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, isExternalName )
23 import OptCoercion ( optCoercion )
24 import FamInstEnv ( topNormaliseType )
25 import DataCon ( DataCon, dataConWorkId, dataConRepStrictness )
26 import CoreMonad ( SimplifierSwitch(..), Tick(..) )
28 import Demand ( isStrictDmd, splitStrictSig )
29 import PprCore ( pprParendExpr, pprCoreExpr )
30 import CoreUnfold ( mkUnfolding, mkCoreUnfolding, mkInlineRule,
31 exprIsConApp_maybe, callSiteInline, CallCtxt(..) )
33 import qualified CoreSubst
34 import CoreArity ( exprArity )
35 import Rules ( lookupRule, getRules )
36 import BasicTypes ( isMarkedStrict, Arity )
37 import CostCentre ( currentCCS, pushCCisNop )
38 import TysPrim ( realWorldStatePrimTy )
39 import PrelInfo ( realWorldPrimId )
40 import BasicTypes ( TopLevelFlag(..), isTopLevel, RecFlag(..) )
41 import MonadUtils ( foldlM, mapAccumLM )
42 import Maybes ( orElse )
43 import Data.List ( mapAccumL )
49 The guts of the simplifier is in this module, but the driver loop for
50 the simplifier is in SimplCore.lhs.
53 -----------------------------------------
54 *** IMPORTANT NOTE ***
55 -----------------------------------------
56 The simplifier used to guarantee that the output had no shadowing, but
57 it does not do so any more. (Actually, it never did!) The reason is
58 documented with simplifyArgs.
61 -----------------------------------------
62 *** IMPORTANT NOTE ***
63 -----------------------------------------
64 Many parts of the simplifier return a bunch of "floats" as well as an
65 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
67 All "floats" are let-binds, not case-binds, but some non-rec lets may
68 be unlifted (with RHS ok-for-speculation).
72 -----------------------------------------
73 ORGANISATION OF FUNCTIONS
74 -----------------------------------------
76 - simplify all top-level binders
77 - for NonRec, call simplRecOrTopPair
78 - for Rec, call simplRecBind
81 ------------------------------
82 simplExpr (applied lambda) ==> simplNonRecBind
83 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
84 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
86 ------------------------------
87 simplRecBind [binders already simplfied]
88 - use simplRecOrTopPair on each pair in turn
90 simplRecOrTopPair [binder already simplified]
91 Used for: recursive bindings (top level and nested)
92 top-level non-recursive bindings
94 - check for PreInlineUnconditionally
98 Used for: non-top-level non-recursive bindings
99 beta reductions (which amount to the same thing)
100 Because it can deal with strict arts, it takes a
101 "thing-inside" and returns an expression
103 - check for PreInlineUnconditionally
104 - simplify binder, including its IdInfo
113 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
114 Used for: binding case-binder and constr args in a known-constructor case
115 - check for PreInLineUnconditionally
119 ------------------------------
120 simplLazyBind: [binder already simplified, RHS not]
121 Used for: recursive bindings (top level and nested)
122 top-level non-recursive bindings
123 non-top-level, but *lazy* non-recursive bindings
124 [must not be strict or unboxed]
125 Returns floats + an augmented environment, not an expression
126 - substituteIdInfo and add result to in-scope
127 [so that rules are available in rec rhs]
130 - float if exposes constructor or PAP
134 completeNonRecX: [binder and rhs both simplified]
135 - if the the thing needs case binding (unlifted and not ok-for-spec)
141 completeBind: [given a simplified RHS]
142 [used for both rec and non-rec bindings, top level and not]
143 - try PostInlineUnconditionally
144 - add unfolding [this is the only place we add an unfolding]
149 Right hand sides and arguments
150 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
151 In many ways we want to treat
152 (a) the right hand side of a let(rec), and
153 (b) a function argument
154 in the same way. But not always! In particular, we would
155 like to leave these arguments exactly as they are, so they
156 will match a RULE more easily.
161 It's harder to make the rule match if we ANF-ise the constructor,
162 or eta-expand the PAP:
164 f (let { a = g x; b = h x } in (a,b))
167 On the other hand if we see the let-defns
172 then we *do* want to ANF-ise and eta-expand, so that p and q
173 can be safely inlined.
175 Even floating lets out is a bit dubious. For let RHS's we float lets
176 out if that exposes a value, so that the value can be inlined more vigorously.
179 r = let x = e in (x,x)
181 Here, if we float the let out we'll expose a nice constructor. We did experiments
182 that showed this to be a generally good thing. But it was a bad thing to float
183 lets out unconditionally, because that meant they got allocated more often.
185 For function arguments, there's less reason to expose a constructor (it won't
186 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
187 So for the moment we don't float lets out of function arguments either.
192 For eta expansion, we want to catch things like
194 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
196 If the \x was on the RHS of a let, we'd eta expand to bring the two
197 lambdas together. And in general that's a good thing to do. Perhaps
198 we should eta expand wherever we find a (value) lambda? Then the eta
199 expansion at a let RHS can concentrate solely on the PAP case.
202 %************************************************************************
204 \subsection{Bindings}
206 %************************************************************************
209 simplTopBinds :: SimplEnv -> [InBind] -> SimplM SimplEnv
211 simplTopBinds env0 binds0
212 = do { -- Put all the top-level binders into scope at the start
213 -- so that if a transformation rule has unexpectedly brought
214 -- anything into scope, then we don't get a complaint about that.
215 -- It's rather as if the top-level binders were imported.
216 ; env1 <- simplRecBndrs env0 (bindersOfBinds binds0)
217 ; dflags <- getDOptsSmpl
218 ; let dump_flag = dopt Opt_D_verbose_core2core dflags
219 ; env2 <- simpl_binds dump_flag env1 binds0
220 ; freeTick SimplifierDone
223 -- We need to track the zapped top-level binders, because
224 -- they should have their fragile IdInfo zapped (notably occurrence info)
225 -- That's why we run down binds and bndrs' simultaneously.
227 -- The dump-flag emits a trace for each top-level binding, which
228 -- helps to locate the tracing for inlining and rule firing
229 simpl_binds :: Bool -> SimplEnv -> [InBind] -> SimplM SimplEnv
230 simpl_binds _ env [] = return env
231 simpl_binds dump env (bind:binds) = do { env' <- trace_bind dump bind $
233 ; simpl_binds dump env' binds }
235 trace_bind True bind = pprTrace "SimplBind" (ppr (bindersOf bind))
236 trace_bind False _ = \x -> x
238 simpl_bind env (Rec pairs) = simplRecBind env TopLevel pairs
239 simpl_bind env (NonRec b r) = simplRecOrTopPair env' TopLevel b b' r
241 (env', b') = addBndrRules env b (lookupRecBndr env b)
245 %************************************************************************
247 \subsection{Lazy bindings}
249 %************************************************************************
251 simplRecBind is used for
252 * recursive bindings only
255 simplRecBind :: SimplEnv -> TopLevelFlag
258 simplRecBind env0 top_lvl pairs0
259 = do { let (env_with_info, triples) = mapAccumL add_rules env0 pairs0
260 ; env1 <- go (zapFloats env_with_info) triples
261 ; return (env0 `addRecFloats` env1) }
262 -- addFloats adds the floats from env1,
263 -- _and_ updates env0 with the in-scope set from env1
265 add_rules :: SimplEnv -> (InBndr,InExpr) -> (SimplEnv, (InBndr, OutBndr, InExpr))
266 -- Add the (substituted) rules to the binder
267 add_rules env (bndr, rhs) = (env', (bndr, bndr', rhs))
269 (env', bndr') = addBndrRules env bndr (lookupRecBndr env bndr)
271 go env [] = return env
273 go env ((old_bndr, new_bndr, rhs) : pairs)
274 = do { env' <- simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
278 simplOrTopPair is used for
279 * recursive bindings (whether top level or not)
280 * top-level non-recursive bindings
282 It assumes the binder has already been simplified, but not its IdInfo.
285 simplRecOrTopPair :: SimplEnv
287 -> InId -> OutBndr -> InExpr -- Binder and rhs
288 -> SimplM SimplEnv -- Returns an env that includes the binding
290 simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
291 | preInlineUnconditionally env top_lvl old_bndr rhs -- Check for unconditional inline
292 = do { tick (PreInlineUnconditionally old_bndr)
293 ; return (extendIdSubst env old_bndr (mkContEx env rhs)) }
296 = simplLazyBind env top_lvl Recursive old_bndr new_bndr rhs env
297 -- May not actually be recursive, but it doesn't matter
301 simplLazyBind is used for
302 * [simplRecOrTopPair] recursive bindings (whether top level or not)
303 * [simplRecOrTopPair] top-level non-recursive bindings
304 * [simplNonRecE] non-top-level *lazy* non-recursive bindings
307 1. It assumes that the binder is *already* simplified,
308 and is in scope, and its IdInfo too, except unfolding
310 2. It assumes that the binder type is lifted.
312 3. It does not check for pre-inline-unconditionallly;
313 that should have been done already.
316 simplLazyBind :: SimplEnv
317 -> TopLevelFlag -> RecFlag
318 -> InId -> OutId -- Binder, both pre-and post simpl
319 -- The OutId has IdInfo, except arity, unfolding
320 -> InExpr -> SimplEnv -- The RHS and its environment
323 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
324 = do { let rhs_env = rhs_se `setInScope` env
325 (tvs, body) = case collectTyBinders rhs of
326 (tvs, body) | not_lam body -> (tvs,body)
327 | otherwise -> ([], rhs)
328 not_lam (Lam _ _) = False
330 -- Do not do the "abstract tyyvar" thing if there's
331 -- a lambda inside, becuase it defeats eta-reduction
332 -- f = /\a. \x. g a x
335 ; (body_env, tvs') <- simplBinders rhs_env tvs
336 -- See Note [Floating and type abstraction] in SimplUtils
339 ; (body_env1, body1) <- simplExprF body_env body mkRhsStop
340 -- ANF-ise a constructor or PAP rhs
341 ; (body_env2, body2) <- prepareRhs body_env1 bndr1 body1
344 <- if not (doFloatFromRhs top_lvl is_rec False body2 body_env2)
345 then -- No floating, just wrap up!
