2 % (c) The AQUA Project, Glasgow University, 1993-1998
4 \section[Simplify]{The main module of the simplifier}
7 module Simplify ( simplTopBinds, simplExpr ) where
9 #include "HsVersions.h"
13 import Type hiding ( substTy, extendTvSubst )
16 import MkId ( rUNTIME_ERROR_ID )
21 import FamInstEnv ( topNormaliseType )
22 import DataCon ( dataConRepStrictness, dataConUnivTyVars )
24 import NewDemand ( isStrictDmd )
25 import PprCore ( pprParendExpr, pprCoreExpr )
26 import CoreUnfold ( mkUnfolding, callSiteInline, CallCtxt(..) )
28 import Rules ( lookupRule, getRules )
29 import BasicTypes ( isMarkedStrict )
30 import CostCentre ( currentCCS )
31 import TysPrim ( realWorldStatePrimTy )
32 import PrelInfo ( realWorldPrimId )
33 import BasicTypes ( TopLevelFlag(..), isTopLevel,
34 RecFlag(..), isNonRuleLoopBreaker )
35 import Maybes ( orElse )
36 import Data.List ( mapAccumL )
42 The guts of the simplifier is in this module, but the driver loop for
43 the simplifier is in SimplCore.lhs.
46 -----------------------------------------
47 *** IMPORTANT NOTE ***
48 -----------------------------------------
49 The simplifier used to guarantee that the output had no shadowing, but
50 it does not do so any more. (Actually, it never did!) The reason is
51 documented with simplifyArgs.
54 -----------------------------------------
55 *** IMPORTANT NOTE ***
56 -----------------------------------------
57 Many parts of the simplifier return a bunch of "floats" as well as an
58 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
60 All "floats" are let-binds, not case-binds, but some non-rec lets may
61 be unlifted (with RHS ok-for-speculation).
65 -----------------------------------------
66 ORGANISATION OF FUNCTIONS
67 -----------------------------------------
69 - simplify all top-level binders
70 - for NonRec, call simplRecOrTopPair
71 - for Rec, call simplRecBind
74 ------------------------------
75 simplExpr (applied lambda) ==> simplNonRecBind
76 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
77 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
79 ------------------------------
80 simplRecBind [binders already simplfied]
81 - use simplRecOrTopPair on each pair in turn
83 simplRecOrTopPair [binder already simplified]
84 Used for: recursive bindings (top level and nested)
85 top-level non-recursive bindings
87 - check for PreInlineUnconditionally
91 Used for: non-top-level non-recursive bindings
92 beta reductions (which amount to the same thing)
93 Because it can deal with strict arts, it takes a
94 "thing-inside" and returns an expression
96 - check for PreInlineUnconditionally
97 - simplify binder, including its IdInfo
106 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
107 Used for: binding case-binder and constr args in a known-constructor case
108 - check for PreInLineUnconditionally
112 ------------------------------
113 simplLazyBind: [binder already simplified, RHS not]
114 Used for: recursive bindings (top level and nested)
115 top-level non-recursive bindings
116 non-top-level, but *lazy* non-recursive bindings
117 [must not be strict or unboxed]
118 Returns floats + an augmented environment, not an expression
119 - substituteIdInfo and add result to in-scope
120 [so that rules are available in rec rhs]
123 - float if exposes constructor or PAP
127 completeNonRecX: [binder and rhs both simplified]
128 - if the the thing needs case binding (unlifted and not ok-for-spec)
134 completeBind: [given a simplified RHS]
135 [used for both rec and non-rec bindings, top level and not]
136 - try PostInlineUnconditionally
137 - add unfolding [this is the only place we add an unfolding]
142 Right hand sides and arguments
143 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
144 In many ways we want to treat
145 (a) the right hand side of a let(rec), and
146 (b) a function argument
147 in the same way. But not always! In particular, we would
148 like to leave these arguments exactly as they are, so they
149 will match a RULE more easily.
154 It's harder to make the rule match if we ANF-ise the constructor,
155 or eta-expand the PAP:
157 f (let { a = g x; b = h x } in (a,b))
160 On the other hand if we see the let-defns
165 then we *do* want to ANF-ise and eta-expand, so that p and q
166 can be safely inlined.
168 Even floating lets out is a bit dubious. For let RHS's we float lets
169 out if that exposes a value, so that the value can be inlined more vigorously.
172 r = let x = e in (x,x)
174 Here, if we float the let out we'll expose a nice constructor. We did experiments
175 that showed this to be a generally good thing. But it was a bad thing to float
176 lets out unconditionally, because that meant they got allocated more often.
178 For function arguments, there's less reason to expose a constructor (it won't
179 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
180 So for the moment we don't float lets out of function arguments either.
185 For eta expansion, we want to catch things like
187 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
189 If the \x was on the RHS of a let, we'd eta expand to bring the two
190 lambdas together. And in general that's a good thing to do. Perhaps
191 we should eta expand wherever we find a (value) lambda? Then the eta
192 expansion at a let RHS can concentrate solely on the PAP case.
195 %************************************************************************
197 \subsection{Bindings}
199 %************************************************************************
202 simplTopBinds :: SimplEnv -> [InBind] -> SimplM [OutBind]
204 simplTopBinds env0 binds0
205 = do { -- Put all the top-level binders into scope at the start
206 -- so that if a transformation rule has unexpectedly brought
207 -- anything into scope, then we don't get a complaint about that.
208 -- It's rather as if the top-level binders were imported.
209 ; env1 <- simplRecBndrs env0 (bindersOfBinds binds0)
210 ; dflags <- getDOptsSmpl
211 ; let dump_flag = dopt Opt_D_dump_inlinings dflags ||
212 dopt Opt_D_dump_rule_firings dflags
213 ; env2 <- simpl_binds dump_flag env1 binds0
214 ; freeTick SimplifierDone
215 ; return (getFloats env2) }
217 -- We need to track the zapped top-level binders, because
218 -- they should have their fragile IdInfo zapped (notably occurrence info)
219 -- That's why we run down binds and bndrs' simultaneously.
221 -- The dump-flag emits a trace for each top-level binding, which
222 -- helps to locate the tracing for inlining and rule firing
223 simpl_binds :: Bool -> SimplEnv -> [InBind] -> SimplM SimplEnv
224 simpl_binds _ env [] = return env
225 simpl_binds dump env (bind:binds) = do { env' <- trace_bind dump bind $
227 ; simpl_binds dump env' binds }
229 trace_bind True bind = pprTrace "SimplBind" (ppr (bindersOf bind))
230 trace_bind False _ = \x -> x
232 simpl_bind env (Rec pairs) = simplRecBind env TopLevel pairs
233 simpl_bind env (NonRec b r) = simplRecOrTopPair env' TopLevel b b' r
235 (env', b') = addBndrRules env b (lookupRecBndr env b)
239 %************************************************************************
241 \subsection{Lazy bindings}
243 %************************************************************************
245 simplRecBind is used for
246 * recursive bindings only
249 simplRecBind :: SimplEnv -> TopLevelFlag
252 simplRecBind env0 top_lvl pairs0
253 = do { let (env_with_info, triples) = mapAccumL add_rules env0 pairs0
254 ; env1 <- go (zapFloats env_with_info) triples
255 ; return (env0 `addRecFloats` env1) }
256 -- addFloats adds the floats from env1,
257 -- _and_ updates env0 with the in-scope set from env1
259 add_rules :: SimplEnv -> (InBndr,InExpr) -> (SimplEnv, (InBndr, OutBndr, InExpr))
260 -- Add the (substituted) rules to the binder
261 add_rules env (bndr, rhs) = (env', (bndr, bndr', rhs))
263 (env', bndr') = addBndrRules env bndr (lookupRecBndr env bndr)
265 go env [] = return env
267 go env ((old_bndr, new_bndr, rhs) : pairs)
268 = do { env' <- simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
272 simplOrTopPair is used for
273 * recursive bindings (whether top level or not)
274 * top-level non-recursive bindings
276 It assumes the binder has already been simplified, but not its IdInfo.
279 simplRecOrTopPair :: SimplEnv
281 -> InId -> OutBndr -> InExpr -- Binder and rhs
282 -> SimplM SimplEnv -- Returns an env that includes the binding
284 simplRecOrTopPair env top_lvl old_bndr new_bndr rhs
285 | preInlineUnconditionally env top_lvl old_bndr rhs -- Check for unconditional inline
286 = do { tick (PreInlineUnconditionally old_bndr)
287 ; return (extendIdSubst env old_bndr (mkContEx env rhs)) }
290 = simplLazyBind env top_lvl Recursive old_bndr new_bndr rhs env
291 -- May not actually be recursive, but it doesn't matter
295 simplLazyBind is used for
296 * [simplRecOrTopPair] recursive bindings (whether top level or not)
297 * [simplRecOrTopPair] top-level non-recursive bindings
298 * [simplNonRecE] non-top-level *lazy* non-recursive bindings
301 1. It assumes that the binder is *already* simplified,
302 and is in scope, and its IdInfo too, except unfolding
304 2. It assumes that the binder type is lifted.
