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 occurAnalysePgm, occurAnalyseExpr
18 #include "HsVersions.h"
22 import CoreUtils ( exprIsTrivial, isDefaultAlt )
23 import Coercion ( mkSymCoercion )
31 import Maybes ( orElse )
32 import Digraph ( SCC(..), stronglyConnCompFromEdgedVerticesR )
33 import PrelNames ( buildIdKey, foldrIdKey, runSTRepIdKey, augmentIdKey )
34 import Unique ( Unique )
35 import UniqFM ( keysUFM, intersectUFM_C, foldUFM_Directly )
36 import Util ( mapAndUnzip )
43 %************************************************************************
45 \subsection[OccurAnal-main]{Counting occurrences: main function}
47 %************************************************************************
49 Here's the externally-callable interface:
52 occurAnalysePgm :: [CoreBind] -> [CoreBind]
54 = snd (go initOccEnv binds)
56 go :: OccEnv -> [CoreBind] -> (UsageDetails, [CoreBind])
60 = (final_usage, bind' ++ binds')
62 (bs_usage, binds') = go env binds
63 (final_usage, bind') = occAnalBind env bind bs_usage
65 occurAnalyseExpr :: CoreExpr -> CoreExpr
66 -- Do occurrence analysis, and discard occurence info returned
67 occurAnalyseExpr expr = snd (occAnal initOccEnv expr)
71 %************************************************************************
73 \subsection[OccurAnal-main]{Counting occurrences: main function}
75 %************************************************************************
83 -> UsageDetails -- Usage details of scope
84 -> (UsageDetails, -- Of the whole let(rec)
87 occAnalBind env (NonRec binder rhs) body_usage
88 | isTyVar binder -- A type let; we don't gather usage info
89 = (body_usage, [NonRec binder rhs])
91 | not (binder `usedIn` body_usage) -- It's not mentioned
94 | otherwise -- It's mentioned in the body
95 = (body_usage' +++ addRuleUsage rhs_usage binder, -- Note [Rules are extra RHSs]
96 [NonRec tagged_binder rhs'])
98 (body_usage', tagged_binder) = tagBinder body_usage binder
99 (rhs_usage, rhs') = occAnalRhs env tagged_binder rhs
104 Dropping dead code for recursive bindings is done in a very simple way:
106 the entire set of bindings is dropped if none of its binders are
107 mentioned in its body; otherwise none are.
109 This seems to miss an obvious improvement.
121 Now 'f' is unused! But it's OK! Dependency analysis will sort this
122 out into a letrec for 'g' and a 'let' for 'f', and then 'f' will get
123 dropped. It isn't easy to do a perfect job in one blow. Consider
134 Note [Loop breaking and RULES]
135 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
136 Loop breaking is surprisingly subtle. First read the section 4 of
137 "Secrets of the GHC inliner". This describes our basic plan.
139 However things are made quite a bit more complicated by RULES. Remember
141 * Note [Rules are extra RHSs]
142 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
143 A RULE for 'f' is like an extra RHS for 'f'. That way the "parent"
144 keeps the specialised "children" alive. If the parent dies
145 (because it isn't referenced any more), then the children will die
146 too (unless they are already referenced directly).
148 To that end, we build a Rec group for each cyclic strongly
150 *treating f's rules as extra RHSs for 'f'*.
152 When we make the Rec groups we include variables free in *either*
153 LHS *or* RHS of the rule. The former might seems silly, but see
154 Note [Rule dependency info].
156 So in Example [eftInt], eftInt and eftIntFB will be put in the
157 same Rec, even though their 'main' RHSs are both non-recursive.
159 * Note [Rules are visible in their own rec group]
160 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
161 We want the rules for 'f' to be visible in f's right-hand side.
162 And we'd like them to be visible in other functions in f's Rec
163 group. E.g. in Example [Specialisation rules] we want f' rule
164 to be visible in both f's RHS, and fs's RHS.
166 This means that we must simplify the RULEs first, before looking
167 at any of the definitions. This is done by Simplify.simplRecBind,
168 when it calls addLetIdInfo.
170 * Note [Choosing loop breakers]
171 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
172 We avoid infinite inlinings by choosing loop breakers, and
173 ensuring that a loop breaker cuts each loop. But what is a
174 "loop"? In particular, a RULES is like an equation for 'f' that
175 is *always* inlined if it are applicable. We do *not* disable
176 rules for loop-breakers. It's up to whoever makes the rules to
177 make sure that the rules themselves alwasys terminate. See Note
178 [Rules for recursive functions] in Simplify.lhs
181 f's RHS mentions g, and
182 g has a RULE that mentions h, and
183 h has a RULE that mentions f
185 then we *must* choose f to be a loop breaker. In general, take the
186 free variables of f's RHS, and augment it with all the variables
187 reachable by RULES from those starting points. That is the whole
188 reason for computing rule_fv_env in occAnalBind. (Of course we
189 only consider free vars that are also binders in this Rec group.)
191 Note that when we compute this rule_fv_env, we only consider variables
192 free in the *RHS* of the rule, in contrast to the way we build the
193 Rec group in the first place (Note [Rule dependency info])
195 Note that in Example [eftInt], *neither* eftInt *nor* eftIntFB is
196 chosen as a loop breaker, because their RHSs don't mention each other.
197 And indeed both can be inlined safely.
199 Note that the edges of the graph we use for computing loop breakers
200 are not the same as the edges we use for computing the Rec blocks.
201 That's why we compute
202 rec_edges for the Rec block analysis
203 loop_breaker_edges for the loop breaker analysis
206 * Note [Weak loop breakers]
207 ~~~~~~~~~~~~~~~~~~~~~~~~~
208 There is a last nasty wrinkle. Suppose we have
218 Remmber that we simplify the RULES before any RHS (see Note
219 [Rules are visible in their own rec group] above).
