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 || not (activeInline env old_bndr)
572 addPolyBind :: TopLevelFlag -> SimplEnv -> OutBind -> SimplEnv
573 -- Add a new binding to the environment, complete with its unfolding
574 -- but *do not* do postInlineUnconditionally, because we have already
575 -- processed some of the scope of the binding
576 -- We still want the unfolding though. Consider
578 -- x = /\a. let y = ... in Just y
580 -- Then we float the y-binding out (via abstractFloats and addPolyBind)
581 -- but 'x' may well then be inlined in 'body' in which case we'd like the
582 -- opportunity to inline 'y' too.
584 addPolyBind top_lvl env (NonRec poly_id rhs)
585 = addNonRecWithUnf env poly_id rhs unfolding NoWorker
587 unfolding | not (activeInline env poly_id) = NoUnfolding
588 | otherwise = mkUnfolding (isTopLevel top_lvl) rhs
589 -- addNonRecWithInfo adds the new binding in the
590 -- proper way (ie complete with unfolding etc),
591 -- and extends the in-scope set
593 addPolyBind _ env bind@(Rec _) = extendFloats env bind
594 -- Hack: letrecs are more awkward, so we extend "by steam"
595 -- without adding unfoldings etc. At worst this leads to
596 -- more simplifier iterations
599 addNonRecWithUnf :: SimplEnv
600 -> OutId -> OutExpr -- New binder and RHS
601 -> Unfolding -> WorkerInfo -- and unfolding
603 -- Add suitable IdInfo to the Id, add the binding to the floats, and extend the in-scope set
604 addNonRecWithUnf env new_bndr rhs unfolding wkr
605 = final_id `seq` -- This seq forces the Id, and hence its IdInfo,
606 -- and hence any inner substitutions
607 addNonRec env final_id rhs
608 -- The addNonRec adds it to the in-scope set too
611 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity rhs
614 -- Add the unfolding *only* for non-loop-breakers
615 -- Making loop breakers not have an unfolding at all
616 -- means that we can avoid tests in exprIsConApp, for example.
617 -- This is important: if exprIsConApp says 'yes' for a recursive
618 -- thing, then we can get into an infinite loop
621 -- If the unfolding is a value, the demand info may
622 -- go pear-shaped, so we nuke it. Example:
624 -- case x of (p,q) -> h p q x
625 -- Here x is certainly demanded. But after we've nuked
626 -- the case, we'll get just
627 -- let x = (a,b) in h a b x
628 -- and now x is not demanded (I'm assuming h is lazy)
629 -- This really happens. Similarly
630 -- let f = \x -> e in ...f..f...
631 -- After inlining f at some of its call sites the original binding may
632 -- (for example) be no longer strictly demanded.
633 -- The solution here is a bit ad hoc...
634 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
637 final_info | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
638 | otherwise = info_w_unf
640 final_id = new_bndr `setIdInfo` final_info
645 %************************************************************************
647 \subsection[Simplify-simplExpr]{The main function: simplExpr}
649 %************************************************************************
651 The reason for this OutExprStuff stuff is that we want to float *after*
652 simplifying a RHS, not before. If we do so naively we get quadratic
653 behaviour as things float out.
655 To see why it's important to do it after, consider this (real) example:
669 a -- Can't inline a this round, cos it appears twice
673 Each of the ==> steps is a round of simplification. We'd save a
674 whole round if we float first. This can cascade. Consider
679 let f = let d1 = ..d.. in \y -> e
683 in \x -> ...(\y ->e)...
685 Only in this second round can the \y be applied, and it
686 might do the same again.
690 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
691 simplExpr env expr = simplExprC env expr mkBoringStop
693 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
694 -- Simplify an expression, given a continuation
695 simplExprC env expr cont
696 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
697 do { (env', expr') <- simplExprF (zapFloats env) expr cont
698 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
699 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
700 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
701 return (wrapFloats env' expr') }
703 --------------------------------------------------
704 simplExprF :: SimplEnv -> InExpr -> SimplCont
705 -> SimplM (SimplEnv, OutExpr)
707 simplExprF env e cont
708 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
709 simplExprF' env e cont
711 simplExprF' :: SimplEnv -> InExpr -> SimplCont
712 -> SimplM (SimplEnv, OutExpr)
713 simplExprF' env (Var v) cont = simplVar env v cont
714 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
715 simplExprF' env (Note n expr) cont = simplNote env n expr cont
716 simplExprF' env (Cast body co) cont = simplCast env body co cont
717 simplExprF' env (App fun arg) cont = simplExprF env fun $
718 ApplyTo NoDup arg env cont
720 simplExprF' env expr@(Lam _ _) cont
721 = simplLam env (map zap bndrs) body cont
722 -- The main issue here is under-saturated lambdas
723 -- (\x1. \x2. e) arg1
724 -- Here x1 might have "occurs-once" occ-info, because occ-info
725 -- is computed assuming that a group of lambdas is applied
726 -- all at once. If there are too few args, we must zap the
729 n_args = countArgs cont
730 n_params = length bndrs
731 (bndrs, body) = collectBinders expr
732 zap | n_args >= n_params = \b -> b
733 | otherwise = \b -> if isTyVar b then b
735 -- NB: we count all the args incl type args
736 -- so we must count all the binders (incl type lambdas)
738 simplExprF' env (Type ty) cont
739 = ASSERT( contIsRhsOrArg cont )
740 do { ty' <- simplType env ty
741 ; rebuild env (Type ty') cont }
743 simplExprF' env (Case scrut bndr _ alts) cont
744 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
745 = -- Simplify the scrutinee with a Select continuation
746 simplExprF env scrut (Select NoDup bndr alts env cont)
749 = -- If case-of-case is off, simply simplify the case expression
750 -- in a vanilla Stop context, and rebuild the result around it
751 do { case_expr' <- simplExprC env scrut case_cont
752 ; rebuild env case_expr' cont }
754 case_cont = Select NoDup bndr alts env mkBoringStop
756 simplExprF' env (Let (Rec pairs) body) cont
757 = do { env' <- simplRecBndrs env (map fst pairs)
758 -- NB: bndrs' don't have unfoldings or rules
759 -- We add them as we go down
761 ; env'' <- simplRecBind env' NotTopLevel pairs
762 ; simplExprF env'' body cont }
764 simplExprF' env (Let (NonRec bndr rhs) body) cont
765 = simplNonRecE env bndr (rhs, env) ([], body) cont
767 ---------------------------------
768 simplType :: SimplEnv -> InType -> SimplM OutType
769 -- Kept monadic just so we can do the seqType
771 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
772 seqType new_ty `seq` return new_ty
774 new_ty = substTy env ty
778 %************************************************************************
780 \subsection{The main rebuilder}
782 %************************************************************************
785 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
786 -- At this point the substitution in the SimplEnv should be irrelevant