306 3. It does not check for pre-inline-unconditionallly;
307 that should have been done already.
310 simplLazyBind :: SimplEnv
311 -> TopLevelFlag -> RecFlag
312 -> InId -> OutId -- Binder, both pre-and post simpl
313 -- The OutId has IdInfo, except arity, unfolding
314 -> InExpr -> SimplEnv -- The RHS and its environment
317 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
318 = do { let rhs_env = rhs_se `setInScope` env
319 (tvs, body) = case collectTyBinders rhs of
320 (tvs, body) | not_lam body -> (tvs,body)
321 | otherwise -> ([], rhs)
322 not_lam (Lam _ _) = False
324 -- Do not do the "abstract tyyvar" thing if there's
325 -- a lambda inside, becuase it defeats eta-reduction
326 -- f = /\a. \x. g a x
329 ; (body_env, tvs') <- simplBinders rhs_env tvs
330 -- See Note [Floating and type abstraction] in SimplUtils
333 ; (body_env1, body1) <- simplExprF body_env body mkBoringStop
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 ; let env' = foldl (addPolyBind top_lvl) env poly_binds
353 ; return (env', rhs') }
355 ; completeBind env' top_lvl bndr bndr1 rhs' }
358 A specialised variant of simplNonRec used when the RHS is already simplified,
359 notably in knownCon. It uses case-binding where necessary.
362 simplNonRecX :: SimplEnv
363 -> InId -- Old binder
364 -> OutExpr -- Simplified RHS
367 simplNonRecX env bndr new_rhs
368 = do { (env', bndr') <- simplBinder env bndr
369 ; completeNonRecX env' (isStrictId bndr) bndr bndr' new_rhs }
371 completeNonRecX :: SimplEnv
373 -> InId -- Old binder
374 -> OutId -- New binder
375 -> OutExpr -- Simplified RHS
378 completeNonRecX env is_strict old_bndr new_bndr new_rhs
379 = do { (env1, rhs1) <- prepareRhs (zapFloats env) new_rhs
381 if doFloatFromRhs NotTopLevel NonRecursive 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, env1, rhs1) <- go 0 env0 rhs0
440 ; return (env1, rhs1) }
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)
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 False var var expr
513 ; return (env', substExpr env' (Var var)) }
517 %************************************************************************
519 \subsection{Completing a lazy binding}
521 %************************************************************************
524 * deals only with Ids, not TyVars
525 * takes an already-simplified binder and RHS
526 * is used for both recursive and non-recursive bindings
527 * is used for both top-level and non-top-level bindings
529 It does the following:
530 - tries discarding a dead binding
531 - tries PostInlineUnconditionally
532 - add unfolding [this is the only place we add an unfolding]
535 It does *not* attempt to do let-to-case. Why? Because it is used for
536 - top-level bindings (when let-to-case is impossible)
537 - many situations where the "rhs" is known to be a WHNF
538 (so let-to-case is inappropriate).
540 Nor does it do the atomic-argument thing
543 completeBind :: SimplEnv
544 -> TopLevelFlag -- Flag stuck into unfolding
545 -> InId -- Old binder
546 -> OutId -> OutExpr -- New binder and RHS
548 -- completeBind may choose to do its work
549 -- * by extending the substitution (e.g. let x = y in ...)
550 -- * or by adding to the floats in the envt
552 completeBind env top_lvl old_bndr new_bndr new_rhs
553 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
554 -- Inline and discard the binding
555 = do { tick (PostInlineUnconditionally old_bndr)
556 ; -- pprTrace "postInlineUnconditionally" (ppr old_bndr <+> ppr new_bndr <+> ppr new_rhs) $
557 return (extendIdSubst env old_bndr (DoneEx new_rhs)) }
558 -- Use the substitution to make quite, quite sure that the
559 -- substitution will happen, since we are going to discard the binding
562 = return (addNonRecWithUnf env new_bndr new_rhs unfolding wkr)
564 unfolding | omit_unfolding = NoUnfolding
565 | otherwise = mkUnfolding (isTopLevel top_lvl) new_rhs
566 old_info = idInfo old_bndr
567 occ_info = occInfo old_info
568 wkr = substWorker env (workerInfo old_info)
569 omit_unfolding = isNonRuleLoopBreaker occ_info
570 -- or not (activeInline env old_bndr)
571 -- Do *not* trim the unfolding in SimplGently, else
572 -- the specialiser can't see it!
575 addPolyBind :: TopLevelFlag -> SimplEnv -> OutBind -> SimplEnv
576 -- Add a new binding to the environment, complete with its unfolding
577 -- but *do not* do postInlineUnconditionally, because we have already
578 -- processed some of the scope of the binding
579 -- We still want the unfolding though. Consider
581 -- x = /\a. let y = ... in Just y
583 -- Then we float the y-binding out (via abstractFloats and addPolyBind)
584 -- but 'x' may well then be inlined in 'body' in which case we'd like the
585 -- opportunity to inline 'y' too.
587 addPolyBind top_lvl env (NonRec poly_id rhs)
588 = addNonRecWithUnf env poly_id rhs unfolding NoWorker
590 unfolding | not (activeInline env poly_id) = NoUnfolding
591 | otherwise = mkUnfolding (isTopLevel top_lvl) rhs
592 -- addNonRecWithInfo adds the new binding in the
593 -- proper way (ie complete with unfolding etc),
594 -- and extends the in-scope set
596 addPolyBind _ env bind@(Rec _) = extendFloats env bind
597 -- Hack: letrecs are more awkward, so we extend "by steam"
598 -- without adding unfoldings etc. At worst this leads to
599 -- more simplifier iterations
602 addNonRecWithUnf :: SimplEnv
603 -> OutId -> OutExpr -- New binder and RHS
604 -> Unfolding -> WorkerInfo -- and unfolding
606 -- Add suitable IdInfo to the Id, add the binding to the floats, and extend the in-scope set
607 addNonRecWithUnf env new_bndr rhs unfolding wkr
608 = ASSERT( isId new_bndr )
609 final_id `seq` -- This seq forces the Id, and hence its IdInfo,
610 -- and hence any inner substitutions
611 addNonRec env final_id rhs
612 -- The addNonRec adds it to the in-scope set too
615 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity rhs
618 -- Add the unfolding *only* for non-loop-breakers
619 -- Making loop breakers not have an unfolding at all
620 -- means that we can avoid tests in exprIsConApp, for example.
621 -- This is important: if exprIsConApp says 'yes' for a recursive
622 -- thing, then we can get into an infinite loop
625 -- If the unfolding is a value, the demand info may
626 -- go pear-shaped, so we nuke it. Example:
628 -- case x of (p,q) -> h p q x
629 -- Here x is certainly demanded. But after we've nuked
630 -- the case, we'll get just
631 -- let x = (a,b) in h a b x
632 -- and now x is not demanded (I'm assuming h is lazy)
633 -- This really happens. Similarly
634 -- let f = \x -> e in ...f..f...
635 -- After inlining f at some of its call sites the original binding may
636 -- (for example) be no longer strictly demanded.
637 -- The solution here is a bit ad hoc...
638 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
641 final_info | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
642 | otherwise = info_w_unf
644 final_id = new_bndr `setIdInfo` final_info
649 %************************************************************************
651 \subsection[Simplify-simplExpr]{The main function: simplExpr}
653 %************************************************************************
655 The reason for this OutExprStuff stuff is that we want to float *after*
656 simplifying a RHS, not before. If we do so naively we get quadratic
657 behaviour as things float out.
659 To see why it's important to do it after, consider this (real) example:
673 a -- Can't inline a this round, cos it appears twice
677 Each of the ==> steps is a round of simplification. We'd save a
678 whole round if we float first. This can cascade. Consider
683 let f = let d1 = ..d.. in \y -> e
687 in \x -> ...(\y ->e)...
689 Only in this second round can the \y be applied, and it
690 might do the same again.