221 So we must *not* postInlineUnconditionally 'g', even though
222 its RHS turns out to be trivial. (I'm assuming that 'g' is
223 not choosen as a loop breaker.)
225 We "solve" this by making g a "weak" or "rules-only" loop breaker,
226 with OccInfo = IAmLoopBreaker True. A normal "strong" loop breaker
227 has IAmLoopBreaker False. So
229 Inline postInlineUnconditinoally
230 IAmLoopBreaker False no no
231 IAmLoopBreaker True yes no
234 The **sole** reason for this kind of loop breaker is so that
235 postInlineUnconditionally does not fire. Ugh.
237 * Note [Rule dependency info]
238 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
239 The VarSet in a SpecInfo is used for dependency analysis in the
240 occurrence analyser. We must track free vars in *both* lhs and rhs. Why both?
244 Then if we substitute y for x, we'd better do so in the
245 rule's LHS too, so we'd better ensure the dependency is respected
250 Example (from GHC.Enum):
252 eftInt :: Int# -> Int# -> [Int]
253 eftInt x y = ...(non-recursive)...
255 {-# INLINE [0] eftIntFB #-}
256 eftIntFB :: (Int -> r -> r) -> r -> Int# -> Int# -> r
257 eftIntFB c n x y = ...(non-recursive)...
260 "eftInt" [~1] forall x y. eftInt x y = build (\ c n -> eftIntFB c n x y)
261 "eftIntList" [1] eftIntFB (:) [] = eftInt
264 Example [Specialisation rules]
265 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
266 Consider this group, which is typical of what SpecConstr builds:
268 fs a = ....f (C a)....
269 f x = ....f (C a)....
270 {-# RULE f (C a) = fs a #-}
272 So 'f' and 'fs' are in the same Rec group (since f refers to fs via its RULE).
274 But watch out! If 'fs' is not chosen as a loop breaker, we may get an infinite loop:
275 - the RULE is applied in f's RHS (see Note [Self-recursive rules] in Simplify
276 - fs is inlined (say it's small)
277 - now there's another opportunity to apply the RULE
279 This showed up when compiling Control.Concurrent.Chan.getChanContents.
283 occAnalBind env (Rec pairs) body_usage
284 = foldr occAnalRec (body_usage, []) sccs
285 -- For a recursive group, we
286 -- * occ-analyse all the RHSs
287 -- * compute strongly-connected components
288 -- * feed those components to occAnalRec
290 -------------Dependency analysis ------------------------------
291 bndr_set = mkVarSet (map fst pairs)
293 sccs :: [SCC (Node Details)]
294 sccs = {-# SCC "occAnalBind.scc" #-} stronglyConnCompFromEdgedVerticesR rec_edges
296 rec_edges :: [Node Details]
297 rec_edges = {-# SCC "occAnalBind.assoc" #-} map make_node pairs
299 make_node (bndr, rhs)
300 = (ND bndr rhs' rhs_usage rhs_fvs, idUnique bndr, out_edges)
302 (rhs_usage, rhs') = occAnalRhs env bndr rhs
303 rhs_fvs = intersectUFM_C (\b _ -> b) bndr_set rhs_usage
304 out_edges = keysUFM (rhs_fvs `unionVarSet` idRuleVars bndr)
305 -- (a -> b) means a mentions b
306 -- Given the usage details (a UFM that gives occ info for each free var of
307 -- the RHS) we can get the list of free vars -- or rather their Int keys --
308 -- by just extracting the keys from the finite map. Grimy, but fast.
309 -- Previously we had this:
310 -- [ bndr | bndr <- bndrs,
311 -- maybeToBool (lookupVarEnv rhs_usage bndr)]
312 -- which has n**2 cost, and this meant that edges_from alone
313 -- consumed 10% of total runtime!
315 -----------------------------
316 occAnalRec :: SCC (Node Details) -> (UsageDetails, [CoreBind])
317 -> (UsageDetails, [CoreBind])
319 -- The NonRec case is just like a Let (NonRec ...) above
320 occAnalRec (AcyclicSCC (ND bndr rhs rhs_usage _, _, _)) (body_usage, binds)
321 | not (bndr `usedIn` body_usage)
322 = (body_usage, binds)
324 | otherwise -- It's mentioned in the body
325 = (body_usage' +++ addRuleUsage rhs_usage bndr, -- Note [Rules are extra RHSs]
326 NonRec tagged_bndr rhs : binds)
328 (body_usage', tagged_bndr) = tagBinder body_usage bndr
331 -- The Rec case is the interesting one
332 -- See Note [Loop breaking]
333 occAnalRec (CyclicSCC nodes) (body_usage, binds)
334 | not (any (`usedIn` body_usage) bndrs) -- NB: look at body_usage, not total_usage
335 = (body_usage, binds) -- Dead code
337 | otherwise -- At this point we always build a single Rec
338 = (final_usage, Rec pairs : binds)
341 bndrs = [b | (ND b _ _ _, _, _) <- nodes]
342 bndr_set = mkVarSet bndrs
344 ----------------------------
