2 % (c) The AQUA Project, Glasgow University, 1993-1998
4 \section[Simplify]{The main module of the simplifier}
8 -- The above warning supression flag is a temporary kludge.
9 -- While working on this module you are encouraged to remove it and fix
10 -- any warnings in the module. See
11 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
14 module Simplify ( simplTopBinds, simplExpr ) where
16 #include "HsVersions.h"
20 import Type hiding ( substTy, extendTvSubst )
27 import FamInstEnv ( topNormaliseType )
28 import DataCon ( dataConRepStrictness, dataConUnivTyVars )
30 import NewDemand ( isStrictDmd )
31 import PprCore ( pprParendExpr, pprCoreExpr )
32 import CoreUnfold ( mkUnfolding, callSiteInline, CallCtxt(..) )
34 import Rules ( lookupRule )
35 import BasicTypes ( isMarkedStrict )
36 import CostCentre ( currentCCS )
37 import TysPrim ( realWorldStatePrimTy )
38 import PrelInfo ( realWorldPrimId )
39 import BasicTypes ( TopLevelFlag(..), isTopLevel,
40 RecFlag(..), isNonRuleLoopBreaker )
41 import Maybes ( orElse )
42 import Data.List ( mapAccumL )
48 The guts of the simplifier is in this module, but the driver loop for
49 the simplifier is in SimplCore.lhs.
52 -----------------------------------------
53 *** IMPORTANT NOTE ***
54 -----------------------------------------
55 The simplifier used to guarantee that the output had no shadowing, but
56 it does not do so any more. (Actually, it never did!) The reason is
57 documented with simplifyArgs.
60 -----------------------------------------
61 *** IMPORTANT NOTE ***
62 -----------------------------------------
63 Many parts of the simplifier return a bunch of "floats" as well as an
64 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
66 All "floats" are let-binds, not case-binds, but some non-rec lets may
67 be unlifted (with RHS ok-for-speculation).
71 -----------------------------------------
72 ORGANISATION OF FUNCTIONS
73 -----------------------------------------
75 - simplify all top-level binders
76 - for NonRec, call simplRecOrTopPair
77 - for Rec, call simplRecBind
80 ------------------------------
81 simplExpr (applied lambda) ==> simplNonRecBind
82 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
83 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
85 ------------------------------
86 simplRecBind [binders already simplfied]
87 - use simplRecOrTopPair on each pair in turn
89 simplRecOrTopPair [binder already simplified]
90 Used for: recursive bindings (top level and nested)
91 top-level non-recursive bindings
93 - check for PreInlineUnconditionally
97 Used for: non-top-level non-recursive bindings
98 beta reductions (which amount to the same thing)
99 Because it can deal with strict arts, it takes a
100 "thing-inside" and returns an expression
102 - check for PreInlineUnconditionally
103 - simplify binder, including its IdInfo
112 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
113 Used for: binding case-binder and constr args in a known-constructor case
114 - check for PreInLineUnconditionally
118 ------------------------------
119 simplLazyBind: [binder already simplified, RHS not]
120 Used for: recursive bindings (top level and nested)
121 top-level non-recursive bindings
122 non-top-level, but *lazy* non-recursive bindings
123 [must not be strict or unboxed]
124 Returns floats + an augmented environment, not an expression
125 - substituteIdInfo and add result to in-scope
126 [so that rules are available in rec rhs]
129 - float if exposes constructor or PAP
133 completeNonRecX: [binder and rhs both simplified]
134 - if the the thing needs case binding (unlifted and not ok-for-spec)
140 completeBind: [given a simplified RHS]
141 [used for both rec and non-rec bindings, top level and not]
142 - try PostInlineUnconditionally
143 - add unfolding [this is the only place we add an unfolding]
148 Right hand sides and arguments
149 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
150 In many ways we want to treat
151 (a) the right hand side of a let(rec), and
152 (b) a function argument
153 in the same way. But not always! In particular, we would
154 like to leave these arguments exactly as they are, so they
155 will match a RULE more easily.
160 It's harder to make the rule match if we ANF-ise the constructor,
161 or eta-expand the PAP:
163 f (let { a = g x; b = h x } in (a,b))
166 On the other hand if we see the let-defns
171 then we *do* want to ANF-ise and eta-expand, so that p and q
172 can be safely inlined.
174 Even floating lets out is a bit dubious. For let RHS's we float lets
175 out if that exposes a value, so that the value can be inlined more vigorously.
178 r = let x = e in (x,x)
180 Here, if we float the let out we'll expose a nice constructor. We did experiments
181 that showed this to be a generally good thing. But it was a bad thing to float
182 lets out unconditionally, because that meant they got allocated more often.
184 For function arguments, there's less reason to expose a constructor (it won't
185 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
186 So for the moment we don't float lets out of function arguments either.
191 For eta expansion, we want to catch things like
193 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
195 If the \x was on the RHS of a let, we'd eta expand to bring the two
196 lambdas together. And in general that's a good thing to do. Perhaps
197 we should eta expand wherever we find a (value) lambda? Then the eta
198 expansion at a let RHS can concentrate solely on the PAP case.
201 %************************************************************************
203 \subsection{Bindings}
205 %************************************************************************
208 simplTopBinds :: SimplEnv -> [InBind] -> SimplM [OutBind]
210 simplTopBinds env binds
211 = do { -- Put all the top-level binders into scope at the start
212 -- so that if a transformation rule has unexpectedly brought
213 -- anything into scope, then we don't get a complaint about that.
214 -- It's rather as if the top-level binders were imported.
215 ; env <- simplRecBndrs env (bindersOfBinds binds)
216 ; dflags <- getDOptsSmpl
217 ; let dump_flag = dopt Opt_D_dump_inlinings dflags ||
218 dopt Opt_D_dump_rule_firings dflags
219 ; env' <- simpl_binds dump_flag env binds
220 ; freeTick SimplifierDone
221 ; return (getFloats env') }
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 dump env [] = return env
231 simpl_binds dump env (bind:binds) = do { env' <- trace dump bind $
233 ; simpl_binds dump env' binds }
235 trace True bind = pprTrace "SimplBind" (ppr (bindersOf bind))
236 trace False bind = \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 env top_lvl pairs
259 = do { let (env_with_info, triples) = mapAccumL add_rules env pairs
260 ; env' <- go (zapFloats env_with_info) triples
261 ; return (env `addRecFloats` env') }
262 -- addFloats adds the floats from env',
263 -- *and* updates env with the in-scope set from env'
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) = collectTyBinders rhs
326 ; (body_env, tvs') <- simplBinders rhs_env tvs
327 -- See Note [Floating and type abstraction]
330 -- Simplify the RHS; note the mkRhsStop, which tells
331 -- the simplifier that this is the RHS of a let.
332 ; let rhs_cont = mkRhsStop (applyTys (idType bndr1) (mkTyVarTys tvs'))
333 ; (body_env1, body1) <- simplExprF body_env body rhs_cont
335 -- ANF-ise a constructor or PAP rhs
336 ; (body_env2, body2) <- prepareRhs body_env1 body1
339 <- if not (doFloatFromRhs top_lvl is_rec False body2 body_env2)
340 then -- No floating, just wrap up!
341 do { rhs' <- mkLam tvs' (wrapFloats body_env2 body2)
342 ; return (env, rhs') }
344 else if null tvs then -- Simple floating
345 do { tick LetFloatFromLet
346 ; return (addFloats env body_env2, body2) }
348 else -- Do type-abstraction first
349 do { tick LetFloatFromLet
350 ; (poly_binds, body3) <- abstractFloats tvs' body_env2 body2
351 ; rhs' <- mkLam tvs' body3
352 ; return (extendFloats env poly_binds, rhs') }
354 ; completeBind env' top_lvl bndr bndr1 rhs' }
357 A specialised variant of simplNonRec used when the RHS is already simplified,
358 notably in knownCon. It uses case-binding where necessary.