787 -- only the in-scope set and floats should matter
788 rebuild env expr cont0
789 = -- pprTrace "rebuild" (ppr expr $$ ppr cont0 $$ ppr (seFloats env)) $
791 Stop {} -> return (env, expr)
792 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
793 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
794 StrictArg fun _ info cont -> rebuildCall env (fun `App` expr) info cont
795 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
796 ; simplLam env' bs body cont }
797 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
798 ; rebuild env (App expr arg') cont }
802 %************************************************************************
806 %************************************************************************
809 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
810 -> SimplM (SimplEnv, OutExpr)
811 simplCast env body co0 cont0
812 = do { co1 <- simplType env co0
813 ; simplExprF env body (addCoerce co1 cont0) }
815 addCoerce co cont = add_coerce co (coercionKind co) cont
817 add_coerce _co (s1, k1) cont -- co :: ty~ty
818 | s1 `coreEqType` k1 = cont -- is a no-op
820 add_coerce co1 (s1, _k2) (CoerceIt co2 cont)
821 | (_l1, t1) <- coercionKind co2
822 -- coerce T1 S1 (coerce S1 K1 e)
825 -- coerce T1 K1 e, otherwise
827 -- For example, in the initial form of a worker
828 -- we may find (coerce T (coerce S (\x.e))) y
829 -- and we'd like it to simplify to e[y/x] in one round
831 , s1 `coreEqType` t1 = cont -- The coerces cancel out
832 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
834 add_coerce co (s1s2, _t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
835 -- (f `cast` g) ty ---> (f ty) `cast` (g @ ty)
836 -- This implements the PushT rule from the paper
837 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
838 , not (isCoVar tyvar)
839 = ApplyTo dup (Type ty') (zapSubstEnv env) (addCoerce (mkInstCoercion co ty') cont)
841 ty' = substTy (arg_se `setInScope` env) arg_ty
843 -- ToDo: the PushC rule is not implemented at all
845 add_coerce co (s1s2, _t1t2) (ApplyTo dup arg arg_se cont)
846 | not (isTypeArg arg) -- This implements the Push rule from the paper
847 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
848 -- co : s1s2 :=: t1t2
849 -- (coerce (T1->T2) (S1->S2) F) E
851 -- coerce T2 S2 (F (coerce S1 T1 E))
853 -- t1t2 must be a function type, T1->T2, because it's applied
854 -- to something but s1s2 might conceivably not be
856 -- When we build the ApplyTo we can't mix the out-types
857 -- with the InExpr in the argument, so we simply substitute
858 -- to make it all consistent. It's a bit messy.
859 -- But it isn't a common case.
861 -- Example of use: Trac #995
862 = ApplyTo dup new_arg (zapSubstEnv env) (addCoerce co2 cont)
864 -- we split coercion t1->t2 :=: s1->s2 into t1 :=: s1 and
865 -- t2 :=: s2 with left and right on the curried form:
866 -- (->) t1 t2 :=: (->) s1 s2
867 [co1, co2] = decomposeCo 2 co
868 new_arg = mkCoerce (mkSymCoercion co1) arg'
869 arg' = substExpr (arg_se `setInScope` env) arg
871 add_coerce co _ cont = CoerceIt co cont
875 %************************************************************************
879 %************************************************************************
882 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
883 -> SimplM (SimplEnv, OutExpr)
885 simplLam env [] body cont = simplExprF env body cont
888 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
889 = do { tick (BetaReduction bndr)
890 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
892 -- Not enough args, so there are real lambdas left to put in the result
893 simplLam env bndrs body cont
894 = do { (env', bndrs') <- simplLamBndrs env bndrs
895 ; body' <- simplExpr env' body
896 ; new_lam <- mkLam bndrs' body'
897 ; rebuild env' new_lam cont }
900 simplNonRecE :: SimplEnv
901 -> InId -- The binder
902 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
903 -> ([InBndr], InExpr) -- Body of the let/lambda
906 -> SimplM (SimplEnv, OutExpr)
908 -- simplNonRecE is used for
909 -- * non-top-level non-recursive lets in expressions
912 -- It deals with strict bindings, via the StrictBind continuation,
913 -- which may abort the whole process
915 -- The "body" of the binding comes as a pair of ([InId],InExpr)
916 -- representing a lambda; so we recurse back to simplLam
917 -- Why? Because of the binder-occ-info-zapping done before
918 -- the call to simplLam in simplExprF (Lam ...)
920 -- First deal with type applications and type lets
921 -- (/\a. e) (Type ty) and (let a = Type ty in e)
922 simplNonRecE env bndr (Type ty_arg, rhs_se) (bndrs, body) cont
923 = ASSERT( isTyVar bndr )
924 do { ty_arg' <- simplType (rhs_se `setInScope` env) ty_arg
925 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
927 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
928 | preInlineUnconditionally env NotTopLevel bndr rhs
929 = do { tick (PreInlineUnconditionally bndr)
930 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
933 = do { simplExprF (rhs_se `setFloats` env) rhs
934 (StrictBind bndr bndrs body env cont) }
937 = do { (env1, bndr1) <- simplNonRecBndr env bndr
938 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
939 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
940 ; simplLam env3 bndrs body cont }
944 %************************************************************************
948 %************************************************************************
951 -- Hack alert: we only distinguish subsumed cost centre stacks for the
952 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
953 simplNote :: SimplEnv -> Note -> CoreExpr -> SimplCont
954 -> SimplM (SimplEnv, OutExpr)
955 simplNote env (SCC cc) e cont
956 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
957 ; rebuild env (mkSCC cc e') cont }
959 -- See notes with SimplMonad.inlineMode
960 simplNote env InlineMe e cont
961 | Just (inside, outside) <- splitInlineCont cont -- Boring boring continuation; see notes above
962 = do { -- Don't inline inside an INLINE expression
963 e' <- simplExprC (setMode inlineMode env) e inside
964 ; rebuild env (mkInlineMe e') outside }
966 | otherwise -- Dissolve the InlineMe note if there's
967 -- an interesting context of any kind to combine with
968 -- (even a type application -- anything except Stop)
969 = simplExprF env e cont
971 simplNote env (CoreNote s) e cont = do
972 e' <- simplExpr env e
973 rebuild env (Note (CoreNote s) e') cont
977 %************************************************************************
979 \subsection{Dealing with calls}
981 %************************************************************************
984 simplVar :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
985 simplVar env var cont
986 = case substId env var of
987 DoneEx e -> simplExprF (zapSubstEnv env) e cont
988 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
989 DoneId var1 -> completeCall (zapSubstEnv env) var1 cont
990 -- Note [zapSubstEnv]
991 -- The template is already simplified, so don't re-substitute.
992 -- This is VITAL. Consider
994 -- let y = \z -> ...x... in
996 -- We'll clone the inner \x, adding x->x' in the id_subst
997 -- Then when we inline y, we must *not* replace x by x' in
998 -- the inlined copy!!