345 -- Tag the binders with their occurrence info
346 total_usage = foldl add_usage body_usage nodes
347 add_usage body_usage (ND bndr _ rhs_usage _, _, _)
348 = body_usage +++ addRuleUsage rhs_usage bndr
349 (final_usage, tagged_nodes) = mapAccumL tag_node total_usage nodes
351 tag_node :: UsageDetails -> Node Details -> (UsageDetails, Node Details)
352 -- (a) Tag the binders in the details with occ info
353 -- (b) Mark the binder with "weak loop-breaker" OccInfo
354 -- saying "no preInlineUnconditionally" if it is used
355 -- in any rule (lhs or rhs) of the recursive group
356 -- See Note [Weak loop breakers]
357 tag_node usage (ND bndr rhs rhs_usage rhs_fvs, k, ks)
358 = (usage `delVarEnv` bndr, (ND bndr2 rhs rhs_usage rhs_fvs, k, ks))
360 bndr2 | bndr `elemVarSet` all_rule_fvs = makeLoopBreaker True bndr1
362 bndr1 = setBinderOcc usage bndr
363 all_rule_fvs = bndr_set `intersectVarSet` foldr (unionVarSet . idRuleVars)
366 ----------------------------
367 -- Now reconstruct the cycle
368 pairs | no_rules = reOrderCycle tagged_nodes
369 | otherwise = concatMap reOrderRec (stronglyConnCompFromEdgedVerticesR loop_breaker_edges)
371 -- See Note [Choosing loop breakers] for looop_breaker_edges
372 loop_breaker_edges = map mk_node tagged_nodes
373 mk_node (details@(ND _ _ _ rhs_fvs), k, _) = (details, k, new_ks)
375 new_ks = keysUFM (extendFvs rule_fv_env rhs_fvs rhs_fvs)
377 ------------------------------------
378 rule_fv_env :: IdEnv IdSet -- Variables from this group mentioned in RHS of rules
379 -- Domain is *subset* of bound vars (others have no rule fvs)
380 rule_fv_env = rule_loop init_rule_fvs
382 no_rules = null init_rule_fvs
383 init_rule_fvs = [(b, rule_fvs)
385 , let rule_fvs = idRuleRhsVars b `intersectVarSet` bndr_set
386 , not (isEmptyVarSet rule_fvs)]
388 rule_loop :: [(Id,IdSet)] -> IdEnv IdSet -- Finds fixpoint
391 | otherwise = rule_loop new_fv_list
393 env = mkVarEnv init_rule_fvs
394 (no_change, new_fv_list) = mapAccumL bump True fv_list
395 bump no_change (b,fvs)
396 | new_fvs `subVarSet` fvs = (no_change, (b,fvs))
397 | otherwise = (False, (b,new_fvs `unionVarSet` fvs))
399 new_fvs = extendFvs env emptyVarSet fvs
401 idRuleRhsVars :: Id -> VarSet
402 -- Just the variables free on the *rhs* of a rule
403 -- See Note [Choosing loop breakers]
404 idRuleRhsVars id = foldr (unionVarSet . ruleRhsFreeVars) emptyVarSet (idCoreRules id)
406 extendFvs :: IdEnv IdSet -> IdSet -> IdSet -> IdSet
407 -- (extendFVs env fvs s) returns (fvs `union` env(s))
408 extendFvs env fvs id_set
409 = foldUFM_Directly add fvs id_set
412 = case lookupVarEnv_Directly env uniq of
413 Just fvs' -> fvs' `unionVarSet` fvs
417 @reOrderRec@ is applied to the list of (binder,rhs) pairs for a cyclic
418 strongly connected component (there's guaranteed to be a cycle). It returns the
420 a) in a better order,
421 b) with some of the Ids having a IAmALoopBreaker pragma
423 The "loop-breaker" Ids are sufficient to break all cycles in the SCC. This means
424 that the simplifier can guarantee not to loop provided it never records an inlining
425 for these no-inline guys.
427 Furthermore, the order of the binds is such that if we neglect dependencies
428 on the no-inline Ids then the binds are topologically sorted. This means
429 that the simplifier will generally do a good job if it works from top bottom,
430 recording inlinings for any Ids which aren't marked as "no-inline" as it goes.
433 [June 98: I don't understand the following paragraphs, and I've
434 changed the a=b case again so that it isn't a special case any more.]
436 Here's a case that bit me:
444 Re-ordering doesn't change the order of bindings, but there was no loop-breaker.
446 My solution was to make a=b bindings record b as Many, rather like INLINE bindings.
447 Perhaps something cleverer would suffice.
452 type Node details = (details, Unique, [Unique]) -- The Ints are gotten from the Unique,
453 -- which is gotten from the Id.
454 data Details = ND Id -- Binder
456 UsageDetails -- Full usage from RHS (*not* including rules)
457 IdSet -- Other binders from this Rec group mentioned on RHS
458 -- (derivable from UsageDetails but cached here)
460 reOrderRec :: SCC (Node Details)
462 -- Sorted into a plausible order. Enough of the Ids have
463 -- IAmALoopBreaker pragmas that there are no loops left.
