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 )
16 import FamInstEnv ( FamInstEnv )
18 import MkId ( mkImpossibleExpr, seqId )
21 import Name ( mkSystemVarName )
23 import FamInstEnv ( topNormaliseType )
24 import DataCon ( DataCon, dataConWorkId, dataConRepStrictness )
26 import NewDemand ( isStrictDmd, splitStrictSig )
27 import PprCore ( pprParendExpr, pprCoreExpr )
28 import CoreUnfold ( mkUnfolding, mkCoreUnfolding, mkInlineRule,
29 exprIsConApp_maybe, callSiteInline, CallCtxt(..) )
31 import qualified CoreSubst
32 import CoreArity ( exprArity )
33 import Rules ( lookupRule, getRules )
34 import BasicTypes ( isMarkedStrict, Arity )
35 import CostCentre ( currentCCS, pushCCisNop )
36 import TysPrim ( realWorldStatePrimTy )
37 import PrelInfo ( realWorldPrimId )
38 import BasicTypes ( TopLevelFlag(..), isTopLevel, RecFlag(..) )
39 import MonadUtils ( foldlM, mapAccumLM )
40 import Maybes ( orElse )
41 import Data.List ( mapAccumL )
47 The guts of the simplifier is in this module, but the driver loop for
48 the simplifier is in SimplCore.lhs.
51 -----------------------------------------
52 *** IMPORTANT NOTE ***
53 -----------------------------------------
54 The simplifier used to guarantee that the output had no shadowing, but
55 it does not do so any more. (Actually, it never did!) The reason is
56 documented with simplifyArgs.
59 -----------------------------------------
60 *** IMPORTANT NOTE ***
61 -----------------------------------------
62 Many parts of the simplifier return a bunch of "floats" as well as an
63 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
65 All "floats" are let-binds, not case-binds, but some non-rec lets may
66 be unlifted (with RHS ok-for-speculation).
70 -----------------------------------------
71 ORGANISATION OF FUNCTIONS
72 -----------------------------------------
74 - simplify all top-level binders
75 - for NonRec, call simplRecOrTopPair
76 - for Rec, call simplRecBind
79 ------------------------------
80 simplExpr (applied lambda) ==> simplNonRecBind
81 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
82 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
84 ------------------------------
85 simplRecBind [binders already simplfied]
86 - use simplRecOrTopPair on each pair in turn
88 simplRecOrTopPair [binder already simplified]
89 Used for: recursive bindings (top level and nested)
90 top-level non-recursive bindings
92 - check for PreInlineUnconditionally
96 Used for: non-top-level non-recursive bindings
97 beta reductions (which amount to the same thing)
98 Because it can deal with strict arts, it takes a
99 "thing-inside" and returns an expression
101 - check for PreInlineUnconditionally
102 - simplify binder, including its IdInfo
111 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
112 Used for: binding case-binder and constr args in a known-constructor case
113 - check for PreInLineUnconditionally
117 ------------------------------
118 simplLazyBind: [binder already simplified, RHS not]
119 Used for: recursive bindings (top level and nested)
120 top-level non-recursive bindings
121 non-top-level, but *lazy* non-recursive bindings
122 [must not be strict or unboxed]
123 Returns floats + an augmented environment, not an expression
124 - substituteIdInfo and add result to in-scope
125 [so that rules are available in rec rhs]
128 - float if exposes constructor or PAP
132 completeNonRecX: [binder and rhs both simplified]
133 - if the the thing needs case binding (unlifted and not ok-for-spec)
139 completeBind: [given a simplified RHS]
140 [used for both rec and non-rec bindings, top level and not]
141 - try PostInlineUnconditionally
142 - add unfolding [this is the only place we add an unfolding]
147 Right hand sides and arguments
148 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
149 In many ways we want to treat
150 (a) the right hand side of a let(rec), and
151 (b) a function argument
152 in the same way. But not always! In particular, we would
153 like to leave these arguments exactly as they are, so they
154 will match a RULE more easily.
159 It's harder to make the rule match if we ANF-ise the constructor,
160 or eta-expand the PAP:
162 f (let { a = g x; b = h x } in (a,b))
165 On the other hand if we see the let-defns
170 then we *do* want to ANF-ise and eta-expand, so that p and q
171 can be safely inlined.
173 Even floating lets out is a bit dubious. For let RHS's we float lets
174 out if that exposes a value, so that the value can be inlined more vigorously.
177 r = let x = e in (x,x)
179 Here, if we float the let out we'll expose a nice constructor. We did experiments
180 that showed this to be a generally good thing. But it was a bad thing to float
181 lets out unconditionally, because that meant they got allocated more often.
183 For function arguments, there's less reason to expose a constructor (it won't
184 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
185 So for the moment we don't float lets out of function arguments either.
190 For eta expansion, we want to catch things like
192 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
194 If the \x was on the RHS of a let, we'd eta expand to bring the two
195 lambdas together. And in general that's a good thing to do. Perhaps
196 we should eta expand wherever we find a (value) lambda? Then the eta
197 expansion at a let RHS can concentrate solely on the PAP case.
200 %************************************************************************
202 \subsection{Bindings}
204 %************************************************************************
207 simplTopBinds :: SimplEnv -> [InBind] -> SimplM SimplEnv
209 simplTopBinds env0 binds0
210 = do { -- Put all the top-level binders into scope at the start
211 -- so that if a transformation rule has unexpectedly brought
212 -- anything into scope, then we don't get a complaint about that.
213 -- It's rather as if the top-level binders were imported.
214 ; env1 <- simplRecBndrs env0 (bindersOfBinds binds0)
215 ; dflags <- getDOptsSmpl
216 ; let dump_flag = dopt Opt_D_dump_inlinings dflags ||
217 dopt Opt_D_dump_rule_firings dflags
218 ; env2 <- simpl_binds dump_flag env1 binds0
219 ; freeTick SimplifierDone
222 -- We need to track the zapped top-level binders, because
223 -- they should have their fragile IdInfo zapped (notably occurrence info)
224 -- That's why we run down binds and bndrs' simultaneously.
226 -- The dump-flag emits a trace for each top-level binding, which
227 -- helps to locate the tracing for inlining and rule firing
228 simpl_binds :: Bool -> SimplEnv -> [InBind] -> SimplM SimplEnv
229 simpl_binds _ env [] = return env
230 simpl_binds dump env (bind:binds) = do { env' <- trace_bind dump bind $
232 ; simpl_binds dump env' binds }
234 trace_bind True bind = pprTrace "SimplBind" (ppr (bindersOf bind))
235 trace_bind False _ = \x -> x
237 simpl_bind env (Rec pairs) = simplRecBind env TopLevel pairs
238 simpl_bind env (NonRec b r) = simplRecOrTopPair env' TopLevel b b' r
240 (env', b') = addBndrRules env b (lookupRecBndr env b)
244 %************************************************************************
246 \subsection{Lazy bindings}
248 %************************************************************************
250 simplRecBind is used for
251 * recursive bindings only
254 simplRecBind :: SimplEnv -> TopLevelFlag
257 simplRecBind env0 top_lvl pairs0
258 = do { let (env_with_info, triples) = mapAccumL add_rules env0 pairs0
259 ; env1 <- go (zapFloats env_with_info) triples
260 ; return (env0 `addRecFloats` env1) }
261 -- addFloats adds the floats from env1,
262 -- _and_ updates env0 with the in-scope set from env1
264 add_rules :: SimplEnv -> (InBndr,InExpr) -> (SimplEnv, (InBndr, OutBndr, InExpr))
265 -- Add the (substituted) rules to the binder
266 add_rules env (bndr, rhs) = (env', (bndr, bndr', rhs))
268 (env', bndr') = addBndrRules env bndr (lookupRecBndr env bndr)
270 go env [] = return env
272 go env ((old_bndr, new_bndr, rhs) : pairs)
273 = do { env' <- simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
277 simplOrTopPair is used for
278 * recursive bindings (whether top level or not)
279 * top-level non-recursive bindings
281 It assumes the binder has already been simplified, but not its IdInfo.
284 simplRecOrTopPair :: SimplEnv
286 -> InId -> OutBndr -> InExpr -- Binder and rhs
287 -> SimplM SimplEnv -- Returns an env that includes the binding
289 simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
290 | preInlineUnconditionally env top_lvl old_bndr rhs -- Check for unconditional inline
291 = do { tick (PreInlineUnconditionally old_bndr)
292 ; return (extendIdSubst env old_bndr (mkContEx env rhs)) }
295 = simplLazyBind env top_lvl Recursive old_bndr new_bndr rhs env
296 -- May not actually be recursive, but it doesn't matter
300 simplLazyBind is used for
301 * [simplRecOrTopPair] recursive bindings (whether top level or not)
302 * [simplRecOrTopPair] top-level non-recursive bindings
303 * [simplNonRecE] non-top-level *lazy* non-recursive bindings
306 1. It assumes that the binder is *already* simplified,
307 and is in scope, and its IdInfo too, except unfolding
309 2. It assumes that the binder type is lifted.
311 3. It does not check for pre-inline-unconditionallly;
312 that should have been done already.
315 simplLazyBind :: SimplEnv
316 -> TopLevelFlag -> RecFlag
317 -> InId -> OutId -- Binder, both pre-and post simpl
318 -- The OutId has IdInfo, except arity, unfolding
319 -> InExpr -> SimplEnv -- The RHS and its environment
322 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
323 = do { let rhs_env = rhs_se `setInScope` env
324 (tvs, body) = case collectTyBinders rhs of
325 (tvs, body) | not_lam body -> (tvs,body)
326 | otherwise -> ([], rhs)
327 not_lam (Lam _ _) = False
329 -- Do not do the "abstract tyyvar" thing if there's
330 -- a lambda inside, becuase it defeats eta-reduction
331 -- f = /\a. \x. g a x
334 ; (body_env, tvs') <- simplBinders rhs_env tvs
335 -- See Note [Floating and type abstraction] in SimplUtils
338 ; (body_env1, body1) <- simplExprF body_env body mkRhsStop
339 -- ANF-ise a constructor or PAP rhs
340 ; (body_env2, body2) <- prepareRhs body_env1 bndr1 body1
343 <- if not (doFloatFromRhs top_lvl is_rec False body2 body_env2)
344 then -- No floating, just wrap up!
