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 Demand ( 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 `setStrictnessInfo` strictnessInfo info
446 `setDemandInfo` demandInfo info
449 prepareRhs env0 _ rhs0
450 = do { (_is_exp, env1, rhs1) <- go 0 env0 rhs0
451 ; return (env1, rhs1) }
453 go n_val_args env (Cast rhs co)
454 = do { (is_exp, env', rhs') <- go n_val_args env rhs
455 ; return (is_exp, env', Cast rhs' co) }
456 go n_val_args env (App fun (Type ty))
457 = do { (is_exp, env', rhs') <- go n_val_args env fun
458 ; return (is_exp, env', App rhs' (Type ty)) }
459 go n_val_args env (App fun arg)
460 = do { (is_exp, 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_exp, env, Var fun)
468 is_exp = isExpandableApp fun n_val_args -- The fun a constructor or PAP
469 -- See Note [CONLIKE pragma] in BasicTypes
470 -- The definition of is_exp should match that in
471 -- OccurAnal.occAnalApp
474 = return (False, env, other)
478 Note [Float coercions]
479 ~~~~~~~~~~~~~~~~~~~~~~
480 When we find the binding
482 we'd like to transform it to
484 x = x `cast` co -- A trivial binding
485 There's a chance that e will be a constructor application or function, or something
486 like that, so moving the coerion to the usage site may well cancel the coersions
487 and lead to further optimisation. Example:
490 data instance T Int = T Int
492 foo :: Int -> Int -> Int
497 go n = case x of { T m -> go (n-m) }
498 -- This case should optimise
500 Note [Preserve strictness when floating coercions]
501 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
502 In the Note [Float coercions] transformation, keep the strictness info.
504 f = e `cast` co -- f has strictness SSL
506 f' = e -- f' also has strictness SSL
507 f = f' `cast` co -- f still has strictness SSL
509 Its not wrong to drop it on the floor, but better to keep it.
511 Note [Float coercions (unlifted)]
512 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
513 BUT don't do [Float coercions] if 'e' has an unlifted type.
516 foo :: Int = (error (# Int,Int #) "urk")
517 `cast` CoUnsafe (# Int,Int #) Int
519 If do the makeTrivial thing to the error call, we'll get
520 foo = case error (# Int,Int #) "urk" of v -> v `cast` ...
521 But 'v' isn't in scope!
523 These strange casts can happen as a result of case-of-case
524 bar = case (case x of { T -> (# 2,3 #); F -> error "urk" }) of
529 makeTrivial :: SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
530 -- Binds the expression to a variable, if it's not trivial, returning the variable
531 makeTrivial env expr = makeTrivialWithInfo env vanillaIdInfo expr
533 makeTrivialWithInfo :: SimplEnv -> IdInfo -> OutExpr -> SimplM (SimplEnv, OutExpr)
534 -- Propagate strictness and demand info to the new binder
535 -- Note [Preserve strictness when floating coercions]
536 makeTrivialWithInfo env info expr
539 | otherwise -- See Note [Take care] below
540 = do { uniq <- getUniqueM
541 ; let name = mkSystemVarName uniq (fsLit "a")
542 var = mkLocalIdWithInfo name (exprType expr) info
543 ; env' <- completeNonRecX env False var var expr
544 ; return (env', substExpr env' (Var var)) }
545 -- The substitution is needed becase we're constructing a new binding
547 -- And if rhs is of form (rhs1 |> co), then we might get
550 -- and now a's RHS is trivial and can be substituted out, and that
551 -- is what completeNonRecX will do
555 %************************************************************************
557 \subsection{Completing a lazy binding}
559 %************************************************************************
562 * deals only with Ids, not TyVars
563 * takes an already-simplified binder and RHS
564 * is used for both recursive and non-recursive bindings
565 * is used for both top-level and non-top-level bindings
567 It does the following:
568 - tries discarding a dead binding
569 - tries PostInlineUnconditionally
570 - add unfolding [this is the only place we add an unfolding]
573 It does *not* attempt to do let-to-case. Why? Because it is used for
574 - top-level bindings (when let-to-case is impossible)
575 - many situations where the "rhs" is known to be a WHNF
576 (so let-to-case is inappropriate).
578 Nor does it do the atomic-argument thing
581 completeBind :: SimplEnv
582 -> TopLevelFlag -- Flag stuck into unfolding
583 -> InId -- Old binder
584 -> OutId -> OutExpr -- New binder and RHS
586 -- completeBind may choose to do its work
587 -- * by extending the substitution (e.g. let x = y in ...)
588 -- * or by adding to the floats in the envt
590 completeBind env top_lvl old_bndr new_bndr new_rhs
591 = do { let old_info = idInfo old_bndr
592 old_unf = unfoldingInfo old_info
593 occ_info = occInfo old_info
595 ; new_unfolding <- simplUnfolding env top_lvl old_bndr occ_info new_rhs old_unf
597 ; if postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs new_unfolding
598 -- Inline and discard the binding
599 then do { tick (PostInlineUnconditionally old_bndr)
600 ; -- pprTrace "postInlineUnconditionally" (ppr old_bndr <+> equals <+> ppr new_rhs) $
601 return (extendIdSubst env old_bndr (DoneEx new_rhs)) }
602 -- Use the substitution to make quite, quite sure that the
603 -- substitution will happen, since we are going to discard the binding
605 else return (addNonRecWithUnf env new_bndr new_rhs new_unfolding) }
607 ------------------------------
608 addPolyBind :: TopLevelFlag -> SimplEnv -> OutBind -> SimplM SimplEnv
609 -- Add a new binding to the environment, complete with its unfolding
610 -- but *do not* do postInlineUnconditionally, because we have already
611 -- processed some of the scope of the binding
612 -- We still want the unfolding though. Consider
614 -- x = /\a. let y = ... in Just y
616 -- Then we float the y-binding out (via abstractFloats and addPolyBind)
617 -- but 'x' may well then be inlined in 'body' in which case we'd like the
618 -- opportunity to inline 'y' too.
620 addPolyBind top_lvl env (NonRec poly_id rhs)
621 = do { unfolding <- simplUnfolding env top_lvl poly_id NoOccInfo rhs noUnfolding
622 -- Assumes that poly_id did not have an INLINE prag
623 -- which is perhaps wrong. ToDo: think about this
624 ; return (addNonRecWithUnf env poly_id rhs unfolding) }
626 addPolyBind _ env bind@(Rec _) = return (extendFloats env bind)
627 -- Hack: letrecs are more awkward, so we extend "by steam"
628 -- without adding unfoldings etc. At worst this leads to
629 -- more simplifier iterations
631 ------------------------------
632 addNonRecWithUnf :: SimplEnv
633 -> OutId -> OutExpr -- New binder and RHS
634 -> Unfolding -- New unfolding
636 addNonRecWithUnf env new_bndr new_rhs new_unfolding
637 = let new_arity = exprArity new_rhs
638 old_arity = idArity new_bndr
639 info1 = idInfo new_bndr `setArityInfo` new_arity
641 -- Unfolding info: Note [Setting the new unfolding]
642 info2 = info1 `setUnfoldingInfo` new_unfolding
644 -- Demand info: Note [Setting the demand info]
645 info3 | isEvaldUnfolding new_unfolding = zapDemandInfo info2 `orElse` info2
648 final_id = new_bndr `setIdInfo` info3
649 dmd_arity = length $ fst $ splitStrictSig $ idStrictness new_bndr
651 ASSERT( isId new_bndr )
652 WARN( new_arity < old_arity || new_arity < dmd_arity,
653 (ptext (sLit "Arity decrease:") <+> ppr final_id <+> ppr old_arity
654 <+> ppr new_arity <+> ppr dmd_arity) )
655 -- Note [Arity decrease]
657 final_id `seq` -- This seq forces the Id, and hence its IdInfo,
658 -- and hence any inner substitutions
659 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
660 addNonRec env final_id new_rhs
661 -- The addNonRec adds it to the in-scope set too
663 ------------------------------
664 simplUnfolding :: SimplEnv-> TopLevelFlag
665 -> Id -- Debug output only
666 -> OccInfo -> OutExpr
667 -> Unfolding -> SimplM Unfolding
668 -- Note [Setting the new unfolding]
669 simplUnfolding env _ _ _ _ (DFunUnfolding con ops)
670 = return (DFunUnfolding con ops')
672 ops' = map (CoreSubst.substExpr (mkCoreSubst env)) ops
674 simplUnfolding env top_lvl _ _ _
675 (CoreUnfolding { uf_tmpl = expr, uf_arity = arity
676 , uf_src = src, uf_guidance = guide })
677 | isInlineRuleSource src
678 = do { expr' <- simplExpr (updMode updModeForInlineRules env) expr
679 -- See Note [Simplifying gently inside InlineRules] in SimplUtils
680 ; let src' = CoreSubst.substUnfoldingSource (mkCoreSubst env) src
681 ; return (mkCoreUnfolding (isTopLevel top_lvl) src' expr' arity guide) }
682 -- See Note [Top-level flag on inline rules] in CoreUnfold
684 simplUnfolding _ top_lvl _ _occ_info new_rhs _
685 = return (mkUnfolding (isTopLevel top_lvl) new_rhs)
686 -- We make an unfolding *even for loop-breakers*.