464 reOrderRec (AcyclicSCC (ND bndr rhs _ _, _, _)) = [(bndr, rhs)]
465 reOrderRec (CyclicSCC cycle) = reOrderCycle cycle
467 reOrderCycle :: [Node Details] -> [(Id,CoreExpr)]
469 = panic "reOrderCycle"
470 reOrderCycle [bind] -- Common case of simple self-recursion
471 = [(makeLoopBreaker False bndr, rhs)]
473 (ND bndr rhs _ _, _, _) = bind
475 reOrderCycle (bind : binds)
476 = -- Choose a loop breaker, mark it no-inline,
477 -- do SCC analysis on the rest, and recursively sort them out
478 concatMap reOrderRec (stronglyConnCompFromEdgedVerticesR unchosen) ++
479 [(makeLoopBreaker False bndr, rhs)]
482 (chosen_bind, unchosen) = choose_loop_breaker bind (score bind) [] binds
483 ND bndr rhs _ _ = chosen_bind
485 -- This loop looks for the bind with the lowest score
486 -- to pick as the loop breaker. The rest accumulate in
487 choose_loop_breaker (details,_,_) _loop_sc acc []
488 = (details, acc) -- Done
490 choose_loop_breaker loop_bind loop_sc acc (bind : binds)
491 | sc < loop_sc -- Lower score so pick this new one
492 = choose_loop_breaker bind sc (loop_bind : acc) binds
494 | otherwise -- No lower so don't pick it
495 = choose_loop_breaker loop_bind loop_sc (bind : acc) binds
499 score :: Node Details -> Int -- Higher score => less likely to be picked as loop breaker
500 score (ND bndr rhs _ _, _, _)
501 | workerExists (idWorkerInfo bndr) = 10
502 -- Note [Worker inline loop]
504 | exprIsTrivial rhs = 5 -- Practically certain to be inlined
505 -- Used to have also: && not (isExportedId bndr)
506 -- But I found this sometimes cost an extra iteration when we have
507 -- rec { d = (a,b); a = ...df...; b = ...df...; df = d }
508 -- where df is the exported dictionary. Then df makes a really
509 -- bad choice for loop breaker
511 | is_con_app rhs = 3 -- Data types help with cases
514 -- If an Id is marked "never inline" then it makes a great loop breaker
515 -- The only reason for not checking that here is that it is rare
516 -- and I've never seen a situation where it makes a difference,
517 -- so it probably isn't worth the time to test on every binder
518 -- | isNeverActive (idInlinePragma bndr) = -10
520 | inlineCandidate bndr rhs = 2 -- Likely to be inlined
521 -- Note [Inline candidates]
523 | not (neverUnfold (idUnfolding bndr)) = 1
524 -- the Id has some kind of unfolding
528 inlineCandidate :: Id -> CoreExpr -> Bool
529 inlineCandidate _ (Note InlineMe _) = True
530 inlineCandidate id _ = isOneOcc (idOccInfo id)
534 -- It's really really important to inline dictionaries. Real
535 -- example (the Enum Ordering instance from GHC.Base):
537 -- rec f = \ x -> case d of (p,q,r) -> p x
538 -- g = \ x -> case d of (p,q,r) -> q x
541 -- Here, f and g occur just once; but we can't inline them into d.
542 -- On the other hand we *could* simplify those case expressions if
543 -- we didn't stupidly choose d as the loop breaker.
544 -- But we won't because constructor args are marked "Many".
545 -- Inlining dictionaries is really essential to unravelling
546 -- the loops in static numeric dictionaries, see GHC.Float.
548 -- Cheap and cheerful; the simplifer moves casts out of the way
549 -- The lambda case is important to spot x = /\a. C (f a)
550 -- which comes up when C is a dictionary constructor and
551 -- f is a default method.
552 -- Example: the instance for Show (ST s a) in GHC.ST
554 -- However we *also* treat (\x. C p q) as a con-app-like thing,
555 -- Note [Closure conversion]
556 is_con_app (Var v) = isDataConWorkId v
557 is_con_app (App f _) = is_con_app f
558 is_con_app (Lam _ e) = is_con_app e
559 is_con_app (Note _ e) = is_con_app e
562 makeLoopBreaker :: Bool -> Id -> Id
563 -- Set the loop-breaker flag
564 -- See Note [Weak loop breakers]
565 makeLoopBreaker weak bndr = setIdOccInfo bndr (IAmALoopBreaker weak)
568 Note [Worker inline loop]
569 ~~~~~~~~~~~~~~~~~~~~~~~~
570 Never choose a wrapper as the loop breaker! Because
571 wrappers get auto-generated inlinings when importing, and
572 that can lead to an infinite inlining loop. For example:
574 $wfoo x = ....foo x....
576 {-loop brk-} foo x = ...$wfoo x...
579 The interface file sees the unfolding for $wfoo, and sees that foo is
580 strict (and hence it gets an auto-generated wrapper). Result: an
581 infinite inlining in the importing scope. So be a bit careful if you
582 change this. A good example is Tree.repTree in
583 nofib/spectral/minimax. If the repTree wrapper is chosen as the loop
584 breaker then compiling Game.hs goes into an infinite loop (this
585 happened when we gave is_con_app a lower score than inline candidates).
587 Note [Closure conversion]
588 ~~~~~~~~~~~~~~~~~~~~~~~~~
589 We treat (\x. C p q) as a high-score candidate in the letrec scoring algorithm.
590 The immediate motivation came from the result of a closure-conversion transformation
591 which generated code like this:
593 data Clo a b = forall c. Clo (c -> a -> b) c
595 ($:) :: Clo a b -> a -> b
596 Clo f env $: x = f env x
598 rec { plus = Clo plus1 ()
600 ; plus1 _ n = Clo plus2 n
603 ; plus2 (Succ m) n = Succ (plus $: m $: n) }
605 If we inline 'plus' and 'plus1', everything unravels nicely. But if
606 we choose 'plus1' as the loop breaker (which is entirely possible
607 otherwise), the loop does not unravel nicely.
610 @occAnalRhs@ deals with the question of bindings where the Id is marked
611 by an INLINE pragma. For these we record that anything which occurs
612 in its RHS occurs many times. This pessimistically assumes that ths
613 inlined binder also occurs many times in its scope, but if it doesn't
614 we'll catch it next time round. At worst this costs an extra simplifier pass.
615 ToDo: try using the occurrence info for the inline'd binder.
617 [March 97] We do the same for atomic RHSs. Reason: see notes with reOrderRec.
618 [June 98, SLPJ] I've undone this change; I don't understand it. See notes with reOrderRec.
623 -> Id -> CoreExpr -- Binder and rhs
624 -- For non-recs the binder is alrady tagged
625 -- with occurrence info
626 -> (UsageDetails, CoreExpr)
628 occAnalRhs env id rhs
631 ctxt | certainly_inline id = env
632 | otherwise = rhsCtxt env
633 -- Note that we generally use an rhsCtxt. This tells the occ anal n
634 -- that it's looking at an RHS, which has an effect in occAnalApp
636 -- But there's a problem. Consider
641 -- First time round, it looks as if x1 and x2 occur as an arg of a
642 -- let-bound constructor ==> give them a many-occurrence.
