2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
6 ***************************
8 ***************************
10 1. We attach binding levels to Core bindings, in preparation for floating
11 outwards (@FloatOut@).
13 2. We also let-ify many expressions (notably case scrutinees), so they
14 will have a fighting chance of being floated sensible.
16 3. We clone the binders of any floatable let-binding, so that when it is
17 floated out it will be unique. (This used to be done by the simplifier
18 but the latter now only ensures that there's no shadowing; indeed, even
19 that may not be true.)
21 NOTE: this can't be done using the uniqAway idea, because the variable
22 must be unique in the whole program, not just its current scope,
23 because two variables in different scopes may float out to the
26 NOTE: Very tiresomely, we must apply this substitution to
27 the rules stored inside a variable too.
29 We do *not* clone top-level bindings, because some of them must not change,
30 but we *do* clone bindings that are heading for the top level
33 case x of wild { p -> ...wild... }
34 we substitute x for wild in the RHS of the case alternatives:
35 case x of wild { p -> ...x... }
36 This means that a sub-expression involving x is not "trapped" inside the RHS.
37 And it's not inconvenient because we already have a substitution.
39 Note that this is EXACTLY BACKWARDS from the what the simplifier does.
40 The simplifier tries to get rid of occurrences of x, in favour of wild,
41 in the hope that there will only be one remaining occurrence of x, namely
42 the scrutinee of the case, and we can inline it.
49 LevelledBind, LevelledExpr,
51 incMinorLvl, ltMajLvl, ltLvl, isTopLvl
54 #include "HsVersions.h"
58 import DynFlags ( FloatOutSwitches(..) )
59 import CoreUtils ( exprType, exprIsTrivial, mkPiTypes )
60 import CoreArity ( exprBotStrictness_maybe )
61 import CoreFVs -- all of it
62 import CoreSubst ( Subst, emptySubst, extendInScope, extendInScopeList,
63 extendIdSubst, cloneIdBndr, cloneRecIdBndrs )
64 import Id ( idType, mkSysLocal, isOneShotLambda,
65 zapDemandIdInfo, transferPolyIdInfo,
66 idSpecialisation, idUnfolding, setIdInfo,
67 setIdStrictness, setIdArity
73 import Name ( getOccName )
74 import OccName ( occNameString )
75 import Type ( isUnLiftedType, Type )
76 import BasicTypes ( TopLevelFlag(..) )
78 import Util ( sortLe, isSingleton, count )
83 %************************************************************************
85 \subsection{Level numbers}
87 %************************************************************************
90 data Level = Level Int -- Level number of enclosing lambdas
91 Int -- Number of big-lambda and/or case expressions between
92 -- here and the nearest enclosing lambda
95 The {\em level number} on a (type-)lambda-bound variable is the
96 nesting depth of the (type-)lambda which binds it. The outermost lambda
97 has level 1, so (Level 0 0) means that the variable is bound outside any lambda.
99 On an expression, it's the maximum level number of its free
100 (type-)variables. On a let(rec)-bound variable, it's the level of its
101 RHS. On a case-bound variable, it's the number of enclosing lambdas.
103 Top-level variables: level~0. Those bound on the RHS of a top-level
104 definition but ``before'' a lambda; e.g., the \tr{x} in (levels shown
105 as ``subscripts'')...
107 a_0 = let b_? = ... in
108 x_1 = ... b ... in ...
111 The main function @lvlExpr@ carries a ``context level'' (@ctxt_lvl@).
112 That's meant to be the level number of the enclosing binder in the
113 final (floated) program. If the level number of a sub-expression is
114 less than that of the context, then it might be worth let-binding the
115 sub-expression so that it will indeed float.
117 If you can float to level @Level 0 0@ worth doing so because then your
118 allocation becomes static instead of dynamic. We always start with
122 Note [FloatOut inside INLINE]
123 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
124 @InlineCtxt@ very similar to @Level 0 0@, but is used for one purpose:
125 to say "don't float anything out of here". That's exactly what we
126 want for the body of an INLINE, where we don't want to float anything
127 out at all. See notes with lvlMFE below.
131 -- At one time I tried the effect of not float anything out of an InlineMe,
132 -- but it sometimes works badly. For example, consider PrelArr.done. It
133 -- has the form __inline (\d. e)
134 -- where e doesn't mention d. If we float this to
135 -- __inline (let x = e in \d. x)
136 -- things are bad. The inliner doesn't even inline it because it doesn't look
137 -- like a head-normal form. So it seems a lesser evil to let things float.
138 -- In SetLevels we do set the context to (Level 0 0) when we get to an InlineMe
139 -- which discourages floating out.