687 -- Reason: (a) It might be useful to know that they are WHNF
688 -- (b) In TidyPgm we currently assume that, if we want to
689 -- expose the unfolding then indeed we *have* an unfolding
690 -- to expose. (We could instead use the RHS, but currently
691 -- we don't.) The simple thing is always to have one.
694 Note [Arity decrease]
695 ~~~~~~~~~~~~~~~~~~~~~
696 Generally speaking the arity of a binding should not decrease. But it *can*
697 legitimately happen becuase of RULES. Eg
699 where g has arity 2, will have arity 2. But if there's a rewrite rule
701 where h has arity 1, then f's arity will decrease. Here's a real-life example,
702 which is in the output of Specialise:
705 $dm {Arity 2} = \d.\x. op d
706 {-# RULES forall d. $dm Int d = $s$dm #-}
708 dInt = MkD .... opInt ...
709 opInt {Arity 1} = $dm dInt
711 $s$dm {Arity 0} = \x. op dInt }
713 Here opInt has arity 1; but when we apply the rule its arity drops to 0.
714 That's why Specialise goes to a little trouble to pin the right arity
715 on specialised functions too.
717 Note [Setting the new unfolding]
718 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
719 * If there's an INLINE pragma, we simplify the RHS gently. Maybe we
720 should do nothing at all, but simplifying gently might get rid of
723 * If not, we make an unfolding from the new RHS. But *only* for
724 non-loop-breakers. Making loop breakers not have an unfolding at all
725 means that we can avoid tests in exprIsConApp, for example. This is
726 important: if exprIsConApp says 'yes' for a recursive thing, then we
727 can get into an infinite loop
729 If there's an InlineRule on a loop breaker, we hang on to the inlining.
730 It's pretty dodgy, but the user did say 'INLINE'. May need to revisit
733 Note [Setting the demand info]
734 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
735 If the unfolding is a value, the demand info may
736 go pear-shaped, so we nuke it. Example:
738 case x of (p,q) -> h p q x
739 Here x is certainly demanded. But after we've nuked
740 the case, we'll get just
741 let x = (a,b) in h a b x
742 and now x is not demanded (I'm assuming h is lazy)
743 This really happens. Similarly
744 let f = \x -> e in ...f..f...
745 After inlining f at some of its call sites the original binding may
746 (for example) be no longer strictly demanded.
747 The solution here is a bit ad hoc...
750 %************************************************************************
752 \subsection[Simplify-simplExpr]{The main function: simplExpr}
754 %************************************************************************
756 The reason for this OutExprStuff stuff is that we want to float *after*
757 simplifying a RHS, not before. If we do so naively we get quadratic
758 behaviour as things float out.
760 To see why it's important to do it after, consider this (real) example:
774 a -- Can't inline a this round, cos it appears twice
778 Each of the ==> steps is a round of simplification. We'd save a
779 whole round if we float first. This can cascade. Consider
784 let f = let d1 = ..d.. in \y -> e
788 in \x -> ...(\y ->e)...
790 Only in this second round can the \y be applied, and it
791 might do the same again.
795 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
796 simplExpr env expr = simplExprC env expr mkBoringStop
798 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
799 -- Simplify an expression, given a continuation
800 simplExprC env expr cont
801 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
802 do { (env', expr') <- simplExprF (zapFloats env) expr cont
803 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
804 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
805 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
806 return (wrapFloats env' expr') }
808 --------------------------------------------------
809 simplExprF :: SimplEnv -> InExpr -> SimplCont
810 -> SimplM (SimplEnv, OutExpr)
812 simplExprF env e cont
813 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
814 simplExprF' env e cont
816 simplExprF' :: SimplEnv -> InExpr -> SimplCont
817 -> SimplM (SimplEnv, OutExpr)
818 simplExprF' env (Var v) cont = simplVar env v cont
819 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
820 simplExprF' env (Note n expr) cont = simplNote env n expr cont
821 simplExprF' env (Cast body co) cont = simplCast env body co cont
822 simplExprF' env (App fun arg) cont = simplExprF env fun $
823 ApplyTo NoDup arg env cont
825 simplExprF' env expr@(Lam _ _) cont
826 = simplLam env (map zap bndrs) body cont
827 -- The main issue here is under-saturated lambdas
828 -- (\x1. \x2. e) arg1
829 -- Here x1 might have "occurs-once" occ-info, because occ-info
830 -- is computed assuming that a group of lambdas is applied
831 -- all at once. If there are too few args, we must zap the
834 n_args = countArgs cont
835 n_params = length bndrs
836 (bndrs, body) = collectBinders expr
837 zap | n_args >= n_params = \b -> b
838 | otherwise = \b -> if isTyVar b then b
840 -- NB: we count all the args incl type args
841 -- so we must count all the binders (incl type lambdas)
843 simplExprF' env (Type ty) cont
844 = ASSERT( contIsRhsOrArg cont )
845 do { ty' <- simplCoercion env ty
846 ; rebuild env (Type ty') cont }
848 simplExprF' env (Case scrut bndr _ alts) cont
849 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
850 = -- Simplify the scrutinee with a Select continuation
851 simplExprF env scrut (Select NoDup bndr alts env cont)
854 = -- If case-of-case is off, simply simplify the case expression
855 -- in a vanilla Stop context, and rebuild the result around it
856 do { case_expr' <- simplExprC env scrut case_cont
857 ; rebuild env case_expr' cont }
859 case_cont = Select NoDup bndr alts env mkBoringStop
861 simplExprF' env (Let (Rec pairs) body) cont
862 = do { env' <- simplRecBndrs env (map fst pairs)
863 -- NB: bndrs' don't have unfoldings or rules
864 -- We add them as we go down
866 ; env'' <- simplRecBind env' NotTopLevel pairs
867 ; simplExprF env'' body cont }
869 simplExprF' env (Let (NonRec bndr rhs) body) cont
870 = simplNonRecE env bndr (rhs, env) ([], body) cont
872 ---------------------------------
873 simplType :: SimplEnv -> InType -> SimplM OutType
874 -- Kept monadic just so we can do the seqType
876 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
877 seqType new_ty `seq` return new_ty
879 new_ty = substTy env ty
881 ---------------------------------
882 simplCoercion :: SimplEnv -> InType -> SimplM OutType
883 -- The InType isn't *necessarily* a coercion, but it might be
884 -- (in a type application, say) and optCoercion is a no-op on types
886 = do { co' <- simplType env co
887 ; return (optCoercion co') }
891 %************************************************************************
893 \subsection{The main rebuilder}
895 %************************************************************************
898 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
899 -- At this point the substitution in the SimplEnv should be irrelevant
900 -- only the in-scope set and floats should matter
901 rebuild env expr cont0
902 = -- pprTrace "rebuild" (ppr expr $$ ppr cont0 $$ ppr (seFloats env)) $
904 Stop {} -> return (env, expr)
905 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
906 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
907 StrictArg info _ cont -> rebuildCall env (info `addArgTo` expr) cont
908 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
909 ; simplLam env' bs body cont }
910 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
911 ; rebuild env (App expr arg') cont }
915 %************************************************************************
919 %************************************************************************
922 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
923 -> SimplM (SimplEnv, OutExpr)
924 simplCast env body co0 cont0
925 = do { co1 <- simplCoercion env co0
926 ; simplExprF env body (addCoerce co1 cont0) }
928 addCoerce co cont = add_coerce co (coercionKind co) cont
930 add_coerce _co (s1, k1) cont -- co :: ty~ty
931 | s1 `coreEqType` k1 = cont -- is a no-op
933 add_coerce co1 (s1, _k2) (CoerceIt co2 cont)
934 | (_l1, t1) <- coercionKind co2
935 -- e |> (g1 :: S1~L) |> (g2 :: L~T1)
938 -- e |> (g1 . g2 :: S1~T1) otherwise
940 -- For example, in the initial form of a worker
941 -- we may find (coerce T (coerce S (\x.e))) y
942 -- and we'd like it to simplify to e[y/x] in one round
944 , s1 `coreEqType` t1 = cont -- The coerces cancel out
945 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
947 add_coerce co (s1s2, _t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
948 -- (f |> g) ty ---> (f ty) |> (g @ ty)
949 -- This implements the PushT and PushC rules from the paper
950 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
952 (new_arg_ty, new_cast)
953 | isCoVar tyvar = (new_arg_co, mkCselRCoercion co) -- PushC rule
954 | otherwise = (ty', mkInstCoercion co ty') -- PushT rule
956 ApplyTo dup (Type new_arg_ty) (zapSubstEnv arg_se) (addCoerce new_cast cont)
958 ty' = substTy (arg_se `setInScope` env) arg_ty
959 new_arg_co = mkCsel1Coercion co `mkTransCoercion`
960 ty' `mkTransCoercion`
961 mkSymCoercion (mkCsel2Coercion co)
963 add_coerce co (s1s2, _t1t2) (ApplyTo dup arg arg_se cont)
964 | not (isTypeArg arg) -- This implements the Push rule from the paper
965 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
966 -- (e |> (g :: s1s2 ~ t1->t2)) f
968 -- (e (f |> (arg g :: t1~s1))
969 -- |> (res g :: s2->t2)
971 -- t1t2 must be a function type, t1->t2, because it's applied
972 -- to something but s1s2 might conceivably not be
974 -- When we build the ApplyTo we can't mix the out-types
975 -- with the InExpr in the argument, so we simply substitute
976 -- to make it all consistent. It's a bit messy.