643 -- But then x3 is inlined (unconditionally as it happens) and
644 -- next time round, x2 will be, and the next time round x1 will be
645 -- Result: multiple simplifier iterations. Sigh.
646 -- Crude solution: use rhsCtxt for things that occur just once...
648 certainly_inline id = case idOccInfo id of
649 OneOcc in_lam one_br _ -> not in_lam && one_br
656 addRuleUsage :: UsageDetails -> Id -> UsageDetails
657 -- Add the usage from RULES in Id to the usage
658 addRuleUsage usage id
659 = foldVarSet add usage (idRuleVars id)
661 add v u = addOneOcc u v NoOccInfo -- Give a non-committal binder info
662 -- (i.e manyOcc) because many copies
663 -- of the specialised thing can appear
671 -> (UsageDetails, -- Gives info only about the "interesting" Ids
674 occAnal _ (Type t) = (emptyDetails, Type t)
675 occAnal env (Var v) = (mkOneOcc env v False, Var v)
676 -- At one stage, I gathered the idRuleVars for v here too,
677 -- which in a way is the right thing to do.
678 -- But that went wrong right after specialisation, when
679 -- the *occurrences* of the overloaded function didn't have any
680 -- rules in them, so the *specialised* versions looked as if they
681 -- weren't used at all.
684 We regard variables that occur as constructor arguments as "dangerousToDup":
688 f x = let y = expensive x in
690 (case z of {(p,q)->q}, case z of {(p,q)->q})
693 We feel free to duplicate the WHNF (True,y), but that means
694 that y may be duplicated thereby.
696 If we aren't careful we duplicate the (expensive x) call!
697 Constructors are rather like lambdas in this way.
700 occAnal _ expr@(Lit _) = (emptyDetails, expr)
704 occAnal env (Note InlineMe body)
705 = case occAnal env body of { (usage, body') ->
706 (mapVarEnv markMany usage, Note InlineMe body')
709 occAnal env (Note note@(SCC _) body)
710 = case occAnal env body of { (usage, body') ->
711 (mapVarEnv markInsideSCC usage, Note note body')
714 occAnal env (Note note body)
715 = case occAnal env body of { (usage, body') ->
716 (usage, Note note body')
719 occAnal env (Cast expr co)
720 = case occAnal env expr of { (usage, expr') ->
721 (markRhsUds env True usage, Cast expr' co)
722 -- If we see let x = y `cast` co
723 -- then mark y as 'Many' so that we don't
724 -- immediately inline y again.
729 occAnal env app@(App _ _)
730 = occAnalApp env (collectArgs app)
732 -- Ignore type variables altogether
733 -- (a) occurrences inside type lambdas only not marked as InsideLam
734 -- (b) type variables not in environment
736 occAnal env (Lam x body) | isTyVar x
737 = case occAnal env body of { (body_usage, body') ->
738 (body_usage, Lam x body')
741 -- For value lambdas we do a special hack. Consider
743 -- If we did nothing, x is used inside the \y, so would be marked
744 -- as dangerous to dup. But in the common case where the abstraction
745 -- is applied to two arguments this is over-pessimistic.
746 -- So instead, we just mark each binder with its occurrence
747 -- info in the *body* of the multiple lambda.
748 -- Then, the simplifier is careful when partially applying lambdas.
750 occAnal env expr@(Lam _ _)
751 = case occAnal env_body body of { (body_usage, body') ->
753 (final_usage, tagged_binders) = tagBinders body_usage binders
754 -- URGH! Sept 99: we don't seem to be able to use binders' here, because
755 -- we get linear-typed things in the resulting program that we can't handle yet.
756 -- (e.g. PrelShow) TODO
758 really_final_usage = if linear then
761 mapVarEnv markInsideLam final_usage
764 mkLams tagged_binders body') }
766 env_body = vanillaCtxt env -- Body is (no longer) an RhsContext
767 (binders, body) = collectBinders expr
768 binders' = oneShotGroup env binders
769 linear = all is_one_shot binders'
770 is_one_shot b = isId b && isOneShotBndr b
772 occAnal env (Case scrut bndr ty alts)
773 = case occ_anal_scrut scrut alts of { (scrut_usage, scrut') ->
774 case mapAndUnzip occ_anal_alt alts of { (alts_usage_s, alts') ->
776 alts_usage = foldr1 combineAltsUsageDetails alts_usage_s
777 alts_usage' = addCaseBndrUsage alts_usage
778 (alts_usage1, tagged_bndr) = tagBinder alts_usage' bndr
779 total_usage = scrut_usage +++ alts_usage1
781 total_usage `seq` (total_usage, Case scrut' tagged_bndr ty alts') }}
783 -- Note [Case binder usage]
784 -- ~~~~~~~~~~~~~~~~~~~~~~~~
785 -- The case binder gets a usage of either "many" or "dead", never "one".
786 -- Reason: we like to inline single occurrences, to eliminate a binding,
787 -- but inlining a case binder *doesn't* eliminate a binding.