141 So the conclusion is: don't do any floating at all inside an InlineMe.
142 (In the above example, don't float the {x=e} out of the \d.)
144 One particular case is that of workers: we don't want to float the
145 call to the worker outside the wrapper, otherwise the worker might get
146 inlined into the floated expression, and an importing module won't see
150 type LevelledExpr = TaggedExpr Level
151 type LevelledBind = TaggedBind Level
154 tOP_LEVEL = Level 0 0
156 incMajorLvl :: Level -> Level
157 incMajorLvl (Level major _) = Level (major + 1) 0
159 incMinorLvl :: Level -> Level
160 incMinorLvl (Level major minor) = Level major (minor+1)
162 maxLvl :: Level -> Level -> Level
163 maxLvl l1@(Level maj1 min1) l2@(Level maj2 min2)
164 | (maj1 > maj2) || (maj1 == maj2 && min1 > min2) = l1
167 ltLvl :: Level -> Level -> Bool
168 ltLvl (Level maj1 min1) (Level maj2 min2)
169 = (maj1 < maj2) || (maj1 == maj2 && min1 < min2)
171 ltMajLvl :: Level -> Level -> Bool
172 -- Tells if one level belongs to a difft *lambda* level to another
173 ltMajLvl (Level maj1 _) (Level maj2 _) = maj1 < maj2
175 isTopLvl :: Level -> Bool
176 isTopLvl (Level 0 0) = True
179 instance Outputable Level where
180 ppr (Level maj min) = hcat [ char '<', int maj, char ',', int min, char '>' ]
182 instance Eq Level where
183 (Level maj1 min1) == (Level maj2 min2) = maj1 == maj2 && min1 == min2
187 %************************************************************************
189 \subsection{Main level-setting code}
191 %************************************************************************
194 setLevels :: FloatOutSwitches
199 setLevels float_lams binds us
200 = initLvl us (do_them init_env binds)
202 init_env = initialEnv float_lams
204 do_them :: LevelEnv -> [CoreBind] -> LvlM [LevelledBind]
205 do_them _ [] = return []
207 = do { (lvld_bind, env') <- lvlTopBind env b
208 ; lvld_binds <- do_them env' bs
209 ; return (lvld_bind : lvld_binds) }
211 lvlTopBind :: LevelEnv -> Bind Id -> LvlM (LevelledBind, LevelEnv)
212 lvlTopBind env (NonRec binder rhs)
213 = lvlBind TopLevel tOP_LEVEL env (AnnNonRec binder (freeVars rhs))
214 -- Rhs can have no free vars!
216 lvlTopBind env (Rec pairs)
217 = lvlBind TopLevel tOP_LEVEL env (AnnRec [(b,freeVars rhs) | (b,rhs) <- pairs])
220 %************************************************************************
222 \subsection{Setting expression levels}
224 %************************************************************************
227 lvlExpr :: Level -- ctxt_lvl: Level of enclosing expression
228 -> LevelEnv -- Level of in-scope names/tyvars
229 -> CoreExprWithFVs -- input expression
230 -> LvlM LevelledExpr -- Result expression
233 The @ctxt_lvl@ is, roughly, the level of the innermost enclosing
234 binder. Here's an example
236 v = \x -> ...\y -> let r = case (..x..) of
240 When looking at the rhs of @r@, @ctxt_lvl@ will be 1 because that's
241 the level of @r@, even though it's inside a level-2 @\y@. It's
242 important that @ctxt_lvl@ is 1 and not 2 in @r@'s rhs, because we
243 don't want @lvlExpr@ to turn the scrutinee of the @case@ into an MFE
244 --- because it isn't a *maximal* free expression.
246 If there were another lambda in @r@'s rhs, it would get level-2 as well.
249 lvlExpr _ _ ( _, AnnType ty) = return (Type ty)
250 lvlExpr _ env (_, AnnVar v) = return (lookupVar env v)
251 lvlExpr _ _ (_, AnnLit lit) = return (Lit lit)
253 lvlExpr ctxt_lvl env (_, AnnApp fun arg) = do
254 fun' <- lvlExpr ctxt_lvl env fun -- We don't do MFE on partial applications
255 arg' <- lvlMFE False ctxt_lvl env arg
256 return (App fun' arg')
258 lvlExpr ctxt_lvl env (_, AnnNote note expr) = do
259 expr' <- lvlExpr ctxt_lvl env expr
260 return (Note note expr')
262 lvlExpr ctxt_lvl env (_, AnnCast expr co) = do
263 expr' <- lvlExpr ctxt_lvl env expr
264 return (Cast expr' co)
266 -- We don't split adjacent lambdas. That is, given
268 -- we don't float to give
269 -- \x -> let v = x+y in \y -> (v,y)
270 -- Why not? Because partial applications are fairly rare, and splitting
271 -- lambdas makes them more expensive.
