2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 %************************************************************************
6 \section[OccurAnal]{Occurrence analysis pass}
8 %************************************************************************
10 The occurrence analyser re-typechecks a core expression, returning a new
11 core expression with (hopefully) improved usage information.
15 -- The above warning supression flag is a temporary kludge.
16 -- While working on this module you are encouraged to remove it and fix
17 -- any warnings in the module. See
18 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
22 occurAnalysePgm, occurAnalyseExpr
25 #include "HsVersions.h"
29 import CoreUtils ( exprIsTrivial, isDefaultAlt )
32 import BasicTypes ( OccInfo(..), isOneOcc, InterestingCxt )
37 import Maybes ( orElse )
38 import Digraph ( stronglyConnCompR, SCC(..) )
39 import PrelNames ( buildIdKey, foldrIdKey, runSTRepIdKey, augmentIdKey )
40 import Unique ( Unique )
41 import UniqFM ( keysUFM, intersectsUFM, intersectUFM_C, foldUFM_Directly )
42 import Util ( mapAndUnzip )
49 %************************************************************************
51 \subsection[OccurAnal-main]{Counting occurrences: main function}
53 %************************************************************************
55 Here's the externally-callable interface:
58 occurAnalysePgm :: [CoreBind] -> [CoreBind]
60 = snd (go initOccEnv binds)
62 go :: OccEnv -> [CoreBind] -> (UsageDetails, [CoreBind])
66 = (final_usage, bind' ++ binds')
68 (bs_usage, binds') = go env binds
69 (final_usage, bind') = occAnalBind env bind bs_usage
71 occurAnalyseExpr :: CoreExpr -> CoreExpr
72 -- Do occurrence analysis, and discard occurence info returned
73 occurAnalyseExpr expr = snd (occAnal initOccEnv expr)
77 %************************************************************************
79 \subsection[OccurAnal-main]{Counting occurrences: main function}
81 %************************************************************************
89 -> UsageDetails -- Usage details of scope
90 -> (UsageDetails, -- Of the whole let(rec)
93 occAnalBind env (NonRec binder rhs) body_usage
94 | not (binder `usedIn` body_usage) -- It's not mentioned
97 | otherwise -- It's mentioned in the body
98 = (body_usage' +++ addRuleUsage rhs_usage binder, -- Note [Rules are extra RHSs]
99 [NonRec tagged_binder rhs'])
101 (body_usage', tagged_binder) = tagBinder body_usage binder
102 (rhs_usage, rhs') = occAnalRhs env tagged_binder rhs
107 Dropping dead code for recursive bindings is done in a very simple way:
109 the entire set of bindings is dropped if none of its binders are
110 mentioned in its body; otherwise none are.
112 This seems to miss an obvious improvement.
124 Now 'f' is unused! But it's OK! Dependency analysis will sort this
125 out into a letrec for 'g' and a 'let' for 'f', and then 'f' will get
126 dropped. It isn't easy to do a perfect job in one blow. Consider
137 Note [Loop breaking and RULES]
138 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
139 Loop breaking is surprisingly subtle. First read the section 4 of
140 "Secrets of the GHC inliner". This describes our basic plan.
142 However things are made quite a bit more complicated by RULES. Remember
144 * Note [Rules are extra RHSs]
145 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
146 A RULE for 'f' is like an extra RHS for 'f'. That way the "parent"
147 keeps the specialised "children" alive. If the parent dies
148 (because it isn't referenced any more), then the children will die
149 too (unless they are already referenced directly).
151 To that end, we build a Rec group for each cyclic strongly
153 *treating f's rules as extra RHSs for 'f'*.
155 When we make the Rec groups we include variables free in *either*
156 LHS *or* RHS of the rule. The former might seems silly, but see
157 Note [Rule dependency info].
159 So in Example [eftInt], eftInt and eftIntFB will be put in the
160 same Rec, even though their 'main' RHSs are both non-recursive.
162 * Note [Rules are visible in their own rec group]
163 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
164 We want the rules for 'f' to be visible in f's right-hand side.
165 And we'd like them to be visible in other functions in f's Rec
166 group. E.g. in Example [Specialisation rules] we want f' rule
167 to be visible in both f's RHS, and fs's RHS.
169 This means that we must simplify the RULEs first, before looking
170 at any of the definitions. This is done by Simplify.simplRecBind,
171 when it calls addLetIdInfo.
173 * Note [Choosing loop breakers]
174 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
175 We avoid infinite inlinings by choosing loop breakers, and
176 ensuring that a loop breaker cuts each loop. But what is a
177 "loop"? In particular, a RULES is like an equation for 'f' that
178 is *always* inlined if it are applicable. We do *not* disable
179 rules for loop-breakers. It's up to whoever makes the rules to
180 make sure that the rules themselves alwasys terminate. See Note
181 [Rules for recursive functions] in Simplify.lhs
184 f's RHS mentions g, and
185 g has a RULE that mentions h, and
186 h has a RULE that mentions f
188 then we *must* choose f to be a loop breaker. In general, take the
189 free variables of f's RHS, and augment it with all the variables
190 reachable by RULES from those starting points. That is the whole
191 reason for computing rule_fv_env in occAnalBind. (Of course we
192 only consider free vars that are also binders in this Rec group.)