788 -- We *don't* want to transform
789 -- case x of w { (p,q) -> f w }
791 -- case x of w { (p,q) -> f (p,q) }
792 addCaseBndrUsage usage = case lookupVarEnv usage bndr of
794 Just _ -> extendVarEnv usage bndr NoOccInfo
796 alt_env = mkAltEnv env bndr_swap
797 -- Consider x = case v of { True -> (p,q); ... }
798 -- Then it's fine to inline p and q
800 bndr_swap = case scrut of
801 Var v -> Just (v, Var bndr)
802 Cast (Var v) co -> Just (v, Cast (Var bndr) (mkSymCoercion co))
805 occ_anal_alt = occAnalAlt alt_env bndr bndr_swap
807 occ_anal_scrut (Var v) (alt1 : other_alts)
808 | not (null other_alts) || not (isDefaultAlt alt1)
809 = (mkOneOcc env v True, Var v) -- The 'True' says that the variable occurs
810 -- in an interesting context; the case has
811 -- at least one non-default alternative
812 occ_anal_scrut scrut _alts
813 = occAnal (vanillaCtxt env) scrut -- No need for rhsCtxt
815 occAnal env (Let bind body)
816 = case occAnal env body of { (body_usage, body') ->
817 case occAnalBind env bind body_usage of { (final_usage, new_binds) ->
818 (final_usage, mkLets new_binds body') }}
820 occAnalArgs :: OccEnv -> [CoreExpr] -> (UsageDetails, [CoreExpr])
822 = case mapAndUnzip (occAnal arg_env) args of { (arg_uds_s, args') ->
823 (foldr (+++) emptyDetails arg_uds_s, args')}
825 arg_env = vanillaCtxt env
828 Applications are dealt with specially because we want
829 the "build hack" to work.
833 -> (Expr CoreBndr, [Arg CoreBndr])
834 -> (UsageDetails, Expr CoreBndr)
835 occAnalApp env (Var fun, args)
836 = case args_stuff of { (args_uds, args') ->
838 final_args_uds = markRhsUds env is_pap args_uds
840 (fun_uds +++ final_args_uds, mkApps (Var fun) args') }
842 fun_uniq = idUnique fun
843 fun_uds = mkOneOcc env fun (valArgCount args > 0)
844 is_pap = isDataConWorkId fun || valArgCount args < idArity fun
846 -- Hack for build, fold, runST
847 args_stuff | fun_uniq == buildIdKey = appSpecial env 2 [True,True] args
848 | fun_uniq == augmentIdKey = appSpecial env 2 [True,True] args
849 | fun_uniq == foldrIdKey = appSpecial env 3 [False,True] args
850 | fun_uniq == runSTRepIdKey = appSpecial env 2 [True] args
851 -- (foldr k z xs) may call k many times, but it never
852 -- shares a partial application of k; hence [False,True]
853 -- This means we can optimise
854 -- foldr (\x -> let v = ...x... in \y -> ...v...) z xs
855 -- by floating in the v
857 | otherwise = occAnalArgs env args
860 occAnalApp env (fun, args)
861 = case occAnal (addAppCtxt env args) fun of { (fun_uds, fun') ->
862 -- The addAppCtxt is a bit cunning. One iteration of the simplifier
863 -- often leaves behind beta redexs like
865 -- Here we would like to mark x,y as one-shot, and treat the whole
866 -- thing much like a let. We do this by pushing some True items
867 -- onto the context stack.
869 case occAnalArgs env args of { (args_uds, args') ->
871 final_uds = fun_uds +++ args_uds
873 (final_uds, mkApps fun' args') }}
876 markRhsUds :: OccEnv -- Check if this is a RhsEnv
877 -> Bool -- and this is true
878 -> UsageDetails -- The do markMany on this
880 -- We mark the free vars of the argument of a constructor or PAP
881 -- as "many", if it is the RHS of a let(rec).
882 -- This means that nothing gets inlined into a constructor argument
883 -- position, which is what we want. Typically those constructor
884 -- arguments are just variables, or trivial expressions.
886 -- This is the *whole point* of the isRhsEnv predicate
887 markRhsUds env is_pap arg_uds
888 | isRhsEnv env && is_pap = mapVarEnv markMany arg_uds
889 | otherwise = arg_uds
893 -> Int -> CtxtTy -- Argument number, and context to use for it
895 -> (UsageDetails, [CoreExpr])
896 appSpecial env n ctxt args
899 arg_env = vanillaCtxt env
901 go _ [] = (emptyDetails, []) -- Too few args
903 go 1 (arg:args) -- The magic arg
904 = case occAnal (setCtxtTy arg_env ctxt) arg of { (arg_uds, arg') ->
905 case occAnalArgs env args of { (args_uds, args') ->
906 (arg_uds +++ args_uds, arg':args') }}
909 = case occAnal arg_env arg of { (arg_uds, arg') ->
910 case go (n-1) args of { (args_uds, args') ->
911 (arg_uds +++ args_uds, arg':args') }}
917 We do these two transformations right here:
919 (1) case x of b { pi -> ri }
921 case x of b { pi -> let x=b in ri }
923 (2) case (x |> co) of b { pi -> ri }
925 case (x |> co) of b { pi -> let x = b |> sym co in ri }
927 Why (2)? See Note [Case of cast]
929 In both cases, in a particular alternative (pi -> ri), we only
931 (a) x occurs free in (pi -> ri)
932 (ie it occurs in ri, but is not bound in pi)
933 (b) the pi does not bind b (or the free vars of co)
934 We need (a) and (b) for the inserted binding to be correct.
936 For the alternatives where we inject the binding, we can transfer
937 all x's OccInfo to b. And that is the point.
940 * The deliberate shadowing of 'x'.
941 * That (a) rapidly becomes false, so no bindings are injected.
943 The reason for doing these transformations here is because it allows
944 us to adjust the OccInfo for 'x' and 'b' as we go.
946 * Suppose the only occurrences of 'x' are the scrutinee and in the
947 ri; then this transformation makes it occur just once, and hence
948 get inlined right away.
950 * If we do this in the Simplifier, we don't know whether 'x' is used
951 in ri, so we are forced to pessimistically zap b's OccInfo even
952 though it is typically dead (ie neither it nor x appear in the
953 ri). There's nothing actually wrong with zapping it, except that
954 it's kind of nice to know which variables are dead. My nose
955 tells me to keep this information as robustly as possible.
957 The Maybe (Id,CoreExpr) passed to occAnalAlt is the extra let-binding
958 {x=b}; it's Nothing if the binder-swap doesn't happen.