361 simplNonRecX :: SimplEnv
362 -> InId -- Old binder
363 -> OutExpr -- Simplified RHS
366 simplNonRecX env bndr new_rhs
367 = do { (env, bndr') <- simplBinder env bndr
368 ; completeNonRecX env NotTopLevel NonRecursive
369 (isStrictId bndr) bndr bndr' new_rhs }
371 completeNonRecX :: SimplEnv
372 -> TopLevelFlag -> RecFlag -> Bool
373 -> InId -- Old binder
374 -> OutId -- New binder
375 -> OutExpr -- Simplified RHS
378 completeNonRecX env top_lvl is_rec is_strict old_bndr new_bndr new_rhs
379 = do { (env1, rhs1) <- prepareRhs (zapFloats env) new_rhs
381 if doFloatFromRhs top_lvl is_rec is_strict rhs1 env1
382 then do { tick LetFloatFromLet
383 ; return (addFloats env env1, rhs1) } -- Add the floats to the main env
384 else return (env, wrapFloats env1 rhs1) -- Wrap the floats around the RHS
385 ; completeBind env2 NotTopLevel old_bndr new_bndr rhs2 }
388 {- No, no, no! Do not try preInlineUnconditionally in completeNonRecX
389 Doing so risks exponential behaviour, because new_rhs has been simplified once already
390 In the cases described by the folowing commment, postInlineUnconditionally will
391 catch many of the relevant cases.
392 -- This happens; for example, the case_bndr during case of
393 -- known constructor: case (a,b) of x { (p,q) -> ... }
394 -- Here x isn't mentioned in the RHS, so we don't want to
395 -- create the (dead) let-binding let x = (a,b) in ...
397 -- Similarly, single occurrences can be inlined vigourously
398 -- e.g. case (f x, g y) of (a,b) -> ....
399 -- If a,b occur once we can avoid constructing the let binding for them.
401 Furthermore in the case-binding case preInlineUnconditionally risks extra thunks
402 -- Consider case I# (quotInt# x y) of
403 -- I# v -> let w = J# v in ...
404 -- If we gaily inline (quotInt# x y) for v, we end up building an
406 -- let w = J# (quotInt# x y) in ...
407 -- because quotInt# can fail.
409 | preInlineUnconditionally env NotTopLevel bndr new_rhs
410 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
413 ----------------------------------
414 prepareRhs takes a putative RHS, checks whether it's a PAP or
415 constructor application and, if so, converts it to ANF, so that the
416 resulting thing can be inlined more easily. Thus
423 We also want to deal well cases like this
424 v = (f e1 `cast` co) e2
425 Here we want to make e1,e2 trivial and get
426 x1 = e1; x2 = e2; v = (f x1 `cast` co) v2
427 That's what the 'go' loop in prepareRhs does
430 prepareRhs :: SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
431 -- Adds new floats to the env iff that allows us to return a good RHS
432 prepareRhs env (Cast rhs co) -- Note [Float coercions]
433 | (ty1, ty2) <- coercionKind co -- Do *not* do this if rhs has an unlifted type
434 , not (isUnLiftedType ty1) -- see Note [Float coercions (unlifted)]
435 = do { (env', rhs') <- makeTrivial env rhs
436 ; return (env', Cast rhs' co) }
439 = do { (is_val, env', rhs') <- go 0 env rhs
440 ; return (env', rhs') }
442 go n_val_args env (Cast rhs co)
443 = do { (is_val, env', rhs') <- go n_val_args env rhs
444 ; return (is_val, env', Cast rhs' co) }
445 go n_val_args env (App fun (Type ty))
446 = do { (is_val, env', rhs') <- go n_val_args env fun
447 ; return (is_val, env', App rhs' (Type ty)) }
448 go n_val_args env (App fun arg)
449 = do { (is_val, env', fun') <- go (n_val_args+1) env fun
451 True -> do { (env'', arg') <- makeTrivial env' arg
452 ; return (True, env'', App fun' arg') }
453 False -> return (False, env, App fun arg) }
454 go n_val_args env (Var fun)
455 = return (is_val, env, Var fun)
457 is_val = n_val_args > 0 -- There is at least one arg
458 -- ...and the fun a constructor or PAP
459 && (isDataConWorkId fun || n_val_args < idArity fun)
460 go n_val_args env other
461 = return (False, env, other)
465 Note [Float coercions]
466 ~~~~~~~~~~~~~~~~~~~~~~
467 When we find the binding
469 we'd like to transform it to
471 x = x `cast` co -- A trivial binding
472 There's a chance that e will be a constructor application or function, or something
473 like that, so moving the coerion to the usage site may well cancel the coersions
474 and lead to further optimisation. Example:
477 data instance T Int = T Int
479 foo :: Int -> Int -> Int
484 go n = case x of { T m -> go (n-m) }
485 -- This case should optimise
487 Note [Float coercions (unlifted)]
488 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
489 BUT don't do [Float coercions] if 'e' has an unlifted type.
492 foo :: Int = (error (# Int,Int #) "urk")
493 `cast` CoUnsafe (# Int,Int #) Int
495 If do the makeTrivial thing to the error call, we'll get
496 foo = case error (# Int,Int #) "urk" of v -> v `cast` ...
497 But 'v' isn't in scope!
499 These strange casts can happen as a result of case-of-case
500 bar = case (case x of { T -> (# 2,3 #); F -> error "urk" }) of
505 makeTrivial :: SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
506 -- Binds the expression to a variable, if it's not trivial, returning the variable
510 | otherwise -- See Note [Take care] below
511 = do { var <- newId FSLIT("a") (exprType expr)
512 ; env <- completeNonRecX env NotTopLevel NonRecursive
514 ; return (env, substExpr env (Var var)) }
518 %************************************************************************
520 \subsection{Completing a lazy binding}
522 %************************************************************************
525 * deals only with Ids, not TyVars
526 * takes an already-simplified binder and RHS
527 * is used for both recursive and non-recursive bindings
528 * is used for both top-level and non-top-level bindings
530 It does the following:
531 - tries discarding a dead binding
532 - tries PostInlineUnconditionally
533 - add unfolding [this is the only place we add an unfolding]
536 It does *not* attempt to do let-to-case. Why? Because it is used for
537 - top-level bindings (when let-to-case is impossible)
538 - many situations where the "rhs" is known to be a WHNF
539 (so let-to-case is inappropriate).
541 Nor does it do the atomic-argument thing
544 completeBind :: SimplEnv
545 -> TopLevelFlag -- Flag stuck into unfolding
546 -> InId -- Old binder
547 -> OutId -> OutExpr -- New binder and RHS
549 -- completeBind may choose to do its work
550 -- * by extending the substitution (e.g. let x = y in ...)
551 -- * or by adding to the floats in the envt
553 completeBind env top_lvl old_bndr new_bndr new_rhs
554 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
555 -- Inline and discard the binding
556 = do { tick (PostInlineUnconditionally old_bndr)
557 ; -- pprTrace "postInlineUnconditionally" (ppr old_bndr <+> ppr new_bndr <+> ppr new_rhs) $
558 return (extendIdSubst env old_bndr (DoneEx new_rhs)) }
559 -- Use the substitution to make quite, quite sure that the
560 -- substitution will happen, since we are going to discard the binding
565 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
568 -- Add the unfolding *only* for non-loop-breakers
569 -- Making loop breakers not have an unfolding at all
570 -- means that we can avoid tests in exprIsConApp, for example.
571 -- This is important: if exprIsConApp says 'yes' for a recursive
572 -- thing, then we can get into an infinite loop
575 -- If the unfolding is a value, the demand info may
576 -- go pear-shaped, so we nuke it. Example:
578 -- case x of (p,q) -> h p q x
579 -- Here x is certainly demanded. But after we've nuked
580 -- the case, we'll get just
581 -- let x = (a,b) in h a b x
582 -- and now x is not demanded (I'm assuming h is lazy)
583 -- This really happens. Similarly
584 -- let f = \x -> e in ...f..f...
585 -- After inlining f at some of its call sites the original binding may
586 -- (for example) be no longer strictly demanded.
587 -- The solution here is a bit ad hoc...