194 Note that when we compute this rule_fv_env, we only consider variables
195 free in the *RHS* of the rule, in contrast to the way we build the
196 Rec group in the first place (Note [Rule dependency info])
198 Note that in Example [eftInt], *neither* eftInt *nor* eftIntFB is
199 chosen as a loop breaker, because their RHSs don't mention each other.
200 And indeed both can be inlined safely.
202 Note that the edges of the graph we use for computing loop breakers
203 are not the same as the edges we use for computing the Rec blocks.
204 That's why we compute
205 rec_edges for the Rec block analysis
206 loop_breaker_edges for the loop breaker analysis
209 * Note [Weak loop breakers]
210 ~~~~~~~~~~~~~~~~~~~~~~~~~
211 There is a last nasty wrinkle. Suppose we have
221 Remmber that we simplify the RULES before any RHS (see Note
222 [Rules are visible in their own rec group] above).
224 So we must *not* postInlineUnconditinoally 'g', even though
225 its RHS turns out to be trivial. (I'm assuming that 'g' is
226 not choosen as a loop breaker.)
228 We "solve" this by making g a "weak" or "rules-only" loop breaker,
229 with OccInfo = IAmLoopBreaker True. A normal "strong" loop breaker
230 has IAmLoopBreaker False. So
232 Inline postInlineUnconditinoally
233 IAmLoopBreaker False no no
234 IAmLoopBreaker True yes no
237 The **sole** reason for this kind of loop breaker is so that
238 postInlineUnconditioanlly does not fire. Ugh.
240 * Note [Rule dependency info]
241 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
242 The VarSet in a SpecInfo is used for dependency analysis in the
243 occurrence analyser. We must track free vars in *both* lhs and rhs. Why both?
247 Then if we substitute y for x, we'd better do so in the
248 rule's LHS too, so we'd better ensure the dependency is respected
253 Example (from GHC.Enum):
255 eftInt :: Int# -> Int# -> [Int]
256 eftInt x y = ...(non-recursive)...
258 {-# INLINE [0] eftIntFB #-}
259 eftIntFB :: (Int -> r -> r) -> r -> Int# -> Int# -> r
260 eftIntFB c n x y = ...(non-recursive)...
263 "eftInt" [~1] forall x y. eftInt x y = build (\ c n -> eftIntFB c n x y)
264 "eftIntList" [1] eftIntFB (:) [] = eftInt
267 Example [Specialisation rules]
268 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
269 Consider this group, which is typical of what SpecConstr builds:
271 fs a = ....f (C a)....
272 f x = ....f (C a)....
273 {-# RULE f (C a) = fs a #-}
275 So 'f' and 'fs' are in the same Rec group (since f refers to fs via its RULE).
277 But watch out! If 'fs' is not chosen as a loop breaker, we may get an infinite loop:
278 - the RULE is applied in f's RHS (see Note [Self-recursive rules] in Simplify
279 - fs is inlined (say it's small)
280 - now there's another opportunity to apply the RULE
282 This showed up when compiling Control.Concurrent.Chan.getChanContents.
286 occAnalBind env (Rec pairs) body_usage
287 | not (any (`usedIn` body_usage) bndrs) -- NB: look at body_usage, not total_usage
288 = (body_usage, []) -- Dead code
290 = (final_usage, map ({-# SCC "occAnalBind.dofinal" #-} do_final_bind) sccs)
292 bndrs = map fst pairs
293 bndr_set = mkVarSet bndrs
295 ---------------------------------------
296 -- See Note [Loop breaking]
297 ---------------------------------------
299 -------------Dependency analysis ------------------------------
300 occ_anald :: [(Id, (UsageDetails, CoreExpr))]
301 -- The UsageDetails here are strictly those arising from the RHS
302 -- *not* from any rules in the Id
303 occ_anald = [(bndr, occAnalRhs env bndr rhs) | (bndr,rhs) <- pairs]
305 total_usage = foldl add_usage body_usage occ_anald
306 add_usage body_usage (bndr, (rhs_usage, _))
307 = body_usage +++ addRuleUsage rhs_usage bndr
309 (final_usage, tagged_bndrs) = tagBinders total_usage bndrs
310 final_bndrs | no_rules = tagged_bndrs
311 | otherwise = map tag_rule_var tagged_bndrs
312 tag_rule_var bndr | bndr `elemVarSet` all_rule_fvs = makeLoopBreaker True bndr
315 ---- stuff for dependency analysis of binds -------------------------------
316 sccs :: [SCC (Node Details)]
317 sccs = {-# SCC "occAnalBind.scc" #-} stronglyConnCompR rec_edges
319 rec_edges :: [Node Details] -- The binders are tagged with correct occ-info
320 rec_edges = {-# SCC "occAnalBind.assoc" #-} zipWith make_node final_bndrs occ_anald
321 make_node tagged_bndr (_bndr, (rhs_usage, rhs))
322 = ((tagged_bndr, rhs, rhs_fvs), idUnique tagged_bndr, out_edges)
324 rhs_fvs = intersectUFM_C (\b _ -> b) bndr_set rhs_usage
325 out_edges = keysUFM (rhs_fvs `unionVarSet` idRuleVars tagged_bndr)
328 -- (a -> b) means a mentions b
329 -- Given the usage details (a UFM that gives occ info for each free var of
330 -- the RHS) we can get the list of free vars -- or rather their Int keys --
331 -- by just extracting the keys from the finite map. Grimy, but fast.
