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, isInlineCtxt
54 #include "HsVersions.h"
58 import CmdLineOpts ( FloatOutSwitches(..) )
59 import CoreUtils ( exprType, exprIsTrivial, exprIsCheap, mkPiTypes )
60 import CoreFVs -- all of it
62 import Id ( Id, idType, mkSysLocalUnencoded,
63 isOneShotLambda, zapDemandIdInfo,
64 idSpecialisation, idWorkerInfo, setIdInfo
66 import IdInfo ( workerExists, vanillaIdInfo, )
70 import Name ( getOccName )
71 import OccName ( occNameUserString )
72 import Type ( isUnLiftedType, Type )
73 import BasicTypes ( TopLevelFlag(..) )
75 import Util ( sortLe, isSingleton, count )
80 %************************************************************************
82 \subsection{Level numbers}
84 %************************************************************************
87 data Level = InlineCtxt -- A level that's used only for
88 -- the context parameter ctxt_lvl
89 | Level Int -- Level number of enclosing lambdas
90 Int -- Number of big-lambda and/or case expressions between
91 -- here and the nearest enclosing lambda
94 The {\em level number} on a (type-)lambda-bound variable is the
95 nesting depth of the (type-)lambda which binds it. The outermost lambda
96 has level 1, so (Level 0 0) means that the variable is bound outside any lambda.
98 On an expression, it's the maximum level number of its free
99 (type-)variables. On a let(rec)-bound variable, it's the level of its
100 RHS. On a case-bound variable, it's the number of enclosing lambdas.
102 Top-level variables: level~0. Those bound on the RHS of a top-level
103 definition but ``before'' a lambda; e.g., the \tr{x} in (levels shown
104 as ``subscripts'')...
106 a_0 = let b_? = ... in
107 x_1 = ... b ... in ...
110 The main function @lvlExpr@ carries a ``context level'' (@ctxt_lvl@).
111 That's meant to be the level number of the enclosing binder in the
112 final (floated) program. If the level number of a sub-expression is
113 less than that of the context, then it might be worth let-binding the
114 sub-expression so that it will indeed float.
116 If you can float to level @Level 0 0@ worth doing so because then your
117 allocation becomes static instead of dynamic. We always start with
123 @InlineCtxt@ very similar to @Level 0 0@, but is used for one purpose:
124 to say "don't float anything out of here". That's exactly what we
125 want for the body of an INLINE, where we don't want to float anything
126 out at all. See notes with lvlMFE below.
130 -- At one time I tried the effect of not float anything out of an InlineMe,
131 -- but it sometimes works badly. For example, consider PrelArr.done. It
132 -- has the form __inline (\d. e)
133 -- where e doesn't mention d. If we float this to
134 -- __inline (let x = e in \d. x)
135 -- things are bad. The inliner doesn't even inline it because it doesn't look
136 -- like a head-normal form. So it seems a lesser evil to let things float.
137 -- In SetLevels we do set the context to (Level 0 0) when we get to an InlineMe
138 -- which discourages floating out.
140 So the conclusion is: don't do any floating at all inside an InlineMe.
141 (In the above example, don't float the {x=e} out of the \d.)