345 do { rhs' <- mkLam env tvs' (wrapFloats body_env2 body2)
346 ; return (env, rhs') }
348 else if null tvs then -- Simple floating
349 do { tick LetFloatFromLet
350 ; return (addFloats env body_env2, body2) }
352 else -- Do type-abstraction first
353 do { tick LetFloatFromLet
354 ; (poly_binds, body3) <- abstractFloats tvs' body_env2 body2
355 ; rhs' <- mkLam env tvs' body3
356 ; env' <- foldlM (addPolyBind top_lvl) env poly_binds
357 ; return (env', rhs') }
359 ; completeBind env' top_lvl bndr bndr1 rhs' }
362 A specialised variant of simplNonRec used when the RHS is already simplified,
363 notably in knownCon. It uses case-binding where necessary.
366 simplNonRecX :: SimplEnv
367 -> InId -- Old binder
368 -> OutExpr -- Simplified RHS
371 simplNonRecX env bndr new_rhs
372 | isDeadBinder bndr -- Not uncommon; e.g. case (a,b) of b { (p,q) -> p }
373 = return env -- Here b is dead, and we avoid creating
374 | otherwise -- the binding b = (a,b)
375 = do { (env', bndr') <- simplBinder env bndr
376 ; completeNonRecX env' (isStrictId bndr) bndr bndr' new_rhs }
378 completeNonRecX :: SimplEnv
380 -> InId -- Old binder
381 -> OutId -- New binder
382 -> OutExpr -- Simplified RHS
385 completeNonRecX env is_strict old_bndr new_bndr new_rhs
386 = do { (env1, rhs1) <- prepareRhs (zapFloats env) new_bndr new_rhs
388 if doFloatFromRhs NotTopLevel NonRecursive is_strict rhs1 env1
389 then do { tick LetFloatFromLet
390 ; return (addFloats env env1, rhs1) } -- Add the floats to the main env
391 else return (env, wrapFloats env1 rhs1) -- Wrap the floats around the RHS
392 ; completeBind env2 NotTopLevel old_bndr new_bndr rhs2 }
395 {- No, no, no! Do not try preInlineUnconditionally in completeNonRecX
396 Doing so risks exponential behaviour, because new_rhs has been simplified once already
397 In the cases described by the folowing commment, postInlineUnconditionally will
398 catch many of the relevant cases.
399 -- This happens; for example, the case_bndr during case of
400 -- known constructor: case (a,b) of x { (p,q) -> ... }
401 -- Here x isn't mentioned in the RHS, so we don't want to
402 -- create the (dead) let-binding let x = (a,b) in ...
404 -- Similarly, single occurrences can be inlined vigourously
405 -- e.g. case (f x, g y) of (a,b) -> ....
406 -- If a,b occur once we can avoid constructing the let binding for them.
408 Furthermore in the case-binding case preInlineUnconditionally risks extra thunks
409 -- Consider case I# (quotInt# x y) of
410 -- I# v -> let w = J# v in ...
411 -- If we gaily inline (quotInt# x y) for v, we end up building an
413 -- let w = J# (quotInt# x y) in ...
414 -- because quotInt# can fail.
416 | preInlineUnconditionally env NotTopLevel bndr new_rhs
417 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
420 ----------------------------------
421 prepareRhs takes a putative RHS, checks whether it's a PAP or
422 constructor application and, if so, converts it to ANF, so that the
423 resulting thing can be inlined more easily. Thus
430 We also want to deal well cases like this
431 v = (f e1 `cast` co) e2
432 Here we want to make e1,e2 trivial and get
433 x1 = e1; x2 = e2; v = (f x1 `cast` co) v2
434 That's what the 'go' loop in prepareRhs does
437 prepareRhs :: SimplEnv -> OutId -> OutExpr -> SimplM (SimplEnv, OutExpr)
438 -- Adds new floats to the env iff that allows us to return a good RHS
439 prepareRhs env id (Cast rhs co) -- Note [Float coercions]
440 | (ty1, _ty2) <- coercionKind co -- Do *not* do this if rhs has an unlifted type
441 , not (isUnLiftedType ty1) -- see Note [Float coercions (unlifted)]
442 = do { (env', rhs') <- makeTrivialWithInfo env sanitised_info rhs
443 ; return (env', Cast rhs' co) }
445 sanitised_info = vanillaIdInfo `setNewStrictnessInfo` newStrictnessInfo info
446 `setNewDemandInfo` newDemandInfo info
449 prepareRhs env0 _ rhs0
450 = do { (_is_val, env1, rhs1) <- go 0 env0 rhs0
451 ; return (env1, rhs1) }
453 go n_val_args env (Cast rhs co)
454 = do { (is_val, env', rhs') <- go n_val_args env rhs
455 ; return (is_val, env', Cast rhs' co) }
456 go n_val_args env (App fun (Type ty))
457 = do { (is_val, env', rhs') <- go n_val_args env fun
458 ; return (is_val, env', App rhs' (Type ty)) }
459 go n_val_args env (App fun arg)
460 = do { (is_val, env', fun') <- go (n_val_args+1) env fun
462 True -> do { (env'', arg') <- makeTrivial env' arg
463 ; return (True, env'', App fun' arg') }
464 False -> return (False, env, App fun arg) }
465 go n_val_args env (Var fun)
466 = return (is_val, env, Var fun)
468 is_val = n_val_args > 0 -- There is at least one arg
469 -- ...and the fun a constructor or PAP
470 && (isConLikeId fun || n_val_args < idArity fun)
471 -- See Note [CONLIKE pragma] in BasicTypes
473 = return (False, env, other)
477 Note [Float coercions]
478 ~~~~~~~~~~~~~~~~~~~~~~
479 When we find the binding
481 we'd like to transform it to
483 x = x `cast` co -- A trivial binding
484 There's a chance that e will be a constructor application or function, or something
485 like that, so moving the coerion to the usage site may well cancel the coersions
486 and lead to further optimisation. Example:
489 data instance T Int = T Int
491 foo :: Int -> Int -> Int
496 go n = case x of { T m -> go (n-m) }
497 -- This case should optimise
499 Note [Preserve strictness when floating coercions]
500 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
501 In the Note [Float coercions] transformation, keep the strictness info.
503 f = e `cast` co -- f has strictness SSL
505 f' = e -- f' also has strictness SSL
506 f = f' `cast` co -- f still has strictness SSL
508 Its not wrong to drop it on the floor, but better to keep it.
510 Note [Float coercions (unlifted)]
511 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
512 BUT don't do [Float coercions] if 'e' has an unlifted type.
515 foo :: Int = (error (# Int,Int #) "urk")
516 `cast` CoUnsafe (# Int,Int #) Int
518 If do the makeTrivial thing to the error call, we'll get
519 foo = case error (# Int,Int #) "urk" of v -> v `cast` ...
520 But 'v' isn't in scope!
522 These strange casts can happen as a result of case-of-case
523 bar = case (case x of { T -> (# 2,3 #); F -> error "urk" }) of
528 makeTrivial :: SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
529 -- Binds the expression to a variable, if it's not trivial, returning the variable
530 makeTrivial env expr = makeTrivialWithInfo env vanillaIdInfo expr
532 makeTrivialWithInfo :: SimplEnv -> IdInfo -> OutExpr -> SimplM (SimplEnv, OutExpr)
533 -- Propagate strictness and demand info to the new binder
534 -- Note [Preserve strictness when floating coercions]
535 makeTrivialWithInfo env info expr
538 | otherwise -- See Note [Take care] below
539 = do { uniq <- getUniqueM
540 ; let name = mkSystemVarName uniq (fsLit "a")
541 var = mkLocalIdWithInfo name (exprType expr) info
542 ; env' <- completeNonRecX env False var var expr
543 ; return (env', substExpr env' (Var var)) }
544 -- The substitution is needed becase we're constructing a new binding
546 -- And if rhs is of form (rhs1 |> co), then we might get
549 -- and now a's RHS is trivial and can be substituted out, and that
550 -- is what completeNonRecX will do
554 %************************************************************************
556 \subsection{Completing a lazy binding}
558 %************************************************************************
561 * deals only with Ids, not TyVars
562 * takes an already-simplified binder and RHS
563 * is used for both recursive and non-recursive bindings
564 * is used for both top-level and non-top-level bindings
566 It does the following:
567 - tries discarding a dead binding
568 - tries PostInlineUnconditionally
569 - add unfolding [this is the only place we add an unfolding]
572 It does *not* attempt to do let-to-case. Why? Because it is used for
573 - top-level bindings (when let-to-case is impossible)
574 - many situations where the "rhs" is known to be a WHNF
575 (so let-to-case is inappropriate).
577 Nor does it do the atomic-argument thing
580 completeBind :: SimplEnv
581 -> TopLevelFlag -- Flag stuck into unfolding
582 -> InId -- Old binder
583 -> OutId -> OutExpr -- New binder and RHS
585 -- completeBind may choose to do its work
586 -- * by extending the substitution (e.g. let x = y in ...)
587 -- * or by adding to the floats in the envt
589 completeBind env top_lvl old_bndr new_bndr new_rhs
590 = do { let old_info = idInfo old_bndr
591 old_unf = unfoldingInfo old_info
592 occ_info = occInfo old_info
594 ; new_unfolding <- simplUnfolding env top_lvl old_bndr occ_info new_rhs old_unf
596 ; if postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs new_unfolding
597 -- Inline and discard the binding
598 then do { tick (PostInlineUnconditionally old_bndr)
599 ; return (extendIdSubst env old_bndr (DoneEx new_rhs)) }
600 -- Use the substitution to make quite, quite sure that the
601 -- substitution will happen, since we are going to discard the binding
603 else return (addNonRecWithUnf env new_bndr new_rhs new_unfolding) }
605 ------------------------------
606 addPolyBind :: TopLevelFlag -> SimplEnv -> OutBind -> SimplM SimplEnv
607 -- Add a new binding to the environment, complete with its unfolding
608 -- but *do not* do postInlineUnconditionally, because we have already
609 -- processed some of the scope of the binding
610 -- We still want the unfolding though. Consider
612 -- x = /\a. let y = ... in Just y
614 -- Then we float the y-binding out (via abstractFloats and addPolyBind)
615 -- but 'x' may well then be inlined in 'body' in which case we'd like the
616 -- opportunity to inline 'y' too.