332 -- Previously we had this:
333 -- [ bndr | bndr <- bndrs,
334 -- maybeToBool (lookupVarEnv rhs_usage bndr)]
335 -- which has n**2 cost, and this meant that edges_from alone
336 -- consumed 10% of total runtime!
338 ---- Stuff to "re-constitute" bindings from dependency-analysis info ------
339 do_final_bind (AcyclicSCC ((bndr, rhs, _), _, _)) = NonRec bndr rhs
340 do_final_bind (CyclicSCC cycle)
341 | no_rules = Rec (reOrderCycle cycle)
342 | otherwise = Rec (concatMap reOrderRec (stronglyConnCompR loop_breaker_edges))
343 where -- See Note [Choosing loop breakers] for looop_breker_edges
344 loop_breaker_edges = map mk_node cycle
345 mk_node (details@(bndr, rhs, rhs_fvs), k, _) = (details, k, new_ks)
347 new_ks = keysUFM (extendFvs rule_fv_env rhs_fvs rhs_fvs)
350 ------------------------------------
351 rule_fv_env :: IdEnv IdSet -- Variables from this group mentioned in RHS of rules
352 -- Domain is *subset* of bound vars (others have no rule fvs)
353 rule_fv_env = rule_loop init_rule_fvs
355 no_rules = null init_rule_fvs
356 all_rule_fvs = foldr (unionVarSet . snd) emptyVarSet init_rule_fvs
357 init_rule_fvs = [(b, rule_fvs)
359 , let rule_fvs = idRuleRhsVars b `intersectVarSet` bndr_set
360 , not (isEmptyVarSet rule_fvs)]
362 rule_loop :: [(Id,IdSet)] -> IdEnv IdSet -- Finds fixpoint
365 | otherwise = rule_loop new_fv_list
367 env = mkVarEnv init_rule_fvs
368 (no_change, new_fv_list) = mapAccumL bump True fv_list
369 bump no_change (b,fvs)
370 | new_fvs `subVarSet` fvs = (no_change, (b,fvs))
371 | otherwise = (False, (b,new_fvs `unionVarSet` fvs))
373 new_fvs = extendFvs env emptyVarSet fvs
375 idRuleRhsVars :: Id -> VarSet
376 -- Just the variables free on the *rhs* of a rule
377 -- See Note [Choosing loop breakers]
378 idRuleRhsVars id = foldr (unionVarSet . ruleRhsFreeVars) emptyVarSet (idCoreRules id)
380 extendFvs :: IdEnv IdSet -> IdSet -> IdSet -> IdSet
381 -- (extendFVs env fvs s) returns (fvs `union` env(s))
382 extendFvs env fvs id_set
383 = foldUFM_Directly add fvs id_set
386 = case lookupVarEnv_Directly env uniq of
387 Just fvs' -> fvs' `unionVarSet` fvs
391 @reOrderRec@ is applied to the list of (binder,rhs) pairs for a cyclic
392 strongly connected component (there's guaranteed to be a cycle). It returns the
394 a) in a better order,
395 b) with some of the Ids having a IAmALoopBreaker pragma
397 The "loop-breaker" Ids are sufficient to break all cycles in the SCC. This means
398 that the simplifier can guarantee not to loop provided it never records an inlining
399 for these no-inline guys.
401 Furthermore, the order of the binds is such that if we neglect dependencies
402 on the no-inline Ids then the binds are topologically sorted. This means
403 that the simplifier will generally do a good job if it works from top bottom,
404 recording inlinings for any Ids which aren't marked as "no-inline" as it goes.
407 [June 98: I don't understand the following paragraphs, and I've
408 changed the a=b case again so that it isn't a special case any more.]