143 One particular case is that of workers: we don't want to float the
144 call to the worker outside the wrapper, otherwise the worker might get
145 inlined into the floated expression, and an importing module won't see
149 type LevelledExpr = TaggedExpr Level
150 type LevelledBind = TaggedBind Level
152 tOP_LEVEL = Level 0 0
153 iNLINE_CTXT = InlineCtxt
155 incMajorLvl :: Level -> Level
156 -- For InlineCtxt we ignore any inc's; we don't want
157 -- to do any floating at all; see notes above
158 incMajorLvl InlineCtxt = InlineCtxt
159 incMajorLvl (Level major minor) = Level (major+1) 0
161 incMinorLvl :: Level -> Level
162 incMinorLvl InlineCtxt = InlineCtxt
163 incMinorLvl (Level major minor) = Level major (minor+1)
165 maxLvl :: Level -> Level -> Level
166 maxLvl InlineCtxt l2 = l2
167 maxLvl l1 InlineCtxt = l1
168 maxLvl l1@(Level maj1 min1) l2@(Level maj2 min2)
169 | (maj1 > maj2) || (maj1 == maj2 && min1 > min2) = l1
172 ltLvl :: Level -> Level -> Bool
173 ltLvl any_lvl InlineCtxt = False
174 ltLvl InlineCtxt (Level _ _) = True
175 ltLvl (Level maj1 min1) (Level maj2 min2)
176 = (maj1 < maj2) || (maj1 == maj2 && min1 < min2)
178 ltMajLvl :: Level -> Level -> Bool
179 -- Tells if one level belongs to a difft *lambda* level to another
180 ltMajLvl any_lvl InlineCtxt = False
181 ltMajLvl InlineCtxt (Level maj2 _) = 0 < maj2
182 ltMajLvl (Level maj1 _) (Level maj2 _) = maj1 < maj2
184 isTopLvl :: Level -> Bool
185 isTopLvl (Level 0 0) = True
186 isTopLvl other = False
188 isInlineCtxt :: Level -> Bool
189 isInlineCtxt InlineCtxt = True
190 isInlineCtxt other = False
192 instance Outputable Level where
193 ppr InlineCtxt = text "<INLINE>"
194 ppr (Level maj min) = hcat [ char '<', int maj, char ',', int min, char '>' ]
196 instance Eq Level where
197 InlineCtxt == InlineCtxt = True
198 (Level maj1 min1) == (Level maj2 min2) = maj1==maj2 && min1==min2
203 %************************************************************************
205 \subsection{Main level-setting code}
207 %************************************************************************
210 setLevels :: FloatOutSwitches
215 setLevels float_lams binds us
216 = initLvl us (do_them binds)
218 -- "do_them"'s main business is to thread the monad along
219 -- It gives each top binding the same empty envt, because
220 -- things unbound in the envt have level number zero implicitly
221 do_them :: [CoreBind] -> LvlM [LevelledBind]
223 do_them [] = returnLvl []
225 = lvlTopBind init_env b `thenLvl` \ (lvld_bind, _) ->
226 do_them bs `thenLvl` \ lvld_binds ->
227 returnLvl (lvld_bind : lvld_binds)
229 init_env = initialEnv float_lams
231 lvlTopBind env (NonRec binder rhs)
232 = lvlBind TopLevel tOP_LEVEL env (AnnNonRec binder (freeVars rhs))
233 -- Rhs can have no free vars!
235 lvlTopBind env (Rec pairs)
236 = lvlBind TopLevel tOP_LEVEL env (AnnRec [(b,freeVars rhs) | (b,rhs) <- pairs])
239 %************************************************************************
241 \subsection{Setting expression levels}
243 %************************************************************************
246 lvlExpr :: Level -- ctxt_lvl: Level of enclosing expression
247 -> LevelEnv -- Level of in-scope names/tyvars
248 -> CoreExprWithFVs -- input expression
249 -> LvlM LevelledExpr -- Result expression
252 The @ctxt_lvl@ is, roughly, the level of the innermost enclosing
253 binder. Here's an example
255 v = \x -> ...\y -> let r = case (..x..) of
259 When looking at the rhs of @r@, @ctxt_lvl@ will be 1 because that's
260 the level of @r@, even though it's inside a level-2 @\y@. It's
261 important that @ctxt_lvl@ is 1 and not 2 in @r@'s rhs, because we
262 don't want @lvlExpr@ to turn the scrutinee of the @case@ into an MFE
263 --- because it isn't a *maximal* free expression.
