2 % (c) The AQUA Project, Glasgow University, 1993-1998
4 \section[Simplify]{The main module of the simplifier}
7 module Simplify ( simplTopBinds, simplExpr ) where
9 #include "HsVersions.h"
13 import Type hiding ( substTy, extendTvSubst, substTyVar )
16 import FamInstEnv ( FamInstEnv )
18 import MkId ( seqId, realWorldPrimId )
19 import MkCore ( mkImpossibleExpr )
22 import Name ( mkSystemVarName, isExternalName )
24 import OptCoercion ( optCoercion )
25 import FamInstEnv ( topNormaliseType )
26 import DataCon ( DataCon, dataConWorkId, dataConRepStrictness )
27 import CoreMonad ( SimplifierSwitch(..), Tick(..) )
29 import Demand ( isStrictDmd, splitStrictSig )
30 import PprCore ( pprParendExpr, pprCoreExpr )
31 import CoreUnfold ( mkUnfolding, mkCoreUnfolding
32 , mkInlineUnfolding, mkSimpleUnfolding
33 , exprIsConApp_maybe, callSiteInline, CallCtxt(..) )
35 import qualified CoreSubst
36 import CoreArity ( exprArity )
37 import Rules ( lookupRule, getRules )
38 import BasicTypes ( isMarkedStrict, Arity )
39 import CostCentre ( currentCCS, pushCCisNop )
40 import TysPrim ( realWorldStatePrimTy )
41 import BasicTypes ( TopLevelFlag(..), isTopLevel, RecFlag(..) )
42 import MonadUtils ( foldlM, mapAccumLM )
43 import Maybes ( orElse )
44 import Data.List ( mapAccumL )
50 The guts of the simplifier is in this module, but the driver loop for
51 the simplifier is in SimplCore.lhs.
54 -----------------------------------------
55 *** IMPORTANT NOTE ***
56 -----------------------------------------
57 The simplifier used to guarantee that the output had no shadowing, but
58 it does not do so any more. (Actually, it never did!) The reason is
59 documented with simplifyArgs.
62 -----------------------------------------
63 *** IMPORTANT NOTE ***
64 -----------------------------------------
65 Many parts of the simplifier return a bunch of "floats" as well as an
66 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
68 All "floats" are let-binds, not case-binds, but some non-rec lets may
69 be unlifted (with RHS ok-for-speculation).
73 -----------------------------------------
74 ORGANISATION OF FUNCTIONS
75 -----------------------------------------
77 - simplify all top-level binders
78 - for NonRec, call simplRecOrTopPair
79 - for Rec, call simplRecBind
82 ------------------------------
83 simplExpr (applied lambda) ==> simplNonRecBind
84 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
85 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
87 ------------------------------
88 simplRecBind [binders already simplfied]
89 - use simplRecOrTopPair on each pair in turn
91 simplRecOrTopPair [binder already simplified]
92 Used for: recursive bindings (top level and nested)
93 top-level non-recursive bindings
95 - check for PreInlineUnconditionally
99 Used for: non-top-level non-recursive bindings
100 beta reductions (which amount to the same thing)
101 Because it can deal with strict arts, it takes a
102 "thing-inside" and returns an expression
104 - check for PreInlineUnconditionally
105 - simplify binder, including its IdInfo
114 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
115 Used for: binding case-binder and constr args in a known-constructor case
116 - check for PreInLineUnconditionally
120 ------------------------------
121 simplLazyBind: [binder already simplified, RHS not]
122 Used for: recursive bindings (top level and nested)
123 top-level non-recursive bindings
124 non-top-level, but *lazy* non-recursive bindings
125 [must not be strict or unboxed]
126 Returns floats + an augmented environment, not an expression
127 - substituteIdInfo and add result to in-scope
128 [so that rules are available in rec rhs]
131 - float if exposes constructor or PAP
135 completeNonRecX: [binder and rhs both simplified]
136 - if the the thing needs case binding (unlifted and not ok-for-spec)
142 completeBind: [given a simplified RHS]
143 [used for both rec and non-rec bindings, top level and not]
144 - try PostInlineUnconditionally
145 - add unfolding [this is the only place we add an unfolding]
150 Right hand sides and arguments
151 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
152 In many ways we want to treat
153 (a) the right hand side of a let(rec), and
154 (b) a function argument
155 in the same way. But not always! In particular, we would
156 like to leave these arguments exactly as they are, so they
157 will match a RULE more easily.
162 It's harder to make the rule match if we ANF-ise the constructor,
163 or eta-expand the PAP:
165 f (let { a = g x; b = h x } in (a,b))
168 On the other hand if we see the let-defns
173 then we *do* want to ANF-ise and eta-expand, so that p and q
174 can be safely inlined.
176 Even floating lets out is a bit dubious. For let RHS's we float lets
177 out if that exposes a value, so that the value can be inlined more vigorously.
180 r = let x = e in (x,x)
182 Here, if we float the let out we'll expose a nice constructor. We did experiments
183 that showed this to be a generally good thing. But it was a bad thing to float
184 lets out unconditionally, because that meant they got allocated more often.
186 For function arguments, there's less reason to expose a constructor (it won't
187 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
188 So for the moment we don't float lets out of function arguments either.
193 For eta expansion, we want to catch things like
195 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
197 If the \x was on the RHS of a let, we'd eta expand to bring the two
198 lambdas together. And in general that's a good thing to do. Perhaps
199 we should eta expand wherever we find a (value) lambda? Then the eta
200 expansion at a let RHS can concentrate solely on the PAP case.
203 %************************************************************************
205 \subsection{Bindings}
207 %************************************************************************
210 simplTopBinds :: SimplEnv -> [InBind] -> SimplM SimplEnv
212 simplTopBinds env0 binds0
213 = do { -- Put all the top-level binders into scope at the start
214 -- so that if a transformation rule has unexpectedly brought
215 -- anything into scope, then we don't get a complaint about that.
216 -- It's rather as if the top-level binders were imported.
217 ; env1 <- simplRecBndrs env0 (bindersOfBinds binds0)
218 ; dflags <- getDOptsSmpl
219 ; let dump_flag = dopt Opt_D_verbose_core2core dflags
220 ; env2 <- simpl_binds dump_flag env1 binds0
221 ; freeTick SimplifierDone
224 -- We need to track the zapped top-level binders, because
225 -- they should have their fragile IdInfo zapped (notably occurrence info)
226 -- That's why we run down binds and bndrs' simultaneously.
228 -- The dump-flag emits a trace for each top-level binding, which
229 -- helps to locate the tracing for inlining and rule firing
230 simpl_binds :: Bool -> SimplEnv -> [InBind] -> SimplM SimplEnv
231 simpl_binds _ env [] = return env
232 simpl_binds dump env (bind:binds) = do { env' <- trace_bind dump bind $
234 ; simpl_binds dump env' binds }
236 trace_bind True bind = pprTrace "SimplBind" (ppr (bindersOf bind))
237 trace_bind False _ = \x -> x
239 simpl_bind env (Rec pairs) = simplRecBind env TopLevel pairs
240 simpl_bind env (NonRec b r) = simplRecOrTopPair env' TopLevel b b' r
242 (env', b') = addBndrRules env b (lookupRecBndr env b)
246 %************************************************************************
248 \subsection{Lazy bindings}
250 %************************************************************************
252 simplRecBind is used for
253 * recursive bindings only
256 simplRecBind :: SimplEnv -> TopLevelFlag
259 simplRecBind env0 top_lvl pairs0
260 = do { let (env_with_info, triples) = mapAccumL add_rules env0 pairs0
261 ; env1 <- go (zapFloats env_with_info) triples
262 ; return (env0 `addRecFloats` env1) }
263 -- addFloats adds the floats from env1,
264 -- _and_ updates env0 with the in-scope set from env1
266 add_rules :: SimplEnv -> (InBndr,InExpr) -> (SimplEnv, (InBndr, OutBndr, InExpr))
267 -- Add the (substituted) rules to the binder
268 add_rules env (bndr, rhs) = (env', (bndr, bndr', rhs))
270 (env', bndr') = addBndrRules env bndr (lookupRecBndr env bndr)
272 go env [] = return env
274 go env ((old_bndr, new_bndr, rhs) : pairs)
275 = do { env' <- simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
279 simplOrTopPair is used for
280 * recursive bindings (whether top level or not)
281 * top-level non-recursive bindings
283 It assumes the binder has already been simplified, but not its IdInfo.
286 simplRecOrTopPair :: SimplEnv
288 -> InId -> OutBndr -> InExpr -- Binder and rhs
289 -> SimplM SimplEnv -- Returns an env that includes the binding
291 simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
292 | preInlineUnconditionally env top_lvl old_bndr rhs -- Check for unconditional inline
293 = do { tick (PreInlineUnconditionally old_bndr)
294 ; return (extendIdSubst env old_bndr (mkContEx env rhs)) }
297 = simplLazyBind env top_lvl Recursive old_bndr new_bndr rhs env
298 -- May not actually be recursive, but it doesn't matter
302 simplLazyBind is used for
303 * [simplRecOrTopPair] recursive bindings (whether top level or not)
304 * [simplRecOrTopPair] top-level non-recursive bindings
305 * [simplNonRecE] non-top-level *lazy* non-recursive bindings
308 1. It assumes that the binder is *already* simplified,
309 and is in scope, and its IdInfo too, except unfolding
311 2. It assumes that the binder type is lifted.
313 3. It does not check for pre-inline-unconditionallly;
314 that should have been done already.
317 simplLazyBind :: SimplEnv
318 -> TopLevelFlag -> RecFlag
319 -> InId -> OutId -- Binder, both pre-and post simpl
320 -- The OutId has IdInfo, except arity, unfolding
321 -> InExpr -> SimplEnv -- The RHS and its environment
324 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
325 = -- pprTrace "simplLazyBind" ((ppr bndr <+> ppr bndr1) $$ ppr rhs $$ ppr (seIdSubst rhs_se)) $
326 do { let rhs_env = rhs_se `setInScope` env
327 (tvs, body) = case collectTyBinders rhs of
328 (tvs, body) | not_lam body -> (tvs,body)
329 | otherwise -> ([], rhs)
330 not_lam (Lam _ _) = False
332 -- Do not do the "abstract tyyvar" thing if there's
333 -- a lambda inside, becuase it defeats eta-reduction
334 -- f = /\a. \x. g a x
337 ; (body_env, tvs') <- simplBinders rhs_env tvs
338 -- See Note [Floating and type abstraction] in SimplUtils
341 ; (body_env1, body1) <- simplExprF body_env body mkRhsStop
342 -- ANF-ise a constructor or PAP rhs
343 ; (body_env2, body2) <- prepareRhs top_lvl body_env1 bndr1 body1
346 <- if not (doFloatFromRhs top_lvl is_rec False body2 body_env2)
347 then -- No floating, revert to body1
348 do { rhs' <- mkLam env tvs' (wrapFloats body_env1 body1)
349 ; return (env, rhs') }
351 else if null tvs then -- Simple floating
352 do { tick LetFloatFromLet
353 ; return (addFloats env body_env2, body2) }
355 else -- Do type-abstraction first
356 do { tick LetFloatFromLet
357 ; (poly_binds, body3) <- abstractFloats tvs' body_env2 body2
358 ; rhs' <- mkLam env tvs' body3
359 ; env' <- foldlM (addPolyBind top_lvl) env poly_binds
360 ; return (env', rhs') }
362 ; completeBind env' top_lvl bndr bndr1 rhs' }
365 A specialised variant of simplNonRec used when the RHS is already simplified,
366 notably in knownCon. It uses case-binding where necessary.