1000 ---------------------------------------------------------
1001 -- Dealing with a call site
1003 completeCall :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1004 completeCall env var cont
1005 = do { dflags <- getDOptsSmpl
1006 ; let (args,call_cont) = contArgs cont
1007 -- The args are OutExprs, obtained by *lazily* substituting
1008 -- in the args found in cont. These args are only examined
1009 -- to limited depth (unless a rule fires). But we must do
1010 -- the substitution; rule matching on un-simplified args would
1013 ------------- First try rules ----------------
1014 -- Do this before trying inlining. Some functions have
1015 -- rules *and* are strict; in this case, we don't want to
1016 -- inline the wrapper of the non-specialised thing; better
1017 -- to call the specialised thing instead.
1019 -- We used to use the black-listing mechanism to ensure that inlining of
1020 -- the wrapper didn't occur for things that have specialisations till a
1021 -- later phase, so but now we just try RULES first
1023 -- Note [Rules for recursive functions]
1024 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1025 -- You might think that we shouldn't apply rules for a loop breaker:
1026 -- doing so might give rise to an infinite loop, because a RULE is
1027 -- rather like an extra equation for the function:
1028 -- RULE: f (g x) y = x+y
1031 -- But it's too drastic to disable rules for loop breakers.
1032 -- Even the foldr/build rule would be disabled, because foldr
1033 -- is recursive, and hence a loop breaker:
1034 -- foldr k z (build g) = g k z
1035 -- So it's up to the programmer: rules can cause divergence
1036 ; rule_base <- getSimplRules
1037 ; let in_scope = getInScope env
1038 rules = getRules rule_base var
1039 maybe_rule = case activeRule dflags env of
1040 Nothing -> Nothing -- No rules apply
1041 Just act_fn -> lookupRule act_fn in_scope
1043 ; case maybe_rule of {
1044 Just (rule, rule_rhs) -> do
1045 tick (RuleFired (ru_name rule))
1046 (if dopt Opt_D_dump_rule_firings dflags then
1047 pprTrace "Rule fired" (vcat [
1048 text "Rule:" <+> ftext (ru_name rule),
1049 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1050 text "After: " <+> pprCoreExpr rule_rhs,
1051 text "Cont: " <+> ppr call_cont])
1054 simplExprF env rule_rhs (dropArgs (ruleArity rule) cont)
1055 -- The ruleArity says how many args the rule consumed
1057 ; Nothing -> do -- No rules
1059 ------------- Next try inlining ----------------
1060 { let arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1061 n_val_args = length arg_infos
1062 interesting_cont = interestingCallContext call_cont
1063 active_inline = activeInline env var
1064 maybe_inline = callSiteInline dflags active_inline var
1065 (null args) arg_infos interesting_cont
1066 ; case maybe_inline of {
1067 Just unfolding -- There is an inlining!
1068 -> do { tick (UnfoldingDone var)
1069 ; (if dopt Opt_D_dump_inlinings dflags then
1070 pprTrace ("Inlining done" ++ showSDoc (ppr var)) (vcat [
1071 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1072 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1073 text "Cont: " <+> ppr call_cont])
1076 simplExprF env unfolding cont }
1078 ; Nothing -> -- No inlining!
1080 ------------- No inlining! ----------------
1081 -- Next, look for rules or specialisations that match
1083 rebuildCall env (Var var)
1084 (mkArgInfo var n_val_args call_cont) cont
1087 rebuildCall :: SimplEnv
1088 -> OutExpr -- Function
1091 -> SimplM (SimplEnv, OutExpr)
1092 rebuildCall env fun (ArgInfo { ai_strs = [] }) cont
1093 -- When we run out of strictness args, it means
1094 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1095 -- Then we want to discard the entire strict continuation. E.g.
1096 -- * case (error "hello") of { ... }
1097 -- * (error "Hello") arg
1098 -- * f (error "Hello") where f is strict
1100 -- Then, especially in the first of these cases, we'd like to discard
1101 -- the continuation, leaving just the bottoming expression. But the
1102 -- type might not be right, so we may have to add a coerce.
1103 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1104 = return (env, mk_coerce fun) -- contination to discard, else we do it
1105 where -- again and again!
1106 fun_ty = exprType fun
1107 cont_ty = contResultType env fun_ty cont
1108 co = mkUnsafeCoercion fun_ty cont_ty
1109 mk_coerce expr | cont_ty `coreEqType` fun_ty = expr
1110 | otherwise = mkCoerce co expr
1112 rebuildCall env fun info (ApplyTo _ (Type arg_ty) se cont)
1113 = do { ty' <- simplType (se `setInScope` env) arg_ty
1114 ; rebuildCall env (fun `App` Type ty') info cont }
1117 (ArgInfo { ai_rules = has_rules, ai_strs = str:strs, ai_discs = disc:discs })
1118 (ApplyTo _ arg arg_se cont)
1119 | str -- Strict argument
1120 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1121 simplExprF (arg_se `setFloats` env) arg
1122 (StrictArg fun cci arg_info' cont)
1125 | otherwise -- Lazy argument
1126 -- DO NOT float anything outside, hence simplExprC
1127 -- There is no benefit (unlike in a let-binding), and we'd
1128 -- have to be very careful about bogus strictness through
1129 -- floating a demanded let.
1130 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1132 ; rebuildCall env (fun `App` arg') arg_info' cont }
1134 arg_info' = ArgInfo { ai_rules = has_rules, ai_strs = strs, ai_discs = discs }
1135 cci | has_rules || disc > 0 = ArgCtxt has_rules disc -- Be keener here
1136 | otherwise = BoringCtxt -- Nothing interesting
1138 rebuildCall env fun _ cont
1139 = rebuild env fun cont
1144 This part of the simplifier may break the no-shadowing invariant
1146 f (...(\a -> e)...) (case y of (a,b) -> e')
1147 where f is strict in its second arg
1148 If we simplify the innermost one first we get (...(\a -> e)...)
1149 Simplifying the second arg makes us float the case out, so we end up with
1150 case y of (a,b) -> f (...(\a -> e)...) e'
1151 So the output does not have the no-shadowing invariant. However, there is
1152 no danger of getting name-capture, because when the first arg was simplified
1153 we used an in-scope set that at least mentioned all the variables free in its
1154 static environment, and that is enough.
1156 We can't just do innermost first, or we'd end up with a dual problem:
1157 case x of (a,b) -> f e (...(\a -> e')...)
1159 I spent hours trying to recover the no-shadowing invariant, but I just could
1160 not think of an elegant way to do it. The simplifier is already knee-deep in
1161 continuations. We have to keep the right in-scope set around; AND we have
1162 to get the effect that finding (error "foo") in a strict arg position will
1163 discard the entire application and replace it with (error "foo"). Getting
1164 all this at once is TOO HARD!