265 If there were another lambda in @r@'s rhs, it would get level-2 as well.
268 lvlExpr _ _ (_, AnnType ty) = returnLvl (Type ty)
269 lvlExpr _ env (_, AnnVar v) = returnLvl (lookupVar env v)
270 lvlExpr _ env (_, AnnLit lit) = returnLvl (Lit lit)
272 lvlExpr ctxt_lvl env (_, AnnApp fun arg)
273 = lvl_fun fun `thenLvl` \ fun' ->
274 lvlMFE False ctxt_lvl env arg `thenLvl` \ arg' ->
275 returnLvl (App fun' arg')
278 lvl_fun (_, AnnCase _ _ _ _) = lvlMFE True ctxt_lvl env fun
279 lvl_fun other = lvlExpr ctxt_lvl env fun
280 -- We don't do MFE on partial applications generally,
281 -- but we do if the function is big and hairy, like a case
283 lvlExpr ctxt_lvl env (_, AnnNote InlineMe expr)
284 -- Don't float anything out of an InlineMe; hence the iNLINE_CTXT
285 = lvlExpr iNLINE_CTXT env expr `thenLvl` \ expr' ->
286 returnLvl (Note InlineMe expr')
288 lvlExpr ctxt_lvl env (_, AnnNote note expr)
289 = lvlExpr ctxt_lvl env expr `thenLvl` \ expr' ->
290 returnLvl (Note note expr')
292 -- We don't split adjacent lambdas. That is, given
294 -- we don't float to give
295 -- \x -> let v = x+y in \y -> (v,y)
296 -- Why not? Because partial applications are fairly rare, and splitting
297 -- lambdas makes them more expensive.
299 lvlExpr ctxt_lvl env expr@(_, AnnLam bndr rhs)
300 = lvlMFE True new_lvl new_env body `thenLvl` \ new_body ->
301 returnLvl (mkLams new_bndrs new_body)
303 (bndrs, body) = collectAnnBndrs expr
304 (new_lvl, new_bndrs) = lvlLamBndrs ctxt_lvl bndrs
305 new_env = extendLvlEnv env new_bndrs
306 -- At one time we called a special verion of collectBinders,
307 -- which ignored coercions, because we don't want to split
308 -- a lambda like this (\x -> coerce t (\s -> ...))
309 -- This used to happen quite a bit in state-transformer programs,
310 -- but not nearly so much now non-recursive newtypes are transparent.
311 -- [See SetLevels rev 1.50 for a version with this approach.]
313 lvlExpr ctxt_lvl env (_, AnnLet (AnnNonRec bndr rhs) body)
314 | isUnLiftedType (idType bndr)
315 -- Treat unlifted let-bindings (let x = b in e) just like (case b of x -> e)
316 -- That is, leave it exactly where it is
317 -- We used to float unlifted bindings too (e.g. to get a cheap primop
318 -- outside a lambda (to see how, look at lvlBind in rev 1.58)
319 -- but an unrelated change meant that these unlifed bindings
320 -- could get to the top level which is bad. And there's not much point;
321 -- unlifted bindings are always cheap, and so hardly worth floating.
322 = lvlExpr ctxt_lvl env rhs `thenLvl` \ rhs' ->
323 lvlExpr incd_lvl env' body `thenLvl` \ body' ->
324 returnLvl (Let (NonRec bndr' rhs') body')
326 incd_lvl = incMinorLvl ctxt_lvl
327 bndr' = TB bndr incd_lvl
328 env' = extendLvlEnv env [bndr']
330 lvlExpr ctxt_lvl env (_, AnnLet bind body)
331 = lvlBind NotTopLevel ctxt_lvl env bind `thenLvl` \ (bind', new_env) ->
332 lvlExpr ctxt_lvl new_env body `thenLvl` \ body' ->
333 returnLvl (Let bind' body')
335 lvlExpr ctxt_lvl env (_, AnnCase expr case_bndr ty alts)
336 = lvlMFE True ctxt_lvl env expr `thenLvl` \ expr' ->
338 alts_env = extendCaseBndrLvlEnv env expr' case_bndr incd_lvl
340 mapLvl (lvl_alt alts_env) alts `thenLvl` \ alts' ->
341 returnLvl (Case expr' (TB case_bndr incd_lvl) ty alts')
343 incd_lvl = incMinorLvl ctxt_lvl
345 lvl_alt alts_env (con, bs, rhs)
346 = lvlMFE True incd_lvl new_env rhs `thenLvl` \ rhs' ->
347 returnLvl (con, bs', rhs')
349 bs' = [ TB b incd_lvl | b <- bs ]
350 new_env = extendLvlEnv alts_env bs'
353 @lvlMFE@ is just like @lvlExpr@, except that it might let-bind
354 the expression, so that it can itself be floated.
