2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section[SpecConstr]{Specialise over constructors}
7 -- The above warning supression flag is a temporary kludge.
8 -- While working on this module you are encouraged to remove it and fix
9 -- any warnings in the module. See
10 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
17 #include "HsVersions.h"
22 import CoreUnfold ( couldBeSmallEnoughToInline )
23 import CoreFVs ( exprsFreeVars )
24 import WwLib ( mkWorkerArgs )
25 import DataCon ( dataConRepArity, dataConUnivTyVars )
28 import Type hiding( substTy )
30 import MkId ( mkImpossibleExpr )
35 import DynFlags ( DynFlags(..) )
36 import StaticFlags ( opt_PprStyle_Debug )
37 import StaticFlags ( opt_SpecInlineJoinPoints )
38 import BasicTypes ( Activation(..) )
39 import Maybes ( orElse, catMaybes, isJust, isNothing )
41 import DmdAnal ( both )
48 import Control.Monad ( zipWithM )
52 -----------------------------------------------------
54 -----------------------------------------------------
59 drop n (x:xs) = drop (n-1) xs
61 After the first time round, we could pass n unboxed. This happens in
62 numerical code too. Here's what it looks like in Core:
64 drop n xs = case xs of
69 _ -> drop (I# (n# -# 1#)) xs
71 Notice that the recursive call has an explicit constructor as argument.
72 Noticing this, we can make a specialised version of drop
74 RULE: drop (I# n#) xs ==> drop' n# xs
76 drop' n# xs = let n = I# n# in ...orig RHS...
78 Now the simplifier will apply the specialisation in the rhs of drop', giving
80 drop' n# xs = case xs of
84 _ -> drop (n# -# 1#) xs
88 We'd also like to catch cases where a parameter is carried along unchanged,
89 but evaluated each time round the loop:
91 f i n = if i>0 || i>n then i else f (i*2) n
93 Here f isn't strict in n, but we'd like to avoid evaluating it each iteration.
94 In Core, by the time we've w/wd (f is strict in i) we get
96 f i# n = case i# ># 0 of
98 True -> case n of n' { I# n# ->
101 True -> f (i# *# 2#) n'
103 At the call to f, we see that the argument, n is know to be (I# n#),
104 and n is evaluated elsewhere in the body of f, so we can play the same
110 We must be careful not to allocate the same constructor twice. Consider
111 f p = (...(case p of (a,b) -> e)...p...,
112 ...let t = (r,s) in ...t...(f t)...)
113 At the recursive call to f, we can see that t is a pair. But we do NOT want
114 to make a specialised copy:
115 f' a b = let p = (a,b) in (..., ...)
116 because now t is allocated by the caller, then r and s are passed to the
117 recursive call, which allocates the (r,s) pair again.
120 (a) the argument p is used in other than a case-scrutinsation way.
121 (b) the argument to the call is not a 'fresh' tuple; you have to
122 look into its unfolding to see that it's a tuple
124 Hence the "OR" part of Note [Good arguments] below.
126 ALTERNATIVE 2: pass both boxed and unboxed versions. This no longer saves
127 allocation, but does perhaps save evals. In the RULE we'd have
130 f (I# x#) = f' (I# x#) x#
132 If at the call site the (I# x) was an unfolding, then we'd have to
133 rely on CSE to eliminate the duplicate allocation.... This alternative
134 doesn't look attractive enough to pursue.
136 ALTERNATIVE 3: ignore the reboxing problem. The trouble is that
137 the conservative reboxing story prevents many useful functions from being
138 specialised. Example:
139 foo :: Maybe Int -> Int -> Int
141 foo x@(Just m) n = foo x (n-m)
142 Here the use of 'x' will clearly not require boxing in the specialised function.
144 The strictness analyser has the same problem, in fact. Example:
146 If we pass just 'a' and 'b' to the worker, it might need to rebox the
147 pair to create (a,b). A more sophisticated analysis might figure out
148 precisely the cases in which this could happen, but the strictness
149 analyser does no such analysis; it just passes 'a' and 'b', and hopes
152 So my current choice is to make SpecConstr similarly aggressive, and
153 ignore the bad potential of reboxing.
156 Note [Good arguments]
157 ~~~~~~~~~~~~~~~~~~~~~
160 * A self-recursive function. Ignore mutual recursion for now,
161 because it's less common, and the code is simpler for self-recursion.
165 a) At a recursive call, one or more parameters is an explicit
166 constructor application
168 That same parameter is scrutinised by a case somewhere in
169 the RHS of the function
173 b) At a recursive call, one or more parameters has an unfolding
174 that is an explicit constructor application
176 That same parameter is scrutinised by a case somewhere in
177 the RHS of the function
179 Those are the only uses of the parameter (see Note [Reboxing])
182 What to abstract over
183 ~~~~~~~~~~~~~~~~~~~~~
184 There's a bit of a complication with type arguments. If the call
187 f p = ...f ((:) [a] x xs)...
189 then our specialised function look like
191 f_spec x xs = let p = (:) [a] x xs in ....as before....
193 This only makes sense if either
194 a) the type variable 'a' is in scope at the top of f, or
195 b) the type variable 'a' is an argument to f (and hence fs)
197 Actually, (a) may hold for value arguments too, in which case
198 we may not want to pass them. Supose 'x' is in scope at f's
199 defn, but xs is not. Then we'd like
201 f_spec xs = let p = (:) [a] x xs in ....as before....
203 Similarly (b) may hold too. If x is already an argument at the
204 call, no need to pass it again.
206 Finally, if 'a' is not in scope at the call site, we could abstract
207 it as we do the term variables:
209 f_spec a x xs = let p = (:) [a] x xs in ...as before...
211 So the grand plan is:
213 * abstract the call site to a constructor-only pattern
214 e.g. C x (D (f p) (g q)) ==> C s1 (D s2 s3)
216 * Find the free variables of the abstracted pattern
218 * Pass these variables, less any that are in scope at
219 the fn defn. But see Note [Shadowing] below.
222 NOTICE that we only abstract over variables that are not in scope,
223 so we're in no danger of shadowing variables used in "higher up"
229 In this pass we gather up usage information that may mention variables
230 that are bound between the usage site and the definition site; or (more
231 seriously) may be bound to something different at the definition site.
234 f x = letrec g y v = let x = ...
237 Since 'x' is in scope at the call site, we may make a rewrite rule that
239 RULE forall a,b. g (a,b) x = ...
240 But this rule will never match, because it's really a different 'x' at
241 the call site -- and that difference will be manifest by the time the
242 simplifier gets to it. [A worry: the simplifier doesn't *guarantee*
243 no-shadowing, so perhaps it may not be distinct?]
245 Anyway, the rule isn't actually wrong, it's just not useful. One possibility
246 is to run deShadowBinds before running SpecConstr, but instead we run the
247 simplifier. That gives the simplest possible program for SpecConstr to
248 chew on; and it virtually guarantees no shadowing.
