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 = final_id `seq` -- This seq forces the Id, and hence its IdInfo,
609 -- and hence any inner substitutions
610 addNonRec env final_id rhs
611 -- The addNonRec adds it to the in-scope set too
614 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity rhs
617 -- Add the unfolding *only* for non-loop-breakers
618 -- Making loop breakers not have an unfolding at all
619 -- means that we can avoid tests in exprIsConApp, for example.
620 -- This is important: if exprIsConApp says 'yes' for a recursive
621 -- thing, then we can get into an infinite loop
624 -- If the unfolding is a value, the demand info may
625 -- go pear-shaped, so we nuke it. Example:
627 -- case x of (p,q) -> h p q x
628 -- Here x is certainly demanded. But after we've nuked
629 -- the case, we'll get just
630 -- let x = (a,b) in h a b x
631 -- and now x is not demanded (I'm assuming h is lazy)
632 -- This really happens. Similarly
633 -- let f = \x -> e in ...f..f...
634 -- After inlining f at some of its call sites the original binding may
635 -- (for example) be no longer strictly demanded.
636 -- The solution here is a bit ad hoc...
637 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
640 final_info | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
641 | otherwise = info_w_unf
643 final_id = new_bndr `setIdInfo` final_info
648 %************************************************************************
650 \subsection[Simplify-simplExpr]{The main function: simplExpr}
652 %************************************************************************
654 The reason for this OutExprStuff stuff is that we want to float *after*
655 simplifying a RHS, not before. If we do so naively we get quadratic
656 behaviour as things float out.
658 To see why it's important to do it after, consider this (real) example:
672 a -- Can't inline a this round, cos it appears twice
676 Each of the ==> steps is a round of simplification. We'd save a
677 whole round if we float first. This can cascade. Consider
682 let f = let d1 = ..d.. in \y -> e
686 in \x -> ...(\y ->e)...
688 Only in this second round can the \y be applied, and it
689 might do the same again.
693 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
694 simplExpr env expr = simplExprC env expr mkBoringStop
696 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
697 -- Simplify an expression, given a continuation
698 simplExprC env expr cont
699 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
700 do { (env', expr') <- simplExprF (zapFloats env) expr cont
701 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
702 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
703 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
704 return (wrapFloats env' expr') }
706 --------------------------------------------------
707 simplExprF :: SimplEnv -> InExpr -> SimplCont
708 -> SimplM (SimplEnv, OutExpr)
710 simplExprF env e cont
711 = -- pprTrace "simplExprF" (ppr e $$ ppr cont $$ ppr (seTvSubst env) $$ ppr (seIdSubst env) {- $$ ppr (seFloats env) -} ) $
712 simplExprF' env e cont
714 simplExprF' :: SimplEnv -> InExpr -> SimplCont
715 -> SimplM (SimplEnv, OutExpr)
716 simplExprF' env (Var v) cont = simplVar env v cont
717 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
718 simplExprF' env (Note n expr) cont = simplNote env n expr cont
719 simplExprF' env (Cast body co) cont = simplCast env body co cont
720 simplExprF' env (App fun arg) cont = simplExprF env fun $
721 ApplyTo NoDup arg env cont
723 simplExprF' env expr@(Lam _ _) cont
724 = simplLam env (map zap bndrs) body cont
725 -- The main issue here is under-saturated lambdas
726 -- (\x1. \x2. e) arg1
727 -- Here x1 might have "occurs-once" occ-info, because occ-info
728 -- is computed assuming that a group of lambdas is applied
729 -- all at once. If there are too few args, we must zap the
732 n_args = countArgs cont
733 n_params = length bndrs
734 (bndrs, body) = collectBinders expr
735 zap | n_args >= n_params = \b -> b
736 | otherwise = \b -> if isTyVar b then b
738 -- NB: we count all the args incl type args
739 -- so we must count all the binders (incl type lambdas)
741 simplExprF' env (Type ty) cont
742 = ASSERT( contIsRhsOrArg cont )
743 do { ty' <- simplType env ty
744 ; rebuild env (Type ty') cont }
746 simplExprF' env (Case scrut bndr _ alts) cont
747 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
748 = -- Simplify the scrutinee with a Select continuation
749 simplExprF env scrut (Select NoDup bndr alts env cont)
752 = -- If case-of-case is off, simply simplify the case expression
753 -- in a vanilla Stop context, and rebuild the result around it
754 do { case_expr' <- simplExprC env scrut case_cont
755 ; rebuild env case_expr' cont }
757 case_cont = Select NoDup bndr alts env mkBoringStop
759 simplExprF' env (Let (Rec pairs) body) cont
760 = do { env' <- simplRecBndrs env (map fst pairs)
761 -- NB: bndrs' don't have unfoldings or rules
762 -- We add them as we go down
764 ; env'' <- simplRecBind env' NotTopLevel pairs
765 ; simplExprF env'' body cont }
767 simplExprF' env (Let (NonRec bndr rhs) body) cont
768 = simplNonRecE env bndr (rhs, env) ([], body) cont
770 ---------------------------------
771 simplType :: SimplEnv -> InType -> SimplM OutType
772 -- Kept monadic just so we can do the seqType
774 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
775 seqType new_ty `seq` return new_ty
777 new_ty = substTy env ty
781 %************************************************************************
783 \subsection{The main rebuilder}
785 %************************************************************************
788 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
789 -- At this point the substitution in the SimplEnv should be irrelevant
790 -- only the in-scope set and floats should matter
791 rebuild env expr cont0
792 = -- pprTrace "rebuild" (ppr expr $$ ppr cont0 $$ ppr (seFloats env)) $
794 Stop {} -> return (env, expr)
795 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
796 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
797 StrictArg fun _ info cont -> rebuildCall env (fun `App` expr) info cont
798 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
799 ; simplLam env' bs body cont }
800 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
801 ; rebuild env (App expr arg') cont }
805 %************************************************************************
809 %************************************************************************
812 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
813 -> SimplM (SimplEnv, OutExpr)
814 simplCast env body co0 cont0
815 = do { co1 <- simplType env co0
816 ; simplExprF env body (addCoerce co1 cont0) }
818 addCoerce co cont = add_coerce co (coercionKind co) cont
820 add_coerce _co (s1, k1) cont -- co :: ty~ty
821 | s1 `coreEqType` k1 = cont -- is a no-op
823 add_coerce co1 (s1, _k2) (CoerceIt co2 cont)
824 | (_l1, t1) <- coercionKind co2
825 -- coerce T1 S1 (coerce S1 K1 e)
828 -- coerce T1 K1 e, otherwise
830 -- For example, in the initial form of a worker
831 -- we may find (coerce T (coerce S (\x.e))) y
832 -- and we'd like it to simplify to e[y/x] in one round
834 , s1 `coreEqType` t1 = cont -- The coerces cancel out
835 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
837 add_coerce co (s1s2, _t1t2) (ApplyTo dup (Type arg_ty) arg_se cont)
838 -- (f `cast` g) ty ---> (f ty) `cast` (g @ ty)
839 -- This implements the PushT rule from the paper
840 | Just (tyvar,_) <- splitForAllTy_maybe s1s2
841 , not (isCoVar tyvar)
842 = ApplyTo dup (Type ty') (zapSubstEnv env) (addCoerce (mkInstCoercion co ty') cont)
844 ty' = substTy (arg_se `setInScope` env) arg_ty
846 -- ToDo: the PushC rule is not implemented at all
848 add_coerce co (s1s2, _t1t2) (ApplyTo dup arg arg_se cont)
849 | not (isTypeArg arg) -- This implements the Push rule from the paper
850 , isFunTy s1s2 -- t1t2 must be a function type, becuase it's applied
851 -- co : s1s2 :=: t1t2
852 -- (coerce (T1->T2) (S1->S2) F) E
854 -- coerce T2 S2 (F (coerce S1 T1 E))
856 -- t1t2 must be a function type, T1->T2, because it's applied
857 -- to something but s1s2 might conceivably not be
859 -- When we build the ApplyTo we can't mix the out-types
860 -- with the InExpr in the argument, so we simply substitute
861 -- to make it all consistent. It's a bit messy.