273 lvlExpr ctxt_lvl env expr@(_, AnnLam {}) = do
274 new_body <- lvlMFE True new_lvl new_env body
275 return (mkLams new_bndrs new_body)
277 (bndrs, body) = collectAnnBndrs expr
278 (new_lvl, new_bndrs) = lvlLamBndrs ctxt_lvl bndrs
279 new_env = extendLvlEnv env new_bndrs
280 -- At one time we called a special verion of collectBinders,
281 -- which ignored coercions, because we don't want to split
282 -- a lambda like this (\x -> coerce t (\s -> ...))
283 -- This used to happen quite a bit in state-transformer programs,
284 -- but not nearly so much now non-recursive newtypes are transparent.
285 -- [See SetLevels rev 1.50 for a version with this approach.]
287 lvlExpr ctxt_lvl env (_, AnnLet (AnnNonRec bndr rhs) body)
288 | isUnLiftedType (idType bndr) = do
289 -- Treat unlifted let-bindings (let x = b in e) just like (case b of x -> e)
290 -- That is, leave it exactly where it is
291 -- We used to float unlifted bindings too (e.g. to get a cheap primop
292 -- outside a lambda (to see how, look at lvlBind in rev 1.58)
293 -- but an unrelated change meant that these unlifed bindings
294 -- could get to the top level which is bad. And there's not much point;
295 -- unlifted bindings are always cheap, and so hardly worth floating.
296 rhs' <- lvlExpr ctxt_lvl env rhs
297 body' <- lvlExpr incd_lvl env' body
298 return (Let (NonRec bndr' rhs') body')
300 incd_lvl = incMinorLvl ctxt_lvl
301 bndr' = TB bndr incd_lvl
302 env' = extendLvlEnv env [bndr']
304 lvlExpr ctxt_lvl env (_, AnnLet bind body) = do
305 (bind', new_env) <- lvlBind NotTopLevel ctxt_lvl env bind
306 body' <- lvlExpr ctxt_lvl new_env body
307 return (Let bind' body')
309 lvlExpr ctxt_lvl env (_, AnnCase expr case_bndr ty alts) = do
310 expr' <- lvlMFE True ctxt_lvl env expr
311 let alts_env = extendCaseBndrLvlEnv env expr' case_bndr incd_lvl
312 alts' <- mapM (lvl_alt alts_env) alts
313 return (Case expr' (TB case_bndr incd_lvl) ty alts')
315 incd_lvl = incMinorLvl ctxt_lvl
317 lvl_alt alts_env (con, bs, rhs) = do
318 rhs' <- lvlMFE True incd_lvl new_env rhs
319 return (con, bs', rhs')
321 bs' = [ TB b incd_lvl | b <- bs ]
322 new_env = extendLvlEnv alts_env bs'
325 @lvlMFE@ is just like @lvlExpr@, except that it might let-bind
326 the expression, so that it can itself be floated.
330 We don't float unlifted MFEs, which potentially loses big opportunites.
333 where h :: Int -> Int# is expensive. We'd like to float the (h y) outside
334 the \x, but we don't because it's unboxed. Possible solution: box it.
336 Note [Bottoming floats]
337 ~~~~~~~~~~~~~~~~~~~~~~~
339 f = \x. g (error "urk")
340 we'd like to float the call to error, to get
343 But, it's very helpful for lvl to get a strictness signature, so that,
344 for example, its unfolding is not exposed in interface files (unnecessary).
345 But this float-out might occur after strictness analysis. So we use the
346 cheap-and-cheerful exprBotStrictness_maybe function.
350 We don't float a case expression as an MFE from a strict context. Why not?
351 Because in doing so we share a tiny bit of computation (the switch) but
352 in exchange we build a thunk, which is bad. This case reduces allocation
353 by 7% in spectral/puzzle (a rather strange benchmark) and 1.2% in real/fem.
354 Doesn't change any other allocation at all.
357 lvlMFE :: Bool -- True <=> strict context [body of case or let]
358 -> Level -- Level of innermost enclosing lambda/tylam
359 -> LevelEnv -- Level of in-scope names/tyvars
360 -> CoreExprWithFVs -- input expression
361 -> LvlM LevelledExpr -- Result expression
363 lvlMFE _ _ _ (_, AnnType ty)
366 -- No point in floating out an expression wrapped in a coercion or note
367 -- If we do we'll transform lvl = e |> co
368 -- to lvl' = e; lvl = lvl' |> co
369 -- and then inline lvl. Better just to float out the payload.
