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 )
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') = addLetIdInfo 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_info 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_info :: SimplEnv -> (InBndr,InExpr) -> (SimplEnv, (InBndr, OutBndr, InExpr))
266 -- Substitute in IdInfo, agument envt
267 add_info env (bndr, rhs) = (env, (bndr, bndr', rhs))
269 (env', bndr') = addLetIdInfo 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 final_info | loop_breaker = new_bndr_info
590 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
591 | otherwise = info_w_unf
593 final_id = new_bndr `setIdInfo` final_info
595 -- These seqs forces the Id, and hence its IdInfo,
596 -- and hence any inner substitutions
598 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
599 return (addNonRec env final_id new_rhs)
601 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
602 loop_breaker = isNonRuleLoopBreaker occ_info
603 old_info = idInfo old_bndr
604 occ_info = occInfo old_info
609 %************************************************************************
611 \subsection[Simplify-simplExpr]{The main function: simplExpr}
613 %************************************************************************
615 The reason for this OutExprStuff stuff is that we want to float *after*
616 simplifying a RHS, not before. If we do so naively we get quadratic
617 behaviour as things float out.
619 To see why it's important to do it after, consider this (real) example:
633 a -- Can't inline a this round, cos it appears twice
637 Each of the ==> steps is a round of simplification. We'd save a
638 whole round if we float first. This can cascade. Consider
643 let f = let d1 = ..d.. in \y -> e
647 in \x -> ...(\y ->e)...
649 Only in this second round can the \y be applied, and it
650 might do the same again.
654 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
655 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
657 expr_ty' = substTy env (exprType expr)
658 -- The type in the Stop continuation, expr_ty', is usually not used
659 -- It's only needed when discarding continuations after finding
660 -- a function that returns bottom.
661 -- Hence the lazy substitution
664 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
665 -- Simplify an expression, given a continuation
666 simplExprC env expr cont
667 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
668 do { (env', expr') <- simplExprF (zapFloats env) expr cont
669 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
670 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
671 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
672 return (wrapFloats env' expr') }
674 --------------------------------------------------
675 simplExprF :: SimplEnv -> InExpr -> SimplCont
676 -> SimplM (SimplEnv, OutExpr)
678 simplExprF env e cont
679 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
680 simplExprF' env e cont
682 simplExprF' env (Var v) cont = simplVar env v cont
683 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
684 simplExprF' env (Note n expr) cont = simplNote env n expr cont
685 simplExprF' env (Cast body co) cont = simplCast env body co cont
686 simplExprF' env (App fun arg) cont = simplExprF env fun $
687 ApplyTo NoDup arg env cont
689 simplExprF' env expr@(Lam _ _) cont
690 = simplLam env (map zap bndrs) body cont
691 -- The main issue here is under-saturated lambdas
692 -- (\x1. \x2. e) arg1
693 -- Here x1 might have "occurs-once" occ-info, because occ-info
694 -- is computed assuming that a group of lambdas is applied
695 -- all at once. If there are too few args, we must zap the
698 n_args = countArgs cont
699 n_params = length bndrs
700 (bndrs, body) = collectBinders expr
701 zap | n_args >= n_params = \b -> b
702 | otherwise = \b -> if isTyVar b then b
704 -- NB: we count all the args incl type args
705 -- so we must count all the binders (incl type lambdas)
707 simplExprF' env (Type ty) cont
708 = ASSERT( contIsRhsOrArg cont )
709 do { ty' <- simplType env ty
710 ; rebuild env (Type ty') cont }
712 simplExprF' env (Case scrut bndr case_ty alts) cont
713 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
714 = -- Simplify the scrutinee with a Select continuation
715 simplExprF env scrut (Select NoDup bndr alts env cont)
718 = -- If case-of-case is off, simply simplify the case expression
719 -- in a vanilla Stop context, and rebuild the result around it
720 do { case_expr' <- simplExprC env scrut case_cont
721 ; rebuild env case_expr' cont }
723 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
724 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
726 simplExprF' env (Let (Rec pairs) body) cont
727 = do { env <- simplRecBndrs env (map fst pairs)
728 -- NB: bndrs' don't have unfoldings or rules
729 -- We add them as we go down
731 ; env <- simplRecBind env NotTopLevel pairs
732 ; simplExprF env body cont }
734 simplExprF' env (Let (NonRec bndr rhs) body) cont
735 = simplNonRecE env bndr (rhs, env) ([], body) cont
737 ---------------------------------
738 simplType :: SimplEnv -> InType -> SimplM OutType
739 -- Kept monadic just so we can do the seqType
741 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
742 seqType new_ty `seq` returnSmpl new_ty
744 new_ty = substTy env ty
748 %************************************************************************
750 \subsection{The main rebuilder}
752 %************************************************************************
755 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
756 -- At this point the substitution in the SimplEnv should be irrelevant
757 -- only the in-scope set and floats should matter
758 rebuild env expr cont
759 = -- pprTrace "rebuild" (ppr expr $$ ppr cont $$ ppr (seFloats env)) $
761 Stop {} -> return (env, expr)
762 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
763 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
764 StrictArg fun ty info cont -> rebuildCall env (fun `App` expr) (funResultTy ty) info cont
765 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
766 ; simplLam env' bs body cont }
767 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
768 ; rebuild env (App expr arg') cont }
772 %************************************************************************
776 %************************************************************************
779 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
780 -> SimplM (SimplEnv, OutExpr)
781 simplCast env body co cont
782 = do { co' <- simplType env co
783 ; simplExprF env body (addCoerce co' cont) }
785 addCoerce co cont = add_coerce co (coercionKind co) cont
787 add_coerce co (s1, k1) cont -- co :: ty~ty
788 | s1 `coreEqType` k1 = cont -- is a no-op
790 add_coerce co1 (s1, k2) (CoerceIt co2 cont)
791 | (l1, t1) <- coercionKind co2
792 -- coerce T1 S1 (coerce S1 K1 e)
795 -- coerce T1 K1 e, otherwise
797 -- For example, in the initial form of a worker
798 -- we may find (coerce T (coerce S (\x.e))) y
799 -- and we'd like it to simplify to e[y/x] in one round
801 , s1 `coreEqType` t1 = cont -- The coerces cancel out
802 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
804 add_coerce co (s1s2, t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
805 -- (f `cast` g) ty ---> (f ty) `cast` (g @ ty)
806 -- This implements the PushT rule from the paper
807 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
808 , not (isCoVar tyvar)
809 = ApplyTo dup (Type ty') (zapSubstEnv env) (addCoerce (mkInstCoercion co ty') cont)
811 ty' = substTy arg_se arg_ty
813 -- ToDo: the PushC rule is not implemented at all
815 add_coerce co (s1s2, t1t2) (ApplyTo dup arg arg_se cont)
816 | not (isTypeArg arg) -- This implements the Push rule from the paper
817 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
818 -- co : s1s2 :=: t1t2
819 -- (coerce (T1->T2) (S1->S2) F) E
821 -- coerce T2 S2 (F (coerce S1 T1 E))
823 -- t1t2 must be a function type, T1->T2, because it's applied
824 -- to something but s1s2 might conceivably not be
826 -- When we build the ApplyTo we can't mix the out-types
827 -- with the InExpr in the argument, so we simply substitute
828 -- to make it all consistent. It's a bit messy.