862 -- But it isn't a common case.
864 -- Example of use: Trac #995
865 = ApplyTo dup new_arg (zapSubstEnv env) (addCoerce co2 cont)
867 -- we split coercion t1->t2 :=: s1->s2 into t1 :=: s1 and
868 -- t2 :=: s2 with left and right on the curried form:
869 -- (->) t1 t2 :=: (->) s1 s2
870 [co1, co2] = decomposeCo 2 co
871 new_arg = mkCoerce (mkSymCoercion co1) arg'
872 arg' = substExpr (arg_se `setInScope` env) arg
874 add_coerce co _ cont = CoerceIt co cont
878 %************************************************************************
882 %************************************************************************
885 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
886 -> SimplM (SimplEnv, OutExpr)
888 simplLam env [] body cont = simplExprF env body cont
891 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
892 = do { tick (BetaReduction bndr)
893 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
895 -- Not enough args, so there are real lambdas left to put in the result
896 simplLam env bndrs body cont
897 = do { (env', bndrs') <- simplLamBndrs env bndrs
898 ; body' <- simplExpr env' body
899 ; new_lam <- mkLam bndrs' body'
900 ; rebuild env' new_lam cont }
903 simplNonRecE :: SimplEnv
904 -> InId -- The binder
905 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
906 -> ([InBndr], InExpr) -- Body of the let/lambda
909 -> SimplM (SimplEnv, OutExpr)
911 -- simplNonRecE is used for
912 -- * non-top-level non-recursive lets in expressions
915 -- It deals with strict bindings, via the StrictBind continuation,
916 -- which may abort the whole process
918 -- The "body" of the binding comes as a pair of ([InId],InExpr)
919 -- representing a lambda; so we recurse back to simplLam
920 -- Why? Because of the binder-occ-info-zapping done before
921 -- the call to simplLam in simplExprF (Lam ...)
923 -- First deal with type applications and type lets
924 -- (/\a. e) (Type ty) and (let a = Type ty in e)
925 simplNonRecE env bndr (Type ty_arg, rhs_se) (bndrs, body) cont
926 = ASSERT( isTyVar bndr )
927 do { ty_arg' <- simplType (rhs_se `setInScope` env) ty_arg
928 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
930 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
931 | preInlineUnconditionally env NotTopLevel bndr rhs
932 = do { tick (PreInlineUnconditionally bndr)
933 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
936 = do { simplExprF (rhs_se `setFloats` env) rhs
937 (StrictBind bndr bndrs body env cont) }
940 = do { (env1, bndr1) <- simplNonRecBndr env bndr
941 ; let (env2, bndr2) = addBndrRules env1 bndr bndr1
942 ; env3 <- simplLazyBind env2 NotTopLevel NonRecursive bndr bndr2 rhs rhs_se
943 ; simplLam env3 bndrs body cont }
947 %************************************************************************
951 %************************************************************************
954 -- Hack alert: we only distinguish subsumed cost centre stacks for the
955 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
956 simplNote :: SimplEnv -> Note -> CoreExpr -> SimplCont
957 -> SimplM (SimplEnv, OutExpr)
958 simplNote env (SCC cc) e cont
959 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
960 ; rebuild env (mkSCC cc e') cont }
962 -- See notes with SimplMonad.inlineMode
963 simplNote env InlineMe e cont
964 | Just (inside, outside) <- splitInlineCont cont -- Boring boring continuation; see notes above
965 = do { -- Don't inline inside an INLINE expression
966 e' <- simplExprC (setMode inlineMode env) e inside
967 ; rebuild env (mkInlineMe e') outside }
969 | otherwise -- Dissolve the InlineMe note if there's
970 -- an interesting context of any kind to combine with
971 -- (even a type application -- anything except Stop)
972 = simplExprF env e cont
974 simplNote env (CoreNote s) e cont = do
975 e' <- simplExpr env e
976 rebuild env (Note (CoreNote s) e') cont
980 %************************************************************************
982 \subsection{Dealing with calls}
984 %************************************************************************
987 simplVar :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
988 simplVar env var cont
989 = case substId env var of
990 DoneEx e -> simplExprF (zapSubstEnv env) e cont
991 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
992 DoneId var1 -> completeCall (zapSubstEnv env) var1 cont
993 -- Note [zapSubstEnv]
994 -- The template is already simplified, so don't re-substitute.
995 -- This is VITAL. Consider
997 -- let y = \z -> ...x... in
999 -- We'll clone the inner \x, adding x->x' in the id_subst
1000 -- Then when we inline y, we must *not* replace x by x' in
1001 -- the inlined copy!!
1003 ---------------------------------------------------------
1004 -- Dealing with a call site
1006 completeCall :: SimplEnv -> Id -> SimplCont -> SimplM (SimplEnv, OutExpr)
1007 completeCall env var cont
1008 = do { dflags <- getDOptsSmpl
1009 ; let (args,call_cont) = contArgs cont
1010 -- The args are OutExprs, obtained by *lazily* substituting
1011 -- in the args found in cont. These args are only examined
1012 -- to limited depth (unless a rule fires). But we must do
1013 -- the substitution; rule matching on un-simplified args would
1016 ------------- First try rules ----------------
1017 -- Do this before trying inlining. Some functions have
1018 -- rules *and* are strict; in this case, we don't want to
1019 -- inline the wrapper of the non-specialised thing; better
1020 -- to call the specialised thing instead.
1022 -- We used to use the black-listing mechanism to ensure that inlining of
1023 -- the wrapper didn't occur for things that have specialisations till a
1024 -- later phase, so but now we just try RULES first
1026 -- Note [Rules for recursive functions]
1027 -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1028 -- You might think that we shouldn't apply rules for a loop breaker:
1029 -- doing so might give rise to an infinite loop, because a RULE is
1030 -- rather like an extra equation for the function:
1031 -- RULE: f (g x) y = x+y
1034 -- But it's too drastic to disable rules for loop breakers.