977 -- But it isn't a common case.
979 -- Example of use: Trac #995
980 = ApplyTo dup new_arg (zapSubstEnv arg_se) (addCoerce co2 cont)
982 -- we split coercion t1->t2 ~ s1->s2 into t1 ~ s1 and
983 -- t2 ~ s2 with left and right on the curried form:
984 -- (->) t1 t2 ~ (->) s1 s2
985 [co1, co2] = decomposeCo 2 co
986 new_arg = mkCoerce (mkSymCoercion co1) arg'
987 arg' = substExpr (arg_se `setInScope` env) arg
989 add_coerce co _ cont = CoerceIt co cont
993 %************************************************************************
997 %************************************************************************
1000 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
1001 -> SimplM (SimplEnv, OutExpr)
1003 simplLam env [] body cont = simplExprF env body cont
1006 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
1007 = do { tick (BetaReduction bndr)
1008 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
1010 -- Not enough args, so there are real lambdas left to put in the result
1011 simplLam env bndrs body cont
1012 = do { (env', bndrs') <- simplLamBndrs env bndrs
1013 ; body' <- simplExpr env' body
1014 ; new_lam <- mkLam env' bndrs' body'
1015 ; rebuild env' new_lam cont }
1018 simplNonRecE :: SimplEnv
1019 -> InBndr -- The binder
1020 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
1021 -> ([InBndr], InExpr) -- Body of the let/lambda
1024 -> SimplM (SimplEnv, OutExpr)
1026 -- simplNonRecE is used for
1027 -- * non-top-level non-recursive lets in expressions
1030 -- It deals with strict bindings, via the StrictBind continuation,
1031 -- which may abort the whole process
1033 -- The "body" of the binding comes as a pair of ([InId],InExpr)
1034 -- representing a lambda; so we recurse back to simplLam
1035 -- Why? Because of the binder-occ-info-zapping done before
1036 -- the call to simplLam in simplExprF (Lam ...)
1038 -- First deal with type applications and type lets
1039 -- (/\a. e) (Type ty) and (let a = Type ty in e)
1040 simplNonRecE env bndr (Type ty_arg, rhs_se) (bndrs, body) cont
1041 = ASSERT( isTyVar bndr )
1042 do { ty_arg' <- simplType (rhs_se `setInScope` env) ty_arg
1043 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
1045 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
1046 | preInlineUnconditionally env NotTopLevel bndr rhs
1047 = do { tick (PreInlineUnconditionally bndr)
1048 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
1051 = do { simplExprF (rhs_se `setFloats` env) rhs
1052 (StrictBind bndr bndrs body env cont) }
1055 = ASSERT( not (isTyVar bndr) )
1056 do { (env1, bndr1) <- simplNonRecBndr env bndr
1057 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
1058 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
1059 ; simplLam env3 bndrs body cont }
1063 %************************************************************************
1067 %************************************************************************
1070 -- Hack alert: we only distinguish subsumed cost centre stacks for the
1071 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
1072 simplNote :: SimplEnv -> Note -> CoreExpr -> SimplCont
1073 -> SimplM (SimplEnv, OutExpr)
1074 simplNote env (SCC cc) e cont
1075 | pushCCisNop cc (getEnclosingCC env) -- scc "f" (...(scc "f" e)...)
1076 = simplExprF env e cont -- ==> scc "f" (...e...)
1078 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
1079 ; rebuild env (mkSCC cc e') cont }
1081 simplNote env (CoreNote s) e cont
1082 = do { e' <- simplExpr env e
1083 ; rebuild env (Note (CoreNote s) e') cont }
1087 %************************************************************************
1089 \subsection{Dealing with calls}
1091 %************************************************************************
1094 simplVar :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1095 simplVar env var cont
1096 = case substId env var of
1097 DoneEx e -> simplExprF (zapSubstEnv env) e cont
1098 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
1099 DoneId var1 -> completeCall env var1 cont
1100 -- Note [zapSubstEnv]
1101 -- The template is already simplified, so don't re-substitute.
1102 -- This is VITAL. Consider
1104 -- let y = \z -> ...x... in
1106 -- We'll clone the inner \x, adding x->x' in the id_subst
1107 -- Then when we inline y, we must *not* replace x by x' in
1108 -- the inlined copy!!
1110 ---------------------------------------------------------
1111 -- Dealing with a call site
1113 completeCall :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1114 completeCall env var cont
1115 = do { ------------- Try inlining ----------------
1116 dflags <- getDOptsSmpl
1117 ; let (args,call_cont) = contArgs cont
1118 -- The args are OutExprs, obtained by *lazily* substituting
1119 -- in the args found in cont. These args are only examined
1120 -- to limited depth (unless a rule fires). But we must do
1121 -- the substitution; rule matching on un-simplified args would
1124 arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1125 n_val_args = length arg_infos
1126 interesting_cont = interestingCallContext call_cont
1127 unfolding = activeUnfolding env var
1128 maybe_inline = callSiteInline dflags var unfolding
1129 (null args) arg_infos interesting_cont
1130 ; case maybe_inline of {
1131 Just unfolding -- There is an inlining!
1132 -> do { tick (UnfoldingDone var)
1133 ; (if dopt Opt_D_dump_inlinings dflags then
1134 pprTrace ("Inlining done: " ++ showSDoc (ppr var)) (vcat [
1135 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1136 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1137 text "Cont: " <+> ppr call_cont])
1140 simplExprF (zapSubstEnv env) unfolding cont }
1142 ; Nothing -> do -- No inlining!
1144 { rule_base <- getSimplRules
1145 ; let info = mkArgInfo var (getRules rule_base var) n_val_args call_cont
1146 ; rebuildCall env info cont
1149 rebuildCall :: SimplEnv
1152 -> SimplM (SimplEnv, OutExpr)
1153 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_strs = [] }) cont
1154 -- When we run out of strictness args, it means
1155 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1156 -- Then we want to discard the entire strict continuation. E.g.