829 -- But it isn't a common case.
831 -- Example of use: Trac #995
832 = ApplyTo dup new_arg (zapSubstEnv env) (addCoerce co2 cont)
834 -- we split coercion t1->t2 :=: s1->s2 into t1 :=: s1 and
835 -- t2 :=: s2 with left and right on the curried form:
836 -- (->) t1 t2 :=: (->) s1 s2
837 [co1, co2] = decomposeCo 2 co
838 new_arg = mkCoerce (mkSymCoercion co1) arg'
839 arg' = substExpr arg_se arg
841 add_coerce co _ cont = CoerceIt co cont
845 %************************************************************************
849 %************************************************************************
852 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
853 -> SimplM (SimplEnv, OutExpr)
855 simplLam env [] body cont = simplExprF env body cont
857 -- Type-beta reduction
858 simplLam env (bndr:bndrs) body (ApplyTo _ (Type ty_arg) arg_se cont)
859 = ASSERT( isTyVar bndr )
860 do { tick (BetaReduction bndr)
861 ; ty_arg' <- simplType (arg_se `setInScope` env) ty_arg
862 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
864 -- Ordinary beta reduction
865 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
866 = do { tick (BetaReduction bndr)
867 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
869 -- Not enough args, so there are real lambdas left to put in the result
870 simplLam env bndrs body cont
871 = do { (env, bndrs') <- simplLamBndrs env bndrs
872 ; body' <- simplExpr env body
873 ; new_lam <- mkLam bndrs' body'
874 ; rebuild env new_lam cont }
877 simplNonRecE :: SimplEnv
878 -> InId -- The binder
879 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
880 -> ([InId], InExpr) -- Body of the let/lambda
883 -> SimplM (SimplEnv, OutExpr)
885 -- simplNonRecE is used for
886 -- * non-top-level non-recursive lets in expressions
889 -- It deals with strict bindings, via the StrictBind continuation,
890 -- which may abort the whole process
892 -- The "body" of the binding comes as a pair of ([InId],InExpr)
893 -- representing a lambda; so we recurse back to simplLam
894 -- Why? Because of the binder-occ-info-zapping done before
895 -- the call to simplLam in simplExprF (Lam ...)
897 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
898 | preInlineUnconditionally env NotTopLevel bndr rhs
899 = do { tick (PreInlineUnconditionally bndr)
900 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
903 = do { simplExprF (rhs_se `setFloats` env) rhs
904 (StrictBind bndr bndrs body env cont) }
907 = do { (env1, bndr1) <- simplNonRecBndr env bndr
908 ; let (env2, bndr2) = addLetIdInfo env1 bndr bndr1
909 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
910 ; simplLam env3 bndrs body cont }
914 %************************************************************************
918 %************************************************************************
921 -- Hack alert: we only distinguish subsumed cost centre stacks for the
922 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
923 simplNote env (SCC cc) e cont
924 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
925 ; rebuild env (mkSCC cc e') cont }
927 -- See notes with SimplMonad.inlineMode
928 simplNote env InlineMe e cont
929 | Just (inside, outside) <- splitInlineCont cont -- Boring boring continuation; see notes above
930 = do { -- Don't inline inside an INLINE expression
931 e' <- simplExprC (setMode inlineMode env) e inside
932 ; rebuild env (mkInlineMe e') outside }
934 | otherwise -- Dissolve the InlineMe note if there's
935 -- an interesting context of any kind to combine with
936 -- (even a type application -- anything except Stop)
937 = simplExprF env e cont
939 simplNote env (CoreNote s) e cont
940 = simplExpr env e `thenSmpl` \ e' ->
941 rebuild env (Note (CoreNote s) e') cont
945 %************************************************************************
947 \subsection{Dealing with calls}
949 %************************************************************************
952 simplVar env var cont
953 = case substId env var of
954 DoneEx e -> simplExprF (zapSubstEnv env) e cont
955 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
956 DoneId var1 -> completeCall (zapSubstEnv env) var1 cont
957 -- Note [zapSubstEnv]
958 -- The template is already simplified, so don't re-substitute.
959 -- This is VITAL. Consider
961 -- let y = \z -> ...x... in
963 -- We'll clone the inner \x, adding x->x' in the id_subst
964 -- Then when we inline y, we must *not* replace x by x' in
965 -- the inlined copy!!
967 ---------------------------------------------------------
968 -- Dealing with a call site
970 completeCall env var cont
971 = do { dflags <- getDOptsSmpl
972 ; let (args,call_cont) = contArgs cont
973 -- The args are OutExprs, obtained by *lazily* substituting
974 -- in the args found in cont. These args are only examined
975 -- to limited depth (unless a rule fires). But we must do
976 -- the substitution; rule matching on un-simplified args would
979 ------------- First try rules ----------------
980 -- Do this before trying inlining. Some functions have
981 -- rules *and* are strict; in this case, we don't want to
982 -- inline the wrapper of the non-specialised thing; better
983 -- to call the specialised thing instead.
985 -- We used to use the black-listing mechanism to ensure that inlining of
986 -- the wrapper didn't occur for things that have specialisations till a
987 -- later phase, so but now we just try RULES first
989 -- Note [Rules for recursive functions]
990 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
991 -- You might think that we shouldn't apply rules for a loop breaker:
992 -- doing so might give rise to an infinite loop, because a RULE is
993 -- rather like an extra equation for the function:
994 -- RULE: f (g x) y = x+y
997 -- But it's too drastic to disable rules for loop breakers.