250 Note [Specialising for constant parameters]
251 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
252 This one is about specialising on a *constant* (but not necessarily
253 constructor) argument
255 foo :: Int -> (Int -> Int) -> Int
257 foo m f = foo (f m) (+1)
261 lvl_rmV :: GHC.Base.Int -> GHC.Base.Int
263 \ (ds_dlk :: GHC.Base.Int) ->
264 case ds_dlk of wild_alH { GHC.Base.I# x_alG ->
265 GHC.Base.I# (GHC.Prim.+# x_alG 1)
267 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
270 \ (ww_sme :: GHC.Prim.Int#) (w_smg :: GHC.Base.Int -> GHC.Base.Int) ->
271 case ww_sme of ds_Xlw {
273 case w_smg (GHC.Base.I# ds_Xlw) of w1_Xmo { GHC.Base.I# ww1_Xmz ->
274 T.$wfoo ww1_Xmz lvl_rmV
279 The recursive call has lvl_rmV as its argument, so we could create a specialised copy
280 with that argument baked in; that is, not passed at all. Now it can perhaps be inlined.
282 When is this worth it? Call the constant 'lvl'
283 - If 'lvl' has an unfolding that is a constructor, see if the corresponding
284 parameter is scrutinised anywhere in the body.
286 - If 'lvl' has an unfolding that is a inlinable function, see if the corresponding
287 parameter is applied (...to enough arguments...?)
289 Also do this is if the function has RULES?
293 Note [Specialising for lambda parameters]
294 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
295 foo :: Int -> (Int -> Int) -> Int
297 foo m f = foo (f m) (\n -> n-m)
299 This is subtly different from the previous one in that we get an
300 explicit lambda as the argument:
302 T.$wfoo :: GHC.Prim.Int# -> (GHC.Base.Int -> GHC.Base.Int) ->
305 \ (ww_sm8 :: GHC.Prim.Int#) (w_sma :: GHC.Base.Int -> GHC.Base.Int) ->
306 case ww_sm8 of ds_Xlr {
308 case w_sma (GHC.Base.I# ds_Xlr) of w1_Xmf { GHC.Base.I# ww1_Xmq ->
311 (\ (n_ad3 :: GHC.Base.Int) ->
312 case n_ad3 of wild_alB { GHC.Base.I# x_alA ->
313 GHC.Base.I# (GHC.Prim.-# x_alA ds_Xlr)
319 I wonder if SpecConstr couldn't be extended to handle this? After all,
320 lambda is a sort of constructor for functions and perhaps it already
321 has most of the necessary machinery?
323 Furthermore, there's an immediate win, because you don't need to allocate the lamda
324 at the call site; and if perchance it's called in the recursive call, then you
325 may avoid allocating it altogether. Just like for constructors.
327 Looks cool, but probably rare...but it might be easy to implement.
330 Note [SpecConstr for casts]
331 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
334 data instance T Int = T Int
339 go (T n) = go (T (n-1))
341 The recursive call ends up looking like
342 go (T (I# ...) `cast` g)
343 So we want to spot the construtor application inside the cast.
344 That's why we have the Cast case in argToPat
346 Note [Local recursive groups]
347 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
348 For a *local* recursive group, we can see all the calls to the
349 function, so we seed the specialisation loop from the calls in the
350 body, not from the calls in the RHS. Consider:
352 bar m n = foo n (n,n) (n,n) (n,n) (n,n)
356 | n > 3000 = case p of { (p1,p2) -> foo (n-1) (p2,p1) q r s }
357 | n > 2000 = case q of { (q1,q2) -> foo (n-1) p (q2,q1) r s }
358 | n > 1000 = case r of { (r1,r2) -> foo (n-1) p q (r2,r1) s }
359 | otherwise = case s of { (s1,s2) -> foo (n-1) p q r (s2,s1) }
361 If we start with the RHSs of 'foo', we get lots and lots of specialisations,
362 most of which are not needed. But if we start with the (single) call
363 in the rhs of 'bar' we get exactly one fully-specialised copy, and all
364 the recursive calls go to this fully-specialised copy. Indeed, the original
365 function is later collected as dead code. This is very important in
366 specialising the loops arising from stream fusion, for example in NDP where
367 we were getting literally hundreds of (mostly unused) specialisations of
370 Note [Do not specialise diverging functions]
371 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
372 Specialising a function that just diverges is a waste of code.
373 Furthermore, it broke GHC (simpl014) thus:
375 f = \x. case x of (a,b) -> f x
376 If we specialise f we get
377 f = \x. case x of (a,b) -> fspec a b
378 But fspec doesn't have decent strictnes info. As it happened,
379 (f x) :: IO t, so the state hack applied and we eta expanded fspec,
380 and hence f. But now f's strictness is less than its arity, which
383 -----------------------------------------------------
384 Stuff not yet handled
385 -----------------------------------------------------
387 Here are notes arising from Roman's work that I don't want to lose.
393 foo :: Int -> T Int -> Int
395 foo x t | even x = case t of { T n -> foo (x-n) t }
396 | otherwise = foo (x-1) t
398 SpecConstr does no specialisation, because the second recursive call
399 looks like a boxed use of the argument. A pity.
401 $wfoo_sFw :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
403 \ (ww_sFo [Just L] :: GHC.Prim.Int#) (w_sFq [Just L] :: T.T GHC.Base.Int) ->
404 case ww_sFo of ds_Xw6 [Just L] {
406 case GHC.Prim.remInt# ds_Xw6 2 of wild1_aEF [Dead Just A] {
407 __DEFAULT -> $wfoo_sFw (GHC.Prim.-# ds_Xw6 1) w_sFq;
409 case w_sFq of wild_Xy [Just L] { T.T n_ad5 [Just U(L)] ->
410 case n_ad5 of wild1_aET [Just A] { GHC.Base.I# y_aES [Just L] ->
411 $wfoo_sFw (GHC.Prim.-# ds_Xw6 y_aES) wild_Xy
417 data a :*: b = !a :*: !b
420 foo :: (Int :*: T Int) -> Int
422 foo (x :*: t) | even x = case t of { T n -> foo ((x-n) :*: t) }
423 | otherwise = foo ((x-1) :*: t)
425 Very similar to the previous one, except that the parameters are now in
426 a strict tuple. Before SpecConstr, we have
428 $wfoo_sG3 :: GHC.Prim.Int# -> T.T GHC.Base.Int -> GHC.Prim.Int#
430 \ (ww_sFU [Just L] :: GHC.Prim.Int#) (ww_sFW [Just L] :: T.T
432 case ww_sFU of ds_Xws [Just L] {
434 case GHC.Prim.remInt# ds_Xws 2 of wild1_aEZ [Dead Just A] {
436 case ww_sFW of tpl_B2 [Just L] { T.T a_sFo [Just A] ->
437 $wfoo_sG3 (GHC.Prim.-# ds_Xws 1) tpl_B2 -- $wfoo1
440 case ww_sFW of wild_XB [Just A] { T.T n_ad7 [Just S(L)] ->
441 case n_ad7 of wild1_aFd [Just L] { GHC.Base.I# y_aFc [Just L] ->
442 $wfoo_sG3 (GHC.Prim.-# ds_Xws y_aFc) wild_XB -- $wfoo2
446 We get two specialisations:
447 "SC:$wfoo1" [0] __forall {a_sFB :: GHC.Base.Int sc_sGC :: GHC.Prim.Int#}
448 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int a_sFB)
449 = Foo.$s$wfoo1 a_sFB sc_sGC ;
450 "SC:$wfoo2" [0] __forall {y_aFp :: GHC.Prim.Int# sc_sGC :: GHC.Prim.Int#}
451 Foo.$wfoo sc_sGC (Foo.T @ GHC.Base.Int (GHC.Base.I# y_aFp))
452 = Foo.$s$wfoo y_aFp sc_sGC ;
454 But perhaps the first one isn't good. After all, we know that tpl_B2 is
455 a T (I# x) really, because T is strict and Int has one constructor. (We can't
456 unbox the strict fields, becuase T is polymorphic!)