1157 -- * case (error "hello") of { ... }
1158 -- * (error "Hello") arg
1159 -- * f (error "Hello") where f is strict
1161 -- Then, especially in the first of these cases, we'd like to discard
1162 -- the continuation, leaving just the bottoming expression. But the
1163 -- type might not be right, so we may have to add a coerce.
1164 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1165 = return (env, mk_coerce res) -- contination to discard, else we do it
1166 where -- again and again!
1167 res = mkApps (Var fun) (reverse rev_args)
1168 res_ty = exprType res
1169 cont_ty = contResultType env res_ty cont
1170 co = mkUnsafeCoercion res_ty cont_ty
1171 mk_coerce expr | cont_ty `coreEqType` res_ty = expr
1172 | otherwise = mkCoerce co expr
1174 rebuildCall env info (ApplyTo _ (Type arg_ty) se cont)
1175 = do { ty' <- simplCoercion (se `setInScope` env) arg_ty
1176 ; rebuildCall env (info `addArgTo` Type ty') cont }
1178 rebuildCall env info@(ArgInfo { ai_encl = encl_rules
1179 , ai_strs = str:strs, ai_discs = disc:discs })
1180 (ApplyTo _ arg arg_se cont)
1181 | str -- Strict argument
1182 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1183 simplExprF (arg_se `setFloats` env) arg
1184 (StrictArg info' cci cont)
1187 | otherwise -- Lazy argument
1188 -- DO NOT float anything outside, hence simplExprC
1189 -- There is no benefit (unlike in a let-binding), and we'd
1190 -- have to be very careful about bogus strictness through
1191 -- floating a demanded let.
1192 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1194 ; rebuildCall env (addArgTo info' arg') cont }
1196 info' = info { ai_strs = strs, ai_discs = discs }
1197 cci | encl_rules || disc > 0 = ArgCtxt encl_rules -- Be keener here
1198 | otherwise = BoringCtxt -- Nothing interesting
1200 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_rules = rules }) cont
1201 = do { -- We've accumulated a simplified call in <fun,rev_args>
1202 -- so try rewrite rules; see Note [RULEs apply to simplified arguments]
1203 -- See also Note [Rules for recursive functions]
1204 ; let args = reverse rev_args
1205 env' = zapSubstEnv env
1206 ; mb_rule <- tryRules env rules fun args cont
1208 Just (n_args, rule_rhs) -> simplExprF env' rule_rhs $
1209 pushArgs env' (drop n_args args) cont ;
1210 -- n_args says how many args the rule consumed
1211 ; Nothing -> rebuild env (mkApps (Var fun) args) cont -- No rules
1215 Note [RULES apply to simplified arguments]
1216 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1217 It's very desirable to try RULES once the arguments have been simplified, because
1218 doing so ensures that rule cascades work in one pass. Consider
1219 {-# RULES g (h x) = k x
1222 Then we want to rewrite (g (h x)) to (k x) and only then try f's rules. If
1223 we match f's rules against the un-simplified RHS, it won't match. This
1224 makes a particularly big difference when superclass selectors are involved:
1225 op ($p1 ($p2 (df d)))
1226 We want all this to unravel in one sweeep.
1230 This part of the simplifier may break the no-shadowing invariant
1232 f (...(\a -> e)...) (case y of (a,b) -> e')
1233 where f is strict in its second arg
1234 If we simplify the innermost one first we get (...(\a -> e)...)
1235 Simplifying the second arg makes us float the case out, so we end up with
1236 case y of (a,b) -> f (...(\a -> e)...) e'
1237 So the output does not have the no-shadowing invariant. However, there is
1238 no danger of getting name-capture, because when the first arg was simplified
1239 we used an in-scope set that at least mentioned all the variables free in its
1240 static environment, and that is enough.
1242 We can't just do innermost first, or we'd end up with a dual problem:
1243 case x of (a,b) -> f e (...(\a -> e')...)
1245 I spent hours trying to recover the no-shadowing invariant, but I just could
1246 not think of an elegant way to do it. The simplifier is already knee-deep in
1247 continuations. We have to keep the right in-scope set around; AND we have
1248 to get the effect that finding (error "foo") in a strict arg position will
1249 discard the entire application and replace it with (error "foo"). Getting
1250 all this at once is TOO HARD!
1253 %************************************************************************
1257 %************************************************************************
1260 tryRules :: SimplEnv -> [CoreRule]
1261 -> Id -> [OutExpr] -> SimplCont
1262 -> SimplM (Maybe (Arity, CoreExpr)) -- The arity is the number of
1263 -- args consumed by the rule
1264 tryRules env rules fn args call_cont
1268 = do { dflags <- getDOptsSmpl
1269 ; case activeRule dflags env of {
1270 Nothing -> return Nothing ; -- No rules apply
1272 case lookupRule act_fn (activeUnfInRule env) (getInScope env) fn args rules of {
1273 Nothing -> return Nothing ; -- No rule matches
1274 Just (rule, rule_rhs) ->
1276 do { tick (RuleFired (ru_name rule))
1277 ; (if dopt Opt_D_dump_rule_firings dflags then
1278 pprTrace "Rule fired" (vcat [
1279 text "Rule:" <+> ftext (ru_name rule),
1280 text "Before:" <+> ppr fn <+> sep (map pprParendExpr args),
1281 text "After: " <+> pprCoreExpr rule_rhs,
1282 text "Cont: " <+> ppr call_cont])
1285 return (Just (ruleArity rule, rule_rhs)) }}}}
1288 Note [Rules for recursive functions]
1289 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1290 You might think that we shouldn't apply rules for a loop breaker:
1291 doing so might give rise to an infinite loop, because a RULE is
1292 rather like an extra equation for the function:
1293 RULE: f (g x) y = x+y
1296 But it's too drastic to disable rules for loop breakers.
1297 Even the foldr/build rule would be disabled, because foldr
1298 is recursive, and hence a loop breaker:
1299 foldr k z (build g) = g k z
1300 So it's up to the programmer: rules can cause divergence
1303 %************************************************************************
1305 Rebuilding a cse expression
1307 %************************************************************************
1309 Note [Case elimination]
1310 ~~~~~~~~~~~~~~~~~~~~~~~
1311 The case-elimination transformation discards redundant case expressions.
1312 Start with a simple situation:
1314 case x# of ===> e[x#/y#]
1317 (when x#, y# are of primitive type, of course). We can't (in general)
1318 do this for algebraic cases, because we might turn bottom into
1321 The code in SimplUtils.prepareAlts has the effect of generalise this
1322 idea to look for a case where we're scrutinising a variable, and we
1323 know that only the default case can match. For example:
1327 DEFAULT -> ...(case x of
1331 Here the inner case is first trimmed to have only one alternative, the
1332 DEFAULT, after which it's an instance of the previous case. This
1333 really only shows up in eliminating error-checking code.
1335 We also make sure that we deal with this very common case:
1340 Here we are using the case as a strict let; if x is used only once
1341 then we want to inline it. We have to be careful that this doesn't
1342 make the program terminate when it would have diverged before, so we
1344 - e is already evaluated (it may so if e is a variable)
1345 - x is used strictly, or
1347 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1349 case e of ===> case e of DEFAULT -> r
1353 Now again the case may be elminated by the CaseElim transformation.
1356 Further notes about case elimination
1357 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1358 Consider: test :: Integer -> IO ()
1361 Turns out that this compiles to:
1364 eta1 :: State# RealWorld ->
1365 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1367 (PrelNum.jtos eta ($w[] @ Char))
1369 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1371 Notice the strange '<' which has no effect at all. This is a funny one.
1372 It started like this:
1374 f x y = if x < 0 then jtos x
1375 else if y==0 then "" else jtos x
1377 At a particular call site we have (f v 1). So we inline to get
1379 if v < 0 then jtos x
1380 else if 1==0 then "" else jtos x
1382 Now simplify the 1==0 conditional:
1384 if v<0 then jtos v else jtos v
1386 Now common-up the two branches of the case:
1388 case (v<0) of DEFAULT -> jtos v
1390 Why don't we drop the case? Because it's strict in v. It's technically
1391 wrong to drop even unnecessary evaluations, and in practice they
1392 may be a result of 'seq' so we *definitely* don't want to drop those.