998 -- Even the foldr/build rule would be disabled, because foldr
999 -- is recursive, and hence a loop breaker:
1000 -- foldr k z (build g) = g k z
1001 -- So it's up to the programmer: rules can cause divergence
1003 ; let in_scope = getInScope env
1004 maybe_rule = case activeRule dflags env of
1005 Nothing -> Nothing -- No rules apply
1006 Just act_fn -> lookupRule act_fn in_scope
1008 ; case maybe_rule of {
1009 Just (rule, rule_rhs) ->
1010 tick (RuleFired (ru_name rule)) `thenSmpl_`
1011 (if dopt Opt_D_dump_rule_firings dflags then
1012 pprTrace "Rule fired" (vcat [
1013 text "Rule:" <+> ftext (ru_name rule),
1014 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1015 text "After: " <+> pprCoreExpr rule_rhs,
1016 text "Cont: " <+> ppr call_cont])
1019 simplExprF env rule_rhs (dropArgs (ruleArity rule) cont)
1020 -- The ruleArity says how many args the rule consumed
1022 ; Nothing -> do -- No rules
1024 ------------- Next try inlining ----------------
1025 { let arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1026 n_val_args = length arg_infos
1027 interesting_cont = interestingCallContext (notNull args)
1030 active_inline = activeInline env var
1031 maybe_inline = callSiteInline dflags active_inline
1032 var arg_infos interesting_cont
1033 ; case maybe_inline of {
1034 Just unfolding -- There is an inlining!
1035 -> do { tick (UnfoldingDone var)
1036 ; (if dopt Opt_D_dump_inlinings dflags then
1037 pprTrace ("Inlining done" ++ showSDoc (ppr var)) (vcat [
1038 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1039 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1040 text "Cont: " <+> ppr call_cont])
1043 simplExprF env unfolding cont }
1045 ; Nothing -> -- No inlining!
1047 ------------- No inlining! ----------------
1048 -- Next, look for rules or specialisations that match
1050 rebuildCall env (Var var) (idType var)
1051 (mkArgInfo var n_val_args call_cont) cont
1054 rebuildCall :: SimplEnv
1055 -> OutExpr -> OutType -- Function and its type
1056 -> (Bool, [Bool]) -- See SimplUtils.mkArgInfo
1058 -> SimplM (SimplEnv, OutExpr)
1059 rebuildCall env fun fun_ty (has_rules, []) cont
1060 -- When we run out of strictness args, it means
1061 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1062 -- Then we want to discard the entire strict continuation. E.g.
1063 -- * case (error "hello") of { ... }
1064 -- * (error "Hello") arg
1065 -- * f (error "Hello") where f is strict
1067 -- Then, especially in the first of these cases, we'd like to discard
1068 -- the continuation, leaving just the bottoming expression. But the
1069 -- type might not be right, so we may have to add a coerce.
1070 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1071 = return (env, mk_coerce fun) -- contination to discard, else we do it
1072 where -- again and again!
1073 cont_ty = contResultType cont
1074 co = mkUnsafeCoercion fun_ty cont_ty
1075 mk_coerce expr | cont_ty `coreEqType` fun_ty = fun
1076 | otherwise = mkCoerce co fun
1078 rebuildCall env fun fun_ty info (ApplyTo _ (Type arg_ty) se cont)
1079 = do { ty' <- simplType (se `setInScope` env) arg_ty
1080 ; rebuildCall env (fun `App` Type ty') (applyTy fun_ty ty') info cont }
1082 rebuildCall env fun fun_ty (has_rules, str:strs) (ApplyTo _ arg arg_se cont)
1083 | str || isStrictType arg_ty -- Strict argument
1084 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1085 simplExprF (arg_se `setFloats` env) arg
1086 (StrictArg fun fun_ty (has_rules, strs) cont)
1089 | otherwise -- Lazy argument
1090 -- DO NOT float anything outside, hence simplExprC
1091 -- There is no benefit (unlike in a let-binding), and we'd
1092 -- have to be very careful about bogus strictness through
1093 -- floating a demanded let.
1094 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1095 (mkLazyArgStop arg_ty has_rules)
1096 ; rebuildCall env (fun `App` arg') res_ty (has_rules, strs) cont }
1098 (arg_ty, res_ty) = splitFunTy fun_ty
1100 rebuildCall env fun fun_ty info cont
1101 = rebuild env fun cont
1106 This part of the simplifier may break the no-shadowing invariant
1108 f (...(\a -> e)...) (case y of (a,b) -> e')
1109 where f is strict in its second arg
1110 If we simplify the innermost one first we get (...(\a -> e)...)
1111 Simplifying the second arg makes us float the case out, so we end up with
1112 case y of (a,b) -> f (...(\a -> e)...) e'
1113 So the output does not have the no-shadowing invariant. However, there is
1114 no danger of getting name-capture, because when the first arg was simplified
1115 we used an in-scope set that at least mentioned all the variables free in its
1116 static environment, and that is enough.
1118 We can't just do innermost first, or we'd end up with a dual problem:
1119 case x of (a,b) -> f e (...(\a -> e')...)
1121 I spent hours trying to recover the no-shadowing invariant, but I just could
1122 not think of an elegant way to do it. The simplifier is already knee-deep in
1123 continuations. We have to keep the right in-scope set around; AND we have
1124 to get the effect that finding (error "foo") in a strict arg position will
1125 discard the entire application and replace it with (error "foo"). Getting
1126 all this at once is TOO HARD!