369 simplNonRecX :: SimplEnv
370 -> InId -- Old binder
371 -> OutExpr -- Simplified RHS
374 simplNonRecX env bndr new_rhs
375 | isDeadBinder bndr -- Not uncommon; e.g. case (a,b) of b { (p,q) -> p }
376 = return env -- Here b is dead, and we avoid creating
377 | otherwise -- the binding b = (a,b)
378 = do { (env', bndr') <- simplBinder env bndr
379 ; completeNonRecX NotTopLevel env' (isStrictId bndr) bndr bndr' new_rhs }
380 -- simplNonRecX is only used for NotTopLevel things
382 completeNonRecX :: TopLevelFlag -> SimplEnv
384 -> InId -- Old binder
385 -> OutId -- New binder
386 -> OutExpr -- Simplified RHS
389 completeNonRecX top_lvl env is_strict old_bndr new_bndr new_rhs
390 = do { (env1, rhs1) <- prepareRhs top_lvl (zapFloats env) new_bndr new_rhs
392 if doFloatFromRhs NotTopLevel NonRecursive is_strict rhs1 env1
393 then do { tick LetFloatFromLet
394 ; return (addFloats env env1, rhs1) } -- Add the floats to the main env
395 else return (env, wrapFloats env1 rhs1) -- Wrap the floats around the RHS
396 ; completeBind env2 NotTopLevel old_bndr new_bndr rhs2 }
399 {- No, no, no! Do not try preInlineUnconditionally in completeNonRecX
400 Doing so risks exponential behaviour, because new_rhs has been simplified once already
401 In the cases described by the folowing commment, postInlineUnconditionally will
402 catch many of the relevant cases.
403 -- This happens; for example, the case_bndr during case of
404 -- known constructor: case (a,b) of x { (p,q) -> ... }
405 -- Here x isn't mentioned in the RHS, so we don't want to
406 -- create the (dead) let-binding let x = (a,b) in ...
408 -- Similarly, single occurrences can be inlined vigourously
409 -- e.g. case (f x, g y) of (a,b) -> ....
410 -- If a,b occur once we can avoid constructing the let binding for them.
412 Furthermore in the case-binding case preInlineUnconditionally risks extra thunks
413 -- Consider case I# (quotInt# x y) of
414 -- I# v -> let w = J# v in ...
415 -- If we gaily inline (quotInt# x y) for v, we end up building an
417 -- let w = J# (quotInt# x y) in ...
418 -- because quotInt# can fail.
420 | preInlineUnconditionally env NotTopLevel bndr new_rhs
421 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
424 ----------------------------------
425 prepareRhs takes a putative RHS, checks whether it's a PAP or
426 constructor application and, if so, converts it to ANF, so that the
427 resulting thing can be inlined more easily. Thus
434 We also want to deal well cases like this
435 v = (f e1 `cast` co) e2
436 Here we want to make e1,e2 trivial and get
437 x1 = e1; x2 = e2; v = (f x1 `cast` co) v2
438 That's what the 'go' loop in prepareRhs does
441 prepareRhs :: TopLevelFlag -> SimplEnv -> OutId -> OutExpr -> SimplM (SimplEnv, OutExpr)
442 -- Adds new floats to the env iff that allows us to return a good RHS
443 prepareRhs top_lvl env id (Cast rhs co) -- Note [Float coercions]
444 | (ty1, _ty2) <- coercionKind co -- Do *not* do this if rhs has an unlifted type
445 , not (isUnLiftedType ty1) -- see Note [Float coercions (unlifted)]
446 = do { (env', rhs') <- makeTrivialWithInfo top_lvl env sanitised_info rhs
447 ; return (env', Cast rhs' co) }
449 sanitised_info = vanillaIdInfo `setStrictnessInfo` strictnessInfo info
450 `setDemandInfo` demandInfo info
453 prepareRhs top_lvl env0 _ rhs0
454 = do { (_is_exp, env1, rhs1) <- go 0 env0 rhs0
455 ; return (env1, rhs1) }
457 go n_val_args env (Cast rhs co)
458 = do { (is_exp, env', rhs') <- go n_val_args env rhs
459 ; return (is_exp, env', Cast rhs' co) }
460 go n_val_args env (App fun (Type ty))
461 = do { (is_exp, env', rhs') <- go n_val_args env fun
462 ; return (is_exp, env', App rhs' (Type ty)) }
463 go n_val_args env (App fun arg)
464 = do { (is_exp, env', fun') <- go (n_val_args+1) env fun
466 True -> do { (env'', arg') <- makeTrivial top_lvl env' arg
467 ; return (True, env'', App fun' arg') }
468 False -> return (False, env, App fun arg) }
469 go n_val_args env (Var fun)
470 = return (is_exp, env, Var fun)
472 is_exp = isExpandableApp fun n_val_args -- The fun a constructor or PAP
473 -- See Note [CONLIKE pragma] in BasicTypes
474 -- The definition of is_exp should match that in
475 -- OccurAnal.occAnalApp
478 = return (False, env, other)
482 Note [Float coercions]
483 ~~~~~~~~~~~~~~~~~~~~~~
484 When we find the binding
486 we'd like to transform it to
488 x = x `cast` co -- A trivial binding
489 There's a chance that e will be a constructor application or function, or something
490 like that, so moving the coerion to the usage site may well cancel the coersions
491 and lead to further optimisation. Example:
494 data instance T Int = T Int
496 foo :: Int -> Int -> Int
501 go n = case x of { T m -> go (n-m) }
502 -- This case should optimise
504 Note [Preserve strictness when floating coercions]
505 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
506 In the Note [Float coercions] transformation, keep the strictness info.
508 f = e `cast` co -- f has strictness SSL
510 f' = e -- f' also has strictness SSL
511 f = f' `cast` co -- f still has strictness SSL
513 Its not wrong to drop it on the floor, but better to keep it.
515 Note [Float coercions (unlifted)]
516 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
517 BUT don't do [Float coercions] if 'e' has an unlifted type.
520 foo :: Int = (error (# Int,Int #) "urk")
521 `cast` CoUnsafe (# Int,Int #) Int
523 If do the makeTrivial thing to the error call, we'll get
524 foo = case error (# Int,Int #) "urk" of v -> v `cast` ...
525 But 'v' isn't in scope!
527 These strange casts can happen as a result of case-of-case
528 bar = case (case x of { T -> (# 2,3 #); F -> error "urk" }) of
533 makeTrivial :: TopLevelFlag -> SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
534 -- Binds the expression to a variable, if it's not trivial, returning the variable
535 makeTrivial top_lvl env expr = makeTrivialWithInfo top_lvl env vanillaIdInfo expr
537 makeTrivialWithInfo :: TopLevelFlag -> SimplEnv -> IdInfo
538 -> OutExpr -> SimplM (SimplEnv, OutExpr)
539 -- Propagate strictness and demand info to the new binder
540 -- Note [Preserve strictness when floating coercions]
541 -- Returned SimplEnv has same substitution as incoming one
542 makeTrivialWithInfo top_lvl env info expr
543 | exprIsTrivial expr -- Already trivial
544 || not (bindingOk top_lvl expr expr_ty) -- Cannot trivialise
545 -- See Note [Cannot trivialise]
547 | otherwise -- See Note [Take care] below
548 = do { uniq <- getUniqueM
549 ; let name = mkSystemVarName uniq (fsLit "a")
550 var = mkLocalIdWithInfo name expr_ty info
551 ; env' <- completeNonRecX top_lvl env False var var expr
552 ; expr' <- simplVar env' var
553 ; return (env', expr') }
554 -- The simplVar is needed becase we're constructing a new binding
556 -- And if rhs is of form (rhs1 |> co), then we might get
559 -- and now a's RHS is trivial and can be substituted out, and that
560 -- is what completeNonRecX will do
561 -- To put it another way, it's as if we'd simplified
562 -- let var = e in var
564 expr_ty = exprType expr
566 bindingOk :: TopLevelFlag -> CoreExpr -> Type -> Bool
567 -- True iff we can have a binding of this expression at this level
568 -- Precondition: the type is the type of the expression
569 bindingOk top_lvl _ expr_ty
570 | isTopLevel top_lvl = not (isUnLiftedType expr_ty)
574 Note [Cannot trivialise]
575 ~~~~~~~~~~~~~~~~~~~~~~~~
582 Then we can't ANF-ise foo, even though we'd like to, because
583 we can't make a top-level binding for the Addr# (f 3). And if
584 so we don't want to turn it into
585 foo = let x = f 3 in Bar x
586 because we'll just end up inlining x back, and that makes the
587 simplifier loop. Better not to ANF-ise it at all.
589 A case in point is literal strings (a MachStr is not regarded as
594 We don't want to ANF-ise this.
596 %************************************************************************
598 \subsection{Completing a lazy binding}
600 %************************************************************************
603 * deals only with Ids, not TyVars
604 * takes an already-simplified binder and RHS
605 * is used for both recursive and non-recursive bindings
606 * is used for both top-level and non-top-level bindings
608 It does the following:
609 - tries discarding a dead binding
610 - tries PostInlineUnconditionally
611 - add unfolding [this is the only place we add an unfolding]
614 It does *not* attempt to do let-to-case. Why? Because it is used for
615 - top-level bindings (when let-to-case is impossible)
616 - many situations where the "rhs" is known to be a WHNF
617 (so let-to-case is inappropriate).
619 Nor does it do the atomic-argument thing
622 completeBind :: SimplEnv
623 -> TopLevelFlag -- Flag stuck into unfolding
624 -> InId -- Old binder
625 -> OutId -> OutExpr -- New binder and RHS
627 -- completeBind may choose to do its work
628 -- * by extending the substitution (e.g. let x = y in ...)