1166 %************************************************************************
1168 Rebuilding a cse expression
1170 %************************************************************************
1172 Blob of helper functions for the "case-of-something-else" situation.
1175 ---------------------------------------------------------
1176 -- Eliminate the case if possible
1178 rebuildCase :: SimplEnv
1179 -> OutExpr -- Scrutinee
1180 -> InId -- Case binder
1181 -> [InAlt] -- Alternatives (inceasing order)
1183 -> SimplM (SimplEnv, OutExpr)
1185 --------------------------------------------------
1186 -- 1. Eliminate the case if there's a known constructor
1187 --------------------------------------------------
1189 rebuildCase env scrut case_bndr alts cont
1190 | Just (con,args) <- exprIsConApp_maybe scrut
1191 -- Works when the scrutinee is a variable with a known unfolding
1192 -- as well as when it's an explicit constructor application
1193 = knownCon env scrut (DataAlt con) args case_bndr alts cont
1195 | Lit lit <- scrut -- No need for same treatment as constructors
1196 -- because literals are inlined more vigorously
1197 = knownCon env scrut (LitAlt lit) [] case_bndr alts cont
1200 --------------------------------------------------
1201 -- 2. Eliminate the case if scrutinee is evaluated
1202 --------------------------------------------------
1204 rebuildCase env scrut case_bndr [(_, bndrs, rhs)] cont
1205 -- See if we can get rid of the case altogether
1206 -- See the extensive notes on case-elimination above
1207 -- mkCase made sure that if all the alternatives are equal,
1208 -- then there is now only one (DEFAULT) rhs
1209 | all isDeadBinder bndrs -- bndrs are [InId]
1211 -- Check that the scrutinee can be let-bound instead of case-bound
1212 , exprOkForSpeculation scrut
1213 -- OK not to evaluate it
1214 -- This includes things like (==# a# b#)::Bool
1215 -- so that we simplify
1216 -- case ==# a# b# of { True -> x; False -> x }
1219 -- This particular example shows up in default methods for
1220 -- comparision operations (e.g. in (>=) for Int.Int32)
1221 || exprIsHNF scrut -- It's already evaluated
1222 || var_demanded_later scrut -- It'll be demanded later
1224 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1225 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1226 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1227 -- its argument: case x of { y -> dataToTag# y }
1228 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1229 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1231 -- Also we don't want to discard 'seq's
1232 = do { tick (CaseElim case_bndr)
1233 ; env' <- simplNonRecX env case_bndr scrut
1234 ; simplExprF env' rhs cont }
1236 -- The case binder is going to be evaluated later,
1237 -- and the scrutinee is a simple variable
1238 var_demanded_later (Var v) = isStrictDmd (idNewDemandInfo case_bndr)
1239 && not (isTickBoxOp v)
1240 -- ugly hack; covering this case is what
1241 -- exprOkForSpeculation was intended for.
1242 var_demanded_later _ = False
1245 --------------------------------------------------
1246 -- 3. Catch-all case
1247 --------------------------------------------------
1249 rebuildCase env scrut case_bndr alts cont
1250 = do { -- Prepare the continuation;
1251 -- The new subst_env is in place
1252 (env', dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1254 -- Simplify the alternatives
1255 ; (scrut', case_bndr', alts') <- simplAlts env' scrut case_bndr alts dup_cont
1257 -- Check for empty alternatives
1258 ; if null alts' then
1259 -- This isn't strictly an error, although it is unusual.
1260 -- It's possible that the simplifer might "see" that
1261 -- an inner case has no accessible alternatives before
1262 -- it "sees" that the entire branch of an outer case is
1263 -- inaccessible. So we simply put an error case here instead.
1264 pprTrace "mkCase: null alts" (ppr case_bndr <+> ppr scrut) $
1265 let res_ty' = contResultType env' (substTy env' (coreAltsType alts)) dup_cont
1266 lit = mkStringLit "Impossible alternative"
1267 in return (env', mkApps (Var rUNTIME_ERROR_ID) [Type res_ty', lit])
1270 { case_expr <- mkCase scrut' case_bndr' alts'
1272 -- Notice that rebuild gets the in-scope set from env, not alt_env
1273 -- The case binder *not* scope over the whole returned case-expression
1274 ; rebuild env' case_expr nodup_cont } }
1277 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1278 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1279 way, there's a chance that v will now only be used once, and hence
1282 Note [no-case-of-case]
1283 ~~~~~~~~~~~~~~~~~~~~~~
1284 We *used* to suppress the binder-swap in case expressoins when
1285 -fno-case-of-case is on. Old remarks:
1286 "This happens in the first simplifier pass,
1287 and enhances full laziness. Here's the bad case:
1288 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1289 If we eliminate the inner case, we trap it inside the I# v -> arm,
1290 which might prevent some full laziness happening. I've seen this
1291 in action in spectral/cichelli/Prog.hs:
1292 [(m,n) | m <- [1..max], n <- [1..max]]
1293 Hence the check for NoCaseOfCase."
1294 However, now the full-laziness pass itself reverses the binder-swap, so this
1295 check is no longer necessary.
1297 Note [Suppressing the case binder-swap]
1298 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1299 There is another situation when it might make sense to suppress the
1300 case-expression binde-swap. If we have
1302 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1303 ...other cases .... }
1305 We'll perform the binder-swap for the outer case, giving
1307 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1308 ...other cases .... }
1310 But there is no point in doing it for the inner case, because w1 can't
1311 be inlined anyway. Furthermore, doing the case-swapping involves
1312 zapping w2's occurrence info (see paragraphs that follow), and that
1313 forces us to bind w2 when doing case merging. So we get
1315 case x of w1 { A -> let w2 = w1 in e1
1316 B -> let w2 = w1 in e2
1317 ...other cases .... }
1319 This is plain silly in the common case where w2 is dead.
1321 Even so, I can't see a good way to implement this idea. I tried
1322 not doing the binder-swap if the scrutinee was already evaluated
1323 but that failed big-time:
1327 case v of w { MkT x ->
1328 case x of x1 { I# y1 ->
1329 case x of x2 { I# y2 -> ...
1331 Notice that because MkT is strict, x is marked "evaluated". But to
1332 eliminate the last case, we must either make sure that x (as well as
1333 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1334 the binder-swap. So this whole note is a no-op.
1338 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1339 any occurrence info (eg IAmDead) in the case binder, because the
1340 case-binder now effectively occurs whenever v does. AND we have to do
1341 the same for the pattern-bound variables! Example:
1343 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1345 Here, b and p are dead. But when we move the argment inside the first
1346 case RHS, and eliminate the second case, we get
1348 case x of { (a,b) -> a b }
1350 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1353 Indeed, this can happen anytime the case binder isn't dead:
1354 case <any> of x { (a,b) ->
1355 case x of { (p,q) -> p } }
1356 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1357 The point is that we bring into the envt a binding
1359 after the outer case, and that makes (a,b) alive. At least we do unless
1360 the case binder is guaranteed dead.