1035 -- Even the foldr/build rule would be disabled, because foldr
1036 -- is recursive, and hence a loop breaker:
1037 -- foldr k z (build g) = g k z
1038 -- So it's up to the programmer: rules can cause divergence
1039 ; rule_base <- getSimplRules
1040 ; let in_scope = getInScope env
1041 rules = getRules rule_base var
1042 maybe_rule = case activeRule dflags env of
1043 Nothing -> Nothing -- No rules apply
1044 Just act_fn -> lookupRule act_fn in_scope
1046 ; case maybe_rule of {
1047 Just (rule, rule_rhs) -> do
1048 tick (RuleFired (ru_name rule))
1049 (if dopt Opt_D_dump_rule_firings dflags then
1050 pprTrace "Rule fired" (vcat [
1051 text "Rule:" <+> ftext (ru_name rule),
1052 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1053 text "After: " <+> pprCoreExpr rule_rhs,
1054 text "Cont: " <+> ppr call_cont])
1057 simplExprF env rule_rhs (dropArgs (ruleArity rule) cont)
1058 -- The ruleArity says how many args the rule consumed
1060 ; Nothing -> do -- No rules
1062 ------------- Next try inlining ----------------
1063 { let arg_infos = [interestingArg arg | arg <- args, isValArg arg]
1064 n_val_args = length arg_infos
1065 interesting_cont = interestingCallContext call_cont
1066 active_inline = activeInline env var
1067 maybe_inline = callSiteInline dflags active_inline var
1068 (null args) arg_infos interesting_cont
1069 ; case maybe_inline of {
1070 Just unfolding -- There is an inlining!
1071 -> do { tick (UnfoldingDone var)
1072 ; (if dopt Opt_D_dump_inlinings dflags then
1073 pprTrace ("Inlining done" ++ showSDoc (ppr var)) (vcat [
1074 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1075 text "Inlined fn: " <+> nest 2 (ppr unfolding),
1076 text "Cont: " <+> ppr call_cont])
1079 simplExprF env unfolding cont }
1081 ; Nothing -> -- No inlining!
1083 ------------- No inlining! ----------------
1084 -- Next, look for rules or specialisations that match
1086 rebuildCall env (Var var)
1087 (mkArgInfo var n_val_args call_cont) cont
1090 rebuildCall :: SimplEnv
1091 -> OutExpr -- Function
1094 -> SimplM (SimplEnv, OutExpr)
1095 rebuildCall env fun (ArgInfo { ai_strs = [] }) cont
1096 -- When we run out of strictness args, it means
1097 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1098 -- Then we want to discard the entire strict continuation. E.g.
1099 -- * case (error "hello") of { ... }
1100 -- * (error "Hello") arg
1101 -- * f (error "Hello") where f is strict
1103 -- Then, especially in the first of these cases, we'd like to discard
1104 -- the continuation, leaving just the bottoming expression. But the
1105 -- type might not be right, so we may have to add a coerce.
1106 | not (contIsTrivial cont) -- Only do this if there is a non-trivial
1107 = return (env, mk_coerce fun) -- contination to discard, else we do it
1108 where -- again and again!
1109 fun_ty = exprType fun
1110 cont_ty = contResultType env fun_ty cont
1111 co = mkUnsafeCoercion fun_ty cont_ty
1112 mk_coerce expr | cont_ty `coreEqType` fun_ty = expr
1113 | otherwise = mkCoerce co expr
1115 rebuildCall env fun info (ApplyTo _ (Type arg_ty) se cont)
1116 = do { ty' <- simplType (se `setInScope` env) arg_ty
1117 ; rebuildCall env (fun `App` Type ty') info cont }
1120 (ArgInfo { ai_rules = has_rules, ai_strs = str:strs, ai_discs = disc:discs })
1121 (ApplyTo _ arg arg_se cont)
1122 | str -- Strict argument
1123 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1124 simplExprF (arg_se `setFloats` env) arg
1125 (StrictArg fun cci arg_info' cont)
1128 | otherwise -- Lazy argument
1129 -- DO NOT float anything outside, hence simplExprC
1130 -- There is no benefit (unlike in a let-binding), and we'd
1131 -- have to be very careful about bogus strictness through
1132 -- floating a demanded let.
1133 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1135 ; rebuildCall env (fun `App` arg') arg_info' cont }
1137 arg_info' = ArgInfo { ai_rules = has_rules, ai_strs = strs, ai_discs = discs }
1138 cci | has_rules || disc > 0 = ArgCtxt has_rules disc -- Be keener here
1139 | otherwise = BoringCtxt -- Nothing interesting
1141 rebuildCall env fun _ cont
1142 = rebuild env fun cont
1147 This part of the simplifier may break the no-shadowing invariant
1149 f (...(\a -> e)...) (case y of (a,b) -> e')
1150 where f is strict in its second arg
1151 If we simplify the innermost one first we get (...(\a -> e)...)
1152 Simplifying the second arg makes us float the case out, so we end up with
1153 case y of (a,b) -> f (...(\a -> e)...) e'
1154 So the output does not have the no-shadowing invariant. However, there is
1155 no danger of getting name-capture, because when the first arg was simplified
1156 we used an in-scope set that at least mentioned all the variables free in its
1157 static environment, and that is enough.
1159 We can't just do innermost first, or we'd end up with a dual problem:
1160 case x of (a,b) -> f e (...(\a -> e')...)
1162 I spent hours trying to recover the no-shadowing invariant, but I just could
1163 not think of an elegant way to do it. The simplifier is already knee-deep in
1164 continuations. We have to keep the right in-scope set around; AND we have
1165 to get the effect that finding (error "foo") in a strict arg position will
1166 discard the entire application and replace it with (error "foo"). Getting
1167 all this at once is TOO HARD!
