) ...
484 The idea is that a SPEC argument says "specialise this argument
485 regardless of whether the function case-analyses it. But this
487 * SPEC must still be a sum type, else the strictness analyser
489 * But that means that 'loop' won't be strict in its real payload
490 This loss of strictness in turn screws up specialisation, because
491 we may end up with calls like
492 loop (SPEC (case z of (p,q) -> (q,p)))
493 Without the SPEC, if 'loop' was strict, the case would move out
494 and we'd see loop applied to a pair. But if 'loop' isn' strict
495 this doesn't look like a specialisable call.
499 The ignoreAltCon stuff allows you to say
500 {-# ANN type T NoSpecConstr #-}
501 to mean "don't specialise on arguments of this type. It was added
502 before we had ForceSpecConstr. Lacking ForceSpecConstr we specialised
503 regardless of size; and then we needed a way to turn that *off*. Now
504 that we have ForceSpecConstr, this NoSpecConstr is probably redundant.
505 (Used only for PArray.)
507 -----------------------------------------------------
508 Stuff not yet handled
509 -----------------------------------------------------
511 Here are notes arising from Roman's work that I don't want to lose.
517 foo :: Int -> T Int -> Int
519 foo x t | even x = case t of { T n -> foo (x-n) t }
520 | otherwise = foo (x-1) t
522 SpecConstr does no specialisation, because the second recursive call
523 looks like a boxed use of the argument. A pity.
525 $wfoo_sFw :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
527 \ (ww_sFo [Just L] :: GHC.Prim.Int#) (w_sFq [Just L] :: T.T GHC.Base.Int) ->
528 case ww_sFo of ds_Xw6 [Just L] {
530 case GHC.Prim.remInt# ds_Xw6 2 of wild1_aEF [Dead Just A] {
531 __DEFAULT -> $wfoo_sFw (GHC.Prim.-# ds_Xw6 1) w_sFq;
533 case w_sFq of wild_Xy [Just L] { T.T n_ad5 [Just U(L)] ->
534 case n_ad5 of wild1_aET [Just A] { GHC.Base.I# y_aES [Just L] ->
535 $wfoo_sFw (GHC.Prim.-# ds_Xw6 y_aES) wild_Xy
541 data a :*: b = !a :*: !b
544 foo :: (Int :*: T Int) -> Int
546 foo (x :*: t) | even x = case t of { T n -> foo ((x-n) :*: t) }
547 | otherwise = foo ((x-1) :*: t)
549 Very similar to the previous one, except that the parameters are now in
550 a strict tuple. Before SpecConstr, we have
552 $wfoo_sG3 :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
554 \ (ww_sFU [Just L] :: GHC.Prim.Int#) (ww_sFW [Just L] :: T.T
556 case ww_sFU of ds_Xws [Just L] {
558 case GHC.Prim.remInt# ds_Xws 2 of wild1_aEZ [Dead Just A] {
560 case ww_sFW of tpl_B2 [Just L] { T.T a_sFo [Just A] ->
561 $wfoo_sG3 (GHC.Prim.-# ds_Xws 1) tpl_B2 -- $wfoo1
564 case ww_sFW of wild_XB [Just A] { T.T n_ad7 [Just S(L)] ->
565 case n_ad7 of wild1_aFd [Just L] { GHC.Base.I# y_aFc [Just L] ->
566 $wfoo_sG3 (GHC.Prim.-# ds_Xws y_aFc) wild_XB -- $wfoo2
570 We get two specialisations:
571 "SC:$wfoo1" [0] __forall {a_sFB :: GHC.Base.Int sc_sGC :: GHC.Prim.Int#}
572 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int a_sFB)
573 = Foo.$s$wfoo1 a_sFB sc_sGC ;
574 "SC:$wfoo2" [0] __forall {y_aFp :: GHC.Prim.Int# sc_sGC :: GHC.Prim.Int#}
575 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int (GHC.Base.I# y_aFp))
576 = Foo.$s$wfoo y_aFp sc_sGC ;
578 But perhaps the first one isn't good. After all, we know that tpl_B2 is
579 a T (I# x) really, because T is strict and Int has one constructor. (We can't
580 unbox the strict fields, becuase T is polymorphic!)
582 %************************************************************************
584 \subsection{Top level wrapper stuff}
586 %************************************************************************
589 specConstrProgram :: ModGuts -> CoreM ModGuts
590 specConstrProgram guts
592 dflags <- getDynFlags
593 us <- getUniqueSupplyM
594 annos <- getFirstAnnotations deserializeWithData guts
595 let binds' = fst $ initUs us (go (initScEnv dflags annos) (mg_binds guts))
596 return (guts { mg_binds = binds' })
599 go env (bind:binds) = do (env', bind') <- scTopBind env bind
600 binds' <- go env' binds
601 return (bind' : binds')
605 %************************************************************************
607 \subsection{Environment: goes downwards}
609 %************************************************************************
612 data ScEnv = SCE { sc_size :: Maybe Int, -- Size threshold
613 sc_count :: Maybe Int, -- Max # of specialisations for any one fn
614 -- See Note [Avoiding exponential blowup]
615 sc_force :: Bool, -- Force specialisation?