1128 %************************************************************************
1130 Rebuilding a cse expression
1132 %************************************************************************
1134 Blob of helper functions for the "case-of-something-else" situation.
1137 ---------------------------------------------------------
1138 -- Eliminate the case if possible
1140 rebuildCase :: SimplEnv
1141 -> OutExpr -- Scrutinee
1142 -> InId -- Case binder
1143 -> [InAlt] -- Alternatives (inceasing order)
1145 -> SimplM (SimplEnv, OutExpr)
1147 --------------------------------------------------
1148 -- 1. Eliminate the case if there's a known constructor
1149 --------------------------------------------------
1151 rebuildCase env scrut case_bndr alts cont
1152 | Just (con,args) <- exprIsConApp_maybe scrut
1153 -- Works when the scrutinee is a variable with a known unfolding
1154 -- as well as when it's an explicit constructor application
1155 = knownCon env scrut (DataAlt con) args case_bndr alts cont
1157 | Lit lit <- scrut -- No need for same treatment as constructors
1158 -- because literals are inlined more vigorously
1159 = knownCon env scrut (LitAlt lit) [] case_bndr alts cont
1162 --------------------------------------------------
1163 -- 2. Eliminate the case if scrutinee is evaluated
1164 --------------------------------------------------
1166 rebuildCase env scrut case_bndr [(con,bndrs,rhs)] cont
1167 -- See if we can get rid of the case altogether
1168 -- See the extensive notes on case-elimination above
1169 -- mkCase made sure that if all the alternatives are equal,
1170 -- then there is now only one (DEFAULT) rhs
1171 | all isDeadBinder bndrs -- bndrs are [InId]
1173 -- Check that the scrutinee can be let-bound instead of case-bound
1174 , exprOkForSpeculation scrut
1175 -- OK not to evaluate it
1176 -- This includes things like (==# a# b#)::Bool
1177 -- so that we simplify
1178 -- case ==# a# b# of { True -> x; False -> x }
1181 -- This particular example shows up in default methods for
1182 -- comparision operations (e.g. in (>=) for Int.Int32)
1183 || exprIsHNF scrut -- It's already evaluated
1184 || var_demanded_later scrut -- It'll be demanded later
1186 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1187 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1188 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1189 -- its argument: case x of { y -> dataToTag# y }
1190 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1191 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1193 -- Also we don't want to discard 'seq's
1194 = do { tick (CaseElim case_bndr)
1195 ; env <- simplNonRecX env case_bndr scrut
1196 ; simplExprF env rhs cont }
1198 -- The case binder is going to be evaluated later,
1199 -- and the scrutinee is a simple variable
1200 var_demanded_later (Var v) = isStrictDmd (idNewDemandInfo case_bndr)
1201 && not (isTickBoxOp v)
1202 -- ugly hack; covering this case is what
1203 -- exprOkForSpeculation was intended for.
1204 var_demanded_later other = False
1207 --------------------------------------------------
1208 -- 3. Catch-all case
1209 --------------------------------------------------
1211 rebuildCase env scrut case_bndr alts cont
1212 = do { -- Prepare the continuation;
1213 -- The new subst_env is in place
1214 (env, dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1216 -- Simplify the alternatives
1217 ; (scrut', case_bndr', alts') <- simplAlts env scrut case_bndr alts dup_cont
1218 ; let res_ty' = contResultType dup_cont
1219 ; case_expr <- mkCase scrut' case_bndr' res_ty' alts'
1221 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1222 -- The case binder *not* scope over the whole returned case-expression
1223 ; rebuild env case_expr nodup_cont }
1226 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1227 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1228 way, there's a chance that v will now only be used once, and hence
1231 Note [no-case-of-case]
1232 ~~~~~~~~~~~~~~~~~~~~~~
1233 There is a time we *don't* want to do that, namely when
1234 -fno-case-of-case is on. This happens in the first simplifier pass,
1235 and enhances full laziness. Here's the bad case:
1236 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1237 If we eliminate the inner case, we trap it inside the I# v -> arm,
1238 which might prevent some full laziness happening. I've seen this
1239 in action in spectral/cichelli/Prog.hs:
1240 [(m,n) | m <- [1..max], n <- [1..max]]
1241 Hence the check for NoCaseOfCase.
1243 Note [Suppressing the case binder-swap]
1244 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1245 There is another situation when it might make sense to suppress the
1246 case-expression binde-swap. If we have
1248 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1249 ...other cases .... }
1251 We'll perform the binder-swap for the outer case, giving
1253 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1254 ...other cases .... }
1256 But there is no point in doing it for the inner case, because w1 can't
1257 be inlined anyway. Furthermore, doing the case-swapping involves
1258 zapping w2's occurrence info (see paragraphs that follow), and that
1259 forces us to bind w2 when doing case merging. So we get
1261 case x of w1 { A -> let w2 = w1 in e1
1262 B -> let w2 = w1 in e2
1263 ...other cases .... }
1265 This is plain silly in the common case where w2 is dead.
1267 Even so, I can't see a good way to implement this idea. I tried
1268 not doing the binder-swap if the scrutinee was already evaluated
1269 but that failed big-time:
1273 case v of w { MkT x ->
1274 case x of x1 { I# y1 ->
1275 case x of x2 { I# y2 -> ...
1277 Notice that because MkT is strict, x is marked "evaluated". But to
1278 eliminate the last case, we must either make sure that x (as well as
1279 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1280 the binder-swap. So this whole note is a no-op.
1284 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1285 any occurrence info (eg IAmDead) in the case binder, because the
1286 case-binder now effectively occurs whenever v does. AND we have to do
1287 the same for the pattern-bound variables! Example:
1289 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1291 Here, b and p are dead. But when we move the argment inside the first
1292 case RHS, and eliminate the second case, we get
1294 case x of { (a,b) -> a b }
1296 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1299 Indeed, this can happen anytime the case binder isn't dead:
1300 case <any> of x { (a,b) ->
1301 case x of { (p,q) -> p } }
1302 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1303 The point is that we bring into the envt a binding
1305 after the outer case, and that makes (a,b) alive. At least we do unless
1306 the case binder is guaranteed dead.
1310 Consider case (v `cast` co) of x { I# ->
1311 ... (case (v `cast` co) of {...}) ...
1312 We'd like to eliminate the inner case. We can get this neatly by
1313 arranging that inside the outer case we add the unfolding
1314 v |-> x `cast` (sym co)
1315 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1317 Note [Improving seq]
1320 type family F :: * -> *
1321 type instance F Int = Int
1323 ... case e of x { DEFAULT -> rhs } ...
1325 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1327 case e `cast` co of x'::Int
1328 I# x# -> let x = x' `cast` sym co
1331 so that 'rhs' can take advantage of hte form of x'. Notice that Note
1332 [Case of cast] may then apply to the result.