410 Here's a case that bit me:
418 Re-ordering doesn't change the order of bindings, but there was no loop-breaker.
420 My solution was to make a=b bindings record b as Many, rather like INLINE bindings.
421 Perhaps something cleverer would suffice.
426 type Node details = (details, Unique, [Unique]) -- The Ints are gotten from the Unique,
427 -- which is gotten from the Id.
428 type Details = (Id, -- Binder
430 IdSet) -- RHS free vars (*not* include rules)
432 reOrderRec :: SCC (Node Details)
434 -- Sorted into a plausible order. Enough of the Ids have
435 -- IAmALoopBreaker pragmas that there are no loops left.
436 reOrderRec (AcyclicSCC ((bndr, rhs, _), _, _)) = [(bndr, rhs)]
437 reOrderRec (CyclicSCC cycle) = reOrderCycle cycle
439 reOrderCycle :: [Node Details] -> [(Id,CoreExpr)]
441 = panic "reOrderCycle"
442 reOrderCycle [bind] -- Common case of simple self-recursion
443 = [(makeLoopBreaker False bndr, rhs)]
445 ((bndr, rhs, _), _, _) = bind
447 reOrderCycle (bind : binds)
448 = -- Choose a loop breaker, mark it no-inline,
449 -- do SCC analysis on the rest, and recursively sort them out
450 concatMap reOrderRec (stronglyConnCompR unchosen) ++
451 [(makeLoopBreaker False bndr, rhs)]
454 (chosen_bind, unchosen) = choose_loop_breaker bind (score bind) [] binds
455 (bndr, rhs, _) = chosen_bind
457 -- This loop looks for the bind with the lowest score
458 -- to pick as the loop breaker. The rest accumulate in
459 choose_loop_breaker (details,_,_) loop_sc acc []
460 = (details, acc) -- Done
462 choose_loop_breaker loop_bind loop_sc acc (bind : binds)
463 | sc < loop_sc -- Lower score so pick this new one
464 = choose_loop_breaker bind sc (loop_bind : acc) binds
466 | otherwise -- No lower so don't pick it
467 = choose_loop_breaker loop_bind loop_sc (bind : acc) binds
471 score :: Node Details -> Int -- Higher score => less likely to be picked as loop breaker
472 score ((bndr, rhs, _), _, _)
473 | workerExists (idWorkerInfo bndr) = 10
474 -- Note [Worker inline loop]
476 | exprIsTrivial rhs = 4 -- Practically certain to be inlined
477 -- Used to have also: && not (isExportedId bndr)
478 -- But I found this sometimes cost an extra iteration when we have
479 -- rec { d = (a,b); a = ...df...; b = ...df...; df = d }
480 -- where df is the exported dictionary. Then df makes a really
481 -- bad choice for loop breaker
483 | is_con_app rhs = 2 -- Data types help with cases
486 | inlineCandidate bndr rhs = 1 -- Likely to be inlined
487 -- Note [Inline candidates]
491 inlineCandidate :: Id -> CoreExpr -> Bool
492 inlineCandidate id (Note InlineMe _) = True
493 inlineCandidate id rhs = isOneOcc (idOccInfo id)
497 -- It's really really important to inline dictionaries. Real
498 -- example (the Enum Ordering instance from GHC.Base):
500 -- rec f = \ x -> case d of (p,q,r) -> p x
501 -- g = \ x -> case d of (p,q,r) -> q x
504 -- Here, f and g occur just once; but we can't inline them into d.
505 -- On the other hand we *could* simplify those case expressions if
506 -- we didn't stupidly choose d as the loop breaker.
507 -- But we won't because constructor args are marked "Many".
508 -- Inlining dictionaries is really essential to unravelling
509 -- the loops in static numeric dictionaries, see GHC.Float.
511 -- Cheap and cheerful; the simplifer moves casts out of the way
512 -- The lambda case is important to spot x = /\a. C (f a)
513 -- which comes up when C is a dictionary constructor and
514 -- f is a default method.
515 -- Example: the instance for Show (ST s a) in GHC.ST
517 -- However we *also* treat (\x. C p q) as a con-app-like thing,
518 -- Note [Closure conversion]
519 is_con_app (Var v) = isDataConWorkId v
520 is_con_app (App f _) = is_con_app f
521 is_con_app (Lam b e) = is_con_app e
522 is_con_app (Note _ e) = is_con_app e
523 is_con_app other = False
525 makeLoopBreaker :: Bool -> Id -> Id
526 -- Set the loop-breaker flag
527 -- See Note [Weak loop breakers]
528 makeLoopBreaker weak bndr = setIdOccInfo bndr (IAmALoopBreaker weak)
531 Note [Worker inline loop]
532 ~~~~~~~~~~~~~~~~~~~~~~~~
533 Never choose a wrapper as the loop breaker! Because
534 wrappers get auto-generated inlinings when importing, and
535 that can lead to an infinite inlining loop. For example:
537 $wfoo x = ....foo x....
539 {-loop brk-} foo x = ...$wfoo x...