1169 %************************************************************************
1171 Rebuilding a cse expression
1173 %************************************************************************
1175 Blob of helper functions for the "case-of-something-else" situation.
1178 ---------------------------------------------------------
1179 -- Eliminate the case if possible
1181 rebuildCase :: SimplEnv
1182 -> OutExpr -- Scrutinee
1183 -> InId -- Case binder
1184 -> [InAlt] -- Alternatives (inceasing order)
1186 -> SimplM (SimplEnv, OutExpr)
1188 --------------------------------------------------
1189 -- 1. Eliminate the case if there's a known constructor
1190 --------------------------------------------------
1192 rebuildCase env scrut case_bndr alts cont
1193 | Just (con,args) <- exprIsConApp_maybe scrut
1194 -- Works when the scrutinee is a variable with a known unfolding
1195 -- as well as when it's an explicit constructor application
1196 = knownCon env scrut (DataAlt con) args case_bndr alts cont
1198 | Lit lit <- scrut -- No need for same treatment as constructors
1199 -- because literals are inlined more vigorously
1200 = knownCon env scrut (LitAlt lit) [] case_bndr alts cont
1203 --------------------------------------------------
1204 -- 2. Eliminate the case if scrutinee is evaluated
1205 --------------------------------------------------
1207 rebuildCase env scrut case_bndr [(_, bndrs, rhs)] cont
1208 -- See if we can get rid of the case altogether
1209 -- See the extensive notes on case-elimination above
1210 -- mkCase made sure that if all the alternatives are equal,
1211 -- then there is now only one (DEFAULT) rhs
1212 | all isDeadBinder bndrs -- bndrs are [InId]
1214 -- Check that the scrutinee can be let-bound instead of case-bound
1215 , exprOkForSpeculation scrut
1216 -- OK not to evaluate it
1217 -- This includes things like (==# a# b#)::Bool
1218 -- so that we simplify
1219 -- case ==# a# b# of { True -> x; False -> x }
1222 -- This particular example shows up in default methods for
1223 -- comparision operations (e.g. in (>=) for Int.Int32)
1224 || exprIsHNF scrut -- It's already evaluated
1225 || var_demanded_later scrut -- It'll be demanded later
1227 -- || not opt_SimplPedanticBottoms) -- Or we don't care!
1228 -- We used to allow improving termination by discarding cases, unless -fpedantic-bottoms was on,
1229 -- but that breaks badly for the dataToTag# primop, which relies on a case to evaluate
1230 -- its argument: case x of { y -> dataToTag# y }
1231 -- Here we must *not* discard the case, because dataToTag# just fetches the tag from
1232 -- the info pointer. So we'll be pedantic all the time, and see if that gives any
1234 -- Also we don't want to discard 'seq's
1235 = do { tick (CaseElim case_bndr)
1236 ; env' <- simplNonRecX env case_bndr scrut
1237 ; simplExprF env' rhs cont }
1239 -- The case binder is going to be evaluated later,
1240 -- and the scrutinee is a simple variable
1241 var_demanded_later (Var v) = isStrictDmd (idNewDemandInfo case_bndr)
1242 && not (isTickBoxOp v)
1243 -- ugly hack; covering this case is what
1244 -- exprOkForSpeculation was intended for.
1245 var_demanded_later _ = False
1248 --------------------------------------------------
1249 -- 3. Catch-all case
1250 --------------------------------------------------
1252 rebuildCase env scrut case_bndr alts cont
1253 = do { -- Prepare the continuation;
1254 -- The new subst_env is in place
1255 (env', dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1257 -- Simplify the alternatives
1258 ; (scrut', case_bndr', alts') <- simplAlts env' scrut case_bndr alts dup_cont
1260 -- Check for empty alternatives
1261 ; if null alts' then
1262 -- This isn't strictly an error, although it is unusual.
1263 -- It's possible that the simplifer might "see" that
1264 -- an inner case has no accessible alternatives before
1265 -- it "sees" that the entire branch of an outer case is
1266 -- inaccessible. So we simply put an error case here instead.
1267 pprTrace "mkCase: null alts" (ppr case_bndr <+> ppr scrut) $
1268 let res_ty' = contResultType env' (substTy env' (coreAltsType alts)) dup_cont
1269 lit = mkStringLit "Impossible alternative"
1270 in return (env', mkApps (Var rUNTIME_ERROR_ID) [Type res_ty', lit])
1273 { case_expr <- mkCase scrut' case_bndr' alts'
1275 -- Notice that rebuild gets the in-scope set from env, not alt_env
1276 -- The case binder *not* scope over the whole returned case-expression
1277 ; rebuild env' case_expr nodup_cont } }
1280 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1281 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1282 way, there's a chance that v will now only be used once, and hence
1285 Note [no-case-of-case]
1286 ~~~~~~~~~~~~~~~~~~~~~~
1287 We *used* to suppress the binder-swap in case expressoins when
1288 -fno-case-of-case is on. Old remarks:
1289 "This happens in the first simplifier pass,
1290 and enhances full laziness. Here's the bad case:
1291 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1292 If we eliminate the inner case, we trap it inside the I# v -> arm,
1293 which might prevent some full laziness happening. I've seen this
1294 in action in spectral/cichelli/Prog.hs:
1295 [(m,n) | m <- [1..max], n <- [1..max]]
1296 Hence the check for NoCaseOfCase."
1297 However, now the full-laziness pass itself reverses the binder-swap, so this
1298 check is no longer necessary.
1300 Note [Suppressing the case binder-swap]
1301 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1302 There is another situation when it might make sense to suppress the
1303 case-expression binde-swap. If we have
1305 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1306 ...other cases .... }
1308 We'll perform the binder-swap for the outer case, giving
1310 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1311 ...other cases .... }
1313 But there is no point in doing it for the inner case, because w1 can't
1314 be inlined anyway. Furthermore, doing the case-swapping involves
1315 zapping w2's occurrence info (see paragraphs that follow), and that
1316 forces us to bind w2 when doing case merging. So we get
1318 case x of w1 { A -> let w2 = w1 in e1
1319 B -> let w2 = w1 in e2
1320 ...other cases .... }
1322 This is plain silly in the common case where w2 is dead.
1324 Even so, I can't see a good way to implement this idea. I tried
1325 not doing the binder-swap if the scrutinee was already evaluated
1326 but that failed big-time:
1330 case v of w { MkT x ->
1331 case x of x1 { I# y1 ->
1332 case x of x2 { I# y2 -> ...
1334 Notice that because MkT is strict, x is marked "evaluated". But to
1335 eliminate the last case, we must either make sure that x (as well as
1336 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1337 the binder-swap. So this whole note is a no-op.
1341 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1342 any occurrence info (eg IAmDead) in the case binder, because the
1343 case-binder now effectively occurs whenever v does. AND we have to do
1344 the same for the pattern-bound variables! Example:
1346 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1348 Here, b and p are dead. But when we move the argment inside the first
1349 case RHS, and eliminate the second case, we get
1351 case x of { (a,b) -> a b }
1353 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1356 Indeed, this can happen anytime the case binder isn't dead:
1357 case <any> of x { (a,b) ->
1358 case x of { (p,q) -> p } }
1359 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1360 The point is that we bring into the envt a binding
1362 after the outer case, and that makes (a,b) alive. At least we do unless
1363 the case binder is guaranteed dead.
1367 Consider case (v `cast` co) of x { I# ->
1368 ... (case (v `cast` co) of {...}) ...
1369 We'd like to eliminate the inner case. We can get this neatly by
1370 arranging that inside the outer case we add the unfolding
1371 v |-> x `cast` (sym co)
1372 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1374 Note [Improving seq]
1377 type family F :: * -> *
1378 type instance F Int = Int
1380 ... case e of x { DEFAULT -> rhs } ...
1382 where x::F Int. Then we'd like to rewrite (F Int) to Int, getting
1384 case e `cast` co of x'::Int
1385 I# x# -> let x = x' `cast` sym co
1388 so that 'rhs' can take advantage of the form of x'. Notice that Note
1389 [Case of cast] may then apply to the result.