629 -- * or by adding to the floats in the envt
631 completeBind env top_lvl old_bndr new_bndr new_rhs
632 = do { let old_info = idInfo old_bndr
633 old_unf = unfoldingInfo old_info
634 occ_info = occInfo old_info
636 ; new_unfolding <- simplUnfolding env top_lvl old_bndr occ_info new_rhs old_unf
638 ; if postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs new_unfolding
639 -- Inline and discard the binding
640 then do { tick (PostInlineUnconditionally old_bndr)
641 ; -- pprTrace "postInlineUnconditionally" (ppr old_bndr <+> equals <+> ppr new_rhs) $
642 return (extendIdSubst env old_bndr (DoneEx new_rhs)) }
643 -- Use the substitution to make quite, quite sure that the
644 -- substitution will happen, since we are going to discard the binding
646 else return (addNonRecWithUnf env new_bndr new_rhs new_unfolding) }
648 ------------------------------
649 addPolyBind :: TopLevelFlag -> SimplEnv -> OutBind -> SimplM SimplEnv
650 -- Add a new binding to the environment, complete with its unfolding
651 -- but *do not* do postInlineUnconditionally, because we have already
652 -- processed some of the scope of the binding
653 -- We still want the unfolding though. Consider
655 -- x = /\a. let y = ... in Just y
657 -- Then we float the y-binding out (via abstractFloats and addPolyBind)
658 -- but 'x' may well then be inlined in 'body' in which case we'd like the
659 -- opportunity to inline 'y' too.
661 addPolyBind top_lvl env (NonRec poly_id rhs)
662 = do { unfolding <- simplUnfolding env top_lvl poly_id NoOccInfo rhs noUnfolding
663 -- Assumes that poly_id did not have an INLINE prag
664 -- which is perhaps wrong. ToDo: think about this
665 ; return (addNonRecWithUnf env poly_id rhs unfolding) }
667 addPolyBind _ env bind@(Rec _) = return (extendFloats env bind)
668 -- Hack: letrecs are more awkward, so we extend "by steam"
669 -- without adding unfoldings etc. At worst this leads to
670 -- more simplifier iterations
672 ------------------------------
673 addNonRecWithUnf :: SimplEnv
674 -> OutId -> OutExpr -- New binder and RHS
675 -> Unfolding -- New unfolding
677 addNonRecWithUnf env new_bndr new_rhs new_unfolding
678 = let new_arity = exprArity new_rhs
679 old_arity = idArity new_bndr
680 info1 = idInfo new_bndr `setArityInfo` new_arity
682 -- Unfolding info: Note [Setting the new unfolding]
683 info2 = info1 `setUnfoldingInfo` new_unfolding
685 -- Demand info: Note [Setting the demand info]
686 info3 | isEvaldUnfolding new_unfolding = zapDemandInfo info2 `orElse` info2
689 final_id = new_bndr `setIdInfo` info3
690 dmd_arity = length $ fst $ splitStrictSig $ idStrictness new_bndr
692 ASSERT( isId new_bndr )
693 WARN( new_arity < old_arity || new_arity < dmd_arity,
694 (ptext (sLit "Arity decrease:") <+> (ppr final_id <+> ppr old_arity
695 <+> ppr new_arity <+> ppr dmd_arity) $$ ppr new_rhs) )
696 -- Note [Arity decrease]
698 final_id `seq` -- This seq forces the Id, and hence its IdInfo,
699 -- and hence any inner substitutions
700 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
701 addNonRec env final_id new_rhs
702 -- The addNonRec adds it to the in-scope set too
704 ------------------------------
705 simplUnfolding :: SimplEnv-> TopLevelFlag
707 -> OccInfo -> OutExpr
708 -> Unfolding -> SimplM Unfolding
709 -- Note [Setting the new unfolding]
710 simplUnfolding env _ _ _ _ (DFunUnfolding ar con ops)
711 = return (DFunUnfolding ar con ops')
713 ops' = map (substExpr (text "simplUnfolding") env) ops
715 simplUnfolding env top_lvl id _ _
716 (CoreUnfolding { uf_tmpl = expr, uf_arity = arity
717 , uf_src = src, uf_guidance = guide })
719 = do { expr' <- simplExpr rule_env expr
720 ; let src' = CoreSubst.substUnfoldingSource (mkCoreSubst (text "inline-unf") env) src
721 is_top_lvl = isTopLevel top_lvl
724 return (mkUnfolding src' is_top_lvl (isBottomingId id) expr')
725 -- If the guidance is UnfIfGoodArgs, this is an INLINABLE
726 -- unfolding, and we need to make sure the guidance is kept up
727 -- to date with respect to any changes in the unfolding.
729 return (mkCoreUnfolding src' is_top_lvl expr' arity guide)
730 -- See Note [Top-level flag on inline rules] in CoreUnfold
733 act = idInlineActivation id
734 rule_env = updMode (updModeForInlineRules act) env
735 -- See Note [Simplifying inside InlineRules] in SimplUtils
737 simplUnfolding _ top_lvl id _occ_info new_rhs _
738 = return (mkUnfolding InlineRhs (isTopLevel top_lvl) (isBottomingId id) new_rhs)
739 -- We make an unfolding *even for loop-breakers*.
740 -- Reason: (a) It might be useful to know that they are WHNF
741 -- (b) In TidyPgm we currently assume that, if we want to
742 -- expose the unfolding then indeed we *have* an unfolding
743 -- to expose. (We could instead use the RHS, but currently
744 -- we don't.) The simple thing is always to have one.
747 Note [Arity decrease]
748 ~~~~~~~~~~~~~~~~~~~~~
749 Generally speaking the arity of a binding should not decrease. But it *can*
750 legitimately happen becuase of RULES. Eg
752 where g has arity 2, will have arity 2. But if there's a rewrite rule
754 where h has arity 1, then f's arity will decrease. Here's a real-life example,
755 which is in the output of Specialise:
758 $dm {Arity 2} = \d.\x. op d
759 {-# RULES forall d. $dm Int d = $s$dm #-}
761 dInt = MkD .... opInt ...
762 opInt {Arity 1} = $dm dInt
764 $s$dm {Arity 0} = \x. op dInt }
766 Here opInt has arity 1; but when we apply the rule its arity drops to 0.
767 That's why Specialise goes to a little trouble to pin the right arity
768 on specialised functions too.
770 Note [Setting the new unfolding]
771 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
772 * If there's an INLINE pragma, we simplify the RHS gently. Maybe we
773 should do nothing at all, but simplifying gently might get rid of
776 * If not, we make an unfolding from the new RHS. But *only* for
777 non-loop-breakers. Making loop breakers not have an unfolding at all
778 means that we can avoid tests in exprIsConApp, for example. This is
779 important: if exprIsConApp says 'yes' for a recursive thing, then we
780 can get into an infinite loop
782 If there's an InlineRule on a loop breaker, we hang on to the inlining.
783 It's pretty dodgy, but the user did say 'INLINE'. May need to revisit
786 Note [Setting the demand info]
787 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
788 If the unfolding is a value, the demand info may
789 go pear-shaped, so we nuke it. Example:
791 case x of (p,q) -> h p q x
792 Here x is certainly demanded. But after we've nuked
793 the case, we'll get just
794 let x = (a,b) in h a b x
795 and now x is not demanded (I'm assuming h is lazy)
796 This really happens. Similarly
797 let f = \x -> e in ...f..f...
798 After inlining f at some of its call sites the original binding may
799 (for example) be no longer strictly demanded.
800 The solution here is a bit ad hoc...
803 %************************************************************************
805 \subsection[Simplify-simplExpr]{The main function: simplExpr}
807 %************************************************************************
809 The reason for this OutExprStuff stuff is that we want to float *after*
810 simplifying a RHS, not before. If we do so naively we get quadratic
811 behaviour as things float out.
813 To see why it's important to do it after, consider this (real) example:
827 a -- Can't inline a this round, cos it appears twice
831 Each of the ==> steps is a round of simplification. We'd save a
832 whole round if we float first. This can cascade. Consider
837 let f = let d1 = ..d.. in \y -> e
841 in \x -> ...(\y ->e)...
843 Only in this second round can the \y be applied, and it
844 might do the same again.
