X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=compiler%2FsimplCore%2FOccurAnal.lhs;h=ba302ff1410d5fe2aac4d11020cd9ffa5cca8439;hb=a5f2ab64f2f1306c803c0c20e21238973070f74b;hp=85da34adfc48dd777b8802dd54b6ab48a0abccb8;hpb=7fc749a43b4b6b85d234fa95d4928648259584f4;p=ghc-hetmet.git diff --git a/compiler/simplCore/OccurAnal.lhs b/compiler/simplCore/OccurAnal.lhs index 85da34a..ba302ff 100644 --- a/compiler/simplCore/OccurAnal.lhs +++ b/compiler/simplCore/OccurAnal.lhs @@ -27,11 +27,8 @@ module OccurAnal ( import CoreSyn import CoreFVs ( idRuleVars ) import CoreUtils ( exprIsTrivial, isDefaultAlt ) -import Id ( isDataConWorkId, isOneShotBndr, setOneShotLambda, - idOccInfo, setIdOccInfo, isLocalId, - isExportedId, idArity, idHasRules, - idUnique, Id - ) +import Id +import IdInfo import BasicTypes ( OccInfo(..), isOneOcc, InterestingCxt ) import VarSet @@ -41,7 +38,7 @@ import Maybes ( orElse ) import Digraph ( stronglyConnCompR, SCC(..) ) import PrelNames ( buildIdKey, foldrIdKey, runSTRepIdKey, augmentIdKey ) import Unique ( Unique ) -import UniqFM ( keysUFM, intersectsUFM ) +import UniqFM ( keysUFM, intersectsUFM, intersectUFM_C, foldUFM_Directly ) import Util ( mapAndUnzip ) import Outputable @@ -98,38 +95,36 @@ occAnalBind env (NonRec binder rhs) body_usage = (body_usage, []) | otherwise -- It's mentioned in the body - = (body_usage' +++ addRuleUsage rhs_usage binder, -- Note [RulesOnly] + = (body_usage' +++ addRuleUsage rhs_usage binder, -- Note [Rules are extra RHSs] [NonRec tagged_binder rhs']) where (body_usage', tagged_binder) = tagBinder body_usage binder (rhs_usage, rhs') = occAnalRhs env tagged_binder rhs \end{code} +Note [Dead code] +~~~~~~~~~~~~~~~~ Dropping dead code for recursive bindings is done in a very simple way: the entire set of bindings is dropped if none of its binders are mentioned in its body; otherwise none are. This seems to miss an obvious improvement. -@ + letrec f = ...g... g = ...f... in ...g... - ===> - letrec f = ...g... g = ...(...g...)... in ...g... -@ -Now @f@ is unused. But dependency analysis will sort this out into a -@letrec@ for @g@ and a @let@ for @f@, and then @f@ will get dropped. -It isn't easy to do a perfect job in one blow. Consider +Now 'f' is unused! But it's OK! Dependency analysis will sort this +out into a letrec for 'g' and a 'let' for 'f', and then 'f' will get +dropped. It isn't easy to do a perfect job in one blow. Consider -@ letrec f = ...g... g = ...h... h = ...k... @@ -137,29 +132,180 @@ It isn't easy to do a perfect job in one blow. Consider m = ...m... in ...m... -@ + + +Note [Loop breaking and RULES] +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +Loop breaking is surprisingly subtle. First read the section 4 of +"Secrets of the GHC inliner". This describes our basic plan. + +However things are made quite a bit more complicated by RULES. Remember + + * Note [Rules are extra RHSs] + ~~~~~~~~~~~~~~~~~~~~~~~~~~~ + A RULE for 'f' is like an extra RHS for 'f'. That way the "parent" + keeps the specialised "children" alive. If the parent dies + (because it isn't referenced any more), then the children will die + too (unless they are already referenced directly). + + To that end, we build a Rec group for each cyclic strongly + connected component, + *treating f's rules as extra RHSs for 'f'*. + + So in Example [eftInt], eftInt and eftIntFB will be put in the + same Rec, even though their 'main' RHSs are both non-recursive. + + * Note [Rules are visible in their own rec group] + ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + We want the rules for 'f' to be visible in f's right-hand side. + And we'd like them to