1391 This showed up in Roman's experiments. Example:
1392 foo :: F Int -> Int -> Int
1393 foo t n = t `seq` bar n
1396 bar n = bar (n - case t of TI i -> i)
1397 Here we'd like to avoid repeated evaluating t inside the loop, by
1398 taking advantage of the `seq`.
1400 At one point I did transformation in LiberateCase, but it's more robust here.
1401 (Otherwise, there's a danger that we'll simply drop the 'seq' altogether, before
1402 LiberateCase gets to see it.)
1404 Note [Case elimination]
1405 ~~~~~~~~~~~~~~~~~~~~~~~
1406 The case-elimination transformation discards redundant case expressions.
1407 Start with a simple situation:
1409 case x# of ===> e[x#/y#]
1412 (when x#, y# are of primitive type, of course). We can't (in general)
1413 do this for algebraic cases, because we might turn bottom into
1416 The code in SimplUtils.prepareAlts has the effect of generalise this
1417 idea to look for a case where we're scrutinising a variable, and we
1418 know that only the default case can match. For example:
1422 DEFAULT -> ...(case x of
1426 Here the inner case is first trimmed to have only one alternative, the
1427 DEFAULT, after which it's an instance of the previous case. This
1428 really only shows up in eliminating error-checking code.
1430 We also make sure that we deal with this very common case:
1435 Here we are using the case as a strict let; if x is used only once
1436 then we want to inline it. We have to be careful that this doesn't
1437 make the program terminate when it would have diverged before, so we
1439 - e is already evaluated (it may so if e is a variable)
1440 - x is used strictly, or
1442 Lastly, the code in SimplUtils.mkCase combines identical RHSs. So
1444 case e of ===> case e of DEFAULT -> r
1448 Now again the case may be elminated by the CaseElim transformation.
1451 Further notes about case elimination
1452 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1453 Consider: test :: Integer -> IO ()
1456 Turns out that this compiles to:
1459 eta1 :: State# RealWorld ->
1460 case PrelNum.< eta PrelNum.zeroInteger of wild { __DEFAULT ->
1462 (PrelNum.jtos eta ($w[] @ Char))
1464 of wild1 { (# new_s, a4 #) -> PrelIO.lvl23 new_s }}
1466 Notice the strange '<' which has no effect at all. This is a funny one.
1467 It started like this:
1469 f x y = if x < 0 then jtos x
1470 else if y==0 then "" else jtos x
1472 At a particular call site we have (f v 1). So we inline to get
1474 if v < 0 then jtos x
1475 else if 1==0 then "" else jtos x
1477 Now simplify the 1==0 conditional:
1479 if v<0 then jtos v else jtos v
1481 Now common-up the two branches of the case:
1483 case (v<0) of DEFAULT -> jtos v
1485 Why don't we drop the case? Because it's strict in v. It's technically
1486 wrong to drop even unnecessary evaluations, and in practice they
1487 may be a result of 'seq' so we *definitely* don't want to drop those.
1488 I don't really know how to improve this situation.
1492 simplCaseBinder :: SimplEnv -> OutExpr -> OutId -> [InAlt]
1493 -> SimplM (SimplEnv, OutExpr, OutId)
1494 simplCaseBinder env0 scrut0 case_bndr0 alts
1495 = do { (env1, case_bndr1) <- simplBinder env0 case_bndr0
1497 ; fam_envs <- getFamEnvs
1498 ; (env2, scrut2, case_bndr2) <- improve_seq fam_envs env1 scrut0
1499 case_bndr0 case_bndr1 alts
1500 -- Note [Improving seq]
1502 ; let (env3, case_bndr3) = improve_case_bndr env2 scrut2 case_bndr2
1503 -- Note [Case of cast]
1505 ; return (env3, scrut2, case_bndr3) }
1508 improve_seq fam_envs env scrut case_bndr case_bndr1 [(DEFAULT,_,_)]
1509 | Just (co, ty2) <- topNormaliseType fam_envs (idType case_bndr1)
1510 = do { case_bndr2 <- newId (fsLit "nt") ty2
1511 ; let rhs = DoneEx (Var case_bndr2 `Cast` mkSymCoercion co)
1512 env2 = extendIdSubst env case_bndr rhs
1513 ; return (env2, scrut `Cast` co, case_bndr2) }
1515 improve_seq _ env scrut _ case_bndr1 _
1516 = return (env, scrut, case_bndr1)
1519 improve_case_bndr env scrut case_bndr
1520 -- See Note [no-case-of-case]
1521 -- | switchIsOn (getSwitchChecker env) NoCaseOfCase
1522 -- = (env, case_bndr)
1524 | otherwise -- Failed try; see Note [Suppressing the case binder-swap]
1525 -- not (isEvaldUnfolding (idUnfolding v))
1527 Var v -> (modifyInScope env1 v case_bndr', case_bndr')
1528 -- Note about using modifyInScope for v here
1529 -- We could extend the substitution instead, but it would be
1530 -- a hack because then the substitution wouldn't be idempotent
1531 -- any more (v is an OutId). And this does just as well.
1533 Cast (Var v) co -> (addBinderUnfolding env1 v rhs, case_bndr')
1535 rhs = Cast (Var case_bndr') (mkSymCoercion co)
1537 _ -> (env, case_bndr)
1539 case_bndr' = zapOccInfo case_bndr
1540 env1 = modifyInScope env case_bndr case_bndr'
1543 zapOccInfo :: InId -> InId -- See Note [zapOccInfo]
1544 zapOccInfo b = b `setIdOccInfo` NoOccInfo
1548 simplAlts does two things:
1550 1. Eliminate alternatives that cannot match, including the
1551 DEFAULT alternative.
1553 2. If the DEFAULT alternative can match only one possible constructor,
1554 then make that constructor explicit.
1556 case e of x { DEFAULT -> rhs }
1558 case e of x { (a,b) -> rhs }
1559 where the type is a single constructor type. This gives better code
1560 when rhs also scrutinises x or e.
1562 Here "cannot match" includes knowledge from GADTs
1564 It's a good idea do do this stuff before simplifying the alternatives, to
1565 avoid simplifying alternatives we know can't happen, and to come up with
1566 the list of constructors that are handled, to put into the IdInfo of the
1567 case binder, for use when simplifying the alternatives.
1569 Eliminating the default alternative in (1) isn't so obvious, but it can
1572 data Colour = Red | Green | Blue
1581 DEFAULT -> [ case y of ... ]
1583 If we inline h into f, the default case of the inlined h can't happen.
1584 If we don't notice this, we may end up filtering out *all* the cases
1585 of the inner case y, which give us nowhere to go!