694 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
695 simplExpr env expr = simplExprC env expr mkBoringStop
697 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
698 -- Simplify an expression, given a continuation
699 simplExprC env expr cont
700 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
701 do { (env', expr') <- simplExprF (zapFloats env) expr cont
702 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
703 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
704 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
705 return (wrapFloats env' expr') }
707 --------------------------------------------------
708 simplExprF :: SimplEnv -> InExpr -> SimplCont
709 -> SimplM (SimplEnv, OutExpr)
711 simplExprF env e cont
712 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
713 simplExprF' env e cont
715 simplExprF' :: SimplEnv -> InExpr -> SimplCont
716 -> SimplM (SimplEnv, OutExpr)
717 simplExprF' env (Var v) cont = simplVar env v cont
718 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
719 simplExprF' env (Note n expr) cont = simplNote env n expr cont
720 simplExprF' env (Cast body co) cont = simplCast env body co cont
721 simplExprF' env (App fun arg) cont = simplExprF env fun $
722 ApplyTo NoDup arg env cont
724 simplExprF' env expr@(Lam _ _) cont
725 = simplLam env (map zap bndrs) body cont
726 -- The main issue here is under-saturated lambdas
727 -- (\x1. \x2. e) arg1
728 -- Here x1 might have "occurs-once" occ-info, because occ-info
729 -- is computed assuming that a group of lambdas is applied
730 -- all at once. If there are too few args, we must zap the
733 n_args = countArgs cont
734 n_params = length bndrs
735 (bndrs, body) = collectBinders expr
736 zap | n_args >= n_params = \b -> b
737 | otherwise = \b -> if isTyVar b then b
739 -- NB: we count all the args incl type args
740 -- so we must count all the binders (incl type lambdas)
742 simplExprF' env (Type ty) cont
743 = ASSERT( contIsRhsOrArg cont )
744 do { ty' <- simplType env ty
745 ; rebuild env (Type ty') cont }
747 simplExprF' env (Case scrut bndr _ alts) cont
748 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
749 = -- Simplify the scrutinee with a Select continuation
750 simplExprF env scrut (Select NoDup bndr alts env cont)
753 = -- If case-of-case is off, simply simplify the case expression
754 -- in a vanilla Stop context, and rebuild the result around it
755 do { case_expr' <- simplExprC env scrut case_cont
756 ; rebuild env case_expr' cont }
758 case_cont = Select NoDup bndr alts env mkBoringStop
760 simplExprF' env (Let (Rec pairs) body) cont
761 = do { env' <- simplRecBndrs env (map fst pairs)
762 -- NB: bndrs' don't have unfoldings or rules
763 -- We add them as we go down
765 ; env'' <- simplRecBind env' NotTopLevel pairs
766 ; simplExprF env'' body cont }
768 simplExprF' env (Let (NonRec bndr rhs) body) cont
769 = simplNonRecE env bndr (rhs, env) ([], body) cont
771 ---------------------------------
772 simplType :: SimplEnv -> InType -> SimplM OutType
773 -- Kept monadic just so we can do the seqType
775 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
776 seqType new_ty `seq` return new_ty
778 new_ty = substTy env ty
782 %************************************************************************
784 \subsection{The main rebuilder}
786 %************************************************************************
789 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
790 -- At this point the substitution in the SimplEnv should be irrelevant
791 -- only the in-scope set and floats should matter
792 rebuild env expr cont0
793 = -- pprTrace "rebuild" (ppr expr $$ ppr cont0 $$ ppr (seFloats env)) $
795 Stop {} -> return (env, expr)
796 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
797 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
798 StrictArg fun _ info cont -> rebuildCall env (fun `App` expr) info cont
799 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
800 ; simplLam env' bs body cont }
801 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
802 ; rebuild env (App expr arg') cont }
806 %************************************************************************
810 %************************************************************************
813 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
814 -> SimplM (SimplEnv, OutExpr)
815 simplCast env body co0 cont0
816 = do { co1 <- simplType env co0
817 ; simplExprF env body (addCoerce co1 cont0) }
819 addCoerce co cont = add_coerce co (coercionKind co) cont
821 add_coerce _co (s1, k1) cont -- co :: ty~ty
822 | s1 `coreEqType` k1 = cont -- is a no-op
824 add_coerce co1 (s1, _k2) (CoerceIt co2 cont)
825 | (_l1, t1) <- coercionKind co2
826 -- e |> (g1 :: S1~L) |> (g2 :: L~T1)
829 -- e |> (g1 . g2 :: T1~T2) otherwise
831 -- For example, in the initial form of a worker
832 -- we may find (coerce T (coerce S (\x.e))) y
833 -- and we'd like it to simplify to e[y/x] in one round
835 , s1 `coreEqType` t1 = cont -- The coerces cancel out
836 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
838 add_coerce co (s1s2, _t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
839 -- (f |> g) ty ---> (f ty) |> (g @ ty)
840 -- This implements the PushT rule from the paper
841 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
842 , not (isCoVar tyvar)
843 = ApplyTo dup (Type ty') (zapSubstEnv env) (addCoerce (mkInstCoercion co ty') cont)
845 ty' = substTy (arg_se `setInScope` env) arg_ty
847 -- ToDo: the PushC rule is not implemented at all
849 add_coerce co (s1s2, _t1t2) (ApplyTo dup arg arg_se cont)
850 | not (isTypeArg arg) -- This implements the Push rule from the paper
851 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
852 -- (e |> (g :: s1s2 ~ t1->t2)) f
854 -- (e (f |> (arg g :: t1~s1))
855 -- |> (res g :: s2->t2)
857 -- t1t2 must be a function type, t1->t2, because it's applied
858 -- to something but s1s2 might conceivably not be
860 -- When we build the ApplyTo we can't mix the out-types
861 -- with the InExpr in the argument, so we simply substitute
862 -- to make it all consistent. It's a bit messy.
863 -- But it isn't a common case.
865 -- Example of use: Trac #995
866 = ApplyTo dup new_arg (zapSubstEnv env) (addCoerce co2 cont)
868 -- we split coercion t1->t2 ~ s1->s2 into t1 ~ s1 and
869 -- t2 ~ s2 with left and right on the curried form:
870 -- (->) t1 t2 ~ (->) s1 s2
871 [co1, co2] = decomposeCo 2 co
872 new_arg = mkCoerce (mkSymCoercion co1) arg'
873 arg' = substExpr (arg_se `setInScope` env) arg
875 add_coerce co _ cont = CoerceIt co cont
879 %************************************************************************
883 %************************************************************************
886 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
887 -> SimplM (SimplEnv, OutExpr)
889 simplLam env [] body cont = simplExprF env body cont
892 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
893 = do { tick (BetaReduction bndr)
894 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
896 -- Not enough args, so there are real lambdas left to put in the result
897 simplLam env bndrs body cont
898 = do { (env', bndrs') <- simplLamBndrs env bndrs
899 ; body' <- simplExpr env' body
900 ; new_lam <- mkLam bndrs' body'
901 ; rebuild env' new_lam cont }
904 simplNonRecE :: SimplEnv
905 -> InId -- The binder
906 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
907 -> ([InBndr], InExpr) -- Body of the let/lambda
910 -> SimplM (SimplEnv, OutExpr)
912 -- simplNonRecE is used for
913 -- * non-top-level non-recursive lets in expressions
916 -- It deals with strict bindings, via the StrictBind continuation,
917 -- which may abort the whole process
919 -- The "body" of the binding comes as a pair of ([InId],InExpr)
920 -- representing a lambda; so we recurse back to simplLam
921 -- Why? Because of the binder-occ-info-zapping done before
922 -- the call to simplLam in simplExprF (Lam ...)