be visible in other function in f's Rec + group. E.g. in Example [Specialisation rules] we want f' rule + to be visible in both f's RHS, and fs's RHS. + + This means that we must simplify the RULEs first, before looking + at any of the definitions. This is done by Simplify.simplRecBind, + when it calls addLetIdInfo. + + * Note [Choosing loop breakers] + ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + We avoid infinite inlinings by choosing loop breakers, and + ensuring that a loop breaker cuts each loop. But what is a + "loop"? In particular, a RULES is like an equation for 'f' that + is *always* inlined if it are applicable. We do *not* disable + rules for loop-breakers. It's up to whoever makes the rules to + make sure that the rules themselves alwasys terminate. See Note + [Rules for recursive functions] in Simplify.lhs + + Hence, if + f's RHS mentions g, and + g has a RULE that mentions h, and + h has a RULE that mentions f + + then we *must* choose f to be a loop breaker. In general, take the + free variables of f's RHS, and augment it with all the variables + reachable by RULES from those starting points. That is the whole + reason for computing rule_fv_env in occAnalBind. (Of course we + only consider free vars that are also binders in this Rec group.) + + Note that in Example [eftInt], *neither* eftInt *nor* eftIntFB is + chosen as a loop breaker, because their RHSs don't mention each other. + And indeed both can be inlined safely. + + Note that the edges of the graph we use for computing loop breakers + are not the same as the edges we use for computing the Rec blocks. + That's why we compute + rec_edges for the Rec block analysis + loop_breaker_edges for the loop breaker analysis + + + * Note [Weak loop breakers] + ~~~~~~~~~~~~~~~~~~~~~~~~~ + There is a last nasty wrinkle. Suppose we have + + Rec { f = f_rhs + RULE f [] = g + + h = h_rhs + g = h + ...more... + } + + Remmber that we simplify the RULES before any RHS (see Note + [Rules are visible in their own rec group] above). + + So we must *not* postInlineUnconditinoally 'g', even though + its RHS turns out to be trivial. (I'm assuming that 'g' is + not choosen as a loop breaker.) + + We "solve" this by making g a "weak" or "rules-only" loop breaker, + with OccInfo = IAmLoopBreaker True. A normal "strong" loop breaker + has IAmLoopBreaker False. So + + Inline postInlineUnconditinoally + IAmLoopBreaker False no no + IAmLoopBreaker True yes no + other yes yes + + The **sole** reason for this kind of loop breaker is so that + postInlineUnconditioanlly does not fire. Ugh. + + +Example [eftInt] +~~~~~~~~~~~~~~~ +Example (from GHC.Enum): + + eftInt :: Int# -> Int# -> [Int] + eftInt x y = ...(non-recursive)... + + {-# INLINE [0] eftIntFB #-} + eftIntFB :: (Int -> r -> r) -> r -> Int# -> Int# -> r + eftIntFB c n x y = ...(non-recursive)... + + {-# RULES + "eftInt" [~1] forall x y. eftInt x y = build (\ c n -> eftIntFB c n x y) + "eftIntList" [1] eftIntFB (:) [] = eftInt + #-} + +Example [Specialisation rules] +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +Consider this group, which is typical of what SpecConstr builds: + + fs a = ....f (C a).... + f x = ....f (C a).... + {-# RULE f (C a) = fs a #-} + +So 'f' and 'fs' are in the same Rec group (since f refers to fs via its RULE). + +But watch out! If 'fs' is not chosen as a loop breaker, we may get an infinite loop: + - the RULE is applied in f's RHS (see Note [Self-recursive rules] in Simplify + - fs is inlined (say it's small) + - now there's another opportunity to apply the RULE + +This showed up when compiling Control.Concurrent.Chan.getChanContents. \begin{code} occAnalBind