542 The interface file sees the unfolding for $wfoo, and sees that foo is
543 strict (and hence it gets an auto-generated wrapper). Result: an
544 infinite inlining in the importing scope. So be a bit careful if you
545 change this. A good example is Tree.repTree in
546 nofib/spectral/minimax. If the repTree wrapper is chosen as the loop
547 breaker then compiling Game.hs goes into an infinite loop (this
548 happened when we gave is_con_app a lower score than inline candidates).
550 Note [Closure conversion]
551 ~~~~~~~~~~~~~~~~~~~~~~~~~
552 We treat (\x. C p q) as a high-score candidate in the letrec scoring algorithm.
553 The immediate motivation came from the result of a closure-conversion transformation
554 which generated code like this:
556 data Clo a b = forall c. Clo (c -> a -> b) c
558 ($:) :: Clo a b -> a -> b
559 Clo f env $: x = f env x
561 rec { plus = Clo plus1 ()
563 ; plus1 _ n = Clo plus2 n
566 ; plus2 (Succ m) n = Succ (plus $: m $: n) }
568 If we inline 'plus' and 'plus1', everything unravels nicely. But if
569 we choose 'plus1' as the loop breaker (which is entirely possible
570 otherwise), the loop does not unravel nicely.
573 @occAnalRhs@ deals with the question of bindings where the Id is marked
574 by an INLINE pragma. For these we record that anything which occurs
575 in its RHS occurs many times. This pessimistically assumes that ths
576 inlined binder also occurs many times in its scope, but if it doesn't
577 we'll catch it next time round. At worst this costs an extra simplifier pass.
578 ToDo: try using the occurrence info for the inline'd binder.
580 [March 97] We do the same for atomic RHSs. Reason: see notes with reOrderRec.
581 [June 98, SLPJ] I've undone this change; I don't understand it. See notes with reOrderRec.
586 -> Id -> CoreExpr -- Binder and rhs
587 -- For non-recs the binder is alrady tagged
588 -- with occurrence info
589 -> (UsageDetails, CoreExpr)
591 occAnalRhs env id rhs
594 ctxt | certainly_inline id = env
595 | otherwise = rhsCtxt
596 -- Note that we generally use an rhsCtxt. This tells the occ anal n
597 -- that it's looking at an RHS, which has an effect in occAnalApp
599 -- But there's a problem. Consider
604 -- First time round, it looks as if x1 and x2 occur as an arg of a
605 -- let-bound constructor ==> give them a many-occurrence.
606 -- But then x3 is inlined (unconditionally as it happens) and
607 -- next time round, x2 will be, and the next time round x1 will be
608 -- Result: multiple simplifier iterations. Sigh.
609 -- Crude solution: use rhsCtxt for things that occur just once...
611 certainly_inline id = case idOccInfo id of
612 OneOcc in_lam one_br _ -> not in_lam && one_br
619 addRuleUsage :: UsageDetails -> Id -> UsageDetails
620 -- Add the usage from RULES in Id to the usage
621 addRuleUsage usage id
622 = foldVarSet add usage (idRuleVars id)
624 add v u = addOneOcc u v NoOccInfo -- Give a non-committal binder info
625 -- (i.e manyOcc) because many copies
626 -- of the specialised thing can appear
634 -> (UsageDetails, -- Gives info only about the "interesting" Ids
637 occAnal env (Type t) = (emptyDetails, Type t)
638 occAnal env (Var v) = (mkOneOcc env v False, Var v)
639 -- At one stage, I gathered the idRuleVars for v here too,
640 -- which in a way is the right thing to do.
641 -- Btu that went wrong right after specialisation, when
642 -- the *occurrences* of the overloaded function didn't have any
643 -- rules in them, so the *specialised* versions looked as if they
644 -- weren't used at all.
647 We regard variables that occur as constructor arguments as "dangerousToDup":
651 f x = let y = expensive x in
653 (case z of {(p,q)->q}, case z of {(p,q)->q})
656 We feel free to duplicate the WHNF (True,y), but that means
657 that y may be duplicated thereby.
659 If we aren't careful we duplicate the (expensive x) call!
660 Constructors are rather like lambdas in this way.
663 occAnal env expr@(Lit lit) = (emptyDetails, expr)
667 occAnal env (Note InlineMe body)
668 = case occAnal env body of { (usage, body') ->
669 (mapVarEnv markMany usage, Note InlineMe body')
672 occAnal env (Note note@(SCC cc) body)
673 = case occAnal env body of { (usage, body') ->
674 (mapVarEnv markInsideSCC usage, Note note body')
677 occAnal env (Note note body)
678 = case occAnal env body of { (usage, body') ->
679 (usage, Note note body')
682 occAnal env (Cast expr co)
683 = case occAnal env expr of { (usage, expr') ->
684 (markRhsUds env True usage, Cast expr' co)
685 -- If we see let x = y `cast` co
686 -- then mark y as 'Many' so that we don't
687 -- immediately inline y again.