588 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
589 `setWorkerInfo` worker_info
591 final_info | loop_breaker = new_bndr_info
592 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
593 | otherwise = info_w_unf
595 final_id = new_bndr `setIdInfo` final_info
597 -- These seqs forces the Id, and hence its IdInfo,
598 -- and hence any inner substitutions
600 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
601 return (addNonRec env final_id new_rhs)
603 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
604 worker_info = substWorker env (workerInfo old_info)
605 loop_breaker = isNonRuleLoopBreaker occ_info
606 old_info = idInfo old_bndr
607 occ_info = occInfo old_info
612 %************************************************************************
614 \subsection[Simplify-simplExpr]{The main function: simplExpr}
616 %************************************************************************
618 The reason for this OutExprStuff stuff is that we want to float *after*
619 simplifying a RHS, not before. If we do so naively we get quadratic
620 behaviour as things float out.
622 To see why it's important to do it after, consider this (real) example:
636 a -- Can't inline a this round, cos it appears twice
640 Each of the ==> steps is a round of simplification. We'd save a
641 whole round if we float first. This can cascade. Consider
646 let f = let d1 = ..d.. in \y -> e
650 in \x -> ...(\y ->e)...
652 Only in this second round can the \y be applied, and it
653 might do the same again.
657 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
658 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
660 expr_ty' = substTy env (exprType expr)
661 -- The type in the Stop continuation, expr_ty', is usually not used
662 -- It's only needed when discarding continuations after finding
663 -- a function that returns bottom.
664 -- Hence the lazy substitution
667 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
668 -- Simplify an expression, given a continuation
669 simplExprC env expr cont
670 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
671 do { (env', expr') <- simplExprF (zapFloats env) expr cont
672 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
673 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
674 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
675 return (wrapFloats env' expr') }
677 --------------------------------------------------
678 simplExprF :: SimplEnv -> InExpr -> SimplCont
679 -> SimplM (SimplEnv, OutExpr)
681 simplExprF env e cont
682 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
683 simplExprF' env e cont
685 simplExprF' env (Var v) cont = simplVar env v cont
686 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
687 simplExprF' env (Note n expr) cont = simplNote env n expr cont
688 simplExprF' env (Cast body co) cont = simplCast env body co cont
689 simplExprF' env (App fun arg) cont = simplExprF env fun $
690 ApplyTo NoDup arg env cont
692 simplExprF' env expr@(Lam _ _) cont
693 = simplLam env (map zap bndrs) body cont
694 -- The main issue here is under-saturated lambdas
695 -- (\x1. \x2. e) arg1
696 -- Here x1 might have "occurs-once" occ-info, because occ-info
697 -- is computed assuming that a group of lambdas is applied
698 -- all at once. If there are too few args, we must zap the
701 n_args = countArgs cont
702 n_params = length bndrs
703 (bndrs, body) = collectBinders expr
704 zap | n_args >= n_params = \b -> b
705 | otherwise = \b -> if isTyVar b then b
707 -- NB: we count all the args incl type args
708 -- so we must count all the binders (incl type lambdas)
710 simplExprF' env (Type ty) cont
711 = ASSERT( contIsRhsOrArg cont )
712 do { ty' <- simplType env ty
713 ; rebuild env (Type ty') cont }
715 simplExprF' env (Case scrut bndr case_ty alts) cont
716 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
717 = -- Simplify the scrutinee with a Select continuation
718 simplExprF env scrut (Select NoDup bndr alts env cont)
721 = -- If case-of-case is off, simply simplify the case expression
722 -- in a vanilla Stop context, and rebuild the result around it
723 do { case_expr' <- simplExprC env scrut case_cont
724 ; rebuild env case_expr' cont }
726 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
727 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
729 simplExprF' env (Let (Rec pairs) body) cont
730 = do { env <- simplRecBndrs env (map fst pairs)
731 -- NB: bndrs' don't have unfoldings or rules
732 -- We add them as we go down
734 ; env <- simplRecBind env NotTopLevel pairs
735 ; simplExprF env body cont }
737 simplExprF' env (Let (NonRec bndr rhs) body) cont
738 = simplNonRecE env bndr (rhs, env) ([], body) cont
740 ---------------------------------
741 simplType :: SimplEnv -> InType -> SimplM OutType
742 -- Kept monadic just so we can do the seqType
744 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
745 seqType new_ty `seq` return new_ty
747 new_ty = substTy env ty
751 %************************************************************************
753 \subsection{The main rebuilder}
755 %************************************************************************
758 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
759 -- At this point the substitution in the SimplEnv should be irrelevant
760 -- only the in-scope set and floats should matter
761 rebuild env expr cont
762 = -- pprTrace "rebuild" (ppr expr $$ ppr cont $$ ppr (seFloats env)) $
764 Stop {} -> return (env, expr)
765 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
766 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
767 StrictArg fun ty _ info cont -> rebuildCall env (fun `App` expr) (funResultTy ty) info cont
768 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
769 ; simplLam env' bs body cont }
770 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
771 ; rebuild env (App expr arg') cont }
775 %************************************************************************
779 %************************************************************************
782 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
783 -> SimplM (SimplEnv, OutExpr)
784 simplCast env body co cont
785 = do { co' <- simplType env co
786 ; simplExprF env body (addCoerce co' cont) }
788 addCoerce co cont = add_coerce co (coercionKind co) cont
790 add_coerce co (s1, k1) cont -- co :: ty~ty
791 | s1 `coreEqType` k1 = cont -- is a no-op
793 add_coerce co1 (s1, k2) (CoerceIt co2 cont)
794 | (l1, t1) <- coercionKind co2
795 -- coerce T1 S1 (coerce S1 K1 e)
798 -- coerce T1 K1 e, otherwise
800 -- For example, in the initial form of a worker
801 -- we may find (coerce T (coerce S (\x.e))) y
802 -- and we'd like it to simplify to e[y/x] in one round
804 , s1 `coreEqType` t1 = cont -- The coerces cancel out
805 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
807 add_coerce co (s1s2, t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
808 -- (f `cast` g) ty ---> (f ty) `cast` (g @ ty)
809 -- This implements the PushT rule from the paper
810 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
811 , not (isCoVar tyvar)
812 = ApplyTo dup (Type ty') (zapSubstEnv env) (addCoerce (mkInstCoercion co ty') cont)
814 ty' = substTy arg_se arg_ty
816 -- ToDo: the PushC rule is not implemented at all
818 add_coerce co (s1s2, t1t2) (ApplyTo dup arg arg_se cont)
819 | not (isTypeArg arg) -- This implements the Push rule from the paper
820 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
821 -- co : s1s2 :=: t1t2
822 -- (coerce (T1->T2) (S1->S2) F) E
824 -- coerce T2 S2 (F (coerce S1 T1 E))
826 -- t1t2 must be a function type, T1->T2, because it's applied
827 -- to something but s1s2 might conceivably not be
829 -- When we build the ApplyTo we can't mix the out-types
830 -- with the InExpr in the argument, so we simply substitute
831 -- to make it all consistent. It's a bit messy.
832 -- But it isn't a common case.
834 -- Example of use: Trac #995
835 = ApplyTo dup new_arg (zapSubstEnv env) (addCoerce co2 cont)
837 -- we split coercion t1->t2 :=: s1->s2 into t1 :=: s1 and
838 -- t2 :=: s2 with left and right on the curried form:
839 -- (->) t1 t2 :=: (->) s1 s2
840 [co1, co2] = decomposeCo 2 co
841 new_arg = mkCoerce (mkSymCoercion co1) arg'
842 arg' = substExpr arg_se arg
844 add_coerce co _ cont = CoerceIt co cont
848 %************************************************************************
852 %************************************************************************
855 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
856 -> SimplM (SimplEnv, OutExpr)
858 simplLam env [] body cont = simplExprF env body cont
860 -- Type-beta reduction
861 simplLam env (bndr:bndrs) body (ApplyTo _ (Type ty_arg) arg_se cont)
862 = ASSERT( isTyVar bndr )
863 do { tick (BetaReduction bndr)
864 ; ty_arg' <- simplType (arg_se `setInScope` env) ty_arg
865 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
867 -- Ordinary beta reduction
868 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
869 = do { tick (BetaReduction bndr)
870 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
872 -- Not enough args, so there are real lambdas left to put in the result
873 simplLam env bndrs body cont
874 = do { (env, bndrs') <- simplLamBndrs env bndrs
875 ; body' <- simplExpr env body
876 ; new_lam <- mkLam bndrs' body'
877 ; rebuild env new_lam cont }
880 simplNonRecE :: SimplEnv
881 -> InId -- The binder
882 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
883 -> ([InId], InExpr) -- Body of the let/lambda
886 -> SimplM (SimplEnv, OutExpr)
888 -- simplNonRecE is used for
889 -- * non-top-level non-recursive lets in expressions
892 -- It deals with strict bindings, via the StrictBind continuation,
893 -- which may abort the whole process
895 -- The "body" of the binding comes as a pair of ([InId],InExpr)
896 -- representing a lambda; so we recurse back to simplLam
897 -- Why? Because of the binder-occ-info-zapping done before
898 -- the call to simplLam in simplExprF (Lam ...)