618 addPolyBind top_lvl env (NonRec poly_id rhs)
619 = do { unfolding <- simplUnfolding env top_lvl poly_id NoOccInfo rhs noUnfolding
620 -- Assumes that poly_id did not have an INLINE prag
621 -- which is perhaps wrong. ToDo: think about this
622 ; return (addNonRecWithUnf env poly_id rhs unfolding) }
624 addPolyBind _ env bind@(Rec _) = return (extendFloats env bind)
625 -- Hack: letrecs are more awkward, so we extend "by steam"
626 -- without adding unfoldings etc. At worst this leads to
627 -- more simplifier iterations
629 ------------------------------
630 addNonRecWithUnf :: SimplEnv
631 -> OutId -> OutExpr -- New binder and RHS
632 -> Unfolding -- New unfolding
634 addNonRecWithUnf env new_bndr new_rhs new_unfolding
635 = let new_arity = exprArity new_rhs
636 old_arity = idArity new_bndr
637 info1 = idInfo new_bndr `setArityInfo` new_arity
639 -- Unfolding info: Note [Setting the new unfolding]
640 info2 = info1 `setUnfoldingInfo` new_unfolding
642 -- Demand info: Note [Setting the demand info]
643 info3 | isEvaldUnfolding new_unfolding = zapDemandInfo info2 `orElse` info2
646 final_id = new_bndr `setIdInfo` info3
647 dmd_arity = length $ fst $ splitStrictSig $ idNewStrictness new_bndr
649 ASSERT( isId new_bndr )
650 WARN( new_arity < old_arity || new_arity < dmd_arity,
651 (ptext (sLit "Arity decrease:") <+> ppr final_id <+> ppr old_arity
652 <+> ppr new_arity <+> ppr dmd_arity) )
653 -- Note [Arity decrease]
655 final_id `seq` -- This seq forces the Id, and hence its IdInfo,
656 -- and hence any inner substitutions
657 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
658 addNonRec env final_id new_rhs
659 -- The addNonRec adds it to the in-scope set too
661 ------------------------------
662 simplUnfolding :: SimplEnv-> TopLevelFlag
663 -> Id -- Debug output only
664 -> OccInfo -> OutExpr
665 -> Unfolding -> SimplM Unfolding
666 -- Note [Setting the new unfolding]
667 simplUnfolding env _ _ _ _ (DFunUnfolding con ops)
668 = return (DFunUnfolding con ops')
670 ops' = map (CoreSubst.substExpr (mkCoreSubst env)) ops
672 simplUnfolding env top_lvl _ _ _
673 (CoreUnfolding { uf_tmpl = expr, uf_arity = arity
674 , uf_guidance = guide@(InlineRule {}) })
675 = do { expr' <- simplExpr (updMode updModeForInlineRules env) expr
676 -- See Note [Simplifying gently inside InlineRules] in SimplUtils
677 ; let mb_wkr' = CoreSubst.substInlineRuleInfo (mkCoreSubst env) (ir_info guide)
678 ; return (mkCoreUnfolding (isTopLevel top_lvl) expr' arity
679 (guide { ir_info = mb_wkr' })) }
680 -- See Note [Top-level flag on inline rules] in CoreUnfold
682 simplUnfolding _ top_lvl _ _occ_info new_rhs _
683 = return (mkUnfolding (isTopLevel top_lvl) new_rhs)
684 -- We make an unfolding *even for loop-breakers*.
685 -- Reason: (a) It might be useful to know that they are WHNF
686 -- (b) In TidyPgm we currently assume that, if we want to
687 -- expose the unfolding then indeed we *have* an unfolding
688 -- to expose. (We could instead use the RHS, but currently
689 -- we don't.) The simple thing is always to have one.
692 Note [Arity decrease]
693 ~~~~~~~~~~~~~~~~~~~~~
694 Generally speaking the arity of a binding should not decrease. But it *can*
695 legitimately happen becuase of RULES. Eg
697 where g has arity 2, will have arity 2. But if there's a rewrite rule
699 where h has arity 1, then f's arity will decrease. Here's a real-life example,
700 which is in the output of Specialise:
703 $dm {Arity 2} = \d.\x. op d
704 {-# RULES forall d. $dm Int d = $s$dm #-}
706 dInt = MkD .... opInt ...
707 opInt {Arity 1} = $dm dInt
709 $s$dm {Arity 0} = \x. op dInt }
711 Here opInt has arity 1; but when we apply the rule its arity drops to 0.
712 That's why Specialise goes to a little trouble to pin the right arity
713 on specialised functions too.
715 Note [Setting the new unfolding]
716 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
717 * If there's an INLINE pragma, we simplify the RHS gently. Maybe we
718 should do nothing at all, but simplifying gently might get rid of
721 * If not, we make an unfolding from the new RHS. But *only* for
722 non-loop-breakers. Making loop breakers not have an unfolding at all
723 means that we can avoid tests in exprIsConApp, for example. This is
724 important: if exprIsConApp says 'yes' for a recursive thing, then we
725 can get into an infinite loop
727 If there's an InlineRule on a loop breaker, we hang on to the inlining.
728 It's pretty dodgy, but the user did say 'INLINE'. May need to revisit
731 Note [Setting the demand info]
732 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
733 If the unfolding is a value, the demand info may
734 go pear-shaped, so we nuke it. Example:
736 case x of (p,q) -> h p q x
737 Here x is certainly demanded. But after we've nuked
738 the case, we'll get just
739 let x = (a,b) in h a b x
740 and now x is not demanded (I'm assuming h is lazy)
741 This really happens. Similarly
742 let f = \x -> e in ...f..f...
743 After inlining f at some of its call sites the original binding may
744 (for example) be no longer strictly demanded.
745 The solution here is a bit ad hoc...
748 %************************************************************************
750 \subsection[Simplify-simplExpr]{The main function: simplExpr}
752 %************************************************************************
754 The reason for this OutExprStuff stuff is that we want to float *after*
755 simplifying a RHS, not before. If we do so naively we get quadratic
756 behaviour as things float out.
758 To see why it's important to do it after, consider this (real) example:
772 a -- Can't inline a this round, cos it appears twice
776 Each of the ==> steps is a round of simplification. We'd save a
777 whole round if we float first. This can cascade. Consider
782 let f = let d1 = ..d.. in \y -> e
786 in \x -> ...(\y ->e)...
788 Only in this second round can the \y be applied, and it
789 might do the same again.