460 %************************************************************************
462 \subsection{Top level wrapper stuff}
464 %************************************************************************
467 specConstrProgram :: DynFlags -> UniqSupply -> [CoreBind] -> [CoreBind]
468 specConstrProgram dflags us binds = fst $ initUs us (go (initScEnv dflags) binds)
471 go env (bind:binds) = do (env', bind') <- scTopBind env bind
472 binds' <- go env' binds
473 return (bind' : binds')
477 %************************************************************************
479 \subsection{Environment: goes downwards}
481 %************************************************************************
484 data ScEnv = SCE { sc_size :: Maybe Int, -- Size threshold
485 sc_count :: Maybe Int, -- Max # of specialisations for any one fn
487 sc_subst :: Subst, -- Current substitution
488 -- Maps InIds to OutExprs
490 sc_how_bound :: HowBoundEnv,
491 -- Binds interesting non-top-level variables
492 -- Domain is OutVars (*after* applying the substitution)
495 -- Domain is OutIds (*after* applying the substitution)
496 -- Used even for top-level bindings (but not imported ones)
499 ---------------------
500 -- As we go, we apply a substitution (sc_subst) to the current term
501 type InExpr = CoreExpr -- _Before_ applying the subst
503 type OutExpr = CoreExpr -- _After_ applying the subst
507 ---------------------
508 type HowBoundEnv = VarEnv HowBound -- Domain is OutVars
510 ---------------------
511 type ValueEnv = IdEnv Value -- Domain is OutIds
512 data Value = ConVal AltCon [CoreArg] -- _Saturated_ constructors
513 | LambdaVal -- Inlinable lambdas or PAPs
515 instance Outputable Value where
516 ppr (ConVal con args) = ppr con <+> interpp'SP args
517 ppr LambdaVal = ptext (sLit "<Lambda>")
519 ---------------------
520 initScEnv :: DynFlags -> ScEnv
522 = SCE { sc_size = specConstrThreshold dflags,
523 sc_count = specConstrCount dflags,
524 sc_subst = emptySubst,
525 sc_how_bound = emptyVarEnv,
526 sc_vals = emptyVarEnv }
528 data HowBound = RecFun -- These are the recursive functions for which
529 -- we seek interesting call patterns
531 | RecArg -- These are those functions' arguments, or their sub-components;
532 -- we gather occurrence information for these
534 instance Outputable HowBound where
535 ppr RecFun = text "RecFun"
536 ppr RecArg = text "RecArg"
538 lookupHowBound :: ScEnv -> Id -> Maybe HowBound
539 lookupHowBound env id = lookupVarEnv (sc_how_bound env) id
541 scSubstId :: ScEnv -> Id -> CoreExpr
542 scSubstId env v = lookupIdSubst (sc_subst env) v
544 scSubstTy :: ScEnv -> Type -> Type
545 scSubstTy env ty = substTy (sc_subst env) ty
547 zapScSubst :: ScEnv -> ScEnv
548 zapScSubst env = env { sc_subst = zapSubstEnv (sc_subst env) }
550 extendScInScope :: ScEnv -> [Var] -> ScEnv
551 -- Bring the quantified variables into scope
552 extendScInScope env qvars = env { sc_subst = extendInScopeList (sc_subst env) qvars }
554 -- Extend the substitution
555 extendScSubst :: ScEnv -> Var -> OutExpr -> ScEnv
556 extendScSubst env var expr = env { sc_subst = extendSubst (sc_subst env) var expr }
558 extendScSubstList :: ScEnv -> [(Var,OutExpr)] -> ScEnv
559 extendScSubstList env prs = env { sc_subst = extendSubstList (sc_subst env) prs }
561 extendHowBound :: ScEnv -> [Var] -> HowBound -> ScEnv
562 extendHowBound env bndrs how_bound
563 = env { sc_how_bound = extendVarEnvList (sc_how_bound env)
564 [(bndr,how_bound) | bndr <- bndrs] }
566 extendBndrsWith :: HowBound -> ScEnv -> [Var] -> (ScEnv, [Var])
567 extendBndrsWith how_bound env bndrs
568 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndrs')
570 (subst', bndrs') = substBndrs (sc_subst env) bndrs
571 hb_env' = sc_how_bound env `extendVarEnvList`
572 [(bndr,how_bound) | bndr <- bndrs']
574 extendBndrWith :: HowBound -> ScEnv -> Var -> (ScEnv, Var)
575 extendBndrWith how_bound env bndr
576 = (env { sc_subst = subst', sc_how_bound = hb_env' }, bndr')
578 (subst', bndr') = substBndr (sc_subst env) bndr
579 hb_env' = extendVarEnv (sc_how_bound env) bndr' how_bound
581 extendRecBndrs :: ScEnv -> [Var] -> (ScEnv, [Var])
582 extendRecBndrs env bndrs = (env { sc_subst = subst' }, bndrs')
584 (subst', bndrs') = substRecBndrs (sc_subst env) bndrs
586 extendBndr :: ScEnv -> Var -> (ScEnv, Var)
587 extendBndr env bndr = (env { sc_subst = subst' }, bndr')
589 (subst', bndr') = substBndr (sc_subst env) bndr
591 extendValEnv :: ScEnv -> Id -> Maybe Value -> ScEnv
592 extendValEnv env _ Nothing = env
593 extendValEnv env id (Just cv) = env { sc_vals = extendVarEnv (sc_vals env) id cv }
595 extendCaseBndrs :: ScEnv -> Id -> AltCon -> [Var] -> (ScEnv, [Var])
599 -- we want to bind b, to (C x y)
600 -- NB1: Extends only the sc_vals part of the envt
601 -- NB2: Kill the dead-ness info on the pattern binders x,y, since
602 -- they are potentially made alive by the [b -> C x y] binding
603 extendCaseBndrs env case_bndr con alt_bndrs
604 | isDeadBinder case_bndr
607 = (env1, map zap alt_bndrs)
608 -- NB: We used to bind v too, if scrut = (Var v); but
609 -- the simplifer has already done this so it seems
610 -- redundant to do so here
612 -- Var v -> extendValEnv env1 v cval
615 zap v | isTyVar v = v -- See NB2 above
616 | otherwise = zapIdOccInfo v
617 env1 = extendValEnv env case_bndr cval
620 LitAlt {} -> Just (ConVal con [])
621 DataAlt {} -> Just (ConVal con vanilla_args)
623 vanilla_args = map Type (tyConAppArgs (idType case_bndr)) ++
624 varsToCoreExprs alt_bndrs
628 %************************************************************************
630 \subsection{Usage information: flows upwards}
632 %************************************************************************
637 scu_calls :: CallEnv, -- Calls
638 -- The functions are a subset of the
639 -- RecFuns in the ScEnv
641 scu_occs :: !(IdEnv ArgOcc) -- Information on argument occurrences
642 } -- The domain is OutIds
644 type CallEnv = IdEnv [Call]
645 type Call = (ValueEnv, [CoreArg])
646 -- The arguments of the call, together with the
647 -- env giving the constructor bindings at the call site
650 nullUsage = SCU { scu_calls = emptyVarEnv, scu_occs = emptyVarEnv }
652 combineCalls :: CallEnv -> CallEnv -> CallEnv
653 combineCalls = plusVarEnv_C (++)
655 combineUsage :: ScUsage -> ScUsage -> ScUsage
656 combineUsage u1 u2 = SCU { scu_calls = combineCalls (scu_calls u1) (scu_calls u2),
657 scu_occs = plusVarEnv_C combineOcc (scu_occs u1) (scu_occs u2) }
659 combineUsages :: [ScUsage] -> ScUsage
660 combineUsages [] = nullUsage
661 combineUsages us = foldr1 combineUsage us
663 lookupOcc :: ScUsage -> OutVar -> (ScUsage, ArgOcc)
664 lookupOcc (SCU { scu_calls = sc_calls, scu_occs = sc_occs }) bndr
665 = (SCU {scu_calls = sc_calls, scu_occs = delVarEnv sc_occs bndr},
666 lookupVarEnv sc_occs bndr `orElse` NoOcc)
668 lookupOccs :: ScUsage -> [OutVar] -> (ScUsage, [ArgOcc])
669 lookupOccs (SCU { scu_calls = sc_calls, scu_occs = sc_occs }) bndrs
670 = (SCU {scu_calls = sc_calls, scu_occs = delVarEnvList sc_occs bndrs},
671 [lookupVarEnv sc_occs b `orElse` NoOcc | b <- bndrs])
673 data ArgOcc = NoOcc -- Doesn't occur at all; or a type argument
674 | UnkOcc -- Used in some unknown way
676 | ScrutOcc (UniqFM [ArgOcc]) -- See Note [ScrutOcc]
678 | BothOcc -- Definitely taken apart, *and* perhaps used in some other way
682 An occurrence of ScrutOcc indicates that the thing, or a `cast` version of the thing,
683 is *only* taken apart or applied.