900 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
901 | preInlineUnconditionally env NotTopLevel bndr rhs
902 = do { tick (PreInlineUnconditionally bndr)
903 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
906 = do { simplExprF (rhs_se `setFloats` env) rhs
907 (StrictBind bndr bndrs body env cont) }
910 = do { (env1, bndr1) <- simplNonRecBndr env bndr
911 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
912 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
913 ; simplLam env3 bndrs body cont }
917 %************************************************************************
921 %************************************************************************
924 -- Hack alert: we only distinguish subsumed cost centre stacks for the
925 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
926 simplNote env (SCC cc) e cont
927 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
928 ; rebuild env (mkSCC cc e') cont }
930 -- See notes with SimplMonad.inlineMode
931 simplNote env InlineMe e cont
932 | Just (inside, outside) <- splitInlineCont cont -- Boring boring continuation; see notes above
933 = do { -- Don't inline inside an INLINE expression
934 e' <- simplExprC (setMode inlineMode env) e inside
935 ; rebuild env (mkInlineMe e') outside }
937 | otherwise -- Dissolve the InlineMe note if there's
938 -- an interesting context of any kind to combine with
939 -- (even a type application -- anything except Stop)
940 = simplExprF env e cont
942 simplNote env (CoreNote s) e cont = do
943 e' <- simplExpr env e
944 rebuild env (Note (CoreNote s) e') cont
948 %************************************************************************
950 \subsection{Dealing with calls}
952 %************************************************************************
955 simplVar env var cont
956 = case substId env var of
957 DoneEx e -> simplExprF (zapSubstEnv env) e cont
958 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
959 DoneId var1 -> completeCall (zapSubstEnv env) var1 cont
960 -- Note [zapSubstEnv]
961 -- The template is already simplified, so don't re-substitute.
962 -- This is VITAL. Consider
964 -- let y = \z -> ...x... in
966 -- We'll clone the inner \x, adding x->x' in the id_subst
967 -- Then when we inline y, we must *not* replace x by x' in
968 -- the inlined copy!!
970 ---------------------------------------------------------
971 -- Dealing with a call site
973 completeCall env var cont
974 = do { dflags <- getDOptsSmpl
975 ; let (args,call_cont) = contArgs cont
976 -- The args are OutExprs, obtained by *lazily* substituting
977 -- in the args found in cont. These args are only examined
978 -- to limited depth (unless a rule fires). But we must do
979 -- the substitution; rule matching on un-simplified args would
982 ------------- First try rules ----------------
983 -- Do this before trying inlining. Some functions have
984 -- rules *and* are strict; in this case, we don't want to
985 -- inline the wrapper of the non-specialised thing; better
986 -- to call the specialised thing instead.
988 -- We used to use the black-listing mechanism to ensure that inlining of
989 -- the wrapper didn't occur for things that have specialisations till a
990 -- later phase, so but now we just try RULES first
992 -- Note [Rules for recursive functions]
993 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
994 -- You might think that we shouldn't apply rules for a loop breaker:
995 -- doing so might give rise to an infinite loop, because a RULE is
996 -- rather like an extra equation for the function:
997 -- RULE: f (g x) y = x+y
1000 -- But it's too drastic to disable rules for loop breakers.
1001 -- Even the foldr/build rule would be disabled, because foldr
1002 -- is recursive, and hence a loop breaker:
1003 -- foldr k z (build g) = g k z
1004 -- So it's up to the programmer: rules can cause divergence
1006 ; let in_scope = getInScope env
1007 maybe_rule = case activeRule dflags env of
1008 Nothing -> Nothing -- No rules apply
1009 Just act_fn -> lookupRule act_fn in_scope
1011 ; case maybe_rule of {
1012 Just (rule, rule_rhs) -> do
1013 tick (RuleFired (ru_name rule))
1014 (if dopt Opt_D_dump_rule_firings dflags then
1015 pprTrace "Rule fired" (vcat [
1016 text "Rule:" <+> ftext (ru_name rule),
1017 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1018 text "After: " <+> pprCoreExpr rule_rhs,
1019 text "Cont: " <+> ppr call_cont])
1022 simplExprF env rule_rhs (dropArgs (ruleArity rule) cont)
1023 -- The ruleArity says how many args the rule consumed
1025 ; Nothing -> do -- No rules
1027 ------------- Next try inlining ----------------
1028 { let arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1029 n_val_args = length arg_infos
1030 interesting_cont = interestingCallContext call_cont
1031 active_inline = activeInline env var
1032 maybe_inline = callSiteInline dflags active_inline var
1033 (null args) arg_infos interesting_cont
1034 ; case maybe_inline of {
1035 Just unfolding -- There is an inlining!
1036 -> do { tick (UnfoldingDone var)
1037 ; (if dopt Opt_D_dump_inlinings dflags then
1038 pprTrace ("Inlining done" ++ showSDoc (ppr var)) (vcat [
1039 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1040 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1041 text "Cont: " <+> ppr call_cont])
1044 simplExprF env unfolding cont }
1046 ; Nothing -> -- No inlining!
1048 ------------- No inlining! ----------------
1049 -- Next, look for rules or specialisations that match
1051 rebuildCall env (Var var) (idType var)
1052 (mkArgInfo var n_val_args call_cont) cont
1055 rebuildCall :: SimplEnv
1056 -> OutExpr -> OutType -- Function and its type
1059 -> SimplM (SimplEnv, OutExpr)
1060 rebuildCall env fun fun_ty (ArgInfo { ai_strs = [] }) cont
1061 -- When we run out of strictness args, it means
1062 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1063 -- Then we want to discard the entire strict continuation. E.g.
1064 -- * case (error "hello") of { ... }
1065 -- * (error "Hello") arg
1066 -- * f (error "Hello") where f is strict
1068 -- Then, especially in the first of these cases, we'd like to discard
1069 -- the continuation, leaving just the bottoming expression. But the
1070 -- type might not be right, so we may have to add a coerce.
1071 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1072 = return (env, mk_coerce fun) -- contination to discard, else we do it
1073 where -- again and again!
1074 cont_ty = contResultType cont
1075 co = mkUnsafeCoercion fun_ty cont_ty
1076 mk_coerce expr | cont_ty `coreEqType` fun_ty = fun
1077 | otherwise = mkCoerce co fun
1079 rebuildCall env fun fun_ty info (ApplyTo _ (Type arg_ty) se cont)
1080 = do { ty' <- simplType (se `setInScope` env) arg_ty
1081 ; rebuildCall env (fun `App` Type ty') (applyTy fun_ty ty') info cont }
1083 rebuildCall env fun fun_ty
1084 (ArgInfo { ai_rules = has_rules, ai_strs = str:strs, ai_discs = disc:discs })
1085 (ApplyTo _ arg arg_se cont)
1086 | str || isStrictType arg_ty -- Strict argument
1087 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1088 simplExprF (arg_se `setFloats` env) arg
1089 (StrictArg fun fun_ty cci arg_info' cont)
1092 | otherwise -- Lazy argument
1093 -- DO NOT float anything outside, hence simplExprC
1094 -- There is no benefit (unlike in a let-binding), and we'd
1095 -- have to be very careful about bogus strictness through
1096 -- floating a demanded let.