793 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
794 simplExpr env expr = simplExprC env expr mkBoringStop
796 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
797 -- Simplify an expression, given a continuation
798 simplExprC env expr cont
799 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
800 do { (env', expr') <- simplExprF (zapFloats env) expr cont
801 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
802 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
803 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
804 return (wrapFloats env' expr') }
806 --------------------------------------------------
807 simplExprF :: SimplEnv -> InExpr -> SimplCont
808 -> SimplM (SimplEnv, OutExpr)
810 simplExprF env e cont
811 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
812 simplExprF' env e cont
814 simplExprF' :: SimplEnv -> InExpr -> SimplCont
815 -> SimplM (SimplEnv, OutExpr)
816 simplExprF' env (Var v) cont = simplVar env v cont
817 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
818 simplExprF' env (Note n expr) cont = simplNote env n expr cont
819 simplExprF' env (Cast body co) cont = simplCast env body co cont
820 simplExprF' env (App fun arg) cont = simplExprF env fun $
821 ApplyTo NoDup arg env cont
823 simplExprF' env expr@(Lam _ _) cont
824 = simplLam env (map zap bndrs) body cont
825 -- The main issue here is under-saturated lambdas
826 -- (\x1. \x2. e) arg1
827 -- Here x1 might have "occurs-once" occ-info, because occ-info
828 -- is computed assuming that a group of lambdas is applied
829 -- all at once. If there are too few args, we must zap the
832 n_args = countArgs cont
833 n_params = length bndrs
834 (bndrs, body) = collectBinders expr
835 zap | n_args >= n_params = \b -> b
836 | otherwise = \b -> if isTyVar b then b
838 -- NB: we count all the args incl type args
839 -- so we must count all the binders (incl type lambdas)
841 simplExprF' env (Type ty) cont
842 = ASSERT( contIsRhsOrArg cont )
843 do { ty' <- simplCoercion env ty
844 ; rebuild env (Type ty') cont }
846 simplExprF' env (Case scrut bndr _ alts) cont
847 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
848 = -- Simplify the scrutinee with a Select continuation
849 simplExprF env scrut (Select NoDup bndr alts env cont)
852 = -- If case-of-case is off, simply simplify the case expression
853 -- in a vanilla Stop context, and rebuild the result around it
854 do { case_expr' <- simplExprC env scrut case_cont
855 ; rebuild env case_expr' cont }
857 case_cont = Select NoDup bndr alts env mkBoringStop
859 simplExprF' env (Let (Rec pairs) body) cont
860 = do { env' <- simplRecBndrs env (map fst pairs)
861 -- NB: bndrs' don't have unfoldings or rules
862 -- We add them as we go down
864 ; env'' <- simplRecBind env' NotTopLevel pairs
865 ; simplExprF env'' body cont }
867 simplExprF' env (Let (NonRec bndr rhs) body) cont
868 = simplNonRecE env bndr (rhs, env) ([], body) cont
870 ---------------------------------
871 simplType :: SimplEnv -> InType -> SimplM OutType
872 -- Kept monadic just so we can do the seqType
874 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
875 seqType new_ty `seq` return new_ty
877 new_ty = substTy env ty
879 ---------------------------------
880 simplCoercion :: SimplEnv -> InType -> SimplM OutType
881 -- The InType isn't *necessarily* a coercion, but it might be
882 -- (in a type application, say) and optCoercion is a no-op on types
884 = do { co' <- simplType env co
885 ; return (optCoercion co') }
889 %************************************************************************
891 \subsection{The main rebuilder}
893 %************************************************************************
896 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
897 -- At this point the substitution in the SimplEnv should be irrelevant
898 -- only the in-scope set and floats should matter
899 rebuild env expr cont0
900 = -- pprTrace "rebuild" (ppr expr $$ ppr cont0 $$ ppr (seFloats env)) $
902 Stop {} -> return (env, expr)
903 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
904 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
905 StrictArg info _ cont -> rebuildCall env (info `addArgTo` expr) cont
906 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
907 ; simplLam env' bs body cont }
908 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
909 ; rebuild env (App expr arg') cont }
913 %************************************************************************
917 %************************************************************************
920 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
921 -> SimplM (SimplEnv, OutExpr)
922 simplCast env body co0 cont0
923 = do { co1 <- simplCoercion env co0
924 ; simplExprF env body (addCoerce co1 cont0) }
926 addCoerce co cont = add_coerce co (coercionKind co) cont
928 add_coerce _co (s1, k1) cont -- co :: ty~ty
929 | s1 `coreEqType` k1 = cont -- is a no-op
931 add_coerce co1 (s1, _k2) (CoerceIt co2 cont)
932 | (_l1, t1) <- coercionKind co2
933 -- e |> (g1 :: S1~L) |> (g2 :: L~T1)
936 -- e |> (g1 . g2 :: S1~T1) otherwise
938 -- For example, in the initial form of a worker
939 -- we may find (coerce T (coerce S (\x.e))) y
940 -- and we'd like it to simplify to e[y/x] in one round
942 , s1 `coreEqType` t1 = cont -- The coerces cancel out
943 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
945 add_coerce co (s1s2, _t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
946 -- (f |> g) ty ---> (f ty) |> (g @ ty)
947 -- This implements the PushT and PushC rules from the paper
948 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
950 (new_arg_ty, new_cast)
951 | isCoVar tyvar = (new_arg_co, mkCselRCoercion co) -- PushC rule
952 | otherwise = (ty', mkInstCoercion co ty') -- PushT rule
954 ApplyTo dup (Type new_arg_ty) (zapSubstEnv arg_se) (addCoerce new_cast cont)
956 ty' = substTy (arg_se `setInScope` env) arg_ty
957 new_arg_co = mkCsel1Coercion co `mkTransCoercion`
958 ty' `mkTransCoercion`
959 mkSymCoercion (mkCsel2Coercion co)
961 add_coerce co (s1s2, _t1t2) (ApplyTo dup arg arg_se cont)
962 | not (isTypeArg arg) -- This implements the Push rule from the paper
963 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
964 -- (e |> (g :: s1s2 ~ t1->t2)) f
966 -- (e (f |> (arg g :: t1~s1))
967 -- |> (res g :: s2->t2)
969 -- t1t2 must be a function type, t1->t2, because it's applied
970 -- to something but s1s2 might conceivably not be
972 -- When we build the ApplyTo we can't mix the out-types
973 -- with the InExpr in the argument, so we simply substitute
974 -- to make it all consistent. It's a bit messy.
975 -- But it isn't a common case.
977 -- Example of use: Trac #995
978 = ApplyTo dup new_arg (zapSubstEnv arg_se) (addCoerce co2 cont)
980 -- we split coercion t1->t2 ~ s1->s2 into t1 ~ s1 and
981 -- t2 ~ s2 with left and right on the curried form:
982 -- (->) t1 t2 ~ (->) s1 s2
983 [co1, co2] = decomposeCo 2 co
984 new_arg = mkCoerce (mkSymCoercion co1) arg'
985 arg' = substExpr (arg_se `setInScope` env) arg
987 add_coerce co _ cont = CoerceIt co cont
991 %************************************************************************
995 %************************************************************************
998 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
999 -> SimplM (SimplEnv, OutExpr)
1001 simplLam env [] body cont = simplExprF env body cont
1004 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
1005 = do { tick (BetaReduction bndr)
1006 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
1008 -- Not enough args, so there are real lambdas left to put in the result
1009 simplLam env bndrs body cont
1010 = do { (env', bndrs') <- simplLamBndrs env bndrs
1011 ; body' <- simplExpr env' body
1012 ; new_lam <- mkLam env' bndrs' body'
1013 ; rebuild env' new_lam cont }
1016 simplNonRecE :: SimplEnv
1017 -> InBndr -- The binder
1018 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
1019 -> ([InBndr], InExpr) -- Body of the let/lambda
1022 -> SimplM (SimplEnv, OutExpr)
1024 -- simplNonRecE is used for
1025 -- * non-top-level non-recursive lets in expressions
1028 -- It deals with strict bindings, via the StrictBind continuation,
1029 -- which may abort the whole process
1031 -- The "body" of the binding comes as a pair of ([InId],InExpr)
1032 -- representing a lambda; so we recurse back to simplLam
1033 -- Why? Because of the binder-occ-info-zapping done before
1034 -- the call to simplLam in simplExprF (Lam ...)
1036 -- First deal with type applications and type lets
1037 -- (/\a. e) (Type ty) and (let a = Type ty in e)
1038 simplNonRecE env bndr (Type ty_arg, rhs_se) (bndrs, body) cont
1039 = ASSERT( isTyVar bndr )
1040 do { ty_arg' <- simplType (rhs_se `setInScope` env) ty_arg
1041 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
1043 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
1044 | preInlineUnconditionally env NotTopLevel bndr rhs
1045 = do { tick (PreInlineUnconditionally bndr)
1046 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
1049 = do { simplExprF (rhs_se `setFloats` env) rhs
1050 (StrictBind bndr bndrs body env cont) }
1053 = ASSERT( not (isTyVar bndr) )
1054 do { (env1, bndr1) <- simplNonRecBndr env bndr
1055 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
1056 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
1057 ; simplLam env3 bndrs body cont }
1061 %************************************************************************
1065 %************************************************************************
1068 -- Hack alert: we only distinguish subsumed cost centre stacks for the
1069 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
1070 simplNote :: SimplEnv -> Note -> CoreExpr -> SimplCont
1071 -> SimplM (SimplEnv, OutExpr)
1072 simplNote env (SCC cc) e cont
1073 | pushCCisNop cc (getEnclosingCC env) -- scc "f" (...(scc "f" e)...)
1074 = simplExprF env e cont -- ==> scc "f" (...e...)
1076 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
1077 ; rebuild env (mkSCC cc e') cont }
1079 simplNote env (CoreNote s) e cont
1080 = do { e' <- simplExpr env e
1081 ; rebuild env (Note (CoreNote s) e') cont }
1085 %************************************************************************
1087 \subsection{Dealing with calls}
1089 %************************************************************************
1092 simplVar :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1093 simplVar env var cont
1094 = case substId env var of
1095 DoneEx e -> simplExprF (zapSubstEnv env) e cont
1096 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
1097 DoneId var1 -> completeCall env var1 cont
1098 -- Note [zapSubstEnv]
1099 -- The template is already simplified, so don't re-substitute.
1100 -- This is VITAL. Consider
1102 -- let y = \z -> ...x... in
1104 -- We'll clone the inner \x, adding x->x' in the id_subst
1105 -- Then when we inline y, we must *not* replace x by x' in
1106 -- the inlined copy!!
1108 ---------------------------------------------------------
1109 -- Dealing with a call site
1111 completeCall :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1112 completeCall env var cont
1113 = do { ------------- Try inlining ----------------
1114 dflags <- getDOptsSmpl
1115 ; let (args,call_cont) = contArgs cont
1116 -- The args are OutExprs, obtained by *lazily* substituting
1117 -- in the args found in cont. These args are only examined
1118 -- to limited depth (unless a rule fires). But we must do
1119 -- the substitution; rule matching on un-simplified args would
1122 arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1123 n_val_args = length arg_infos
1124 interesting_cont = interestingCallContext call_cont
1125 active_inline = activeInline env var
1126 maybe_inline = callSiteInline dflags active_inline var
1127 (null args) arg_infos interesting_cont
1128 ; case maybe_inline of {
1129 Just unfolding -- There is an inlining!
1130 -> do { tick (UnfoldingDone var)
1131 ; (if dopt Opt_D_dump_inlinings dflags then
1132 pprTrace ("Inlining done: " ++ showSDoc (ppr var)) (vcat [
1133 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1134 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1135 text "Cont: " <+> ppr call_cont])
1138 simplExprF (zapSubstEnv env) unfolding cont }
1140 ; Nothing -> do -- No inlining!
1142 { rule_base <- getSimplRules
1143 ; let info = mkArgInfo var (getRules rule_base var) n_val_args call_cont
1144 ; rebuildCall env info cont
1147 rebuildCall :: SimplEnv
1150 -> SimplM (SimplEnv, OutExpr)
1151 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_strs = [] }) cont
1152 -- When we run out of strictness args, it means
1153 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1154 -- Then we want to discard the entire strict continuation. E.g.
1155 -- * case (error "hello") of { ... }
1156 -- * (error "Hello") arg
1157 -- * f (error "Hello") where f is strict
1159 -- Then, especially in the first of these cases, we'd like to discard
1160 -- the continuation, leaving just the bottoming expression. But the
1161 -- type might not be right, so we may have to add a coerce.
1162 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1163 = return (env, mk_coerce res) -- contination to discard, else we do it
1164 where -- again and again!