1334 This showed up in Roman's experiments. Example:
1335 foo :: F Int -> Int -> Int
1336 foo t n = t `seq` bar n
1339 bar n = bar (n - case t of TI i -> i)
1340 Here we'd like to avoid repeated evaluating t inside the loop, by
1341 taking advantage of the `seq`.
1343 At one point I did transformation in LiberateCase, but it's more robust here.
1344 (Otherwise, there's a danger that we'll simply drop the 'seq' altogether, before
1345 LiberateCase gets to see it.)
1347 Note [Case elimination]
1348 ~~~~~~~~~~~~~~~~~~~~~~~
1349 The case-elimination transformation discards redundant case expressions.
1350 Start with a simple situation:
1352 case x# of ===> e[x#/y#]
1355 (when x#, y# are of primitive type, of course). We can't (in general)
1356 do this for algebraic cases, because we might turn bottom into
1359 The code in SimplUtils.prepareAlts has the effect of generalise this
1360 idea to look for a case where we're scrutinising a variable, and we
1361 know that only the default case can match. For example:
1365 DEFAULT -> ...(case x of
1369 Here the inner case is first trimmed to have only one alternative, the
1370 DEFAULT, after which it's an instance of the previous case. This
1371 really only shows up in eliminating error-checking code.
1373 We also make sure that we deal with this very common case:
1378 Here we are using the case as a strict let; if x is used only once
1379 then we want to inline it. We have to be careful that this doesn't
1380 make the program terminate when it would have diverged before, so we
1382 - e is already evaluated (it may so if e is a variable)
1383 - x is used strictly, or
1385 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1387 case e of ===> case e of DEFAULT -> r
1391 Now again the case may be elminated by the CaseElim transformation.
1394 Further notes about case elimination
1395 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1396 Consider: test :: Integer -> IO ()
1399 Turns out that this compiles to:
1402 eta1 :: State# RealWorld ->
1403 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1405 (PrelNum.jtos eta ($w[] @ Char))
1407 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1409 Notice the strange '<' which has no effect at all. This is a funny one.
1410 It started like this:
1412 f x y = if x < 0 then jtos x
1413 else if y==0 then "" else jtos x
1415 At a particular call site we have (f v 1). So we inline to get
1417 if v < 0 then jtos x
1418 else if 1==0 then "" else jtos x
1420 Now simplify the 1==0 conditional:
1422 if v<0 then jtos v else jtos v
1424 Now common-up the two branches of the case:
1426 case (v<0) of DEFAULT -> jtos v
1428 Why don't we drop the case? Because it's strict in v. It's technically
1429 wrong to drop even unnecessary evaluations, and in practice they
1430 may be a result of 'seq' so we *definitely* don't want to drop those.
1431 I don't really know how to improve this situation.
1435 simplCaseBinder :: SimplEnv -> OutExpr -> OutId -> [InAlt]
1436 -> SimplM (SimplEnv, OutExpr, OutId)
1437 simplCaseBinder env scrut case_bndr alts
1438 = do { (env1, case_bndr1) <- simplBinder env case_bndr
1440 ; fam_envs <- getFamEnvs
1441 ; (env2, scrut2, case_bndr2) <- improve_seq fam_envs env1 scrut
1442 case_bndr case_bndr1 alts
1443 -- Note [Improving seq]
1445 ; let (env3, case_bndr3) = improve_case_bndr env2 scrut2 case_bndr2
1446 -- Note [Case of cast]
1448 ; return (env3, scrut2, case_bndr3) }
1451 improve_seq fam_envs env1 scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1452 | Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1453 = do { case_bndr2 <- newId FSLIT("nt") ty2
1454 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1455 env2 = extendIdSubst env1 case_bndr rhs
1456 ; return (env2, scrut `Cast` co, case_bndr2) }
1458 improve_seq fam_envs env1 scrut case_bndr case_bndr1 alts
1459 = return (env1, scrut, case_bndr1)
1462 improve_case_bndr env scrut case_bndr
1463 | switchIsOn (getSwitchChecker env) NoCaseOfCase
1464 -- See Note [no-case-of-case]
1467 | otherwise -- Failed try [see Note 2 above]
1468 -- not (isEvaldUnfolding (idUnfolding v))
1470 Var v -> (modifyInScope env1 v case_bndr', case_bndr')
1471 -- Note about using modifyInScope for v here
1472 -- We could extend the substitution instead, but it would be
1473 -- a hack because then the substitution wouldn't be idempotent
1474 -- any more (v is an OutId). And this does just as well.
1476 Cast (Var v) co -> (addBinderUnfolding env1 v rhs, case_bndr')
1478 rhs = Cast (Var case_bndr') (mkSymCoercion co)
1480 other -> (env, case_bndr)
1482 case_bndr' = zapOccInfo case_bndr
1483 env1 = modifyInScope env case_bndr case_bndr'
1486 zapOccInfo :: InId -> InId -- See Note [zapOccInfo]
1487 zapOccInfo b = b `setIdOccInfo` NoOccInfo
1491 simplAlts does two things:
1493 1. Eliminate alternatives that cannot match, including the
1494 DEFAULT alternative.
1496 2. If the DEFAULT alternative can match only one possible constructor,
1497 then make that constructor explicit.
1499 case e of x { DEFAULT -> rhs }
1501 case e of x { (a,b) -> rhs }
1502 where the type is a single constructor type. This gives better code
1503 when rhs also scrutinises x or e.
1505 Here "cannot match" includes knowledge from GADTs
1507 It's a good idea do do this stuff before simplifying the alternatives, to
1508 avoid simplifying alternatives we know can't happen, and to come up with
1509 the list of constructors that are handled, to put into the IdInfo of the
1510 case binder, for use when simplifying the alternatives.
1512 Eliminating the default alternative in (1) isn't so obvious, but it can
1515 data Colour = Red | Green | Blue
1524 DEFAULT -> [ case y of ... ]
1526 If we inline h into f, the default case of the inlined h can't happen.
1527 If we don't notice this, we may end up filtering out *all* the cases
1528 of the inner case y, which give us nowhere to go!