356 [NOTE: unlifted MFEs]
357 We don't float unlifted MFEs, which potentially loses big opportunites.
360 where h :: Int -> Int# is expensive. We'd like to float the (h y) outside
361 the \x, but we don't because it's unboxed. Possible solution: box it.
364 lvlMFE :: Bool -- True <=> strict context [body of case or let]
365 -> Level -- Level of innermost enclosing lambda/tylam
366 -> LevelEnv -- Level of in-scope names/tyvars
367 -> CoreExprWithFVs -- input expression
368 -> LvlM LevelledExpr -- Result expression
370 lvlMFE strict_ctxt ctxt_lvl env (_, AnnType ty)
371 = returnLvl (Type ty)
374 lvlMFE strict_ctxt ctxt_lvl env ann_expr@(fvs, _)
375 | isUnLiftedType ty -- Can't let-bind it; see [NOTE: unlifted MFEs]
376 || isInlineCtxt ctxt_lvl -- Don't float out of an __inline__ context
377 || exprIsTrivial expr -- Never float if it's trivial
378 || not good_destination
379 = -- Don't float it out
380 lvlExpr ctxt_lvl env ann_expr
382 | otherwise -- Float it out!
383 = lvlFloatRhs abs_vars dest_lvl env ann_expr `thenLvl` \ expr' ->
384 newLvlVar "lvl" abs_vars ty `thenLvl` \ var ->
385 returnLvl (Let (NonRec (TB var dest_lvl) expr')
386 (mkVarApps (Var var) abs_vars))
388 expr = deAnnotate ann_expr
390 dest_lvl = destLevel env fvs (isFunction ann_expr)
391 abs_vars = abstractVars dest_lvl env fvs
393 -- A decision to float entails let-binding this thing, and we only do
394 -- that if we'll escape a value lambda, or will go to the top level.
396 | dest_lvl `ltMajLvl` ctxt_lvl -- Escapes a value lambda
397 = not (exprIsCheap expr) || isTopLvl dest_lvl
398 -- Even if it escapes a value lambda, we only
399 -- float if it's not cheap (unless it'll get all the
400 -- way to the top). I've seen cases where we
401 -- float dozens of tiny free expressions, which cost
402 -- more to allocate than to evaluate.
403 -- NB: exprIsCheap is also true of bottom expressions, which
404 -- is good; we don't want to share them
406 -- It's only Really Bad to float a cheap expression out of a
407 -- strict context, because that builds a thunk that otherwise
408 -- would never be built. So another alternative would be to
410 -- || (strict_ctxt && not (exprIsBottom expr))
411 -- to the condition above. We should really try this out.
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
432 %************************************************************************
434 \subsection{Bindings}
436 %************************************************************************
438 The binding stuff works for top level too.
441 lvlBind :: TopLevelFlag -- Used solely to decide whether to clone
442 -> Level -- Context level; might be Top even for bindings nested in the RHS
443 -- of a top level binding
446 -> LvlM (LevelledBind, LevelEnv)
448 lvlBind top_lvl ctxt_lvl env (AnnNonRec bndr rhs@(rhs_fvs,_))
449 | isInlineCtxt ctxt_lvl -- Don't do anything inside InlineMe
450 = lvlExpr ctxt_lvl env rhs `thenLvl` \ rhs' ->
451 returnLvl (NonRec (TB bndr ctxt_lvl) rhs', env)
454 = -- No type abstraction; clone existing binder
455 lvlExpr dest_lvl env rhs `thenLvl` \ rhs' ->
456 cloneVar top_lvl env bndr ctxt_lvl dest_lvl `thenLvl` \ (env', bndr') ->
457 returnLvl (NonRec (TB bndr' dest_lvl) rhs', env')
460 = -- Yes, type abstraction; create a new binder, extend substitution, etc
461 lvlFloatRhs abs_vars dest_lvl env rhs `thenLvl` \ rhs' ->
462 newPolyBndrs dest_lvl env abs_vars [bndr] `thenLvl` \ (env', [bndr']) ->
463 returnLvl (NonRec (TB bndr' dest_lvl) rhs', env')
466 bind_fvs = rhs_fvs `unionVarSet` idFreeVars bndr
467 abs_vars = abstractVars dest_lvl env bind_fvs
468 dest_lvl = destLevel env bind_fvs (isFunction rhs)
473 lvlBind top_lvl ctxt_lvl env (AnnRec pairs)
474 | isInlineCtxt ctxt_lvl -- Don't do anything inside InlineMe