960 Note [Binder swap on GlobalId scrutinees]
961 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
962 When the scrutinee is a GlobalId we must take care in two ways
964 i) In order to *know* whether 'x' occurs free in the RHS, we need its
965 occurrence info. BUT, we don't gather occurrence info for
966 GlobalIds. That's what the (small) occ_scrut_ids set in OccEnv is
967 for: it says "gather occurrence info for these.
969 ii) We must call localiseId on 'x' first, in case it's a GlobalId, or
970 has an External Name. See, for example, SimplEnv Note [Global Ids in
975 Consider case (x `cast` co) of b { I# ->
976 ... (case (x `cast` co) of {...}) ...
977 We'd like to eliminate the inner case. That is the motivation for
978 equation (2) in Note [Binder swap]. When we get to the inner case, we
979 inline x, cancel the casts, and away we go.
981 Note [Binders in case alternatives]
982 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
984 case x of y { (a,b) -> f y }
985 We treat 'a', 'b' as dead, because they don't physically occur in the
986 case alternative. (Indeed, a variable is dead iff it doesn't occur in
987 its scope in the output of OccAnal.) This invariant is It really
988 helpe to know when binders are unused. See esp the call to
989 isDeadBinder in Simplify.mkDupableAlt
991 In this example, though, the Simplifier will bring 'a' and 'b' back to
992 life, beause it binds 'y' to (a,b) (imagine got inlined and
998 -> Maybe (Id, CoreExpr) -- Note [Binder swap]
1000 -> (UsageDetails, Alt IdWithOccInfo)
1001 occAnalAlt env case_bndr mb_scrut_var (con, bndrs, rhs)
1002 = case occAnal env rhs of { (rhs_usage, rhs') ->
1004 (alt_usg, tagged_bndrs) = tagBinders rhs_usage bndrs
1005 bndrs' = tagged_bndrs -- See Note [Binders in case alternatives]
1007 case mb_scrut_var of
1008 Just (scrut_var, scrut_rhs) -- See Note [Binder swap]
1009 | scrut_var `localUsedIn` alt_usg -- (a) Fast path, usually false
1010 , not (any shadowing bndrs) -- (b)
1011 -> (addOneOcc usg_wo_scrut case_bndr NoOccInfo,
1012 -- See Note [Case binder usage] for the NoOccInfo
1013 (con, bndrs', Let (NonRec scrut_var' scrut_rhs) rhs'))
1015 (usg_wo_scrut, scrut_var') = tagBinder alt_usg (localiseId scrut_var)
1016 -- Note the localiseId; we're making a new binding
1017 -- for it, and it might have an External Name, or
1018 -- even be a GlobalId; Note [Binder swap on GlobalId scrutinees]
1019 shadowing bndr = bndr `elemVarSet` rhs_fvs
1020 rhs_fvs = exprFreeVars scrut_rhs
1022 _other -> (alt_usg, (con, bndrs', rhs')) }
1026 %************************************************************************
1028 \subsection[OccurAnal-types]{OccEnv}
1030 %************************************************************************
1034 = OccEnv { occ_encl :: !OccEncl -- Enclosing context information
1035 , occ_ctxt :: !CtxtTy -- Tells about linearity
1036 , occ_scrut_ids :: !GblScrutIds }
1038 type GblScrutIds = IdSet -- GlobalIds that are scrutinised, and for which
1039 -- we want to gather occurence info; see
1040 -- Note [Binder swap for GlobalId scrutinee]
1041 -- No need to prune this if there's a shadowing binding
1042 -- because it's OK for it to be too big
1044 -- OccEncl is used to control whether to inline into constructor arguments
1046 -- x = (p,q) -- Don't inline p or q
1047 -- y = /\a -> (p a, q a) -- Still don't inline p or q
1048 -- z = f (p,q) -- Do inline p,q; it may make a rule fire
1049 -- So OccEncl tells enought about the context to know what to do when
1050 -- we encounter a contructor application or PAP.
1053 = OccRhs -- RHS of let(rec), albeit perhaps inside a type lambda
1054 -- Don't inline into constructor args here
1055 | OccVanilla -- Argument of function, body of lambda, scruintee of case etc.
1056 -- Do inline into constructor args here
1058 type CtxtTy = [Bool]
1061 -- True:ctxt Analysing a function-valued expression that will be
1062 -- applied just once
1064 -- False:ctxt Analysing a function-valued expression that may
1065 -- be applied many times; but when it is,
1066 -- the CtxtTy inside applies
1068 initOccEnv :: OccEnv
1069 initOccEnv = OccEnv { occ_encl = OccRhs
1071 , occ_scrut_ids = emptyVarSet }
1073 vanillaCtxt :: OccEnv -> OccEnv
1074 vanillaCtxt env = OccEnv { occ_encl = OccVanilla, occ_ctxt = []
1075 , occ_scrut_ids = occ_scrut_ids env }
1077 rhsCtxt :: OccEnv -> OccEnv
1078 rhsCtxt env = OccEnv { occ_encl = OccRhs, occ_ctxt = []
1079 , occ_scrut_ids = occ_scrut_ids env }
1081 mkAltEnv :: OccEnv -> Maybe (Id, CoreExpr) -> OccEnv
1082 -- Does two things: a) makes the occ_ctxt = OccVanilla
1083 -- b) extends the scrut_ids if necessary
1084 mkAltEnv env (Just (scrut_id, _))
1085 | not (isLocalId scrut_id)
1086 = OccEnv { occ_encl = OccVanilla
1087 , occ_scrut_ids = extendVarSet (occ_scrut_ids env) scrut_id
1088 , occ_ctxt = occ_ctxt env }
1090 | isRhsEnv env = env { occ_encl = OccVanilla }
1093 setCtxtTy :: OccEnv -> CtxtTy -> OccEnv
1094 setCtxtTy env ctxt = env { occ_ctxt = ctxt }
1096 isRhsEnv :: OccEnv -> Bool
1097 isRhsEnv (OccEnv { occ_encl = OccRhs }) = True
1098 isRhsEnv (OccEnv { occ_encl = OccVanilla }) = False
1100 oneShotGroup :: OccEnv -> [CoreBndr] -> [CoreBndr]
1101 -- The result binders have one-shot-ness set that they might not have had originally.