370 lvlMFE strict_ctxt ctxt_lvl env (_, AnnNote n e)
371 = do { e' <- lvlMFE strict_ctxt ctxt_lvl env e
372 ; return (Note n e') }
374 lvlMFE strict_ctxt ctxt_lvl env (_, AnnCast e co)
375 = do { e' <- lvlMFE strict_ctxt ctxt_lvl env e
376 ; return (Cast e' co) }
379 lvlMFE True ctxt_lvl env e@(_, AnnCase {})
380 = lvlExpr ctxt_lvl env e -- Don't share cases
382 lvlMFE strict_ctxt ctxt_lvl env ann_expr@(fvs, _)
383 | isUnLiftedType ty -- Can't let-bind it; see Note [Unlifted MFEs]
384 || exprIsTrivial expr -- Never float if it's trivial
385 || not good_destination
386 = -- Don't float it out
387 lvlExpr ctxt_lvl env ann_expr
389 | otherwise -- Float it out!
390 = do expr' <- lvlFloatRhs abs_vars dest_lvl env ann_expr
391 var <- newLvlVar "lvl" abs_vars ty
392 -- Note [Bottoming floats]
393 let var_w_str = case exprBotStrictness_maybe expr of
394 Just (arity,str) -> var `setIdArity` arity
395 `setIdStrictness` str
397 return (Let (NonRec (TB var_w_str dest_lvl) expr')
398 (mkVarApps (Var var_w_str) abs_vars))
400 expr = deAnnotate ann_expr
402 dest_lvl = destLevel env fvs (isFunction ann_expr)
403 abs_vars = abstractVars dest_lvl env fvs
405 -- A decision to float entails let-binding this thing, and we only do
406 -- that if we'll escape a value lambda, or will go to the top level.
408 | dest_lvl `ltMajLvl` ctxt_lvl -- Escapes a value lambda
410 -- OLD CODE: not (exprIsCheap expr) || isTopLvl dest_lvl
411 -- see Note [Escaping a value lambda]
413 | otherwise -- Does not escape a value lambda
414 = isTopLvl dest_lvl -- Only float if we are going to the top level
415 && floatConsts env -- and the floatConsts flag is on
416 && not strict_ctxt -- Don't float from a strict context
417 -- We are keen to float something to the top level, even if it does not
418 -- escape a lambda, because then it needs no allocation. But it's controlled
419 -- by a flag, because doing this too early loses opportunities for RULES
420 -- which (needless to say) are important in some nofib programs
421 -- (gcd is an example).
424 -- concat = /\ a -> foldr ..a.. (++) []
425 -- was getting turned into
426 -- concat = /\ a -> lvl a
427 -- lvl = /\ a -> foldr ..a.. (++) []
428 -- which is pretty stupid. Hence the strict_ctxt test
431 Note [Escaping a value lambda]
432 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
433 We want to float even cheap expressions out of value lambdas,
434 because that saves allocation. Consider
435 f = \x. .. (\y.e) ...
436 Then we'd like to avoid allocating the (\y.e) every time we call f,
437 (assuming e does not mention x).
439 An example where this really makes a difference is simplrun009.
441 Another reason it's good is because it makes SpecContr fire on functions.
443 f = \x. ....(f (\y.e))....
444 After floating we get
446 f = \x. ....(f lvl)...
447 and that is much easier for SpecConstr to generate a robust specialisation for.
449 The OLD CODE (given where this Note is referred to) prevents floating
450 of the example above, so I just don't understand the old code. I
451 don't understand the old comment either (which appears below). I
452 measured the effect on nofib of changing OLD CODE to 'True', and got
453 zeros everywhere, but a 4% win for 'puzzle'. Very small 0.5% loss for
454 'cse'; turns out to be because our arity analysis isn't good enough
455 yet (mentioned in Simon-nofib-notes).
458 Even if it escapes a value lambda, we only
459 float if it's not cheap (unless it'll get all the
460 way to the top). I've seen cases where we
461 float dozens of tiny free expressions, which cost
462 more to allocate than to evaluate.
463 NB: exprIsCheap is also true of bottom expressions, which
464 is good; we don't want to share them
466 It's only Really Bad to float a cheap expression out of a
467 strict context, because that builds a thunk that otherwise
468 would never be built. So another alternative would be to
470 || (strict_ctxt && not (exprIsBottom expr))
471 to the condition above. We should really try this out.