475 = mapLvl (lvlExpr ctxt_lvl env) rhss `thenLvl` \ rhss' ->
476 returnLvl (Rec ([TB b ctxt_lvl | b <- bndrs] `zip` rhss'), env)
479 = cloneRecVars top_lvl env bndrs ctxt_lvl dest_lvl `thenLvl` \ (new_env, new_bndrs) ->
480 mapLvl (lvlExpr ctxt_lvl new_env) rhss `thenLvl` \ new_rhss ->
481 returnLvl (Rec ([TB b dest_lvl | b <- new_bndrs] `zip` new_rhss), new_env)
483 | isSingleton pairs && count isId abs_vars > 1
484 = -- Special case for self recursion where there are
485 -- several variables carried around: build a local loop:
486 -- poly_f = \abs_vars. \lam_vars . letrec f = \lam_vars. rhs in f lam_vars
487 -- This just makes the closures a bit smaller. If we don't do
488 -- this, allocation rises significantly on some programs
490 -- We could elaborate it for the case where there are several
491 -- mutually functions, but it's quite a bit more complicated
493 -- This all seems a bit ad hoc -- sigh
495 (bndr,rhs) = head pairs
496 (rhs_lvl, abs_vars_w_lvls) = lvlLamBndrs dest_lvl abs_vars
497 rhs_env = extendLvlEnv env abs_vars_w_lvls
499 cloneVar NotTopLevel rhs_env bndr rhs_lvl rhs_lvl `thenLvl` \ (rhs_env', new_bndr) ->
501 (lam_bndrs, rhs_body) = collectAnnBndrs rhs
502 (body_lvl, new_lam_bndrs) = lvlLamBndrs rhs_lvl lam_bndrs
503 body_env = extendLvlEnv rhs_env' new_lam_bndrs
505 lvlExpr body_lvl body_env rhs_body `thenLvl` \ new_rhs_body ->
506 newPolyBndrs dest_lvl env abs_vars [bndr] `thenLvl` \ (poly_env, [poly_bndr]) ->
507 returnLvl (Rec [(TB poly_bndr dest_lvl,
508 mkLams abs_vars_w_lvls $
509 mkLams new_lam_bndrs $
510 Let (Rec [(TB new_bndr rhs_lvl, mkLams new_lam_bndrs new_rhs_body)])
511 (mkVarApps (Var new_bndr) lam_bndrs))],
514 | otherwise -- Non-null abs_vars
515 = newPolyBndrs dest_lvl env abs_vars bndrs `thenLvl` \ (new_env, new_bndrs) ->
516 mapLvl (lvlFloatRhs abs_vars dest_lvl new_env) rhss `thenLvl` \ new_rhss ->
517 returnLvl (Rec ([TB b dest_lvl | b <- new_bndrs] `zip` new_rhss), new_env)
520 (bndrs,rhss) = unzip pairs
522 -- Finding the free vars of the binding group is annoying
523 bind_fvs = (unionVarSets [ idFreeVars bndr `unionVarSet` rhs_fvs
524 | (bndr, (rhs_fvs,_)) <- pairs])
528 dest_lvl = destLevel env bind_fvs (all isFunction rhss)
529 abs_vars = abstractVars dest_lvl env bind_fvs
531 ----------------------------------------------------
532 -- Three help functons for the type-abstraction case
534 lvlFloatRhs abs_vars dest_lvl env rhs
535 = lvlExpr rhs_lvl rhs_env rhs `thenLvl` \ rhs' ->
536 returnLvl (mkLams abs_vars_w_lvls rhs')
538 (rhs_lvl, abs_vars_w_lvls) = lvlLamBndrs dest_lvl abs_vars
539 rhs_env = extendLvlEnv env abs_vars_w_lvls
543 %************************************************************************
545 \subsection{Deciding floatability}
547 %************************************************************************
550 lvlLamBndrs :: Level -> [CoreBndr] -> (Level, [TaggedBndr Level])
551 -- Compute the levels for the binders of a lambda group
552 -- The binders returned are exactly the same as the ones passed,
553 -- but they are now paired with a level
557 lvlLamBndrs lvl bndrs
558 = go (incMinorLvl lvl)
559 False -- Havn't bumped major level in this group
562 go old_lvl bumped_major rev_lvld_bndrs (bndr:bndrs)
563 | isId bndr && -- Go to the next major level if this is a value binder,
564 not bumped_major && -- and we havn't already gone to the next level (one jump per group)
565 not (isOneShotLambda bndr) -- and it isn't a one-shot lambda
566 = go new_lvl True (TB bndr new_lvl : rev_lvld_bndrs) bndrs
569 = go old_lvl bumped_major (TB bndr old_lvl : rev_lvld_bndrs) bndrs
572 new_lvl = incMajorLvl old_lvl
574 go old_lvl _ rev_lvld_bndrs []
575 = (old_lvl, reverse rev_lvld_bndrs)
576 -- a lambda like this (\x -> coerce t (\s -> ...))