616 -- See Note [Forcing specialisation]
618 sc_subst :: Subst, -- Current substitution
619 -- Maps InIds to OutExprs
621 sc_how_bound :: HowBoundEnv,
622 -- Binds interesting non-top-level variables
623 -- Domain is OutVars (*after* applying the substitution)
626 -- Domain is OutIds (*after* applying the substitution)
627 -- Used even for top-level bindings (but not imported ones)
629 sc_annotations :: UniqFM SpecConstrAnnotation
632 ---------------------
633 -- As we go, we apply a substitution (sc_subst) to the current term
634 type InExpr = CoreExpr -- _Before_ applying the subst
637 type OutExpr = CoreExpr -- _After_ applying the subst
641 ---------------------
642 type HowBoundEnv = VarEnv HowBound -- Domain is OutVars
644 ---------------------
645 type ValueEnv = IdEnv Value -- Domain is OutIds
646 data Value = ConVal AltCon [CoreArg] -- _Saturated_ constructors
647 -- The AltCon is never DEFAULT
648 | LambdaVal -- Inlinable lambdas or PAPs
650 instance Outputable Value where
651 ppr (ConVal con args) = ppr con <+> interpp'SP args
652 ppr LambdaVal = ptext (sLit "<Lambda>")
654 ---------------------
655 initScEnv :: DynFlags -> UniqFM SpecConstrAnnotation -> ScEnv
656 initScEnv dflags anns
657 = SCE { sc_size = specConstrThreshold dflags,
658 sc_count = specConstrCount dflags,
660 sc_subst = emptySubst,
661 sc_how_bound = emptyVarEnv,
662 sc_vals = emptyVarEnv,
663 sc_annotations = anns }
665 data HowBound = RecFun -- These are the recursive functions for which
666 -- we seek interesting call patterns
668 | RecArg -- These are those functions' arguments, or their sub-components;
669 -- we gather occurrence information for these
671 instance Outputable HowBound where
672 ppr RecFun = text "RecFun"
673 ppr RecArg = text "RecArg"
675 scForce :: ScEnv -> Bool -> ScEnv
676 scForce env b = env { sc_force = b }
678 lookupHowBound :: ScEnv -> Id -> Maybe HowBound
679 lookupHowBound env id = lookupVarEnv (sc_how_bound env) id
681 scSubstId :: ScEnv -> Id -> CoreExpr
682 scSubstId env v = lookupIdSubst (text "scSubstId") (sc_subst env) v
684 scSubstTy :: ScEnv -> Type -> Type
685 scSubstTy env ty = substTy (sc_subst env) ty
687 zapScSubst :: ScEnv -> ScEnv
688 zapScSubst env = env { sc_subst = zapSubstEnv (sc_subst env) }
690 extendScInScope :: ScEnv -> [Var] -> ScEnv
691 -- Bring the quantified variables into scope
692 extendScInScope env qvars = env { sc_subst = extendInScopeList (sc_subst env) qvars }
694 -- Extend the substitution
695 extendScSubst :: ScEnv -> Var -> OutExpr -> ScEnv
696 extendScSubst env var expr = env { sc_subst = extendSubst (sc_subst env) var expr }
698 extendScSubstList :: ScEnv -> [(Var,OutExpr)] -> ScEnv
699 extendScSubstList env prs = env { sc_subst = extendSubstList (sc_subst env) prs }
701 extendHowBound :: ScEnv -> [Var] -> HowBound -> ScEnv
702 extendHowBound env bndrs how_bound
703 = env { sc_how_bound = extendVarEnvList (sc_how_bound env)
704 [(bndr,how_bound) | bndr <- bndrs] }
706 extendBndrsWith :: HowBound -> ScEnv -> [Var] -> (ScEnv, [Var])
707 extendBndrsWith how_bound env bndrs
708 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndrs')
710 (subst', bndrs') = substBndrs (sc_subst env) bndrs
711 hb_env' = sc_how_bound env `extendVarEnvList`
712 [(bndr,how_bound) | bndr <- bndrs']
714 extendBndrWith :: HowBound -> ScEnv -> Var -> (ScEnv, Var)
715 extendBndrWith how_bound env bndr
716 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndr')
718 (subst', bndr') = substBndr (sc_subst env) bndr
719 hb_env' = extendVarEnv (sc_how_bound env) bndr' how_bound
721 extendRecBndrs :: ScEnv -> [Var] -> (ScEnv, [Var])
722 extendRecBndrs env bndrs = (env { sc_subst = subst' }, bndrs')
724 (subst', bndrs') = substRecBndrs (sc_subst env) bndrs
726 extendBndr :: ScEnv -> Var -> (ScEnv, Var)
727 extendBndr env bndr = (env { sc_subst = subst' }, bndr')
729 (subst', bndr') = substBndr (sc_subst env) bndr
731 extendValEnv :: ScEnv -> Id -> Maybe Value -> ScEnv
732 extendValEnv env _ Nothing = env
733 extendValEnv env id (Just cv) = env { sc_vals = extendVarEnv (sc_vals env) id cv }
735 extendCaseBndrs :: ScEnv -> Id -> AltCon -> [Var] -> (ScEnv, [Var])
739 -- we want to bind b, to (C x y)
740 -- NB1: Extends only the sc_vals part of the envt
741 -- NB2: Kill the dead-ness info on the pattern binders x,y, since
742 -- they are potentially made alive by the [b -> C x y] binding
743 extendCaseBndrs env case_bndr con alt_bndrs
744 | isDeadBinder case_bndr
747 = (env1, map zap alt_bndrs)
748 -- NB: We used to bind v too, if scrut = (Var v); but
749 -- the simplifer has already done this so it seems
750 -- redundant to do so here
752 -- Var v -> extendValEnv env1 v cval
755 zap v | isTyCoVar v = v -- See NB2 above
756 | otherwise = zapIdOccInfo v
757 env1 = extendValEnv env case_bndr cval
760 LitAlt {} -> Just (ConVal con [])
761 DataAlt {} -> Just (ConVal con vanilla_args)
763 vanilla_args = map Type (tyConAppArgs (idType case_bndr)) ++
764 varsToCoreExprs alt_bndrs
767 decreaseSpecCount :: ScEnv -> Int -> ScEnv
768 -- See Note [Avoiding exponential blowup]
769 decreaseSpecCount env n_specs
770 = env { sc_count = case sc_count env of
772 Just n -> Just (n `div` (n_specs + 1)) }
773 -- The "+1" takes account of the original function;
774 -- See Note [Avoiding exponential blowup]
776 ---------------------------------------------------
777 -- See Note [SpecConstrAnnotation]
778 ignoreType :: ScEnv -> Type -> Bool
779 ignoreAltCon :: ScEnv -> AltCon -> Bool
780 forceSpecBndr :: ScEnv -> Var -> Bool
782 ignoreType _ _ = False
783 ignoreAltCon _ _ = False
784 forceSpecBndr _ _ = False
788 ignoreAltCon env (DataAlt dc) = ignoreTyCon env (dataConTyCon dc)
789 ignoreAltCon env (LitAlt lit) = ignoreType env (literalType lit)
790 ignoreAltCon _ DEFAULT = panic "ignoreAltCon" -- DEFAULT cannot be in a ConVal
793 = case splitTyConApp_maybe ty of
794 Just (tycon, _) -> ignoreTyCon env tycon
797 ignoreTyCon :: ScEnv -> TyCon -> Bool
798 ignoreTyCon env tycon
799 = lookupUFM (sc_annotations env) tycon == Just NoSpecConstr
801 forceSpecBndr env var = forceSpecFunTy env . snd . splitForAllTys . varType $ var
803 forceSpecFunTy :: ScEnv -> Type -> Bool
804 forceSpecFunTy env = any (forceSpecArgTy env) . fst . splitFunTys
806 forceSpecArgTy :: ScEnv -> Type -> Bool
807 forceSpecArgTy env ty
808 | Just ty' <- coreView ty = forceSpecArgTy env ty'
810 forceSpecArgTy env ty
811 | Just (tycon, tys) <- splitTyConApp_maybe ty
813 = lookupUFM (sc_annotations env) tycon == Just ForceSpecConstr
814 || any (forceSpecArgTy env) tys
816 forceSpecArgTy _ _ = False
820 Note [Avoiding exponential blowup]
821 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
822 The sc_count field of the ScEnv says how many times we are prepared to
823 duplicate a single function. But we must take care with recursive
824 specialiations. Consider
826 let $j1 = let $j2 = let $j3 = ...