685 Functions, literal: ScrutOcc emptyUFM
686 Data constructors: ScrutOcc subs,
688 where (subs :: UniqFM [ArgOcc]) gives usage of the *pattern-bound* components,
689 The domain of the UniqFM is the Unique of the data constructor
691 The [ArgOcc] is the occurrences of the *pattern-bound* components
692 of the data structure. E.g.
693 data T a = forall b. MkT a b (b->a)
694 A pattern binds b, x::a, y::b, z::b->a, but not 'a'!
698 instance Outputable ArgOcc where
699 ppr (ScrutOcc xs) = ptext (sLit "scrut-occ") <> ppr xs
700 ppr UnkOcc = ptext (sLit "unk-occ")
701 ppr BothOcc = ptext (sLit "both-occ")
702 ppr NoOcc = ptext (sLit "no-occ")
704 -- Experimentally, this vesion of combineOcc makes ScrutOcc "win", so
705 -- that if the thing is scrutinised anywhere then we get to see that
706 -- in the overall result, even if it's also used in a boxed way
707 -- This might be too agressive; see Note [Reboxing] Alternative 3
708 combineOcc :: ArgOcc -> ArgOcc -> ArgOcc
709 combineOcc NoOcc occ = occ
710 combineOcc occ NoOcc = occ
711 combineOcc (ScrutOcc xs) (ScrutOcc ys) = ScrutOcc (plusUFM_C combineOccs xs ys)
712 combineOcc _occ (ScrutOcc ys) = ScrutOcc ys
713 combineOcc (ScrutOcc xs) _occ = ScrutOcc xs
714 combineOcc UnkOcc UnkOcc = UnkOcc
715 combineOcc _ _ = BothOcc
717 combineOccs :: [ArgOcc] -> [ArgOcc] -> [ArgOcc]
718 combineOccs xs ys = zipWithEqual "combineOccs" combineOcc xs ys
720 setScrutOcc :: ScEnv -> ScUsage -> OutExpr -> ArgOcc -> ScUsage
721 -- _Overwrite_ the occurrence info for the scrutinee, if the scrutinee
722 -- is a variable, and an interesting variable
723 setScrutOcc env usg (Cast e _) occ = setScrutOcc env usg e occ
724 setScrutOcc env usg (Note _ e) occ = setScrutOcc env usg e occ
725 setScrutOcc env usg (Var v) occ
726 | Just RecArg <- lookupHowBound env v = usg { scu_occs = extendVarEnv (scu_occs usg) v occ }
728 setScrutOcc _env usg _other _occ -- Catch-all
731 conArgOccs :: ArgOcc -> AltCon -> [ArgOcc]
732 -- Find usage of components of data con; returns [UnkOcc...] if unknown
733 -- See Note [ScrutOcc] for the extra UnkOccs in the vanilla datacon case
735 conArgOccs (ScrutOcc fm) (DataAlt dc)
736 | Just pat_arg_occs <- lookupUFM fm dc
737 = [UnkOcc | _ <- dataConUnivTyVars dc] ++ pat_arg_occs
739 conArgOccs _other _con = repeat UnkOcc
742 %************************************************************************
744 \subsection{The main recursive function}
746 %************************************************************************
748 The main recursive function gathers up usage information, and
749 creates specialised versions of functions.