577 -- This happens quite a bit in state-transformer programs
581 -- Destintion level is the max Id level of the expression
582 -- (We'll abstract the type variables, if any.)
583 destLevel :: LevelEnv -> VarSet -> Bool -> Level
584 destLevel env fvs is_function
586 && is_function = tOP_LEVEL -- Send functions to top level; see
587 -- the comments with isFunction
588 | otherwise = maxIdLevel env fvs
590 isFunction :: CoreExprWithFVs -> Bool
591 -- The idea here is that we want to float *functions* to
592 -- the top level. This saves no work, but
593 -- (a) it can make the host function body a lot smaller,
594 -- and hence inlinable.
595 -- (b) it can also save allocation when the function is recursive:
596 -- h = \x -> letrec f = \y -> ...f...y...x...
599 -- f = \x y -> ...(f x)...y...x...
601 -- No allocation for f now.
602 -- We may only want to do this if there are sufficiently few free
603 -- variables. We certainly only want to do it for values, and not for
604 -- constructors. So the simple thing is just to look for lambdas
605 isFunction (_, AnnLam b e) | isId b = True
606 | otherwise = isFunction e
607 isFunction (_, AnnNote n e) = isFunction e
608 isFunction other = False
612 %************************************************************************
614 \subsection{Free-To-Level Monad}
616 %************************************************************************
619 type LevelEnv = (FloatOutSwitches,
620 VarEnv Level, -- Domain is *post-cloned* TyVars and Ids
621 Subst, -- Domain is pre-cloned Ids; tracks the in-scope set
622 -- so that subtitution is capture-avoiding
623 IdEnv ([Var], LevelledExpr)) -- Domain is pre-cloned Ids
624 -- We clone let-bound variables so that they are still
625 -- distinct when floated out; hence the SubstEnv/IdEnv.
626 -- (see point 3 of the module overview comment).
627 -- We also use these envs when making a variable polymorphic
628 -- because we want to float it out past a big lambda.
630 -- The SubstEnv and IdEnv always implement the same mapping, but the
631 -- SubstEnv maps to CoreExpr and the IdEnv to LevelledExpr
632 -- Since the range is always a variable or type application,
633 -- there is never any difference between the two, but sadly
634 -- the types differ. The SubstEnv is used when substituting in
635 -- a variable's IdInfo; the IdEnv when we find a Var.
637 -- In addition the IdEnv records a list of tyvars free in the
638 -- type application, just so we don't have to call freeVars on
639 -- the type application repeatedly.