834 If we specialise $j1 then in each specialisation (as well as the original)
835 we can specialise $j2, and similarly $j3. Even if we make just *one*
836 specialisation of each, becuase we also have the original we'll get 2^n
837 copies of $j3, which is not good.
839 So when recursively specialising we divide the sc_count by the number of
840 copies we are making at this level, including the original.
843 %************************************************************************
845 \subsection{Usage information: flows upwards}
847 %************************************************************************
852 scu_calls :: CallEnv, -- Calls
853 -- The functions are a subset of the
854 -- RecFuns in the ScEnv
856 scu_occs :: !(IdEnv ArgOcc) -- Information on argument occurrences
857 } -- The domain is OutIds
859 type CallEnv = IdEnv [Call]
860 type Call = (ValueEnv, [CoreArg])
861 -- The arguments of the call, together with the
862 -- env giving the constructor bindings at the call site
865 nullUsage = SCU { scu_calls = emptyVarEnv, scu_occs = emptyVarEnv }
867 combineCalls :: CallEnv -> CallEnv -> CallEnv
868 combineCalls = plusVarEnv_C (++)
870 combineUsage :: ScUsage -> ScUsage -> ScUsage
871 combineUsage u1 u2 = SCU { scu_calls = combineCalls (scu_calls u1) (scu_calls u2),
872 scu_occs = plusVarEnv_C combineOcc (scu_occs u1) (scu_occs u2) }
874 combineUsages :: [ScUsage] -> ScUsage
875 combineUsages [] = nullUsage
876 combineUsages us = foldr1 combineUsage us
878 lookupOcc :: ScUsage -> OutVar -> (ScUsage, ArgOcc)
879 lookupOcc (SCU { scu_calls = sc_calls, scu_occs = sc_occs }) bndr
880 = (SCU {scu_calls = sc_calls, scu_occs = delVarEnv sc_occs bndr},
881 lookupVarEnv sc_occs bndr `orElse` NoOcc)
883 lookupOccs :: ScUsage -> [OutVar] -> (ScUsage, [ArgOcc])
884 lookupOccs (SCU { scu_calls = sc_calls, scu_occs = sc_occs }) bndrs
885 = (SCU {scu_calls = sc_calls, scu_occs = delVarEnvList sc_occs bndrs},
886 [lookupVarEnv sc_occs b `orElse` NoOcc | b <- bndrs])
888 data ArgOcc = NoOcc -- Doesn't occur at all; or a type argument
889 | UnkOcc -- Used in some unknown way
891 | ScrutOcc (UniqFM [ArgOcc]) -- See Note [ScrutOcc]
893 | BothOcc -- Definitely taken apart, *and* perhaps used in some other way
897 An occurrence of ScrutOcc indicates that the thing, or a `cast` version of the thing,
898 is *only* taken apart or applied.
900 Functions, literal: ScrutOcc emptyUFM
901 Data constructors: ScrutOcc subs,
903 where (subs :: UniqFM [ArgOcc]) gives usage of the *pattern-bound* components,
904 The domain of the UniqFM is the Unique of the data constructor
906 The [ArgOcc] is the occurrences of the *pattern-bound* components
907 of the data structure. E.g.
908 data T a = forall b. MkT a b (b->a)
909 A pattern binds b, x::a, y::b, z::b->a, but not 'a'!
913 instance Outputable ArgOcc where
914 ppr (ScrutOcc xs) = ptext (sLit "scrut-occ") <> ppr xs
915 ppr UnkOcc = ptext (sLit "unk-occ")
916 ppr BothOcc = ptext (sLit "both-occ")
917 ppr NoOcc = ptext (sLit "no-occ")
919 -- Experimentally, this vesion of combineOcc makes ScrutOcc "win", so
920 -- that if the thing is scrutinised anywhere then we get to see that
921 -- in the overall result, even if it's also used in a boxed way
922 -- This might be too agressive; see Note [Reboxing] Alternative 3
923 combineOcc :: ArgOcc -> ArgOcc -> ArgOcc
924 combineOcc NoOcc occ = occ
925 combineOcc occ NoOcc = occ
926 combineOcc (ScrutOcc xs) (ScrutOcc ys) = ScrutOcc (plusUFM_C combineOccs xs ys)
927 combineOcc _occ (ScrutOcc ys) = ScrutOcc ys
928 combineOcc (ScrutOcc xs) _occ = ScrutOcc xs
929 combineOcc UnkOcc UnkOcc = UnkOcc
930 combineOcc _ _ = BothOcc
932 combineOccs :: [ArgOcc] -> [ArgOcc] -> [ArgOcc]
933 combineOccs xs ys = zipWithEqual "combineOccs" combineOcc xs ys
935 setScrutOcc :: ScEnv -> ScUsage -> OutExpr -> ArgOcc -> ScUsage
936 -- _Overwrite_ the occurrence info for the scrutinee, if the scrutinee
937 -- is a variable, and an interesting variable
938 setScrutOcc env usg (Cast e _) occ = setScrutOcc env usg e occ
939 setScrutOcc env usg (Note _ e) occ = setScrutOcc env usg e occ
940 setScrutOcc env usg (Var v) occ
941 | Just RecArg <- lookupHowBound env v = usg { scu_occs = extendVarEnv (scu_occs usg) v occ }
943 setScrutOcc _env usg _other _occ -- Catch-all
946 conArgOccs :: ArgOcc -> AltCon -> [ArgOcc]
947 -- Find usage of components of data con; returns [UnkOcc...] if unknown
948 -- See Note [ScrutOcc] for the extra UnkOccs in the vanilla datacon case
950 conArgOccs (ScrutOcc fm) (DataAlt dc)
951 | Just pat_arg_occs <- lookupUFM fm dc
952 = [UnkOcc | _ <- dataConUnivTyVars dc] ++ pat_arg_occs
954 conArgOccs _other _con = repeat UnkOcc
957 %************************************************************************
959 \subsection{The main recursive function}
961 %************************************************************************
963 The main recursive function gathers up usage information, and
964 creates specialised versions of functions.