1393 I don't really know how to improve this situation.
1396 ---------------------------------------------------------
1397 -- Eliminate the case if possible
1399 rebuildCase, reallyRebuildCase
1401 -> OutExpr -- Scrutinee
1402 -> InId -- Case binder
1403 -> [InAlt] -- Alternatives (inceasing order)
1405 -> SimplM (SimplEnv, OutExpr)
1407 --------------------------------------------------
1408 -- 1. Eliminate the case if there's a known constructor
1409 --------------------------------------------------
1411 rebuildCase env scrut case_bndr alts cont
1412 | Lit lit <- scrut -- No need for same treatment as constructors
1413 -- because literals are inlined more vigorously
1414 = do { tick (KnownBranch case_bndr)
1415 ; case findAlt (LitAlt lit) alts of
1416 Nothing -> missingAlt env case_bndr alts cont
1417 Just (_, bs, rhs) -> simple_rhs bs rhs }
1419 | Just (con, ty_args, other_args) <- exprIsConApp_maybe (activeUnfInRule env) scrut
1420 -- Works when the scrutinee is a variable with a known unfolding
1421 -- as well as when it's an explicit constructor application
1422 = do { tick (KnownBranch case_bndr)
1423 ; case findAlt (DataAlt con) alts of
1424 Nothing -> missingAlt env case_bndr alts cont
1425 Just (DEFAULT, bs, rhs) -> simple_rhs bs rhs
1426 Just (_, bs, rhs) -> knownCon env scrut con ty_args other_args
1427 case_bndr bs rhs cont
1430 simple_rhs bs rhs = ASSERT( null bs )
1431 do { env' <- simplNonRecX env case_bndr scrut
1432 ; simplExprF env' rhs cont }
1435 --------------------------------------------------
1436 -- 2. Eliminate the case if scrutinee is evaluated
1437 --------------------------------------------------
1439 rebuildCase env scrut case_bndr [(_, bndrs, rhs)] cont
1440 -- See if we can get rid of the case altogether
1441 -- See Note [Case eliminiation]
1442 -- mkCase made sure that if all the alternatives are equal,
1443 -- then there is now only one (DEFAULT) rhs
1444 | all isDeadBinder bndrs -- bndrs are [InId]
1446 -- Check that the scrutinee can be let-bound instead of case-bound
1447 , exprOkForSpeculation scrut
1448 -- OK not to evaluate it
1449 -- This includes things like (==# a# b#)::Bool
1450 -- so that we simplify
1451 -- case ==# a# b# of { True -> x; False -> x }
1454 -- This particular example shows up in default methods for
1455 -- comparision operations (e.g. in (>=) for Int.Int32)
1456 || exprIsHNF scrut -- It's already evaluated
1457 || var_demanded_later scrut -- It'll be demanded later
1459 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1460 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1461 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1462 -- its argument: case x of { y -> dataToTag# y }
1463 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1464 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1466 -- Also we don't want to discard 'seq's
1467 = do { tick (CaseElim case_bndr)
1468 ; env' <- simplNonRecX env case_bndr scrut
1469 ; simplExprF env' rhs cont }
1471 -- The case binder is going to be evaluated later,
1472 -- and the scrutinee is a simple variable
1473 var_demanded_later (Var v) = isStrictDmd (idDemandInfo case_bndr)
1474 && not (isTickBoxOp v)
1475 -- ugly hack; covering this case is what
1476 -- exprOkForSpeculation was intended for.
1477 var_demanded_later _ = False
1479 --------------------------------------------------
1480 -- 3. Try seq rules; see Note [User-defined RULES for seq] in MkId
1481 --------------------------------------------------
1483 rebuildCase env scrut case_bndr alts@[(_, bndrs, rhs)] cont
1484 | all isDeadBinder (case_bndr : bndrs) -- So this is just 'seq'
1485 = do { let rhs' = substExpr env rhs
1486 out_args = [Type (substTy env (idType case_bndr)),
1487 Type (exprType rhs'), scrut, rhs']
1488 -- Lazily evaluated, so we don't do most of this
1490 ; rule_base <- getSimplRules
1491 ; mb_rule <- tryRules env (getRules rule_base seqId) seqId out_args cont
1493 Just (n_args, res) -> simplExprF (zapSubstEnv env)
1494 (mkApps res (drop n_args out_args))
1496 Nothing -> reallyRebuildCase env scrut case_bndr alts cont }
1498 rebuildCase env scrut case_bndr alts cont
1499 = reallyRebuildCase env scrut case_bndr alts cont
1501 --------------------------------------------------
1502 -- 3. Catch-all case
1503 --------------------------------------------------
1505 reallyRebuildCase env scrut case_bndr alts cont
1506 = do { -- Prepare the continuation;
1507 -- The new subst_env is in place
1508 (env', dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1510 -- Simplify the alternatives
1511 ; (scrut', case_bndr', alts') <- simplAlts env' scrut case_bndr alts dup_cont
1513 -- Check for empty alternatives
1514 ; if null alts' then missingAlt env case_bndr alts cont
1516 { dflags <- getDOptsSmpl
1517 ; case_expr <- mkCase dflags scrut' case_bndr' alts'
1519 -- Notice that rebuild gets the in-scope set from env', not alt_env
1520 -- (which in any case is only build in simplAlts)
1521 -- The case binder *not* scope over the whole returned case-expression
1522 ; rebuild env' case_expr nodup_cont } }
1525 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1526 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1527 way, there's a chance that v will now only be used once, and hence
1530 Historical note: we use to do the "case binder swap" in the Simplifier
1531 so there were additional complications if the scrutinee was a variable.
1532 Now the binder-swap stuff is done in the occurrence analyer; see
1533 OccurAnal Note [Binder swap].
1537 If the case binder is not dead, then neither are the pattern bound
1539 case <any> of x { (a,b) ->
1540 case x of { (p,q) -> p } }
1541 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1542 The point is that we bring into the envt a binding
1544 after the outer case, and that makes (a,b) alive. At least we do unless
1545 the case binder is guaranteed dead.
1547 In practice, the scrutinee is almost always a variable, so we pretty
1548 much always zap the OccInfo of the binders. It doesn't matter much though.
1553 Consider case (v `cast` co) of x { I# y ->
1554 ... (case (v `cast` co) of {...}) ...
1555 We'd like to eliminate the inner case. We can get this neatly by
1556 arranging that inside the outer case we add the unfolding
1557 v |-> x `cast` (sym co)
1558 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1560 Note [Improving seq]
1563 type family F :: * -> *
1564 type instance F Int = Int
1566 ... case e of x { DEFAULT -> rhs } ...
1568 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1570 case e `cast` co of x'::Int
1571 I# x# -> let x = x' `cast` sym co
1574 so that 'rhs' can take advantage of the form of x'.
1576 Notice that Note [Case of cast] may then apply to the result.
1578 Nota Bene: We only do the [Improving seq] transformation if the
1579 case binder 'x' is actually used in the rhs; that is, if the case
1580 is *not* a *pure* seq.
1581 a) There is no point in adding the cast to a pure seq.
1582 b) There is a good reason not to: doing so would interfere
1583 with seq rules (Note [Built-in RULES for seq] in MkId).
1584 In particular, this [Improving seq] thing *adds* a cast
1585 while [Built-in RULES for seq] *removes* one, so they
1588 You might worry about
1589 case v of x { __DEFAULT ->
1590 ... case (v `cast` co) of y { I# -> ... }}
1591 This is a pure seq (since x is unused), so [Improving seq] won't happen.
1592 But it's ok: the simplifier will replace 'v' by 'x' in the rhs to get
1593 case v of x { __DEFAULT ->
1594 ... case (x `cast` co) of y { I# -> ... }}
1595 Now the outer case is not a pure seq, so [Improving seq] will happen,
1596 and then the inner case will disappear.
1598 The need for [Improving seq] showed up in Roman's experiments. Example:
1599 foo :: F Int -> Int -> Int
1600 foo t n = t `seq` bar n
1603 bar n = bar (n - case t of TI i -> i)
1604 Here we'd like to avoid repeated evaluating t inside the loop, by
1605 taking advantage of the `seq`.