641 -- The domain of the both envs is *pre-cloned* Ids, though
643 -- The domain of the VarEnv Level is the *post-cloned* Ids
645 initialEnv :: FloatOutSwitches -> LevelEnv
646 initialEnv float_lams = (float_lams, emptyVarEnv, emptySubst, emptyVarEnv)
648 floatLams :: LevelEnv -> Bool
649 floatLams (FloatOutSw float_lams _, _, _, _) = float_lams
651 floatConsts :: LevelEnv -> Bool
652 floatConsts (FloatOutSw _ float_consts, _, _, _) = float_consts
654 extendLvlEnv :: LevelEnv -> [TaggedBndr Level] -> LevelEnv
655 -- Used when *not* cloning
656 extendLvlEnv (float_lams, lvl_env, subst, id_env) prs
658 foldl add_lvl lvl_env prs,
659 foldl del_subst subst prs,
660 foldl del_id id_env prs)
662 add_lvl env (TB v l) = extendVarEnv env v l
663 del_subst env (TB v _) = extendInScope env v
664 del_id env (TB v _) = delVarEnv env v
665 -- We must remove any clone for this variable name in case of
666 -- shadowing. This bit me in the following case
667 -- (in nofib/real/gg/Spark.hs):
670 -- ... -> case e of wild {
671 -- ... -> ... wild ...
675 -- The inside occurrence of @wild@ was being replaced with @ds@,
676 -- incorrectly, because the SubstEnv was still lying around. Ouch!
679 -- extendCaseBndrLvlEnv adds the mapping case-bndr->scrut-var if it can
680 -- (see point 4 of the module overview comment)
681 extendCaseBndrLvlEnv (float_lams, lvl_env, subst, id_env) (Var scrut_var) case_bndr lvl
683 extendVarEnv lvl_env case_bndr lvl,
684 extendIdSubst subst case_bndr (DoneEx (Var scrut_var)),
685 extendVarEnv id_env case_bndr ([scrut_var], Var scrut_var))
687 extendCaseBndrLvlEnv env scrut case_bndr lvl
688 = extendLvlEnv env [TB case_bndr lvl]
690 extendPolyLvlEnv dest_lvl (float_lams, lvl_env, subst, id_env) abs_vars bndr_pairs
692 foldl add_lvl lvl_env bndr_pairs,
693 foldl add_subst subst bndr_pairs,
694 foldl add_id id_env bndr_pairs)
696 add_lvl env (v,v') = extendVarEnv env v' dest_lvl
697 add_subst env (v,v') = extendIdSubst env v (DoneEx (mkVarApps (Var v') abs_vars))
698 add_id env (v,v') = extendVarEnv env v ((v':abs_vars), mkVarApps (Var v') abs_vars)
700 extendCloneLvlEnv lvl (float_lams, lvl_env, _, id_env) new_subst bndr_pairs
702 foldl add_lvl lvl_env bndr_pairs,
704 foldl add_id id_env bndr_pairs)
706 add_lvl env (v,v') = extendVarEnv env v' lvl
707 add_id env (v,v') = extendVarEnv env v ([v'], Var v')
710 maxIdLevel :: LevelEnv -> VarSet -> Level
711 maxIdLevel (_, lvl_env,_,id_env) var_set
712 = foldVarSet max_in tOP_LEVEL var_set
714 max_in in_var lvl = foldr max_out lvl (case lookupVarEnv id_env in_var of
715 Just (abs_vars, _) -> abs_vars
719 | isId out_var = case lookupVarEnv lvl_env out_var of
720 Just lvl' -> maxLvl lvl' lvl
722 | otherwise = lvl -- Ignore tyvars in *maxIdLevel*
724 lookupVar :: LevelEnv -> Id -> LevelledExpr
725 lookupVar (_, _, _, id_env) v = case lookupVarEnv id_env v of
726 Just (_, expr) -> expr
729 abstractVars :: Level -> LevelEnv -> VarSet -> [Var]
730 -- Find the variables in fvs, free vars of the target expresion,
731 -- whose level is greater than the destination level
732 -- These are the ones we are going to abstract out
733 abstractVars dest_lvl env fvs
734 = uniq (sortLe le [var | fv <- varSetElems fvs, var <- absVarsOf dest_lvl env fv])