346 do { rhs' <- mkLam env tvs' (wrapFloats body_env2 body2)
347 ; return (env, rhs') }
349 else if null tvs then -- Simple floating
350 do { tick LetFloatFromLet
351 ; return (addFloats env body_env2, body2) }
353 else -- Do type-abstraction first
354 do { tick LetFloatFromLet
355 ; (poly_binds, body3) <- abstractFloats tvs' body_env2 body2
356 ; rhs' <- mkLam env tvs' body3
357 ; env' <- foldlM (addPolyBind top_lvl) env poly_binds
358 ; return (env', rhs') }
360 ; completeBind env' top_lvl bndr bndr1 rhs' }
363 A specialised variant of simplNonRec used when the RHS is already simplified,
364 notably in knownCon. It uses case-binding where necessary.
367 simplNonRecX :: SimplEnv
368 -> InId -- Old binder
369 -> OutExpr -- Simplified RHS
372 simplNonRecX env bndr new_rhs
373 | isDeadBinder bndr -- Not uncommon; e.g. case (a,b) of b { (p,q) -> p }
374 = return env -- Here b is dead, and we avoid creating
375 | otherwise -- the binding b = (a,b)
376 = do { (env', bndr') <- simplBinder env bndr
377 ; completeNonRecX env' (isStrictId bndr) bndr bndr' new_rhs }
379 completeNonRecX :: SimplEnv
381 -> InId -- Old binder
382 -> OutId -- New binder
383 -> OutExpr -- Simplified RHS
386 completeNonRecX env is_strict old_bndr new_bndr new_rhs
387 = do { (env1, rhs1) <- prepareRhs (zapFloats env) new_bndr new_rhs
389 if doFloatFromRhs NotTopLevel NonRecursive is_strict rhs1 env1
390 then do { tick LetFloatFromLet
391 ; return (addFloats env env1, rhs1) } -- Add the floats to the main env
392 else return (env, wrapFloats env1 rhs1) -- Wrap the floats around the RHS
393 ; completeBind env2 NotTopLevel old_bndr new_bndr rhs2 }
396 {- No, no, no! Do not try preInlineUnconditionally in completeNonRecX
397 Doing so risks exponential behaviour, because new_rhs has been simplified once already
398 In the cases described by the folowing commment, postInlineUnconditionally will
399 catch many of the relevant cases.
400 -- This happens; for example, the case_bndr during case of
401 -- known constructor: case (a,b) of x { (p,q) -> ... }
402 -- Here x isn't mentioned in the RHS, so we don't want to
403 -- create the (dead) let-binding let x = (a,b) in ...
405 -- Similarly, single occurrences can be inlined vigourously
406 -- e.g. case (f x, g y) of (a,b) -> ....
407 -- If a,b occur once we can avoid constructing the let binding for them.
409 Furthermore in the case-binding case preInlineUnconditionally risks extra thunks
410 -- Consider case I# (quotInt# x y) of
411 -- I# v -> let w = J# v in ...
412 -- If we gaily inline (quotInt# x y) for v, we end up building an
414 -- let w = J# (quotInt# x y) in ...
415 -- because quotInt# can fail.
417 | preInlineUnconditionally env NotTopLevel bndr new_rhs
418 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
421 ----------------------------------
422 prepareRhs takes a putative RHS, checks whether it's a PAP or
423 constructor application and, if so, converts it to ANF, so that the
424 resulting thing can be inlined more easily. Thus
431 We also want to deal well cases like this
432 v = (f e1 `cast` co) e2
433 Here we want to make e1,e2 trivial and get
434 x1 = e1; x2 = e2; v = (f x1 `cast` co) v2
435 That's what the 'go' loop in prepareRhs does
438 prepareRhs :: SimplEnv -> OutId -> OutExpr -> SimplM (SimplEnv, OutExpr)
439 -- Adds new floats to the env iff that allows us to return a good RHS
440 prepareRhs env id (Cast rhs co) -- Note [Float coercions]
441 | (ty1, _ty2) <- coercionKind co -- Do *not* do this if rhs has an unlifted type
442 , not (isUnLiftedType ty1) -- see Note [Float coercions (unlifted)]
443 = do { (env', rhs') <- makeTrivialWithInfo env sanitised_info rhs
444 ; return (env', Cast rhs' co) }
446 sanitised_info = vanillaIdInfo `setStrictnessInfo` strictnessInfo info
447 `setDemandInfo` demandInfo info
450 prepareRhs env0 _ rhs0
451 = do { (_is_exp, env1, rhs1) <- go 0 env0 rhs0
452 ; return (env1, rhs1) }
454 go n_val_args env (Cast rhs co)
455 = do { (is_exp, env', rhs') <- go n_val_args env rhs
456 ; return (is_exp, env', Cast rhs' co) }
457 go n_val_args env (App fun (Type ty))
458 = do { (is_exp, env', rhs') <- go n_val_args env fun
459 ; return (is_exp, env', App rhs' (Type ty)) }
460 go n_val_args env (App fun arg)
461 = do { (is_exp, env', fun') <- go (n_val_args+1) env fun
463 True -> do { (env'', arg') <- makeTrivial env' arg
464 ; return (True, env'', App fun' arg') }
465 False -> return (False, env, App fun arg) }
466 go n_val_args env (Var fun)
467 = return (is_exp, env, Var fun)
469 is_exp = isExpandableApp fun n_val_args -- The fun a constructor or PAP
470 -- See Note [CONLIKE pragma] in BasicTypes
471 -- The definition of is_exp should match that in
472 -- OccurAnal.occAnalApp
475 = return (False, env, other)
479 Note [Float coercions]
480 ~~~~~~~~~~~~~~~~~~~~~~
481 When we find the binding
483 we'd like to transform it to
485 x = x `cast` co -- A trivial binding
486 There's a chance that e will be a constructor application or function, or something
487 like that, so moving the coerion to the usage site may well cancel the coersions
488 and lead to further optimisation. Example:
491 data instance T Int = T Int
493 foo :: Int -> Int -> Int
498 go n = case x of { T m -> go (n-m) }
499 -- This case should optimise
501 Note [Preserve strictness when floating coercions]
502 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
503 In the Note [Float coercions] transformation, keep the strictness info.
505 f = e `cast` co -- f has strictness SSL
507 f' = e -- f' also has strictness SSL
508 f = f' `cast` co -- f still has strictness SSL
510 Its not wrong to drop it on the floor, but better to keep it.
512 Note [Float coercions (unlifted)]
513 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
514 BUT don't do [Float coercions] if 'e' has an unlifted type.
517 foo :: Int = (error (# Int,Int #) "urk")
518 `cast` CoUnsafe (# Int,Int #) Int
520 If do the makeTrivial thing to the error call, we'll get
521 foo = case error (# Int,Int #) "urk" of v -> v `cast` ...
522 But 'v' isn't in scope!
524 These strange casts can happen as a result of case-of-case
525 bar = case (case x of { T -> (# 2,3 #); F -> error "urk" }) of
530 makeTrivial :: SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
531 -- Binds the expression to a variable, if it's not trivial, returning the variable
532 makeTrivial env expr = makeTrivialWithInfo env vanillaIdInfo expr
534 makeTrivialWithInfo :: SimplEnv -> IdInfo -> OutExpr -> SimplM (SimplEnv, OutExpr)
535 -- Propagate strictness and demand info to the new binder
536 -- Note [Preserve strictness when floating coercions]
537 makeTrivialWithInfo env info expr
540 | otherwise -- See Note [Take care] below
541 = do { uniq <- getUniqueM
542 ; let name = mkSystemVarName uniq (fsLit "a")
543 var = mkLocalIdWithInfo name (exprType expr) info
544 ; env' <- completeNonRecX env False var var expr
545 ; return (env', substExpr env' (Var var)) }
546 -- The substitution is needed becase we're constructing a new binding
548 -- And if rhs is of form (rhs1 |> co), then we might get
551 -- and now a's RHS is trivial and can be substituted out, and that
552 -- is what completeNonRecX will do
556 %************************************************************************
558 \subsection{Completing a lazy binding}
560 %************************************************************************
563 * deals only with Ids, not TyVars
564 * takes an already-simplified binder and RHS
565 * is used for both recursive and non-recursive bindings
566 * is used for both top-level and non-top-level bindings
568 It does the following:
569 - tries discarding a dead binding
570 - tries PostInlineUnconditionally
571 - add unfolding [this is the only place we add an unfolding]
574 It does *not* attempt to do let-to-case. Why? Because it is used for
575 - top-level bindings (when let-to-case is impossible)
576 - many situations where the "rhs" is known to be a WHNF
577 (so let-to-case is inappropriate).
579 Nor does it do the atomic-argument thing
582 completeBind :: SimplEnv
583 -> TopLevelFlag -- Flag stuck into unfolding
584 -> InId -- Old binder
585 -> OutId -> OutExpr -- New binder and RHS
587 -- completeBind may choose to do its work
588 -- * by extending the substitution (e.g. let x = y in ...)
589 -- * or by adding to the floats in the envt
591 completeBind env top_lvl old_bndr new_bndr new_rhs
592 = do { let old_info = idInfo old_bndr
593 old_unf = unfoldingInfo old_info
594 occ_info = occInfo old_info
596 ; new_unfolding <- simplUnfolding env top_lvl old_bndr occ_info new_rhs old_unf
598 ; if postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs new_unfolding
599 -- Inline and discard the binding
600 then do { tick (PostInlineUnconditionally old_bndr)
601 ; -- pprTrace "postInlineUnconditionally" (ppr old_bndr <+> equals <+> ppr new_rhs) $
602 return (extendIdSubst env old_bndr (DoneEx new_rhs)) }
603 -- Use the substitution to make quite, quite sure that the
604 -- substitution will happen, since we are going to discard the binding
606 else return (addNonRecWithUnf env new_bndr new_rhs new_unfolding) }
608 ------------------------------
609 addPolyBind :: TopLevelFlag -> SimplEnv -> OutBind -> SimplM SimplEnv
610 -- Add a new binding to the environment, complete with its unfolding
611 -- but *do not* do postInlineUnconditionally, because we have already
612 -- processed some of the scope of the binding
613 -- We still want the unfolding though. Consider
615 -- x = /\a. let y = ... in Just y
617 -- Then we float the y-binding out (via abstractFloats and addPolyBind)
618 -- but 'x' may well then be inlined in 'body' in which case we'd like the
619 -- opportunity to inline 'y' too.