1102 -- This happens in (build (\cn -> e)). Here the occurrence analyser
1103 -- linearity context knows that c,n are one-shot, and it records that fact in
1104 -- the binder. This is useful to guide subsequent float-in/float-out tranformations
1106 oneShotGroup (OccEnv { occ_ctxt = ctxt }) bndrs
1109 go _ [] rev_bndrs = reverse rev_bndrs
1111 go (lin_ctxt:ctxt) (bndr:bndrs) rev_bndrs
1112 | isId bndr = go ctxt bndrs (bndr':rev_bndrs)
1114 bndr' | lin_ctxt = setOneShotLambda bndr
1117 go ctxt (bndr:bndrs) rev_bndrs = go ctxt bndrs (bndr:rev_bndrs)
1119 addAppCtxt :: OccEnv -> [Arg CoreBndr] -> OccEnv
1120 addAppCtxt env@(OccEnv { occ_ctxt = ctxt }) args
1121 = env { occ_ctxt = replicate (valArgCount args) True ++ ctxt }
1124 %************************************************************************
1126 \subsection[OccurAnal-types]{OccEnv}
1128 %************************************************************************
1131 type UsageDetails = IdEnv OccInfo -- A finite map from ids to their usage
1132 -- INVARIANT: never IAmDead
1133 -- (Deadness is signalled by not being in the map at all)
1135 (+++), combineAltsUsageDetails
1136 :: UsageDetails -> UsageDetails -> UsageDetails
1139 = plusVarEnv_C addOccInfo usage1 usage2
1141 combineAltsUsageDetails usage1 usage2
1142 = plusVarEnv_C orOccInfo usage1 usage2
1144 addOneOcc :: UsageDetails -> Id -> OccInfo -> UsageDetails
1145 addOneOcc usage id info
1146 = plusVarEnv_C addOccInfo usage (unitVarEnv id info)
1147 -- ToDo: make this more efficient
1149 emptyDetails :: UsageDetails
1150 emptyDetails = (emptyVarEnv :: UsageDetails)
1152 localUsedIn, usedIn :: Id -> UsageDetails -> Bool
1153 v `localUsedIn` details = v `elemVarEnv` details
1154 v `usedIn` details = isExportedId v || v `localUsedIn` details
1156 type IdWithOccInfo = Id
1158 tagBinders :: UsageDetails -- Of scope
1160 -> (UsageDetails, -- Details with binders removed
1161 [IdWithOccInfo]) -- Tagged binders
1163 tagBinders usage binders
1165 usage' = usage `delVarEnvList` binders
1166 uss = map (setBinderOcc usage) binders
1168 usage' `seq` (usage', uss)
1170 tagBinder :: UsageDetails -- Of scope
1172 -> (UsageDetails, -- Details with binders removed
1173 IdWithOccInfo) -- Tagged binders
1175 tagBinder usage binder
1177 usage' = usage `delVarEnv` binder
1178 binder' = setBinderOcc usage binder
1180 usage' `seq` (usage', binder')
1182 setBinderOcc :: UsageDetails -> CoreBndr -> CoreBndr
1183 setBinderOcc usage bndr
1184 | isTyVar bndr = bndr
1185 | isExportedId bndr = case idOccInfo bndr of
1187 _ -> setIdOccInfo bndr NoOccInfo
1188 -- Don't use local usage info for visible-elsewhere things
1189 -- BUT *do* erase any IAmALoopBreaker annotation, because we're
1190 -- about to re-generate it and it shouldn't be "sticky"
1192 | otherwise = setIdOccInfo bndr occ_info
1194 occ_info = lookupVarEnv usage bndr `orElse` IAmDead
1198 %************************************************************************
1200 \subsection{Operations over OccInfo}
1202 %************************************************************************
1205 mkOneOcc :: OccEnv -> Id -> InterestingCxt -> UsageDetails
1206 mkOneOcc env id int_cxt
1207 | isLocalId id = unitVarEnv id (OneOcc False True int_cxt)
1208 | id `elemVarSet` occ_scrut_ids env = unitVarEnv id NoOccInfo
1209 | otherwise = emptyDetails
1211 markMany, markInsideLam, markInsideSCC :: OccInfo -> OccInfo
1213 markMany _ = NoOccInfo
1215 markInsideSCC occ = markMany occ
1217 markInsideLam (OneOcc _ one_br int_cxt) = OneOcc True one_br int_cxt
1218 markInsideLam occ = occ
1220 addOccInfo, orOccInfo :: OccInfo -> OccInfo -> OccInfo
1222 addOccInfo a1 a2 = ASSERT( not (isDeadOcc a1 || isDeadOcc a2) )
1223 NoOccInfo -- Both branches are at least One
1224 -- (Argument is never IAmDead)
1226 -- (orOccInfo orig new) is used
1227 -- when combining occurrence info from branches of a case
1229 orOccInfo (OneOcc in_lam1 _ int_cxt1)
1230 (OneOcc in_lam2 _ int_cxt2)
1231 = OneOcc (in_lam1 || in_lam2)
1232 False -- False, because it occurs in both branches
1233 (int_cxt1 && int_cxt2)
1234 orOccInfo a1 a2 = ASSERT( not (isDeadOcc a1 || isDeadOcc a2) )