967 scExpr, scExpr' :: ScEnv -> CoreExpr -> UniqSM (ScUsage, CoreExpr)
968 -- The unique supply is needed when we invent
969 -- a new name for the specialised function and its args
971 scExpr env e = scExpr' env e
974 scExpr' env (Var v) = case scSubstId env v of
975 Var v' -> return (varUsage env v' UnkOcc, Var v')
976 e' -> scExpr (zapScSubst env) e'
978 scExpr' env (Type t) = return (nullUsage, Type (scSubstTy env t))
979 scExpr' _ e@(Lit {}) = return (nullUsage, e)
980 scExpr' env (Note n e) = do (usg,e') <- scExpr env e
981 return (usg, Note n e')
982 scExpr' env (Cast e co) = do (usg, e') <- scExpr env e
983 return (usg, Cast e' (scSubstTy env co))
984 scExpr' env e@(App _ _) = scApp env (collectArgs e)
985 scExpr' env (Lam b e) = do let (env', b') = extendBndr env b
986 (usg, e') <- scExpr env' e
987 return (usg, Lam b' e')
989 scExpr' env (Case scrut b ty alts)
990 = do { (scrut_usg, scrut') <- scExpr env scrut
991 ; case isValue (sc_vals env) scrut' of
992 Just (ConVal con args) -> sc_con_app con args scrut'
993 _other -> sc_vanilla scrut_usg scrut'
996 sc_con_app con args scrut' -- Known constructor; simplify
997 = do { let (_, bs, rhs) = findAlt con alts
998 `orElse` (DEFAULT, [], mkImpossibleExpr (coreAltsType alts))
999 alt_env' = extendScSubstList env ((b,scrut') : bs `zip` trimConArgs con args)
1000 ; scExpr alt_env' rhs }
1002 sc_vanilla scrut_usg scrut' -- Normal case
1003 = do { let (alt_env,b') = extendBndrWith RecArg env b
1004 -- Record RecArg for the components
1006 ; (alt_usgs, alt_occs, alts')
1007 <- mapAndUnzip3M (sc_alt alt_env scrut' b') alts
1009 ; let (alt_usg, b_occ) = lookupOcc (combineUsages alt_usgs) b'
1010 scrut_occ = foldr combineOcc b_occ alt_occs
1011 scrut_usg' = setScrutOcc env scrut_usg scrut' scrut_occ
1012 -- The combined usage of the scrutinee is given
1013 -- by scrut_occ, which is passed to scScrut, which
1014 -- in turn treats a bare-variable scrutinee specially
1016 ; return (alt_usg `combineUsage` scrut_usg',
1017 Case scrut' b' (scSubstTy env ty) alts') }
1019 sc_alt env _scrut' b' (con,bs,rhs)
1020 = do { let (env1, bs1) = extendBndrsWith RecArg env bs
1021 (env2, bs2) = extendCaseBndrs env1 b' con bs1
1022 ; (usg,rhs') <- scExpr env2 rhs
1023 ; let (usg', arg_occs) = lookupOccs usg bs2
1024 scrut_occ = case con of
1025 DataAlt dc -> ScrutOcc (unitUFM dc arg_occs)
1026 _ -> ScrutOcc emptyUFM
1027 ; return (usg', scrut_occ, (con, bs2, rhs')) }
1029 scExpr' env (Let (NonRec bndr rhs) body)
1030 | isTyCoVar bndr -- Type-lets may be created by doBeta
1031 = scExpr' (extendScSubst env bndr rhs) body
1034 = do { let (body_env, bndr') = extendBndr env bndr
1035 ; (rhs_usg, rhs_info) <- scRecRhs env (bndr',rhs)
1037 ; let body_env2 = extendHowBound body_env [bndr'] RecFun
1038 -- Note [Local let bindings]
1039 RI _ rhs' _ _ _ = rhs_info
1040 body_env3 = extendValEnv body_env2 bndr' (isValue (sc_vals env) rhs')
1042 ; (body_usg, body') <- scExpr body_env3 body
1044 -- NB: We don't use the ForceSpecConstr mechanism (see
1045 -- Note [Forcing specialisation]) for non-recursive bindings
1046 -- at the moment. I'm not sure if this is the right thing to do.
1047 ; let env' = scForce env False
1048 ; (spec_usg, specs) <- specialise env'
1049 (scu_calls body_usg)
1051 (SI [] 0 (Just rhs_usg))
1053 ; return (body_usg { scu_calls = scu_calls body_usg `delVarEnv` bndr' }
1054 `combineUsage` spec_usg,
1055 mkLets [NonRec b r | (b,r) <- specInfoBinds rhs_info specs] body')
1059 -- A *local* recursive group: see Note [Local recursive groups]
1060 scExpr' env (Let (Rec prs) body)
1061 = do { let (bndrs,rhss) = unzip prs
1062 (rhs_env1,bndrs') = extendRecBndrs env bndrs
1063 rhs_env2 = extendHowBound rhs_env1 bndrs' RecFun
1064 force_spec = any (forceSpecBndr env) bndrs'
1065 -- Note [Forcing specialisation]
1067 ; (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs rhs_env2) (bndrs' `zip` rhss)
1068 ; (body_usg, body') <- scExpr rhs_env2 body
1070 -- NB: start specLoop from body_usg
1071 ; (spec_usg, specs) <- specLoop (scForce rhs_env2 force_spec)
1072 (scu_calls body_usg) rhs_infos nullUsage
1073 [SI [] 0 (Just usg) | usg <- rhs_usgs]
1074 -- Do not unconditionally use rhs_usgs.
1075 -- Instead use them only if we find an unspecialised call
1076 -- See Note [Local recursive groups]
1078 ; let all_usg = spec_usg `combineUsage` body_usg
1079 bind' = Rec (concat (zipWith specInfoBinds rhs_infos specs))
1081 ; return (all_usg { scu_calls = scu_calls all_usg `delVarEnvList` bndrs' },
1085 Note [Local let bindings]
1086 ~~~~~~~~~~~~~~~~~~~~~~~~~
1087 It is not uncommon to find this
1089 let $j = \x. <blah> in ...$j True...$j True...
1091 Here $j is an arbitrary let-bound function, but it often comes up for
1092 join points. We might like to specialise $j for its call patterns.
1093 Notice the difference from a letrec, where we look for call patterns
1094 in the *RHS* of the function. Here we look for call patterns in the
1097 At one point I predicated this on the RHS mentioning the outer
1098 recursive function, but that's not essential and might even be
1099 harmful. I'm not sure.