621 addPolyBind top_lvl env (NonRec poly_id rhs)
622 = do { unfolding <- simplUnfolding env top_lvl poly_id NoOccInfo rhs noUnfolding
623 -- Assumes that poly_id did not have an INLINE prag
624 -- which is perhaps wrong. ToDo: think about this
625 ; return (addNonRecWithUnf env poly_id rhs unfolding) }
627 addPolyBind _ env bind@(Rec _) = return (extendFloats env bind)
628 -- Hack: letrecs are more awkward, so we extend "by steam"
629 -- without adding unfoldings etc. At worst this leads to
630 -- more simplifier iterations
632 ------------------------------
633 addNonRecWithUnf :: SimplEnv
634 -> OutId -> OutExpr -- New binder and RHS
635 -> Unfolding -- New unfolding
637 addNonRecWithUnf env new_bndr new_rhs new_unfolding
638 = let new_arity = exprArity new_rhs
639 old_arity = idArity new_bndr
640 info1 = idInfo new_bndr `setArityInfo` new_arity
642 -- Unfolding info: Note [Setting the new unfolding]
643 info2 = info1 `setUnfoldingInfo` new_unfolding
645 -- Demand info: Note [Setting the demand info]
646 info3 | isEvaldUnfolding new_unfolding = zapDemandInfo info2 `orElse` info2
649 final_id = new_bndr `setIdInfo` info3
650 dmd_arity = length $ fst $ splitStrictSig $ idStrictness new_bndr
652 ASSERT( isId new_bndr )
653 WARN( new_arity < old_arity || new_arity < dmd_arity,
654 (ptext (sLit "Arity decrease:") <+> ppr final_id <+> ppr old_arity
655 <+> ppr new_arity <+> ppr dmd_arity) )
656 -- Note [Arity decrease]
658 final_id `seq` -- This seq forces the Id, and hence its IdInfo,
659 -- and hence any inner substitutions
660 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
661 addNonRec env final_id new_rhs
662 -- The addNonRec adds it to the in-scope set too
664 ------------------------------
665 simplUnfolding :: SimplEnv-> TopLevelFlag
667 -> OccInfo -> OutExpr
668 -> Unfolding -> SimplM Unfolding
669 -- Note [Setting the new unfolding]
670 simplUnfolding env _ _ _ _ (DFunUnfolding con ops)
671 = return (DFunUnfolding con ops')
673 ops' = map (CoreSubst.substExpr (mkCoreSubst env)) ops
675 simplUnfolding env top_lvl id _ _
676 (CoreUnfolding { uf_tmpl = expr, uf_arity = arity
677 , uf_src = src, uf_guidance = guide })
678 | isInlineRuleSource src
679 = -- pprTrace "su" (vcat [ppr id, ppr act, ppr (getMode env), ppr (getMode rule_env)]) $
680 do { expr' <- simplExpr rule_env expr
681 ; let src' = CoreSubst.substUnfoldingSource (mkCoreSubst env) src
682 ; return (mkCoreUnfolding (isTopLevel top_lvl) src' expr' arity guide) }
683 -- See Note [Top-level flag on inline rules] in CoreUnfold
685 act = idInlineActivation id
686 rule_env = updMode (updModeForInlineRules act) env
687 -- See Note [Simplifying gently inside InlineRules] in SimplUtils
689 simplUnfolding _ top_lvl id _occ_info new_rhs _
690 = return (mkUnfolding (isTopLevel top_lvl) (isBottomingId id) new_rhs)
691 -- We make an unfolding *even for loop-breakers*.
692 -- Reason: (a) It might be useful to know that they are WHNF
693 -- (b) In TidyPgm we currently assume that, if we want to
694 -- expose the unfolding then indeed we *have* an unfolding
695 -- to expose. (We could instead use the RHS, but currently
696 -- we don't.) The simple thing is always to have one.
699 Note [Arity decrease]
700 ~~~~~~~~~~~~~~~~~~~~~
701 Generally speaking the arity of a binding should not decrease. But it *can*
702 legitimately happen becuase of RULES. Eg
704 where g has arity 2, will have arity 2. But if there's a rewrite rule
706 where h has arity 1, then f's arity will decrease. Here's a real-life example,
707 which is in the output of Specialise:
710 $dm {Arity 2} = \d.\x. op d
711 {-# RULES forall d. $dm Int d = $s$dm #-}
713 dInt = MkD .... opInt ...
714 opInt {Arity 1} = $dm dInt
716 $s$dm {Arity 0} = \x. op dInt }
718 Here opInt has arity 1; but when we apply the rule its arity drops to 0.
719 That's why Specialise goes to a little trouble to pin the right arity
720 on specialised functions too.
722 Note [Setting the new unfolding]
723 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
724 * If there's an INLINE pragma, we simplify the RHS gently. Maybe we
725 should do nothing at all, but simplifying gently might get rid of
728 * If not, we make an unfolding from the new RHS. But *only* for
729 non-loop-breakers. Making loop breakers not have an unfolding at all
730 means that we can avoid tests in exprIsConApp, for example. This is
731 important: if exprIsConApp says 'yes' for a recursive thing, then we
732 can get into an infinite loop
734 If there's an InlineRule on a loop breaker, we hang on to the inlining.
735 It's pretty dodgy, but the user did say 'INLINE'. May need to revisit
738 Note [Setting the demand info]
739 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
740 If the unfolding is a value, the demand info may
741 go pear-shaped, so we nuke it. Example:
743 case x of (p,q) -> h p q x
744 Here x is certainly demanded. But after we've nuked
745 the case, we'll get just
746 let x = (a,b) in h a b x
747 and now x is not demanded (I'm assuming h is lazy)
748 This really happens. Similarly
749 let f = \x -> e in ...f..f...
750 After inlining f at some of its call sites the original binding may
751 (for example) be no longer strictly demanded.
752 The solution here is a bit ad hoc...
755 %************************************************************************
757 \subsection[Simplify-simplExpr]{The main function: simplExpr}
759 %************************************************************************
761 The reason for this OutExprStuff stuff is that we want to float *after*
762 simplifying a RHS, not before. If we do so naively we get quadratic
763 behaviour as things float out.
765 To see why it's important to do it after, consider this (real) example:
779 a -- Can't inline a this round, cos it appears twice
783 Each of the ==> steps is a round of simplification. We'd save a
784 whole round if we float first. This can cascade. Consider
789 let f = let d1 = ..d.. in \y -> e
793 in \x -> ...(\y ->e)...
795 Only in this second round can the \y be applied, and it
796 might do the same again.
800 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
801 simplExpr env expr = simplExprC env expr mkBoringStop
803 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
804 -- Simplify an expression, given a continuation
805 simplExprC env expr cont
806 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
807 do { (env', expr') <- simplExprF (zapFloats env) expr cont
808 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
809 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
810 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
811 return (wrapFloats env' expr') }
813 --------------------------------------------------
814 simplExprF :: SimplEnv -> InExpr -> SimplCont
815 -> SimplM (SimplEnv, OutExpr)
817 simplExprF env e cont
818 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
819 simplExprF' env e cont
821 simplExprF' :: SimplEnv -> InExpr -> SimplCont
822 -> SimplM (SimplEnv, OutExpr)
823 simplExprF' env (Var v) cont = simplVar env v cont
824 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
825 simplExprF' env (Note n expr) cont = simplNote env n expr cont
826 simplExprF' env (Cast body co) cont = simplCast env body co cont
827 simplExprF' env (App fun arg) cont = simplExprF env fun $
828 ApplyTo NoDup arg env cont
830 simplExprF' env expr@(Lam _ _) cont
831 = simplLam env (map zap bndrs) body cont
832 -- The main issue here is under-saturated lambdas
833 -- (\x1. \x2. e) arg1
834 -- Here x1 might have "occurs-once" occ-info, because occ-info
835 -- is computed assuming that a group of lambdas is applied
836 -- all at once. If there are too few args, we must zap the
839 n_args = countArgs cont
840 n_params = length bndrs
841 (bndrs, body) = collectBinders expr
842 zap | n_args >= n_params = \b -> b
843 | otherwise = \b -> if isTyVar b then b
845 -- NB: we count all the args incl type args
846 -- so we must count all the binders (incl type lambdas)
848 simplExprF' env (Type ty) cont
849 = ASSERT( contIsRhsOrArg cont )
850 do { ty' <- simplCoercion env ty
851 ; rebuild env (Type ty') cont }
853 simplExprF' env (Case scrut bndr _ alts) cont
854 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
855 = -- Simplify the scrutinee with a Select continuation
856 simplExprF env scrut (Select NoDup bndr alts env cont)
859 = -- If case-of-case is off, simply simplify the case expression
860 -- in a vanilla Stop context, and rebuild the result around it
861 do { case_expr' <- simplExprC env scrut case_cont
862 ; rebuild env case_expr' cont }
864 case_cont = Select NoDup bndr alts env mkBoringStop
866 simplExprF' env (Let (Rec pairs) body) cont
867 = do { env' <- simplRecBndrs env (map fst pairs)
868 -- NB: bndrs' don't have unfoldings or rules
869 -- We add them as we go down
871 ; env'' <- simplRecBind env' NotTopLevel pairs
872 ; simplExprF env'' body cont }
874 simplExprF' env (Let (NonRec bndr rhs) body) cont
875 = simplNonRecE env bndr (rhs, env) ([], body) cont
877 ---------------------------------
878 simplType :: SimplEnv -> InType -> SimplM OutType
879 -- Kept monadic just so we can do the seqType
881 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
882 seqType new_ty `seq` return new_ty
884 new_ty = substTy env ty
886 ---------------------------------
887 simplCoercion :: SimplEnv -> InType -> SimplM OutType
888 -- The InType isn't *necessarily* a coercion, but it might be
889 -- (in a type application, say) and optCoercion is a no-op on types
891 = seqType new_co `seq` return new_co
893 new_co = optCoercion (getTvSubst env) co
897 %************************************************************************
899 \subsection{The main rebuilder}