924 -- First deal with type applications and type lets
925 -- (/\a. e) (Type ty) and (let a = Type ty in e)
926 simplNonRecE env bndr (Type ty_arg, rhs_se) (bndrs, body) cont
927 = ASSERT( isTyVar bndr )
928 do { ty_arg' <- simplType (rhs_se `setInScope` env) ty_arg
929 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
931 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
932 | preInlineUnconditionally env NotTopLevel bndr rhs
933 = do { tick (PreInlineUnconditionally bndr)
934 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
937 = do { simplExprF (rhs_se `setFloats` env) rhs
938 (StrictBind bndr bndrs body env cont) }
941 = ASSERT( not (isTyVar bndr) )
942 do { (env1, bndr1) <- simplNonRecBndr env bndr
943 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
944 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
945 ; simplLam env3 bndrs body cont }
949 %************************************************************************
953 %************************************************************************
956 -- Hack alert: we only distinguish subsumed cost centre stacks for the
957 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
958 simplNote :: SimplEnv -> Note -> CoreExpr -> SimplCont
959 -> SimplM (SimplEnv, OutExpr)
960 simplNote env (SCC cc) e cont
961 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
962 ; rebuild env (mkSCC cc e') cont }
964 -- See notes with SimplMonad.inlineMode
965 simplNote env InlineMe e cont
966 | Just (inside, outside) <- splitInlineCont cont -- Boring boring continuation; see notes above
967 = do { -- Don't inline inside an INLINE expression
968 e' <- simplExprC (setMode inlineMode env) e inside
969 ; rebuild env (mkInlineMe e') outside }
971 | otherwise -- Dissolve the InlineMe note if there's
972 -- an interesting context of any kind to combine with
973 -- (even a type application -- anything except Stop)
974 = simplExprF env e cont
976 simplNote env (CoreNote s) e cont = do
977 e' <- simplExpr env e
978 rebuild env (Note (CoreNote s) e') cont
982 %************************************************************************
984 \subsection{Dealing with calls}
986 %************************************************************************
989 simplVar :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
990 simplVar env var cont
991 = case substId env var of
992 DoneEx e -> simplExprF (zapSubstEnv env) e cont
993 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
994 DoneId var1 -> completeCall (zapSubstEnv env) var1 cont
995 -- Note [zapSubstEnv]
996 -- The template is already simplified, so don't re-substitute.
997 -- This is VITAL. Consider
999 -- let y = \z -> ...x... in
1001 -- We'll clone the inner \x, adding x->x' in the id_subst
1002 -- Then when we inline y, we must *not* replace x by x' in
1003 -- the inlined copy!!
1005 ---------------------------------------------------------
1006 -- Dealing with a call site
1008 completeCall :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1009 completeCall env var cont
1010 = do { dflags <- getDOptsSmpl
1011 ; let (args,call_cont) = contArgs cont
1012 -- The args are OutExprs, obtained by *lazily* substituting
1013 -- in the args found in cont. These args are only examined
1014 -- to limited depth (unless a rule fires). But we must do
1015 -- the substitution; rule matching on un-simplified args would
1018 ------------- First try rules ----------------
1019 -- Do this before trying inlining. Some functions have
1020 -- rules *and* are strict; in this case, we don't want to
1021 -- inline the wrapper of the non-specialised thing; better
1022 -- to call the specialised thing instead.
1024 -- We used to use the black-listing mechanism to ensure that inlining of
1025 -- the wrapper didn't occur for things that have specialisations till a
1026 -- later phase, so but now we just try RULES first
1028 -- Note [Rules for recursive functions]
1029 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1030 -- You might think that we shouldn't apply rules for a loop breaker:
1031 -- doing so might give rise to an infinite loop, because a RULE is
1032 -- rather like an extra equation for the function:
1033 -- RULE: f (g x) y = x+y
1036 -- But it's too drastic to disable rules for loop breakers.
1037 -- Even the foldr/build rule would be disabled, because foldr
1038 -- is recursive, and hence a loop breaker:
1039 -- foldr k z (build g) = g k z
1040 -- So it's up to the programmer: rules can cause divergence
1041 ; rule_base <- getSimplRules
1042 ; let in_scope = getInScope env
1043 rules = getRules rule_base var
1044 maybe_rule = case activeRule dflags env of
1045 Nothing -> Nothing -- No rules apply
1046 Just act_fn -> lookupRule act_fn in_scope
1048 ; case maybe_rule of {
1049 Just (rule, rule_rhs) -> do
1050 tick (RuleFired (ru_name rule))
1051 (if dopt Opt_D_dump_rule_firings dflags then
1052 pprTrace "Rule fired" (vcat [
1053 text "Rule:" <+> ftext (ru_name rule),
1054 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1055 text "After: " <+> pprCoreExpr rule_rhs,
1056 text "Cont: " <+> ppr call_cont])
1059 simplExprF env rule_rhs (dropArgs (ruleArity rule) cont)
1060 -- The ruleArity says how many args the rule consumed
1062 ; Nothing -> do -- No rules
1064 ------------- Next try inlining ----------------
1065 { let arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1066 n_val_args = length arg_infos
1067 interesting_cont = interestingCallContext call_cont
1068 active_inline = activeInline env var
1069 maybe_inline = callSiteInline dflags active_inline var
1070 (null args) arg_infos interesting_cont
1071 ; case maybe_inline of {
1072 Just unfolding -- There is an inlining!
1073 -> do { tick (UnfoldingDone var)
1074 ; (if dopt Opt_D_dump_inlinings dflags then
1075 pprTrace ("Inlining done" ++ showSDoc (ppr var)) (vcat [
1076 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1077 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1078 text "Cont: " <+> ppr call_cont])
1081 simplExprF env unfolding cont }
1083 ; Nothing -> -- No inlining!
1085 ------------- No inlining! ----------------
1086 -- Next, look for rules or specialisations that match
1088 rebuildCall env (Var var)
1089 (mkArgInfo var n_val_args call_cont) cont
1092 rebuildCall :: SimplEnv
1093 -> OutExpr -- Function
1096 -> SimplM (SimplEnv, OutExpr)
1097 rebuildCall env fun (ArgInfo { ai_strs = [] }) cont
1098 -- When we run out of strictness args, it means
1099 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1100 -- Then we want to discard the entire strict continuation. E.g.
1101 -- * case (error "hello") of { ... }
1102 -- * (error "Hello") arg
1103 -- * f (error "Hello") where f is strict
1105 -- Then, especially in the first of these cases, we'd like to discard
1106 -- the continuation, leaving just the bottoming expression. But the
1107 -- type might not be right, so we may have to add a coerce.
1108 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1109 = return (env, mk_coerce fun) -- contination to discard, else we do it
1110 where -- again and again!
1111 fun_ty = exprType fun
1112 cont_ty = contResultType env fun_ty cont
1113 co = mkUnsafeCoercion fun_ty cont_ty
1114 mk_coerce expr | cont_ty `coreEqType` fun_ty = expr
1115 | otherwise = mkCoerce co expr
1117 rebuildCall env fun info (ApplyTo _ (Type arg_ty) se cont)
1118 = do { ty' <- simplType (se `setInScope` env) arg_ty
1119 ; rebuildCall env (fun `App` Type ty') info cont }
1122 (ArgInfo { ai_rules = has_rules, ai_strs = str:strs, ai_discs = disc:discs })
1123 (ApplyTo _ arg arg_se cont)
1124 | str -- Strict argument
1125 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1126 simplExprF (arg_se `setFloats` env) arg
1127 (StrictArg fun cci arg_info' cont)
1130 | otherwise -- Lazy argument
1131 -- DO NOT float anything outside, hence simplExprC
1132 -- There is no benefit (unlike in a let-binding), and we'd
1133 -- have to be very careful about bogus strictness through
1134 -- floating a demanded let.
1135 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1137 ; rebuildCall env (fun `App` arg') arg_info' cont }
1139 arg_info' = ArgInfo { ai_rules = has_rules, ai_strs = strs, ai_discs = discs }
1140 cci | has_rules || disc > 0 = ArgCtxt has_rules disc -- Be keener here
1141 | otherwise = BoringCtxt -- Nothing interesting
1143 rebuildCall env fun _ cont
1144 = rebuild env fun cont
1149 This part of the simplifier may break the no-shadowing invariant
1151 f (...(\a -> e)...) (case y of (a,b) -> e')
1152 where f is strict in its second arg
1153 If we simplify the innermost one first we get (...(\a -> e)...)
1154 Simplifying the second arg makes us float the case out, so we end up with
1155 case y of (a,b) -> f (...(\a -> e)...) e'
1156 So the output does not have the no-shadowing invariant. However, there is
1157 no danger of getting name-capture, because when the first arg was simplified
1158 we used an in-scope set that at least mentioned all the variables free in its
1159 static environment, and that is enough.
1161 We can't just do innermost first, or we'd end up with a dual problem:
1162 case x of (a,b) -> f e (...(\a -> e')...)
1164 I spent hours trying to recover the no-shadowing invariant, but I just could
1165 not think of an elegant way to do it. The simplifier is already knee-deep in
1166 continuations. We have to keep the right in-scope set around; AND we have
1167 to get the effect that finding (error "foo") in a strict arg position will
1168 discard the entire application and replace it with (error "foo"). Getting
1169 all this at once is TOO HARD!