1589 simplAlts :: SimplEnv
1591 -> InId -- Case binder
1592 -> [InAlt] -- Non-empty
1594 -> SimplM (OutExpr, OutId, [OutAlt]) -- Includes the continuation
1595 -- Like simplExpr, this just returns the simplified alternatives;
1596 -- it not return an environment
1598 simplAlts env scrut case_bndr alts cont'
1599 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1600 do { let alt_env = zapFloats env
1601 ; (alt_env', scrut', case_bndr') <- simplCaseBinder alt_env scrut case_bndr alts
1603 ; (imposs_deflt_cons, in_alts) <- prepareAlts alt_env' scrut case_bndr' alts
1605 ; alts' <- mapM (simplAlt alt_env' imposs_deflt_cons case_bndr' cont') in_alts
1606 ; return (scrut', case_bndr', alts') }
1608 ------------------------------------
1609 simplAlt :: SimplEnv
1610 -> [AltCon] -- These constructors can't be present when
1611 -- matching the DEFAULT alternative
1612 -> OutId -- The case binder
1617 simplAlt env imposs_deflt_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1618 = ASSERT( null bndrs )
1619 do { let env' = addBinderOtherCon env case_bndr' imposs_deflt_cons
1620 -- Record the constructors that the case-binder *can't* be.
1621 ; rhs' <- simplExprC env' rhs cont'
1622 ; return (DEFAULT, [], rhs') }
1624 simplAlt env _ case_bndr' cont' (LitAlt lit, bndrs, rhs)
1625 = ASSERT( null bndrs )
1626 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1627 ; rhs' <- simplExprC env' rhs cont'
1628 ; return (LitAlt lit, [], rhs') }
1630 simplAlt env _ case_bndr' cont' (DataAlt con, vs, rhs)
1631 = do { -- Deal with the pattern-bound variables
1632 -- Mark the ones that are in ! positions in the
1633 -- data constructor as certainly-evaluated.
1634 -- NB: simplLamBinders preserves this eval info
1635 let vs_with_evals = add_evals (dataConRepStrictness con)
1636 ; (env', vs') <- simplLamBndrs env vs_with_evals
1638 -- Bind the case-binder to (con args)
1639 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1640 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1641 env'' = addBinderUnfolding env' case_bndr'
1642 (mkConApp con con_args)
1644 ; rhs' <- simplExprC env'' rhs cont'
1645 ; return (DataAlt con, vs', rhs') }
1647 -- add_evals records the evaluated-ness of the bound variables of
1648 -- a case pattern. This is *important*. Consider
1649 -- data T = T !Int !Int
1651 -- case x of { T a b -> T (a+1) b }
1653 -- We really must record that b is already evaluated so that we don't
1654 -- go and re-evaluate it when constructing the result.
1655 -- See Note [Data-con worker strictness] in MkId.lhs
1660 go (v:vs') strs | isTyVar v = v : go vs' strs
1661 go (v:vs') (str:strs)
1662 | isMarkedStrict str = evald_v : go vs' strs
1663 | otherwise = zapped_v : go vs' strs
1665 zapped_v = zap_occ_info v
1666 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1667 go _ _ = pprPanic "cat_evals" (ppr con $$ ppr vs $$ ppr the_strs)
1669 -- zap_occ_info: if the case binder is alive, then we add the unfolding
1671 -- to the envt; so vs are now very much alive
1672 -- Note [Aug06] I can't see why this actually matters, but it's neater
1673 -- case e of t { (a,b) -> ...(case t of (p,q) -> p)... }
1674 -- ==> case e of t { (a,b) -> ...(a)... }
1675 -- Look, Ma, a is alive now.
1676 zap_occ_info | isDeadBinder case_bndr' = \ident -> ident
1677 | otherwise = zapOccInfo
1679 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1680 addBinderUnfolding env bndr rhs
1681 = modifyInScope env bndr (bndr `setIdUnfolding` mkUnfolding False rhs)
1683 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1684 addBinderOtherCon env bndr cons
1685 = modifyInScope env bndr (bndr `setIdUnfolding` mkOtherCon cons)
1689 %************************************************************************
1691 \subsection{Known constructor}
1693 %************************************************************************
1695 We are a bit careful with occurrence info. Here's an example
1697 (\x* -> case x of (a*, b) -> f a) (h v, e)
1699 where the * means "occurs once". This effectively becomes
1700 case (h v, e) of (a*, b) -> f a)
1702 let a* = h v; b = e in f a
1706 All this should happen in one sweep.
1709 knownCon :: SimplEnv -> OutExpr -> AltCon
1710 -> [OutExpr] -- Args *including* the universal args
1711 -> InId -> [InAlt] -> SimplCont
1712 -> SimplM (SimplEnv, OutExpr)
1714 knownCon env scrut con args bndr alts cont
1715 = do { tick (KnownBranch bndr)
1716 ; knownAlt env scrut args bndr (findAlt con alts) cont }
1718 knownAlt :: SimplEnv -> OutExpr -> [OutExpr]
1719 -> InId -> (AltCon, [CoreBndr], InExpr) -> SimplCont
1720 -> SimplM (SimplEnv, OutExpr)
1721 knownAlt env scrut _ bndr (DEFAULT, bs, rhs) cont
1723 do { env' <- simplNonRecX env bndr scrut
1724 -- This might give rise to a binding with non-atomic args
1725 -- like x = Node (f x) (g x)
1726 -- but simplNonRecX will atomic-ify it
1727 ; simplExprF env' rhs cont }
1729 knownAlt env scrut _ bndr (LitAlt _, bs, rhs) cont
1731 do { env' <- simplNonRecX env bndr scrut
1732 ; simplExprF env' rhs cont }
1734 knownAlt env scrut the_args bndr (DataAlt dc, bs, rhs) cont
1735 = do { let dead_bndr = isDeadBinder bndr -- bndr is an InId
1736 n_drop_tys = length (dataConUnivTyVars dc)
1737 ; env' <- bind_args env dead_bndr bs (drop n_drop_tys the_args)
1739 -- It's useful to bind bndr to scrut, rather than to a fresh
1740 -- binding x = Con arg1 .. argn
1741 -- because very often the scrut is a variable, so we avoid
1742 -- creating, and then subsequently eliminating, a let-binding
1743 -- BUT, if scrut is a not a variable, we must be careful
1744 -- about duplicating the arg redexes; in that case, make
1745 -- a new con-app from the args
1746 bndr_rhs = case scrut of
1749 con_app = mkConApp dc (take n_drop_tys the_args ++ con_args)
1750 con_args = [substExpr env' (varToCoreExpr b) | b <- bs]
1751 -- args are aready OutExprs, but bs are InIds
1753 ; env'' <- simplNonRecX env' bndr bndr_rhs
1754 ; -- pprTrace "knownCon2" (ppr bs $$ ppr rhs $$ ppr (seIdSubst env'')) $
1755 simplExprF env'' rhs cont }
1758 bind_args env' _ [] _ = return env'
1760 bind_args env' dead_bndr (b:bs') (Type ty : args)
1761 = ASSERT( isTyVar b )
1762 bind_args (extendTvSubst env' b ty) dead_bndr bs' args
1764 bind_args env' dead_bndr (b:bs') (arg : args)
1766 do { let b' = if dead_bndr then b else zapOccInfo b
1767 -- Note that the binder might be "dead", because it doesn't
1768 -- occur in the RHS; and simplNonRecX may therefore discard
1769 -- it via postInlineUnconditionally.