1364 Consider case (v `cast` co) of x { I# ->
1365 ... (case (v `cast` co) of {...}) ...
1366 We'd like to eliminate the inner case. We can get this neatly by
1367 arranging that inside the outer case we add the unfolding
1368 v |-> x `cast` (sym co)
1369 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1371 Note [Improving seq]
1374 type family F :: * -> *
1375 type instance F Int = Int
1377 ... case e of x { DEFAULT -> rhs } ...
1379 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1381 case e `cast` co of x'::Int
1382 I# x# -> let x = x' `cast` sym co
1385 so that 'rhs' can take advantage of the form of x'. Notice that Note
1386 [Case of cast] may then apply to the result.
1388 This showed up in Roman's experiments. Example:
1389 foo :: F Int -> Int -> Int
1390 foo t n = t `seq` bar n
1393 bar n = bar (n - case t of TI i -> i)
1394 Here we'd like to avoid repeated evaluating t inside the loop, by
1395 taking advantage of the `seq`.
1397 At one point I did transformation in LiberateCase, but it's more robust here.
1398 (Otherwise, there's a danger that we'll simply drop the 'seq' altogether, before
1399 LiberateCase gets to see it.)
1401 Note [Case elimination]
1402 ~~~~~~~~~~~~~~~~~~~~~~~
1403 The case-elimination transformation discards redundant case expressions.
1404 Start with a simple situation:
1406 case x# of ===> e[x#/y#]
1409 (when x#, y# are of primitive type, of course). We can't (in general)
1410 do this for algebraic cases, because we might turn bottom into
1413 The code in SimplUtils.prepareAlts has the effect of generalise this
1414 idea to look for a case where we're scrutinising a variable, and we
1415 know that only the default case can match. For example:
1419 DEFAULT -> ...(case x of
1423 Here the inner case is first trimmed to have only one alternative, the
1424 DEFAULT, after which it's an instance of the previous case. This
1425 really only shows up in eliminating error-checking code.
1427 We also make sure that we deal with this very common case:
1432 Here we are using the case as a strict let; if x is used only once
1433 then we want to inline it. We have to be careful that this doesn't
1434 make the program terminate when it would have diverged before, so we
1436 - e is already evaluated (it may so if e is a variable)
1437 - x is used strictly, or
1439 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1441 case e of ===> case e of DEFAULT -> r
1445 Now again the case may be elminated by the CaseElim transformation.
1448 Further notes about case elimination
1449 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1450 Consider: test :: Integer -> IO ()
1453 Turns out that this compiles to:
1456 eta1 :: State# RealWorld ->
1457 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1459 (PrelNum.jtos eta ($w[] @ Char))
1461 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1463 Notice the strange '<' which has no effect at all. This is a funny one.
1464 It started like this:
1466 f x y = if x < 0 then jtos x
1467 else if y==0 then "" else jtos x
1469 At a particular call site we have (f v 1). So we inline to get
1471 if v < 0 then jtos x
1472 else if 1==0 then "" else jtos x
1474 Now simplify the 1==0 conditional:
1476 if v<0 then jtos v else jtos v
1478 Now common-up the two branches of the case:
1480 case (v<0) of DEFAULT -> jtos v
1482 Why don't we drop the case? Because it's strict in v. It's technically
1483 wrong to drop even unnecessary evaluations, and in practice they
1484 may be a result of 'seq' so we *definitely* don't want to drop those.
1485 I don't really know how to improve this situation.
1489 simplCaseBinder :: SimplEnv -> OutExpr -> OutId -> [InAlt]
1490 -> SimplM (SimplEnv, OutExpr, OutId)
1491 simplCaseBinder env0 scrut0 case_bndr0 alts
1492 = do { (env1, case_bndr1) <- simplBinder env0 case_bndr0
1494 ; fam_envs <- getFamEnvs
1495 ; (env2, scrut2, case_bndr2) <- improve_seq fam_envs env1 scrut0
1496 case_bndr0 case_bndr1 alts
1497 -- Note [Improving seq]
1499 ; let (env3, case_bndr3) = improve_case_bndr env2 scrut2 case_bndr2
1500 -- Note [Case of cast]
1502 ; return (env3, scrut2, case_bndr3) }
1505 improve_seq fam_envs env scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1506 | Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1507 = do { case_bndr2 <- newId (fsLit "nt") ty2
1508 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1509 env2 = extendIdSubst env case_bndr rhs
1510 ; return (env2, scrut `Cast` co, case_bndr2) }
1512 improve_seq _ env scrut _ case_bndr1 _
1513 = return (env, scrut, case_bndr1)
1516 improve_case_bndr env scrut case_bndr
1517 -- See Note [no-case-of-case]
1518 -- | switchIsOn (getSwitchChecker env) NoCaseOfCase
1519 -- = (env, case_bndr)
1521 | otherwise -- Failed try; see Note [Suppressing the case binder-swap]
1522 -- not (isEvaldUnfolding (idUnfolding v))
1524 Var v -> (modifyInScope env1 v case_bndr', case_bndr')
1525 -- Note about using modifyInScope for v here
1526 -- We could extend the substitution instead, but it would be
1527 -- a hack because then the substitution wouldn't be idempotent
1528 -- any more (v is an OutId). And this does just as well.
1530 Cast (Var v) co -> (addBinderUnfolding env1 v rhs, case_bndr')
1532 rhs = Cast (Var case_bndr') (mkSymCoercion co)
1534 _ -> (env, case_bndr)
1536 case_bndr' = zapOccInfo case_bndr
1537 env1 = modifyInScope env case_bndr case_bndr'
1540 zapOccInfo :: InId -> InId -- See Note [zapOccInfo]
1541 zapOccInfo b = b `setIdOccInfo` NoOccInfo
1545 simplAlts does two things:
1547 1. Eliminate alternatives that cannot match, including the
1548 DEFAULT alternative.
1550 2. If the DEFAULT alternative can match only one possible constructor,
1551 then make that constructor explicit.
1553 case e of x { DEFAULT -> rhs }
1555 case e of x { (a,b) -> rhs }
1556 where the type is a single constructor type. This gives better code
1557 when rhs also scrutinises x or e.
1559 Here "cannot match" includes knowledge from GADTs
1561 It's a good idea do do this stuff before simplifying the alternatives, to
1562 avoid simplifying alternatives we know can't happen, and to come up with
1563 the list of constructors that are handled, to put into the IdInfo of the
1564 case binder, for use when simplifying the alternatives.
1566 Eliminating the default alternative in (1) isn't so obvious, but it can
1569 data Colour = Red | Green | Blue
1578 DEFAULT -> [ case y of ... ]
1580 If we inline h into f, the default case of the inlined h can't happen.
1581 If we don't notice this, we may end up filtering out *all* the cases
1582 of the inner case y, which give us nowhere to go!