692 occAnal env app@(App fun arg)
693 = occAnalApp env (collectArgs app) False
695 -- Ignore type variables altogether
696 -- (a) occurrences inside type lambdas only not marked as InsideLam
697 -- (b) type variables not in environment
699 occAnal env expr@(Lam x body) | isTyVar x
700 = case occAnal env body of { (body_usage, body') ->
701 (body_usage, Lam x body')
704 -- For value lambdas we do a special hack. Consider
706 -- If we did nothing, x is used inside the \y, so would be marked
707 -- as dangerous to dup. But in the common case where the abstraction
708 -- is applied to two arguments this is over-pessimistic.
709 -- So instead, we just mark each binder with its occurrence
710 -- info in the *body* of the multiple lambda.
711 -- Then, the simplifier is careful when partially applying lambdas.
713 occAnal env expr@(Lam _ _)
714 = case occAnal env_body body of { (body_usage, body') ->
716 (final_usage, tagged_binders) = tagBinders body_usage binders
717 -- URGH! Sept 99: we don't seem to be able to use binders' here, because
718 -- we get linear-typed things in the resulting program that we can't handle yet.
719 -- (e.g. PrelShow) TODO
721 really_final_usage = if linear then
724 mapVarEnv markInsideLam final_usage
727 mkLams tagged_binders body') }
729 env_body = vanillaCtxt -- Body is (no longer) an RhsContext
730 (binders, body) = collectBinders expr
731 binders' = oneShotGroup env binders
732 linear = all is_one_shot binders'
733 is_one_shot b = isId b && isOneShotBndr b
735 occAnal env (Case scrut bndr ty alts)
736 = case occ_anal_scrut scrut alts of { (scrut_usage, scrut') ->
737 case mapAndUnzip (occAnalAlt alt_env bndr) alts of { (alts_usage_s, alts') ->
739 alts_usage = foldr1 combineAltsUsageDetails alts_usage_s
740 alts_usage' = addCaseBndrUsage alts_usage
741 (alts_usage1, tagged_bndr) = tagBinder alts_usage' bndr
742 total_usage = scrut_usage +++ alts_usage1
744 total_usage `seq` (total_usage, Case scrut' tagged_bndr ty alts') }}
746 -- The case binder gets a usage of either "many" or "dead", never "one".
747 -- Reason: we like to inline single occurrences, to eliminate a binding,
748 -- but inlining a case binder *doesn't* eliminate a binding.
749 -- We *don't* want to transform
750 -- case x of w { (p,q) -> f w }
752 -- case x of w { (p,q) -> f (p,q) }
753 addCaseBndrUsage usage = case lookupVarEnv usage bndr of
755 Just occ -> extendVarEnv usage bndr (markMany occ)
757 alt_env = setVanillaCtxt env
758 -- Consider x = case v of { True -> (p,q); ... }
759 -- Then it's fine to inline p and q
761 occ_anal_scrut (Var v) (alt1 : other_alts)
762 | not (null other_alts) || not (isDefaultAlt alt1)
763 = (mkOneOcc env v True, Var v)
764 occ_anal_scrut scrut alts = occAnal vanillaCtxt scrut
765 -- No need for rhsCtxt
767 occAnal env (Let bind body)
768 = case occAnal env body of { (body_usage, body') ->
769 case occAnalBind env bind body_usage of { (final_usage, new_binds) ->
770 (final_usage, mkLets new_binds body') }}
773 = case mapAndUnzip (occAnal arg_env) args of { (arg_uds_s, args') ->
774 (foldr (+++) emptyDetails arg_uds_s, args')}
776 arg_env = vanillaCtxt
779 Applications are dealt with specially because we want
780 the "build hack" to work.
783 occAnalApp env (Var fun, args) is_rhs
784 = case args_stuff of { (args_uds, args') ->
786 final_args_uds = markRhsUds env is_pap args_uds
788 (fun_uds +++ final_args_uds, mkApps (Var fun) args') }
790 fun_uniq = idUnique fun
791 fun_uds = mkOneOcc env fun (valArgCount args > 0)
792 is_pap = isDataConWorkId fun || valArgCount args < idArity fun
794 -- Hack for build, fold, runST
795 args_stuff | fun_uniq == buildIdKey = appSpecial env 2 [True,True] args
796 | fun_uniq == augmentIdKey = appSpecial env 2 [True,True] args
797 | fun_uniq == foldrIdKey = appSpecial env 3 [False,True] args
798 | fun_uniq == runSTRepIdKey = appSpecial env 2 [True] args
799 -- (foldr k z xs) may call k many times, but it never
800 -- shares a partial application of k; hence [False,True]
801 -- This means we can optimise
802 -- foldr (\x -> let v = ...x... in \y -> ...v...) z xs
803 -- by floating in the v
805 | otherwise = occAnalArgs env args
808 occAnalApp env (fun, args) is_rhs
809 = case occAnal (addAppCtxt env args) fun of { (fun_uds, fun') ->
810 -- The addAppCtxt is a bit cunning. One iteration of the simplifier
811 -- often leaves behind beta redexs like
813 -- Here we would like to mark x,y as one-shot, and treat the whole
814 -- thing much like a let. We do this by pushing some True items
815 -- onto the context stack.