752 scExpr, scExpr' :: ScEnv -> CoreExpr -> UniqSM (ScUsage, CoreExpr)
753 -- The unique supply is needed when we invent
754 -- a new name for the specialised function and its args
756 scExpr env e = scExpr' env e
759 scExpr' env (Var v) = case scSubstId env v of
760 Var v' -> return (varUsage env v' UnkOcc, Var v')
761 e' -> scExpr (zapScSubst env) e'
763 scExpr' env (Type t) = return (nullUsage, Type (scSubstTy env t))
764 scExpr' _ e@(Lit {}) = return (nullUsage, e)
765 scExpr' env (Note n e) = do (usg,e') <- scExpr env e
766 return (usg, Note n e')
767 scExpr' env (Cast e co) = do (usg, e') <- scExpr env e
768 return (usg, Cast e' (scSubstTy env co))
769 scExpr' env e@(App _ _) = scApp env (collectArgs e)
770 scExpr' env (Lam b e) = do let (env', b') = extendBndr env b
771 (usg, e') <- scExpr env' e
772 return (usg, Lam b' e')
774 scExpr' env (Case scrut b ty alts)
775 = do { (scrut_usg, scrut') <- scExpr env scrut
776 ; case isValue (sc_vals env) scrut' of
777 Just (ConVal con args) -> sc_con_app con args scrut'
778 _other -> sc_vanilla scrut_usg scrut'
781 sc_con_app con args scrut' -- Known constructor; simplify
782 = do { let (_, bs, rhs) = findAlt con alts
783 `orElse` (DEFAULT, [], mkImpossibleExpr (coreAltsType alts))
784 alt_env' = extendScSubstList env ((b,scrut') : bs `zip` trimConArgs con args)
785 ; scExpr alt_env' rhs }
787 sc_vanilla scrut_usg scrut' -- Normal case
788 = do { let (alt_env,b') = extendBndrWith RecArg env b
789 -- Record RecArg for the components
791 ; (alt_usgs, alt_occs, alts')
792 <- mapAndUnzip3M (sc_alt alt_env scrut' b') alts
794 ; let (alt_usg, b_occ) = lookupOcc (combineUsages alt_usgs) b'
795 scrut_occ = foldr combineOcc b_occ alt_occs
796 scrut_usg' = setScrutOcc env scrut_usg scrut' scrut_occ
797 -- The combined usage of the scrutinee is given
798 -- by scrut_occ, which is passed to scScrut, which
799 -- in turn treats a bare-variable scrutinee specially
801 ; return (alt_usg `combineUsage` scrut_usg',
802 Case scrut' b' (scSubstTy env ty) alts') }
804 sc_alt env _scrut' b' (con,bs,rhs)
805 = do { let (env1, bs1) = extendBndrsWith RecArg env bs
806 (env2, bs2) = extendCaseBndrs env1 b' con bs1
807 ; (usg,rhs') <- scExpr env2 rhs
808 ; let (usg', arg_occs) = lookupOccs usg bs2
809 scrut_occ = case con of
810 DataAlt dc -> ScrutOcc (unitUFM dc arg_occs)
811 _ -> ScrutOcc emptyUFM
812 ; return (usg', scrut_occ, (con, bs2, rhs')) }
814 scExpr' env (Let (NonRec bndr rhs) body)
815 | isTyVar bndr -- Type-lets may be created by doBeta
816 = scExpr' (extendScSubst env bndr rhs) body
818 = do { let (body_env, bndr') = extendBndr env bndr
819 ; (rhs_usg, (_, args', rhs_body', _)) <- scRecRhs env (bndr',rhs)
820 ; let rhs' = mkLams args' rhs_body'
822 ; if not opt_SpecInlineJoinPoints || null args' || isEmptyVarEnv (scu_calls rhs_usg) then do
824 let body_env2 = extendValEnv body_env bndr' (isValue (sc_vals env) rhs')
825 -- Record if the RHS is a value
826 ; (body_usg, body') <- scExpr body_env2 body
827 ; return (body_usg `combineUsage` rhs_usg, Let (NonRec bndr' rhs') body') }
828 else -- For now, just brutally inline the join point
829 do { let body_env2 = extendScSubst env bndr rhs'
830 ; scExpr body_env2 body } }
834 do { -- Join-point case
835 let body_env2 = extendHowBound body_env [bndr'] RecFun
836 -- If the RHS of this 'let' contains calls
837 -- to recursive functions that we're trying
838 -- to specialise, then treat this let too
839 -- as one to specialise
840 ; (body_usg, body') <- scExpr body_env2 body
842 ; (spec_usg, _, specs) <- specialise env (scu_calls body_usg) ([], rhs_info)
844 ; return (body_usg { scu_calls = scu_calls body_usg `delVarEnv` bndr' }
845 `combineUsage` rhs_usg `combineUsage` spec_usg,
846 mkLets [NonRec b r | (b,r) <- specInfoBinds rhs_info specs] body')
850 -- A *local* recursive group: see Note [Local recursive groups]
851 scExpr' env (Let (Rec prs) body)
852 = do { let (bndrs,rhss) = unzip prs
853 (rhs_env1,bndrs') = extendRecBndrs env bndrs
854 rhs_env2 = extendHowBound rhs_env1 bndrs' RecFun
856 ; (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs rhs_env2) (bndrs' `zip` rhss)
857 ; (body_usg, body') <- scExpr rhs_env2 body
859 -- NB: start specLoop from body_usg
860 ; (spec_usg, specs) <- specLoop rhs_env2 (scu_calls body_usg) rhs_infos nullUsage
861 [SI [] 0 (Just usg) | usg <- rhs_usgs]
863 ; let all_usg = spec_usg `combineUsage` body_usg
864 bind' = Rec (concat (zipWith specInfoBinds rhs_infos specs))
866 ; return (all_usg { scu_calls = scu_calls all_usg `delVarEnvList` bndrs' },
869 -----------------------------------
870 scApp :: ScEnv -> (InExpr, [InExpr]) -> UniqSM (ScUsage, CoreExpr)
872 scApp env (Var fn, args) -- Function is a variable
873 = ASSERT( not (null args) )
874 do { args_w_usgs <- mapM (scExpr env) args
875 ; let (arg_usgs, args') = unzip args_w_usgs
876 arg_usg = combineUsages arg_usgs
877 ; case scSubstId env fn of
878 fn'@(Lam {}) -> scExpr (zapScSubst env) (doBeta fn' args')
879 -- Do beta-reduction and try again
881 Var fn' -> return (arg_usg `combineUsage` fn_usg, mkApps (Var fn') args')
883 fn_usg = case lookupHowBound env fn' of
884 Just RecFun -> SCU { scu_calls = unitVarEnv fn' [(sc_vals env, args')],
885 scu_occs = emptyVarEnv }
886 Just RecArg -> SCU { scu_calls = emptyVarEnv,
887 scu_occs = unitVarEnv fn' (ScrutOcc emptyUFM) }
891 other_fn' -> return (arg_usg, mkApps other_fn' args') }
892 -- NB: doing this ignores any usage info from the substituted
893 -- function, but I don't think that matters. If it does
896 doBeta :: OutExpr -> [OutExpr] -> OutExpr
897 -- ToDo: adjust for System IF
898 doBeta (Lam bndr body) (arg : args) = Let (NonRec bndr arg) (doBeta body args)
899 doBeta fn args = mkApps fn args
901 -- The function is almost always a variable, but not always.