1103 scApp :: ScEnv -> (InExpr, [InExpr]) -> UniqSM (ScUsage, CoreExpr)
1105 scApp env (Var fn, args) -- Function is a variable
1106 = ASSERT( not (null args) )
1107 do { args_w_usgs <- mapM (scExpr env) args
1108 ; let (arg_usgs, args') = unzip args_w_usgs
1109 arg_usg = combineUsages arg_usgs
1110 ; case scSubstId env fn of
1111 fn'@(Lam {}) -> scExpr (zapScSubst env) (doBeta fn' args')
1112 -- Do beta-reduction and try again
1114 Var fn' -> return (arg_usg `combineUsage` fn_usg, mkApps (Var fn') args')
1116 fn_usg = case lookupHowBound env fn' of
1117 Just RecFun -> SCU { scu_calls = unitVarEnv fn' [(sc_vals env, args')],
1118 scu_occs = emptyVarEnv }
1119 Just RecArg -> SCU { scu_calls = emptyVarEnv,
1120 scu_occs = unitVarEnv fn' (ScrutOcc emptyUFM) }
1121 Nothing -> nullUsage
1124 other_fn' -> return (arg_usg, mkApps other_fn' args') }
1125 -- NB: doing this ignores any usage info from the substituted
1126 -- function, but I don't think that matters. If it does
1129 doBeta :: OutExpr -> [OutExpr] -> OutExpr
1130 -- ToDo: adjust for System IF
1131 doBeta (Lam bndr body) (arg : args) = Let (NonRec bndr arg) (doBeta body args)
1132 doBeta fn args = mkApps fn args
1134 -- The function is almost always a variable, but not always.
1135 -- In particular, if this pass follows float-in,
1136 -- which it may, we can get
1137 -- (let f = ...f... in f) arg1 arg2
1138 scApp env (other_fn, args)
1139 = do { (fn_usg, fn') <- scExpr env other_fn
1140 ; (arg_usgs, args') <- mapAndUnzipM (scExpr env) args
1141 ; return (combineUsages arg_usgs `combineUsage` fn_usg, mkApps fn' args') }
1143 ----------------------
1144 scTopBind :: ScEnv -> CoreBind -> UniqSM (ScEnv, CoreBind)
1145 scTopBind env (Rec prs)
1146 | Just threshold <- sc_size env
1148 , not (all (couldBeSmallEnoughToInline threshold) rhss)
1149 -- No specialisation
1150 = do { let (rhs_env,bndrs') = extendRecBndrs env bndrs
1151 ; (_, rhss') <- mapAndUnzipM (scExpr rhs_env) rhss
1152 ; return (rhs_env, Rec (bndrs' `zip` rhss')) }
1153 | otherwise -- Do specialisation
1154 = do { let (rhs_env1,bndrs') = extendRecBndrs env bndrs
1155 rhs_env2 = extendHowBound rhs_env1 bndrs' RecFun
1157 ; (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs rhs_env2) (bndrs' `zip` rhss)
1158 ; let rhs_usg = combineUsages rhs_usgs
1160 ; (_, specs) <- specLoop (scForce rhs_env2 force_spec)
1161 (scu_calls rhs_usg) rhs_infos nullUsage
1162 [SI [] 0 Nothing | _ <- bndrs]
1164 ; return (rhs_env1, -- For the body of the letrec, delete the RecFun business
1165 Rec (concat (zipWith specInfoBinds rhs_infos specs))) }
1167 (bndrs,rhss) = unzip prs
1168 force_spec = any (forceSpecBndr env) bndrs
1169 -- Note [Forcing specialisation]
1171 scTopBind env (NonRec bndr rhs)
1172 = do { (_, rhs') <- scExpr env rhs
1173 ; let (env1, bndr') = extendBndr env bndr
1174 env2 = extendValEnv env1 bndr' (isValue (sc_vals env) rhs')
1175 ; return (env2, NonRec bndr' rhs') }
1177 ----------------------
1178 scRecRhs :: ScEnv -> (OutId, InExpr) -> UniqSM (ScUsage, RhsInfo)
1179 scRecRhs env (bndr,rhs)
1180 = do { let (arg_bndrs,body) = collectBinders rhs
1181 (body_env, arg_bndrs') = extendBndrsWith RecArg env arg_bndrs
1182 ; (body_usg, body') <- scExpr body_env body
1183 ; let (rhs_usg, arg_occs) = lookupOccs body_usg arg_bndrs'
1184 ; return (rhs_usg, RI bndr (mkLams arg_bndrs' body')
1185 arg_bndrs body arg_occs) }
1186 -- The arg_occs says how the visible,
1187 -- lambda-bound binders of the RHS are used
1188 -- (including the TyVar binders)
1189 -- Two pats are the same if they match both ways
1191 ----------------------
1192 specInfoBinds :: RhsInfo -> SpecInfo -> [(Id,CoreExpr)]
1193 specInfoBinds (RI fn new_rhs _ _ _) (SI specs _ _)
1194 = [(id,rhs) | OS _ _ id rhs <- specs] ++
1195 [(fn `addIdSpecialisations` rules, new_rhs)]
1197 rules = [r | OS _ r _ _ <- specs]
1199 ----------------------
1200 varUsage :: ScEnv -> OutVar -> ArgOcc -> ScUsage
1202 | Just RecArg <- lookupHowBound env v = SCU { scu_calls = emptyVarEnv
1203 , scu_occs = unitVarEnv v use }
1204 | otherwise = nullUsage
1208 %************************************************************************
1210 The specialiser itself
1212 %************************************************************************
1215 data RhsInfo = RI OutId -- The binder
1216 OutExpr -- The new RHS
1217 [InVar] InExpr -- The *original* RHS (\xs.body)
1218 -- Note [Specialise original body]
1219 [ArgOcc] -- Info on how the xs occur in body
1221 data SpecInfo = SI [OneSpec] -- The specialisations we have generated
1223 Int -- Length of specs; used for numbering them
1225 (Maybe ScUsage) -- Nothing => we have generated specialisations
1226 -- from calls in the *original* RHS
1227 -- Just cs => we haven't, and this is the usage
1228 -- of the original RHS
1229 -- See Note [Local recursive groups]
1231 -- One specialisation: Rule plus definition
1232 data OneSpec = OS CallPat -- Call pattern that generated this specialisation
1233 CoreRule -- Rule connecting original id with the specialisation
1234 OutId OutExpr -- Spec id + its rhs
1240 -> ScUsage -> [SpecInfo] -- One per binder; acccumulating parameter
1241 -> UniqSM (ScUsage, [SpecInfo]) -- ...ditto...
1242 specLoop env all_calls rhs_infos usg_so_far specs_so_far
1243 = do { specs_w_usg <- zipWithM (specialise env all_calls) rhs_infos specs_so_far
1244 ; let (new_usg_s, all_specs) = unzip specs_w_usg
1245 new_usg = combineUsages new_usg_s
1246 new_calls = scu_calls new_usg
1247 all_usg = usg_so_far `combineUsage` new_usg
1248 ; if isEmptyVarEnv new_calls then
1249 return (all_usg, all_specs)
1251 specLoop env new_calls rhs_infos all_usg all_specs }
1255 -> CallEnv -- Info on calls
1257 -> SpecInfo -- Original RHS plus patterns dealt with
1258 -> UniqSM (ScUsage, SpecInfo) -- New specialised versions and their usage
1260 -- Note: the rhs here is the optimised version of the original rhs
1261 -- So when we make a specialised copy of the RHS, we're starting
1262 -- from an RHS whose nested functions have been optimised already.