1770 -- Nevertheless we must keep it if the case-binder is alive,
1771 -- because it may be used in the con_app. See Note [zapOccInfo]
1772 ; env'' <- simplNonRecX env' b' arg
1773 ; bind_args env'' dead_bndr bs' args }
1776 pprPanic "bind_args" $ ppr dc $$ ppr bs $$ ppr the_args $$
1777 text "scrut:" <+> ppr scrut
1781 %************************************************************************
1783 \subsection{Duplicating continuations}
1785 %************************************************************************
1788 prepareCaseCont :: SimplEnv
1789 -> [InAlt] -> SimplCont
1790 -> SimplM (SimplEnv, SimplCont,SimplCont)
1791 -- Return a duplicatable continuation, a non-duplicable part
1792 -- plus some extra bindings (that scope over the entire
1795 -- No need to make it duplicatable if there's only one alternative
1796 prepareCaseCont env [_] cont = return (env, cont, mkBoringStop)
1797 prepareCaseCont env _ cont = mkDupableCont env cont
1801 mkDupableCont :: SimplEnv -> SimplCont
1802 -> SimplM (SimplEnv, SimplCont, SimplCont)
1804 mkDupableCont env cont
1805 | contIsDupable cont
1806 = return (env, cont, mkBoringStop)
1808 mkDupableCont _ (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1810 mkDupableCont env (CoerceIt ty cont)
1811 = do { (env', dup, nodup) <- mkDupableCont env cont
1812 ; return (env', CoerceIt ty dup, nodup) }
1814 mkDupableCont env cont@(StrictBind {})
1815 = return (env, mkBoringStop, cont)
1816 -- See Note [Duplicating strict continuations]
1818 mkDupableCont env cont@(StrictArg {})
1819 = return (env, mkBoringStop, cont)
1820 -- See Note [Duplicating strict continuations]
1822 mkDupableCont env (ApplyTo _ arg se cont)
1823 = -- e.g. [...hole...] (...arg...)
1825 -- let a = ...arg...
1826 -- in [...hole...] a
1827 do { (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1828 ; arg' <- simplExpr (se `setInScope` env') arg
1829 ; (env'', arg'') <- makeTrivial env' arg'
1830 ; let app_cont = ApplyTo OkToDup arg'' (zapSubstEnv env') dup_cont
1831 ; return (env'', app_cont, nodup_cont) }
1833 mkDupableCont env cont@(Select _ case_bndr [(_, bs, _rhs)] _ _)
1834 -- See Note [Single-alternative case]
1835 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1836 -- | not (isDeadBinder case_bndr)
1837 | all isDeadBinder bs -- InIds
1838 && not (isUnLiftedType (idType case_bndr))
1839 -- Note [Single-alternative-unlifted]
1840 = return (env, mkBoringStop, cont)
1842 mkDupableCont env (Select _ case_bndr alts se cont)
1843 = -- e.g. (case [...hole...] of { pi -> ei })
1845 -- let ji = \xij -> ei
1846 -- in case [...hole...] of { pi -> ji xij }
1847 do { tick (CaseOfCase case_bndr)
1848 ; (env', dup_cont, nodup_cont) <- mkDupableCont env cont
1849 -- NB: call mkDupableCont here, *not* prepareCaseCont
1850 -- We must make a duplicable continuation, whereas prepareCaseCont
1851 -- doesn't when there is a single case branch
1853 ; let alt_env = se `setInScope` env'
1854 ; (alt_env', case_bndr') <- simplBinder alt_env case_bndr
1855 ; alts' <- mapM (simplAlt alt_env' [] case_bndr' dup_cont) alts
1856 -- Safe to say that there are no handled-cons for the DEFAULT case
1857 -- NB: simplBinder does not zap deadness occ-info, so
1858 -- a dead case_bndr' will still advertise its deadness
1859 -- This is really important because in
1860 -- case e of b { (# p,q #) -> ... }
1861 -- b is always dead, and indeed we are not allowed to bind b to (# p,q #),
1862 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1863 -- In the new alts we build, we have the new case binder, so it must retain
1865 -- NB: we don't use alt_env further; it has the substEnv for
1866 -- the alternatives, and we don't want that
1868 ; (env'', alts'') <- mkDupableAlts env' case_bndr' alts'
1869 ; return (env'', -- Note [Duplicated env]
1870 Select OkToDup case_bndr' alts'' (zapSubstEnv env'') mkBoringStop,
1874 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1875 -> SimplM (SimplEnv, [InAlt])
1876 -- Absorbs the continuation into the new alternatives
1878 mkDupableAlts env case_bndr' the_alts
1881 go env0 [] = return (env0, [])
1883 = do { (env1, alt') <- mkDupableAlt env0 case_bndr' alt
1884 ; (env2, alts') <- go env1 alts
1885 ; return (env2, alt' : alts' ) }
1887 mkDupableAlt :: SimplEnv -> OutId -> (AltCon, [CoreBndr], CoreExpr)
1888 -> SimplM (SimplEnv, (AltCon, [CoreBndr], CoreExpr))
1889 mkDupableAlt env case_bndr' (con, bndrs', rhs')
1890 | exprIsDupable rhs' -- Note [Small alternative rhs]
1891 = return (env, (con, bndrs', rhs'))
1893 = do { let rhs_ty' = exprType rhs'
1894 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1896 | isTyVar bndr = True -- Abstract over all type variables just in case
1897 | otherwise = not (isDeadBinder bndr)
1898 -- The deadness info on the new Ids is preserved by simplBinders
1900 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1901 <- if (any isId used_bndrs')
1902 then return (used_bndrs', varsToCoreExprs used_bndrs')
1903 else do { rw_id <- newId (fsLit "w") realWorldStatePrimTy
1904 ; return ([rw_id], [Var realWorldPrimId]) }
1906 ; join_bndr <- newId (fsLit "$j") (mkPiTypes final_bndrs' rhs_ty')
1907 -- Note [Funky mkPiTypes]
1909 ; let -- We make the lambdas into one-shot-lambdas. The
1910 -- join point is sure to be applied at most once, and doing so
1911 -- prevents the body of the join point being floated out by
1912 -- the full laziness pass
1913 really_final_bndrs = map one_shot final_bndrs'
1914 one_shot v | isId v = setOneShotLambda v
1916 join_rhs = mkLams really_final_bndrs rhs'
1917 join_call = mkApps (Var join_bndr) final_args
1919 ; return (addPolyBind NotTopLevel env (NonRec join_bndr join_rhs), (con, bndrs', join_call)) }
1920 -- See Note [Duplicated env]
1923 Note [Duplicated env]
1924 ~~~~~~~~~~~~~~~~~~~~~
1925 Some of the alternatives are simplified, but have not been turned into a join point
1926 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
1927 bind the join point, because it might to do PostInlineUnconditionally, and
1928 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
1929 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
1930 at worst delays the join-point inlining.