902 -- In particular, if this pass follows float-in,
903 -- which it may, we can get
904 -- (let f = ...f... in f) arg1 arg2
905 scApp env (other_fn, args)
906 = do { (fn_usg, fn') <- scExpr env other_fn
907 ; (arg_usgs, args') <- mapAndUnzipM (scExpr env) args
908 ; return (combineUsages arg_usgs `combineUsage` fn_usg, mkApps fn' args') }
910 ----------------------
911 scTopBind :: ScEnv -> CoreBind -> UniqSM (ScEnv, CoreBind)
912 scTopBind env (Rec prs)
913 | Just threshold <- sc_size env
914 , not (all (couldBeSmallEnoughToInline threshold) rhss)
916 = do { let (rhs_env,bndrs') = extendRecBndrs env bndrs
917 ; (_, rhss') <- mapAndUnzipM (scExpr rhs_env) rhss
918 ; return (rhs_env, Rec (bndrs' `zip` rhss')) }
919 | otherwise -- Do specialisation
920 = do { let (rhs_env1,bndrs') = extendRecBndrs env bndrs
921 rhs_env2 = extendHowBound rhs_env1 bndrs' RecFun
923 ; (rhs_usgs, rhs_infos) <- mapAndUnzipM (scRecRhs rhs_env2) (bndrs' `zip` rhss)
924 ; let rhs_usg = combineUsages rhs_usgs
926 ; (_, specs) <- specLoop rhs_env2 (scu_calls rhs_usg) rhs_infos nullUsage
927 [SI [] 0 Nothing | _ <- bndrs]
929 ; return (rhs_env1, -- For the body of the letrec, delete the RecFun business
930 Rec (concat (zipWith specInfoBinds rhs_infos specs))) }
932 (bndrs,rhss) = unzip prs
934 scTopBind env (NonRec bndr rhs)
935 = do { (_, rhs') <- scExpr env rhs
936 ; let (env1, bndr') = extendBndr env bndr
937 env2 = extendValEnv env1 bndr' (isValue (sc_vals env) rhs')
938 ; return (env2, NonRec bndr' rhs') }
940 ----------------------
941 scRecRhs :: ScEnv -> (OutId, InExpr) -> UniqSM (ScUsage, RhsInfo)
942 scRecRhs env (bndr,rhs)
943 = do { let (arg_bndrs,body) = collectBinders rhs
944 (body_env, arg_bndrs') = extendBndrsWith RecArg env arg_bndrs
945 ; (body_usg, body') <- scExpr body_env body
946 ; let (rhs_usg, arg_occs) = lookupOccs body_usg arg_bndrs'
947 ; return (rhs_usg, (bndr, arg_bndrs', body', arg_occs)) }
949 -- The arg_occs says how the visible,
950 -- lambda-bound binders of the RHS are used
951 -- (including the TyVar binders)
952 -- Two pats are the same if they match both ways
954 ----------------------
955 specInfoBinds :: RhsInfo -> SpecInfo -> [(Id,CoreExpr)]
956 specInfoBinds (fn, args, body, _) (SI specs _ _)
957 = [(id,rhs) | OS _ _ id rhs <- specs] ++
958 [(fn `addIdSpecialisations` rules, mkLams args body)]
960 rules = [r | OS _ r _ _ <- specs]
962 ----------------------
963 varUsage :: ScEnv -> OutVar -> ArgOcc -> ScUsage
965 | Just RecArg <- lookupHowBound env v = SCU { scu_calls = emptyVarEnv
966 , scu_occs = unitVarEnv v use }
967 | otherwise = nullUsage
971 %************************************************************************
973 The specialiser itself
975 %************************************************************************
978 type RhsInfo = (OutId, [OutVar], OutExpr, [ArgOcc])
979 -- Info about the *original* RHS of a binding we are specialising
980 -- Original binding f = \xs.body
981 -- Plus info about usage of arguments
983 data SpecInfo = SI [OneSpec] -- The specialisations we have generated
984 Int -- Length of specs; used for numbering them
985 (Maybe ScUsage) -- Nothing => we have generated specialisations
986 -- from calls in the *original* RHS
987 -- Just cs => we haven't, and this is the usage
988 -- of the original RHS
990 -- One specialisation: Rule plus definition
991 data OneSpec = OS CallPat -- Call pattern that generated this specialisation
992 CoreRule -- Rule connecting original id with the specialisation
993 OutId OutExpr -- Spec id + its rhs
999 -> ScUsage -> [SpecInfo] -- One per binder; acccumulating parameter
1000 -> UniqSM (ScUsage, [SpecInfo]) -- ...ditto...
1001 specLoop env all_calls rhs_infos usg_so_far specs_so_far
1002 = do { specs_w_usg <- zipWithM (specialise env all_calls) rhs_infos specs_so_far
1003 ; let (new_usg_s, all_specs) = unzip specs_w_usg
1004 new_usg = combineUsages new_usg_s
1005 new_calls = scu_calls new_usg
1006 all_usg = usg_so_far `combineUsage` new_usg
1007 ; if isEmptyVarEnv new_calls then
1008 return (all_usg, all_specs)
1010 specLoop env new_calls rhs_infos all_usg all_specs }
1014 -> CallEnv -- Info on calls
1016 -> SpecInfo -- Original RHS plus patterns dealt with
1017 -> UniqSM (ScUsage, SpecInfo) -- New specialised versions and their usage
1019 -- Note: the rhs here is the optimised version of the original rhs
1020 -- So when we make a specialised copy of the RHS, we're starting
1021 -- from an RHS whose nested functions have been optimised already.
1023 specialise env bind_calls (fn, arg_bndrs, body, arg_occs)
1024 spec_info@(SI specs spec_count mb_unspec)
1025 | not (isBottomingId fn) -- Note [Do not specialise diverging functions]
1026 , notNull arg_bndrs -- Only specialise functions
1027 , Just all_calls <- lookupVarEnv bind_calls fn
1028 = do { (boring_call, pats) <- callsToPats env specs arg_occs all_calls
1029 -- ; pprTrace "specialise" (vcat [ppr fn <+> ppr arg_occs,
1030 -- text "calls" <+> ppr all_calls,
1031 -- text "good pats" <+> ppr pats]) $
1034 -- Bale out if too many specialisations
1035 -- Rather a hacky way to do so, but it'll do for now
1036 ; let spec_count' = length pats + spec_count
1037 ; case sc_count env of
1038 Just max | spec_count' > max
1039 -> WARN( True, msg ) return (nullUsage, spec_info)
1041 msg = vcat [ sep [ ptext (sLit "SpecConstr: specialisation of") <+> quotes (ppr fn)
1042 , nest 2 (ptext (sLit "limited by bound of")) <+> int max ]
1043 , ptext (sLit "Use -fspec-constr-count=n to set the bound")
1045 extra | not opt_PprStyle_Debug = ptext (sLit "Use -dppr-debug to see specialisations")
1046 | otherwise = ptext (sLit "Specialisations:") <+> ppr (pats ++ [p | OS p _ _ _ <- specs])
1048 _normal_case -> do {
1050 (spec_usgs, new_specs) <- mapAndUnzipM (spec_one env fn arg_bndrs body)
1051 (pats `zip` [spec_count..])
1053 ; let spec_usg = combineUsages spec_usgs
1054 (new_usg, mb_unspec')
1056 Just rhs_usg | boring_call -> (spec_usg `combineUsage` rhs_usg, Nothing)
1057 _ -> (spec_usg, mb_unspec)
1059 ; return (new_usg, SI (new_specs ++ specs) spec_count' mb_unspec') } }
1061 = return (nullUsage, spec_info) -- The boring case
1064 ---------------------
1066 -> OutId -- Function
1067 -> [Var] -- Lambda-binders of RHS; should match patterns
1068 -> CoreExpr -- Body of the original function
1070 -> UniqSM (ScUsage, OneSpec) -- Rule and binding
1072 -- spec_one creates a specialised copy of the function, together
1073 -- with a rule for using it. I'm very proud of how short this
1074 -- function is, considering what it does :-).
1080 f = /\b \y::[(a,b)] -> ....f (b,c) ((:) (a,(b,c)) (x,v) (h w))...