1264 specialise env bind_calls (RI fn _ arg_bndrs body arg_occs)
1265 spec_info@(SI specs spec_count mb_unspec)
1266 | not (isBottomingId fn) -- Note [Do not specialise diverging functions]
1267 , not (isNeverActive (idInlineActivation fn)) -- See Note [Transfer activation]
1268 , notNull arg_bndrs -- Only specialise functions
1269 , Just all_calls <- lookupVarEnv bind_calls fn
1270 = do { (boring_call, pats) <- callsToPats env specs arg_occs all_calls
1271 -- ; pprTrace "specialise" (vcat [ ppr fn <+> text "with" <+> int (length pats) <+> text "good patterns"
1272 -- , text "arg_occs" <+> ppr arg_occs
1273 -- , text "calls" <+> ppr all_calls
1274 -- , text "good pats" <+> ppr pats]) $
1277 -- Bale out if too many specialisations
1278 ; let n_pats = length pats
1279 spec_count' = n_pats + spec_count
1280 ; case sc_count env of
1281 Just max | not (sc_force env) && spec_count' > max
1282 -> pprTrace "SpecConstr" msg $
1283 return (nullUsage, spec_info)
1285 msg = vcat [ sep [ ptext (sLit "Function") <+> quotes (ppr fn)
1286 , nest 2 (ptext (sLit "has") <+>
1287 speakNOf spec_count' (ptext (sLit "call pattern")) <> comma <+>
1288 ptext (sLit "but the limit is") <+> int max) ]
1289 , ptext (sLit "Use -fspec-constr-count=n to set the bound")
1291 extra | not opt_PprStyle_Debug = ptext (sLit "Use -dppr-debug to see specialisations")
1292 | otherwise = ptext (sLit "Specialisations:") <+> ppr (pats ++ [p | OS p _ _ _ <- specs])
1294 _normal_case -> do {
1296 let spec_env = decreaseSpecCount env n_pats
1297 ; (spec_usgs, new_specs) <- mapAndUnzipM (spec_one spec_env fn arg_bndrs body)
1298 (pats `zip` [spec_count..])
1299 -- See Note [Specialise original body]
1301 ; let spec_usg = combineUsages spec_usgs
1302 (new_usg, mb_unspec')
1304 Just rhs_usg | boring_call -> (spec_usg `combineUsage` rhs_usg, Nothing)
1305 _ -> (spec_usg, mb_unspec)
1307 ; return (new_usg, SI (new_specs ++ specs) spec_count' mb_unspec') } }
1309 = return (nullUsage, spec_info) -- The boring case
1312 ---------------------
1314 -> OutId -- Function
1315 -> [InVar] -- Lambda-binders of RHS; should match patterns
1316 -> InExpr -- Body of the original function
1318 -> UniqSM (ScUsage, OneSpec) -- Rule and binding
1320 -- spec_one creates a specialised copy of the function, together
1321 -- with a rule for using it. I'm very proud of how short this
1322 -- function is, considering what it does :-).
1328 f = /\b \y::[(a,b)] -> ....f (b,c) ((:) (a,(b,c)) (x,v) (h w))...
1329 [c::*, v::(b,c) are presumably bound by the (...) part]
1331 f_spec = /\ b c \ v::(b,c) hw::[(a,(b,c))] ->
1332 (...entire body of f...) [b -> (b,c),
1333 y -> ((:) (a,(b,c)) (x,v) hw)]
1335 RULE: forall b::* c::*, -- Note, *not* forall a, x
1339 f (b,c) ((:) (a,(b,c)) (x,v) hw) = f_spec b c v hw
1342 spec_one env fn arg_bndrs body (call_pat@(qvars, pats), rule_number)
1343 = do { spec_uniq <- getUniqueUs
1344 ; let spec_env = extendScSubstList (extendScInScope env qvars)
1345 (arg_bndrs `zip` pats)
1347 fn_loc = nameSrcSpan fn_name
1348 spec_occ = mkSpecOcc (nameOccName fn_name)
1349 rule_name = mkFastString ("SC:" ++ showSDoc (ppr fn <> int rule_number))
1350 spec_name = mkInternalName spec_uniq spec_occ fn_loc
1351 -- ; pprTrace "{spec_one" (ppr (sc_count env) <+> ppr fn <+> ppr pats <+> text "-->" <+> ppr spec_name) $
1354 -- Specialise the body
1355 ; (spec_usg, spec_body) <- scExpr spec_env body
1357 -- ; pprTrace "done spec_one}" (ppr fn) $
1360 -- And build the results
1361 ; let spec_id = mkLocalId spec_name (mkPiTypes spec_lam_args body_ty)
1362 `setIdStrictness` spec_str -- See Note [Transfer strictness]
1363 `setIdArity` count isId spec_lam_args
1364 spec_str = calcSpecStrictness fn spec_lam_args pats
1365 (spec_lam_args, spec_call_args) = mkWorkerArgs qvars body_ty
1366 -- Usual w/w hack to avoid generating
1367 -- a spec_rhs of unlifted type and no args
1369 spec_rhs = mkLams spec_lam_args spec_body
1370 body_ty = exprType spec_body
1371 rule_rhs = mkVarApps (Var spec_id) spec_call_args
1372 inline_act = idInlineActivation fn
1373 rule = mkRule True {- Auto -} True {- Local -}
1374 rule_name inline_act fn_name qvars pats rule_rhs
1375 -- See Note [Transfer activation]
1376 ; return (spec_usg, OS call_pat rule spec_id spec_rhs) }
1378 calcSpecStrictness :: Id -- The original function
1379 -> [Var] -> [CoreExpr] -- Call pattern
1380 -> StrictSig -- Strictness of specialised thing
1381 -- See Note [Transfer strictness]
1382 calcSpecStrictness fn qvars pats
1383 = StrictSig (mkTopDmdType spec_dmds TopRes)
1385 spec_dmds = [ lookupVarEnv dmd_env qv `orElse` lazyDmd | qv <- qvars, isId qv ]
1386 StrictSig (DmdType _ dmds _) = idStrictness fn
1388 dmd_env = go emptyVarEnv dmds pats
1390 go env ds (Type {} : pats) = go env ds pats
1391 go env (d:ds) (pat : pats) = go (go_one env d pat) ds pats
1394 go_one env d (Var v) = extendVarEnv_C both env v d
1395 go_one env (Box d) e = go_one env d e
1396 go_one env (Eval (Prod ds)) e
1397 | (Var _, args) <- collectArgs e = go env ds args
1398 go_one env _ _ = env
1402 Note [Specialise original body]
1403 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1404 The RhsInfo for a binding keeps the *original* body of the binding. We
1405 must specialise that, *not* the result of applying specExpr to the RHS
1406 (which is also kept in RhsInfo). Otherwise we end up specialising a
1407 specialised RHS, and that can lead directly to exponential behaviour.