1097 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1098 (mkLazyArgStop arg_ty cci)
1099 ; rebuildCall env (fun `App` arg') res_ty arg_info' cont }
1101 (arg_ty, res_ty) = splitFunTy fun_ty
1102 arg_info' = ArgInfo { ai_rules = has_rules, ai_strs = strs, ai_discs = discs }
1103 cci | has_rules || disc > 0 = ArgCtxt has_rules disc -- Be keener here
1104 | otherwise = BoringCtxt -- Nothing interesting
1106 rebuildCall env fun fun_ty info cont
1107 = rebuild env fun cont
1112 This part of the simplifier may break the no-shadowing invariant
1114 f (...(\a -> e)...) (case y of (a,b) -> e')
1115 where f is strict in its second arg
1116 If we simplify the innermost one first we get (...(\a -> e)...)
1117 Simplifying the second arg makes us float the case out, so we end up with
1118 case y of (a,b) -> f (...(\a -> e)...) e'
1119 So the output does not have the no-shadowing invariant. However, there is
1120 no danger of getting name-capture, because when the first arg was simplified
1121 we used an in-scope set that at least mentioned all the variables free in its
1122 static environment, and that is enough.
1124 We can't just do innermost first, or we'd end up with a dual problem:
1125 case x of (a,b) -> f e (...(\a -> e')...)
1127 I spent hours trying to recover the no-shadowing invariant, but I just could
1128 not think of an elegant way to do it. The simplifier is already knee-deep in
1129 continuations. We have to keep the right in-scope set around; AND we have
1130 to get the effect that finding (error "foo") in a strict arg position will
1131 discard the entire application and replace it with (error "foo"). Getting
1132 all this at once is TOO HARD!
1134 %************************************************************************
1136 Rebuilding a cse expression
1138 %************************************************************************
1140 Blob of helper functions for the "case-of-something-else" situation.
1143 ---------------------------------------------------------
1144 -- Eliminate the case if possible
1146 rebuildCase :: SimplEnv
1147 -> OutExpr -- Scrutinee
1148 -> InId -- Case binder
1149 -> [InAlt] -- Alternatives (inceasing order)
1151 -> SimplM (SimplEnv, OutExpr)
1153 --------------------------------------------------
1154 -- 1. Eliminate the case if there's a known constructor
1155 --------------------------------------------------
1157 rebuildCase env scrut case_bndr alts cont
1158 | Just (con,args) <- exprIsConApp_maybe scrut
1159 -- Works when the scrutinee is a variable with a known unfolding
1160 -- as well as when it's an explicit constructor application
1161 = knownCon env scrut (DataAlt con) args case_bndr alts cont
1163 | Lit lit <- scrut -- No need for same treatment as constructors
1164 -- because literals are inlined more vigorously
1165 = knownCon env scrut (LitAlt lit) [] case_bndr alts cont
1168 --------------------------------------------------
1169 -- 2. Eliminate the case if scrutinee is evaluated
1170 --------------------------------------------------
1172 rebuildCase env scrut case_bndr [(con,bndrs,rhs)] cont
1173 -- See if we can get rid of the case altogether
1174 -- See the extensive notes on case-elimination above
1175 -- mkCase made sure that if all the alternatives are equal,
1176 -- then there is now only one (DEFAULT) rhs
1177 | all isDeadBinder bndrs -- bndrs are [InId]
1179 -- Check that the scrutinee can be let-bound instead of case-bound
1180 , exprOkForSpeculation scrut
1181 -- OK not to evaluate it
1182 -- This includes things like (==# a# b#)::Bool
1183 -- so that we simplify
1184 -- case ==# a# b# of { True -> x; False -> x }
1187 -- This particular example shows up in default methods for
1188 -- comparision operations (e.g. in (>=) for Int.Int32)
1189 || exprIsHNF scrut -- It's already evaluated
1190 || var_demanded_later scrut -- It'll be demanded later
1192 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1193 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1194 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1195 -- its argument: case x of { y -> dataToTag# y }
1196 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1197 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1199 -- Also we don't want to discard 'seq's
1200 = do { tick (CaseElim case_bndr)
1201 ; env <- simplNonRecX env case_bndr scrut
1202 ; simplExprF env rhs cont }
1204 -- The case binder is going to be evaluated later,
1205 -- and the scrutinee is a simple variable
1206 var_demanded_later (Var v) = isStrictDmd (idNewDemandInfo case_bndr)
1207 && not (isTickBoxOp v)
1208 -- ugly hack; covering this case is what
1209 -- exprOkForSpeculation was intended for.
1210 var_demanded_later other = False
1213 --------------------------------------------------
1214 -- 3. Catch-all case
1215 --------------------------------------------------
1217 rebuildCase env scrut case_bndr alts cont
1218 = do { -- Prepare the continuation;
1219 -- The new subst_env is in place
1220 (env, dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1222 -- Simplify the alternatives
1223 ; (scrut', case_bndr', alts') <- simplAlts env scrut case_bndr alts dup_cont
1224 ; let res_ty' = contResultType dup_cont
1225 ; case_expr <- mkCase scrut' case_bndr' res_ty' alts'
1227 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1228 -- The case binder *not* scope over the whole returned case-expression
1229 ; rebuild env case_expr nodup_cont }
1232 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1233 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1234 way, there's a chance that v will now only be used once, and hence
1237 Note [no-case-of-case]
1238 ~~~~~~~~~~~~~~~~~~~~~~
1239 There is a time we *don't* want to do that, namely when
1240 -fno-case-of-case is on. This happens in the first simplifier pass,
1241 and enhances full laziness. Here's the bad case:
1242 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1243 If we eliminate the inner case, we trap it inside the I# v -> arm,
1244 which might prevent some full laziness happening. I've seen this
1245 in action in spectral/cichelli/Prog.hs:
1246 [(m,n) | m <- [1..max], n <- [1..max]]
1247 Hence the check for NoCaseOfCase.
1249 Note [Suppressing the case binder-swap]
1250 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1251 There is another situation when it might make sense to suppress the
1252 case-expression binde-swap. If we have
1254 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1255 ...other cases .... }
1257 We'll perform the binder-swap for the outer case, giving
1259 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1260 ...other cases .... }
1262 But there is no point in doing it for the inner case, because w1 can't
1263 be inlined anyway. Furthermore, doing the case-swapping involves
1264 zapping w2's occurrence info (see paragraphs that follow), and that
1265 forces us to bind w2 when doing case merging. So we get
1267 case x of w1 { A -> let w2 = w1 in e1
1268 B -> let w2 = w1 in e2
1269 ...other cases .... }
1271 This is plain silly in the common case where w2 is dead.
1273 Even so, I can't see a good way to implement this idea. I tried
1274 not doing the binder-swap if the scrutinee was already evaluated
1275 but that failed big-time:
1279 case v of w { MkT x ->
1280 case x of x1 { I# y1 ->
1281 case x of x2 { I# y2 -> ...
1283 Notice that because MkT is strict, x is marked "evaluated". But to
1284 eliminate the last case, we must either make sure that x (as well as
1285 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1286 the binder-swap. So this whole note is a no-op.
1290 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1291 any occurrence info (eg IAmDead) in the case binder, because the
1292 case-binder now effectively occurs whenever v does. AND we have to do
1293 the same for the pattern-bound variables! Example:
1295 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1297 Here, b and p are dead. But when we move the argment inside the first
1298 case RHS, and eliminate the second case, we get
1300 case x of { (a,b) -> a b }
1302 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1305 Indeed, this can happen anytime the case binder isn't dead:
1306 case <any> of x { (a,b) ->
1307 case x of { (p,q) -> p } }
1308 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1309 The point is that we bring into the envt a binding
1311 after the outer case, and that makes (a,b) alive. At least we do unless
1312 the case binder is guaranteed dead.
1316 Consider case (v `cast` co) of x { I# ->
1317 ... (case (v `cast` co) of {...}) ...
1318 We'd like to eliminate the inner case. We can get this neatly by
1319 arranging that inside the outer case we add the unfolding
1320 v |-> x `cast` (sym co)
1321 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1323 Note [Improving seq]
1326 type family F :: * -> *
1327 type instance F Int = Int
1329 ... case e of x { DEFAULT -> rhs } ...
1331 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1333 case e `cast` co of x'::Int
1334 I# x# -> let x = x' `cast` sym co
1337 so that 'rhs' can take advantage of the form of x'. Notice that Note
1338 [Case of cast] may then apply to the result.
1340 This showed up in Roman's experiments. Example:
1341 foo :: F Int -> Int -> Int
1342 foo t n = t `seq` bar n
1345 bar n = bar (n - case t of TI i -> i)
1346 Here we'd like to avoid repeated evaluating t inside the loop, by
1347 taking advantage of the `seq`.