1081 [c::*, v::(b,c) are presumably bound by the (...) part]
1083 f_spec = /\ b c \ v::(b,c) hw::[(a,(b,c))] ->
1084 (...entire body of f...) [b -> (b,c),
1085 y -> ((:) (a,(b,c)) (x,v) hw)]
1087 RULE: forall b::* c::*, -- Note, *not* forall a, x
1091 f (b,c) ((:) (a,(b,c)) (x,v) hw) = f_spec b c v hw
1094 spec_one env fn arg_bndrs body (call_pat@(qvars, pats), rule_number)
1095 = do { -- Specialise the body
1096 let spec_env = extendScSubstList (extendScInScope env qvars)
1097 (arg_bndrs `zip` pats)
1098 ; (spec_usg, spec_body) <- scExpr spec_env body
1100 -- ; pprTrace "spec_one" (ppr fn <+> vcat [text "pats" <+> ppr pats,
1101 -- text "calls" <+> (ppr (scu_calls spec_usg))])
1104 -- And build the results
1105 ; spec_uniq <- getUniqueUs
1106 ; let (spec_lam_args, spec_call_args) = mkWorkerArgs qvars body_ty
1107 -- Usual w/w hack to avoid generating
1108 -- a spec_rhs of unlifted type and no args
1111 fn_loc = nameSrcSpan fn_name
1112 spec_occ = mkSpecOcc (nameOccName fn_name)
1113 rule_name = mkFastString ("SC:" ++ showSDoc (ppr fn <> int rule_number))
1114 spec_rhs = mkLams spec_lam_args spec_body
1115 spec_str = calcSpecStrictness fn spec_lam_args pats
1116 spec_id = mkUserLocal spec_occ spec_uniq (mkPiTypes spec_lam_args body_ty) fn_loc
1117 `setIdNewStrictness` spec_str -- See Note [Transfer strictness]
1118 `setIdArity` count isId spec_lam_args
1119 body_ty = exprType spec_body
1120 rule_rhs = mkVarApps (Var spec_id) spec_call_args
1121 rule = mkLocalRule rule_name specConstrActivation fn_name qvars pats rule_rhs
1122 ; return (spec_usg, OS call_pat rule spec_id spec_rhs) }
1124 calcSpecStrictness :: Id -- The original function
1125 -> [Var] -> [CoreExpr] -- Call pattern
1126 -> StrictSig -- Strictness of specialised thing
1127 -- See Note [Transfer strictness]
1128 calcSpecStrictness fn qvars pats
1129 = StrictSig (mkTopDmdType spec_dmds TopRes)
1131 spec_dmds = [ lookupVarEnv dmd_env qv `orElse` lazyDmd | qv <- qvars, isId qv ]
1132 StrictSig (DmdType _ dmds _) = idNewStrictness fn
1134 dmd_env = go emptyVarEnv dmds pats
1136 go env ds (Type {} : pats) = go env ds pats
1137 go env (d:ds) (pat : pats) = go (go_one env d pat) ds pats
1140 go_one env d (Var v) = extendVarEnv_C both env v d
1141 go_one env (Box d) e = go_one env d e
1142 go_one env (Eval (Prod ds)) e
1143 | (Var _, args) <- collectArgs e = go env ds args
1144 go_one env _ _ = env
1146 -- In which phase should the specialise-constructor rules be active?
1147 -- Originally I made them always-active, but Manuel found that
1148 -- this defeated some clever user-written rules. So Plan B
1149 -- is to make them active only in Phase 0; after all, currently,
1150 -- the specConstr transformation is only run after the simplifier
1151 -- has reached Phase 0. In general one would want it to be
1152 -- flag-controllable, but for now I'm leaving it baked in
1154 specConstrActivation :: Activation
1155 specConstrActivation = ActiveAfter 0 -- Baked in; see comments above
1158 Note [Transfer strictness]
1159 ~~~~~~~~~~~~~~~~~~~~~~~~~~
1160 We must transfer strictness information from the original function to
1161 the specialised one. Suppose, for example
1164 and a RULE f (a:as) b = f_spec a as b
1166 Now we want f_spec to have strictess LLS, otherwise we'll use call-by-need
1167 when calling f_spec instead of call-by-value. And that can result in
1168 unbounded worsening in space (cf the classic foldl vs foldl')
1170 See Trac #3437 for a good example.
1172 The function calcSpecStrictness performs the calculation.
1175 %************************************************************************
1177 \subsection{Argument analysis}
1179 %************************************************************************
1181 This code deals with analysing call-site arguments to see whether
1182 they are constructor applications.
1186 type CallPat = ([Var], [CoreExpr]) -- Quantified variables and arguments
1189 callsToPats :: ScEnv -> [OneSpec] -> [ArgOcc] -> [Call] -> UniqSM (Bool, [CallPat])
1190 -- Result has no duplicate patterns,
1191 -- nor ones mentioned in done_pats
1192 -- Bool indicates that there was at least one boring pattern
1193 callsToPats env done_specs bndr_occs calls
1194 = do { mb_pats <- mapM (callToPats env bndr_occs) calls
1196 ; let good_pats :: [([Var], [CoreArg])]
1197 good_pats = catMaybes mb_pats
1198 done_pats = [p | OS p _ _ _ <- done_specs]
1199 is_done p = any (samePat p) done_pats
1201 ; return (any isNothing mb_pats,
1202 filterOut is_done (nubBy samePat good_pats)) }
1204 callToPats :: ScEnv -> [ArgOcc] -> Call -> UniqSM (Maybe CallPat)
1205 -- The [Var] is the variables to quantify over in the rule
1206 -- Type variables come first, since they may scope
1207 -- over the following term variables
1208 -- The [CoreExpr] are the argument patterns for the rule
1209 callToPats env bndr_occs (con_env, args)
1210 | length args < length bndr_occs -- Check saturated
1213 = do { let in_scope = substInScope (sc_subst env)
1214 ; prs <- argsToPats in_scope con_env (args `zip` bndr_occs)
1215 ; let (interesting_s, pats) = unzip prs
1216 pat_fvs = varSetElems (exprsFreeVars pats)
1217 qvars = filterOut (`elemInScopeSet` in_scope) pat_fvs
1218 -- Quantify over variables that are not in sccpe
1220 -- See Note [Shadowing] at the top
1222 (tvs, ids) = partition isTyVar qvars
1224 -- Put the type variables first; the type of a term
1225 -- variable may mention a type variable
1227 ; -- pprTrace "callToPats" (ppr args $$ ppr prs $$ ppr bndr_occs) $
1229 then return (Just (qvars', pats))
1230 else return Nothing }
1232 -- argToPat takes an actual argument, and returns an abstracted
1233 -- version, consisting of just the "constructor skeleton" of the
1234 -- argument, with non-constructor sub-expression replaced by new
1235 -- placeholder variables. For example:
1236 -- C a (D (f x) (g y)) ==> C p1 (D p2 p3)
1238 argToPat :: InScopeSet -- What's in scope at the fn defn site
1239 -> ValueEnv -- ValueEnv at the call site
1240 -> CoreArg -- A call arg (or component thereof)
1242 -> UniqSM (Bool, CoreArg)
1243 -- Returns (interesting, pat),
1244 -- where pat is the pattern derived from the argument
1245 -- intersting=True if the pattern is non-trivial (not a variable or type)
1246 -- E.g. x:xs --> (True, x:xs)
1247 -- f xs --> (False, w) where w is a fresh wildcard
1248 -- (f xs, 'c') --> (True, (w, 'c')) where w is a fresh wildcard
1249 -- \x. x+y --> (True, \x. x+y)
1250 -- lvl7 --> (True, lvl7) if lvl7 is bound
1251 -- somewhere further out
1253 argToPat _in_scope _val_env arg@(Type {}) _arg_occ
1254 = return (False, arg)
1256 argToPat in_scope val_env (Note _ arg) arg_occ
1257 = argToPat in_scope val_env arg arg_occ
1258 -- Note [Notes in call patterns]
1259 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1260 -- Ignore Notes. In particular, we want to ignore any InlineMe notes
1261 -- Perhaps we should not ignore profiling notes, but I'm going to
1262 -- ride roughshod over them all for now.