1607 At one point I did transformation in LiberateCase, but it's more
1608 robust here. (Otherwise, there's a danger that we'll simply drop the
1609 'seq' altogether, before LiberateCase gets to see it.)
1612 simplAlts :: SimplEnv
1614 -> InId -- Case binder
1615 -> [InAlt] -- Non-empty
1617 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1618 -- Like simplExpr, this just returns the simplified alternatives;
1619 -- it does not return an environment
1621 simplAlts env scrut case_bndr alts cont'
1622 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1623 do { let env0 = zapFloats env
1625 ; (env1, case_bndr1) <- simplBinder env0 case_bndr
1627 ; fam_envs <- getFamEnvs
1628 ; (alt_env', scrut', case_bndr') <- improveSeq fam_envs env1 scrut
1629 case_bndr case_bndr1 alts
1631 ; (imposs_deflt_cons, in_alts) <- prepareAlts scrut' case_bndr' alts
1633 ; alts' <- mapM (simplAlt alt_env' imposs_deflt_cons case_bndr' cont') in_alts
1634 ; return (scrut', case_bndr', alts') }
1637 ------------------------------------
1638 improveSeq :: (FamInstEnv, FamInstEnv) -> SimplEnv
1639 -> OutExpr -> InId -> OutId -> [InAlt]
1640 -> SimplM (SimplEnv, OutExpr, OutId)
1641 -- Note [Improving seq]
1642 improveSeq fam_envs env scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1643 | not (isDeadBinder case_bndr) -- Not a pure seq! See the Note!
1644 , Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1645 = do { case_bndr2 <- newId (fsLit "nt") ty2
1646 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1647 env2 = extendIdSubst env case_bndr rhs
1648 ; return (env2, scrut `Cast` co, case_bndr2) }
1650 improveSeq _ env scrut _ case_bndr1 _
1651 = return (env, scrut, case_bndr1)
1654 ------------------------------------
1655 simplAlt :: SimplEnv
1656 -> [AltCon] -- These constructors can't be present when
1657 -- matching the DEFAULT alternative
1658 -> OutId -- The case binder
1663 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1664 = ASSERT( null bndrs )
1665 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1666 -- Record the constructors that the case-binder *can't* be.
1667 ; rhs' <- simplExprC env' rhs cont'
1668 ; return (DEFAULT, [], rhs') }
1670 simplAlt env _ case_bndr' cont' (LitAlt lit, bndrs, rhs)
1671 = ASSERT( null bndrs )
1672 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1673 ; rhs' <- simplExprC env' rhs cont'
1674 ; return (LitAlt lit, [], rhs') }
1676 simplAlt env _ case_bndr' cont' (DataAlt con, vs, rhs)
1677 = do { -- Deal with the pattern-bound variables
1678 -- Mark the ones that are in ! positions in the
1679 -- data constructor as certainly-evaluated.
1680 -- NB: simplLamBinders preserves this eval info
1681 let vs_with_evals = add_evals (dataConRepStrictness con)
1682 ; (env', vs') <- simplLamBndrs env vs_with_evals
1684 -- Bind the case-binder to (con args)
1685 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1686 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1687 env'' = addBinderUnfolding env' case_bndr'
1688 (mkConApp con con_args)
1690 ; rhs' <- simplExprC env'' rhs cont'
1691 ; return (DataAlt con, vs', rhs') }
1693 -- add_evals records the evaluated-ness of the bound variables of
1694 -- a case pattern. This is *important*. Consider
1695 -- data T = T !Int !Int
1697 -- case x of { T a b -> T (a+1) b }
1699 -- We really must record that b is already evaluated so that we don't
1700 -- go and re-evaluate it when constructing the result.
1701 -- See Note [Data-con worker strictness] in MkId.lhs
1706 go (v:vs') strs | isTyVar v = v : go vs' strs
1707 go (v:vs') (str:strs)
1708 | isMarkedStrict str = evald_v : go vs' strs
1709 | otherwise = zapped_v : go vs' strs
1711 zapped_v = zap_occ_info v
1712 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1713 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr the_strs)
1715 -- See Note [zapOccInfo]
1716 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1718 -- to the envt; so vs are now very much alive
1719 -- Note [Aug06] I can't see why this actually matters, but it's neater
1720 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1721 -- ==> case e of t { (a,b) -> ...(a)... }
1722 -- Look, Ma, a is alive now.
1723 zap_occ_info = zapCasePatIdOcc case_bndr'
1725 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1726 addBinderUnfolding env bndr rhs
1727 = modifyInScope env (bndr `setIdUnfolding` mkUnfolding False rhs)
1729 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1730 addBinderOtherCon env bndr cons
1731 = modifyInScope env (bndr `setIdUnfolding` mkOtherCon cons)
1733 zapCasePatIdOcc :: Id -> Id -> Id
1734 -- Consider case e of b { (a,b) -> ... }
1735 -- Then if we bind b to (a,b) in "...", and b is not dead,
1736 -- then we must zap the deadness info on a,b
1737 zapCasePatIdOcc case_bndr
1738 | isDeadBinder case_bndr = \ pat_id -> pat_id
1739 | otherwise = \ pat_id -> zapIdOccInfo pat_id
1743 %************************************************************************
1745 \subsection{Known constructor}
1747 %************************************************************************
1749 We are a bit careful with occurrence info. Here's an example
1751 (\x* -> case x of (a*, b) -> f a) (h v, e)
1753 where the * means "occurs once". This effectively becomes
1754 case (h v, e) of (a*, b) -> f a)
1756 let a* = h v; b = e in f a
1760 All this should happen in one sweep.
1763 knownCon :: SimplEnv
1764 -> OutExpr -- The scrutinee
1765 -> DataCon -> [OutType] -> [OutExpr] -- The scrutinee (in pieces)
1766 -> InId -> [InBndr] -> InExpr -- The alternative
1768 -> SimplM (SimplEnv, OutExpr)
1770 knownCon env scrut dc dc_ty_args dc_args bndr bs rhs cont
1771 = do { env' <- bind_args env bs dc_args
1773 -- It's useful to bind bndr to scrut, rather than to a fresh
1774 -- binding x = Con arg1 .. argn
1775 -- because very often the scrut is a variable, so we avoid
1776 -- creating, and then subsequently eliminating, a let-binding
1777 -- BUT, if scrut is a not a variable, we must be careful
1778 -- about duplicating the arg redexes; in that case, make
1779 -- a new con-app from the args
1780 bndr_rhs | exprIsTrivial scrut = scrut
1781 | otherwise = con_app
1782 con_app = Var (dataConWorkId dc)
1783 `mkTyApps` dc_ty_args
1784 `mkApps` [substExpr env' (varToCoreExpr b) | b <- bs]
1785 -- dc_ty_args are aready OutTypes, but bs are InBndrs
1787 ; env'' <- simplNonRecX env' bndr bndr_rhs
1788 ; simplExprF env'' rhs cont }
1790 zap_occ = zapCasePatIdOcc bndr -- bndr is an InId
1793 bind_args env' [] _ = return env'
1795 bind_args env' (b:bs') (Type ty : args)
1796 = ASSERT( isTyVar b )
1797 bind_args (extendTvSubst env' b ty) bs' args
1799 bind_args env' (b:bs') (arg : args)
1801 do { let b' = zap_occ b
1802 -- Note that the binder might be "dead", because it doesn't
1803 -- occur in the RHS; and simplNonRecX may therefore discard
1804 -- it via postInlineUnconditionally.
1805 -- Nevertheless we must keep it if the case-binder is alive,
1806 -- because it may be used in the con_app. See Note [zapOccInfo]
1807 ; env'' <- simplNonRecX env' b' arg
1808 ; bind_args env'' bs' args }
1811 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr dc_args $$
1812 text "scrut:" <+> ppr scrut
1815 missingAlt :: SimplEnv -> Id -> [InAlt] -> SimplCont -> SimplM (SimplEnv, OutExpr)
1816 -- This isn't strictly an error, although it is unusual.