1165 res = mkApps (Var fun) (reverse rev_args)
1166 res_ty = exprType res
1167 cont_ty = contResultType env res_ty cont
1168 co = mkUnsafeCoercion res_ty cont_ty
1169 mk_coerce expr | cont_ty `coreEqType` res_ty = expr
1170 | otherwise = mkCoerce co expr
1172 rebuildCall env info (ApplyTo _ (Type arg_ty) se cont)
1173 = do { ty' <- simplCoercion (se `setInScope` env) arg_ty
1174 ; rebuildCall env (info `addArgTo` Type ty') cont }
1176 rebuildCall env info@(ArgInfo { ai_encl = encl_rules
1177 , ai_strs = str:strs, ai_discs = disc:discs })
1178 (ApplyTo _ arg arg_se cont)
1179 | str -- Strict argument
1180 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1181 simplExprF (arg_se `setFloats` env) arg
1182 (StrictArg info' cci cont)
1185 | otherwise -- Lazy argument
1186 -- DO NOT float anything outside, hence simplExprC
1187 -- There is no benefit (unlike in a let-binding), and we'd
1188 -- have to be very careful about bogus strictness through
1189 -- floating a demanded let.
1190 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1192 ; rebuildCall env (addArgTo info' arg') cont }
1194 info' = info { ai_strs = strs, ai_discs = discs }
1195 cci | encl_rules || disc > 0 = ArgCtxt encl_rules -- Be keener here
1196 | otherwise = BoringCtxt -- Nothing interesting
1198 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_rules = rules }) cont
1199 = do { -- We've accumulated a simplified call in <fun,rev_args>
1200 -- so try rewrite rules; see Note [RULEs apply to simplified arguments]
1201 -- See also Note [Rules for recursive functions]
1202 ; let args = reverse rev_args
1203 env' = zapSubstEnv env
1204 ; mb_rule <- tryRules env rules fun args cont
1206 Just (n_args, rule_rhs) -> simplExprF env' rule_rhs $
1207 pushArgs env' (drop n_args args) cont ;
1208 -- n_args says how many args the rule consumed
1209 ; Nothing -> rebuild env (mkApps (Var fun) args) cont -- No rules
1213 Note [RULES apply to simplified arguments]
1214 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1215 It's very desirable to try RULES once the arguments have been simplified, because
1216 doing so ensures that rule cascades work in one pass. Consider
1217 {-# RULES g (h x) = k x
1220 Then we want to rewrite (g (h x)) to (k x) and only then try f's rules. If
1221 we match f's rules against the un-simplified RHS, it won't match. This
1222 makes a particularly big difference when superclass selectors are involved:
1223 op ($p1 ($p2 (df d)))
1224 We want all this to unravel in one sweeep.
1228 This part of the simplifier may break the no-shadowing invariant
1230 f (...(\a -> e)...) (case y of (a,b) -> e')
1231 where f is strict in its second arg
1232 If we simplify the innermost one first we get (...(\a -> e)...)
1233 Simplifying the second arg makes us float the case out, so we end up with
1234 case y of (a,b) -> f (...(\a -> e)...) e'
1235 So the output does not have the no-shadowing invariant. However, there is
1236 no danger of getting name-capture, because when the first arg was simplified
1237 we used an in-scope set that at least mentioned all the variables free in its
1238 static environment, and that is enough.
1240 We can't just do innermost first, or we'd end up with a dual problem:
1241 case x of (a,b) -> f e (...(\a -> e')...)
1243 I spent hours trying to recover the no-shadowing invariant, but I just could
1244 not think of an elegant way to do it. The simplifier is already knee-deep in
1245 continuations. We have to keep the right in-scope set around; AND we have
1246 to get the effect that finding (error "foo") in a strict arg position will
1247 discard the entire application and replace it with (error "foo"). Getting
1248 all this at once is TOO HARD!
1251 %************************************************************************
1255 %************************************************************************
1258 tryRules :: SimplEnv -> [CoreRule]
1259 -> Id -> [OutExpr] -> SimplCont
1260 -> SimplM (Maybe (Arity, CoreExpr)) -- The arity is the number of
1261 -- args consumed by the rule
1262 tryRules env rules fn args call_cont
1266 = do { dflags <- getDOptsSmpl
1267 ; case activeRule dflags env of {
1268 Nothing -> return Nothing ; -- No rules apply
1270 case lookupRule act_fn (getInScope env) fn args rules of {
1271 Nothing -> return Nothing ; -- No rule matches
1272 Just (rule, rule_rhs) ->
1274 do { tick (RuleFired (ru_name rule))
1275 ; (if dopt Opt_D_dump_rule_firings dflags then
1276 pprTrace "Rule fired" (vcat [
1277 text "Rule:" <+> ftext (ru_name rule),
1278 text "Before:" <+> ppr fn <+> sep (map pprParendExpr args),
1279 text "After: " <+> pprCoreExpr rule_rhs,
1280 text "Cont: " <+> ppr call_cont])
1283 return (Just (ruleArity rule, rule_rhs)) }}}}
1286 Note [Rules for recursive functions]
1287 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1288 You might think that we shouldn't apply rules for a loop breaker:
1289 doing so might give rise to an infinite loop, because a RULE is
1290 rather like an extra equation for the function:
1291 RULE: f (g x) y = x+y
1294 But it's too drastic to disable rules for loop breakers.
1295 Even the foldr/build rule would be disabled, because foldr
1296 is recursive, and hence a loop breaker:
1297 foldr k z (build g) = g k z
1298 So it's up to the programmer: rules can cause divergence
1301 %************************************************************************
1303 Rebuilding a cse expression
1305 %************************************************************************
1307 Note [Case elimination]
1308 ~~~~~~~~~~~~~~~~~~~~~~~
1309 The case-elimination transformation discards redundant case expressions.
1310 Start with a simple situation:
1312 case x# of ===> e[x#/y#]
1315 (when x#, y# are of primitive type, of course). We can't (in general)
1316 do this for algebraic cases, because we might turn bottom into
1319 The code in SimplUtils.prepareAlts has the effect of generalise this
1320 idea to look for a case where we're scrutinising a variable, and we
1321 know that only the default case can match. For example:
1325 DEFAULT -> ...(case x of
1329 Here the inner case is first trimmed to have only one alternative, the
1330 DEFAULT, after which it's an instance of the previous case. This
1331 really only shows up in eliminating error-checking code.
1333 We also make sure that we deal with this very common case:
1338 Here we are using the case as a strict let; if x is used only once
1339 then we want to inline it. We have to be careful that this doesn't
1340 make the program terminate when it would have diverged before, so we
1342 - e is already evaluated (it may so if e is a variable)
1343 - x is used strictly, or
1345 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1347 case e of ===> case e of DEFAULT -> r
1351 Now again the case may be elminated by the CaseElim transformation.
1354 Further notes about case elimination
1355 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1356 Consider: test :: Integer -> IO ()
1359 Turns out that this compiles to:
1362 eta1 :: State# RealWorld ->
1363 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1365 (PrelNum.jtos eta ($w[] @ Char))
1367 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1369 Notice the strange '<' which has no effect at all. This is a funny one.
1370 It started like this:
1372 f x y = if x < 0 then jtos x
1373 else if y==0 then "" else jtos x
1375 At a particular call site we have (f v 1). So we inline to get
1377 if v < 0 then jtos x
1378 else if 1==0 then "" else jtos x
1380 Now simplify the 1==0 conditional:
1382 if v<0 then jtos v else jtos v
1384 Now common-up the two branches of the case:
1386 case (v<0) of DEFAULT -> jtos v
1388 Why don't we drop the case? Because it's strict in v. It's technically
1389 wrong to drop even unnecessary evaluations, and in practice they
1390 may be a result of 'seq' so we *definitely* don't want to drop those.
1391 I don't really know how to improve this situation.
1394 ---------------------------------------------------------
1395 -- Eliminate the case if possible
1397 rebuildCase, reallyRebuildCase
1399 -> OutExpr -- Scrutinee
1400 -> InId -- Case binder
1401 -> [InAlt] -- Alternatives (inceasing order)
1403 -> SimplM (SimplEnv, OutExpr)
1405 --------------------------------------------------
1406 -- 1. Eliminate the case if there's a known constructor
1407 --------------------------------------------------
1409 rebuildCase env scrut case_bndr alts cont
1410 | Lit lit <- scrut -- No need for same treatment as constructors
1411 -- because literals are inlined more vigorously
1412 = do { tick (KnownBranch case_bndr)
1413 ; case findAlt (LitAlt lit) alts of
1414 Nothing -> missingAlt env case_bndr alts cont
1415 Just (_, bs, rhs) -> simple_rhs bs rhs }
1417 | Just (con, ty_args, other_args) <- exprIsConApp_maybe scrut
1418 -- Works when the scrutinee is a variable with a known unfolding
1419 -- as well as when it's an explicit constructor application
1420 = do { tick (KnownBranch case_bndr)
1421 ; case findAlt (DataAlt con) alts of
1422 Nothing -> missingAlt env case_bndr alts cont
1423 Just (DEFAULT, bs, rhs) -> simple_rhs bs rhs
1424 Just (_, bs, rhs) -> knownCon env scrut con ty_args other_args
1425 case_bndr bs rhs cont
1428 simple_rhs bs rhs = ASSERT( null bs )
1429 do { env' <- simplNonRecX env case_bndr scrut
1430 ; simplExprF env' rhs cont }
1433 --------------------------------------------------
1434 -- 2. Eliminate the case if scrutinee is evaluated
1435 --------------------------------------------------
1437 rebuildCase env scrut case_bndr [(_, bndrs, rhs)] cont
1438 -- See if we can get rid of the case altogether
1439 -- See Note [Case eliminiation]
1440 -- mkCase made sure that if all the alternatives are equal,
1441 -- then there is now only one (DEFAULT) rhs
1442 | all isDeadBinder bndrs -- bndrs are [InId]
1444 -- Check that the scrutinee can be let-bound instead of case-bound
1445 , exprOkForSpeculation scrut
1446 -- OK not to evaluate it
1447 -- This includes things like (==# a# b#)::Bool
1448 -- so that we simplify
1449 -- case ==# a# b# of { True -> x; False -> x }
1452 -- This particular example shows up in default methods for
1453 -- comparision operations (e.g. in (>=) for Int.Int32)
1454 || exprIsHNF scrut -- It's already evaluated
1455 || var_demanded_later scrut -- It'll be demanded later
1457 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1458 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1459 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1460 -- its argument: case x of { y -> dataToTag# y }
1461 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1462 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1464 -- Also we don't want to discard 'seq's
1465 = do { tick (CaseElim case_bndr)
1466 ; env' <- simplNonRecX env case_bndr scrut
1467 ; simplExprF env' rhs cont }
1469 -- The case binder is going to be evaluated later,
1470 -- and the scrutinee is a simple variable
1471 var_demanded_later (Var v) = isStrictDmd (idNewDemandInfo case_bndr)
1472 && not (isTickBoxOp v)
1473 -- ugly hack; covering this case is what
1474 -- exprOkForSpeculation was intended for.