1409 Note [Transfer activation]
1410 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1411 This note is for SpecConstr, but exactly the same thing
1412 happens in the overloading specialiser; see
1413 Note [Auto-specialisation and RULES] in Specialise.
1415 In which phase should the specialise-constructor rules be active?
1416 Originally I made them always-active, but Manuel found that this
1417 defeated some clever user-written rules. Then I made them active only
1418 in Phase 0; after all, currently, the specConstr transformation is
1419 only run after the simplifier has reached Phase 0, but that meant
1420 that specialisations didn't fire inside wrappers; see test
1421 simplCore/should_compile/spec-inline.
1423 So now I just use the inline-activation of the parent Id, as the
1424 activation for the specialiation RULE, just like the main specialiser;
1426 This in turn means there is no point in specialising NOINLINE things,
1427 so we test for that.
1429 Note [Transfer strictness]
1430 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1431 We must transfer strictness information from the original function to
1432 the specialised one. Suppose, for example
1435 and a RULE f (a:as) b = f_spec a as b
1437 Now we want f_spec to have strictess LLS, otherwise we'll use call-by-need
1438 when calling f_spec instead of call-by-value. And that can result in
1439 unbounded worsening in space (cf the classic foldl vs foldl')
1441 See Trac #3437 for a good example.
1443 The function calcSpecStrictness performs the calculation.
1446 %************************************************************************
1448 \subsection{Argument analysis}
1450 %************************************************************************
1452 This code deals with analysing call-site arguments to see whether
1453 they are constructor applications.
1457 type CallPat = ([Var], [CoreExpr]) -- Quantified variables and arguments
1460 callsToPats :: ScEnv -> [OneSpec] -> [ArgOcc] -> [Call] -> UniqSM (Bool, [CallPat])
1461 -- Result has no duplicate patterns,
1462 -- nor ones mentioned in done_pats
1463 -- Bool indicates that there was at least one boring pattern
1464 callsToPats env done_specs bndr_occs calls
1465 = do { mb_pats <- mapM (callToPats env bndr_occs) calls
1467 ; let good_pats :: [([Var], [CoreArg])]
1468 good_pats = catMaybes mb_pats
1469 done_pats = [p | OS p _ _ _ <- done_specs]
1470 is_done p = any (samePat p) done_pats
1472 ; return (any isNothing mb_pats,
1473 filterOut is_done (nubBy samePat good_pats)) }
1475 callToPats :: ScEnv -> [ArgOcc] -> Call -> UniqSM (Maybe CallPat)
1476 -- The [Var] is the variables to quantify over in the rule
1477 -- Type variables come first, since they may scope
1478 -- over the following term variables
1479 -- The [CoreExpr] are the argument patterns for the rule
1480 callToPats env bndr_occs (con_env, args)
1481 | length args < length bndr_occs -- Check saturated
1484 = do { let in_scope = substInScope (sc_subst env)
1485 ; prs <- argsToPats env in_scope con_env (args `zip` bndr_occs)
1486 ; let (interesting_s, pats) = unzip prs
1487 pat_fvs = varSetElems (exprsFreeVars pats)
1488 qvars = filterOut (`elemInScopeSet` in_scope) pat_fvs
1489 -- Quantify over variables that are not in sccpe
1491 -- See Note [Shadowing] at the top
1493 (tvs, ids) = partition isTyCoVar qvars
1495 -- Put the type variables first; the type of a term
1496 -- variable may mention a type variable
1498 ; -- pprTrace "callToPats" (ppr args $$ ppr prs $$ ppr bndr_occs) $
1500 then return (Just (qvars', pats))
1501 else return Nothing }
1503 -- argToPat takes an actual argument, and returns an abstracted
1504 -- version, consisting of just the "constructor skeleton" of the
1505 -- argument, with non-constructor sub-expression replaced by new
1506 -- placeholder variables. For example:
1507 -- C a (D (f x) (g y)) ==> C p1 (D p2 p3)
1510 -> InScopeSet -- What's in scope at the fn defn site
1511 -> ValueEnv -- ValueEnv at the call site
1512 -> CoreArg -- A call arg (or component thereof)
1514 -> UniqSM (Bool, CoreArg)
1515 -- Returns (interesting, pat),
1516 -- where pat is the pattern derived from the argument
1517 -- intersting=True if the pattern is non-trivial (not a variable or type)
1518 -- E.g. x:xs --> (True, x:xs)
1519 -- f xs --> (False, w) where w is a fresh wildcard
1520 -- (f xs, 'c') --> (True, (w, 'c')) where w is a fresh wildcard
1521 -- \x. x+y --> (True, \x. x+y)
1522 -- lvl7 --> (True, lvl7) if lvl7 is bound
1523 -- somewhere further out
1525 argToPat _env _in_scope _val_env arg@(Type {}) _arg_occ
1526 = return (False, arg)
1528 argToPat env in_scope val_env (Note _ arg) arg_occ
1529 = argToPat env in_scope val_env arg arg_occ
1530 -- Note [Notes in call patterns]
1531 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1532 -- Ignore Notes. In particular, we want to ignore any InlineMe notes
1533 -- Perhaps we should not ignore profiling notes, but I'm going to
1534 -- ride roughshod over them all for now.
1535 --- See Note [Notes in RULE matching] in Rules
1537 argToPat env in_scope val_env (Let _ arg) arg_occ
1538 = argToPat env in_scope val_env arg arg_occ
1539 -- See Note [Matching lets] in Rule.lhs
1540 -- Look through let expressions
1541 -- e.g. f (let v = rhs in (v,w))
1542 -- Here we can specialise for f (v,w)
1543 -- because the rule-matcher will look through the let.