1349 At one point I did transformation in LiberateCase, but it's more robust here.
1350 (Otherwise, there's a danger that we'll simply drop the 'seq' altogether, before
1351 LiberateCase gets to see it.)
1353 Note [Case elimination]
1354 ~~~~~~~~~~~~~~~~~~~~~~~
1355 The case-elimination transformation discards redundant case expressions.
1356 Start with a simple situation:
1358 case x# of ===> e[x#/y#]
1361 (when x#, y# are of primitive type, of course). We can't (in general)
1362 do this for algebraic cases, because we might turn bottom into
1365 The code in SimplUtils.prepareAlts has the effect of generalise this
1366 idea to look for a case where we're scrutinising a variable, and we
1367 know that only the default case can match. For example:
1371 DEFAULT -> ...(case x of
1375 Here the inner case is first trimmed to have only one alternative, the
1376 DEFAULT, after which it's an instance of the previous case. This
1377 really only shows up in eliminating error-checking code.
1379 We also make sure that we deal with this very common case:
1384 Here we are using the case as a strict let; if x is used only once
1385 then we want to inline it. We have to be careful that this doesn't
1386 make the program terminate when it would have diverged before, so we
1388 - e is already evaluated (it may so if e is a variable)
1389 - x is used strictly, or
1391 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1393 case e of ===> case e of DEFAULT -> r
1397 Now again the case may be elminated by the CaseElim transformation.
1400 Further notes about case elimination
1401 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1402 Consider: test :: Integer -> IO ()
1405 Turns out that this compiles to:
1408 eta1 :: State# RealWorld ->
1409 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1411 (PrelNum.jtos eta ($w[] @ Char))
1413 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1415 Notice the strange '<' which has no effect at all. This is a funny one.
1416 It started like this:
1418 f x y = if x < 0 then jtos x
1419 else if y==0 then "" else jtos x
1421 At a particular call site we have (f v 1). So we inline to get
1423 if v < 0 then jtos x
1424 else if 1==0 then "" else jtos x
1426 Now simplify the 1==0 conditional:
1428 if v<0 then jtos v else jtos v
1430 Now common-up the two branches of the case:
1432 case (v<0) of DEFAULT -> jtos v
1434 Why don't we drop the case? Because it's strict in v. It's technically
1435 wrong to drop even unnecessary evaluations, and in practice they
1436 may be a result of 'seq' so we *definitely* don't want to drop those.
1437 I don't really know how to improve this situation.
1441 simplCaseBinder :: SimplEnv -> OutExpr -> OutId -> [InAlt]
1442 -> SimplM (SimplEnv, OutExpr, OutId)
1443 simplCaseBinder env scrut case_bndr alts
1444 = do { (env1, case_bndr1) <- simplBinder env case_bndr
1446 ; fam_envs <- getFamEnvs
1447 ; (env2, scrut2, case_bndr2) <- improve_seq fam_envs env1 scrut
1448 case_bndr case_bndr1 alts
1449 -- Note [Improving seq]
1451 ; let (env3, case_bndr3) = improve_case_bndr env2 scrut2 case_bndr2
1452 -- Note [Case of cast]
1454 ; return (env3, scrut2, case_bndr3) }
1457 improve_seq fam_envs env1 scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1458 | Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1459 = do { case_bndr2 <- newId FSLIT("nt") ty2
1460 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1461 env2 = extendIdSubst env1 case_bndr rhs
1462 ; return (env2, scrut `Cast` co, case_bndr2) }
1464 improve_seq fam_envs env1 scrut case_bndr case_bndr1 alts
1465 = return (env1, scrut, case_bndr1)
1468 improve_case_bndr env scrut case_bndr
1469 | switchIsOn (getSwitchChecker env) NoCaseOfCase
1470 -- See Note [no-case-of-case]
1473 | otherwise -- Failed try; see Note [Suppressing the case binder-swap]
1474 -- not (isEvaldUnfolding (idUnfolding v))
1476 Var v -> (modifyInScope env1 v case_bndr', case_bndr')
1477 -- Note about using modifyInScope for v here
1478 -- We could extend the substitution instead, but it would be
1479 -- a hack because then the substitution wouldn't be idempotent
1480 -- any more (v is an OutId). And this does just as well.
1482 Cast (Var v) co -> (addBinderUnfolding env1 v rhs, case_bndr')
1484 rhs = Cast (Var case_bndr') (mkSymCoercion co)
1486 other -> (env, case_bndr)
1488 case_bndr' = zapOccInfo case_bndr
1489 env1 = modifyInScope env case_bndr case_bndr'
1492 zapOccInfo :: InId -> InId -- See Note [zapOccInfo]
1493 zapOccInfo b = b `setIdOccInfo` NoOccInfo
1497 simplAlts does two things:
1499 1. Eliminate alternatives that cannot match, including the
1500 DEFAULT alternative.
1502 2. If the DEFAULT alternative can match only one possible constructor,
1503 then make that constructor explicit.
1505 case e of x { DEFAULT -> rhs }
1507 case e of x { (a,b) -> rhs }
1508 where the type is a single constructor type. This gives better code
1509 when rhs also scrutinises x or e.
1511 Here "cannot match" includes knowledge from GADTs
1513 It's a good idea do do this stuff before simplifying the alternatives, to
1514 avoid simplifying alternatives we know can't happen, and to come up with
1515 the list of constructors that are handled, to put into the IdInfo of the
1516 case binder, for use when simplifying the alternatives.
1518 Eliminating the default alternative in (1) isn't so obvious, but it can
1521 data Colour = Red | Green | Blue
1530 DEFAULT -> [ case y of ... ]
1532 If we inline h into f, the default case of the inlined h can't happen.
1533 If we don't notice this, we may end up filtering out *all* the cases
1534 of the inner case y, which give us nowhere to go!
1538 simplAlts :: SimplEnv
1540 -> InId -- Case binder
1541 -> [InAlt] -> SimplCont
1542 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1543 -- Like simplExpr, this just returns the simplified alternatives;
1544 -- it not return an environment
1546 simplAlts env scrut case_bndr alts cont'
1547 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1548 do { let alt_env = zapFloats env
1549 ; (alt_env, scrut', case_bndr') <- simplCaseBinder alt_env scrut case_bndr alts
1551 ; (imposs_deflt_cons, in_alts) <- prepareAlts alt_env scrut case_bndr' alts
1553 ; alts' <- mapM (simplAlt alt_env imposs_deflt_cons case_bndr' cont') in_alts
1554 ; return (scrut', case_bndr', alts') }
1556 ------------------------------------
1557 simplAlt :: SimplEnv
1558 -> [AltCon] -- These constructors can't be present when
1559 -- matching the DEFAULT alternative
1560 -> OutId -- The case binder
1565 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1566 = ASSERT( null bndrs )
1567 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1568 -- Record the constructors that the case-binder *can't* be.
1569 ; rhs' <- simplExprC env' rhs cont'
1570 ; return (DEFAULT, [], rhs') }
1572 simplAlt env imposs_deflt_cons case_bndr' cont' (LitAlt lit, bndrs, rhs)
1573 = ASSERT( null bndrs )
1574 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1575 ; rhs' <- simplExprC env' rhs cont'
1576 ; return (LitAlt lit, [], rhs') }
1578 simplAlt env imposs_deflt_cons case_bndr' cont' (DataAlt con, vs, rhs)
1579 = do { -- Deal with the pattern-bound variables
1580 -- Mark the ones that are in ! positions in the
1581 -- data constructor as certainly-evaluated.
1582 -- NB: simplLamBinders preserves this eval info
1583 let vs_with_evals = add_evals vs (dataConRepStrictness con)
1584 ; (env, vs') <- simplLamBndrs env vs_with_evals
1586 -- Bind the case-binder to (con args)
1587 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1588 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1589 env' = addBinderUnfolding env case_bndr' (mkConApp con con_args)
1591 ; rhs' <- simplExprC env' rhs cont'
1592 ; return (DataAlt con, vs', rhs') }
1594 -- add_evals records the evaluated-ness of the bound variables of
1595 -- a case pattern. This is *important*. Consider
1596 -- data T = T !Int !Int
1598 -- case x of { T a b -> T (a+1) b }
1600 -- We really must record that b is already evaluated so that we don't
1601 -- go and re-evaluate it when constructing the result.