848 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
849 simplExpr env expr = simplExprC env expr mkBoringStop
851 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
852 -- Simplify an expression, given a continuation
853 simplExprC env expr cont
854 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
855 do { (env', expr') <- simplExprF (zapFloats env) expr cont
856 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
857 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
858 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
859 return (wrapFloats env' expr') }
861 --------------------------------------------------
862 simplExprF :: SimplEnv -> InExpr -> SimplCont
863 -> SimplM (SimplEnv, OutExpr)
865 simplExprF env e cont
866 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
867 simplExprF' env e cont
869 simplExprF' :: SimplEnv -> InExpr -> SimplCont
870 -> SimplM (SimplEnv, OutExpr)
871 simplExprF' env (Var v) cont = simplVarF env v cont
872 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
873 simplExprF' env (Note n expr) cont = simplNote env n expr cont
874 simplExprF' env (Cast body co) cont = simplCast env body co cont
875 simplExprF' env (App fun arg) cont = simplExprF env fun $
876 ApplyTo NoDup arg env cont
878 simplExprF' env expr@(Lam _ _) cont
879 = simplLam env (map zap bndrs) body cont
880 -- The main issue here is under-saturated lambdas
881 -- (\x1. \x2. e) arg1
882 -- Here x1 might have "occurs-once" occ-info, because occ-info
883 -- is computed assuming that a group of lambdas is applied
884 -- all at once. If there are too few args, we must zap the
887 n_args = countArgs cont
888 n_params = length bndrs
889 (bndrs, body) = collectBinders expr
890 zap | n_args >= n_params = \b -> b
891 | otherwise = \b -> if isTyCoVar b then b
893 -- NB: we count all the args incl type args
894 -- so we must count all the binders (incl type lambdas)
896 simplExprF' env (Type ty) cont
897 = ASSERT( contIsRhsOrArg cont )
898 do { ty' <- simplCoercion env ty
899 ; rebuild env (Type ty') cont }
901 simplExprF' env (Case scrut bndr _ alts) cont
902 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
903 = -- Simplify the scrutinee with a Select continuation
904 simplExprF env scrut (Select NoDup bndr alts env cont)
907 = -- If case-of-case is off, simply simplify the case expression
908 -- in a vanilla Stop context, and rebuild the result around it
909 do { case_expr' <- simplExprC env scrut
910 (Select NoDup bndr alts env mkBoringStop)
911 ; rebuild env case_expr' cont }
913 simplExprF' env (Let (Rec pairs) body) cont
914 = do { env' <- simplRecBndrs env (map fst pairs)
915 -- NB: bndrs' don't have unfoldings or rules
916 -- We add them as we go down
918 ; env'' <- simplRecBind env' NotTopLevel pairs
919 ; simplExprF env'' body cont }
921 simplExprF' env (Let (NonRec bndr rhs) body) cont
922 = simplNonRecE env bndr (rhs, env) ([], body) cont
924 ---------------------------------
925 simplType :: SimplEnv -> InType -> SimplM OutType
926 -- Kept monadic just so we can do the seqType
928 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
929 seqType new_ty `seq` return new_ty
931 new_ty = substTy env ty
933 ---------------------------------
934 simplCoercion :: SimplEnv -> InType -> SimplM OutType
935 -- The InType isn't *necessarily* a coercion, but it might be
936 -- (in a type application, say) and optCoercion is a no-op on types
938 = seqType new_co `seq` return new_co
940 new_co = optCoercion (getTvSubst env) co
944 %************************************************************************
946 \subsection{The main rebuilder}
948 %************************************************************************
951 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
952 -- At this point the substitution in the SimplEnv should be irrelevant
953 -- only the in-scope set and floats should matter
954 rebuild env expr cont0
955 = -- pprTrace "rebuild" (ppr expr $$ ppr cont0 $$ ppr (seFloats env)) $
957 Stop {} -> return (env, expr)
958 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
959 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
960 StrictArg info _ cont -> rebuildCall env (info `addArgTo` expr) cont
961 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
962 ; simplLam env' bs body cont }
963 ApplyTo dup_flag arg se cont -- See Note [Avoid redundant simplification]
964 | isSimplified dup_flag -> rebuild env (App expr arg) cont
965 | otherwise -> do { arg' <- simplExpr (se `setInScope` env) arg
966 ; rebuild env (App expr arg') cont }
970 %************************************************************************
974 %************************************************************************
977 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
978 -> SimplM (SimplEnv, OutExpr)
979 simplCast env body co0 cont0
980 = do { co1 <- simplCoercion env co0
981 ; simplExprF env body (addCoerce co1 cont0) }
983 addCoerce co cont = add_coerce co (coercionKind co) cont
985 add_coerce _co (s1, k1) cont -- co :: ty~ty
986 | s1 `coreEqType` k1 = cont -- is a no-op
988 add_coerce co1 (s1, _k2) (CoerceIt co2 cont)
989 | (_l1, t1) <- coercionKind co2
990 -- e |> (g1 :: S1~L) |> (g2 :: L~T1)
993 -- e |> (g1 . g2 :: S1~T1) otherwise
995 -- For example, in the initial form of a worker
996 -- we may find (coerce T (coerce S (\x.e))) y
997 -- and we'd like it to simplify to e[y/x] in one round
999 , s1 `coreEqType` t1 = cont -- The coerces cancel out
1000 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
1002 add_coerce co (s1s2, _t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
1003 -- (f |> g) ty ---> (f ty) |> (g @ ty)
1004 -- This implements the PushT and PushC rules from the paper
1005 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
1007 (new_arg_ty, new_cast)
1008 | isCoVar tyvar = (new_arg_co, mkCselRCoercion co) -- PushC rule
1009 | otherwise = (ty', mkInstCoercion co ty') -- PushT rule
1011 ApplyTo dup (Type new_arg_ty) (zapSubstEnv arg_se) (addCoerce new_cast cont)
1013 ty' = substTy (arg_se `setInScope` env) arg_ty
1014 new_arg_co = mkCsel1Coercion co `mkTransCoercion`
1015 ty' `mkTransCoercion`
1016 mkSymCoercion (mkCsel2Coercion co)
1018 add_coerce co (s1s2, _t1t2) (ApplyTo dup arg arg_se cont)
1019 | not (isTypeArg arg) -- This implements the Push rule from the paper
1020 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
1021 -- (e |> (g :: s1s2 ~ t1->t2)) f
1023 -- (e (f |> (arg g :: t1~s1))
1024 -- |> (res g :: s2->t2)
1026 -- t1t2 must be a function type, t1->t2, because it's applied
1027 -- to something but s1s2 might conceivably not be
1029 -- When we build the ApplyTo we can't mix the out-types
1030 -- with the InExpr in the argument, so we simply substitute
1031 -- to make it all consistent. It's a bit messy.
1032 -- But it isn't a common case.
1034 -- Example of use: Trac #995
1035 = ApplyTo dup new_arg (zapSubstEnv arg_se) (addCoerce co2 cont)
1037 -- we split coercion t1->t2 ~ s1->s2 into t1 ~ s1 and
1038 -- t2 ~ s2 with left and right on the curried form:
1039 -- (->) t1 t2 ~ (->) s1 s2
1040 [co1, co2] = decomposeCo 2 co
1041 new_arg = mkCoerce (mkSymCoercion co1) arg'
1042 arg' = substExpr (text "move-cast") (arg_se `setInScope` env) arg
1044 add_coerce co _ cont = CoerceIt co cont
1048 %************************************************************************
1050 \subsection{Lambdas}
1052 %************************************************************************
1055 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
1056 -> SimplM (SimplEnv, OutExpr)
1058 simplLam env [] body cont = simplExprF env body cont
1061 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
1062 = do { tick (BetaReduction bndr)
1063 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
1065 -- Not enough args, so there are real lambdas left to put in the result
1066 simplLam env bndrs body cont
1067 = do { (env', bndrs') <- simplLamBndrs env bndrs
1068 ; body' <- simplExpr env' body
1069 ; new_lam <- mkLam env' bndrs' body'
1070 ; rebuild env' new_lam cont }
1073 simplNonRecE :: SimplEnv
1074 -> InBndr -- The binder
1075 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
1076 -> ([InBndr], InExpr) -- Body of the let/lambda
1079 -> SimplM (SimplEnv, OutExpr)
1081 -- simplNonRecE is used for
1082 -- * non-top-level non-recursive lets in expressions
1085 -- It deals with strict bindings, via the StrictBind continuation,
1086 -- which may abort the whole process
1088 -- The "body" of the binding comes as a pair of ([InId],InExpr)
1089 -- representing a lambda; so we recurse back to simplLam
1090 -- Why? Because of the binder-occ-info-zapping done before
1091 -- the call to simplLam in simplExprF (Lam ...)
1093 -- First deal with type applications and type lets
1094 -- (/\a. e) (Type ty) and (let a = Type ty in e)
1095 simplNonRecE env bndr (Type ty_arg, rhs_se) (bndrs, body) cont
1096 = ASSERT( isTyCoVar bndr )
1097 do { ty_arg' <- simplType (rhs_se `setInScope` env) ty_arg
1098 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
1100 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
1101 | preInlineUnconditionally env NotTopLevel bndr rhs
1102 = do { tick (PreInlineUnconditionally bndr)
1103 ; -- pprTrace "preInlineUncond" (ppr bndr <+> ppr rhs) $
1104 simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
1107 = do { simplExprF (rhs_se `setFloats` env) rhs
1108 (StrictBind bndr bndrs body env cont) }
1111 = ASSERT( not (isTyCoVar bndr) )
1112 do { (env1, bndr1) <- simplNonRecBndr env bndr
1113 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
1114 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
1115 ; simplLam env3 bndrs body cont }
1119 %************************************************************************
1123 %************************************************************************
1126 -- Hack alert: we only distinguish subsumed cost centre stacks for the
1127 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
1128 simplNote :: SimplEnv -> Note -> CoreExpr -> SimplCont
1129 -> SimplM (SimplEnv, OutExpr)
1130 simplNote env (SCC cc) e cont
1131 | pushCCisNop cc (getEnclosingCC env) -- scc "f" (...(scc "f" e)...)
1132 = simplExprF env e cont -- ==> scc "f" (...e...)
1134 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
1135 ; rebuild env (mkSCC cc e') cont }
1137 simplNote env (CoreNote s) e cont
1138 = do { e' <- simplExpr env e
1139 ; rebuild env (Note (CoreNote s) e') cont }
1143 %************************************************************************
1147 %************************************************************************
1150 simplVar :: SimplEnv -> InVar -> SimplM OutExpr
1151 -- Look up an InVar in the environment
1154 = return (Type (substTyVar env var))
1156 = case substId env var of
1157 DoneId var1 -> return (Var var1)
1158 DoneEx e -> return e
1159 ContEx tvs ids e -> simplExpr (setSubstEnv env tvs ids) e
1161 simplVarF :: SimplEnv -> InId -> SimplCont -> SimplM (SimplEnv, OutExpr)
1162 simplVarF env var cont
1163 = case substId env var of
1164 DoneEx e -> simplExprF (zapSubstEnv env) e cont
1165 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
1166 DoneId var1 -> completeCall env var1 cont
1167 -- Note [zapSubstEnv]
1168 -- The template is already simplified, so don't re-substitute.
1169 -- This is VITAL. Consider
1171 -- let y = \z -> ...x... in
1173 -- We'll clone the inner \x, adding x->x' in the id_subst
1174 -- Then when we inline y, we must *not* replace x by x' in
1175 -- the inlined copy!!
1177 ---------------------------------------------------------
1178 -- Dealing with a call site
1180 completeCall :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1181 completeCall env var cont
1182 = do { ------------- Try inlining ----------------
1183 dflags <- getDOptsSmpl
1184 ; let (lone_variable, arg_infos, call_cont) = contArgs cont
1185 -- The args are OutExprs, obtained by *lazily* substituting
1186 -- in the args found in cont. These args are only examined
1187 -- to limited depth (unless a rule fires). But we must do
1188 -- the substitution; rule matching on un-simplified args would
1191 n_val_args = length arg_infos
1192 interesting_cont = interestingCallContext call_cont
1193 unfolding = activeUnfolding env var
1194 maybe_inline = callSiteInline dflags var unfolding
1195 lone_variable arg_infos interesting_cont
1196 ; case maybe_inline of {
1197 Just expr -- There is an inlining!
1198 -> do { tick (UnfoldingDone var)
1199 ; trace_inline dflags expr cont $
1200 simplExprF (zapSubstEnv env) expr cont }
1202 ; Nothing -> do -- No inlining!
1204 { rule_base <- getSimplRules
1205 ; let info = mkArgInfo var (getRules rule_base var) n_val_args call_cont
1206 ; rebuildCall env info cont
1209 trace_inline dflags unfolding cont stuff
1210 | not (dopt Opt_D_dump_inlinings dflags) = stuff
1211 | not (dopt Opt_D_verbose_core2core dflags)
1212 = if isExternalName (idName var) then
1213 pprTrace "Inlining done:" (ppr var) stuff
1216 = pprTrace ("Inlining done: " ++ showSDoc (ppr var))
1217 (vcat [text "Inlined fn: " <+> nest 2 (ppr unfolding),
1218 text "Cont: " <+> ppr cont])
1221 rebuildCall :: SimplEnv
1224 -> SimplM (SimplEnv, OutExpr)
1225 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_strs = [] }) cont
1226 -- When we run out of strictness args, it means
1227 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1228 -- Then we want to discard the entire strict continuation. E.g.