901 %************************************************************************
904 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
905 -- At this point the substitution in the SimplEnv should be irrelevant
906 -- only the in-scope set and floats should matter
907 rebuild env expr cont0
908 = -- pprTrace "rebuild" (ppr expr $$ ppr cont0 $$ ppr (seFloats env)) $
910 Stop {} -> return (env, expr)
911 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
912 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
913 StrictArg info _ cont -> rebuildCall env (info `addArgTo` expr) cont
914 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
915 ; simplLam env' bs body cont }
916 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
917 ; rebuild env (App expr arg') cont }
921 %************************************************************************
925 %************************************************************************
928 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
929 -> SimplM (SimplEnv, OutExpr)
930 simplCast env body co0 cont0
931 = do { co1 <- simplCoercion env co0
932 ; simplExprF env body (addCoerce co1 cont0) }
934 addCoerce co cont = add_coerce co (coercionKind co) cont
936 add_coerce _co (s1, k1) cont -- co :: ty~ty
937 | s1 `coreEqType` k1 = cont -- is a no-op
939 add_coerce co1 (s1, _k2) (CoerceIt co2 cont)
940 | (_l1, t1) <- coercionKind co2
941 -- e |> (g1 :: S1~L) |> (g2 :: L~T1)
944 -- e |> (g1 . g2 :: S1~T1) otherwise
946 -- For example, in the initial form of a worker
947 -- we may find (coerce T (coerce S (\x.e))) y
948 -- and we'd like it to simplify to e[y/x] in one round
950 , s1 `coreEqType` t1 = cont -- The coerces cancel out
951 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
953 add_coerce co (s1s2, _t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
954 -- (f |> g) ty ---> (f ty) |> (g @ ty)
955 -- This implements the PushT and PushC rules from the paper
956 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
958 (new_arg_ty, new_cast)
959 | isCoVar tyvar = (new_arg_co, mkCselRCoercion co) -- PushC rule
960 | otherwise = (ty', mkInstCoercion co ty') -- PushT rule
962 ApplyTo dup (Type new_arg_ty) (zapSubstEnv arg_se) (addCoerce new_cast cont)
964 ty' = substTy (arg_se `setInScope` env) arg_ty
965 new_arg_co = mkCsel1Coercion co `mkTransCoercion`
966 ty' `mkTransCoercion`
967 mkSymCoercion (mkCsel2Coercion co)
969 add_coerce co (s1s2, _t1t2) (ApplyTo dup arg arg_se cont)
970 | not (isTypeArg arg) -- This implements the Push rule from the paper
971 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
972 -- (e |> (g :: s1s2 ~ t1->t2)) f
974 -- (e (f |> (arg g :: t1~s1))
975 -- |> (res g :: s2->t2)
977 -- t1t2 must be a function type, t1->t2, because it's applied
978 -- to something but s1s2 might conceivably not be
980 -- When we build the ApplyTo we can't mix the out-types
981 -- with the InExpr in the argument, so we simply substitute
982 -- to make it all consistent. It's a bit messy.
983 -- But it isn't a common case.
985 -- Example of use: Trac #995
986 = ApplyTo dup new_arg (zapSubstEnv arg_se) (addCoerce co2 cont)
988 -- we split coercion t1->t2 ~ s1->s2 into t1 ~ s1 and
989 -- t2 ~ s2 with left and right on the curried form:
990 -- (->) t1 t2 ~ (->) s1 s2
991 [co1, co2] = decomposeCo 2 co
992 new_arg = mkCoerce (mkSymCoercion co1) arg'
993 arg' = substExpr (arg_se `setInScope` env) arg
995 add_coerce co _ cont = CoerceIt co cont
999 %************************************************************************
1001 \subsection{Lambdas}
1003 %************************************************************************
1006 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
1007 -> SimplM (SimplEnv, OutExpr)
1009 simplLam env [] body cont = simplExprF env body cont
1012 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
1013 = do { tick (BetaReduction bndr)
1014 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
1016 -- Not enough args, so there are real lambdas left to put in the result
1017 simplLam env bndrs body cont
1018 = do { (env', bndrs') <- simplLamBndrs env bndrs
1019 ; body' <- simplExpr env' body
1020 ; new_lam <- mkLam env' bndrs' body'
1021 ; rebuild env' new_lam cont }
1024 simplNonRecE :: SimplEnv
1025 -> InBndr -- The binder
1026 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
1027 -> ([InBndr], InExpr) -- Body of the let/lambda
1030 -> SimplM (SimplEnv, OutExpr)
1032 -- simplNonRecE is used for
1033 -- * non-top-level non-recursive lets in expressions
1036 -- It deals with strict bindings, via the StrictBind continuation,
1037 -- which may abort the whole process
1039 -- The "body" of the binding comes as a pair of ([InId],InExpr)
1040 -- representing a lambda; so we recurse back to simplLam
1041 -- Why? Because of the binder-occ-info-zapping done before
1042 -- the call to simplLam in simplExprF (Lam ...)
1044 -- First deal with type applications and type lets
1045 -- (/\a. e) (Type ty) and (let a = Type ty in e)
1046 simplNonRecE env bndr (Type ty_arg, rhs_se) (bndrs, body) cont
1047 = ASSERT( isTyVar bndr )
1048 do { ty_arg' <- simplType (rhs_se `setInScope` env) ty_arg
1049 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
1051 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
1052 | preInlineUnconditionally env NotTopLevel bndr rhs
1053 = do { tick (PreInlineUnconditionally bndr)
1054 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
1057 = do { simplExprF (rhs_se `setFloats` env) rhs
1058 (StrictBind bndr bndrs body env cont) }
1061 = ASSERT( not (isTyVar bndr) )
1062 do { (env1, bndr1) <- simplNonRecBndr env bndr
1063 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
1064 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
1065 ; simplLam env3 bndrs body cont }
1069 %************************************************************************
1073 %************************************************************************
1076 -- Hack alert: we only distinguish subsumed cost centre stacks for the
1077 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
1078 simplNote :: SimplEnv -> Note -> CoreExpr -> SimplCont
1079 -> SimplM (SimplEnv, OutExpr)
1080 simplNote env (SCC cc) e cont
1081 | pushCCisNop cc (getEnclosingCC env) -- scc "f" (...(scc "f" e)...)
1082 = simplExprF env e cont -- ==> scc "f" (...e...)
1084 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
1085 ; rebuild env (mkSCC cc e') cont }
1087 simplNote env (CoreNote s) e cont
1088 = do { e' <- simplExpr env e
1089 ; rebuild env (Note (CoreNote s) e') cont }
1093 %************************************************************************
1095 \subsection{Dealing with calls}
1097 %************************************************************************
1100 simplVar :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1101 simplVar env var cont
1102 = case substId env var of
1103 DoneEx e -> simplExprF (zapSubstEnv env) e cont
1104 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
1105 DoneId var1 -> completeCall env var1 cont
1106 -- Note [zapSubstEnv]
1107 -- The template is already simplified, so don't re-substitute.
1108 -- This is VITAL. Consider
1110 -- let y = \z -> ...x... in
1112 -- We'll clone the inner \x, adding x->x' in the id_subst
1113 -- Then when we inline y, we must *not* replace x by x' in
1114 -- the inlined copy!!
1116 ---------------------------------------------------------
1117 -- Dealing with a call site
1119 completeCall :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1120 completeCall env var cont
1121 = do { ------------- Try inlining ----------------
1122 dflags <- getDOptsSmpl
1123 ; let (args,call_cont) = contArgs cont
1124 -- The args are OutExprs, obtained by *lazily* substituting
1125 -- in the args found in cont. These args are only examined
1126 -- to limited depth (unless a rule fires). But we must do
1127 -- the substitution; rule matching on un-simplified args would
1130 arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1131 n_val_args = length arg_infos
1132 interesting_cont = interestingCallContext call_cont
1133 unfolding = activeUnfolding env var
1134 maybe_inline = callSiteInline dflags var unfolding
1135 (null args) arg_infos interesting_cont
1136 ; case maybe_inline of {
1137 Just unfolding -- There is an inlining!
1138 -> do { tick (UnfoldingDone var)
1139 ; trace_inline dflags unfolding args 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 trace_inline dflags unfolding args call_cont stuff
1150 | not (dopt Opt_D_dump_inlinings dflags) = stuff
1151 | not (dopt Opt_D_verbose_core2core dflags)
1152 = if isExternalName (idName var) then
1153 pprTrace "Inlining done:" (ppr var) stuff
1156 = pprTrace ("Inlining done: " ++ showSDoc (ppr var))
1157 (vcat [text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1158 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1159 text "Cont: " <+> ppr call_cont])
1162 rebuildCall :: SimplEnv
1165 -> SimplM (SimplEnv, OutExpr)
1166 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_strs = [] }) cont
1167 -- When we run out of strictness args, it means
1168 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1169 -- Then we want to discard the entire strict continuation. E.g.
1170 -- * case (error "hello") of { ... }
1171 -- * (error "Hello") arg
1172 -- * f (error "Hello") where f is strict
1174 -- Then, especially in the first of these cases, we'd like to discard
1175 -- the continuation, leaving just the bottoming expression. But the
1176 -- type might not be right, so we may have to add a coerce.
1177 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1178 = return (env, mk_coerce res) -- contination to discard, else we do it
1179 where -- again and again!