1263 --- See Note [Notes in RULE matching] in Rules
1265 argToPat in_scope val_env (Let _ arg) arg_occ
1266 = argToPat in_scope val_env arg arg_occ
1267 -- Look through let expressions
1268 -- e.g. f (let v = rhs in \y -> ...v...)
1269 -- Here we can specialise for f (\y -> ...)
1270 -- because the rule-matcher will look through the let.
1272 argToPat in_scope val_env (Cast arg co) arg_occ
1273 = do { (interesting, arg') <- argToPat in_scope val_env arg arg_occ
1274 ; let (ty1,ty2) = coercionKind co
1275 ; if not interesting then
1278 { -- Make a wild-card pattern for the coercion
1280 ; let co_name = mkSysTvName uniq (fsLit "sg")
1281 co_var = mkCoVar co_name (mkCoKind ty1 ty2)
1282 ; return (interesting, Cast arg' (mkTyVarTy co_var)) } }
1284 {- Disabling lambda specialisation for now
1285 It's fragile, and the spec_loop can be infinite
1286 argToPat in_scope val_env arg arg_occ
1288 = return (True, arg)
1290 is_value_lam (Lam v e) -- Spot a value lambda, even if
1291 | isId v = True -- it is inside a type lambda
1292 | otherwise = is_value_lam e
1293 is_value_lam other = False
1296 -- Check for a constructor application
1297 -- NB: this *precedes* the Var case, so that we catch nullary constrs
1298 argToPat in_scope val_env arg arg_occ
1299 | Just (ConVal dc args) <- isValue val_env arg
1301 ScrutOcc _ -> True -- Used only by case scrutinee
1302 BothOcc -> case arg of -- Used elsewhere
1303 App {} -> True -- see Note [Reboxing]
1305 _other -> False -- No point; the arg is not decomposed
1306 = do { args' <- argsToPats in_scope val_env (args `zip` conArgOccs arg_occ dc)
1307 ; return (True, mk_con_app dc (map snd args')) }
1309 -- Check if the argument is a variable that
1310 -- is in scope at the function definition site
1311 -- It's worth specialising on this if
1312 -- (a) it's used in an interesting way in the body
1313 -- (b) we know what its value is
1314 argToPat in_scope val_env (Var v) arg_occ
1315 | case arg_occ of { UnkOcc -> False; _other -> True }, -- (a)
1317 = return (True, Var v)
1320 | isLocalId v = v `elemInScopeSet` in_scope
1321 && isJust (lookupVarEnv val_env v)
1322 -- Local variables have values in val_env
1323 | otherwise = isValueUnfolding (idUnfolding v)
1324 -- Imports have unfoldings
1326 -- I'm really not sure what this comment means
1327 -- And by not wild-carding we tend to get forall'd
1328 -- variables that are in soope, which in turn can
1329 -- expose the weakness in let-matching
1330 -- See Note [Matching lets] in Rules
1332 -- Check for a variable bound inside the function.
1333 -- Don't make a wild-card, because we may usefully share
1334 -- e.g. f a = let x = ... in f (x,x)
1335 -- NB: this case follows the lambda and con-app cases!!
1336 -- argToPat _in_scope _val_env (Var v) _arg_occ
1337 -- = return (False, Var v)
1338 -- SLPJ : disabling this to avoid proliferation of versions
1339 -- also works badly when thinking about seeding the loop
1340 -- from the body of the let
1341 -- f x y = letrec g z = ... in g (x,y)
1342 -- We don't want to specialise for that *particular* x,y
1344 -- The default case: make a wild-card
1345 argToPat _in_scope _val_env arg _arg_occ
1346 = wildCardPat (exprType arg)
1348 wildCardPat :: Type -> UniqSM (Bool, CoreArg)
1349 wildCardPat ty = do { uniq <- getUniqueUs
1350 ; let id = mkSysLocal (fsLit "sc") uniq ty
1351 ; return (False, Var id) }
1353 argsToPats :: InScopeSet -> ValueEnv
1354 -> [(CoreArg, ArgOcc)]
1355 -> UniqSM [(Bool, CoreArg)]
1356 argsToPats in_scope val_env args
1359 do_one (arg,occ) = argToPat in_scope val_env arg occ
1364 isValue :: ValueEnv -> CoreExpr -> Maybe Value
1365 isValue _env (Lit lit)
1366 = Just (ConVal (LitAlt lit) [])
1369 | Just stuff <- lookupVarEnv env v
1370 = Just stuff -- You might think we could look in the idUnfolding here
1371 -- but that doesn't take account of which branch of a
1372 -- case we are in, which is the whole point
1374 | not (isLocalId v) && isCheapUnfolding unf
1375 = isValue env (unfoldingTemplate unf)
1378 -- However we do want to consult the unfolding
1379 -- as well, for let-bound constructors!
1381 isValue env (Lam b e)
1382 | isTyVar b = case isValue env e of
1383 Just _ -> Just LambdaVal
1385 | otherwise = Just LambdaVal
1387 isValue _env expr -- Maybe it's a constructor application
1388 | (Var fun, args) <- collectArgs expr
1389 = case isDataConWorkId_maybe fun of
1391 Just con | args `lengthAtLeast` dataConRepArity con
1392 -- Check saturated; might be > because the
1393 -- arity excludes type args
1394 -> Just (ConVal (DataAlt con) args)
1396 _other | valArgCount args < idArity fun
1397 -- Under-applied function
1398 -> Just LambdaVal -- Partial application
1402 isValue _env _expr = Nothing
1404 mk_con_app :: AltCon -> [CoreArg] -> CoreExpr
1405 mk_con_app (LitAlt lit) [] = Lit lit
1406 mk_con_app (DataAlt con) args = mkConApp con args
1407 mk_con_app _other _args = panic "SpecConstr.mk_con_app"
1409 samePat :: CallPat -> CallPat -> Bool
1410 samePat (vs1, as1) (vs2, as2)
1413 same (Var v1) (Var v2)
1414 | v1 `elem` vs1 = v2 `elem` vs2
1415 | v2 `elem` vs2 = False
1416 | otherwise = v1 == v2
1418 same (Lit l1) (Lit l2) = l1==l2
1419 same (App f1 a1) (App f2 a2) = same f1 f2 && same a1 a2
1421 same (Type {}) (Type {}) = True -- Note [Ignore type differences]
1422 same (Note _ e1) e2 = same e1 e2 -- Ignore casts and notes
1423 same (Cast e1 _) e2 = same e1 e2
1424 same e1 (Note _ e2) = same e1 e2
1425 same e1 (Cast e2 _) = same e1 e2
1427 same e1 e2 = WARN( bad e1 || bad e2, ppr e1 $$ ppr e2)
1428 False -- Let, lambda, case should not occur
1429 bad (Case {}) = True
1435 Note [Ignore type differences]
1436 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1437 We do not want to generate specialisations where the call patterns
1438 differ only in their type arguments! Not only is it utterly useless,
1439 but it also means that (with polymorphic recursion) we can generate
1440 an infinite number of specialisations. Example is Data.Sequence.adjustTree,