1229 -- * case (error "hello") of { ... }
1230 -- * (error "Hello") arg
1231 -- * f (error "Hello") where f is strict
1233 -- Then, especially in the first of these cases, we'd like to discard
1234 -- the continuation, leaving just the bottoming expression. But the
1235 -- type might not be right, so we may have to add a coerce.
1236 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1237 = return (env, mk_coerce res) -- contination to discard, else we do it
1238 where -- again and again!
1239 res = mkApps (Var fun) (reverse rev_args)
1240 res_ty = exprType res
1241 cont_ty = contResultType env res_ty cont
1242 co = mkUnsafeCoercion res_ty cont_ty
1243 mk_coerce expr | cont_ty `coreEqType` res_ty = expr
1244 | otherwise = mkCoerce co expr
1246 rebuildCall env info (ApplyTo _ (Type arg_ty) se cont)
1247 = do { ty' <- simplCoercion (se `setInScope` env) arg_ty
1248 ; rebuildCall env (info `addArgTo` Type ty') cont }
1250 rebuildCall env info@(ArgInfo { ai_encl = encl_rules
1251 , ai_strs = str:strs, ai_discs = disc:discs })
1252 (ApplyTo dup_flag arg arg_se cont)
1253 | isSimplified dup_flag -- See Note [Avoid redundant simplification]
1254 = rebuildCall env (addArgTo info' arg) cont
1256 | str -- Strict argument
1257 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1258 simplExprF (arg_se `setFloats` env) arg
1259 (StrictArg info' cci cont)
1262 | otherwise -- Lazy argument
1263 -- DO NOT float anything outside, hence simplExprC
1264 -- There is no benefit (unlike in a let-binding), and we'd
1265 -- have to be very careful about bogus strictness through
1266 -- floating a demanded let.
1267 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1269 ; rebuildCall env (addArgTo info' arg') cont }
1271 info' = info { ai_strs = strs, ai_discs = discs }
1272 cci | encl_rules || disc > 0 = ArgCtxt encl_rules -- Be keener here
1273 | otherwise = BoringCtxt -- Nothing interesting
1275 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_rules = rules }) cont
1276 = do { -- We've accumulated a simplified call in <fun,rev_args>
1277 -- so try rewrite rules; see Note [RULEs apply to simplified arguments]
1278 -- See also Note [Rules for recursive functions]
1279 ; let args = reverse rev_args
1280 env' = zapSubstEnv env
1281 ; mb_rule <- tryRules env rules fun args cont
1283 Just (n_args, rule_rhs) -> simplExprF env' rule_rhs $
1284 pushSimplifiedArgs env' (drop n_args args) cont ;
1285 -- n_args says how many args the rule consumed
1286 ; Nothing -> rebuild env (mkApps (Var fun) args) cont -- No rules
1290 Note [RULES apply to simplified arguments]
1291 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1292 It's very desirable to try RULES once the arguments have been simplified, because
1293 doing so ensures that rule cascades work in one pass. Consider
1294 {-# RULES g (h x) = k x
1297 Then we want to rewrite (g (h x)) to (k x) and only then try f's rules. If
1298 we match f's rules against the un-simplified RHS, it won't match. This
1299 makes a particularly big difference when superclass selectors are involved:
1300 op ($p1 ($p2 (df d)))
1301 We want all this to unravel in one sweeep.
1303 Note [Avoid redundant simplification]
1304 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1305 Because RULES apply to simplified arguments, there's a danger of repeatedly
1306 simplifying already-simplified arguments. An important example is that of
1308 Here e1, e2 are simplified before the rule is applied, but don't really
1309 participate in the rule firing. So we mark them as Simplified to avoid
1310 re-simplifying them.
1314 This part of the simplifier may break the no-shadowing invariant
1316 f (...(\a -> e)...) (case y of (a,b) -> e')
1317 where f is strict in its second arg
1318 If we simplify the innermost one first we get (...(\a -> e)...)
1319 Simplifying the second arg makes us float the case out, so we end up with
1320 case y of (a,b) -> f (...(\a -> e)...) e'
1321 So the output does not have the no-shadowing invariant. However, there is
1322 no danger of getting name-capture, because when the first arg was simplified
1323 we used an in-scope set that at least mentioned all the variables free in its
1324 static environment, and that is enough.
1326 We can't just do innermost first, or we'd end up with a dual problem:
1327 case x of (a,b) -> f e (...(\a -> e')...)
1329 I spent hours trying to recover the no-shadowing invariant, but I just could
1330 not think of an elegant way to do it. The simplifier is already knee-deep in
1331 continuations. We have to keep the right in-scope set around; AND we have
1332 to get the effect that finding (error "foo") in a strict arg position will
1333 discard the entire application and replace it with (error "foo"). Getting
1334 all this at once is TOO HARD!
1337 %************************************************************************
1341 %************************************************************************
1344 tryRules :: SimplEnv -> [CoreRule]
1345 -> Id -> [OutExpr] -> SimplCont
1346 -> SimplM (Maybe (Arity, CoreExpr)) -- The arity is the number of
1347 -- args consumed by the rule
1348 tryRules env rules fn args call_cont
1352 = do { dflags <- getDOptsSmpl
1353 ; case activeRule dflags env of {
1354 Nothing -> return Nothing ; -- No rules apply
1356 case lookupRule act_fn (activeUnfInRule env) (getInScope env) fn args rules of {
1357 Nothing -> return Nothing ; -- No rule matches
1358 Just (rule, rule_rhs) ->
1360 do { tick (RuleFired (ru_name rule))
1361 ; trace_dump dflags rule rule_rhs $
1362 return (Just (ruleArity rule, rule_rhs)) }}}}
1364 trace_dump dflags rule rule_rhs stuff
1365 | not (dopt Opt_D_dump_rule_firings dflags) = stuff
1366 | not (dopt Opt_D_verbose_core2core dflags)
1368 = pprTrace "Rule fired:" (ftext (ru_name rule)) stuff
1370 = pprTrace "Rule fired"
1371 (vcat [text "Rule:" <+> ftext (ru_name rule),
1372 text "Before:" <+> ppr fn <+> sep (map pprParendExpr args),
1373 text "After: " <+> pprCoreExpr rule_rhs,
1374 text "Cont: " <+> ppr call_cont])
1378 Note [Rules for recursive functions]
1379 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1380 You might think that we shouldn't apply rules for a loop breaker:
1381 doing so might give rise to an infinite loop, because a RULE is
1382 rather like an extra equation for the function:
1383 RULE: f (g x) y = x+y
1386 But it's too drastic to disable rules for loop breakers.
1387 Even the foldr/build rule would be disabled, because foldr
1388 is recursive, and hence a loop breaker:
1389 foldr k z (build g) = g k z
1390 So it's up to the programmer: rules can cause divergence
1393 %************************************************************************
1395 Rebuilding a cse expression
1397 %************************************************************************
1399 Note [Case elimination]
1400 ~~~~~~~~~~~~~~~~~~~~~~~
1401 The case-elimination transformation discards redundant case expressions.
1402 Start with a simple situation:
1404 case x# of ===> e[x#/y#]
1407 (when x#, y# are of primitive type, of course). We can't (in general)
1408 do this for algebraic cases, because we might turn bottom into
1411 The code in SimplUtils.prepareAlts has the effect of generalise this
1412 idea to look for a case where we're scrutinising a variable, and we
1413 know that only the default case can match. For example:
1417 DEFAULT -> ...(case x of
1421 Here the inner case is first trimmed to have only one alternative, the
1422 DEFAULT, after which it's an instance of the previous case. This
1423 really only shows up in eliminating error-checking code.
1425 We also make sure that we deal with this very common case:
1430 Here we are using the case as a strict let; if x is used only once
1431 then we want to inline it. We have to be careful that this doesn't
1432 make the program terminate when it would have diverged before, so we
1434 - e is already evaluated (it may so if e is a variable)
1435 - x is used strictly, or
1437 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1439 case e of ===> case e of DEFAULT -> r
1443 Now again the case may be elminated by the CaseElim transformation.
1446 Further notes about case elimination
1447 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1448 Consider: test :: Integer -> IO ()
1451 Turns out that this compiles to:
1454 eta1 :: State# RealWorld ->
1455 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1457 (PrelNum.jtos eta ($w[] @ Char))
1459 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1461 Notice the strange '<' which has no effect at all. This is a funny one.
1462 It started like this:
1464 f x y = if x < 0 then jtos x
1465 else if y==0 then "" else jtos x
1467 At a particular call site we have (f v 1). So we inline to get
1469 if v < 0 then jtos x
1470 else if 1==0 then "" else jtos x
1472 Now simplify the 1==0 conditional:
1474 if v<0 then jtos v else jtos v
1476 Now common-up the two branches of the case:
1478 case (v<0) of DEFAULT -> jtos v
1480 Why don't we drop the case? Because it's strict in v. It's technically
1481 wrong to drop even unnecessary evaluations, and in practice they
1482 may be a result of 'seq' so we *definitely* don't want to drop those.