1817 -- It's possible that the simplifer might "see" that
1818 -- an inner case has no accessible alternatives before
1819 -- it "sees" that the entire branch of an outer case is
1820 -- inaccessible. So we simply put an error case here instead.
1821 missingAlt env case_bndr alts cont
1822 = WARN( True, ptext (sLit "missingAlt") <+> ppr case_bndr )
1823 return (env, mkImpossibleExpr res_ty)
1825 res_ty = contResultType env (substTy env (coreAltsType alts)) cont
1829 %************************************************************************
1831 \subsection{Duplicating continuations}
1833 %************************************************************************
1836 prepareCaseCont :: SimplEnv
1837 -> [InAlt] -> SimplCont
1838 -> SimplM (SimplEnv, SimplCont,SimplCont)
1839 -- Return a duplicatable continuation, a non-duplicable part
1840 -- plus some extra bindings (that scope over the entire
1843 -- No need to make it duplicatable if there's only one alternative
1844 prepareCaseCont env [_] cont = return (env, cont, mkBoringStop)
1845 prepareCaseCont env _ cont = mkDupableCont env cont
1849 mkDupableCont :: SimplEnv -> SimplCont
1850 -> SimplM (SimplEnv, SimplCont, SimplCont)
1852 mkDupableCont env cont
1853 | contIsDupable cont
1854 = return (env, cont, mkBoringStop)
1856 mkDupableCont _ (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1858 mkDupableCont env (CoerceIt ty cont)
1859 = do { (env', dup, nodup) <- mkDupableCont env cont
1860 ; return (env', CoerceIt ty dup, nodup) }
1862 mkDupableCont env cont@(StrictBind {})
1863 = return (env, mkBoringStop, cont)
1864 -- See Note [Duplicating StrictBind]
1866 mkDupableCont env (StrictArg info cci cont)
1867 -- See Note [Duplicating StrictArg]
1868 = do { (env', dup, nodup) <- mkDupableCont env cont
1869 ; (env'', args') <- mapAccumLM makeTrivial env' (ai_args info)
1870 ; return (env'', StrictArg (info { ai_args = args' }) cci dup, nodup) }
1872 mkDupableCont env (ApplyTo _ arg se cont)
1873 = -- e.g. [...hole...] (...arg...)
1875 -- let a = ...arg...
1876 -- in [...hole...] a
1877 do { (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1878 ; arg' <- simplExpr (se `setInScope` env') arg
1879 ; (env'', arg'') <- makeTrivial env' arg'
1880 ; let app_cont = ApplyTo OkToDup arg'' (zapSubstEnv env'') dup_cont
1881 ; return (env'', app_cont, nodup_cont) }
1883 mkDupableCont env cont@(Select _ case_bndr [(_, bs, _rhs)] _ _)
1884 -- See Note [Single-alternative case]
1885 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1886 -- | not (isDeadBinder case_bndr)
1887 | all isDeadBinder bs -- InIds
1888 && not (isUnLiftedType (idType case_bndr))
1889 -- Note [Single-alternative-unlifted]
1890 = return (env, mkBoringStop, cont)
1892 mkDupableCont env (Select _ case_bndr alts se cont)
1893 = -- e.g. (case [...hole...] of { pi -> ei })
1895 -- let ji = \xij -> ei
1896 -- in case [...hole...] of { pi -> ji xij }
1897 do { tick (CaseOfCase case_bndr)
1898 ; (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1899 -- NB: call mkDupableCont here, *not* prepareCaseCont
1900 -- We must make a duplicable continuation, whereas prepareCaseCont
1901 -- doesn't when there is a single case branch
1903 ; let alt_env = se `setInScope` env'
1904 ; (alt_env', case_bndr') <- simplBinder alt_env case_bndr
1905 ; alts' <- mapM (simplAlt alt_env' [] case_bndr' dup_cont) alts
1906 -- Safe to say that there are no handled-cons for the DEFAULT case
1907 -- NB: simplBinder does not zap deadness occ-info, so
1908 -- a dead case_bndr' will still advertise its deadness
1909 -- This is really important because in
1910 -- case e of b { (# p,q #) -> ... }
1911 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1912 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1913 -- In the new alts we build, we have the new case binder, so it must retain
1915 -- NB: we don't use alt_env further; it has the substEnv for
1916 -- the alternatives, and we don't want that
1918 ; (env'', alts'') <- mkDupableAlts env' case_bndr' alts'
1919 ; return (env'', -- Note [Duplicated env]
1920 Select OkToDup case_bndr' alts'' (zapSubstEnv env'') mkBoringStop,
1924 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1925 -> SimplM (SimplEnv, [InAlt])
1926 -- Absorbs the continuation into the new alternatives
1928 mkDupableAlts env case_bndr' the_alts
1931 go env0 [] = return (env0, [])
1933 = do { (env1, alt') <- mkDupableAlt env0 case_bndr' alt
1934 ; (env2, alts') <- go env1 alts
1935 ; return (env2, alt' : alts' ) }
1937 mkDupableAlt :: SimplEnv -> OutId -> (AltCon, [CoreBndr], CoreExpr)
1938 -> SimplM (SimplEnv, (AltCon, [CoreBndr], CoreExpr))
1939 mkDupableAlt env case_bndr (con, bndrs', rhs')
1940 | exprIsDupable rhs' -- Note [Small alternative rhs]
1941 = return (env, (con, bndrs', rhs'))
1943 = do { let rhs_ty' = exprType rhs'
1944 scrut_ty = idType case_bndr
1947 DEFAULT -> case_bndr
1948 DataAlt dc -> setIdUnfolding case_bndr unf
1950 -- See Note [Case binders and join points]
1951 unf = mkInlineRule needSaturated rhs 0
1952 rhs = mkConApp dc (map Type (tyConAppArgs scrut_ty)
1953 ++ varsToCoreExprs bndrs')
1955 LitAlt {} -> WARN( True, ptext (sLit "mkDupableAlt")
1956 <+> ppr case_bndr <+> ppr con )
1958 -- The case binder is alive but trivial, so why has
1959 -- it not been substituted away?
1961 used_bndrs' | isDeadBinder case_bndr = filter abstract_over bndrs'
1962 | otherwise = bndrs' ++ [case_bndr_w_unf]
1965 | isTyVar bndr = True -- Abstract over all type variables just in case
1966 | otherwise = not (isDeadBinder bndr)
1967 -- The deadness info on the new Ids is preserved by simplBinders
1969 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1970 <- if (any isId used_bndrs')
1971 then return (used_bndrs', varsToCoreExprs used_bndrs')
1972 else do { rw_id <- newId (fsLit "w") realWorldStatePrimTy
1973 ; return ([rw_id], [Var realWorldPrimId]) }
1975 ; join_bndr <- newId (fsLit "$j") (mkPiTypes final_bndrs' rhs_ty')
1976 -- Note [Funky mkPiTypes]
1978 ; let -- We make the lambdas into one-shot-lambdas. The
1979 -- join point is sure to be applied at most once, and doing so
1980 -- prevents the body of the join point being floated out by
1981 -- the full laziness pass
1982 really_final_bndrs = map one_shot final_bndrs'
1983 one_shot v | isId v = setOneShotLambda v
1985 join_rhs = mkLams really_final_bndrs rhs'
1986 join_call = mkApps (Var join_bndr) final_args
1988 ; env' <- addPolyBind NotTopLevel env (NonRec join_bndr join_rhs)
1989 ; return (env', (con, bndrs', join_call)) }
1990 -- See Note [Duplicated env]
1993 Note [Case binders and join points]
1994 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1996 case (case .. ) of c {
1999 If we make a join point with c but not c# we get
2000 $j = \c -> ....c....
2002 But if later inlining scrutines the c, thus
2004 $j = \c -> ... case c of { I# y -> ... } ...
2006 we won't see that 'c' has already been scrutinised. This actually
2007 happens in the 'tabulate' function in wave4main, and makes a significant
2008 difference to allocation.
2010 An alternative plan is this:
2012 $j = \c# -> let c = I# c# in ...c....
2014 but that is bad if 'c' is *not* later scrutinised.
2016 So instead we do both: we pass 'c' and 'c#' , and record in c's inlining
2017 that it's really I# c#, thus
2019 $j = \c# -> \c[=I# c#] -> ...c....