1602 -- See Note [Data-con worker strictness] in MkId.lhs
1607 go (v:vs) strs | isTyVar v = v : go vs strs
1608 go (v:vs) (str:strs)
1609 | isMarkedStrict str = evald_v : go vs strs
1610 | otherwise = zapped_v : go vs strs
1612 zapped_v = zap_occ_info v
1613 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1614 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr strs)
1616 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1618 -- to the envt; so vs are now very much alive
1619 -- Note [Aug06] I can't see why this actually matters, but it's neater
1620 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1621 -- ==> case e of t { (a,b) -> ...(a)... }
1622 -- Look, Ma, a is alive now.
1623 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1624 | otherwise = zapOccInfo
1626 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1627 addBinderUnfolding env bndr rhs
1628 = modifyInScope env bndr (bndr `setIdUnfolding` mkUnfolding False rhs)
1630 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1631 addBinderOtherCon env bndr cons
1632 = modifyInScope env bndr (bndr `setIdUnfolding` mkOtherCon cons)
1636 %************************************************************************
1638 \subsection{Known constructor}
1640 %************************************************************************
1642 We are a bit careful with occurrence info. Here's an example
1644 (\x* -> case x of (a*, b) -> f a) (h v, e)
1646 where the * means "occurs once". This effectively becomes
1647 case (h v, e) of (a*, b) -> f a)
1649 let a* = h v; b = e in f a
1653 All this should happen in one sweep.
1656 knownCon :: SimplEnv -> OutExpr -> AltCon -> [OutExpr]
1657 -> InId -> [InAlt] -> SimplCont
1658 -> SimplM (SimplEnv, OutExpr)
1660 knownCon env scrut con args bndr alts cont
1661 = do { tick (KnownBranch bndr)
1662 ; knownAlt env scrut args bndr (findAlt con alts) cont }
1664 knownAlt env scrut args bndr (DEFAULT, bs, rhs) cont
1666 do { env <- simplNonRecX env bndr scrut
1667 -- This might give rise to a binding with non-atomic args
1668 -- like x = Node (f x) (g x)
1669 -- but simplNonRecX will atomic-ify it
1670 ; simplExprF env rhs cont }
1672 knownAlt env scrut args bndr (LitAlt lit, bs, rhs) cont
1674 do { env <- simplNonRecX env bndr scrut
1675 ; simplExprF env rhs cont }
1677 knownAlt env scrut args bndr (DataAlt dc, bs, rhs) cont
1678 = do { let dead_bndr = isDeadBinder bndr -- bndr is an InId
1679 n_drop_tys = length (dataConUnivTyVars dc)
1680 ; env <- bind_args env dead_bndr bs (drop n_drop_tys args)
1682 -- It's useful to bind bndr to scrut, rather than to a fresh
1683 -- binding x = Con arg1 .. argn
1684 -- because very often the scrut is a variable, so we avoid
1685 -- creating, and then subsequently eliminating, a let-binding
1686 -- BUT, if scrut is a not a variable, we must be careful
1687 -- about duplicating the arg redexes; in that case, make
1688 -- a new con-app from the args
1689 bndr_rhs = case scrut of
1692 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1693 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1694 -- args are aready OutExprs, but bs are InIds
1696 ; env <- simplNonRecX env bndr bndr_rhs
1697 ; -- pprTrace "knownCon2" (ppr bs $$ ppr rhs $$ ppr (seIdSubst env)) $
1698 simplExprF env rhs cont }
1701 bind_args env dead_bndr [] _ = return env
1703 bind_args env dead_bndr (b:bs) (Type ty : args)
1704 = ASSERT( isTyVar b )
1705 bind_args (extendTvSubst env b ty) dead_bndr bs args
1707 bind_args env dead_bndr (b:bs) (arg : args)
1709 do { let b' = if dead_bndr then b else zapOccInfo b
1710 -- Note that the binder might be "dead", because it doesn't occur
1711 -- in the RHS; and simplNonRecX may therefore discard it via postInlineUnconditionally
1712 -- Nevertheless we must keep it if the case-binder is alive, because it may
1713 -- be used in the con_app. See Note [zapOccInfo]
1714 ; env <- simplNonRecX env b' arg
1715 ; bind_args env dead_bndr bs args }
1718 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr args $$
1719 text "scrut:" <+> ppr scrut
1723 %************************************************************************
1725 \subsection{Duplicating continuations}
1727 %************************************************************************
1730 prepareCaseCont :: SimplEnv
1731 -> [InAlt] -> SimplCont
1732 -> SimplM (SimplEnv, SimplCont,SimplCont)
1733 -- Return a duplicatable continuation, a non-duplicable part
1734 -- plus some extra bindings (that scope over the entire
1737 -- No need to make it duplicatable if there's only one alternative
1738 prepareCaseCont env [alt] cont = return (env, cont, mkBoringStop (contResultType cont))
1739 prepareCaseCont env alts cont = mkDupableCont env cont
1743 mkDupableCont :: SimplEnv -> SimplCont
1744 -> SimplM (SimplEnv, SimplCont, SimplCont)
1746 mkDupableCont env cont
1747 | contIsDupable cont
1748 = return (env, cont, mkBoringStop (contResultType cont))
1750 mkDupableCont env (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1752 mkDupableCont env (CoerceIt ty cont)
1753 = do { (env, dup, nodup) <- mkDupableCont env cont
1754 ; return (env, CoerceIt ty dup, nodup) }
1756 mkDupableCont env cont@(StrictBind bndr _ _ se _)
1757 = return (env, mkBoringStop (substTy se (idType bndr)), cont)
1758 -- See Note [Duplicating strict continuations]
1760 mkDupableCont env cont@(StrictArg _ fun_ty _ _ _)
1761 = return (env, mkBoringStop (funArgTy fun_ty), cont)
1762 -- See Note [Duplicating strict continuations]
1764 mkDupableCont env (ApplyTo _ arg se cont)
1765 = -- e.g. [...hole...] (...arg...)
1767 -- let a = ...arg...