1586 simplAlts :: SimplEnv
1588 -> InId -- Case binder
1589 -> [InAlt] -- Non-empty
1591 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1592 -- Like simplExpr, this just returns the simplified alternatives;
1593 -- it not return an environment
1595 simplAlts env scrut case_bndr alts cont'
1596 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1597 do { let alt_env = zapFloats env
1598 ; (alt_env', scrut', case_bndr') <- simplCaseBinder alt_env scrut case_bndr alts
1600 ; (imposs_deflt_cons, in_alts) <- prepareAlts alt_env' scrut case_bndr' alts
1602 ; alts' <- mapM (simplAlt alt_env' imposs_deflt_cons case_bndr' cont') in_alts
1603 ; return (scrut', case_bndr', alts') }
1605 ------------------------------------
1606 simplAlt :: SimplEnv
1607 -> [AltCon] -- These constructors can't be present when
1608 -- matching the DEFAULT alternative
1609 -> OutId -- The case binder
1614 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1615 = ASSERT( null bndrs )
1616 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1617 -- Record the constructors that the case-binder *can't* be.
1618 ; rhs' <- simplExprC env' rhs cont'
1619 ; return (DEFAULT, [], rhs') }
1621 simplAlt env _ case_bndr' cont' (LitAlt lit, bndrs, rhs)
1622 = ASSERT( null bndrs )
1623 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1624 ; rhs' <- simplExprC env' rhs cont'
1625 ; return (LitAlt lit, [], rhs') }
1627 simplAlt env _ case_bndr' cont' (DataAlt con, vs, rhs)
1628 = do { -- Deal with the pattern-bound variables
1629 -- Mark the ones that are in ! positions in the
1630 -- data constructor as certainly-evaluated.
1631 -- NB: simplLamBinders preserves this eval info
1632 let vs_with_evals = add_evals (dataConRepStrictness con)
1633 ; (env', vs') <- simplLamBndrs env vs_with_evals
1635 -- Bind the case-binder to (con args)
1636 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1637 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1638 env'' = addBinderUnfolding env' case_bndr'
1639 (mkConApp con con_args)
1641 ; rhs' <- simplExprC env'' rhs cont'
1642 ; return (DataAlt con, vs', rhs') }
1644 -- add_evals records the evaluated-ness of the bound variables of
1645 -- a case pattern. This is *important*. Consider
1646 -- data T = T !Int !Int
1648 -- case x of { T a b -> T (a+1) b }
1650 -- We really must record that b is already evaluated so that we don't
1651 -- go and re-evaluate it when constructing the result.
1652 -- See Note [Data-con worker strictness] in MkId.lhs
1657 go (v:vs') strs | isTyVar v = v : go vs' strs
1658 go (v:vs') (str:strs)
1659 | isMarkedStrict str = evald_v : go vs' strs
1660 | otherwise = zapped_v : go vs' strs
1662 zapped_v = zap_occ_info v
1663 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1664 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr the_strs)
1666 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1668 -- to the envt; so vs are now very much alive
1669 -- Note [Aug06] I can't see why this actually matters, but it's neater
1670 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1671 -- ==> case e of t { (a,b) -> ...(a)... }
1672 -- Look, Ma, a is alive now.
1673 zap_occ_info | isDeadBinder case_bndr' = \ident -> ident
1674 | otherwise = zapOccInfo
1676 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1677 addBinderUnfolding env bndr rhs
1678 = modifyInScope env bndr (bndr `setIdUnfolding` mkUnfolding False rhs)
1680 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1681 addBinderOtherCon env bndr cons
1682 = modifyInScope env bndr (bndr `setIdUnfolding` mkOtherCon cons)
1686 %************************************************************************
1688 \subsection{Known constructor}
1690 %************************************************************************
1692 We are a bit careful with occurrence info. Here's an example
1694 (\x* -> case x of (a*, b) -> f a) (h v, e)
1696 where the * means "occurs once". This effectively becomes
1697 case (h v, e) of (a*, b) -> f a)
1699 let a* = h v; b = e in f a
1703 All this should happen in one sweep.
1706 knownCon :: SimplEnv -> OutExpr -> AltCon
1707 -> [OutExpr] -- Args *including* the universal args
1708 -> InId -> [InAlt] -> SimplCont
1709 -> SimplM (SimplEnv, OutExpr)
1711 knownCon env scrut con args bndr alts cont
1712 = do { tick (KnownBranch bndr)
1713 ; knownAlt env scrut args bndr (findAlt con alts) cont }
1715 knownAlt :: SimplEnv -> OutExpr -> [OutExpr]
1716 -> InId -> (AltCon, [CoreBndr], InExpr) -> SimplCont
1717 -> SimplM (SimplEnv, OutExpr)
1718 knownAlt env scrut _ bndr (DEFAULT, bs, rhs) cont
1720 do { env' <- simplNonRecX env bndr scrut
1721 -- This might give rise to a binding with non-atomic args
1722 -- like x = Node (f x) (g x)
1723 -- but simplNonRecX will atomic-ify it
1724 ; simplExprF env' rhs cont }
1726 knownAlt env scrut _ bndr (LitAlt _, bs, rhs) cont
1728 do { env' <- simplNonRecX env bndr scrut
1729 ; simplExprF env' rhs cont }
1731 knownAlt env scrut the_args bndr (DataAlt dc, bs, rhs) cont
1732 = do { let dead_bndr = isDeadBinder bndr -- bndr is an InId
1733 n_drop_tys = length (dataConUnivTyVars dc)
1734 ; env' <- bind_args env dead_bndr bs (drop n_drop_tys the_args)
1736 -- It's useful to bind bndr to scrut, rather than to a fresh
1737 -- binding x = Con arg1 .. argn
1738 -- because very often the scrut is a variable, so we avoid
1739 -- creating, and then subsequently eliminating, a let-binding
1740 -- BUT, if scrut is a not a variable, we must be careful
1741 -- about duplicating the arg redexes; in that case, make
1742 -- a new con-app from the args
1743 bndr_rhs = case scrut of
1746 con_app = mkConApp dc (take n_drop_tys the_args ++ con_args)
1747 con_args = [substExpr env' (varToCoreExpr b) | b <- bs]
1748 -- args are aready OutExprs, but bs are InIds
1750 ; env'' <- simplNonRecX env' bndr bndr_rhs
1751 ; -- pprTrace "knownCon2" (ppr bs $$ ppr rhs $$ ppr (seIdSubst env'')) $
1752 simplExprF env'' rhs cont }
1755 bind_args env' _ [] _ = return env'
1757 bind_args env' dead_bndr (b:bs') (Type ty : args)
1758 = ASSERT( isTyVar b )
1759 bind_args (extendTvSubst env' b ty) dead_bndr bs' args
1761 bind_args env' dead_bndr (b:bs') (arg : args)
1763 do { let b' = if dead_bndr then b else zapOccInfo b
1764 -- Note that the binder might be "dead", because it doesn't
1765 -- occur in the RHS; and simplNonRecX may therefore discard
1766 -- it via postInlineUnconditionally.