474 %************************************************************************
476 \subsection{Bindings}
478 %************************************************************************
480 The binding stuff works for top level too.
483 lvlBind :: TopLevelFlag -- Used solely to decide whether to clone
484 -> Level -- Context level; might be Top even for bindings nested in the RHS
485 -- of a top level binding
488 -> LvlM (LevelledBind, LevelEnv)
490 lvlBind top_lvl ctxt_lvl env (AnnNonRec bndr rhs@(rhs_fvs,_))
491 | isTyVar bndr -- Don't do anything for TyVar binders
492 -- (simplifier gets rid of them pronto)
493 = do rhs' <- lvlExpr ctxt_lvl env rhs
494 return (NonRec (TB bndr ctxt_lvl) rhs', env)
497 = do -- No type abstraction; clone existing binder
498 rhs' <- lvlExpr dest_lvl env rhs
499 (env', bndr') <- cloneVar top_lvl env bndr ctxt_lvl dest_lvl
500 return (NonRec (TB bndr' dest_lvl) rhs', env')
503 = do -- Yes, type abstraction; create a new binder, extend substitution, etc
504 rhs' <- lvlFloatRhs abs_vars dest_lvl env rhs
505 (env', [bndr']) <- newPolyBndrs dest_lvl env abs_vars [bndr]
506 return (NonRec (TB bndr' dest_lvl) rhs', env')
509 bind_fvs = rhs_fvs `unionVarSet` idFreeVars bndr
510 abs_vars = abstractVars dest_lvl env bind_fvs
511 dest_lvl = destLevel env bind_fvs (isFunction rhs)
516 lvlBind top_lvl ctxt_lvl env (AnnRec pairs)
518 = do (new_env, new_bndrs) <- cloneRecVars top_lvl env bndrs ctxt_lvl dest_lvl
519 new_rhss <- mapM (lvlExpr ctxt_lvl new_env) rhss
520 return (Rec ([TB b dest_lvl | b <- new_bndrs] `zip` new_rhss), new_env)
522 | isSingleton pairs && count isId abs_vars > 1
523 = do -- Special case for self recursion where there are
524 -- several variables carried around: build a local loop:
525 -- poly_f = \abs_vars. \lam_vars . letrec f = \lam_vars. rhs in f lam_vars
526 -- This just makes the closures a bit smaller. If we don't do
527 -- this, allocation rises significantly on some programs
529 -- We could elaborate it for the case where there are several
530 -- mutually functions, but it's quite a bit more complicated
532 -- This all seems a bit ad hoc -- sigh
534 (bndr,rhs) = head pairs
535 (rhs_lvl, abs_vars_w_lvls) = lvlLamBndrs dest_lvl abs_vars
536 rhs_env = extendLvlEnv env abs_vars_w_lvls
537 (rhs_env', new_bndr) <- cloneVar NotTopLevel rhs_env bndr rhs_lvl rhs_lvl
539 (lam_bndrs, rhs_body) = collectAnnBndrs rhs
540 (body_lvl, new_lam_bndrs) = lvlLamBndrs rhs_lvl lam_bndrs
541 body_env = extendLvlEnv rhs_env' new_lam_bndrs
542 new_rhs_body <- lvlExpr body_lvl body_env rhs_body
543 (poly_env, [poly_bndr]) <- newPolyBndrs dest_lvl env abs_vars [bndr]
544 return (Rec [(TB poly_bndr dest_lvl,
545 mkLams abs_vars_w_lvls $
546 mkLams new_lam_bndrs $
547 Let (Rec [(TB new_bndr rhs_lvl, mkLams new_lam_bndrs new_rhs_body)])
548 (mkVarApps (Var new_bndr) lam_bndrs))],
551 | otherwise = do -- Non-null abs_vars
552 (new_env, new_bndrs) <- newPolyBndrs dest_lvl env abs_vars bndrs
553 new_rhss <- mapM (lvlFloatRhs abs_vars dest_lvl new_env) rhss
554 return (Rec ([TB b dest_lvl | b <- new_bndrs] `zip` new_rhss), new_env)
557 (bndrs,rhss) = unzip pairs
559 -- Finding the free vars of the binding group is annoying
560 bind_fvs = (unionVarSets [ idFreeVars bndr `unionVarSet` rhs_fvs
561 | (bndr, (rhs_fvs,_)) <- pairs])
565 dest_lvl = destLevel env bind_fvs (all isFunction rhss)
566 abs_vars = abstractVars dest_lvl env bind_fvs
568 ----------------------------------------------------
569 -- Three help functons for the type-abstraction case
571 lvlFloatRhs :: [CoreBndr] -> Level -> LevelEnv -> CoreExprWithFVs
572 -> UniqSM (Expr (TaggedBndr Level))
573 lvlFloatRhs abs_vars dest_lvl env rhs = do
574 rhs' <- lvlExpr rhs_lvl rhs_env rhs
575 return (mkLams abs_vars_w_lvls rhs')
577 (rhs_lvl, abs_vars_w_lvls) = lvlLamBndrs dest_lvl abs_vars
578 rhs_env = extendLvlEnv env abs_vars_w_lvls
582 %************************************************************************
584 \subsection{Deciding floatability}
586 %************************************************************************
589 lvlLamBndrs :: Level -> [CoreBndr] -> (Level, [TaggedBndr Level])
590 -- Compute the levels for the binders of a lambda group
591 -- The binders returned are exactly the same as the ones passed,
592 -- but they are now paired with a level
596 lvlLamBndrs lvl bndrs
597 = go (incMinorLvl lvl)
598 False -- Havn't bumped major level in this group
601 go old_lvl bumped_major rev_lvld_bndrs (bndr:bndrs)
602 | isId bndr && -- Go to the next major level if this is a value binder,
603 not bumped_major && -- and we havn't already gone to the next level (one jump per group)
604 not (isOneShotLambda bndr) -- and it isn't a one-shot lambda
605 = go new_lvl True (TB bndr new_lvl : rev_lvld_bndrs) bndrs
608 = go old_lvl bumped_major (TB bndr old_lvl : rev_lvld_bndrs) bndrs
611 new_lvl = incMajorLvl old_lvl
613 go old_lvl _ rev_lvld_bndrs []
614 = (old_lvl, reverse rev_lvld_bndrs)
615 -- a lambda like this (\x -> coerce t (\s -> ...))