1932 Note [Small alterantive rhs]
1933 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1934 It is worth checking for a small RHS because otherwise we
1935 get extra let bindings that may cause an extra iteration of the simplifier to
1936 inline back in place. Quite often the rhs is just a variable or constructor.
1937 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1938 iterations because the version with the let bindings looked big, and so wasn't
1939 inlined, but after the join points had been inlined it looked smaller, and so
1942 NB: we have to check the size of rhs', not rhs.
1943 Duplicating a small InAlt might invalidate occurrence information
1944 However, if it *is* dupable, we return the *un* simplified alternative,
1945 because otherwise we'd need to pair it up with an empty subst-env....
1946 but we only have one env shared between all the alts.
1947 (Remember we must zap the subst-env before re-simplifying something).
1948 Rather than do this we simply agree to re-simplify the original (small) thing later.
1950 Note [Funky mkPiTypes]
1951 ~~~~~~~~~~~~~~~~~~~~~~
1952 Notice the funky mkPiTypes. If the contructor has existentials
1953 it's possible that the join point will be abstracted over
1954 type varaibles as well as term variables.
1955 Example: Suppose we have
1956 data T = forall t. C [t]
1958 case (case e of ...) of
1960 We get the join point
1961 let j :: forall t. [t] -> ...
1962 j = /\t \xs::[t] -> rhs
1964 case (case e of ...) of
1965 C t xs::[t] -> j t xs
1967 Note [Join point abstaction]
1968 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1969 If we try to lift a primitive-typed something out
1970 for let-binding-purposes, we will *caseify* it (!),
1971 with potentially-disastrous strictness results. So
1972 instead we turn it into a function: \v -> e
1973 where v::State# RealWorld#. The value passed to this function
1974 is realworld#, which generates (almost) no code.
1976 There's a slight infelicity here: we pass the overall
1977 case_bndr to all the join points if it's used in *any* RHS,
1978 because we don't know its usage in each RHS separately
1980 We used to say "&& isUnLiftedType rhs_ty'" here, but now
1981 we make the join point into a function whenever used_bndrs'
1982 is empty. This makes the join-point more CPR friendly.
1983 Consider: let j = if .. then I# 3 else I# 4
1984 in case .. of { A -> j; B -> j; C -> ... }
1986 Now CPR doesn't w/w j because it's a thunk, so
1987 that means that the enclosing function can't w/w either,
1988 which is a lose. Here's the example that happened in practice:
1989 kgmod :: Int -> Int -> Int
1990 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1994 I have seen a case alternative like this:
1996 It's a bit silly to add the realWorld dummy arg in this case, making
1999 (the \v alone is enough to make CPR happy) but I think it's rare
2001 Note [Duplicating strict continuations]
2002 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2003 Do *not* duplicate StrictBind and StritArg continuations. We gain
2004 nothing by propagating them into the expressions, and we do lose a
2005 lot. Here's an example:
2006 && (case x of { T -> F; F -> T }) E
2007 Now, && is strict so we end up simplifying the case with
2008 an ArgOf continuation. If we let-bind it, we get
2010 let $j = \v -> && v E
2011 in simplExpr (case x of { T -> F; F -> T })
2013 And after simplifying more we get
2015 let $j = \v -> && v E
2016 in case x of { T -> $j F; F -> $j T }
2017 Which is a Very Bad Thing
2019 The desire not to duplicate is the entire reason that
2020 mkDupableCont returns a pair of continuations.
2022 The original plan had:
2023 e.g. (...strict-fn...) [...hole...]
2025 let $j = \a -> ...strict-fn...
2028 Note [Single-alternative cases]
2029 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2030 This case is just like the ArgOf case. Here's an example:
2034 case (case x of I# x' ->
2036 True -> I# (negate# x')
2037 False -> I# x') of y {
2039 Because the (case x) has only one alternative, we'll transform to
2041 case (case x' <# 0# of
2042 True -> I# (negate# x')
2043 False -> I# x') of y {
2045 But now we do *NOT* want to make a join point etc, giving
2047 let $j = \y -> MkT y
2049 True -> $j (I# (negate# x'))
2051 In this case the $j will inline again, but suppose there was a big
2052 strict computation enclosing the orginal call to MkT. Then, it won't
2053 "see" the MkT any more, because it's big and won't get duplicated.
2054 And, what is worse, nothing was gained by the case-of-case transform.
2056 When should use this case of mkDupableCont?
2057 However, matching on *any* single-alternative case is a *disaster*;
2058 e.g. case (case ....) of (a,b) -> (# a,b #)
2059 We must push the outer case into the inner one!
2062 * Match [(DEFAULT,_,_)], but in the common case of Int,
2063 the alternative-filling-in code turned the outer case into
2064 case (...) of y { I# _ -> MkT y }
2066 * Match on single alternative plus (not (isDeadBinder case_bndr))
2067 Rationale: pushing the case inwards won't eliminate the construction.
2068 But there's a risk of
2069 case (...) of y { (a,b) -> let z=(a,b) in ... }
2070 Now y looks dead, but it'll come alive again. Still, this
2071 seems like the best option at the moment.
2073 * Match on single alternative plus (all (isDeadBinder bndrs))
2074 Rationale: this is essentially seq.
2076 * Match when the rhs is *not* duplicable, and hence would lead to a
2077 join point. This catches the disaster-case above. We can test
2078 the *un-simplified* rhs, which is fine. It might get bigger or
2079 smaller after simplification; if it gets smaller, this case might
2080 fire next time round. NB also that we must test contIsDupable
2081 case_cont *btoo, because case_cont might be big!
2083 HOWEVER: I found that this version doesn't work well, because
2084 we can get let x = case (...) of { small } in ...case x...
2085 When x is inlined into its full context, we find that it was a bad
2086 idea to have pushed the outer case inside the (...) case.
2088 Note [Single-alternative-unlifted]
2089 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2090 Here's another single-alternative where we really want to do case-of-case:
2098 case y_s6X of tpl_s7m {
2099 M1.Mk1 ipv_s70 -> ipv_s70;
2100 M1.Mk2 ipv_s72 -> ipv_s72;
2106 case x_s74 of tpl_s7n {
2107 M1.Mk1 ipv_s77 -> ipv_s77;
2108 M1.Mk2 ipv_s79 -> ipv_s79;
2112 { __DEFAULT -> ==# [wild1_s7b wild_s7c];
2116 So the outer case is doing *nothing at all*, other than serving as a
2117 join-point. In this case we really want to do case-of-case and decide
2118 whether to use a real join point or just duplicate the continuation.
2120 Hence: check whether the case binder's type is unlifted, because then
2121 the outer case is *not* a seq.