1475 var_demanded_later _ = False
1477 --------------------------------------------------
1478 -- 3. Try seq rules; see Note [User-defined RULES for seq] in MkId
1479 --------------------------------------------------
1481 rebuildCase env scrut case_bndr alts@[(_, bndrs, rhs)] cont
1482 | all isDeadBinder (case_bndr : bndrs) -- So this is just 'seq'
1483 = do { let rhs' = substExpr env rhs
1484 out_args = [Type (substTy env (idType case_bndr)),
1485 Type (exprType rhs'), scrut, rhs']
1486 -- Lazily evaluated, so we don't do most of this
1488 ; rule_base <- getSimplRules
1489 ; mb_rule <- tryRules env (getRules rule_base seqId) seqId out_args cont
1491 Just (n_args, res) -> simplExprF (zapSubstEnv env)
1492 (mkApps res (drop n_args out_args))
1494 Nothing -> reallyRebuildCase env scrut case_bndr alts cont }
1496 rebuildCase env scrut case_bndr alts cont
1497 = reallyRebuildCase env scrut case_bndr alts cont
1499 --------------------------------------------------
1500 -- 3. Catch-all case
1501 --------------------------------------------------
1503 reallyRebuildCase env scrut case_bndr alts cont
1504 = do { -- Prepare the continuation;
1505 -- The new subst_env is in place
1506 (env', dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1508 -- Simplify the alternatives
1509 ; (scrut', case_bndr', alts') <- simplAlts env' scrut case_bndr alts dup_cont
1511 -- Check for empty alternatives
1512 ; if null alts' then missingAlt env case_bndr alts cont
1514 { dflags <- getDOptsSmpl
1515 ; case_expr <- mkCase dflags scrut' case_bndr' alts'
1517 -- Notice that rebuild gets the in-scope set from env', not alt_env
1518 -- (which in any case is only build in simplAlts)
1519 -- The case binder *not* scope over the whole returned case-expression
1520 ; rebuild env' case_expr nodup_cont } }
1523 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1524 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1525 way, there's a chance that v will now only be used once, and hence
1528 Historical note: we use to do the "case binder swap" in the Simplifier
1529 so there were additional complications if the scrutinee was a variable.
1530 Now the binder-swap stuff is done in the occurrence analyer; see
1531 OccurAnal Note [Binder swap].
1535 If the case binder is not dead, then neither are the pattern bound
1537 case <any> of x { (a,b) ->
1538 case x of { (p,q) -> p } }
1539 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1540 The point is that we bring into the envt a binding
1542 after the outer case, and that makes (a,b) alive. At least we do unless
1543 the case binder is guaranteed dead.
1545 In practice, the scrutinee is almost always a variable, so we pretty
1546 much always zap the OccInfo of the binders. It doesn't matter much though.
1551 Consider case (v `cast` co) of x { I# y ->
1552 ... (case (v `cast` co) of {...}) ...
1553 We'd like to eliminate the inner case. We can get this neatly by
1554 arranging that inside the outer case we add the unfolding
1555 v |-> x `cast` (sym co)
1556 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1558 Note [Improving seq]
1561 type family F :: * -> *
1562 type instance F Int = Int
1564 ... case e of x { DEFAULT -> rhs } ...
1566 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1568 case e `cast` co of x'::Int
1569 I# x# -> let x = x' `cast` sym co
1572 so that 'rhs' can take advantage of the form of x'.
1574 Notice that Note [Case of cast] may then apply to the result.
1576 Nota Bene: We only do the [Improving seq] transformation if the
1577 case binder 'x' is actually used in the rhs; that is, if the case
1578 is *not* a *pure* seq.
1579 a) There is no point in adding the cast to a pure seq.
1580 b) There is a good reason not to: doing so would interfere
1581 with seq rules (Note [Built-in RULES for seq] in MkId).
1582 In particular, this [Improving seq] thing *adds* a cast
1583 while [Built-in RULES for seq] *removes* one, so they
1586 You might worry about
1587 case v of x { __DEFAULT ->
1588 ... case (v `cast` co) of y { I# -> ... }}
1589 This is a pure seq (since x is unused), so [Improving seq] won't happen.
1590 But it's ok: the simplifier will replace 'v' by 'x' in the rhs to get
1591 case v of x { __DEFAULT ->
1592 ... case (x `cast` co) of y { I# -> ... }}
1593 Now the outer case is not a pure seq, so [Improving seq] will happen,
1594 and then the inner case will disappear.
1596 The need for [Improving seq] showed up in Roman's experiments. Example:
1597 foo :: F Int -> Int -> Int
1598 foo t n = t `seq` bar n
1601 bar n = bar (n - case t of TI i -> i)
1602 Here we'd like to avoid repeated evaluating t inside the loop, by
1603 taking advantage of the `seq`.
1605 At one point I did transformation in LiberateCase, but it's more
1606 robust here. (Otherwise, there's a danger that we'll simply drop the
1607 'seq' altogether, before LiberateCase gets to see it.)
1610 simplAlts :: SimplEnv
1612 -> InId -- Case binder
1613 -> [InAlt] -- Non-empty
1615 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1616 -- Like simplExpr, this just returns the simplified alternatives;
1617 -- it does not return an environment
1619 simplAlts env scrut case_bndr alts cont'
1620 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1621 do { let env0 = zapFloats env
1623 ; (env1, case_bndr1) <- simplBinder env0 case_bndr
1625 ; fam_envs <- getFamEnvs
1626 ; (alt_env', scrut', case_bndr') <- improveSeq fam_envs env1 scrut
1627 case_bndr case_bndr1 alts
1629 ; (imposs_deflt_cons, in_alts) <- prepareAlts scrut' case_bndr' alts
1631 ; alts' <- mapM (simplAlt alt_env' imposs_deflt_cons case_bndr' cont') in_alts
1632 ; return (scrut', case_bndr', alts') }
1635 ------------------------------------
1636 improveSeq :: (FamInstEnv, FamInstEnv) -> SimplEnv
1637 -> OutExpr -> InId -> OutId -> [InAlt]
1638 -> SimplM (SimplEnv, OutExpr, OutId)
1639 -- Note [Improving seq]
1640 improveSeq fam_envs env scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1641 | not (isDeadBinder case_bndr) -- Not a pure seq! See the Note!
1642 , Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1643 = do { case_bndr2 <- newId (fsLit "nt") ty2
1644 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1645 env2 = extendIdSubst env case_bndr rhs
1646 ; return (env2, scrut `Cast` co, case_bndr2) }
1648 improveSeq _ env scrut _ case_bndr1 _
1649 = return (env, scrut, case_bndr1)
1652 ------------------------------------
1653 simplAlt :: SimplEnv
1654 -> [AltCon] -- These constructors can't be present when
1655 -- matching the DEFAULT alternative
1656 -> OutId -- The case binder
1661 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1662 = ASSERT( null bndrs )
1663 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1664 -- Record the constructors that the case-binder *can't* be.
1665 ; rhs' <- simplExprC env' rhs cont'
1666 ; return (DEFAULT, [], rhs') }
1668 simplAlt env _ case_bndr' cont' (LitAlt lit, bndrs, rhs)
1669 = ASSERT( null bndrs )
1670 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1671 ; rhs' <- simplExprC env' rhs cont'
1672 ; return (LitAlt lit, [], rhs') }
1674 simplAlt env _ case_bndr' cont' (DataAlt con, vs, rhs)
1675 = do { -- Deal with the pattern-bound variables
1676 -- Mark the ones that are in ! positions in the
1677 -- data constructor as certainly-evaluated.
1678 -- NB: simplLamBinders preserves this eval info
1679 let vs_with_evals = add_evals (dataConRepStrictness con)
1680 ; (env', vs') <- simplLamBndrs env vs_with_evals
1682 -- Bind the case-binder to (con args)
1683 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1684 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1685 env'' = addBinderUnfolding env' case_bndr'
1686 (mkConApp con con_args)
1688 ; rhs' <- simplExprC env'' rhs cont'
1689 ; return (DataAlt con, vs', rhs') }
1691 -- add_evals records the evaluated-ness of the bound variables of
1692 -- a case pattern. This is *important*. Consider
1693 -- data T = T !Int !Int
1695 -- case x of { T a b -> T (a+1) b }
1697 -- We really must record that b is already evaluated so that we don't
1698 -- go and re-evaluate it when constructing the result.
1699 -- See Note [Data-con worker strictness] in MkId.lhs
1704 go (v:vs') strs | isTyVar v = v : go vs' strs
1705 go (v:vs') (str:strs)
1706 | isMarkedStrict str = evald_v : go vs' strs
1707 | otherwise = zapped_v : go vs' strs
1709 zapped_v = zap_occ_info v
1710 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1711 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr the_strs)
1713 -- See Note [zapOccInfo]
1714 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1716 -- to the envt; so vs are now very much alive
1717 -- Note [Aug06] I can't see why this actually matters, but it's neater
1718 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1719 -- ==> case e of t { (a,b) -> ...(a)... }
1720 -- Look, Ma, a is alive now.