616 -- This happens quite a bit in state-transformer programs
620 -- Destintion level is the max Id level of the expression
621 -- (We'll abstract the type variables, if any.)
622 destLevel :: LevelEnv -> VarSet -> Bool -> Level
623 destLevel env fvs is_function
625 && is_function = tOP_LEVEL -- Send functions to top level; see
626 -- the comments with isFunction
627 | otherwise = maxIdLevel env fvs
629 isFunction :: CoreExprWithFVs -> Bool
630 -- The idea here is that we want to float *functions* to
631 -- the top level. This saves no work, but
632 -- (a) it can make the host function body a lot smaller,
633 -- and hence inlinable.
634 -- (b) it can also save allocation when the function is recursive:
635 -- h = \x -> letrec f = \y -> ...f...y...x...
638 -- f = \x y -> ...(f x)...y...x...
640 -- No allocation for f now.
641 -- We may only want to do this if there are sufficiently few free
642 -- variables. We certainly only want to do it for values, and not for
643 -- constructors. So the simple thing is just to look for lambdas
644 isFunction (_, AnnLam b e) | isId b = True
645 | otherwise = isFunction e
646 isFunction (_, AnnNote _ e) = isFunction e
651 %************************************************************************
653 \subsection{Free-To-Level Monad}
655 %************************************************************************
658 type LevelEnv = (FloatOutSwitches,
659 VarEnv Level, -- Domain is *post-cloned* TyVars and Ids
660 Subst, -- Domain is pre-cloned Ids; tracks the in-scope set
661 -- so that subtitution is capture-avoiding
662 IdEnv ([Var], LevelledExpr)) -- Domain is pre-cloned Ids
663 -- We clone let-bound variables so that they are still
664 -- distinct when floated out; hence the SubstEnv/IdEnv.
665 -- (see point 3 of the module overview comment).
666 -- We also use these envs when making a variable polymorphic
667 -- because we want to float it out past a big lambda.
669 -- The Subst and IdEnv always implement the same mapping, but the
670 -- Subst maps to CoreExpr and the IdEnv to LevelledExpr
671 -- Since the range is always a variable or type application,
672 -- there is never any difference between the two, but sadly
673 -- the types differ. The SubstEnv is used when substituting in
674 -- a variable's IdInfo; the IdEnv when we find a Var.
676 -- In addition the IdEnv records a list of tyvars free in the
677 -- type application, just so we don't have to call freeVars on
678 -- the type application repeatedly.
680 -- The domain of the both envs is *pre-cloned* Ids, though
682 -- The domain of the VarEnv Level is the *post-cloned* Ids
684 initialEnv :: FloatOutSwitches -> LevelEnv
685 initialEnv float_lams = (float_lams, emptyVarEnv, emptySubst, emptyVarEnv)
687 floatLams :: LevelEnv -> Bool
688 floatLams (fos, _, _, _) = floatOutLambdas fos
690 floatConsts :: LevelEnv -> Bool
691 floatConsts (fos, _, _, _) = floatOutConstants fos
693 extendLvlEnv :: LevelEnv -> [TaggedBndr Level] -> LevelEnv
694 -- Used when *not* cloning
695 extendLvlEnv (float_lams, lvl_env, subst, id_env) prs
697 foldl add_lvl lvl_env prs,
698 foldl del_subst subst prs,
699 foldl del_id id_env prs)
701 add_lvl env (TB v l) = extendVarEnv env v l
702 del_subst env (TB v _) = extendInScope env v
703 del_id env (TB v _) = delVarEnv env v
704 -- We must remove any clone for this variable name in case of
705 -- shadowing. This bit me in the following case
706 -- (in nofib/real/gg/Spark.hs):
709 -- ... -> case e of wild {
710 -- ... -> ... wild ...