1767 -- Nevertheless we must keep it if the case-binder is alive,
1768 -- because it may be used in the con_app. See Note [zapOccInfo]
1769 ; env'' <- simplNonRecX env' b' arg
1770 ; bind_args env'' dead_bndr bs' args }
1773 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr the_args $$
1774 text "scrut:" <+> ppr scrut
1778 %************************************************************************
1780 \subsection{Duplicating continuations}
1782 %************************************************************************
1785 prepareCaseCont :: SimplEnv
1786 -> [InAlt] -> SimplCont
1787 -> SimplM (SimplEnv, SimplCont,SimplCont)
1788 -- Return a duplicatable continuation, a non-duplicable part
1789 -- plus some extra bindings (that scope over the entire
1792 -- No need to make it duplicatable if there's only one alternative
1793 prepareCaseCont env [_] cont = return (env, cont, mkBoringStop)
1794 prepareCaseCont env _ cont = mkDupableCont env cont
1798 mkDupableCont :: SimplEnv -> SimplCont
1799 -> SimplM (SimplEnv, SimplCont, SimplCont)
1801 mkDupableCont env cont
1802 | contIsDupable cont
1803 = return (env, cont, mkBoringStop)
1805 mkDupableCont _ (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1807 mkDupableCont env (CoerceIt ty cont)
1808 = do { (env', dup, nodup) <- mkDupableCont env cont
1809 ; return (env', CoerceIt ty dup, nodup) }
1811 mkDupableCont env cont@(StrictBind {})
1812 = return (env, mkBoringStop, cont)
1813 -- See Note [Duplicating strict continuations]
1815 mkDupableCont env cont@(StrictArg {})
1816 = return (env, mkBoringStop, cont)
1817 -- See Note [Duplicating strict continuations]
1819 mkDupableCont env (ApplyTo _ arg se cont)
1820 = -- e.g. [...hole...] (...arg...)
1822 -- let a = ...arg...
1823 -- in [...hole...] a
1824 do { (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1825 ; arg' <- simplExpr (se `setInScope` env') arg
1826 ; (env'', arg'') <- makeTrivial env' arg'
1827 ; let app_cont = ApplyTo OkToDup arg'' (zapSubstEnv env') dup_cont
1828 ; return (env'', app_cont, nodup_cont) }
1830 mkDupableCont env cont@(Select _ case_bndr [(_, bs, _rhs)] _ _)
1831 -- See Note [Single-alternative case]
1832 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1833 -- | not (isDeadBinder case_bndr)
1834 | all isDeadBinder bs -- InIds
1835 && not (isUnLiftedType (idType case_bndr))
1836 -- Note [Single-alternative-unlifted]
1837 = return (env, mkBoringStop, cont)
1839 mkDupableCont env (Select _ case_bndr alts se cont)
1840 = -- e.g. (case [...hole...] of { pi -> ei })
1842 -- let ji = \xij -> ei
1843 -- in case [...hole...] of { pi -> ji xij }
1844 do { tick (CaseOfCase case_bndr)
1845 ; (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1846 -- NB: call mkDupableCont here, *not* prepareCaseCont
1847 -- We must make a duplicable continuation, whereas prepareCaseCont
1848 -- doesn't when there is a single case branch
1850 ; let alt_env = se `setInScope` env'
1851 ; (alt_env', case_bndr') <- simplBinder alt_env case_bndr
1852 ; alts' <- mapM (simplAlt alt_env' [] case_bndr' dup_cont) alts
1853 -- Safe to say that there are no handled-cons for the DEFAULT case
1854 -- NB: simplBinder does not zap deadness occ-info, so
1855 -- a dead case_bndr' will still advertise its deadness
1856 -- This is really important because in
1857 -- case e of b { (# p,q #) -> ... }
1858 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1859 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1860 -- In the new alts we build, we have the new case binder, so it must retain
1862 -- NB: we don't use alt_env further; it has the substEnv for
1863 -- the alternatives, and we don't want that
1865 ; (env'', alts'') <- mkDupableAlts env' case_bndr' alts'
1866 ; return (env'', -- Note [Duplicated env]
1867 Select OkToDup case_bndr' alts'' (zapSubstEnv env'') mkBoringStop,
1871 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1872 -> SimplM (SimplEnv, [InAlt])
1873 -- Absorbs the continuation into the new alternatives
1875 mkDupableAlts env case_bndr' the_alts
1878 go env0 [] = return (env0, [])
1880 = do { (env1, alt') <- mkDupableAlt env0 case_bndr' alt
1881 ; (env2, alts') <- go env1 alts
1882 ; return (env2, alt' : alts' ) }
1884 mkDupableAlt :: SimplEnv -> OutId -> (AltCon, [CoreBndr], CoreExpr)
1885 -> SimplM (SimplEnv, (AltCon, [CoreBndr], CoreExpr))
1886 mkDupableAlt env case_bndr' (con, bndrs', rhs')
1887 | exprIsDupable rhs' -- Note [Small alternative rhs]
1888 = return (env, (con, bndrs', rhs'))
1890 = do { let rhs_ty' = exprType rhs'
1891 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1893 | isTyVar bndr = True -- Abstract over all type variables just in case
1894 | otherwise = not (isDeadBinder bndr)
1895 -- The deadness info on the new Ids is preserved by simplBinders
1897 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1898 <- if (any isId used_bndrs')
1899 then return (used_bndrs', varsToCoreExprs used_bndrs')
1900 else do { rw_id <- newId (fsLit "w") realWorldStatePrimTy
1901 ; return ([rw_id], [Var realWorldPrimId]) }
1903 ; join_bndr <- newId (fsLit "$j") (mkPiTypes final_bndrs' rhs_ty')
1904 -- Note [Funky mkPiTypes]
1906 ; let -- We make the lambdas into one-shot-lambdas. The
1907 -- join point is sure to be applied at most once, and doing so
1908 -- prevents the body of the join point being floated out by
1909 -- the full laziness pass
1910 really_final_bndrs = map one_shot final_bndrs'
1911 one_shot v | isId v = setOneShotLambda v
1913 join_rhs = mkLams really_final_bndrs rhs'
1914 join_call = mkApps (Var join_bndr) final_args
1916 ; return (addPolyBind NotTopLevel env (NonRec join_bndr join_rhs), (con, bndrs', join_call)) }
1917 -- See Note [Duplicated env]
1920 Note [Duplicated env]
1921 ~~~~~~~~~~~~~~~~~~~~~
1922 Some of the alternatives are simplified, but have not been turned into a join point
1923 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
1924 bind the join point, because it might to do PostInlineUnconditionally, and
1925 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
1926 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
1927 at worst delays the join-point inlining.