1721 zap_occ_info = zapCasePatIdOcc case_bndr'
1723 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1724 addBinderUnfolding env bndr rhs
1725 = modifyInScope env (bndr `setIdUnfolding` mkUnfolding False rhs)
1727 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1728 addBinderOtherCon env bndr cons
1729 = modifyInScope env (bndr `setIdUnfolding` mkOtherCon cons)
1731 zapCasePatIdOcc :: Id -> Id -> Id
1732 -- Consider case e of b { (a,b) -> ... }
1733 -- Then if we bind b to (a,b) in "...", and b is not dead,
1734 -- then we must zap the deadness info on a,b
1735 zapCasePatIdOcc case_bndr
1736 | isDeadBinder case_bndr = \ pat_id -> pat_id
1737 | otherwise = \ pat_id -> zapIdOccInfo pat_id
1741 %************************************************************************
1743 \subsection{Known constructor}
1745 %************************************************************************
1747 We are a bit careful with occurrence info. Here's an example
1749 (\x* -> case x of (a*, b) -> f a) (h v, e)
1751 where the * means "occurs once". This effectively becomes
1752 case (h v, e) of (a*, b) -> f a)
1754 let a* = h v; b = e in f a
1758 All this should happen in one sweep.
1761 knownCon :: SimplEnv
1762 -> OutExpr -- The scrutinee
1763 -> DataCon -> [OutType] -> [OutExpr] -- The scrutinee (in pieces)
1764 -> InId -> [InBndr] -> InExpr -- The alternative
1766 -> SimplM (SimplEnv, OutExpr)
1768 knownCon env scrut dc dc_ty_args dc_args bndr bs rhs cont
1769 = do { env' <- bind_args env bs dc_args
1771 -- It's useful to bind bndr to scrut, rather than to a fresh
1772 -- binding x = Con arg1 .. argn
1773 -- because very often the scrut is a variable, so we avoid
1774 -- creating, and then subsequently eliminating, a let-binding
1775 -- BUT, if scrut is a not a variable, we must be careful
1776 -- about duplicating the arg redexes; in that case, make
1777 -- a new con-app from the args
1778 bndr_rhs | exprIsTrivial scrut = scrut
1779 | otherwise = con_app
1780 con_app = Var (dataConWorkId dc)
1781 `mkTyApps` dc_ty_args
1782 `mkApps` [substExpr env' (varToCoreExpr b) | b <- bs]
1783 -- dc_ty_args are aready OutTypes, but bs are InBndrs
1785 ; env'' <- simplNonRecX env' bndr bndr_rhs
1786 ; simplExprF env'' rhs cont }
1788 zap_occ = zapCasePatIdOcc bndr -- bndr is an InId
1791 bind_args env' [] _ = return env'
1793 bind_args env' (b:bs') (Type ty : args)
1794 = ASSERT( isTyVar b )
1795 bind_args (extendTvSubst env' b ty) bs' args
1797 bind_args env' (b:bs') (arg : args)
1799 do { let b' = zap_occ b
1800 -- Note that the binder might be "dead", because it doesn't
1801 -- occur in the RHS; and simplNonRecX may therefore discard
1802 -- it via postInlineUnconditionally.
1803 -- Nevertheless we must keep it if the case-binder is alive,
1804 -- because it may be used in the con_app. See Note [zapOccInfo]
1805 ; env'' <- simplNonRecX env' b' arg
1806 ; bind_args env'' bs' args }
1809 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr dc_args $$
1810 text "scrut:" <+> ppr scrut
1813 missingAlt :: SimplEnv -> Id -> [InAlt] -> SimplCont -> SimplM (SimplEnv, OutExpr)
1814 -- This isn't strictly an error, although it is unusual.
1815 -- It's possible that the simplifer might "see" that
1816 -- an inner case has no accessible alternatives before
1817 -- it "sees" that the entire branch of an outer case is
1818 -- inaccessible. So we simply put an error case here instead.
1819 missingAlt env case_bndr alts cont
1820 = WARN( True, ptext (sLit "missingAlt") <+> ppr case_bndr )
1821 return (env, mkImpossibleExpr res_ty)
1823 res_ty = contResultType env (substTy env (coreAltsType alts)) cont
1827 %************************************************************************
1829 \subsection{Duplicating continuations}
1831 %************************************************************************
1834 prepareCaseCont :: SimplEnv
1835 -> [InAlt] -> SimplCont
1836 -> SimplM (SimplEnv, SimplCont,SimplCont)
1837 -- Return a duplicatable continuation, a non-duplicable part
1838 -- plus some extra bindings (that scope over the entire
1841 -- No need to make it duplicatable if there's only one alternative
1842 prepareCaseCont env [_] cont = return (env, cont, mkBoringStop)
1843 prepareCaseCont env _ cont = mkDupableCont env cont
1847 mkDupableCont :: SimplEnv -> SimplCont
1848 -> SimplM (SimplEnv, SimplCont, SimplCont)
1850 mkDupableCont env cont
1851 | contIsDupable cont
1852 = return (env, cont, mkBoringStop)
1854 mkDupableCont _ (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1856 mkDupableCont env (CoerceIt ty cont)
1857 = do { (env', dup, nodup) <- mkDupableCont env cont
1858 ; return (env', CoerceIt ty dup, nodup) }
1860 mkDupableCont env cont@(StrictBind {})
1861 = return (env, mkBoringStop, cont)
1862 -- See Note [Duplicating StrictBind]
1864 mkDupableCont env (StrictArg info cci cont)
1865 -- See Note [Duplicating StrictArg]
1866 = do { (env', dup, nodup) <- mkDupableCont env cont
1867 ; (env'', args') <- mapAccumLM makeTrivial env' (ai_args info)
1868 ; return (env'', StrictArg (info { ai_args = args' }) cci dup, nodup) }
1870 mkDupableCont env (ApplyTo _ arg se cont)
1871 = -- e.g. [...hole...] (...arg...)
1873 -- let a = ...arg...
1874 -- in [...hole...] a
1875 do { (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1876 ; arg' <- simplExpr (se `setInScope` env') arg
1877 ; (env'', arg'') <- makeTrivial env' arg'
1878 ; let app_cont = ApplyTo OkToDup arg'' (zapSubstEnv env'') dup_cont
1879 ; return (env'', app_cont, nodup_cont) }
1881 mkDupableCont env cont@(Select _ case_bndr [(_, bs, _rhs)] _ _)
1882 -- See Note [Single-alternative case]
1883 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1884 -- | not (isDeadBinder case_bndr)
1885 | all isDeadBinder bs -- InIds
1886 && not (isUnLiftedType (idType case_bndr))
1887 -- Note [Single-alternative-unlifted]
1888 = return (env, mkBoringStop, cont)
1890 mkDupableCont env (Select _ case_bndr alts se cont)
1891 = -- e.g. (case [...hole...] of { pi -> ei })
1893 -- let ji = \xij -> ei
1894 -- in case [...hole...] of { pi -> ji xij }
1895 do { tick (CaseOfCase case_bndr)
1896 ; (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1897 -- NB: call mkDupableCont here, *not* prepareCaseCont
1898 -- We must make a duplicable continuation, whereas prepareCaseCont
1899 -- doesn't when there is a single case branch
1901 ; let alt_env = se `setInScope` env'
1902 ; (alt_env', case_bndr') <- simplBinder alt_env case_bndr
1903 ; alts' <- mapM (simplAlt alt_env' [] case_bndr' dup_cont) alts
1904 -- Safe to say that there are no handled-cons for the DEFAULT case
1905 -- NB: simplBinder does not zap deadness occ-info, so
1906 -- a dead case_bndr' will still advertise its deadness
1907 -- This is really important because in
1908 -- case e of b { (# p,q #) -> ... }
1909 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1910 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1911 -- In the new alts we build, we have the new case binder, so it must retain
1913 -- NB: we don't use alt_env further; it has the substEnv for
1914 -- the alternatives, and we don't want that
1916 ; (env'', alts'') <- mkDupableAlts env' case_bndr' alts'
1917 ; return (env'', -- Note [Duplicated env]
1918 Select OkToDup case_bndr' alts'' (zapSubstEnv env'') mkBoringStop,
1922 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1923 -> SimplM (SimplEnv, [InAlt])
1924 -- Absorbs the continuation into the new alternatives
1926 mkDupableAlts env case_bndr' the_alts
1929 go env0 [] = return (env0, [])
1931 = do { (env1, alt') <- mkDupableAlt env0 case_bndr' alt
1932 ; (env2, alts') <- go env1 alts
1933 ; return (env2, alt' : alts' ) }
1935 mkDupableAlt :: SimplEnv -> OutId -> (AltCon, [CoreBndr], CoreExpr)
1936 -> SimplM (SimplEnv, (AltCon, [CoreBndr], CoreExpr))
1937 mkDupableAlt env case_bndr (con, bndrs', rhs')
1938 | exprIsDupable rhs' -- Note [Small alternative rhs]
1939 = return (env, (con, bndrs', rhs'))
1941 = do { let rhs_ty' = exprType rhs'
1942 scrut_ty = idType case_bndr
1945 DEFAULT -> case_bndr
1946 DataAlt dc -> setIdUnfolding case_bndr unf
1948 -- See Note [Case binders and join points]
1949 unf = mkInlineRule InlSat rhs 0
1950 rhs = mkConApp dc (map Type (tyConAppArgs scrut_ty)
1951 ++ varsToCoreExprs bndrs')
1953 LitAlt {} -> WARN( True, ptext (sLit "mkDupableAlt")
1954 <+> ppr case_bndr <+> ppr con )
1956 -- The case binder is alive but trivial, so why has
1957 -- it not been substituted away?