714 -- The inside occurrence of @wild@ was being replaced with @ds@,
715 -- incorrectly, because the SubstEnv was still lying around. Ouch!
718 extendInScopeEnv :: LevelEnv -> Var -> LevelEnv
719 extendInScopeEnv (fl, le, subst, ids) v = (fl, le, extendInScope subst v, ids)
721 extendInScopeEnvList :: LevelEnv -> [Var] -> LevelEnv
722 extendInScopeEnvList (fl, le, subst, ids) vs = (fl, le, extendInScopeList subst vs, ids)
724 -- extendCaseBndrLvlEnv adds the mapping case-bndr->scrut-var if it can
725 -- (see point 4 of the module overview comment)
726 extendCaseBndrLvlEnv :: LevelEnv -> Expr (TaggedBndr Level) -> Var -> Level
728 extendCaseBndrLvlEnv (float_lams, lvl_env, subst, id_env) (Var scrut_var) case_bndr lvl
730 extendVarEnv lvl_env case_bndr lvl,
731 extendIdSubst subst case_bndr (Var scrut_var),
732 extendVarEnv id_env case_bndr ([scrut_var], Var scrut_var))
734 extendCaseBndrLvlEnv env _scrut case_bndr lvl
735 = extendLvlEnv env [TB case_bndr lvl]
737 extendPolyLvlEnv :: Level -> LevelEnv -> [Var] -> [(Var, Var)] -> LevelEnv
738 extendPolyLvlEnv dest_lvl (float_lams, lvl_env, subst, id_env) abs_vars bndr_pairs
740 foldl add_lvl lvl_env bndr_pairs,
741 foldl add_subst subst bndr_pairs,
742 foldl add_id id_env bndr_pairs)
744 add_lvl env (_, v') = extendVarEnv env v' dest_lvl
745 add_subst env (v, v') = extendIdSubst env v (mkVarApps (Var v') abs_vars)
746 add_id env (v, v') = extendVarEnv env v ((v':abs_vars), mkVarApps (Var v') abs_vars)
748 extendCloneLvlEnv :: Level -> LevelEnv -> Subst -> [(Var, Var)] -> LevelEnv
749 extendCloneLvlEnv lvl (float_lams, lvl_env, _, id_env) new_subst bndr_pairs
751 foldl add_lvl lvl_env bndr_pairs,
753 foldl add_id id_env bndr_pairs)
755 add_lvl env (_, v') = extendVarEnv env v' lvl
756 add_id env (v, v') = extendVarEnv env v ([v'], Var v')
759 maxIdLevel :: LevelEnv -> VarSet -> Level
760 maxIdLevel (_, lvl_env,_,id_env) var_set
761 = foldVarSet max_in tOP_LEVEL var_set
763 max_in in_var lvl = foldr max_out lvl (case lookupVarEnv id_env in_var of
764 Just (abs_vars, _) -> abs_vars
768 | isId out_var = case lookupVarEnv lvl_env out_var of
769 Just lvl' -> maxLvl lvl' lvl
771 | otherwise = lvl -- Ignore tyvars in *maxIdLevel*
773 lookupVar :: LevelEnv -> Id -> LevelledExpr
774 lookupVar (_, _, _, id_env) v = case lookupVarEnv id_env v of
775 Just (_, expr) -> expr
778 abstractVars :: Level -> LevelEnv -> VarSet -> [Var]
779 -- Find the variables in fvs, free vars of the target expresion,
780 -- whose level is greater than the destination level
781 -- These are the ones we are going to abstract out
782 abstractVars dest_lvl (_, lvl_env, _, id_env) fvs
783 = map zap $ uniq $ sortLe le
784 [var | fv <- varSetElems fvs
785 , var <- absVarsOf id_env fv
787 -- NB: it's important to call abstract_me only on the OutIds the
788 -- come from absVarsOf (not on fv, which is an InId)
790 -- Sort the variables so the true type variables come first;
791 -- the tyvars scope over Ids and coercion vars
792 v1 `le` v2 = case (is_tv v1, is_tv v2) of
793 (True, False) -> True
794 (False, True) -> False
795 _ -> v1 <= v2 -- Same family
797 is_tv v = isTyVar v && not (isCoVar v)
799 uniq :: [Var] -> [Var]
800 -- Remove adjacent duplicates; the sort will have brought them together
801 uniq (v1:v2:vs) | v1 == v2 = uniq (v2:vs)
802 | otherwise = v1 : uniq (v2:vs)
805 abstract_me v = case lookupVarEnv lvl_env v of
806 Just lvl -> dest_lvl `ltLvl` lvl
809 -- We are going to lambda-abstract, so nuke any IdInfo,
810 -- and add the tyvars of the Id (if necessary)
811 zap v | isId v = WARN( isInlineRule (idUnfolding v) ||
812 not (isEmptySpecInfo (idSpecialisation v)),
813 text "absVarsOf: discarding info on" <+> ppr v )
814 setIdInfo v vanillaIdInfo
817 absVarsOf :: IdEnv ([Var], LevelledExpr) -> Var -> [Var]
818 -- If f is free in the expression, and f maps to poly_f a b c in the
819 -- current substitution, then we must report a b c as candidate type
822 -- Also, if x::a is an abstracted variable, then so is a; that is,
823 -- we must look in x's type
824 -- And similarly if x is a coercion variable.