1180 res = mkApps (Var fun) (reverse rev_args)
1181 res_ty = exprType res
1182 cont_ty = contResultType env res_ty cont
1183 co = mkUnsafeCoercion res_ty cont_ty
1184 mk_coerce expr | cont_ty `coreEqType` res_ty = expr
1185 | otherwise = mkCoerce co expr
1187 rebuildCall env info (ApplyTo _ (Type arg_ty) se cont)
1188 = do { ty' <- simplCoercion (se `setInScope` env) arg_ty
1189 ; rebuildCall env (info `addArgTo` Type ty') cont }
1191 rebuildCall env info@(ArgInfo { ai_encl = encl_rules
1192 , ai_strs = str:strs, ai_discs = disc:discs })
1193 (ApplyTo _ arg arg_se cont)
1194 | str -- Strict argument
1195 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1196 simplExprF (arg_se `setFloats` env) arg
1197 (StrictArg info' cci cont)
1200 | otherwise -- Lazy argument
1201 -- DO NOT float anything outside, hence simplExprC
1202 -- There is no benefit (unlike in a let-binding), and we'd
1203 -- have to be very careful about bogus strictness through
1204 -- floating a demanded let.
1205 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1207 ; rebuildCall env (addArgTo info' arg') cont }
1209 info' = info { ai_strs = strs, ai_discs = discs }
1210 cci | encl_rules || disc > 0 = ArgCtxt encl_rules -- Be keener here
1211 | otherwise = BoringCtxt -- Nothing interesting
1213 rebuildCall env (ArgInfo { ai_fun = fun, ai_args = rev_args, ai_rules = rules }) cont
1214 = do { -- We've accumulated a simplified call in <fun,rev_args>
1215 -- so try rewrite rules; see Note [RULEs apply to simplified arguments]
1216 -- See also Note [Rules for recursive functions]
1217 ; let args = reverse rev_args
1218 env' = zapSubstEnv env
1219 ; mb_rule <- tryRules env rules fun args cont
1221 Just (n_args, rule_rhs) -> simplExprF env' rule_rhs $
1222 pushArgs env' (drop n_args args) cont ;
1223 -- n_args says how many args the rule consumed
1224 ; Nothing -> rebuild env (mkApps (Var fun) args) cont -- No rules
1228 Note [RULES apply to simplified arguments]
1229 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1230 It's very desirable to try RULES once the arguments have been simplified, because
1231 doing so ensures that rule cascades work in one pass. Consider
1232 {-# RULES g (h x) = k x
1235 Then we want to rewrite (g (h x)) to (k x) and only then try f's rules. If
1236 we match f's rules against the un-simplified RHS, it won't match. This
1237 makes a particularly big difference when superclass selectors are involved:
1238 op ($p1 ($p2 (df d)))
1239 We want all this to unravel in one sweeep.
1243 This part of the simplifier may break the no-shadowing invariant
1245 f (...(\a -> e)...) (case y of (a,b) -> e')
1246 where f is strict in its second arg
1247 If we simplify the innermost one first we get (...(\a -> e)...)
1248 Simplifying the second arg makes us float the case out, so we end up with
1249 case y of (a,b) -> f (...(\a -> e)...) e'
1250 So the output does not have the no-shadowing invariant. However, there is
1251 no danger of getting name-capture, because when the first arg was simplified
1252 we used an in-scope set that at least mentioned all the variables free in its
1253 static environment, and that is enough.
1255 We can't just do innermost first, or we'd end up with a dual problem:
1256 case x of (a,b) -> f e (...(\a -> e')...)
1258 I spent hours trying to recover the no-shadowing invariant, but I just could
1259 not think of an elegant way to do it. The simplifier is already knee-deep in
1260 continuations. We have to keep the right in-scope set around; AND we have
1261 to get the effect that finding (error "foo") in a strict arg position will
1262 discard the entire application and replace it with (error "foo"). Getting
1263 all this at once is TOO HARD!
1266 %************************************************************************
1270 %************************************************************************
1273 tryRules :: SimplEnv -> [CoreRule]
1274 -> Id -> [OutExpr] -> SimplCont
1275 -> SimplM (Maybe (Arity, CoreExpr)) -- The arity is the number of
1276 -- args consumed by the rule
1277 tryRules env rules fn args call_cont
1281 = do { dflags <- getDOptsSmpl
1282 ; case activeRule dflags env of {
1283 Nothing -> return Nothing ; -- No rules apply
1285 case lookupRule act_fn (activeUnfInRule env) (getInScope env) fn args rules of {
1286 Nothing -> return Nothing ; -- No rule matches
1287 Just (rule, rule_rhs) ->
1289 do { tick (RuleFired (ru_name rule))
1290 ; trace_dump dflags rule rule_rhs $
1291 return (Just (ruleArity rule, rule_rhs)) }}}}
1293 trace_dump dflags rule rule_rhs stuff
1294 | not (dopt Opt_D_dump_rule_firings dflags) = stuff
1295 | not (dopt Opt_D_verbose_core2core dflags)
1297 = pprTrace "Rule fired:" (ftext (ru_name rule)) stuff
1299 = pprTrace "Rule fired"
1300 (vcat [text "Rule:" <+> ftext (ru_name rule),
1301 text "Before:" <+> ppr fn <+> sep (map pprParendExpr args),
1302 text "After: " <+> pprCoreExpr rule_rhs,
1303 text "Cont: " <+> ppr call_cont])
1307 Note [Rules for recursive functions]
1308 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1309 You might think that we shouldn't apply rules for a loop breaker:
1310 doing so might give rise to an infinite loop, because a RULE is
1311 rather like an extra equation for the function:
1312 RULE: f (g x) y = x+y
1315 But it's too drastic to disable rules for loop breakers.
1316 Even the foldr/build rule would be disabled, because foldr
1317 is recursive, and hence a loop breaker:
1318 foldr k z (build g) = g k z
1319 So it's up to the programmer: rules can cause divergence
1322 %************************************************************************
1324 Rebuilding a cse expression
1326 %************************************************************************
1328 Note [Case elimination]
1329 ~~~~~~~~~~~~~~~~~~~~~~~
1330 The case-elimination transformation discards redundant case expressions.
1331 Start with a simple situation:
1333 case x# of ===> e[x#/y#]
1336 (when x#, y# are of primitive type, of course). We can't (in general)
1337 do this for algebraic cases, because we might turn bottom into
1340 The code in SimplUtils.prepareAlts has the effect of generalise this
1341 idea to look for a case where we're scrutinising a variable, and we
1342 know that only the default case can match. For example:
1346 DEFAULT -> ...(case x of
1350 Here the inner case is first trimmed to have only one alternative, the
1351 DEFAULT, after which it's an instance of the previous case. This
1352 really only shows up in eliminating error-checking code.
1354 We also make sure that we deal with this very common case:
1359 Here we are using the case as a strict let; if x is used only once
1360 then we want to inline it. We have to be careful that this doesn't
1361 make the program terminate when it would have diverged before, so we
1363 - e is already evaluated (it may so if e is a variable)
1364 - x is used strictly, or
1366 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1368 case e of ===> case e of DEFAULT -> r
1372 Now again the case may be elminated by the CaseElim transformation.
1375 Further notes about case elimination
1376 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1377 Consider: test :: Integer -> IO ()
1380 Turns out that this compiles to:
1383 eta1 :: State# RealWorld ->
1384 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1386 (PrelNum.jtos eta ($w[] @ Char))
1388 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1390 Notice the strange '<' which has no effect at all. This is a funny one.
1391 It started like this:
1393 f x y = if x < 0 then jtos x
1394 else if y==0 then "" else jtos x
1396 At a particular call site we have (f v 1). So we inline to get
1398 if v < 0 then jtos x
1399 else if 1==0 then "" else jtos x
1401 Now simplify the 1==0 conditional:
1403 if v<0 then jtos v else jtos v
1405 Now common-up the two branches of the case:
1407 case (v<0) of DEFAULT -> jtos v
1409 Why don't we drop the case? Because it's strict in v. It's technically
1410 wrong to drop even unnecessary evaluations, and in practice they
1411 may be a result of 'seq' so we *definitely* don't want to drop those.
1412 I don't really know how to improve this situation.
1415 ---------------------------------------------------------
1416 -- Eliminate the case if possible
1418 rebuildCase, reallyRebuildCase
1420 -> OutExpr -- Scrutinee
1421 -> InId -- Case binder
1422 -> [InAlt] -- Alternatives (inceasing order)
1424 -> SimplM (SimplEnv, OutExpr)
1426 --------------------------------------------------
1427 -- 1. Eliminate the case if there's a known constructor
1428 --------------------------------------------------
1430 rebuildCase env scrut case_bndr alts cont
1431 | Lit lit <- scrut -- No need for same treatment as constructors
1432 -- because literals are inlined more vigorously
1433 = do { tick (KnownBranch case_bndr)
1434 ; case findAlt (LitAlt lit) alts of
1435 Nothing -> missingAlt env case_bndr alts cont
1436 Just (_, bs, rhs) -> simple_rhs bs rhs }
1438 | Just (con, ty_args, other_args) <- exprIsConApp_maybe (activeUnfInRule env) scrut
1439 -- Works when the scrutinee is a variable with a known unfolding
1440 -- as well as when it's an explicit constructor application
1441 = do { tick (KnownBranch case_bndr)
1442 ; case findAlt (DataAlt con) alts of
1443 Nothing -> missingAlt env case_bndr alts cont
1444 Just (DEFAULT, bs, rhs) -> simple_rhs bs rhs
1445 Just (_, bs, rhs) -> knownCon env scrut con ty_args other_args
1446 case_bndr bs rhs cont
1449 simple_rhs bs rhs = ASSERT( null bs )
1450 do { env' <- simplNonRecX env case_bndr scrut
1451 ; simplExprF env' rhs cont }
1454 --------------------------------------------------
1455 -- 2. Eliminate the case if scrutinee is evaluated
1456 --------------------------------------------------
1458 rebuildCase env scrut case_bndr [(_, bndrs, rhs)] cont
1459 -- See if we can get rid of the case altogether
1460 -- See Note [Case eliminiation]
1461 -- mkCase made sure that if all the alternatives are equal,
1462 -- then there is now only one (DEFAULT) rhs
1463 | all isDeadBinder bndrs -- bndrs are [InId]
1465 -- Check that the scrutinee can be let-bound instead of case-bound
1466 , exprOkForSpeculation scrut
1467 -- OK not to evaluate it
1468 -- This includes things like (==# a# b#)::Bool
1469 -- so that we simplify
1470 -- case ==# a# b# of { True -> x; False -> x }
1473 -- This particular example shows up in default methods for
1474 -- comparision operations (e.g. in (>=) for Int.Int32)
1475 || exprIsHNF scrut -- It's already evaluated
1476 || var_demanded_later scrut -- It'll be demanded later
1478 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1479 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1480 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1481 -- its argument: case x of { y -> dataToTag# y }
1482 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1483 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1485 -- Also we don't want to discard 'seq's
1486 = do { tick (CaseElim case_bndr)
1487 ; env' <- simplNonRecX env case_bndr scrut
1488 ; simplExprF env' rhs cont }
1490 -- The case binder is going to be evaluated later,
1491 -- and the scrutinee is a simple variable
1492 var_demanded_later (Var v) = isStrictDmd (idDemandInfo case_bndr)
1493 && not (isTickBoxOp v)
1494 -- ugly hack; covering this case is what
1495 -- exprOkForSpeculation was intended for.