1768 -- in [...hole...] a
1769 do { (env, dup_cont, nodup_cont) <- mkDupableCont env cont
1770 ; arg <- simplExpr (se `setInScope` env) arg
1771 ; (env, arg) <- makeTrivial env arg
1772 ; let app_cont = ApplyTo OkToDup arg (zapSubstEnv env) dup_cont
1773 ; return (env, app_cont, nodup_cont) }
1775 mkDupableCont env cont@(Select _ case_bndr [(_,bs,rhs)] se case_cont)
1776 -- See Note [Single-alternative case]
1777 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1778 -- | not (isDeadBinder case_bndr)
1779 | all isDeadBinder bs -- InIds
1780 = return (env, mkBoringStop scrut_ty, cont)
1782 scrut_ty = substTy se (idType case_bndr)
1784 mkDupableCont env (Select _ case_bndr alts se cont)
1785 = -- e.g. (case [...hole...] of { pi -> ei })
1787 -- let ji = \xij -> ei
1788 -- in case [...hole...] of { pi -> ji xij }
1789 do { tick (CaseOfCase case_bndr)
1790 ; (env, dup_cont, nodup_cont) <- mkDupableCont env cont
1791 -- NB: call mkDupableCont here, *not* prepareCaseCont
1792 -- We must make a duplicable continuation, whereas prepareCaseCont
1793 -- doesn't when there is a single case branch
1795 ; let alt_env = se `setInScope` env
1796 ; (alt_env, case_bndr') <- simplBinder alt_env case_bndr
1797 ; alts' <- mapM (simplAlt alt_env [] case_bndr' dup_cont) alts
1798 -- Safe to say that there are no handled-cons for the DEFAULT case
1799 -- NB: simplBinder does not zap deadness occ-info, so
1800 -- a dead case_bndr' will still advertise its deadness
1801 -- This is really important because in
1802 -- case e of b { (# p,q #) -> ... }
1803 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1804 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1805 -- In the new alts we build, we have the new case binder, so it must retain
1807 -- NB: we don't use alt_env further; it has the substEnv for
1808 -- the alternatives, and we don't want that
1810 ; (env, alts') <- mkDupableAlts env case_bndr' alts'
1811 ; return (env, -- Note [Duplicated env]
1812 Select OkToDup case_bndr' alts' (zapSubstEnv env)
1813 (mkBoringStop (contResultType dup_cont)),
1817 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1818 -> SimplM (SimplEnv, [InAlt])
1819 -- Absorbs the continuation into the new alternatives
1821 mkDupableAlts env case_bndr' alts
1824 go env [] = return (env, [])
1826 = do { (env, alt') <- mkDupableAlt env case_bndr' alt
1827 ; (env, alts') <- go env alts
1828 ; return (env, alt' : alts' ) }
1830 mkDupableAlt env case_bndr' (con, bndrs', rhs')
1831 | exprIsDupable rhs' -- Note [Small alternative rhs]
1832 = return (env, (con, bndrs', rhs'))
1834 = do { let rhs_ty' = exprType rhs'
1835 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1837 | isTyVar bndr = True -- Abstract over all type variables just in case
1838 | otherwise = not (isDeadBinder bndr)
1839 -- The deadness info on the new Ids is preserved by simplBinders
1841 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1842 <- if (any isId used_bndrs')
1843 then return (used_bndrs', varsToCoreExprs used_bndrs')
1844 else do { rw_id <- newId FSLIT("w") realWorldStatePrimTy
1845 ; return ([rw_id], [Var realWorldPrimId]) }
1847 ; join_bndr <- newId FSLIT("$j") (mkPiTypes final_bndrs' rhs_ty')
1848 -- Note [Funky mkPiTypes]
1850 ; let -- We make the lambdas into one-shot-lambdas. The
1851 -- join point is sure to be applied at most once, and doing so
1852 -- prevents the body of the join point being floated out by
1853 -- the full laziness pass
1854 really_final_bndrs = map one_shot final_bndrs'
1855 one_shot v | isId v = setOneShotLambda v
1857 join_rhs = mkLams really_final_bndrs rhs'
1858 join_call = mkApps (Var join_bndr) final_args
1860 ; return (addNonRec env join_bndr join_rhs, (con, bndrs', join_call)) }
1861 -- See Note [Duplicated env]
1864 Note [Duplicated env]
1865 ~~~~~~~~~~~~~~~~~~~~~
1866 Some of the alternatives are simplified, but have not been turned into a join point
1867 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
1868 bind the join point, because it might to do PostInlineUnconditionally, and
1869 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
1870 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
1871 at worst delays the join-point inlining.
1873 Note [Small alterantive rhs]
1874 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1875 It is worth checking for a small RHS because otherwise we
1876 get extra let bindings that may cause an extra iteration of the simplifier to
1877 inline back in place. Quite often the rhs is just a variable or constructor.
1878 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1879 iterations because the version with the let bindings looked big, and so wasn't
1880 inlined, but after the join points had been inlined it looked smaller, and so
1883 NB: we have to check the size of rhs', not rhs.
1884 Duplicating a small InAlt might invalidate occurrence information
1885 However, if it *is* dupable, we return the *un* simplified alternative,
1886 because otherwise we'd need to pair it up with an empty subst-env....
1887 but we only have one env shared between all the alts.
1888 (Remember we must zap the subst-env before re-simplifying something).
1889 Rather than do this we simply agree to re-simplify the original (small) thing later.
1891 Note [Funky mkPiTypes]
1892 ~~~~~~~~~~~~~~~~~~~~~~
1893 Notice the funky mkPiTypes. If the contructor has existentials
1894 it's possible that the join point will be abstracted over
1895 type varaibles as well as term variables.
1896 Example: Suppose we have
1897 data T = forall t. C [t]
1899 case (case e of ...) of
1901 We get the join point
1902 let j :: forall t. [t] -> ...
1903 j = /\t \xs::[t] -> rhs
1905 case (case e of ...) of
1906 C t xs::[t] -> j t xs
1908 Note [Join point abstaction]
1909 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1910 If we try to lift a primitive-typed something out
1911 for let-binding-purposes, we will *caseify* it (!),
1912 with potentially-disastrous strictness results. So
1913 instead we turn it into a function: \v -> e
1914 where v::State# RealWorld#. The value passed to this function
1915 is realworld#, which generates (almost) no code.
1917 There's a slight infelicity here: we pass the overall
1918 case_bndr to all the join points if it's used in *any* RHS,
1919 because we don't know its usage in each RHS separately
1921 We used to say "&& isUnLiftedType rhs_ty'" here, but now
1922 we make the join point into a function whenever used_bndrs'
1923 is empty. This makes the join-point more CPR friendly.
1924 Consider: let j = if .. then I# 3 else I# 4
1925 in case .. of { A -> j; B -> j; C -> ... }
1927 Now CPR doesn't w/w j because it's a thunk, so
1928 that means that the enclosing function can't w/w either,
1929 which is a lose. Here's the example that happened in practice:
1930 kgmod :: Int -> Int -> Int
1931 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1935 I have seen a case alternative like this:
1937 It's a bit silly to add the realWorld dummy arg in this case, making
1940 (the \v alone is enough to make CPR happy) but I think it's rare
1942 Note [Duplicating strict continuations]
1943 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1944 Do *not* duplicate StrictBind and StritArg continuations. We gain
1945 nothing by propagating them into the expressions, and we do lose a
1946 lot. Here's an example:
1947 && (case x of { T -> F; F -> T }) E
1948 Now, && is strict so we end up simplifying the case with
1949 an ArgOf continuation. If we let-bind it, we get
1951 let $j = \v -> && v E
1952 in simplExpr (case x of { T -> F; F -> T })
1954 And after simplifying more we get
1956 let $j = \v -> && v E
1957 in case x of { T -> $j F; F -> $j T }
1958 Which is a Very Bad Thing
1960 The desire not to duplicate is the entire reason that
1961 mkDupableCont returns a pair of continuations.
1963 The original plan had:
1964 e.g. (...strict-fn...) [...hole...]
1966 let $j = \a -> ...strict-fn...
1969 Note [Single-alternative cases]
1970 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1971 This case is just like the ArgOf case. Here's an example:
1975 case (case x of I# x' ->
1977 True -> I# (negate# x')
1978 False -> I# x') of y {
1980 Because the (case x) has only one alternative, we'll transform to
1982 case (case x' <# 0# of
1983 True -> I# (negate# x')
1984 False -> I# x') of y {
1986 But now we do *NOT* want to make a join point etc, giving
1988 let $j = \y -> MkT y
1990 True -> $j (I# (negate# x'))
1992 In this case the $j will inline again, but suppose there was a big
1993 strict computation enclosing the orginal call to MkT. Then, it won't
1994 "see" the MkT any more, because it's big and won't get duplicated.
1995 And, what is worse, nothing was gained by the case-of-case transform.
1997 When should use this case of mkDupableCont?
1998 However, matching on *any* single-alternative case is a *disaster*;
1999 e.g. case (case ....) of (a,b) -> (# a,b #)
2000 We must push the outer case into the inner one!
2003 * Match [(DEFAULT,_,_)], but in the common case of Int,
2004 the alternative-filling-in code turned the outer case into
2005 case (...) of y { I# _ -> MkT y }
2007 * Match on single alternative plus (not (isDeadBinder case_bndr))
2008 Rationale: pushing the case inwards won't eliminate the construction.
2009 But there's a risk of
2010 case (...) of y { (a,b) -> let z=(a,b) in ... }
2011 Now y looks dead, but it'll come alive again. Still, this
2012 seems like the best option at the moment.
2014 * Match on single alternative plus (all (isDeadBinder bndrs))
2015 Rationale: this is essentially seq.
2017 * Match when the rhs is *not* duplicable, and hence would lead to a
2018 join point. This catches the disaster-case above. We can test
2019 the *un-simplified* rhs, which is fine. It might get bigger or
2020 smaller after simplification; if it gets smaller, this case might
2021 fire next time round. NB also that we must test contIsDupable
2022 case_cont *btoo, because case_cont might be big!
2024 HOWEVER: I found that this version doesn't work well, because
2025 we can get let x = case (...) of { small } in ...case x...
2026 When x is inlined into its full context, we find that it was a bad
2027 idea to have pushed the outer case inside the (...) case.