826 | isId v = [av2 | av1 <- lookup_avs v
827 , av2 <- add_tyvars av1]
828 | isCoVar v = add_tyvars v
832 lookup_avs v = case lookupVarEnv id_env v of
833 Just (abs_vars, _) -> abs_vars
836 add_tyvars v = v : varSetElems (varTypeTyVars v)
840 type LvlM result = UniqSM result
842 initLvl :: UniqSupply -> UniqSM a -> a
848 newPolyBndrs :: Level -> LevelEnv -> [Var] -> [Id] -> UniqSM (LevelEnv, [Id])
849 newPolyBndrs dest_lvl env abs_vars bndrs = do
851 let new_bndrs = zipWith mk_poly_bndr bndrs uniqs
852 return (extendPolyLvlEnv dest_lvl env abs_vars (bndrs `zip` new_bndrs), new_bndrs)
854 mk_poly_bndr bndr uniq = transferPolyIdInfo bndr abs_vars $ -- Note [transferPolyIdInfo] in Id.lhs
855 mkSysLocal (mkFastString str) uniq poly_ty
857 str = "poly_" ++ occNameString (getOccName bndr)
858 poly_ty = mkPiTypes abs_vars (idType bndr)
861 -> [CoreBndr] -> Type -- Abstract wrt these bndrs
863 newLvlVar str vars body_ty = do
865 return (mkSysLocal (mkFastString str) uniq (mkPiTypes vars body_ty))
867 -- The deeply tiresome thing is that we have to apply the substitution
868 -- to the rules inside each Id. Grr. But it matters.
870 cloneVar :: TopLevelFlag -> LevelEnv -> Id -> Level -> Level -> LvlM (LevelEnv, Id)
871 cloneVar TopLevel env v _ _
872 = return (extendInScopeEnv env v, v) -- Don't clone top level things
873 -- But do extend the in-scope env, to satisfy the in-scope invariant
875 cloneVar NotTopLevel env@(_,_,subst,_) v ctxt_lvl dest_lvl
876 = ASSERT( isId v ) do
877 us <- getUniqueSupplyM
879 (subst', v1) = cloneIdBndr subst us v
880 v2 = zap_demand ctxt_lvl dest_lvl v1
881 env' = extendCloneLvlEnv dest_lvl env subst' [(v,v2)]
884 cloneRecVars :: TopLevelFlag -> LevelEnv -> [Id] -> Level -> Level -> LvlM (LevelEnv, [Id])
885 cloneRecVars TopLevel env vs _ _
886 = return (extendInScopeEnvList env vs, vs) -- Don't clone top level things
887 cloneRecVars NotTopLevel env@(_,_,subst,_) vs ctxt_lvl dest_lvl
888 = ASSERT( all isId vs ) do
889 us <- getUniqueSupplyM
891 (subst', vs1) = cloneRecIdBndrs subst us vs
892 vs2 = map (zap_demand ctxt_lvl dest_lvl) vs1
893 env' = extendCloneLvlEnv dest_lvl env subst' (vs `zip` vs2)
896 -- VERY IMPORTANT: we must zap the demand info
897 -- if the thing is going to float out past a lambda,
898 -- or if it's going to top level (where things can't be strict)
899 zap_demand :: Level -> Level -> Id -> Id
900 zap_demand dest_lvl ctxt_lvl id
901 | ctxt_lvl == dest_lvl,
902 not (isTopLvl dest_lvl) = id -- Stays, and not going to top level
903 | otherwise = zapDemandIdInfo id -- Floats out