1171 %************************************************************************
1173 Rebuilding a cse expression
1175 %************************************************************************
1177 Blob of helper functions for the "case-of-something-else" situation.
1180 ---------------------------------------------------------
1181 -- Eliminate the case if possible
1183 rebuildCase :: SimplEnv
1184 -> OutExpr -- Scrutinee
1185 -> InId -- Case binder
1186 -> [InAlt] -- Alternatives (inceasing order)
1188 -> SimplM (SimplEnv, OutExpr)
1190 --------------------------------------------------
1191 -- 1. Eliminate the case if there's a known constructor
1192 --------------------------------------------------
1194 rebuildCase env scrut case_bndr alts cont
1195 | Just (con,args) <- exprIsConApp_maybe scrut
1196 -- Works when the scrutinee is a variable with a known unfolding
1197 -- as well as when it's an explicit constructor application
1198 = knownCon env scrut (DataAlt con) args case_bndr alts cont
1200 | Lit lit <- scrut -- No need for same treatment as constructors
1201 -- because literals are inlined more vigorously
1202 = knownCon env scrut (LitAlt lit) [] case_bndr alts cont
1205 --------------------------------------------------
1206 -- 2. Eliminate the case if scrutinee is evaluated
1207 --------------------------------------------------
1209 rebuildCase env scrut case_bndr [(_, bndrs, rhs)] cont
1210 -- See if we can get rid of the case altogether
1211 -- See the extensive notes on case-elimination above
1212 -- mkCase made sure that if all the alternatives are equal,
1213 -- then there is now only one (DEFAULT) rhs
1214 | all isDeadBinder bndrs -- bndrs are [InId]
1216 -- Check that the scrutinee can be let-bound instead of case-bound
1217 , exprOkForSpeculation scrut
1218 -- OK not to evaluate it
1219 -- This includes things like (==# a# b#)::Bool
1220 -- so that we simplify
1221 -- case ==# a# b# of { True -> x; False -> x }
1224 -- This particular example shows up in default methods for
1225 -- comparision operations (e.g. in (>=) for Int.Int32)
1226 || exprIsHNF scrut -- It's already evaluated
1227 || var_demanded_later scrut -- It'll be demanded later
1229 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1230 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1231 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1232 -- its argument: case x of { y -> dataToTag# y }
1233 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1234 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1236 -- Also we don't want to discard 'seq's
1237 = do { tick (CaseElim case_bndr)
1238 ; env' <- simplNonRecX env case_bndr scrut
1239 ; simplExprF env' rhs cont }
1241 -- The case binder is going to be evaluated later,
1242 -- and the scrutinee is a simple variable
1243 var_demanded_later (Var v) = isStrictDmd (idNewDemandInfo case_bndr)
1244 && not (isTickBoxOp v)
1245 -- ugly hack; covering this case is what
1246 -- exprOkForSpeculation was intended for.
1247 var_demanded_later _ = False
1250 --------------------------------------------------
1251 -- 3. Catch-all case
1252 --------------------------------------------------
1254 rebuildCase env scrut case_bndr alts cont
1255 = do { -- Prepare the continuation;
1256 -- The new subst_env is in place
1257 (env', dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1259 -- Simplify the alternatives
1260 ; (scrut', case_bndr', alts') <- simplAlts env' scrut case_bndr alts dup_cont
1262 -- Check for empty alternatives
1263 ; if null alts' then
1264 -- This isn't strictly an error, although it is unusual.
1265 -- It's possible that the simplifer might "see" that
1266 -- an inner case has no accessible alternatives before
1267 -- it "sees" that the entire branch of an outer case is
1268 -- inaccessible. So we simply put an error case here instead.
1269 pprTrace "mkCase: null alts" (ppr case_bndr <+> ppr scrut) $
1270 let res_ty' = contResultType env' (substTy env' (coreAltsType alts)) dup_cont
1271 lit = mkStringLit "Impossible alternative"
1272 in return (env', mkApps (Var rUNTIME_ERROR_ID) [Type res_ty', lit])
1275 { case_expr <- mkCase scrut' case_bndr' alts'
1277 -- Notice that rebuild gets the in-scope set from env, not alt_env
1278 -- The case binder *not* scope over the whole returned case-expression
1279 ; rebuild env' case_expr nodup_cont } }
1282 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1283 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1284 way, there's a chance that v will now only be used once, and hence
1287 Note [no-case-of-case]
1288 ~~~~~~~~~~~~~~~~~~~~~~
1289 We *used* to suppress the binder-swap in case expressoins when
1290 -fno-case-of-case is on. Old remarks:
1291 "This happens in the first simplifier pass,
1292 and enhances full laziness. Here's the bad case:
1293 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1294 If we eliminate the inner case, we trap it inside the I# v -> arm,
1295 which might prevent some full laziness happening. I've seen this
1296 in action in spectral/cichelli/Prog.hs:
1297 [(m,n) | m <- [1..max], n <- [1..max]]
1298 Hence the check for NoCaseOfCase."
1299 However, now the full-laziness pass itself reverses the binder-swap, so this
1300 check is no longer necessary.
1302 Note [Suppressing the case binder-swap]
1303 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1304 There is another situation when it might make sense to suppress the
1305 case-expression binde-swap. If we have
1307 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1308 ...other cases .... }
1310 We'll perform the binder-swap for the outer case, giving
1312 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1313 ...other cases .... }
1315 But there is no point in doing it for the inner case, because w1 can't
1316 be inlined anyway. Furthermore, doing the case-swapping involves
1317 zapping w2's occurrence info (see paragraphs that follow), and that
1318 forces us to bind w2 when doing case merging. So we get
1320 case x of w1 { A -> let w2 = w1 in e1
1321 B -> let w2 = w1 in e2
1322 ...other cases .... }
1324 This is plain silly in the common case where w2 is dead.
1326 Even so, I can't see a good way to implement this idea. I tried
1327 not doing the binder-swap if the scrutinee was already evaluated
1328 but that failed big-time:
1332 case v of w { MkT x ->
1333 case x of x1 { I# y1 ->
1334 case x of x2 { I# y2 -> ...
1336 Notice that because MkT is strict, x is marked "evaluated". But to
1337 eliminate the last case, we must either make sure that x (as well as
1338 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1339 the binder-swap. So this whole note is a no-op.
1343 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1344 any occurrence info (eg IAmDead) in the case binder, because the
1345 case-binder now effectively occurs whenever v does. AND we have to do
1346 the same for the pattern-bound variables! Example:
1348 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1350 Here, b and p are dead. But when we move the argment inside the first
1351 case RHS, and eliminate the second case, we get
1353 case x of { (a,b) -> a b }
1355 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1358 Indeed, this can happen anytime the case binder isn't dead:
1359 case <any> of x { (a,b) ->
1360 case x of { (p,q) -> p } }
1361 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1362 The point is that we bring into the envt a binding
1364 after the outer case, and that makes (a,b) alive. At least we do unless
1365 the case binder is guaranteed dead.
1369 Consider case (v `cast` co) of x { I# ->
1370 ... (case (v `cast` co) of {...}) ...
1371 We'd like to eliminate the inner case. We can get this neatly by
1372 arranging that inside the outer case we add the unfolding
1373 v |-> x `cast` (sym co)
1374 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1376 Note [Improving seq]
1379 type family F :: * -> *
1380 type instance F Int = Int
1382 ... case e of x { DEFAULT -> rhs } ...
1384 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1386 case e `cast` co of x'::Int
1387 I# x# -> let x = x' `cast` sym co
1390 so that 'rhs' can take advantage of the form of x'. Notice that Note
1391 [Case of cast] may then apply to the result.
1393 This showed up in Roman's experiments. Example:
1394 foo :: F Int -> Int -> Int
1395 foo t n = t `seq` bar n
1398 bar n = bar (n - case t of TI i -> i)
1399 Here we'd like to avoid repeated evaluating t inside the loop, by
1400 taking advantage of the `seq`.
1402 At one point I did transformation in LiberateCase, but it's more robust here.
1403 (Otherwise, there's a danger that we'll simply drop the 'seq' altogether, before
1404 LiberateCase gets to see it.)
1406 Note [Case elimination]
1407 ~~~~~~~~~~~~~~~~~~~~~~~
1408 The case-elimination transformation discards redundant case expressions.
1409 Start with a simple situation:
1411 case x# of ===> e[x#/y#]
1414 (when x#, y# are of primitive type, of course). We can't (in general)
1415 do this for algebraic cases, because we might turn bottom into
1418 The code in SimplUtils.prepareAlts has the effect of generalise this
1419 idea to look for a case where we're scrutinising a variable, and we
1420 know that only the default case can match. For example:
1424 DEFAULT -> ...(case x of
1428 Here the inner case is first trimmed to have only one alternative, the
1429 DEFAULT, after which it's an instance of the previous case. This
1430 really only shows up in eliminating error-checking code.