1929 Note [Small alterantive rhs]
1930 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1931 It is worth checking for a small RHS because otherwise we
1932 get extra let bindings that may cause an extra iteration of the simplifier to
1933 inline back in place. Quite often the rhs is just a variable or constructor.
1934 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1935 iterations because the version with the let bindings looked big, and so wasn't
1936 inlined, but after the join points had been inlined it looked smaller, and so
1939 NB: we have to check the size of rhs', not rhs.
1940 Duplicating a small InAlt might invalidate occurrence information
1941 However, if it *is* dupable, we return the *un* simplified alternative,
1942 because otherwise we'd need to pair it up with an empty subst-env....
1943 but we only have one env shared between all the alts.
1944 (Remember we must zap the subst-env before re-simplifying something).
1945 Rather than do this we simply agree to re-simplify the original (small) thing later.
1947 Note [Funky mkPiTypes]
1948 ~~~~~~~~~~~~~~~~~~~~~~
1949 Notice the funky mkPiTypes. If the contructor has existentials
1950 it's possible that the join point will be abstracted over
1951 type varaibles as well as term variables.
1952 Example: Suppose we have
1953 data T = forall t. C [t]
1955 case (case e of ...) of
1957 We get the join point
1958 let j :: forall t. [t] -> ...
1959 j = /\t \xs::[t] -> rhs
1961 case (case e of ...) of
1962 C t xs::[t] -> j t xs
1964 Note [Join point abstaction]
1965 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1966 If we try to lift a primitive-typed something out
1967 for let-binding-purposes, we will *caseify* it (!),
1968 with potentially-disastrous strictness results. So
1969 instead we turn it into a function: \v -> e
1970 where v::State# RealWorld#. The value passed to this function
1971 is realworld#, which generates (almost) no code.
1973 There's a slight infelicity here: we pass the overall
1974 case_bndr to all the join points if it's used in *any* RHS,
1975 because we don't know its usage in each RHS separately
1977 We used to say "&& isUnLiftedType rhs_ty'" here, but now
1978 we make the join point into a function whenever used_bndrs'
1979 is empty. This makes the join-point more CPR friendly.
1980 Consider: let j = if .. then I# 3 else I# 4
1981 in case .. of { A -> j; B -> j; C -> ... }
1983 Now CPR doesn't w/w j because it's a thunk, so
1984 that means that the enclosing function can't w/w either,
1985 which is a lose. Here's the example that happened in practice:
1986 kgmod :: Int -> Int -> Int
1987 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1991 I have seen a case alternative like this:
1993 It's a bit silly to add the realWorld dummy arg in this case, making
1996 (the \v alone is enough to make CPR happy) but I think it's rare
1998 Note [Duplicating strict continuations]
1999 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2000 Do *not* duplicate StrictBind and StritArg continuations. We gain
2001 nothing by propagating them into the expressions, and we do lose a
2002 lot. Here's an example:
2003 && (case x of { T -> F; F -> T }) E
2004 Now, && is strict so we end up simplifying the case with
2005 an ArgOf continuation. If we let-bind it, we get
2007 let $j = \v -> && v E
2008 in simplExpr (case x of { T -> F; F -> T })
2010 And after simplifying more we get
2012 let $j = \v -> && v E
2013 in case x of { T -> $j F; F -> $j T }
2014 Which is a Very Bad Thing
2016 The desire not to duplicate is the entire reason that
2017 mkDupableCont returns a pair of continuations.
2019 The original plan had:
2020 e.g. (...strict-fn...) [...hole...]
2022 let $j = \a -> ...strict-fn...
2025 Note [Single-alternative cases]
2026 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2027 This case is just like the ArgOf case. Here's an example:
2031 case (case x of I# x' ->
2033 True -> I# (negate# x')
2034 False -> I# x') of y {
2036 Because the (case x) has only one alternative, we'll transform to
2038 case (case x' <# 0# of
2039 True -> I# (negate# x')
2040 False -> I# x') of y {
2042 But now we do *NOT* want to make a join point etc, giving
2044 let $j = \y -> MkT y
2046 True -> $j (I# (negate# x'))
2048 In this case the $j will inline again, but suppose there was a big
2049 strict computation enclosing the orginal call to MkT. Then, it won't
2050 "see" the MkT any more, because it's big and won't get duplicated.
2051 And, what is worse, nothing was gained by the case-of-case transform.
2053 When should use this case of mkDupableCont?
2054 However, matching on *any* single-alternative case is a *disaster*;
2055 e.g. case (case ....) of (a,b) -> (# a,b #)
2056 We must push the outer case into the inner one!
2059 * Match [(DEFAULT,_,_)], but in the common case of Int,
2060 the alternative-filling-in code turned the outer case into
2061 case (...) of y { I# _ -> MkT y }
2063 * Match on single alternative plus (not (isDeadBinder case_bndr))
2064 Rationale: pushing the case inwards won't eliminate the construction.
2065 But there's a risk of
2066 case (...) of y { (a,b) -> let z=(a,b) in ... }
2067 Now y looks dead, but it'll come alive again. Still, this
2068 seems like the best option at the moment.
2070 * Match on single alternative plus (all (isDeadBinder bndrs))
2071 Rationale: this is essentially seq.
2073 * Match when the rhs is *not* duplicable, and hence would lead to a
2074 join point. This catches the disaster-case above. We can test
2075 the *un-simplified* rhs, which is fine. It might get bigger or
2076 smaller after simplification; if it gets smaller, this case might
2077 fire next time round. NB also that we must test contIsDupable
2078 case_cont *btoo, because case_cont might be big!
2080 HOWEVER: I found that this version doesn't work well, because
2081 we can get let x = case (...) of { small } in ...case x...
2082 When x is inlined into its full context, we find that it was a bad
2083 idea to have pushed the outer case inside the (...) case.
2085 Note [Single-alternative-unlifted]
2086 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2087 Here's another single-alternative where we really want to do case-of-case:
2095 case y_s6X of tpl_s7m {
2096 M1.Mk1 ipv_s70 -> ipv_s70;
2097 M1.Mk2 ipv_s72 -> ipv_s72;
2103 case x_s74 of tpl_s7n {
2104 M1.Mk1 ipv_s77 -> ipv_s77;
2105 M1.Mk2 ipv_s79 -> ipv_s79;
2109 { __DEFAULT -> ==# [wild1_s7b wild_s7c];
2113 So the outer case is doing *nothing at all*, other than serving as a
2114 join-point. In this case we really want to do case-of-case and decide
2115 whether to use a real join point or just duplicate the continuation.
2117 Hence: check whether the case binder's type is unlifted, because then
2118 the outer case is *not* a seq.