1532 simplAlts :: SimplEnv
1534 -> InId -- Case binder
1535 -> [InAlt] -> SimplCont
1536 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1537 -- Like simplExpr, this just returns the simplified alternatives;
1538 -- it not return an environment
1540 simplAlts env scrut case_bndr alts cont'
1541 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1542 do { let alt_env = zapFloats env
1543 ; (alt_env, scrut', case_bndr') <- simplCaseBinder alt_env scrut case_bndr alts
1545 ; (imposs_deflt_cons, in_alts) <- prepareAlts scrut case_bndr' alts
1547 ; alts' <- mapM (simplAlt alt_env imposs_deflt_cons case_bndr' cont') in_alts
1548 ; return (scrut', case_bndr', alts') }
1550 ------------------------------------
1551 simplAlt :: SimplEnv
1552 -> [AltCon] -- These constructors can't be present when
1553 -- matching the DEFAULT alternative
1554 -> OutId -- The case binder
1559 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1560 = ASSERT( null bndrs )
1561 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1562 -- Record the constructors that the case-binder *can't* be.
1563 ; rhs' <- simplExprC env' rhs cont'
1564 ; return (DEFAULT, [], rhs') }
1566 simplAlt env imposs_deflt_cons case_bndr' cont' (LitAlt lit, bndrs, rhs)
1567 = ASSERT( null bndrs )
1568 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1569 ; rhs' <- simplExprC env' rhs cont'
1570 ; return (LitAlt lit, [], rhs') }
1572 simplAlt env imposs_deflt_cons case_bndr' cont' (DataAlt con, vs, rhs)
1573 = do { -- Deal with the pattern-bound variables
1574 (env, vs') <- simplBinders env (add_evals con vs)
1576 -- Mark the ones that are in ! positions in the
1577 -- data constructor as certainly-evaluated.
1578 ; let vs'' = add_evals con vs'
1580 -- Bind the case-binder to (con args)
1581 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1582 con_args = map Type inst_tys' ++ varsToCoreExprs vs''
1583 env' = addBinderUnfolding env case_bndr' (mkConApp con con_args)
1585 ; rhs' <- simplExprC env' rhs cont'
1586 ; return (DataAlt con, vs'', rhs') }
1588 -- add_evals records the evaluated-ness of the bound variables of
1589 -- a case pattern. This is *important*. Consider
1590 -- data T = T !Int !Int
1592 -- case x of { T a b -> T (a+1) b }
1594 -- We really must record that b is already evaluated so that we don't
1595 -- go and re-evaluate it when constructing the result.
1596 -- See Note [Data-con worker strictness] in MkId.lhs
1597 add_evals dc vs = cat_evals dc vs (dataConRepStrictness dc)
1599 cat_evals dc vs strs
1603 go (v:vs) strs | isTyVar v = v : go vs strs
1604 go (v:vs) (str:strs)
1605 | isMarkedStrict str = evald_v : go vs strs
1606 | otherwise = zapped_v : go vs strs
1608 zapped_v = zap_occ_info v
1609 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1610 go _ _ = pprPanic "cat_evals" (ppr dc $$ ppr vs $$ ppr strs)
1612 -- If the case binder is alive, then we add the unfolding
1614 -- to the envt; so vs are now very much alive
1615 -- Note [Aug06] I can't see why this actually matters
1616 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1617 | otherwise = zapOccInfo
1619 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1620 addBinderUnfolding env bndr rhs
1621 = modifyInScope env bndr (bndr `setIdUnfolding` mkUnfolding False rhs)
1623 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1624 addBinderOtherCon env bndr cons
1625 = modifyInScope env bndr (bndr `setIdUnfolding` mkOtherCon cons)
1629 %************************************************************************
1631 \subsection{Known constructor}
1633 %************************************************************************
1635 We are a bit careful with occurrence info. Here's an example
1637 (\x* -> case x of (a*, b) -> f a) (h v, e)
1639 where the * means "occurs once". This effectively becomes
1640 case (h v, e) of (a*, b) -> f a)
1642 let a* = h v; b = e in f a
1646 All this should happen in one sweep.
1649 knownCon :: SimplEnv -> OutExpr -> AltCon -> [OutExpr]
1650 -> InId -> [InAlt] -> SimplCont
1651 -> SimplM (SimplEnv, OutExpr)
1653 knownCon env scrut con args bndr alts cont
1654 = do { tick (KnownBranch bndr)
1655 ; knownAlt env scrut args bndr (findAlt con alts) cont }
1657 knownAlt env scrut args bndr (DEFAULT, bs, rhs) cont
1659 do { env <- simplNonRecX env bndr scrut
1660 -- This might give rise to a binding with non-atomic args
1661 -- like x = Node (f x) (g x)
1662 -- but simplNonRecX will atomic-ify it
1663 ; simplExprF env rhs cont }
1665 knownAlt env scrut args bndr (LitAlt lit, bs, rhs) cont
1667 do { env <- simplNonRecX env bndr scrut
1668 ; simplExprF env rhs cont }
1670 knownAlt env scrut args bndr (DataAlt dc, bs, rhs) cont
1671 = do { let dead_bndr = isDeadBinder bndr -- bndr is an InId
1672 n_drop_tys = length (dataConUnivTyVars dc)
1673 ; env <- bind_args env dead_bndr bs (drop n_drop_tys args)
1675 -- It's useful to bind bndr to scrut, rather than to a fresh
1676 -- binding x = Con arg1 .. argn
1677 -- because very often the scrut is a variable, so we avoid
1678 -- creating, and then subsequently eliminating, a let-binding
1679 -- BUT, if scrut is a not a variable, we must be careful
1680 -- about duplicating the arg redexes; in that case, make
1681 -- a new con-app from the args
1682 bndr_rhs = case scrut of
1685 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1686 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1687 -- args are aready OutExprs, but bs are InIds
1689 ; env <- simplNonRecX env bndr bndr_rhs
1690 ; -- pprTrace "knownCon2" (ppr bs $$ ppr rhs $$ ppr (seIdSubst env)) $
1691 simplExprF env rhs cont }
1694 bind_args env dead_bndr [] _ = return env
1696 bind_args env dead_bndr (b:bs) (Type ty : args)
1697 = ASSERT( isTyVar b )
1698 bind_args (extendTvSubst env b ty) dead_bndr bs args
1700 bind_args env dead_bndr (b:bs) (arg : args)
1702 do { let b' = if dead_bndr then b else zapOccInfo b
1703 -- Note that the binder might be "dead", because it doesn't occur
1704 -- in the RHS; and simplNonRecX may therefore discard it via postInlineUnconditionally