env (Rec pairs) body_usage - = foldr ({-# SCC "occAnalBind.dofinal" #-} do_final_bind) (body_usage, []) sccs + | not (any (`usedIn` body_usage) bndrs) -- NB: look at body_usage, not total_usage + = (body_usage, []) -- Dead code + | otherwise + = (final_usage, map ({-# SCC "occAnalBind.dofinal" #-} do_final_bind) sccs) where - analysed_pairs :: [Details] - analysed_pairs = [ (bndr, rhs_usage, rhs') - | (bndr, rhs) <- pairs, - let (rhs_usage, rhs') = occAnalRhs env bndr rhs - ] + bndrs = map fst pairs + bndr_set = mkVarSet bndrs - sccs :: [SCC (Node Details)] - sccs = {-# SCC "occAnalBind.scc" #-} stronglyConnCompR edges + --------------------------------------- + -- See Note [Loop breaking] + --------------------------------------- + + -------------Dependency analysis ------------------------------ + occ_anald :: [(Id, (UsageDetails, CoreExpr))] + -- The UsageDetails here are strictly those arising from the RHS + -- *not* from any rules in the Id + occ_anald = [(bndr, occAnalRhs env bndr rhs) | (bndr,rhs) <- pairs] + total_usage = foldl add_usage body_usage occ_anald + add_usage body_usage (bndr, (rhs_usage, _)) + = body_usage +++ addRuleUsage rhs_usage bndr + + (final_usage, tagged_bndrs) = tagBinders total_usage bndrs + final_bndrs | no_rules = tagged_bndrs + | otherwise = map tag_rule_var tagged_bndrs + tag_rule_var bndr | bndr `elemVarSet` all_rule_fvs = makeLoopBreaker True bndr + | otherwise = bndr ---- stuff for dependency analysis of binds ------------------------------- - edges :: [Node Details] - edges = {-# SCC "occAnalBind.assoc" #-} - [ (details, idUnique id, edges_from id rhs_usage) - | details@(id, rhs_usage, rhs) <- analysed_pairs - ] + sccs :: [SCC (Node Details)] + sccs = {-# SCC "occAnalBind.scc" #-} stronglyConnCompR rec_edges + + rec_edges :: [Node Details] -- The binders are tagged with correct occ-info + rec_edges = {-# SCC "occAnalBind.assoc" #-} zipWith make_node final_bndrs occ_anald + make_node tagged_bndr (_bndr, (rhs_usage, rhs)) + = ((tagged_bndr, rhs, rhs_fvs), idUnique tagged_bndr, out_edges) + where + rhs_fvs = intersectUFM_C (\b _ -> b) bndr_set rhs_usage + out_edges = keysUFM (rhs_fvs `unionVarSet` idRuleVars tagged_bndr) + -- (a -> b) means a mentions b -- Given the usage details (a UFM that gives occ info for each free var of @@ -170,55 +316,53 @@ occAnalBind env (Rec pairs) body_usage -- maybeToBool (lookupVarEnv rhs_usage bndr)] -- which has n**2 cost, and this meant that edges_from alone -- consumed 10% of total runtime! - edges_from :: Id -> UsageDetails -> [Unique] - edges_from bndr rhs_usage = {-# SCC "occAnalBind.edges_from" #-} - keysUFM (addRuleUsage rhs_usage bndr) ---- Stuff to "re-constitute" bindings from dependency-analysis info ------ + do_final_bind (AcyclicSCC ((bndr, rhs, _), _, _)) = NonRec bndr rhs + do_final_bind (CyclicSCC cycle) + | no_rules = Rec (reOrderCycle cycle) + | otherwise = Rec (concatMap reOrderRec (stronglyConnCompR loop_breaker_edges)) + where -- See Note [Loop breaking for reason for looop_breker_edges] + loop_breaker_edges = map mk_node cycle + mk_node (details@(bndr, rhs, rhs_fvs), k, _) = (details, k, new_ks) + where + new_ks = keysUFM (extendFvs rule_fv_env rhs_fvs rhs_fvs) - -- Non-recursive SCC - do_final_bind (AcyclicSCC ((bndr, rhs_usage, rhs'), _, _)) (body_usage, binds_so_far) - | not (bndr `usedIn` body_usage) - = (body_usage, binds_so_far) -- Dead code - | otherwise - = (body_usage' +++ addRuleUsage rhs_usage bndr, new_bind : binds_so_far) - where - (body_usage', tagged_bndr) = tagBinder body_usage bndr - new_bind = NonRec tagged_bndr rhs' - - -- Recursive SCC - do_final_bind (CyclicSCC cycle) (body_usage, binds_so_far) - | not (any (`usedIn` body_usage) bndrs) -- NB: look at body_usage, not total_usage - = (body_usage, binds_so_far) -- Dead code - | otherwise -- If any is used, they all are - = (final_usage, final_bind : binds_so_far) - where - details = [details | (details, _, _) <- cycle] - bndrs = [bndr | (bndr, _, _) <- details] - bndr_usages = [addRuleUsage rhs_usage bndr | (bndr, rhs_usage, _) <- details] - total_usage = foldr (+++) body_usage bndr_usages - (final_usage, tagged_cycle) = mapAccumL tag_bind total_usage cycle - tag_bind usg ((bndr,rhs_usg,rhs),k,ks) = (usg', ((bndr',rhs_usg,rhs),k,ks)) - where - (usg', bndr') = tagBinder usg bndr - final_bind = Rec (reOrderCycle (mkVarSet bndrs) tagged_cycle) - -{- An alternative; rebuild the edges. No semantic difference, but perf might change - - -- Hopefully 'bndrs' is a relatively small group now - -- Now get ready for the loop-breaking phase - -- We've done dead-code elimination already, so no worries about un-referenced binders - keys = map idUnique bndrs - mk_node tagged_bndr (_, rhs_usage, rhs') - = ((tagged_bndr, rhs'), idUnique tagged_bndr, used) - where - used = [key | key <- keys, used_outside_rule rhs_usage key ] - - used_outside_rule usage uniq = case lookupUFM_Directly usage uniq of - Nothing -> False - Just RulesOnly -> False -- Ignore rules - other -> True --} + + ------------------------------------ + rule_fv_env :: IdEnv IdSet -- Variables from this group mentioned in RHS of rules + -- Domain is *subset* of bound vars (others have no rule fvs) + rule_fv_env = rule_loop init_rule_fvs + + no_rules = null init_rule_fvs + all_rule_fvs = foldr (unionVarSet . snd) emptyVarSet init_rule_fvs + init_rule_fvs = [(b, rule_fvs) + | b <- bndrs + , let rule_fvs = idRuleVars b `intersectVarSet` bndr_set + , not (isEmptyVarSet rule_fvs)] + + rule_loop :: [(Id,IdSet)] -> IdEnv IdSet -- Finds fixpoint + rule_loop fv_list + | no_change = env + | otherwise = rule_loop new_fv_list + where + env = mkVarEnv init_rule_fvs + (no_change, new_fv_list) = mapAccumL bump True fv_list + bump no_change (b,fvs) + | new_fvs `subVarSet` fvs = (no_change, (b,fvs)) + | otherwise = (False, (b,new_fvs `unionVarSet` fvs)) + where + new_fvs = extendFvs env emptyVarSet fvs + +extendFvs :: IdEnv IdSet -> IdSet -> IdSet -> IdSet +-- (extendFVs env fvs s) returns (fvs `union` env(s)) +extendFvs env fvs id_set + = foldUFM_Directly add fvs id_set + where + add uniq _ fvs + = case lookupVarEnv_Directly env uniq of + Just fvs' -> fvs' `unionVarSet` fvs + Nothing -> fvs \end{code} @reOrderRec@ is applied to the list of (binder,rhs) pairs for a cyclic @@ -258,33 +402,34 @@ Perhaps something cleverer would suffice. \begin{code} type Node details = (details, Unique, [Unique]) -- The Ints are gotten from the Unique, -- which is gotten from the Id. -type Details = (Id, UsageDetails, CoreExpr) +type Details = (Id, -- Binder + CoreExpr, -- RHS + IdSet) -- RHS free vars (*not* include rules) -reOrderRec :: IdSet -- Binders of this group - -> SCC (Node Details) +reOrderRec :: SCC (Node Details) -> [(Id,CoreExpr)] -- Sorted into a plausible order. Enough of the Ids have -- IAmALoopBreaker pragmas that there are no loops left. -reOrderRec bndrs (AcyclicSCC ((bndr, _, rhs), _, _)) = [(bndr, rhs)] -reOrderRec bndrs (CyclicSCC cycle) = reOrderCycle bndrs cycle +reOrderRec (AcyclicSCC ((bndr, rhs, _), _, _)) = [(bndr, rhs)] +reOrderRec (CyclicSCC cycle) = reOrderCycle cycle -reOrderCycle :: IdSet -> [Node Details] -> [(Id,CoreExpr)] -reOrderCycle bndrs [] +reOrderCycle :: [Node Details] -> [(Id,CoreExpr)] +reOrderCycle [] = panic "reOrderCycle" -reOrderCycle