1959 used_bndrs' | isDeadBinder case_bndr = filter abstract_over bndrs'
1960 | otherwise = bndrs' ++ [case_bndr_w_unf]
1963 | isTyVar bndr = True -- Abstract over all type variables just in case
1964 | otherwise = not (isDeadBinder bndr)
1965 -- The deadness info on the new Ids is preserved by simplBinders
1967 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1968 <- if (any isId used_bndrs')
1969 then return (used_bndrs', varsToCoreExprs used_bndrs')
1970 else do { rw_id <- newId (fsLit "w") realWorldStatePrimTy
1971 ; return ([rw_id], [Var realWorldPrimId]) }
1973 ; join_bndr <- newId (fsLit "$j") (mkPiTypes final_bndrs' rhs_ty')
1974 -- Note [Funky mkPiTypes]
1976 ; let -- We make the lambdas into one-shot-lambdas. The
1977 -- join point is sure to be applied at most once, and doing so
1978 -- prevents the body of the join point being floated out by
1979 -- the full laziness pass
1980 really_final_bndrs = map one_shot final_bndrs'
1981 one_shot v | isId v = setOneShotLambda v
1983 join_rhs = mkLams really_final_bndrs rhs'
1984 join_call = mkApps (Var join_bndr) final_args
1986 ; env' <- addPolyBind NotTopLevel env (NonRec join_bndr join_rhs)
1987 ; return (env', (con, bndrs', join_call)) }
1988 -- See Note [Duplicated env]
1991 Note [Case binders and join points]
1992 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1994 case (case .. ) of c {
1997 If we make a join point with c but not c# we get
1998 $j = \c -> ....c....
2000 But if later inlining scrutines the c, thus
2002 $j = \c -> ... case c of { I# y -> ... } ...
2004 we won't see that 'c' has already been scrutinised. This actually
2005 happens in the 'tabulate' function in wave4main, and makes a significant
2006 difference to allocation.
2008 An alternative plan is this:
2010 $j = \c# -> let c = I# c# in ...c....
2012 but that is bad if 'c' is *not* later scrutinised.
2014 So instead we do both: we pass 'c' and 'c#' , and record in c's inlining
2015 that it's really I# c#, thus
2017 $j = \c# -> \c[=I# c#] -> ...c....
2019 Absence analysis may later discard 'c'.
2022 Note [Duplicated env]
2023 ~~~~~~~~~~~~~~~~~~~~~
2024 Some of the alternatives are simplified, but have not been turned into a join point
2025 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
2026 bind the join point, because it might to do PostInlineUnconditionally, and
2027 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
2028 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
2029 at worst delays the join-point inlining.
2031 Note [Small alternative rhs]
2032 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2033 It is worth checking for a small RHS because otherwise we
2034 get extra let bindings that may cause an extra iteration of the simplifier to
2035 inline back in place. Quite often the rhs is just a variable or constructor.
2036 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
2037 iterations because the version with the let bindings looked big, and so wasn't
2038 inlined, but after the join points had been inlined it looked smaller, and so
2041 NB: we have to check the size of rhs', not rhs.
2042 Duplicating a small InAlt might invalidate occurrence information
2043 However, if it *is* dupable, we return the *un* simplified alternative,
2044 because otherwise we'd need to pair it up with an empty subst-env....
2045 but we only have one env shared between all the alts.
2046 (Remember we must zap the subst-env before re-simplifying something).
2047 Rather than do this we simply agree to re-simplify the original (small) thing later.
2049 Note [Funky mkPiTypes]
2050 ~~~~~~~~~~~~~~~~~~~~~~
2051 Notice the funky mkPiTypes. If the contructor has existentials
2052 it's possible that the join point will be abstracted over
2053 type varaibles as well as term variables.
2054 Example: Suppose we have
2055 data T = forall t. C [t]
2057 case (case e of ...) of
2059 We get the join point
2060 let j :: forall t. [t] -> ...
2061 j = /\t \xs::[t] -> rhs
2063 case (case e of ...) of
2064 C t xs::[t] -> j t xs
2066 Note [Join point abstaction]
2067 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2068 If we try to lift a primitive-typed something out
2069 for let-binding-purposes, we will *caseify* it (!),
2070 with potentially-disastrous strictness results. So
2071 instead we turn it into a function: \v -> e
2072 where v::State# RealWorld#. The value passed to this function
2073 is realworld#, which generates (almost) no code.
2075 There's a slight infelicity here: we pass the overall
2076 case_bndr to all the join points if it's used in *any* RHS,
2077 because we don't know its usage in each RHS separately
2079 We used to say "&& isUnLiftedType rhs_ty'" here, but now
2080 we make the join point into a function whenever used_bndrs'
2081 is empty. This makes the join-point more CPR friendly.
2082 Consider: let j = if .. then I# 3 else I# 4
2083 in case .. of { A -> j; B -> j; C -> ... }
2085 Now CPR doesn't w/w j because it's a thunk, so
2086 that means that the enclosing function can't w/w either,
2087 which is a lose. Here's the example that happened in practice:
2088 kgmod :: Int -> Int -> Int
2089 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
2093 I have seen a case alternative like this:
2095 It's a bit silly to add the realWorld dummy arg in this case, making
2098 (the \v alone is enough to make CPR happy) but I think it's rare
2100 Note [Duplicating StrictArg]
2101 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2102 The original plan had (where E is a big argument)
2104 ==> let $j = \a -> f E a
2107 But this is terrible! Here's an example:
2108 && E (case x of { T -> F; F -> T })
2109 Now, && is strict so we end up simplifying the case with
2110 an ArgOf continuation. If we let-bind it, we get
2111 let $j = \v -> && E v
2112 in simplExpr (case x of { T -> F; F -> T })
2114 And after simplifying more we get
2115 let $j = \v -> && E v
2116 in case x of { T -> $j F; F -> $j T }
2117 Which is a Very Bad Thing
2119 What we do now is this
2123 Now if the thing in the hole is a case expression (which is when
2124 we'll call mkDupableCont), we'll push the function call into the
2125 branches, which is what we want. Now RULES for f may fire, and
2126 call-pattern specialisation. Here's an example from Trac #3116
2129 _ -> Chunk p fpc (o+1) (l-1) bs')
2130 If we can push the call for 'go' inside the case, we get
2131 call-pattern specialisation for 'go', which is *crucial* for
2134 Here is the (&&) example:
2135 && E (case x of { T -> F; F -> T })
2137 case x of { T -> && a F; F -> && a T }
2141 * Arguments to f *after* the strict one are handled by
2142 the ApplyTo case of mkDupableCont. Eg
2145 * We can only do the let-binding of E because the function
2146 part of a StrictArg continuation is an explicit syntax
2147 tree. In earlier versions we represented it as a function
2148 (CoreExpr -> CoreEpxr) which we couldn't take apart.
2150 Do *not* duplicate StrictBind and StritArg continuations. We gain
2151 nothing by propagating them into the expressions, and we do lose a
2154 The desire not to duplicate is the entire reason that
2155 mkDupableCont returns a pair of continuations.
2157 Note [Duplicating StrictBind]
2158 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2159 Unlike StrictArg, there doesn't seem anything to gain from
2160 duplicating a StrictBind continuation, so we don't.
2162 The desire not to duplicate is the entire reason that
2163 mkDupableCont returns a pair of continuations.
2166 Note [Single-alternative cases]
2167 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2168 This case is just like the ArgOf case. Here's an example:
2172 case (case x of I# x' ->
2174 True -> I# (negate# x')
2175 False -> I# x') of y {
2177 Because the (case x) has only one alternative, we'll transform to
2179 case (case x' <# 0# of
2180 True -> I# (negate# x')
2181 False -> I# x') of y {
2183 But now we do *NOT* want to make a join point etc, giving
2185 let $j = \y -> MkT y
2187 True -> $j (I# (negate# x'))
2189 In this case the $j will inline again, but suppose there was a big
2190 strict computation enclosing the orginal call to MkT. Then, it won't
2191 "see" the MkT any more, because it's big and won't get duplicated.
2192 And, what is worse, nothing was gained by the case-of-case transform.
2194 When should use this case of mkDupableCont?
2195 However, matching on *any* single-alternative case is a *disaster*;
2196 e.g. case (case ....) of (a,b) -> (# a,b #)
2197 We must push the outer case into the inner one!
2200 * Match [(DEFAULT,_,_)], but in the common case of Int,
2201 the alternative-filling-in code turned the outer case into
2202 case (...) of y { I# _ -> MkT y }
2204 * Match on single alternative plus (not (isDeadBinder case_bndr))
2205 Rationale: pushing the case inwards won't eliminate the construction.
2206 But there's a risk of
2207 case (...) of y { (a,b) -> let z=(a,b) in ... }
2208 Now y looks dead, but it'll come alive again. Still, this
2209 seems like the best option at the moment.
2211 * Match on single alternative plus (all (isDeadBinder bndrs))
2212 Rationale: this is essentially seq.
2214 * Match when the rhs is *not* duplicable, and hence would lead to a
2215 join point. This catches the disaster-case above. We can test
2216 the *un-simplified* rhs, which is fine. It might get bigger or
2217 smaller after simplification; if it gets smaller, this case might
2218 fire next time round. NB also that we must test contIsDupable
2219 case_cont *btoo, because case_cont might be big!
2221 HOWEVER: I found that this version doesn't work well, because
2222 we can get let x = case (...) of { small } in ...case x...
2223 When x is inlined into its full context, we find that it was a bad
2224 idea to have pushed the outer case inside the (...) case.
2226 Note [Single-alternative-unlifted]
2227 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2228 Here's another single-alternative where we really want to do case-of-case:
2236 case y_s6X of tpl_s7m {
2237 M1.Mk1 ipv_s70 -> ipv_s70;
2238 M1.Mk2 ipv_s72 -> ipv_s72;
2244 case x_s74 of tpl_s7n {
2245 M1.Mk1 ipv_s77 -> ipv_s77;
2246 M1.Mk2 ipv_s79 -> ipv_s79;
2250 { __DEFAULT -> ==# [wild1_s7b wild_s7c];
2254 So the outer case is doing *nothing at all*, other than serving as a
2255 join-point. In this case we really want to do case-of-case and decide
2256 whether to use a real join point or just duplicate the continuation.
2258 Hence: check whether the case binder's type is unlifted, because then
2259 the outer case is *not* a seq.