1483 I don't really know how to improve this situation.
1486 ---------------------------------------------------------
1487 -- Eliminate the case if possible
1489 rebuildCase, reallyRebuildCase
1491 -> OutExpr -- Scrutinee
1492 -> InId -- Case binder
1493 -> [InAlt] -- Alternatives (inceasing order)
1495 -> SimplM (SimplEnv, OutExpr)
1497 --------------------------------------------------
1498 -- 1. Eliminate the case if there's a known constructor
1499 --------------------------------------------------
1501 rebuildCase env scrut case_bndr alts cont
1502 | Lit lit <- scrut -- No need for same treatment as constructors
1503 -- because literals are inlined more vigorously
1504 = do { tick (KnownBranch case_bndr)
1505 ; case findAlt (LitAlt lit) alts of
1506 Nothing -> missingAlt env case_bndr alts cont
1507 Just (_, bs, rhs) -> simple_rhs bs rhs }
1509 | Just (con, ty_args, other_args) <- exprIsConApp_maybe (activeUnfInRule env) scrut
1510 -- Works when the scrutinee is a variable with a known unfolding
1511 -- as well as when it's an explicit constructor application
1512 = do { tick (KnownBranch case_bndr)
1513 ; case findAlt (DataAlt con) alts of
1514 Nothing -> missingAlt env case_bndr alts cont
1515 Just (DEFAULT, bs, rhs) -> simple_rhs bs rhs
1516 Just (_, bs, rhs) -> knownCon env scrut con ty_args other_args
1517 case_bndr bs rhs cont
1520 simple_rhs bs rhs = ASSERT( null bs )
1521 do { env' <- simplNonRecX env case_bndr scrut
1522 ; simplExprF env' rhs cont }
1525 --------------------------------------------------
1526 -- 2. Eliminate the case if scrutinee is evaluated
1527 --------------------------------------------------
1529 rebuildCase env scrut case_bndr [(_, bndrs, rhs)] cont
1530 -- See if we can get rid of the case altogether
1531 -- See Note [Case elimination]
1532 -- mkCase made sure that if all the alternatives are equal,
1533 -- then there is now only one (DEFAULT) rhs
1534 | all isDeadBinder bndrs -- bndrs are [InId]
1536 -- Check that the scrutinee can be let-bound instead of case-bound
1537 , exprOkForSpeculation scrut
1538 -- OK not to evaluate it
1539 -- This includes things like (==# a# b#)::Bool
1540 -- so that we simplify
1541 -- case ==# a# b# of { True -> x; False -> x }
1544 -- This particular example shows up in default methods for
1545 -- comparision operations (e.g. in (>=) for Int.Int32)
1546 || exprIsHNF scrut -- It's already evaluated
1547 || var_demanded_later scrut -- It'll be demanded later
1549 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1550 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1551 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1552 -- its argument: case x of { y -> dataToTag# y }
1553 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1554 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1556 -- Also we don't want to discard 'seq's
1557 = do { tick (CaseElim case_bndr)
1558 ; env' <- simplNonRecX env case_bndr scrut
1559 ; simplExprF env' rhs cont }
1561 -- The case binder is going to be evaluated later,
1562 -- and the scrutinee is a simple variable
1563 var_demanded_later (Var v) = isStrictDmd (idDemandInfo case_bndr)
1564 && not (isTickBoxOp v)
1565 -- ugly hack; covering this case is what
1566 -- exprOkForSpeculation was intended for.
1567 var_demanded_later _ = False
1569 --------------------------------------------------
1570 -- 3. Try seq rules; see Note [User-defined RULES for seq] in MkId
1571 --------------------------------------------------
1573 rebuildCase env scrut case_bndr alts@[(_, bndrs, rhs)] cont
1574 | all isDeadBinder (case_bndr : bndrs) -- So this is just 'seq'
1575 = do { let rhs' = substExpr (text "rebuild-case") env rhs
1576 out_args = [Type (substTy env (idType case_bndr)),
1577 Type (exprType rhs'), scrut, rhs']
1578 -- Lazily evaluated, so we don't do most of this
1580 ; rule_base <- getSimplRules
1581 ; mb_rule <- tryRules env (getRules rule_base seqId) seqId out_args cont
1583 Just (n_args, res) -> simplExprF (zapSubstEnv env)
1584 (mkApps res (drop n_args out_args))
1586 Nothing -> reallyRebuildCase env scrut case_bndr alts cont }
1588 rebuildCase env scrut case_bndr alts cont
1589 = reallyRebuildCase env scrut case_bndr alts cont
1591 --------------------------------------------------
1592 -- 3. Catch-all case
1593 --------------------------------------------------
1595 reallyRebuildCase env scrut case_bndr alts cont
1596 = do { -- Prepare the continuation;
1597 -- The new subst_env is in place
1598 (env', dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1600 -- Simplify the alternatives
1601 ; (scrut', case_bndr', alts') <- simplAlts env' scrut case_bndr alts dup_cont
1603 -- Check for empty alternatives
1604 ; if null alts' then missingAlt env case_bndr alts cont
1606 { dflags <- getDOptsSmpl
1607 ; case_expr <- mkCase dflags scrut' case_bndr' alts'
1609 -- Notice that rebuild gets the in-scope set from env', not alt_env
1610 -- (which in any case is only build in simplAlts)
1611 -- The case binder *not* scope over the whole returned case-expression
1612 ; rebuild env' case_expr nodup_cont } }
1615 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1616 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1617 way, there's a chance that v will now only be used once, and hence
1620 Historical note: we use to do the "case binder swap" in the Simplifier
1621 so there were additional complications if the scrutinee was a variable.
1622 Now the binder-swap stuff is done in the occurrence analyer; see
1623 OccurAnal Note [Binder swap].
1627 If the case binder is not dead, then neither are the pattern bound
1629 case <any> of x { (a,b) ->
1630 case x of { (p,q) -> p } }
1631 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1632 The point is that we bring into the envt a binding
1634 after the outer case, and that makes (a,b) alive. At least we do unless
1635 the case binder is guaranteed dead.
1637 In practice, the scrutinee is almost always a variable, so we pretty
1638 much always zap the OccInfo of the binders. It doesn't matter much though.
1643 Consider case (v `cast` co) of x { I# y ->
1644 ... (case (v `cast` co) of {...}) ...
1645 We'd like to eliminate the inner case. We can get this neatly by
1646 arranging that inside the outer case we add the unfolding
1647 v |-> x `cast` (sym co)
1648 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1650 Note [Improving seq]
1653 type family F :: * -> *
1654 type instance F Int = Int
1656 ... case e of x { DEFAULT -> rhs } ...
1658 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1660 case e `cast` co of x'::Int
1661 I# x# -> let x = x' `cast` sym co
1664 so that 'rhs' can take advantage of the form of x'.
1666 Notice that Note [Case of cast] may then apply to the result.
1668 Nota Bene: We only do the [Improving seq] transformation if the
1669 case binder 'x' is actually used in the rhs; that is, if the case
1670 is *not* a *pure* seq.
1671 a) There is no point in adding the cast to a pure seq.
1672 b) There is a good reason not to: doing so would interfere
1673 with seq rules (Note [Built-in RULES for seq] in MkId).
1674 In particular, this [Improving seq] thing *adds* a cast
1675 while [Built-in RULES for seq] *removes* one, so they
1678 You might worry about
1679 case v of x { __DEFAULT ->
1680 ... case (v `cast` co) of y { I# -> ... }}
1681 This is a pure seq (since x is unused), so [Improving seq] won't happen.
1682 But it's ok: the simplifier will replace 'v' by 'x' in the rhs to get
1683 case v of x { __DEFAULT ->
1684 ... case (x `cast` co) of y { I# -> ... }}
1685 Now the outer case is not a pure seq, so [Improving seq] will happen,
1686 and then the inner case will disappear.
1688 The need for [Improving seq] showed up in Roman's experiments. Example:
1689 foo :: F Int -> Int -> Int
1690 foo t n = t `seq` bar n
1693 bar n = bar (n - case t of TI i -> i)
1694 Here we'd like to avoid repeated evaluating t inside the loop, by
1695 taking advantage of the `seq`.
1697 At one point I did transformation in LiberateCase, but it's more
1698 robust here. (Otherwise, there's a danger that we'll simply drop the
1699 'seq' altogether, before LiberateCase gets to see it.)
1702 simplAlts :: SimplEnv
1704 -> InId -- Case binder
1705 -> [InAlt] -- Non-empty
1707 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1708 -- Like simplExpr, this just returns the simplified alternatives;
1709 -- it does not return an environment
1711 simplAlts env scrut case_bndr alts cont'
1712 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seTvSubst env)) $
1713 do { let env0 = zapFloats env
1715 ; (env1, case_bndr1) <- simplBinder env0 case_bndr
1717 ; fam_envs <- getFamEnvs
1718 ; (alt_env', scrut', case_bndr') <- improveSeq fam_envs env1 scrut
1719 case_bndr case_bndr1 alts
1721 ; (imposs_deflt_cons, in_alts) <- prepareAlts scrut' case_bndr' alts
1723 ; alts' <- mapM (simplAlt alt_env' imposs_deflt_cons case_bndr' cont') in_alts
1724 ; return (scrut', case_bndr', alts') }
1727 ------------------------------------
1728 improveSeq :: (FamInstEnv, FamInstEnv) -> SimplEnv
1729 -> OutExpr -> InId -> OutId -> [InAlt]
1730 -> SimplM (SimplEnv, OutExpr, OutId)
1731 -- Note [Improving seq]
1732 improveSeq fam_envs env scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1733 | not (isDeadBinder case_bndr) -- Not a pure seq! See the Note!
1734 , Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1735 = do { case_bndr2 <- newId (fsLit "nt") ty2
1736 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1737 env2 = extendIdSubst env case_bndr rhs
1738 ; return (env2, scrut `Cast` co, case_bndr2) }
1740 improveSeq _ env scrut _ case_bndr1 _
1741 = return (env, scrut, case_bndr1)
1744 ------------------------------------
1745 simplAlt :: SimplEnv
1746 -> [AltCon] -- These constructors can't be present when
1747 -- matching the DEFAULT alternative
1748 -> OutId -- The case binder
1753 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1754 = ASSERT( null bndrs )
1755 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1756 -- Record the constructors that the case-binder *can't* be.
1757 ; rhs' <- simplExprC env' rhs cont'
1758 ; return (DEFAULT, [], rhs') }
1760 simplAlt env _ case_bndr' cont' (LitAlt lit, bndrs, rhs)
1761 = ASSERT( null bndrs )
1762 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1763 ; rhs' <- simplExprC env' rhs cont'
1764 ; return (LitAlt lit, [], rhs') }
1766 simplAlt env _ case_bndr' cont' (DataAlt con, vs, rhs)
1767 = do { -- Deal with the pattern-bound variables
1768 -- Mark the ones that are in ! positions in the
1769 -- data constructor as certainly-evaluated.
1770 -- NB: simplLamBinders preserves this eval info
1771 let vs_with_evals = add_evals (dataConRepStrictness con)
1772 ; (env', vs') <- simplLamBndrs env vs_with_evals
1774 -- Bind the case-binder to (con args)
1775 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1776 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1777 env'' = addBinderUnfolding env' case_bndr'
1778 (mkConApp con con_args)
1780 ; rhs' <- simplExprC env'' rhs cont'
1781 ; return (DataAlt con, vs', rhs') }
1783 -- add_evals records the evaluated-ness of the bound variables of
1784 -- a case pattern. This is *important*. Consider
1785 -- data T = T !Int !Int
1787 -- case x of { T a b -> T (a+1) b }
1789 -- We really must record that b is already evaluated so that we don't
1790 -- go and re-evaluate it when constructing the result.