1496 var_demanded_later _ = False
1498 --------------------------------------------------
1499 -- 3. Try seq rules; see Note [User-defined RULES for seq] in MkId
1500 --------------------------------------------------
1502 rebuildCase env scrut case_bndr alts@[(_, bndrs, rhs)] cont
1503 | all isDeadBinder (case_bndr : bndrs) -- So this is just 'seq'
1504 = do { let rhs' = substExpr env rhs
1505 out_args = [Type (substTy env (idType case_bndr)),
1506 Type (exprType rhs'), scrut, rhs']
1507 -- Lazily evaluated, so we don't do most of this
1509 ; rule_base <- getSimplRules
1510 ; mb_rule <- tryRules env (getRules rule_base seqId) seqId out_args cont
1512 Just (n_args, res) -> simplExprF (zapSubstEnv env)
1513 (mkApps res (drop n_args out_args))
1515 Nothing -> reallyRebuildCase env scrut case_bndr alts cont }
1517 rebuildCase env scrut case_bndr alts cont
1518 = reallyRebuildCase env scrut case_bndr alts cont
1520 --------------------------------------------------
1521 -- 3. Catch-all case
1522 --------------------------------------------------
1524 reallyRebuildCase env scrut case_bndr alts cont
1525 = do { -- Prepare the continuation;
1526 -- The new subst_env is in place
1527 (env', dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1529 -- Simplify the alternatives
1530 ; (scrut', case_bndr', alts') <- simplAlts env' scrut case_bndr alts dup_cont
1532 -- Check for empty alternatives
1533 ; if null alts' then missingAlt env case_bndr alts cont
1535 { dflags <- getDOptsSmpl
1536 ; case_expr <- mkCase dflags scrut' case_bndr' alts'
1538 -- Notice that rebuild gets the in-scope set from env', not alt_env
1539 -- (which in any case is only build in simplAlts)
1540 -- The case binder *not* scope over the whole returned case-expression
1541 ; rebuild env' case_expr nodup_cont } }
1544 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1545 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1546 way, there's a chance that v will now only be used once, and hence
1549 Historical note: we use to do the "case binder swap" in the Simplifier
1550 so there were additional complications if the scrutinee was a variable.
1551 Now the binder-swap stuff is done in the occurrence analyer; see
1552 OccurAnal Note [Binder swap].
1556 If the case binder is not dead, then neither are the pattern bound
1558 case <any> of x { (a,b) ->
1559 case x of { (p,q) -> p } }
1560 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1561 The point is that we bring into the envt a binding
1563 after the outer case, and that makes (a,b) alive. At least we do unless
1564 the case binder is guaranteed dead.
1566 In practice, the scrutinee is almost always a variable, so we pretty
1567 much always zap the OccInfo of the binders. It doesn't matter much though.
1572 Consider case (v `cast` co) of x { I# y ->
1573 ... (case (v `cast` co) of {...}) ...
1574 We'd like to eliminate the inner case. We can get this neatly by
1575 arranging that inside the outer case we add the unfolding
1576 v |-> x `cast` (sym co)
1577 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1579 Note [Improving seq]
1582 type family F :: * -> *
1583 type instance F Int = Int
1585 ... case e of x { DEFAULT -> rhs } ...
1587 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1589 case e `cast` co of x'::Int
1590 I# x# -> let x = x' `cast` sym co
1593 so that 'rhs' can take advantage of the form of x'.
1595 Notice that Note [Case of cast] may then apply to the result.
1597 Nota Bene: We only do the [Improving seq] transformation if the
1598 case binder 'x' is actually used in the rhs; that is, if the case
1599 is *not* a *pure* seq.
1600 a) There is no point in adding the cast to a pure seq.
1601 b) There is a good reason not to: doing so would interfere
1602 with seq rules (Note [Built-in RULES for seq] in MkId).
1603 In particular, this [Improving seq] thing *adds* a cast
1604 while [Built-in RULES for seq] *removes* one, so they
1607 You might worry about
1608 case v of x { __DEFAULT ->
1609 ... case (v `cast` co) of y { I# -> ... }}
1610 This is a pure seq (since x is unused), so [Improving seq] won't happen.
1611 But it's ok: the simplifier will replace 'v' by 'x' in the rhs to get
1612 case v of x { __DEFAULT ->
1613 ... case (x `cast` co) of y { I# -> ... }}
1614 Now the outer case is not a pure seq, so [Improving seq] will happen,
1615 and then the inner case will disappear.
1617 The need for [Improving seq] showed up in Roman's experiments. Example:
1618 foo :: F Int -> Int -> Int
1619 foo t n = t `seq` bar n
1622 bar n = bar (n - case t of TI i -> i)
1623 Here we'd like to avoid repeated evaluating t inside the loop, by
1624 taking advantage of the `seq`.
1626 At one point I did transformation in LiberateCase, but it's more
1627 robust here. (Otherwise, there's a danger that we'll simply drop the
1628 'seq' altogether, before LiberateCase gets to see it.)
1631 simplAlts :: SimplEnv
1633 -> InId -- Case binder
1634 -> [InAlt] -- Non-empty
1636 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1637 -- Like simplExpr, this just returns the simplified alternatives;
1638 -- it does not return an environment
1640 simplAlts env scrut case_bndr alts cont'
1641 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1642 do { let env0 = zapFloats env
1644 ; (env1, case_bndr1) <- simplBinder env0 case_bndr
1646 ; fam_envs <- getFamEnvs
1647 ; (alt_env', scrut', case_bndr') <- improveSeq fam_envs env1 scrut
1648 case_bndr case_bndr1 alts
1650 ; (imposs_deflt_cons, in_alts) <- prepareAlts scrut' case_bndr' alts
1652 ; alts' <- mapM (simplAlt alt_env' imposs_deflt_cons case_bndr' cont') in_alts
1653 ; return (scrut', case_bndr', alts') }
1656 ------------------------------------
1657 improveSeq :: (FamInstEnv, FamInstEnv) -> SimplEnv
1658 -> OutExpr -> InId -> OutId -> [InAlt]
1659 -> SimplM (SimplEnv, OutExpr, OutId)
1660 -- Note [Improving seq]
1661 improveSeq fam_envs env scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1662 | not (isDeadBinder case_bndr) -- Not a pure seq! See the Note!
1663 , Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1664 = do { case_bndr2 <- newId (fsLit "nt") ty2
1665 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1666 env2 = extendIdSubst env case_bndr rhs
1667 ; return (env2, scrut `Cast` co, case_bndr2) }
1669 improveSeq _ env scrut _ case_bndr1 _
1670 = return (env, scrut, case_bndr1)
1673 ------------------------------------
1674 simplAlt :: SimplEnv
1675 -> [AltCon] -- These constructors can't be present when
1676 -- matching the DEFAULT alternative
1677 -> OutId -- The case binder
1682 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1683 = ASSERT( null bndrs )
1684 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1685 -- Record the constructors that the case-binder *can't* be.
1686 ; rhs' <- simplExprC env' rhs cont'
1687 ; return (DEFAULT, [], rhs') }
1689 simplAlt env _ case_bndr' cont' (LitAlt lit, bndrs, rhs)
1690 = ASSERT( null bndrs )
1691 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1692 ; rhs' <- simplExprC env' rhs cont'
1693 ; return (LitAlt lit, [], rhs') }
1695 simplAlt env _ case_bndr' cont' (DataAlt con, vs, rhs)
1696 = do { -- Deal with the pattern-bound variables
1697 -- Mark the ones that are in ! positions in the
1698 -- data constructor as certainly-evaluated.
1699 -- NB: simplLamBinders preserves this eval info
1700 let vs_with_evals = add_evals (dataConRepStrictness con)
1701 ; (env', vs') <- simplLamBndrs env vs_with_evals
1703 -- Bind the case-binder to (con args)
1704 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1705 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1706 env'' = addBinderUnfolding env' case_bndr'
1707 (mkConApp con con_args)
1709 ; rhs' <- simplExprC env'' rhs cont'
1710 ; return (DataAlt con, vs', rhs') }
1712 -- add_evals records the evaluated-ness of the bound variables of
1713 -- a case pattern. This is *important*. Consider
1714 -- data T = T !Int !Int
1716 -- case x of { T a b -> T (a+1) b }
1718 -- We really must record that b is already evaluated so that we don't
1719 -- go and re-evaluate it when constructing the result.
1720 -- See Note [Data-con worker strictness] in MkId.lhs
1725 go (v:vs') strs | isTyVar v = v : go vs' strs
1726 go (v:vs') (str:strs)
1727 | isMarkedStrict str = evald_v : go vs' strs
1728 | otherwise = zapped_v : go vs' strs
1730 zapped_v = zap_occ_info v
1731 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1732 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr the_strs)
1734 -- See Note [zapOccInfo]
1735 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1737 -- to the envt; so vs are now very much alive
1738 -- Note [Aug06] I can't see why this actually matters, but it's neater
1739 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1740 -- ==> case e of t { (a,b) -> ...(a)... }
1741 -- Look, Ma, a is alive now.