1545 {- Disabled; see Note [Matching cases] in Rule.lhs
1546 argToPat env in_scope val_env (Case scrut _ _ [(_, _, rhs)]) arg_occ
1547 | exprOkForSpeculation scrut -- See Note [Matching cases] in Rule.hhs
1548 = argToPat env in_scope val_env rhs arg_occ
1551 argToPat env in_scope val_env (Cast arg co) arg_occ
1552 | not (ignoreType env ty2)
1553 = do { (interesting, arg') <- argToPat env in_scope val_env arg arg_occ
1554 ; if not interesting then
1557 { -- Make a wild-card pattern for the coercion
1559 ; let co_name = mkSysTvName uniq (fsLit "sg")
1560 co_var = mkCoVar co_name (mkCoKind ty1 ty2)
1561 ; return (interesting, Cast arg' (mkTyVarTy co_var)) } }
1563 (ty1, ty2) = coercionKind co
1567 {- Disabling lambda specialisation for now
1568 It's fragile, and the spec_loop can be infinite
1569 argToPat in_scope val_env arg arg_occ
1571 = return (True, arg)
1573 is_value_lam (Lam v e) -- Spot a value lambda, even if
1574 | isId v = True -- it is inside a type lambda
1575 | otherwise = is_value_lam e
1576 is_value_lam other = False
1579 -- Check for a constructor application
1580 -- NB: this *precedes* the Var case, so that we catch nullary constrs
1581 argToPat env in_scope val_env arg arg_occ
1582 | Just (ConVal dc args) <- isValue val_env arg
1583 , not (ignoreAltCon env dc) -- See Note [NoSpecConstr]
1584 , sc_force env || scrutinised
1585 = do { args' <- argsToPats env in_scope val_env (args `zip` conArgOccs arg_occ dc)
1586 ; return (True, mk_con_app dc (map snd args')) }
1590 ScrutOcc _ -> True -- Used only by case scrutinee
1591 BothOcc -> case arg of -- Used elsewhere
1592 App {} -> True -- see Note [Reboxing]
1594 _other -> False -- No point; the arg is not decomposed
1597 -- Check if the argument is a variable that
1598 -- is in scope at the function definition site
1599 -- It's worth specialising on this if
1600 -- (a) it's used in an interesting way in the body
1601 -- (b) we know what its value is
1602 argToPat env in_scope val_env (Var v) arg_occ
1603 | sc_force env || case arg_occ of { UnkOcc -> False; _other -> True }, -- (a)
1605 not (ignoreType env (varType v))
1606 = return (True, Var v)
1609 | isLocalId v = v `elemInScopeSet` in_scope
1610 && isJust (lookupVarEnv val_env v)
1611 -- Local variables have values in val_env
1612 | otherwise = isValueUnfolding (idUnfolding v)
1613 -- Imports have unfoldings
1615 -- I'm really not sure what this comment means
1616 -- And by not wild-carding we tend to get forall'd
1617 -- variables that are in soope, which in turn can
1618 -- expose the weakness in let-matching
1619 -- See Note [Matching lets] in Rules
1621 -- Check for a variable bound inside the function.
1622 -- Don't make a wild-card, because we may usefully share
1623 -- e.g. f a = let x = ... in f (x,x)
1624 -- NB: this case follows the lambda and con-app cases!!
1625 -- argToPat _in_scope _val_env (Var v) _arg_occ
1626 -- = return (False, Var v)
1627 -- SLPJ : disabling this to avoid proliferation of versions
1628 -- also works badly when thinking about seeding the loop
1629 -- from the body of the let
1630 -- f x y = letrec g z = ... in g (x,y)
1631 -- We don't want to specialise for that *particular* x,y
1633 -- The default case: make a wild-card
1634 argToPat _env _in_scope _val_env arg _arg_occ
1635 = wildCardPat (exprType arg)
1637 wildCardPat :: Type -> UniqSM (Bool, CoreArg)
1638 wildCardPat ty = do { uniq <- getUniqueUs
1639 ; let id = mkSysLocal (fsLit "sc") uniq ty
1640 ; return (False, Var id) }
1642 argsToPats :: ScEnv -> InScopeSet -> ValueEnv
1643 -> [(CoreArg, ArgOcc)]
1644 -> UniqSM [(Bool, CoreArg)]
1645 argsToPats env in_scope val_env args
1648 do_one (arg,occ) = argToPat env in_scope val_env arg occ
1653 isValue :: ValueEnv -> CoreExpr -> Maybe Value
1654 isValue _env (Lit lit)
1655 = Just (ConVal (LitAlt lit) [])
1658 | Just stuff <- lookupVarEnv env v
1659 = Just stuff -- You might think we could look in the idUnfolding here
1660 -- but that doesn't take account of which branch of a
1661 -- case we are in, which is the whole point
1663 | not (isLocalId v) && isCheapUnfolding unf
1664 = isValue env (unfoldingTemplate unf)
1667 -- However we do want to consult the unfolding
1668 -- as well, for let-bound constructors!
1670 isValue env (Lam b e)
1671 | isTyCoVar b = case isValue env e of
1672 Just _ -> Just LambdaVal
1674 | otherwise = Just LambdaVal
1676 isValue _env expr -- Maybe it's a constructor application
1677 | (Var fun, args) <- collectArgs expr
1678 = case isDataConWorkId_maybe fun of
1680 Just con | args `lengthAtLeast` dataConRepArity con
1681 -- Check saturated; might be > because the
1682 -- arity excludes type args
1683 -> Just (ConVal (DataAlt con) args)
1685 _other | valArgCount args < idArity fun
1686 -- Under-applied function
1687 -> Just LambdaVal -- Partial application
1691 isValue _env _expr = Nothing
1693 mk_con_app :: AltCon -> [CoreArg] -> CoreExpr
1694 mk_con_app (LitAlt lit) [] = Lit lit
1695 mk_con_app (DataAlt con) args = mkConApp con args
1696 mk_con_app _other _args = panic "SpecConstr.mk_con_app"
1698 samePat :: CallPat -> CallPat -> Bool
1699 samePat (vs1, as1) (vs2, as2)
1702 same (Var v1) (Var v2)
1703 | v1 `elem` vs1 = v2 `elem` vs2
1704 | v2 `elem` vs2 = False
1705 | otherwise = v1 == v2
1707 same (Lit l1) (Lit l2) = l1==l2
1708 same (App f1 a1) (App f2 a2) = same f1 f2 && same a1 a2
1710 same (Type {}) (Type {}) = True -- Note [Ignore type differences]
1711 same (Note _ e1) e2 = same e1 e2 -- Ignore casts and notes
1712 same (Cast e1 _) e2 = same e1 e2
1713 same e1 (Note _ e2) = same e1 e2
1714 same e1 (Cast e2 _) = same e1 e2
1716 same e1 e2 = WARN( bad e1 || bad e2, ppr e1 $$ ppr e2)
1717 False -- Let, lambda, case should not occur
1718 bad (Case {}) = True
1724 Note [Ignore type differences]
1725 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1726 We do not want to generate specialisations where the call patterns
1727 differ only in their type arguments! Not only is it utterly useless,
1728 but it also means that (with polymorphic recursion) we can generate
1729 an infinite number of specialisations. Example is Data.Sequence.adjustTree,