1791 -- See Note [Data-con worker strictness] in MkId.lhs
1796 go (v:vs') strs | isTyCoVar v = v : go vs' strs
1797 go (v:vs') (str:strs)
1798 | isMarkedStrict str = evald_v : go vs' strs
1799 | otherwise = zapped_v : go vs' strs
1801 zapped_v = zap_occ_info v
1802 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1803 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr the_strs)
1805 -- See Note [zapOccInfo]
1806 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1808 -- to the envt; so vs are now very much alive
1809 -- Note [Aug06] I can't see why this actually matters, but it's neater
1810 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1811 -- ==> case e of t { (a,b) -> ...(a)... }
1812 -- Look, Ma, a is alive now.
1813 zap_occ_info = zapCasePatIdOcc case_bndr'
1815 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1816 addBinderUnfolding env bndr rhs
1817 = modifyInScope env (bndr `setIdUnfolding` mkSimpleUnfolding rhs)
1819 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1820 addBinderOtherCon env bndr cons
1821 = modifyInScope env (bndr `setIdUnfolding` mkOtherCon cons)
1823 zapCasePatIdOcc :: Id -> Id -> Id
1824 -- Consider case e of b { (a,b) -> ... }
1825 -- Then if we bind b to (a,b) in "...", and b is not dead,
1826 -- then we must zap the deadness info on a,b
1827 zapCasePatIdOcc case_bndr
1828 | isDeadBinder case_bndr = \ pat_id -> pat_id
1829 | otherwise = \ pat_id -> zapIdOccInfo pat_id
1833 %************************************************************************
1835 \subsection{Known constructor}
1837 %************************************************************************
1839 We are a bit careful with occurrence info. Here's an example
1841 (\x* -> case x of (a*, b) -> f a) (h v, e)
1843 where the * means "occurs once". This effectively becomes
1844 case (h v, e) of (a*, b) -> f a)
1846 let a* = h v; b = e in f a
1850 All this should happen in one sweep.
1853 knownCon :: SimplEnv
1854 -> OutExpr -- The scrutinee
1855 -> DataCon -> [OutType] -> [OutExpr] -- The scrutinee (in pieces)
1856 -> InId -> [InBndr] -> InExpr -- The alternative
1858 -> SimplM (SimplEnv, OutExpr)
1860 knownCon env scrut dc dc_ty_args dc_args bndr bs rhs cont
1861 = do { env' <- bind_args env bs dc_args
1862 ; env'' <- bind_case_bndr env'
1863 ; simplExprF env'' rhs cont }
1865 zap_occ = zapCasePatIdOcc bndr -- bndr is an InId
1868 bind_args env' [] _ = return env'
1870 bind_args env' (b:bs') (Type ty : args)
1871 = ASSERT( isTyCoVar b )
1872 bind_args (extendTvSubst env' b ty) bs' args
1874 bind_args env' (b:bs') (arg : args)
1876 do { let b' = zap_occ b
1877 -- Note that the binder might be "dead", because it doesn't
1878 -- occur in the RHS; and simplNonRecX may therefore discard
1879 -- it via postInlineUnconditionally.
1880 -- Nevertheless we must keep it if the case-binder is alive,
1881 -- because it may be used in the con_app. See Note [zapOccInfo]
1882 ; env'' <- simplNonRecX env' b' arg
1883 ; bind_args env'' bs' args }
1886 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr dc_args $$
1887 text "scrut:" <+> ppr scrut
1889 -- It's useful to bind bndr to scrut, rather than to a fresh
1890 -- binding x = Con arg1 .. argn
1891 -- because very often the scrut is a variable, so we avoid
1892 -- creating, and then subsequently eliminating, a let-binding
1893 -- BUT, if scrut is a not a variable, we must be careful
1894 -- about duplicating the arg redexes; in that case, make
1895 -- a new con-app from the args
1897 | isDeadBinder bndr = return env
1898 | exprIsTrivial scrut = return (extendIdSubst env bndr (DoneEx scrut))
1899 | otherwise = do { dc_args <- mapM (simplVar env) bs
1900 -- dc_ty_args are aready OutTypes,
1901 -- but bs are InBndrs
1902 ; let con_app = Var (dataConWorkId dc)
1903 `mkTyApps` dc_ty_args
1905 ; simplNonRecX env bndr con_app }
1908 missingAlt :: SimplEnv -> Id -> [InAlt] -> SimplCont -> SimplM (SimplEnv, OutExpr)
1909 -- This isn't strictly an error, although it is unusual.
1910 -- It's possible that the simplifer might "see" that
1911 -- an inner case has no accessible alternatives before
1912 -- it "sees" that the entire branch of an outer case is
1913 -- inaccessible. So we simply put an error case here instead.
1914 missingAlt env case_bndr alts cont
1915 = WARN( True, ptext (sLit "missingAlt") <+> ppr case_bndr )
1916 return (env, mkImpossibleExpr res_ty)
1918 res_ty = contResultType env (substTy env (coreAltsType alts)) cont
1922 %************************************************************************
1924 \subsection{Duplicating continuations}
1926 %************************************************************************
1929 prepareCaseCont :: SimplEnv
1930 -> [InAlt] -> SimplCont
1931 -> SimplM (SimplEnv, SimplCont,SimplCont)
1932 -- Return a duplicatable continuation, a non-duplicable part
1933 -- plus some extra bindings (that scope over the entire
1936 -- No need to make it duplicatable if there's only one alternative
1937 prepareCaseCont env [_] cont = return (env, cont, mkBoringStop)
1938 prepareCaseCont env _ cont = mkDupableCont env cont
1942 mkDupableCont :: SimplEnv -> SimplCont
1943 -> SimplM (SimplEnv, SimplCont, SimplCont)
1945 mkDupableCont env cont
1946 | contIsDupable cont
1947 = return (env, cont, mkBoringStop)
1949 mkDupableCont _ (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1951 mkDupableCont env (CoerceIt ty cont)
1952 = do { (env', dup, nodup) <- mkDupableCont env cont
1953 ; return (env', CoerceIt ty dup, nodup) }
1955 mkDupableCont env cont@(StrictBind {})
1956 = return (env, mkBoringStop, cont)
1957 -- See Note [Duplicating StrictBind]
1959 mkDupableCont env (StrictArg info cci cont)
1960 -- See Note [Duplicating StrictArg]
1961 = do { (env', dup, nodup) <- mkDupableCont env cont
1962 ; (env'', args') <- mapAccumLM (makeTrivial NotTopLevel) env' (ai_args info)
1963 ; return (env'', StrictArg (info { ai_args = args' }) cci dup, nodup) }
1965 mkDupableCont env (ApplyTo _ arg se cont)
1966 = -- e.g. [...hole...] (...arg...)
1968 -- let a = ...arg...
1969 -- in [...hole...] a
1970 do { (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1971 ; arg' <- simplExpr (se `setInScope` env') arg
1972 ; (env'', arg'') <- makeTrivial NotTopLevel env' arg'
1973 ; let app_cont = ApplyTo OkToDup arg'' (zapSubstEnv env'') dup_cont
1974 ; return (env'', app_cont, nodup_cont) }
1976 mkDupableCont env cont@(Select _ case_bndr [(_, bs, _rhs)] _ _)
1977 -- See Note [Single-alternative case]
1978 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1979 -- | not (isDeadBinder case_bndr)
1980 | all isDeadBinder bs -- InIds
1981 && not (isUnLiftedType (idType case_bndr))
1982 -- Note [Single-alternative-unlifted]
1983 = return (env, mkBoringStop, cont)
1985 mkDupableCont env (Select _ case_bndr alts se cont)
1986 = -- e.g. (case [...hole...] of { pi -> ei })
1988 -- let ji = \xij -> ei
1989 -- in case [...hole...] of { pi -> ji xij }
1990 do { tick (CaseOfCase case_bndr)
1991 ; (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1992 -- NB: call mkDupableCont here, *not* prepareCaseCont
1993 -- We must make a duplicable continuation, whereas prepareCaseCont
1994 -- doesn't when there is a single case branch
1996 ; let alt_env = se `setInScope` env'
1997 ; (alt_env', case_bndr') <- simplBinder alt_env case_bndr
1998 ; alts' <- mapM (simplAlt alt_env' [] case_bndr' dup_cont) alts
1999 -- Safe to say that there are no handled-cons for the DEFAULT case
2000 -- NB: simplBinder does not zap deadness occ-info, so
2001 -- a dead case_bndr' will still advertise its deadness
2002 -- This is really important because in
2003 -- case e of b { (# p,q #) -> ... }
2004 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
2005 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
2006 -- In the new alts we build, we have the new case binder, so it must retain
2008 -- NB: we don't use alt_env further; it has the substEnv for
2009 -- the alternatives, and we don't want that
2011 ; (env'', alts'') <- mkDupableAlts env' case_bndr' alts'
2012 ; return (env'', -- Note [Duplicated env]
2013 Select OkToDup case_bndr' alts'' (zapSubstEnv env'') mkBoringStop,
2017 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
2018 -> SimplM (SimplEnv, [InAlt])
2019 -- Absorbs the continuation into the new alternatives
2021 mkDupableAlts env case_bndr' the_alts
2024 go env0 [] = return (env0, [])
2026 = do { (env1, alt') <- mkDupableAlt env0 case_bndr' alt
2027 ; (env2, alts') <- go env1 alts
2028 ; return (env2, alt' : alts' ) }
2030 mkDupableAlt :: SimplEnv -> OutId -> (AltCon, [CoreBndr], CoreExpr)
2031 -> SimplM (SimplEnv, (AltCon, [CoreBndr], CoreExpr))
2032 mkDupableAlt env case_bndr (con, bndrs', rhs')
2033 | exprIsDupable rhs' -- Note [Small alternative rhs]
2034 = return (env, (con, bndrs', rhs'))
2036 = do { let rhs_ty' = exprType rhs'
2037 scrut_ty = idType case_bndr
2040 DEFAULT -> case_bndr
2041 DataAlt dc -> setIdUnfolding case_bndr unf
2043 -- See Note [Case binders and join points]
2044 unf = mkInlineUnfolding Nothing rhs
2045 rhs = mkConApp dc (map Type (tyConAppArgs scrut_ty)
2046 ++ varsToCoreExprs bndrs')
2048 LitAlt {} -> WARN( True, ptext (sLit "mkDupableAlt")
2049 <+> ppr case_bndr <+> ppr con )
2051 -- The case binder is alive but trivial, so why has
2052 -- it not been substituted away?