1432 We also make sure that we deal with this very common case:
1437 Here we are using the case as a strict let; if x is used only once
1438 then we want to inline it. We have to be careful that this doesn't
1439 make the program terminate when it would have diverged before, so we
1441 - e is already evaluated (it may so if e is a variable)
1442 - x is used strictly, or
1444 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1446 case e of ===> case e of DEFAULT -> r
1450 Now again the case may be elminated by the CaseElim transformation.
1453 Further notes about case elimination
1454 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1455 Consider: test :: Integer -> IO ()
1458 Turns out that this compiles to:
1461 eta1 :: State# RealWorld ->
1462 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1464 (PrelNum.jtos eta ($w[] @ Char))
1466 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1468 Notice the strange '<' which has no effect at all. This is a funny one.
1469 It started like this:
1471 f x y = if x < 0 then jtos x
1472 else if y==0 then "" else jtos x
1474 At a particular call site we have (f v 1). So we inline to get
1476 if v < 0 then jtos x
1477 else if 1==0 then "" else jtos x
1479 Now simplify the 1==0 conditional:
1481 if v<0 then jtos v else jtos v
1483 Now common-up the two branches of the case:
1485 case (v<0) of DEFAULT -> jtos v
1487 Why don't we drop the case? Because it's strict in v. It's technically
1488 wrong to drop even unnecessary evaluations, and in practice they
1489 may be a result of 'seq' so we *definitely* don't want to drop those.
1490 I don't really know how to improve this situation.
1494 simplCaseBinder :: SimplEnv -> OutExpr -> OutId -> [InAlt]
1495 -> SimplM (SimplEnv, OutExpr, OutId)
1496 simplCaseBinder env0 scrut0 case_bndr0 alts
1497 = do { (env1, case_bndr1) <- simplBinder env0 case_bndr0
1499 ; fam_envs <- getFamEnvs
1500 ; (env2, scrut2, case_bndr2) <- improve_seq fam_envs env1 scrut0
1501 case_bndr0 case_bndr1 alts
1502 -- Note [Improving seq]
1504 ; let (env3, case_bndr3) = improve_case_bndr env2 scrut2 case_bndr2
1505 -- Note [Case of cast]
1507 ; return (env3, scrut2, case_bndr3) }
1510 improve_seq fam_envs env scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1511 | Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1512 = do { case_bndr2 <- newId (fsLit "nt") ty2
1513 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1514 env2 = extendIdSubst env case_bndr rhs
1515 ; return (env2, scrut `Cast` co, case_bndr2) }
1517 improve_seq _ env scrut _ case_bndr1 _
1518 = return (env, scrut, case_bndr1)
1521 improve_case_bndr env scrut case_bndr
1522 -- See Note [no-case-of-case]
1523 -- | switchIsOn (getSwitchChecker env) NoCaseOfCase
1524 -- = (env, case_bndr)
1526 | otherwise -- Failed try; see Note [Suppressing the case binder-swap]
1527 -- not (isEvaldUnfolding (idUnfolding v))
1529 Var v -> (modifyInScope env1 v case_bndr', case_bndr')
1530 -- Note about using modifyInScope for v here
1531 -- We could extend the substitution instead, but it would be
1532 -- a hack because then the substitution wouldn't be idempotent
1533 -- any more (v is an OutId). And this does just as well.
1535 Cast (Var v) co -> (addBinderUnfolding env1 v rhs, case_bndr')
1537 rhs = Cast (Var case_bndr') (mkSymCoercion co)
1539 _ -> (env, case_bndr)
1541 case_bndr' = zapOccInfo case_bndr
1542 env1 = modifyInScope env case_bndr case_bndr'
1545 zapOccInfo :: InId -> InId -- See Note [zapOccInfo]
1546 zapOccInfo b = b `setIdOccInfo` NoOccInfo
1550 simplAlts does two things:
1552 1. Eliminate alternatives that cannot match, including the
1553 DEFAULT alternative.
1555 2. If the DEFAULT alternative can match only one possible constructor,
1556 then make that constructor explicit.
1558 case e of x { DEFAULT -> rhs }
1560 case e of x { (a,b) -> rhs }
1561 where the type is a single constructor type. This gives better code
1562 when rhs also scrutinises x or e.
1564 Here "cannot match" includes knowledge from GADTs
1566 It's a good idea do do this stuff before simplifying the alternatives, to
1567 avoid simplifying alternatives we know can't happen, and to come up with
1568 the list of constructors that are handled, to put into the IdInfo of the
1569 case binder, for use when simplifying the alternatives.
1571 Eliminating the default alternative in (1) isn't so obvious, but it can
1574 data Colour = Red | Green | Blue
1583 DEFAULT -> [ case y of ... ]
1585 If we inline h into f, the default case of the inlined h can't happen.
1586 If we don't notice this, we may end up filtering out *all* the cases
1587 of the inner case y, which give us nowhere to go!
1591 simplAlts :: SimplEnv
1593 -> InId -- Case binder
1594 -> [InAlt] -- Non-empty
1596 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1597 -- Like simplExpr, this just returns the simplified alternatives;
1598 -- it not return an environment
1600 simplAlts env scrut case_bndr alts cont'
1601 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1602 do { let alt_env = zapFloats env
1603 ; (alt_env', scrut', case_bndr') <- simplCaseBinder alt_env scrut case_bndr alts
1605 ; (imposs_deflt_cons, in_alts) <- prepareAlts alt_env' scrut case_bndr' alts
1607 ; alts' <- mapM (simplAlt alt_env' imposs_deflt_cons case_bndr' cont') in_alts
1608 ; return (scrut', case_bndr', alts') }
1610 ------------------------------------
1611 simplAlt :: SimplEnv
1612 -> [AltCon] -- These constructors can't be present when
1613 -- matching the DEFAULT alternative
1614 -> OutId -- The case binder
1619 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1620 = ASSERT( null bndrs )
1621 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1622 -- Record the constructors that the case-binder *can't* be.
1623 ; rhs' <- simplExprC env' rhs cont'
1624 ; return (DEFAULT, [], rhs') }
1626 simplAlt env _ case_bndr' cont' (LitAlt lit, bndrs, rhs)
1627 = ASSERT( null bndrs )
1628 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1629 ; rhs' <- simplExprC env' rhs cont'
1630 ; return (LitAlt lit, [], rhs') }
1632 simplAlt env _ case_bndr' cont' (DataAlt con, vs, rhs)
1633 = do { -- Deal with the pattern-bound variables
1634 -- Mark the ones that are in ! positions in the
1635 -- data constructor as certainly-evaluated.
1636 -- NB: simplLamBinders preserves this eval info
1637 let vs_with_evals = add_evals (dataConRepStrictness con)
1638 ; (env', vs') <- simplLamBndrs env vs_with_evals
1640 -- Bind the case-binder to (con args)
1641 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1642 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1643 env'' = addBinderUnfolding env' case_bndr'
1644 (mkConApp con con_args)
1646 ; rhs' <- simplExprC env'' rhs cont'
1647 ; return (DataAlt con, vs', rhs') }
1649 -- add_evals records the evaluated-ness of the bound variables of
1650 -- a case pattern. This is *important*. Consider
1651 -- data T = T !Int !Int
1653 -- case x of { T a b -> T (a+1) b }
1655 -- We really must record that b is already evaluated so that we don't
1656 -- go and re-evaluate it when constructing the result.
1657 -- See Note [Data-con worker strictness] in MkId.lhs
1662 go (v:vs') strs | isTyVar v = v : go vs' strs
1663 go (v:vs') (str:strs)
1664 | isMarkedStrict str = evald_v : go vs' strs
1665 | otherwise = zapped_v : go vs' strs
1667 zapped_v = zap_occ_info v
1668 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1669 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr the_strs)
1671 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1673 -- to the envt; so vs are now very much alive
1674 -- Note [Aug06] I can't see why this actually matters, but it's neater
1675 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1676 -- ==> case e of t { (a,b) -> ...(a)... }
1677 -- Look, Ma, a is alive now.