1705 -- Nevertheless we must keep it if the case-binder is alive, because it may
1706 -- be used in the con_app. See Note [zapOccInfo]
1707 ; env <- simplNonRecX env b' arg
1708 ; bind_args env dead_bndr bs args }
1711 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr args $$
1712 text "scrut:" <+> ppr scrut
1716 %************************************************************************
1718 \subsection{Duplicating continuations}
1720 %************************************************************************
1723 prepareCaseCont :: SimplEnv
1724 -> [InAlt] -> SimplCont
1725 -> SimplM (SimplEnv, SimplCont,SimplCont)
1726 -- Return a duplicatable continuation, a non-duplicable part
1727 -- plus some extra bindings (that scope over the entire
1730 -- No need to make it duplicatable if there's only one alternative
1731 prepareCaseCont env [alt] cont = return (env, cont, mkBoringStop (contResultType cont))
1732 prepareCaseCont env alts cont = mkDupableCont env cont
1736 mkDupableCont :: SimplEnv -> SimplCont
1737 -> SimplM (SimplEnv, SimplCont, SimplCont)
1739 mkDupableCont env cont
1740 | contIsDupable cont
1741 = returnSmpl (env, cont, mkBoringStop (contResultType cont))
1743 mkDupableCont env (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1745 mkDupableCont env (CoerceIt ty cont)
1746 = do { (env, dup, nodup) <- mkDupableCont env cont
1747 ; return (env, CoerceIt ty dup, nodup) }
1749 mkDupableCont env cont@(StrictBind bndr _ _ se _)
1750 = return (env, mkBoringStop (substTy se (idType bndr)), cont)
1751 -- See Note [Duplicating strict continuations]
1753 mkDupableCont env cont@(StrictArg _ fun_ty _ _)
1754 = return (env, mkBoringStop (funArgTy fun_ty), cont)
1755 -- See Note [Duplicating strict continuations]
1757 mkDupableCont env (ApplyTo _ arg se cont)
1758 = -- e.g. [...hole...] (...arg...)
1760 -- let a = ...arg...
1761 -- in [...hole...] a
1762 do { (env, dup_cont, nodup_cont) <- mkDupableCont env cont
1763 ; arg <- simplExpr (se `setInScope` env) arg
1764 ; (env, arg) <- makeTrivial env arg
1765 ; let app_cont = ApplyTo OkToDup arg (zapSubstEnv env) dup_cont
1766 ; return (env, app_cont, nodup_cont) }
1768 mkDupableCont env cont@(Select _ case_bndr [(_,bs,rhs)] se case_cont)
1769 -- See Note [Single-alternative case]
1770 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1771 -- | not (isDeadBinder case_bndr)
1772 | all isDeadBinder bs -- InIds
1773 = return (env, mkBoringStop scrut_ty, cont)
1775 scrut_ty = substTy se (idType case_bndr)
1777 mkDupableCont env (Select _ case_bndr alts se cont)
1778 = -- e.g. (case [...hole...] of { pi -> ei })
1780 -- let ji = \xij -> ei
1781 -- in case [...hole...] of { pi -> ji xij }
1782 do { tick (CaseOfCase case_bndr)
1783 ; (env, dup_cont, nodup_cont) <- mkDupableCont env cont
1784 -- NB: call mkDupableCont here, *not* prepareCaseCont
1785 -- We must make a duplicable continuation, whereas prepareCaseCont
1786 -- doesn't when there is a single case branch
1788 ; let alt_env = se `setInScope` env
1789 ; (alt_env, case_bndr') <- simplBinder alt_env case_bndr
1790 ; alts' <- mapM (simplAlt alt_env [] case_bndr' dup_cont) alts
1791 -- Safe to say that there are no handled-cons for the DEFAULT case
1792 -- NB: simplBinder does not zap deadness occ-info, so
1793 -- a dead case_bndr' will still advertise its deadness
1794 -- This is really important because in
1795 -- case e of b { (# p,q #) -> ... }
1796 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1797 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1798 -- In the new alts we build, we have the new case binder, so it must retain
1800 -- NB: we don't use alt_env further; it has the substEnv for
1801 -- the alternatives, and we don't want that
1803 ; (env, alts') <- mkDupableAlts env case_bndr' alts'
1804 ; return (env, -- Note [Duplicated env]
1805 Select OkToDup case_bndr' alts' (zapSubstEnv env)
1806 (mkBoringStop (contResultType dup_cont)),
1810 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1811 -> SimplM (SimplEnv, [InAlt])
1812 -- Absorbs the continuation into the new alternatives
1814 mkDupableAlts env case_bndr' alts
1817 go env [] = return (env, [])
1819 = do { (env, alt') <- mkDupableAlt env case_bndr' alt
1820 ; (env, alts') <- go env alts
1821 ; return (env, alt' : alts' ) }
1823 mkDupableAlt env case_bndr' (con, bndrs', rhs')
1824 | exprIsDupable rhs' -- Note [Small alternative rhs]
1825 = return (env, (con, bndrs', rhs'))
1827 = do { let rhs_ty' = exprType rhs'
1828 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1830 | isTyVar bndr = True -- Abstract over all type variables just in case
1831 | otherwise = not (isDeadBinder bndr)
1832 -- The deadness info on the new Ids is preserved by simplBinders
1834 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1835 <- if (any isId used_bndrs')
1836 then return (used_bndrs', varsToCoreExprs used_bndrs')
1837 else do { rw_id <- newId FSLIT("w") realWorldStatePrimTy
1838 ; return ([rw_id], [Var realWorldPrimId]) }
1840 ; join_bndr <- newId FSLIT("$j") (mkPiTypes final_bndrs' rhs_ty')
1841 -- Note [Funky mkPiTypes]
1843 ; let -- We make the lambdas into one-shot-lambdas. The
1844 -- join point is sure to be applied at most once, and doing so
1845 -- prevents the body of the join point being floated out by
1846 -- the full laziness pass
1847 really_final_bndrs = map one_shot final_bndrs'
1848 one_shot v | isId v = setOneShotLambda v
1850 join_rhs = mkLams really_final_bndrs rhs'
1851 join_call = mkApps (Var join_bndr) final_args
1853 ; return (addNonRec env join_bndr join_rhs, (con, bndrs', join_call)) }
1854 -- See Note [Duplicated env]
1857 Note [Duplicated env]
1858 ~~~~~~~~~~~~~~~~~~~~~
1859 Some of the alternatives are simplified, but have not been turned into a join point
1860 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
1861 bind the join point, because it might to do PostInlineUnconditionally, and
1862 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
1863 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
1864 at worst delays the join-point inlining.