1742 zap_occ_info = zapCasePatIdOcc case_bndr'
1744 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1745 addBinderUnfolding env bndr rhs
1746 = modifyInScope env (bndr `setIdUnfolding` mkUnfolding False False rhs)
1748 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1749 addBinderOtherCon env bndr cons
1750 = modifyInScope env (bndr `setIdUnfolding` mkOtherCon cons)
1752 zapCasePatIdOcc :: Id -> Id -> Id
1753 -- Consider case e of b { (a,b) -> ... }
1754 -- Then if we bind b to (a,b) in "...", and b is not dead,
1755 -- then we must zap the deadness info on a,b
1756 zapCasePatIdOcc case_bndr
1757 | isDeadBinder case_bndr = \ pat_id -> pat_id
1758 | otherwise = \ pat_id -> zapIdOccInfo pat_id
1762 %************************************************************************
1764 \subsection{Known constructor}
1766 %************************************************************************
1768 We are a bit careful with occurrence info. Here's an example
1770 (\x* -> case x of (a*, b) -> f a) (h v, e)
1772 where the * means "occurs once". This effectively becomes
1773 case (h v, e) of (a*, b) -> f a)
1775 let a* = h v; b = e in f a
1779 All this should happen in one sweep.
1782 knownCon :: SimplEnv
1783 -> OutExpr -- The scrutinee
1784 -> DataCon -> [OutType] -> [OutExpr] -- The scrutinee (in pieces)
1785 -> InId -> [InBndr] -> InExpr -- The alternative
1787 -> SimplM (SimplEnv, OutExpr)
1789 knownCon env scrut dc dc_ty_args dc_args bndr bs rhs cont
1790 = do { env' <- bind_args env bs dc_args
1792 -- It's useful to bind bndr to scrut, rather than to a fresh
1793 -- binding x = Con arg1 .. argn
1794 -- because very often the scrut is a variable, so we avoid
1795 -- creating, and then subsequently eliminating, a let-binding
1796 -- BUT, if scrut is a not a variable, we must be careful
1797 -- about duplicating the arg redexes; in that case, make
1798 -- a new con-app from the args
1799 bndr_rhs | exprIsTrivial scrut = scrut
1800 | otherwise = con_app
1801 con_app = Var (dataConWorkId dc)
1802 `mkTyApps` dc_ty_args
1803 `mkApps` [substExpr env' (varToCoreExpr b) | b <- bs]
1804 -- dc_ty_args are aready OutTypes, but bs are InBndrs
1806 ; env'' <- simplNonRecX env' bndr bndr_rhs
1807 ; simplExprF env'' rhs cont }
1809 zap_occ = zapCasePatIdOcc bndr -- bndr is an InId
1812 bind_args env' [] _ = return env'
1814 bind_args env' (b:bs') (Type ty : args)
1815 = ASSERT( isTyVar b )
1816 bind_args (extendTvSubst env' b ty) bs' args
1818 bind_args env' (b:bs') (arg : args)
1820 do { let b' = zap_occ b
1821 -- Note that the binder might be "dead", because it doesn't
1822 -- occur in the RHS; and simplNonRecX may therefore discard
1823 -- it via postInlineUnconditionally.
1824 -- Nevertheless we must keep it if the case-binder is alive,
1825 -- because it may be used in the con_app. See Note [zapOccInfo]
1826 ; env'' <- simplNonRecX env' b' arg
1827 ; bind_args env'' bs' args }
1830 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr dc_args $$
1831 text "scrut:" <+> ppr scrut
1834 missingAlt :: SimplEnv -> Id -> [InAlt] -> SimplCont -> SimplM (SimplEnv, OutExpr)
1835 -- This isn't strictly an error, although it is unusual.
1836 -- It's possible that the simplifer might "see" that
1837 -- an inner case has no accessible alternatives before
1838 -- it "sees" that the entire branch of an outer case is
1839 -- inaccessible. So we simply put an error case here instead.
1840 missingAlt env case_bndr alts cont
1841 = WARN( True, ptext (sLit "missingAlt") <+> ppr case_bndr )
1842 return (env, mkImpossibleExpr res_ty)
1844 res_ty = contResultType env (substTy env (coreAltsType alts)) cont
1848 %************************************************************************
1850 \subsection{Duplicating continuations}
1852 %************************************************************************
1855 prepareCaseCont :: SimplEnv
1856 -> [InAlt] -> SimplCont
1857 -> SimplM (SimplEnv, SimplCont,SimplCont)
1858 -- Return a duplicatable continuation, a non-duplicable part
1859 -- plus some extra bindings (that scope over the entire
1862 -- No need to make it duplicatable if there's only one alternative
1863 prepareCaseCont env [_] cont = return (env, cont, mkBoringStop)
1864 prepareCaseCont env _ cont = mkDupableCont env cont
1868 mkDupableCont :: SimplEnv -> SimplCont
1869 -> SimplM (SimplEnv, SimplCont, SimplCont)
1871 mkDupableCont env cont
1872 | contIsDupable cont
1873 = return (env, cont, mkBoringStop)
1875 mkDupableCont _ (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1877 mkDupableCont env (CoerceIt ty cont)
1878 = do { (env', dup, nodup) <- mkDupableCont env cont
1879 ; return (env', CoerceIt ty dup, nodup) }
1881 mkDupableCont env cont@(StrictBind {})
1882 = return (env, mkBoringStop, cont)
1883 -- See Note [Duplicating StrictBind]
1885 mkDupableCont env (StrictArg info cci cont)
1886 -- See Note [Duplicating StrictArg]
1887 = do { (env', dup, nodup) <- mkDupableCont env cont
1888 ; (env'', args') <- mapAccumLM makeTrivial env' (ai_args info)
1889 ; return (env'', StrictArg (info { ai_args = args' }) cci dup, nodup) }
1891 mkDupableCont env (ApplyTo _ arg se cont)
1892 = -- e.g. [...hole...] (...arg...)
1894 -- let a = ...arg...
1895 -- in [...hole...] a
1896 do { (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1897 ; arg' <- simplExpr (se `setInScope` env') arg
1898 ; (env'', arg'') <- makeTrivial env' arg'
1899 ; let app_cont = ApplyTo OkToDup arg'' (zapSubstEnv env'') dup_cont
1900 ; return (env'', app_cont, nodup_cont) }
1902 mkDupableCont env cont@(Select _ case_bndr [(_, bs, _rhs)] _ _)
1903 -- See Note [Single-alternative case]
1904 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1905 -- | not (isDeadBinder case_bndr)
1906 | all isDeadBinder bs -- InIds
1907 && not (isUnLiftedType (idType case_bndr))
1908 -- Note [Single-alternative-unlifted]
1909 = return (env, mkBoringStop, cont)
1911 mkDupableCont env (Select _ case_bndr alts se cont)
1912 = -- e.g. (case [...hole...] of { pi -> ei })
1914 -- let ji = \xij -> ei
1915 -- in case [...hole...] of { pi -> ji xij }
1916 do { tick (CaseOfCase case_bndr)
1917 ; (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1918 -- NB: call mkDupableCont here, *not* prepareCaseCont
1919 -- We must make a duplicable continuation, whereas prepareCaseCont
1920 -- doesn't when there is a single case branch
1922 ; let alt_env = se `setInScope` env'
1923 ; (alt_env', case_bndr') <- simplBinder alt_env case_bndr
1924 ; alts' <- mapM (simplAlt alt_env' [] case_bndr' dup_cont) alts
1925 -- Safe to say that there are no handled-cons for the DEFAULT case
1926 -- NB: simplBinder does not zap deadness occ-info, so
1927 -- a dead case_bndr' will still advertise its deadness
1928 -- This is really important because in
1929 -- case e of b { (# p,q #) -> ... }
1930 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1931 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1932 -- In the new alts we build, we have the new case binder, so it must retain
1934 -- NB: we don't use alt_env further; it has the substEnv for
1935 -- the alternatives, and we don't want that
1937 ; (env'', alts'') <- mkDupableAlts env' case_bndr' alts'
1938 ; return (env'', -- Note [Duplicated env]
1939 Select OkToDup case_bndr' alts'' (zapSubstEnv env'') mkBoringStop,
1943 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1944 -> SimplM (SimplEnv, [InAlt])
1945 -- Absorbs the continuation into the new alternatives
1947 mkDupableAlts env case_bndr' the_alts
1950 go env0 [] = return (env0, [])
1952 = do { (env1, alt') <- mkDupableAlt env0 case_bndr' alt
1953 ; (env2, alts') <- go env1 alts
1954 ; return (env2, alt' : alts' ) }
1956 mkDupableAlt :: SimplEnv -> OutId -> (AltCon, [CoreBndr], CoreExpr)
1957 -> SimplM (SimplEnv, (AltCon, [CoreBndr], CoreExpr))
1958 mkDupableAlt env case_bndr (con, bndrs', rhs')
1959 | exprIsDupable rhs' -- Note [Small alternative rhs]
1960 = return (env, (con, bndrs', rhs'))
1962 = do { let rhs_ty' = exprType rhs'
1963 scrut_ty = idType case_bndr
1966 DEFAULT -> case_bndr
1967 DataAlt dc -> setIdUnfolding case_bndr unf
1969 -- See Note [Case binders and join points]
1970 unf = mkInlineRule needSaturated rhs 0
1971 rhs = mkConApp dc (map Type (tyConAppArgs scrut_ty)
1972 ++ varsToCoreExprs bndrs')
1974 LitAlt {} -> WARN( True, ptext (sLit "mkDupableAlt")
1975 <+> ppr case_bndr <+> ppr con )
1977 -- The case binder is alive but trivial, so why has
1978 -- it not been substituted away?