1678 zap_occ_info | isDeadBinder case_bndr' = \ident -> ident
1679 | otherwise = zapOccInfo
1681 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1682 addBinderUnfolding env bndr rhs
1683 = modifyInScope env bndr (bndr `setIdUnfolding` mkUnfolding False rhs)
1685 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1686 addBinderOtherCon env bndr cons
1687 = modifyInScope env bndr (bndr `setIdUnfolding` mkOtherCon cons)
1691 %************************************************************************
1693 \subsection{Known constructor}
1695 %************************************************************************
1697 We are a bit careful with occurrence info. Here's an example
1699 (\x* -> case x of (a*, b) -> f a) (h v, e)
1701 where the * means "occurs once". This effectively becomes
1702 case (h v, e) of (a*, b) -> f a)
1704 let a* = h v; b = e in f a
1708 All this should happen in one sweep.
1711 knownCon :: SimplEnv -> OutExpr -> AltCon
1712 -> [OutExpr] -- Args *including* the universal args
1713 -> InId -> [InAlt] -> SimplCont
1714 -> SimplM (SimplEnv, OutExpr)
1716 knownCon env scrut con args bndr alts cont
1717 = do { tick (KnownBranch bndr)
1718 ; knownAlt env scrut args bndr (findAlt con alts) cont }
1720 knownAlt :: SimplEnv -> OutExpr -> [OutExpr]
1721 -> InId -> (AltCon, [CoreBndr], InExpr) -> SimplCont
1722 -> SimplM (SimplEnv, OutExpr)
1723 knownAlt env scrut _ bndr (DEFAULT, bs, rhs) cont
1725 do { env' <- simplNonRecX env bndr scrut
1726 -- This might give rise to a binding with non-atomic args
1727 -- like x = Node (f x) (g x)
1728 -- but simplNonRecX will atomic-ify it
1729 ; simplExprF env' rhs cont }
1731 knownAlt env scrut _ bndr (LitAlt _, bs, rhs) cont
1733 do { env' <- simplNonRecX env bndr scrut
1734 ; simplExprF env' rhs cont }
1736 knownAlt env scrut the_args bndr (DataAlt dc, bs, rhs) cont
1737 = do { let dead_bndr = isDeadBinder bndr -- bndr is an InId
1738 n_drop_tys = length (dataConUnivTyVars dc)
1739 ; env' <- bind_args env dead_bndr bs (drop n_drop_tys the_args)
1741 -- It's useful to bind bndr to scrut, rather than to a fresh
1742 -- binding x = Con arg1 .. argn
1743 -- because very often the scrut is a variable, so we avoid
1744 -- creating, and then subsequently eliminating, a let-binding
1745 -- BUT, if scrut is a not a variable, we must be careful
1746 -- about duplicating the arg redexes; in that case, make
1747 -- a new con-app from the args
1748 bndr_rhs = case scrut of
1751 con_app = mkConApp dc (take n_drop_tys the_args ++ con_args)
1752 con_args = [substExpr env' (varToCoreExpr b) | b <- bs]
1753 -- args are aready OutExprs, but bs are InIds
1755 ; env'' <- simplNonRecX env' bndr bndr_rhs
1756 ; -- pprTrace "knownCon2" (ppr bs $$ ppr rhs $$ ppr (seIdSubst env'')) $
1757 simplExprF env'' rhs cont }
1760 bind_args env' _ [] _ = return env'
1762 bind_args env' dead_bndr (b:bs') (Type ty : args)
1763 = ASSERT( isTyVar b )
1764 bind_args (extendTvSubst env' b ty) dead_bndr bs' args
1766 bind_args env' dead_bndr (b:bs') (arg : args)
1768 do { let b' = if dead_bndr then b else zapOccInfo b
1769 -- Note that the binder might be "dead", because it doesn't
1770 -- occur in the RHS; and simplNonRecX may therefore discard
1771 -- it via postInlineUnconditionally.
1772 -- Nevertheless we must keep it if the case-binder is alive,
1773 -- because it may be used in the con_app. See Note [zapOccInfo]
1774 ; env'' <- simplNonRecX env' b' arg
1775 ; bind_args env'' dead_bndr bs' args }
1778 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr the_args $$
1779 text "scrut:" <+> ppr scrut
1783 %************************************************************************
1785 \subsection{Duplicating continuations}
1787 %************************************************************************
1790 prepareCaseCont :: SimplEnv
1791 -> [InAlt] -> SimplCont
1792 -> SimplM (SimplEnv, SimplCont,SimplCont)
1793 -- Return a duplicatable continuation, a non-duplicable part
1794 -- plus some extra bindings (that scope over the entire
1797 -- No need to make it duplicatable if there's only one alternative
1798 prepareCaseCont env [_] cont = return (env, cont, mkBoringStop)
1799 prepareCaseCont env _ cont = mkDupableCont env cont
1803 mkDupableCont :: SimplEnv -> SimplCont
1804 -> SimplM (SimplEnv, SimplCont, SimplCont)
1806 mkDupableCont env cont
1807 | contIsDupable cont
1808 = return (env, cont, mkBoringStop)
1810 mkDupableCont _ (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1812 mkDupableCont env (CoerceIt ty cont)
1813 = do { (env', dup, nodup) <- mkDupableCont env cont
1814 ; return (env', CoerceIt ty dup, nodup) }
1816 mkDupableCont env cont@(StrictBind {})
1817 = return (env, mkBoringStop, cont)
1818 -- See Note [Duplicating strict continuations]
1820 mkDupableCont env cont@(StrictArg {})
1821 = return (env, mkBoringStop, cont)
1822 -- See Note [Duplicating strict continuations]
1824 mkDupableCont env (ApplyTo _ arg se cont)
1825 = -- e.g. [...hole...] (...arg...)
1827 -- let a = ...arg...
1828 -- in [...hole...] a
1829 do { (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1830 ; arg' <- simplExpr (se `setInScope` env') arg
1831 ; (env'', arg'') <- makeTrivial env' arg'
1832 ; let app_cont = ApplyTo OkToDup arg'' (zapSubstEnv env') dup_cont
1833 ; return (env'', app_cont, nodup_cont) }
1835 mkDupableCont env cont@(Select _ case_bndr [(_, bs, _rhs)] _ _)
1836 -- See Note [Single-alternative case]
1837 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1838 -- | not (isDeadBinder case_bndr)
1839 | all isDeadBinder bs -- InIds
1840 && not (isUnLiftedType (idType case_bndr))
1841 -- Note [Single-alternative-unlifted]
1842 = return (env, mkBoringStop, cont)
1844 mkDupableCont env (Select _ case_bndr alts se cont)
1845 = -- e.g. (case [...hole...] of { pi -> ei })
1847 -- let ji = \xij -> ei
1848 -- in case [...hole...] of { pi -> ji xij }
1849 do { tick (CaseOfCase case_bndr)
1850 ; (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1851 -- NB: call mkDupableCont here, *not* prepareCaseCont
1852 -- We must make a duplicable continuation, whereas prepareCaseCont
1853 -- doesn't when there is a single case branch
1855 ; let alt_env = se `setInScope` env'
1856 ; (alt_env', case_bndr') <- simplBinder alt_env case_bndr
1857 ; alts' <- mapM (simplAlt alt_env' [] case_bndr' dup_cont) alts
1858 -- Safe to say that there are no handled-cons for the DEFAULT case
1859 -- NB: simplBinder does not zap deadness occ-info, so
1860 -- a dead case_bndr' will still advertise its deadness
1861 -- This is really important because in
1862 -- case e of b { (# p,q #) -> ... }
1863 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1864 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1865 -- In the new alts we build, we have the new case binder, so it must retain
1867 -- NB: we don't use alt_env further; it has the substEnv for
1868 -- the alternatives, and we don't want that
1870 ; (env'', alts'') <- mkDupableAlts env' case_bndr' alts'
1871 ; return (env'', -- Note [Duplicated env]
1872 Select OkToDup case_bndr' alts'' (zapSubstEnv env'') mkBoringStop,
1876 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1877 -> SimplM (SimplEnv, [InAlt])
1878 -- Absorbs the continuation into the new alternatives
1880 mkDupableAlts env case_bndr' the_alts
1883 go env0 [] = return (env0, [])
1885 = do { (env1, alt') <- mkDupableAlt env0 case_bndr' alt
1886 ; (env2, alts') <- go env1 alts
1887 ; return (env2, alt' : alts' ) }
1889 mkDupableAlt :: SimplEnv -> OutId -> (AltCon, [CoreBndr], CoreExpr)
1890 -> SimplM (SimplEnv, (AltCon, [CoreBndr], CoreExpr))
1891 mkDupableAlt env case_bndr' (con, bndrs', rhs')
1892 | exprIsDupable rhs' -- Note [Small alternative rhs]
1893 = return (env, (con, bndrs', rhs'))
1895 = do { let rhs_ty' = exprType rhs'
1896 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1898 | isTyVar bndr = True -- Abstract over all type variables just in case
1899 | otherwise = not (isDeadBinder bndr)
1900 -- The deadness info on the new Ids is preserved by simplBinders
1902 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1903 <- if (any isId used_bndrs')
1904 then return (used_bndrs', varsToCoreExprs used_bndrs')
1905 else do { rw_id <- newId (fsLit "w") realWorldStatePrimTy
1906 ; return ([rw_id], [Var realWorldPrimId]) }
1908 ; join_bndr <- newId (fsLit "$j") (mkPiTypes final_bndrs' rhs_ty')
1909 -- Note [Funky mkPiTypes]
1911 ; let -- We make the lambdas into one-shot-lambdas. The
1912 -- join point is sure to be applied at most once, and doing so
1913 -- prevents the body of the join point being floated out by
1914 -- the full laziness pass
1915 really_final_bndrs = map one_shot final_bndrs'
1916 one_shot v | isId v = setOneShotLambda v
1918 join_rhs = mkLams really_final_bndrs rhs'
1919 join_call = mkApps (Var join_bndr) final_args
1921 ; return (addPolyBind NotTopLevel env (NonRec join_bndr join_rhs), (con, bndrs', join_call)) }
1922 -- See Note [Duplicated env]
1925 Note [Duplicated env]
1926 ~~~~~~~~~~~~~~~~~~~~~
1927 Some of the alternatives are simplified, but have not been turned into a join point
1928 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
1929 bind the join point, because it might to do PostInlineUnconditionally, and
1930 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
1931 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
1932 at worst delays the join-point inlining.