817 case occAnalArgs env args of { (args_uds, args') ->
819 final_uds = fun_uds +++ args_uds
821 (final_uds, mkApps fun' args') }}
824 markRhsUds :: OccEnv -- Check if this is a RhsEnv
825 -> Bool -- and this is true
826 -> UsageDetails -- The do markMany on this
828 -- We mark the free vars of the argument of a constructor or PAP
829 -- as "many", if it is the RHS of a let(rec).
830 -- This means that nothing gets inlined into a constructor argument
831 -- position, which is what we want. Typically those constructor
832 -- arguments are just variables, or trivial expressions.
834 -- This is the *whole point* of the isRhsEnv predicate
835 markRhsUds env is_pap arg_uds
836 | isRhsEnv env && is_pap = mapVarEnv markMany arg_uds
837 | otherwise = arg_uds
841 -> Int -> CtxtTy -- Argument number, and context to use for it
843 -> (UsageDetails, [CoreExpr])
844 appSpecial env n ctxt args
847 arg_env = vanillaCtxt
849 go n [] = (emptyDetails, []) -- Too few args
851 go 1 (arg:args) -- The magic arg
852 = case occAnal (setCtxt arg_env ctxt) arg of { (arg_uds, arg') ->
853 case occAnalArgs env args of { (args_uds, args') ->
854 (arg_uds +++ args_uds, arg':args') }}
857 = case occAnal arg_env arg of { (arg_uds, arg') ->
858 case go (n-1) args of { (args_uds, args') ->
859 (arg_uds +++ args_uds, arg':args') }}
865 If the case binder occurs at all, the other binders effectively do too.
867 case e of x { (a,b) -> rhs }
870 If e turns out to be (e1,e2) we indeed get something like
871 let a = e1; b = e2; x = (a,b) in rhs
873 Note [Aug 06]: I don't think this is necessary any more, and it helpe
874 to know when binders are unused. See esp the call to
875 isDeadBinder in Simplify.mkDupableAlt
878 occAnalAlt env case_bndr (con, bndrs, rhs)
879 = case occAnal env rhs of { (rhs_usage, rhs') ->
881 (final_usage, tagged_bndrs) = tagBinders rhs_usage bndrs
882 final_bndrs = tagged_bndrs -- See Note [Aug06] above
884 final_bndrs | case_bndr `elemVarEnv` final_usage = bndrs
885 | otherwise = tagged_bndrs
886 -- Leave the binders untagged if the case
887 -- binder occurs at all; see note above
890 (final_usage, (con, final_bndrs, rhs')) }
894 %************************************************************************
896 \subsection[OccurAnal-types]{OccEnv}
898 %************************************************************************
902 = OccEnv OccEncl -- Enclosing context information
903 CtxtTy -- Tells about linearity
905 -- OccEncl is used to control whether to inline into constructor arguments
907 -- x = (p,q) -- Don't inline p or q
908 -- y = /\a -> (p a, q a) -- Still don't inline p or q
909 -- z = f (p,q) -- Do inline p,q; it may make a rule fire
910 -- So OccEncl tells enought about the context to know what to do when
911 -- we encounter a contructor application or PAP.
914 = OccRhs -- RHS of let(rec), albeit perhaps inside a type lambda
915 -- Don't inline into constructor args here
916 | OccVanilla -- Argument of function, body of lambda, scruintee of case etc.
917 -- Do inline into constructor args here
922 -- True:ctxt Analysing a function-valued expression that will be
925 -- False:ctxt Analysing a function-valued expression that may
926 -- be applied many times; but when it is,
927 -- the CtxtTy inside applies
930 initOccEnv = OccEnv OccRhs []
932 vanillaCtxt = OccEnv OccVanilla []
933 rhsCtxt = OccEnv OccRhs []
935 isRhsEnv (OccEnv OccRhs _) = True
936 isRhsEnv (OccEnv OccVanilla _) = False
938 setVanillaCtxt :: OccEnv -> OccEnv
939 setVanillaCtxt (OccEnv OccRhs ctxt_ty) = OccEnv OccVanilla ctxt_ty
940 setVanillaCtxt other_env = other_env
942 setCtxt :: OccEnv -> CtxtTy -> OccEnv
943 setCtxt (OccEnv encl _) ctxt = OccEnv encl ctxt
945 oneShotGroup :: OccEnv -> [CoreBndr] -> [CoreBndr]
946 -- The result binders have one-shot-ness set that they might not have had originally.