bndrs [bind] -- Common case of simple self-recursion - = [(makeLoopBreaker bndrs rhs_usg bndr, rhs)] +reOrderCycle [bind] -- Common case of simple self-recursion + = [(makeLoopBreaker False bndr, rhs)] where - ((bndr, rhs_usg, rhs), _, _) = bind + ((bndr, rhs, _), _, _) = bind -reOrderCycle bndrs (bind : binds) +reOrderCycle (bind : binds) = -- Choose a loop breaker, mark it no-inline, -- do SCC analysis on the rest, and recursively sort them out - concatMap (reOrderRec bndrs) (stronglyConnCompR unchosen) ++ - [(makeLoopBreaker bndrs rhs_usg bndr, rhs)] + concatMap reOrderRec (stronglyConnCompR unchosen) ++ + [(makeLoopBreaker False bndr, rhs)] where (chosen_bind, unchosen) = choose_loop_breaker bind (score bind) [] binds - (bndr, rhs_usg, rhs) = chosen_bind + (bndr, rhs, _) = chosen_bind -- This loop looks for the bind with the lowest score -- to pick as the loop breaker. The rest accumulate in @@ -301,7 +446,10 @@ reOrderCycle bndrs (bind : binds) sc = score bind score :: Node Details -> Int -- Higher score => less likely to be picked as loop breaker - score ((bndr, _, rhs), _, _) + score ((bndr, rhs, _), _, _) + | workerExists (idWorkerInfo bndr) = 10 + -- Note [Worker inline loop] + | exprIsTrivial rhs = 4 -- Practically certain to be inlined -- Used to have also: && not (isExportedId bndr) -- But I found this sometimes cost an extra iteration when we have @@ -309,24 +457,23 @@ reOrderCycle bndrs (bind : binds) -- where df is the exported dictionary. Then df makes a really -- bad choice for loop breaker - | idHasRules bndr = 3 - -- Avoid things with specialisations; we'd like - -- to take advantage of them in the subsequent bindings - -- Also vital to avoid risk of divergence: - -- Note [Recursive rules] + | is_con_app rhs = 2 -- Data types help with cases + -- Note [conapp] - | inlineCandidate bndr rhs = 2 -- Likely to be inlined + | inlineCandidate bndr rhs = 1 -- Likely to be inlined -- Note [Inline candidates] - | is_con_app rhs = 1 -- Data types help with cases - | otherwise = 0 inlineCandidate :: Id -> CoreExpr -> Bool inlineCandidate id (Note InlineMe _) = True inlineCandidate id rhs = isOneOcc (idOccInfo id) - -- Real example (the Enum Ordering instance from PrelBase): + -- Note [conapp] + -- + -- It's really really important to inline dictionaries. Real + -- example (the Enum Ordering instance from GHC.Base): + -- -- rec f = \ x -> case d of (p,q,r) -> p x -- g = \ x -> case d of (p,q,r) -> q x -- d = (v, f, g) @@ -335,6 +482,8 @@ reOrderCycle bndrs (bind : binds) -- On the other hand we *could* simplify those case expressions if -- we didn't stupidly choose d as the loop breaker. -- But we won't because constructor args are marked "Many". + -- Inlining dictionaries is really essential to unravelling + -- the loops in static numeric dictionaries, see GHC.Float. -- Cheap and cheerful; the simplifer moves casts out of the way -- The lambda case is important to spot x = /\a. C (f a) @@ -350,54 +499,30 @@ reOrderCycle bndrs (bind : binds) is_con_app (Note _ e) = is_con_app e is_con_app other = False -makeLoopBreaker :: VarSet -- Binders of this group - -> UsageDetails -- Usage of this rhs (neglecting rules) - -> Id -> Id --- Set the loop-breaker flag, recording whether the thing occurs only in --- the RHS of a RULE (in this recursive group) -makeLoopBreaker bndrs rhs_usg bndr - = setIdOccInfo bndr (IAmALoopBreaker rules_only) - where - rules_only = bndrs `intersectsUFM` rhs_usg +makeLoopBreaker :: Bool -> Id -> Id +-- Set the loop-breaker flag +-- See Note [Weak loop breakers] +makeLoopBreaker weak bndr = setIdOccInfo bndr (IAmALoopBreaker weak) \end{code} -Note [Inline candidates] +Note [Worker inline loop] ~~~~~~~~~~~~~~~~~~~~~~~~ -At one point I gave is_con_app a higher score than inline-candidate, -on the grounds that "it's *really* helpful if dictionaries get inlined fast". -However a nofib run revealed no change if they were swapped so that -inline-candidate has the higher score. And it's important that it does, -else you can get a bad worker-wrapper split thus: +Never choose a wrapper as the loop breaker! Because +wrappers get auto-generated inlinings when importing, and +that can lead to an infinite inlining loop. For example: rec { $wfoo x = ....foo x.... {-loop brk-} foo x = ...$wfoo x... } -But we *want* the wrapper to be inlined! If it isn't, the interface -file sees the unfolding for $wfoo, and sees that foo is strict (and -hence it gets an auto-generated wrapper. Result: an infinite inlining -in the importing scope. So be a bit careful if you change this. A -good example is Tree.repTree in nofib/spectral/minimax. If is_con_app -has the higher score, then compiling Game.hs goes into an infinite loop. - -Note [Recursive rules] -~~~~~~~~~~~~~~~~~~~~~~ -Consider this group, which is typical of what SpecConstr builds: - fs a = ....f (C a).... - f x = ....f (C a).... - {-# RULE f (C a) = fs a #-} - -So 'f' and 'fs' are mutually recursive. If we choose 'fs' as the loop breaker, -all is well; the RULE is applied, and 'fs' becomes self-recursive. - -But if we choose 'f' as the loop breaker, we may get an infinite loop: - - the RULE is applied in f's RHS (see Note [Self-recursive rules] in Simplify - - fs is inlined (say it's small) - - now there's another opportunity to apply the RULE - -So it's very important to choose the RULE-variable as the loop breaker. -This showed up when compiling Control.Concurrent.Chan.getChanContents. +The interface file sees the unfolding for $wfoo, and sees that foo is +strict (and hence it gets an auto-generated wrapper). Result: an +infinite inlining in the importing scope. So be a bit careful if you +change this. A good example is Tree.repTree in +nofib/spectral/minimax. If the repTree wrapper is chosen as the loop +breaker then compiling Game.hs goes into an infinite loop (this +happened when we gave is_con_app a lower score than inline candidates). Note [Closure conversion] ~~~~~~~~~~~~~~~~~~~~~~~~~ @@ -465,39 +590,9 @@ occAnalRhs env id rhs other -> False \end{code} -Note [RulesOnly] -~~~~~~~~~~~~~~~~~~ -If the binder has RULES inside it then we count the specialised Ids as -"extra rhs's". That way the "parent" keeps the specialised "children" -alive. If the parent dies (because it isn't referenced any more), -then the children will die too unless they are already referenced -directly. - -That's the basic idea. However in a recursive situation we want to be a bit -cleverer. Example (from GHC.Enum): - - eftInt :: Int# -> Int# -> [Int] - eftInt x y = ...(non-recursive)... - - {-# INLINE [0] eftIntFB #-} - eftIntFB :: (Int -> r -> r) -> r -> Int# -> Int# -> r - eftIntFB c n x y = ...(non-recursive)... - - {-# RULES - "eftInt" [~1] forall x y. eftInt x y = build (\ c n -> eftIntFB c n x y) - "eftIntList" [1] eftIntFB (:) [] = eftInt - #-} - -The two look mutually recursive only because of their RULES; we don't want -that to inhibit inlining! - -So when we identify a LoopBreaker, we mark it to say whether it only mentions -the other binders in its recursive group in a RULE. If so, we can inline it, -because doing so will not expose new occurrences of binders in its group. \begin{code} - addRuleUsage :: UsageDetails -> Id -> UsageDetails -- Add the usage from RULES in Id to the usage addRuleUsage usage id