1934 Note [Small alterantive rhs]
1935 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1936 It is worth checking for a small RHS because otherwise we
1937 get extra let bindings that may cause an extra iteration of the simplifier to
1938 inline back in place. Quite often the rhs is just a variable or constructor.
1939 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1940 iterations because the version with the let bindings looked big, and so wasn't
1941 inlined, but after the join points had been inlined it looked smaller, and so
1944 NB: we have to check the size of rhs', not rhs.
1945 Duplicating a small InAlt might invalidate occurrence information
1946 However, if it *is* dupable, we return the *un* simplified alternative,
1947 because otherwise we'd need to pair it up with an empty subst-env....
1948 but we only have one env shared between all the alts.
1949 (Remember we must zap the subst-env before re-simplifying something).
1950 Rather than do this we simply agree to re-simplify the original (small) thing later.
1952 Note [Funky mkPiTypes]
1953 ~~~~~~~~~~~~~~~~~~~~~~
1954 Notice the funky mkPiTypes. If the contructor has existentials
1955 it's possible that the join point will be abstracted over
1956 type varaibles as well as term variables.
1957 Example: Suppose we have
1958 data T = forall t. C [t]
1960 case (case e of ...) of
1962 We get the join point
1963 let j :: forall t. [t] -> ...
1964 j = /\t \xs::[t] -> rhs
1966 case (case e of ...) of
1967 C t xs::[t] -> j t xs
1969 Note [Join point abstaction]
1970 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1971 If we try to lift a primitive-typed something out
1972 for let-binding-purposes, we will *caseify* it (!),
1973 with potentially-disastrous strictness results. So
1974 instead we turn it into a function: \v -> e
1975 where v::State# RealWorld#. The value passed to this function
1976 is realworld#, which generates (almost) no code.
1978 There's a slight infelicity here: we pass the overall
1979 case_bndr to all the join points if it's used in *any* RHS,
1980 because we don't know its usage in each RHS separately
1982 We used to say "&& isUnLiftedType rhs_ty'" here, but now
1983 we make the join point into a function whenever used_bndrs'
1984 is empty. This makes the join-point more CPR friendly.
1985 Consider: let j = if .. then I# 3 else I# 4
1986 in case .. of { A -> j; B -> j; C -> ... }
1988 Now CPR doesn't w/w j because it's a thunk, so
1989 that means that the enclosing function can't w/w either,
1990 which is a lose. Here's the example that happened in practice:
1991 kgmod :: Int -> Int -> Int
1992 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1996 I have seen a case alternative like this:
1998 It's a bit silly to add the realWorld dummy arg in this case, making
2001 (the \v alone is enough to make CPR happy) but I think it's rare
2003 Note [Duplicating strict continuations]
2004 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2005 Do *not* duplicate StrictBind and StritArg continuations. We gain
2006 nothing by propagating them into the expressions, and we do lose a
2007 lot. Here's an example:
2008 && (case x of { T -> F; F -> T }) E
2009 Now, && is strict so we end up simplifying the case with
2010 an ArgOf continuation. If we let-bind it, we get
2012 let $j = \v -> && v E
2013 in simplExpr (case x of { T -> F; F -> T })
2015 And after simplifying more we get
2017 let $j = \v -> && v E
2018 in case x of { T -> $j F; F -> $j T }
2019 Which is a Very Bad Thing
2021 The desire not to duplicate is the entire reason that
2022 mkDupableCont returns a pair of continuations.
2024 The original plan had:
2025 e.g. (...strict-fn...) [...hole...]
2027 let $j = \a -> ...strict-fn...
2030 Note [Single-alternative cases]
2031 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2032 This case is just like the ArgOf case. Here's an example:
2036 case (case x of I# x' ->
2038 True -> I# (negate# x')
2039 False -> I# x') of y {
2041 Because the (case x) has only one alternative, we'll transform to
2043 case (case x' <# 0# of
2044 True -> I# (negate# x')
2045 False -> I# x') of y {
2047 But now we do *NOT* want to make a join point etc, giving
2049 let $j = \y -> MkT y
2051 True -> $j (I# (negate# x'))
2053 In this case the $j will inline again, but suppose there was a big
2054 strict computation enclosing the orginal call to MkT. Then, it won't
2055 "see" the MkT any more, because it's big and won't get duplicated.
2056 And, what is worse, nothing was gained by the case-of-case transform.
2058 When should use this case of mkDupableCont?
2059 However, matching on *any* single-alternative case is a *disaster*;
2060 e.g. case (case ....) of (a,b) -> (# a,b #)
2061 We must push the outer case into the inner one!
2064 * Match [(DEFAULT,_,_)], but in the common case of Int,
2065 the alternative-filling-in code turned the outer case into
2066 case (...) of y { I# _ -> MkT y }
2068 * Match on single alternative plus (not (isDeadBinder case_bndr))
2069 Rationale: pushing the case inwards won't eliminate the construction.
2070 But there's a risk of
2071 case (...) of y { (a,b) -> let z=(a,b) in ... }
2072 Now y looks dead, but it'll come alive again. Still, this
2073 seems like the best option at the moment.
2075 * Match on single alternative plus (all (isDeadBinder bndrs))
2076 Rationale: this is essentially seq.
2078 * Match when the rhs is *not* duplicable, and hence would lead to a
2079 join point. This catches the disaster-case above. We can test
2080 the *un-simplified* rhs, which is fine. It might get bigger or
2081 smaller after simplification; if it gets smaller, this case might
2082 fire next time round. NB also that we must test contIsDupable
2083 case_cont *btoo, because case_cont might be big!
2085 HOWEVER: I found that this version doesn't work well, because
2086 we can get let x = case (...) of { small } in ...case x...
2087 When x is inlined into its full context, we find that it was a bad
2088 idea to have pushed the outer case inside the (...) case.
2090 Note [Single-alternative-unlifted]
2091 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2092 Here's another single-alternative where we really want to do case-of-case:
2100 case y_s6X of tpl_s7m {
2101 M1.Mk1 ipv_s70 -> ipv_s70;
2102 M1.Mk2 ipv_s72 -> ipv_s72;
2108 case x_s74 of tpl_s7n {
2109 M1.Mk1 ipv_s77 -> ipv_s77;
2110 M1.Mk2 ipv_s79 -> ipv_s79;
2114 { __DEFAULT -> ==# [wild1_s7b wild_s7c];
2118 So the outer case is doing *nothing at all*, other than serving as a
2119 join-point. In this case we really want to do case-of-case and decide
2120 whether to use a real join point or just duplicate the continuation.
2122 Hence: check whether the case binder's type is unlifted, because then
2123 the outer case is *not* a seq.