947 -- This happens in (build (\cn -> e)). Here the occurrence analyser
948 -- linearity context knows that c,n are one-shot, and it records that fact in
949 -- the binder. This is useful to guide subsequent float-in/float-out tranformations
951 oneShotGroup (OccEnv encl ctxt) bndrs
954 go ctxt [] rev_bndrs = reverse rev_bndrs
956 go (lin_ctxt:ctxt) (bndr:bndrs) rev_bndrs
957 | isId bndr = go ctxt bndrs (bndr':rev_bndrs)
959 bndr' | lin_ctxt = setOneShotLambda bndr
962 go ctxt (bndr:bndrs) rev_bndrs = go ctxt bndrs (bndr:rev_bndrs)
964 addAppCtxt (OccEnv encl ctxt) args
965 = OccEnv encl (replicate (valArgCount args) True ++ ctxt)
968 %************************************************************************
970 \subsection[OccurAnal-types]{OccEnv}
972 %************************************************************************
975 type UsageDetails = IdEnv OccInfo -- A finite map from ids to their usage
977 (+++), combineAltsUsageDetails
978 :: UsageDetails -> UsageDetails -> UsageDetails
981 = plusVarEnv_C addOccInfo usage1 usage2
983 combineAltsUsageDetails usage1 usage2
984 = plusVarEnv_C orOccInfo usage1 usage2
986 addOneOcc :: UsageDetails -> Id -> OccInfo -> UsageDetails
987 addOneOcc usage id info
988 = plusVarEnv_C addOccInfo usage (unitVarEnv id info)
989 -- ToDo: make this more efficient
991 emptyDetails = (emptyVarEnv :: UsageDetails)
993 usedIn :: Id -> UsageDetails -> Bool
994 v `usedIn` details = isExportedId v || v `elemVarEnv` details
996 type IdWithOccInfo = Id
998 tagBinders :: UsageDetails -- Of scope
1000 -> (UsageDetails, -- Details with binders removed
1001 [IdWithOccInfo]) -- Tagged binders
1003 tagBinders usage binders
1005 usage' = usage `delVarEnvList` binders
1006 uss = map (setBinderOcc usage) binders
1008 usage' `seq` (usage', uss)
1010 tagBinder :: UsageDetails -- Of scope
1012 -> (UsageDetails, -- Details with binders removed
1013 IdWithOccInfo) -- Tagged binders
1015 tagBinder usage binder
1017 usage' = usage `delVarEnv` binder
1018 binder' = setBinderOcc usage binder
1020 usage' `seq` (usage', binder')
1022 setBinderOcc :: UsageDetails -> CoreBndr -> CoreBndr
1023 setBinderOcc usage bndr
1024 | isTyVar bndr = bndr
1025 | isExportedId bndr = case idOccInfo bndr of
1027 other -> setIdOccInfo bndr NoOccInfo
1028 -- Don't use local usage info for visible-elsewhere things
1029 -- BUT *do* erase any IAmALoopBreaker annotation, because we're
1030 -- about to re-generate it and it shouldn't be "sticky"
1032 | otherwise = setIdOccInfo bndr occ_info
1034 occ_info = lookupVarEnv usage bndr `orElse` IAmDead
1038 %************************************************************************
1040 \subsection{Operations over OccInfo}
1042 %************************************************************************
1045 mkOneOcc :: OccEnv -> Id -> InterestingCxt -> UsageDetails
1046 mkOneOcc env id int_cxt
1047 | isLocalId id = unitVarEnv id (OneOcc False True int_cxt)
1048 | otherwise = emptyDetails
1050 markMany, markInsideLam, markInsideSCC :: OccInfo -> OccInfo
1052 markMany IAmDead = IAmDead
1053 markMany other = NoOccInfo
1055 markInsideSCC occ = markMany occ
1057 markInsideLam (OneOcc _ one_br int_cxt) = OneOcc True one_br int_cxt
1058 markInsideLam occ = occ
1060 addOccInfo, orOccInfo :: OccInfo -> OccInfo -> OccInfo
1062 addOccInfo IAmDead info2 = info2
1063 addOccInfo info1 IAmDead = info1
1064 addOccInfo info1 info2 = NoOccInfo
1066 -- (orOccInfo orig new) is used
1067 -- when combining occurrence info from branches of a case
1069 orOccInfo IAmDead info2 = info2
1070 orOccInfo info1 IAmDead = info1
1071 orOccInfo (OneOcc in_lam1 one_branch1 int_cxt1)
1072 (OneOcc in_lam2 one_branch2 int_cxt2)
1073 = OneOcc (in_lam1 || in_lam2)
1074 False -- False, because it occurs in both branches
1075 (int_cxt1 && int_cxt2)
1076 orOccInfo info1 info2 = NoOccInfo