2054 used_bndrs' | isDeadBinder case_bndr = filter abstract_over bndrs'
2055 | otherwise = bndrs' ++ [case_bndr_w_unf]
2058 | isTyCoVar bndr = True -- Abstract over all type variables just in case
2059 | otherwise = not (isDeadBinder bndr)
2060 -- The deadness info on the new Ids is preserved by simplBinders
2062 ; (final_bndrs', final_args) -- Note [Join point abstraction]
2063 <- if (any isId used_bndrs')
2064 then return (used_bndrs', varsToCoreExprs used_bndrs')
2065 else do { rw_id <- newId (fsLit "w") realWorldStatePrimTy
2066 ; return ([rw_id], [Var realWorldPrimId]) }
2068 ; join_bndr <- newId (fsLit "$j") (mkPiTypes final_bndrs' rhs_ty')
2069 -- Note [Funky mkPiTypes]
2071 ; let -- We make the lambdas into one-shot-lambdas. The
2072 -- join point is sure to be applied at most once, and doing so
2073 -- prevents the body of the join point being floated out by
2074 -- the full laziness pass
2075 really_final_bndrs = map one_shot final_bndrs'
2076 one_shot v | isId v = setOneShotLambda v
2078 join_rhs = mkLams really_final_bndrs rhs'
2079 join_call = mkApps (Var join_bndr) final_args
2081 ; env' <- addPolyBind NotTopLevel env (NonRec join_bndr join_rhs)
2082 ; return (env', (con, bndrs', join_call)) }
2083 -- See Note [Duplicated env]
2086 Note [Case binders and join points]
2087 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2089 case (case .. ) of c {
2092 If we make a join point with c but not c# we get
2093 $j = \c -> ....c....
2095 But if later inlining scrutines the c, thus
2097 $j = \c -> ... case c of { I# y -> ... } ...
2099 we won't see that 'c' has already been scrutinised. This actually
2100 happens in the 'tabulate' function in wave4main, and makes a significant
2101 difference to allocation.
2103 An alternative plan is this:
2105 $j = \c# -> let c = I# c# in ...c....
2107 but that is bad if 'c' is *not* later scrutinised.
2109 So instead we do both: we pass 'c' and 'c#' , and record in c's inlining
2110 (an InlineRule) that it's really I# c#, thus
2112 $j = \c# -> \c[=I# c#] -> ...c....
2114 Absence analysis may later discard 'c'.
2116 NB: take great care when doing strictness analysis;
2117 see Note [Lamba-bound unfoldings] in DmdAnal.
2119 Also note that we can still end up passing stuff that isn't used. Before
2120 strictness analysis we have
2121 let $j x y c{=(x,y)} = (h c, ...)
2123 After strictness analysis we see that h is strict, we end up with
2124 let $j x y c{=(x,y)} = ($wh x y, ...)
2127 Note [Duplicated env]
2128 ~~~~~~~~~~~~~~~~~~~~~
2129 Some of the alternatives are simplified, but have not been turned into a join point
2130 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
2131 bind the join point, because it might to do PostInlineUnconditionally, and
2132 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
2133 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
2134 at worst delays the join-point inlining.
2136 Note [Small alternative rhs]
2137 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2138 It is worth checking for a small RHS because otherwise we
2139 get extra let bindings that may cause an extra iteration of the simplifier to
2140 inline back in place. Quite often the rhs is just a variable or constructor.
2141 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
2142 iterations because the version with the let bindings looked big, and so wasn't
2143 inlined, but after the join points had been inlined it looked smaller, and so
2146 NB: we have to check the size of rhs', not rhs.
2147 Duplicating a small InAlt might invalidate occurrence information
2148 However, if it *is* dupable, we return the *un* simplified alternative,
2149 because otherwise we'd need to pair it up with an empty subst-env....
2150 but we only have one env shared between all the alts.
2151 (Remember we must zap the subst-env before re-simplifying something).
2152 Rather than do this we simply agree to re-simplify the original (small) thing later.
2154 Note [Funky mkPiTypes]
2155 ~~~~~~~~~~~~~~~~~~~~~~
2156 Notice the funky mkPiTypes. If the contructor has existentials
2157 it's possible that the join point will be abstracted over
2158 type varaibles as well as term variables.
2159 Example: Suppose we have
2160 data T = forall t. C [t]
2162 case (case e of ...) of
2164 We get the join point
2165 let j :: forall t. [t] -> ...
2166 j = /\t \xs::[t] -> rhs
2168 case (case e of ...) of
2169 C t xs::[t] -> j t xs
2171 Note [Join point abstaction]
2172 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2173 If we try to lift a primitive-typed something out
2174 for let-binding-purposes, we will *caseify* it (!),
2175 with potentially-disastrous strictness results. So
2176 instead we turn it into a function: \v -> e
2177 where v::State# RealWorld#. The value passed to this function
2178 is realworld#, which generates (almost) no code.
2180 There's a slight infelicity here: we pass the overall
2181 case_bndr to all the join points if it's used in *any* RHS,
2182 because we don't know its usage in each RHS separately
2184 We used to say "&& isUnLiftedType rhs_ty'" here, but now
2185 we make the join point into a function whenever used_bndrs'
2186 is empty. This makes the join-point more CPR friendly.
2187 Consider: let j = if .. then I# 3 else I# 4
2188 in case .. of { A -> j; B -> j; C -> ... }
2190 Now CPR doesn't w/w j because it's a thunk, so
2191 that means that the enclosing function can't w/w either,
2192 which is a lose. Here's the example that happened in practice:
2193 kgmod :: Int -> Int -> Int
2194 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
2198 I have seen a case alternative like this:
2200 It's a bit silly to add the realWorld dummy arg in this case, making
2203 (the \v alone is enough to make CPR happy) but I think it's rare
2205 Note [Duplicating StrictArg]
2206 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2207 The original plan had (where E is a big argument)
2209 ==> let $j = \a -> f E a
2212 But this is terrible! Here's an example:
2213 && E (case x of { T -> F; F -> T })
2214 Now, && is strict so we end up simplifying the case with
2215 an ArgOf continuation. If we let-bind it, we get
2216 let $j = \v -> && E v
2217 in simplExpr (case x of { T -> F; F -> T })
2219 And after simplifying more we get
2220 let $j = \v -> && E v
2221 in case x of { T -> $j F; F -> $j T }
2222 Which is a Very Bad Thing
2224 What we do now is this
2228 Now if the thing in the hole is a case expression (which is when
2229 we'll call mkDupableCont), we'll push the function call into the
2230 branches, which is what we want. Now RULES for f may fire, and
2231 call-pattern specialisation. Here's an example from Trac #3116
2234 _ -> Chunk p fpc (o+1) (l-1) bs')
2235 If we can push the call for 'go' inside the case, we get
2236 call-pattern specialisation for 'go', which is *crucial* for
2239 Here is the (&&) example:
2240 && E (case x of { T -> F; F -> T })
2242 case x of { T -> && a F; F -> && a T }
2246 * Arguments to f *after* the strict one are handled by
2247 the ApplyTo case of mkDupableCont. Eg
2250 * We can only do the let-binding of E because the function
2251 part of a StrictArg continuation is an explicit syntax
2252 tree. In earlier versions we represented it as a function
2253 (CoreExpr -> CoreEpxr) which we couldn't take apart.
2255 Do *not* duplicate StrictBind and StritArg continuations. We gain
2256 nothing by propagating them into the expressions, and we do lose a
2259 The desire not to duplicate is the entire reason that
2260 mkDupableCont returns a pair of continuations.
2262 Note [Duplicating StrictBind]
2263 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2264 Unlike StrictArg, there doesn't seem anything to gain from
2265 duplicating a StrictBind continuation, so we don't.
2267 The desire not to duplicate is the entire reason that
2268 mkDupableCont returns a pair of continuations.
2271 Note [Single-alternative cases]
2272 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2273 This case is just like the ArgOf case. Here's an example:
2277 case (case x of I# x' ->
2279 True -> I# (negate# x')
2280 False -> I# x') of y {
2282 Because the (case x) has only one alternative, we'll transform to
2284 case (case x' <# 0# of
2285 True -> I# (negate# x')
2286 False -> I# x') of y {
2288 But now we do *NOT* want to make a join point etc, giving
2290 let $j = \y -> MkT y
2292 True -> $j (I# (negate# x'))
2294 In this case the $j will inline again, but suppose there was a big
2295 strict computation enclosing the orginal call to MkT. Then, it won't
2296 "see" the MkT any more, because it's big and won't get duplicated.
2297 And, what is worse, nothing was gained by the case-of-case transform.
2299 So, in circumstances like these, we don't want to build join points
2300 and push the outer case into the branches of the inner one. Instead,
2301 don't duplicate the continuation.
2303 When should we use this strategy? We should not use it on *every*
2304 single-alternative case:
2305 e.g. case (case ....) of (a,b) -> (# a,b #)
2306 Here we must push the outer case into the inner one!
2309 * Match [(DEFAULT,_,_)], but in the common case of Int,
2310 the alternative-filling-in code turned the outer case into
2311 case (...) of y { I# _ -> MkT y }
2313 * Match on single alternative plus (not (isDeadBinder case_bndr))
2314 Rationale: pushing the case inwards won't eliminate the construction.
2315 But there's a risk of
2316 case (...) of y { (a,b) -> let z=(a,b) in ... }
2317 Now y looks dead, but it'll come alive again. Still, this
2318 seems like the best option at the moment.
2320 * Match on single alternative plus (all (isDeadBinder bndrs))
2321 Rationale: this is essentially seq.
2323 * Match when the rhs is *not* duplicable, and hence would lead to a
2324 join point. This catches the disaster-case above. We can test
2325 the *un-simplified* rhs, which is fine. It might get bigger or
2326 smaller after simplification; if it gets smaller, this case might
2327 fire next time round. NB also that we must test contIsDupable
2328 case_cont *too, because case_cont might be big!
2330 HOWEVER: I found that this version doesn't work well, because
2331 we can get let x = case (...) of { small } in ...case x...
2332 When x is inlined into its full context, we find that it was a bad
2333 idea to have pushed the outer case inside the (...) case.
2335 Note [Single-alternative-unlifted]
2336 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2337 Here's another single-alternative where we really want to do case-of-case:
2345 case y_s6X of tpl_s7m {
2346 M1.Mk1 ipv_s70 -> ipv_s70;
2347 M1.Mk2 ipv_s72 -> ipv_s72;
2353 case x_s74 of tpl_s7n {
2354 M1.Mk1 ipv_s77 -> ipv_s77;
2355 M1.Mk2 ipv_s79 -> ipv_s79;
2359 { __DEFAULT -> ==# [wild1_s7b wild_s7c];
2363 So the outer case is doing *nothing at all*, other than serving as a
2364 join-point. In this case we really want to do case-of-case and decide
2365 whether to use a real join point or just duplicate the continuation.
2367 Hence: check whether the case binder's type is unlifted, because then
2368 the outer case is *not* a seq.