2021 Absence analysis may later discard 'c'.
2024 Note [Duplicated env]
2025 ~~~~~~~~~~~~~~~~~~~~~
2026 Some of the alternatives are simplified, but have not been turned into a join point
2027 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
2028 bind the join point, because it might to do PostInlineUnconditionally, and
2029 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
2030 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
2031 at worst delays the join-point inlining.
2033 Note [Small alternative rhs]
2034 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2035 It is worth checking for a small RHS because otherwise we
2036 get extra let bindings that may cause an extra iteration of the simplifier to
2037 inline back in place. Quite often the rhs is just a variable or constructor.
2038 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
2039 iterations because the version with the let bindings looked big, and so wasn't
2040 inlined, but after the join points had been inlined it looked smaller, and so
2043 NB: we have to check the size of rhs', not rhs.
2044 Duplicating a small InAlt might invalidate occurrence information
2045 However, if it *is* dupable, we return the *un* simplified alternative,
2046 because otherwise we'd need to pair it up with an empty subst-env....
2047 but we only have one env shared between all the alts.
2048 (Remember we must zap the subst-env before re-simplifying something).
2049 Rather than do this we simply agree to re-simplify the original (small) thing later.
2051 Note [Funky mkPiTypes]
2052 ~~~~~~~~~~~~~~~~~~~~~~
2053 Notice the funky mkPiTypes. If the contructor has existentials
2054 it's possible that the join point will be abstracted over
2055 type varaibles as well as term variables.
2056 Example: Suppose we have
2057 data T = forall t. C [t]
2059 case (case e of ...) of
2061 We get the join point
2062 let j :: forall t. [t] -> ...
2063 j = /\t \xs::[t] -> rhs
2065 case (case e of ...) of
2066 C t xs::[t] -> j t xs
2068 Note [Join point abstaction]
2069 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2070 If we try to lift a primitive-typed something out
2071 for let-binding-purposes, we will *caseify* it (!),
2072 with potentially-disastrous strictness results. So
2073 instead we turn it into a function: \v -> e
2074 where v::State# RealWorld#. The value passed to this function
2075 is realworld#, which generates (almost) no code.
2077 There's a slight infelicity here: we pass the overall
2078 case_bndr to all the join points if it's used in *any* RHS,
2079 because we don't know its usage in each RHS separately
2081 We used to say "&& isUnLiftedType rhs_ty'" here, but now
2082 we make the join point into a function whenever used_bndrs'
2083 is empty. This makes the join-point more CPR friendly.
2084 Consider: let j = if .. then I# 3 else I# 4
2085 in case .. of { A -> j; B -> j; C -> ... }
2087 Now CPR doesn't w/w j because it's a thunk, so
2088 that means that the enclosing function can't w/w either,
2089 which is a lose. Here's the example that happened in practice:
2090 kgmod :: Int -> Int -> Int
2091 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
2095 I have seen a case alternative like this:
2097 It's a bit silly to add the realWorld dummy arg in this case, making
2100 (the \v alone is enough to make CPR happy) but I think it's rare
2102 Note [Duplicating StrictArg]
2103 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2104 The original plan had (where E is a big argument)
2106 ==> let $j = \a -> f E a
2109 But this is terrible! Here's an example:
2110 && E (case x of { T -> F; F -> T })
2111 Now, && is strict so we end up simplifying the case with
2112 an ArgOf continuation. If we let-bind it, we get
2113 let $j = \v -> && E v
2114 in simplExpr (case x of { T -> F; F -> T })
2116 And after simplifying more we get
2117 let $j = \v -> && E v
2118 in case x of { T -> $j F; F -> $j T }
2119 Which is a Very Bad Thing
2121 What we do now is this
2125 Now if the thing in the hole is a case expression (which is when
2126 we'll call mkDupableCont), we'll push the function call into the
2127 branches, which is what we want. Now RULES for f may fire, and
2128 call-pattern specialisation. Here's an example from Trac #3116
2131 _ -> Chunk p fpc (o+1) (l-1) bs')
2132 If we can push the call for 'go' inside the case, we get
2133 call-pattern specialisation for 'go', which is *crucial* for
2136 Here is the (&&) example:
2137 && E (case x of { T -> F; F -> T })
2139 case x of { T -> && a F; F -> && a T }
2143 * Arguments to f *after* the strict one are handled by
2144 the ApplyTo case of mkDupableCont. Eg
2147 * We can only do the let-binding of E because the function
2148 part of a StrictArg continuation is an explicit syntax
2149 tree. In earlier versions we represented it as a function
2150 (CoreExpr -> CoreEpxr) which we couldn't take apart.
2152 Do *not* duplicate StrictBind and StritArg continuations. We gain
2153 nothing by propagating them into the expressions, and we do lose a
2156 The desire not to duplicate is the entire reason that
2157 mkDupableCont returns a pair of continuations.
2159 Note [Duplicating StrictBind]
2160 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2161 Unlike StrictArg, there doesn't seem anything to gain from
2162 duplicating a StrictBind continuation, so we don't.
2164 The desire not to duplicate is the entire reason that
2165 mkDupableCont returns a pair of continuations.
2168 Note [Single-alternative cases]
2169 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2170 This case is just like the ArgOf case. Here's an example:
2174 case (case x of I# x' ->
2176 True -> I# (negate# x')
2177 False -> I# x') of y {
2179 Because the (case x) has only one alternative, we'll transform to
2181 case (case x' <# 0# of
2182 True -> I# (negate# x')
2183 False -> I# x') of y {
2185 But now we do *NOT* want to make a join point etc, giving
2187 let $j = \y -> MkT y
2189 True -> $j (I# (negate# x'))
2191 In this case the $j will inline again, but suppose there was a big
2192 strict computation enclosing the orginal call to MkT. Then, it won't
2193 "see" the MkT any more, because it's big and won't get duplicated.
2194 And, what is worse, nothing was gained by the case-of-case transform.
2196 When should use this case of mkDupableCont?
2197 However, matching on *any* single-alternative case is a *disaster*;
2198 e.g. case (case ....) of (a,b) -> (# a,b #)
2199 We must push the outer case into the inner one!
2202 * Match [(DEFAULT,_,_)], but in the common case of Int,
2203 the alternative-filling-in code turned the outer case into
2204 case (...) of y { I# _ -> MkT y }
2206 * Match on single alternative plus (not (isDeadBinder case_bndr))
2207 Rationale: pushing the case inwards won't eliminate the construction.
2208 But there's a risk of
2209 case (...) of y { (a,b) -> let z=(a,b) in ... }
2210 Now y looks dead, but it'll come alive again. Still, this
2211 seems like the best option at the moment.
2213 * Match on single alternative plus (all (isDeadBinder bndrs))
2214 Rationale: this is essentially seq.
2216 * Match when the rhs is *not* duplicable, and hence would lead to a
2217 join point. This catches the disaster-case above. We can test
2218 the *un-simplified* rhs, which is fine. It might get bigger or
2219 smaller after simplification; if it gets smaller, this case might
2220 fire next time round. NB also that we must test contIsDupable
2221 case_cont *btoo, because case_cont might be big!
2223 HOWEVER: I found that this version doesn't work well, because
2224 we can get let x = case (...) of { small } in ...case x...
2225 When x is inlined into its full context, we find that it was a bad
2226 idea to have pushed the outer case inside the (...) case.
2228 Note [Single-alternative-unlifted]
2229 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2230 Here's another single-alternative where we really want to do case-of-case:
2238 case y_s6X of tpl_s7m {
2239 M1.Mk1 ipv_s70 -> ipv_s70;
2240 M1.Mk2 ipv_s72 -> ipv_s72;
2246 case x_s74 of tpl_s7n {
2247 M1.Mk1 ipv_s77 -> ipv_s77;
2248 M1.Mk2 ipv_s79 -> ipv_s79;
2252 { __DEFAULT -> ==# [wild1_s7b wild_s7c];
2256 So the outer case is doing *nothing at all*, other than serving as a
2257 join-point. In this case we really want to do case-of-case and decide
2258 whether to use a real join point or just duplicate the continuation.
2260 Hence: check whether the case binder's type is unlifted, because then
2261 the outer case is *not* a seq.