1980 used_bndrs' | isDeadBinder case_bndr = filter abstract_over bndrs'
1981 | otherwise = bndrs' ++ [case_bndr_w_unf]
1984 | isTyVar bndr = True -- Abstract over all type variables just in case
1985 | otherwise = not (isDeadBinder bndr)
1986 -- The deadness info on the new Ids is preserved by simplBinders
1988 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1989 <- if (any isId used_bndrs')
1990 then return (used_bndrs', varsToCoreExprs used_bndrs')
1991 else do { rw_id <- newId (fsLit "w") realWorldStatePrimTy
1992 ; return ([rw_id], [Var realWorldPrimId]) }
1994 ; join_bndr <- newId (fsLit "$j") (mkPiTypes final_bndrs' rhs_ty')
1995 -- Note [Funky mkPiTypes]
1997 ; let -- We make the lambdas into one-shot-lambdas. The
1998 -- join point is sure to be applied at most once, and doing so
1999 -- prevents the body of the join point being floated out by
2000 -- the full laziness pass
2001 really_final_bndrs = map one_shot final_bndrs'
2002 one_shot v | isId v = setOneShotLambda v
2004 join_rhs = mkLams really_final_bndrs rhs'
2005 join_call = mkApps (Var join_bndr) final_args
2007 ; env' <- addPolyBind NotTopLevel env (NonRec join_bndr join_rhs)
2008 ; return (env', (con, bndrs', join_call)) }
2009 -- See Note [Duplicated env]
2012 Note [Case binders and join points]
2013 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2015 case (case .. ) of c {
2018 If we make a join point with c but not c# we get
2019 $j = \c -> ....c....
2021 But if later inlining scrutines the c, thus
2023 $j = \c -> ... case c of { I# y -> ... } ...
2025 we won't see that 'c' has already been scrutinised. This actually
2026 happens in the 'tabulate' function in wave4main, and makes a significant
2027 difference to allocation.
2029 An alternative plan is this:
2031 $j = \c# -> let c = I# c# in ...c....
2033 but that is bad if 'c' is *not* later scrutinised.
2035 So instead we do both: we pass 'c' and 'c#' , and record in c's inlining
2036 that it's really I# c#, thus
2038 $j = \c# -> \c[=I# c#] -> ...c....
2040 Absence analysis may later discard 'c'.
2043 Note [Duplicated env]
2044 ~~~~~~~~~~~~~~~~~~~~~
2045 Some of the alternatives are simplified, but have not been turned into a join point
2046 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
2047 bind the join point, because it might to do PostInlineUnconditionally, and
2048 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
2049 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
2050 at worst delays the join-point inlining.
2052 Note [Small alternative rhs]
2053 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2054 It is worth checking for a small RHS because otherwise we
2055 get extra let bindings that may cause an extra iteration of the simplifier to
2056 inline back in place. Quite often the rhs is just a variable or constructor.
2057 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
2058 iterations because the version with the let bindings looked big, and so wasn't
2059 inlined, but after the join points had been inlined it looked smaller, and so
2062 NB: we have to check the size of rhs', not rhs.
2063 Duplicating a small InAlt might invalidate occurrence information
2064 However, if it *is* dupable, we return the *un* simplified alternative,
2065 because otherwise we'd need to pair it up with an empty subst-env....
2066 but we only have one env shared between all the alts.
2067 (Remember we must zap the subst-env before re-simplifying something).
2068 Rather than do this we simply agree to re-simplify the original (small) thing later.
2070 Note [Funky mkPiTypes]
2071 ~~~~~~~~~~~~~~~~~~~~~~
2072 Notice the funky mkPiTypes. If the contructor has existentials
2073 it's possible that the join point will be abstracted over
2074 type varaibles as well as term variables.
2075 Example: Suppose we have
2076 data T = forall t. C [t]
2078 case (case e of ...) of
2080 We get the join point
2081 let j :: forall t. [t] -> ...
2082 j = /\t \xs::[t] -> rhs
2084 case (case e of ...) of
2085 C t xs::[t] -> j t xs
2087 Note [Join point abstaction]
2088 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2089 If we try to lift a primitive-typed something out
2090 for let-binding-purposes, we will *caseify* it (!),
2091 with potentially-disastrous strictness results. So
2092 instead we turn it into a function: \v -> e
2093 where v::State# RealWorld#. The value passed to this function
2094 is realworld#, which generates (almost) no code.
2096 There's a slight infelicity here: we pass the overall
2097 case_bndr to all the join points if it's used in *any* RHS,
2098 because we don't know its usage in each RHS separately
2100 We used to say "&& isUnLiftedType rhs_ty'" here, but now
2101 we make the join point into a function whenever used_bndrs'
2102 is empty. This makes the join-point more CPR friendly.
2103 Consider: let j = if .. then I# 3 else I# 4
2104 in case .. of { A -> j; B -> j; C -> ... }
2106 Now CPR doesn't w/w j because it's a thunk, so
2107 that means that the enclosing function can't w/w either,
2108 which is a lose. Here's the example that happened in practice:
2109 kgmod :: Int -> Int -> Int
2110 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
2114 I have seen a case alternative like this:
2116 It's a bit silly to add the realWorld dummy arg in this case, making
2119 (the \v alone is enough to make CPR happy) but I think it's rare
2121 Note [Duplicating StrictArg]
2122 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2123 The original plan had (where E is a big argument)
2125 ==> let $j = \a -> f E a
2128 But this is terrible! Here's an example:
2129 && E (case x of { T -> F; F -> T })
2130 Now, && is strict so we end up simplifying the case with
2131 an ArgOf continuation. If we let-bind it, we get
2132 let $j = \v -> && E v
2133 in simplExpr (case x of { T -> F; F -> T })
2135 And after simplifying more we get
2136 let $j = \v -> && E v
2137 in case x of { T -> $j F; F -> $j T }
2138 Which is a Very Bad Thing
2140 What we do now is this
2144 Now if the thing in the hole is a case expression (which is when
2145 we'll call mkDupableCont), we'll push the function call into the
2146 branches, which is what we want. Now RULES for f may fire, and
2147 call-pattern specialisation. Here's an example from Trac #3116
2150 _ -> Chunk p fpc (o+1) (l-1) bs')
2151 If we can push the call for 'go' inside the case, we get
2152 call-pattern specialisation for 'go', which is *crucial* for
2155 Here is the (&&) example:
2156 && E (case x of { T -> F; F -> T })
2158 case x of { T -> && a F; F -> && a T }
2162 * Arguments to f *after* the strict one are handled by
2163 the ApplyTo case of mkDupableCont. Eg
2166 * We can only do the let-binding of E because the function
2167 part of a StrictArg continuation is an explicit syntax
2168 tree. In earlier versions we represented it as a function
2169 (CoreExpr -> CoreEpxr) which we couldn't take apart.
2171 Do *not* duplicate StrictBind and StritArg continuations. We gain
2172 nothing by propagating them into the expressions, and we do lose a
2175 The desire not to duplicate is the entire reason that
2176 mkDupableCont returns a pair of continuations.
2178 Note [Duplicating StrictBind]
2179 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2180 Unlike StrictArg, there doesn't seem anything to gain from
2181 duplicating a StrictBind continuation, so we don't.
2183 The desire not to duplicate is the entire reason that
2184 mkDupableCont returns a pair of continuations.
2187 Note [Single-alternative cases]
2188 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2189 This case is just like the ArgOf case. Here's an example:
2193 case (case x of I# x' ->
2195 True -> I# (negate# x')
2196 False -> I# x') of y {
2198 Because the (case x) has only one alternative, we'll transform to
2200 case (case x' <# 0# of
2201 True -> I# (negate# x')
2202 False -> I# x') of y {
2204 But now we do *NOT* want to make a join point etc, giving
2206 let $j = \y -> MkT y
2208 True -> $j (I# (negate# x'))
2210 In this case the $j will inline again, but suppose there was a big
2211 strict computation enclosing the orginal call to MkT. Then, it won't
2212 "see" the MkT any more, because it's big and won't get duplicated.
2213 And, what is worse, nothing was gained by the case-of-case transform.
2215 When should use this case of mkDupableCont?
2216 However, matching on *any* single-alternative case is a *disaster*;
2217 e.g. case (case ....) of (a,b) -> (# a,b #)
2218 We must push the outer case into the inner one!
2221 * Match [(DEFAULT,_,_)], but in the common case of Int,
2222 the alternative-filling-in code turned the outer case into
2223 case (...) of y { I# _ -> MkT y }
2225 * Match on single alternative plus (not (isDeadBinder case_bndr))
2226 Rationale: pushing the case inwards won't eliminate the construction.
2227 But there's a risk of
2228 case (...) of y { (a,b) -> let z=(a,b) in ... }
2229 Now y looks dead, but it'll come alive again. Still, this
2230 seems like the best option at the moment.
2232 * Match on single alternative plus (all (isDeadBinder bndrs))
2233 Rationale: this is essentially seq.
2235 * Match when the rhs is *not* duplicable, and hence would lead to a
2236 join point. This catches the disaster-case above. We can test
2237 the *un-simplified* rhs, which is fine. It might get bigger or
2238 smaller after simplification; if it gets smaller, this case might
2239 fire next time round. NB also that we must test contIsDupable
2240 case_cont *btoo, because case_cont might be big!
2242 HOWEVER: I found that this version doesn't work well, because
2243 we can get let x = case (...) of { small } in ...case x...
2244 When x is inlined into its full context, we find that it was a bad
2245 idea to have pushed the outer case inside the (...) case.
2247 Note [Single-alternative-unlifted]
2248 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2249 Here's another single-alternative where we really want to do case-of-case:
2257 case y_s6X of tpl_s7m {
2258 M1.Mk1 ipv_s70 -> ipv_s70;
2259 M1.Mk2 ipv_s72 -> ipv_s72;
2265 case x_s74 of tpl_s7n {
2266 M1.Mk1 ipv_s77 -> ipv_s77;
2267 M1.Mk2 ipv_s79 -> ipv_s79;
2271 { __DEFAULT -> ==# [wild1_s7b wild_s7c];
2275 So the outer case is doing *nothing at all*, other than serving as a
2276 join-point. In this case we really want to do case-of-case and decide
2277 whether to use a real join point or just duplicate the continuation.
2279 Hence: check whether the case binder's type is unlifted, because then
2280 the outer case is *not* a seq.