1866 Note [Small alterantive rhs]
1867 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1868 It is worth checking for a small RHS because otherwise we
1869 get extra let bindings that may cause an extra iteration of the simplifier to
1870 inline back in place. Quite often the rhs is just a variable or constructor.
1871 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1872 iterations because the version with the let bindings looked big, and so wasn't
1873 inlined, but after the join points had been inlined it looked smaller, and so
1876 NB: we have to check the size of rhs', not rhs.
1877 Duplicating a small InAlt might invalidate occurrence information
1878 However, if it *is* dupable, we return the *un* simplified alternative,
1879 because otherwise we'd need to pair it up with an empty subst-env....
1880 but we only have one env shared between all the alts.
1881 (Remember we must zap the subst-env before re-simplifying something).
1882 Rather than do this we simply agree to re-simplify the original (small) thing later.
1884 Note [Funky mkPiTypes]
1885 ~~~~~~~~~~~~~~~~~~~~~~
1886 Notice the funky mkPiTypes. If the contructor has existentials
1887 it's possible that the join point will be abstracted over
1888 type varaibles as well as term variables.
1889 Example: Suppose we have
1890 data T = forall t. C [t]
1892 case (case e of ...) of
1894 We get the join point
1895 let j :: forall t. [t] -> ...
1896 j = /\t \xs::[t] -> rhs
1898 case (case e of ...) of
1899 C t xs::[t] -> j t xs
1901 Note [Join point abstaction]
1902 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1903 If we try to lift a primitive-typed something out
1904 for let-binding-purposes, we will *caseify* it (!),
1905 with potentially-disastrous strictness results. So
1906 instead we turn it into a function: \v -> e
1907 where v::State# RealWorld#. The value passed to this function
1908 is realworld#, which generates (almost) no code.
1910 There's a slight infelicity here: we pass the overall
1911 case_bndr to all the join points if it's used in *any* RHS,
1912 because we don't know its usage in each RHS separately
1914 We used to say "&& isUnLiftedType rhs_ty'" here, but now
1915 we make the join point into a function whenever used_bndrs'
1916 is empty. This makes the join-point more CPR friendly.
1917 Consider: let j = if .. then I# 3 else I# 4
1918 in case .. of { A -> j; B -> j; C -> ... }
1920 Now CPR doesn't w/w j because it's a thunk, so
1921 that means that the enclosing function can't w/w either,
1922 which is a lose. Here's the example that happened in practice:
1923 kgmod :: Int -> Int -> Int
1924 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1928 I have seen a case alternative like this:
1930 It's a bit silly to add the realWorld dummy arg in this case, making
1933 (the \v alone is enough to make CPR happy) but I think it's rare
1935 Note [Duplicating strict continuations]
1936 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1937 Do *not* duplicate StrictBind and StritArg continuations. We gain
1938 nothing by propagating them into the expressions, and we do lose a
1939 lot. Here's an example:
1940 && (case x of { T -> F; F -> T }) E
1941 Now, && is strict so we end up simplifying the case with
1942 an ArgOf continuation. If we let-bind it, we get
1944 let $j = \v -> && v E
1945 in simplExpr (case x of { T -> F; F -> T })
1947 And after simplifying more we get
1949 let $j = \v -> && v E
1950 in case x of { T -> $j F; F -> $j T }
1951 Which is a Very Bad Thing
1953 The desire not to duplicate is the entire reason that
1954 mkDupableCont returns a pair of continuations.
1956 The original plan had:
1957 e.g. (...strict-fn...) [...hole...]
1959 let $j = \a -> ...strict-fn...
1962 Note [Single-alternative cases]
1963 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1964 This case is just like the ArgOf case. Here's an example:
1968 case (case x of I# x' ->
1970 True -> I# (negate# x')
1971 False -> I# x') of y {
1973 Because the (case x) has only one alternative, we'll transform to
1975 case (case x' <# 0# of
1976 True -> I# (negate# x')
1977 False -> I# x') of y {
1979 But now we do *NOT* want to make a join point etc, giving
1981 let $j = \y -> MkT y
1983 True -> $j (I# (negate# x'))
1985 In this case the $j will inline again, but suppose there was a big
1986 strict computation enclosing the orginal call to MkT. Then, it won't
1987 "see" the MkT any more, because it's big and won't get duplicated.
1988 And, what is worse, nothing was gained by the case-of-case transform.
1990 When should use this case of mkDupableCont?
1991 However, matching on *any* single-alternative case is a *disaster*;
1992 e.g. case (case ....) of (a,b) -> (# a,b #)
1993 We must push the outer case into the inner one!
1996 * Match [(DEFAULT,_,_)], but in the common case of Int,
1997 the alternative-filling-in code turned the outer case into
1998 case (...) of y { I# _ -> MkT y }
2000 * Match on single alternative plus (not (isDeadBinder case_bndr))
2001 Rationale: pushing the case inwards won't eliminate the construction.
2002 But there's a risk of
2003 case (...) of y { (a,b) -> let z=(a,b) in ... }
2004 Now y looks dead, but it'll come alive again. Still, this
2005 seems like the best option at the moment.
2007 * Match on single alternative plus (all (isDeadBinder bndrs))
2008 Rationale: this is essentially seq.
2010 * Match when the rhs is *not* duplicable, and hence would lead to a
2011 join point. This catches the disaster-case above. We can test
2012 the *un-simplified* rhs, which is fine. It might get bigger or
2013 smaller after simplification; if it gets smaller, this case might
2014 fire next time round. NB also that we must test contIsDupable
2015 case_cont *btoo, because case_cont might be big!
2017 HOWEVER: I found that this version doesn't work well, because
2018 we can get let x = case (...) of { small } in ...case x...
2019 When x is inlined into its full context, we find that it was a bad
2020 idea to have pushed the outer case inside the (...) case.