736 -- Sort the variables so we don't get
737 -- mixed-up tyvars and Ids; it's just messy
738 v1 `le` v2 = case (isId v1, isId v2) of
739 (True, False) -> False
740 (False, True) -> True
741 other -> v1 <= v2 -- Same family
743 uniq :: [Var] -> [Var]
744 -- Remove adjacent duplicates; the sort will have brought them together
745 uniq (v1:v2:vs) | v1 == v2 = uniq (v2:vs)
746 | otherwise = v1 : uniq (v2:vs)
749 absVarsOf :: Level -> LevelEnv -> Var -> [Var]
750 -- If f is free in the expression, and f maps to poly_f a b c in the
751 -- current substitution, then we must report a b c as candidate type
753 absVarsOf dest_lvl (_, lvl_env, _, id_env) v
755 = [zap av2 | av1 <- lookup_avs v, av2 <- add_tyvars av1, abstract_me av2]
758 = if abstract_me v then [v] else []
761 abstract_me v = case lookupVarEnv lvl_env v of
762 Just lvl -> dest_lvl `ltLvl` lvl
765 lookup_avs v = case lookupVarEnv id_env v of
766 Just (abs_vars, _) -> abs_vars
769 add_tyvars v | isId v = v : varSetElems (idFreeTyVars v)
772 -- We are going to lambda-abstract, so nuke any IdInfo,
773 -- and add the tyvars of the Id (if necessary)
774 zap v | isId v = WARN( workerExists (idWorkerInfo v) ||
775 not (isEmptyCoreRules (idSpecialisation v)),
776 text "absVarsOf: discarding info on" <+> ppr v )
777 setIdInfo v vanillaIdInfo
782 type LvlM result = UniqSM result
791 newPolyBndrs dest_lvl env abs_vars bndrs
792 = getUniquesUs `thenLvl` \ uniqs ->
794 new_bndrs = zipWith mk_poly_bndr bndrs uniqs
796 returnLvl (extendPolyLvlEnv dest_lvl env abs_vars (bndrs `zip` new_bndrs), new_bndrs)
798 mk_poly_bndr bndr uniq = mkSysLocalUnencoded (mkFastString str) uniq poly_ty
800 str = "poly_" ++ occNameUserString (getOccName bndr)
801 poly_ty = mkPiTypes abs_vars (idType bndr)
805 -> [CoreBndr] -> Type -- Abstract wrt these bndrs
807 newLvlVar str vars body_ty
808 = getUniqueUs `thenLvl` \ uniq ->
809 returnUs (mkSysLocalUnencoded (mkFastString str) uniq (mkPiTypes vars body_ty))
811 -- The deeply tiresome thing is that we have to apply the substitution
812 -- to the rules inside each Id. Grr. But it matters.
814 cloneVar :: TopLevelFlag -> LevelEnv -> Id -> Level -> Level -> LvlM (LevelEnv, Id)
815 cloneVar TopLevel env v ctxt_lvl dest_lvl
816 = returnUs (env, v) -- Don't clone top level things
817 cloneVar NotTopLevel env@(_,_,subst,_) v ctxt_lvl dest_lvl
819 getUs `thenLvl` \ us ->
821 (subst', v1) = substAndCloneId subst us v
822 v2 = zap_demand ctxt_lvl dest_lvl v1
823 env' = extendCloneLvlEnv dest_lvl env subst' [(v,v2)]
827 cloneRecVars :: TopLevelFlag -> LevelEnv -> [Id] -> Level -> Level -> LvlM (LevelEnv, [Id])
828 cloneRecVars TopLevel env vs ctxt_lvl dest_lvl
829 = returnUs (env, vs) -- Don't clone top level things
830 cloneRecVars NotTopLevel env@(_,_,subst,_) vs ctxt_lvl dest_lvl
831 = ASSERT( all isId vs )
832 getUs `thenLvl` \ us ->
834 (subst', vs1) = substAndCloneRecIds subst us vs
835 vs2 = map (zap_demand ctxt_lvl dest_lvl) vs1
836 env' = extendCloneLvlEnv dest_lvl env subst' (vs `zip` vs2)
840 -- VERY IMPORTANT: we must zap the demand info
841 -- if the thing is going to float out past a lambda
842 zap_demand dest_lvl ctxt_lvl id
843 | ctxt_lvl == dest_lvl = id -- Stays put
844 | otherwise = zapDemandIdInfo id -- Floats out