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"
11 import DynFlags ( dopt, DynFlag(Opt_D_dump_inlinings),
15 import Type hiding ( substTy, extendTvSubst )
21 import TcGadt ( dataConCanMatch )
22 import DataCon ( dataConTyCon, dataConRepStrictness )
23 import TyCon ( tyConArity, isAlgTyCon, isNewTyCon, tyConDataCons_maybe )
25 import PprCore ( pprParendExpr, pprCoreExpr )
26 import CoreUnfold ( mkUnfolding, callSiteInline )
28 import Rules ( lookupRule )
29 import BasicTypes ( isMarkedStrict )
30 import CostCentre ( currentCCS )
31 import TysPrim ( realWorldStatePrimTy )
32 import PrelInfo ( realWorldPrimId )
33 import BasicTypes ( TopLevelFlag(..), isTopLevel,
34 RecFlag(..), isNonRuleLoopBreaker )
36 import Maybes ( orElse )
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 env binds
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 ; env <- simplRecBndrs env (bindersOfBinds binds)
210 ; dflags <- getDOptsSmpl
211 ; let dump_flag = dopt Opt_D_dump_inlinings dflags
212 ; env' <- simpl_binds dump_flag env binds
213 ; freeTick SimplifierDone
214 ; return (getFloats env') }
216 -- We need to track the zapped top-level binders, because
217 -- they should have their fragile IdInfo zapped (notably occurrence info)
218 -- That's why we run down binds and bndrs' simultaneously.
219 simpl_binds :: Bool -> SimplEnv -> [InBind] -> SimplM SimplEnv
220 simpl_binds dump env [] = return env
221 simpl_binds dump env (bind:binds) = do { env' <- trace dump bind $
223 ; simpl_binds dump env' binds }
225 trace True bind = pprTrace "SimplBind" (ppr (bindersOf bind))
226 trace False bind = \x -> x
228 simpl_bind env (NonRec b r) = simplRecOrTopPair env TopLevel b r
229 simpl_bind env (Rec pairs) = simplRecBind env TopLevel pairs
233 %************************************************************************
235 \subsection{Lazy bindings}
237 %************************************************************************
239 simplRecBind is used for
240 * recursive bindings only
243 simplRecBind :: SimplEnv -> TopLevelFlag
246 simplRecBind env top_lvl pairs
247 = do { env' <- go (zapFloats env) pairs
248 ; return (env `addRecFloats` env') }
249 -- addFloats adds the floats from env',
250 -- *and* updates env with the in-scope set from env'
252 go env [] = return env
254 go env ((bndr, rhs) : pairs)
255 = do { env <- simplRecOrTopPair env top_lvl bndr rhs
259 simplOrTopPair is used for
260 * recursive bindings (whether top level or not)
261 * top-level non-recursive bindings
263 It assumes the binder has already been simplified, but not its IdInfo.
266 simplRecOrTopPair :: SimplEnv
268 -> InId -> InExpr -- Binder and rhs
269 -> SimplM SimplEnv -- Returns an env that includes the binding
271 simplRecOrTopPair env top_lvl bndr rhs
272 | preInlineUnconditionally env top_lvl bndr rhs -- Check for unconditional inline
273 = do { tick (PreInlineUnconditionally bndr)
274 ; return (extendIdSubst env bndr (mkContEx env rhs)) }
277 = do { let bndr' = lookupRecBndr env bndr
278 (env', bndr'') = addLetIdInfo env bndr bndr'
279 ; simplLazyBind env' top_lvl Recursive bndr bndr'' rhs env' }
280 -- May not actually be recursive, but it doesn't matter
284 simplLazyBind is used for
285 * [simplRecOrTopPair] recursive bindings (whether top level or not)
286 * [simplRecOrTopPair] top-level non-recursive bindings
287 * [simplNonRecE] non-top-level *lazy* non-recursive bindings
290 1. It assumes that the binder is *already* simplified,
291 and is in scope, and its IdInfo too, except unfolding
293 2. It assumes that the binder type is lifted.
295 3. It does not check for pre-inline-unconditionallly;
296 that should have been done already.
299 simplLazyBind :: SimplEnv
300 -> TopLevelFlag -> RecFlag
301 -> InId -> OutId -- Binder, both pre-and post simpl
302 -- The OutId has IdInfo, except arity, unfolding
303 -> InExpr -> SimplEnv -- The RHS and its environment
306 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
307 = do { let rhs_env = rhs_se `setInScope` env
308 rhs_cont = mkRhsStop (idType bndr1)
310 -- Simplify the RHS; note the mkRhsStop, which tells
311 -- the simplifier that this is the RHS of a let.
312 ; (rhs_env1, rhs1) <- simplExprF rhs_env rhs rhs_cont
314 -- If any of the floats can't be floated, give up now
315 -- (The canFloat predicate says True for empty floats.)
316 ; if (not (canFloat top_lvl is_rec False rhs_env1))
317 then completeBind env top_lvl bndr bndr1
318 (wrapFloats rhs_env1 rhs1)
320 -- ANF-ise a constructor or PAP rhs
321 { (rhs_env2, rhs2) <- prepareRhs rhs_env1 rhs1
322 ; (env', rhs3) <- chooseRhsFloats top_lvl is_rec False env rhs_env2 rhs2
323 ; completeBind env' top_lvl bndr bndr1 rhs3 } }
325 chooseRhsFloats :: TopLevelFlag -> RecFlag -> Bool
326 -> SimplEnv -- Env for the let
327 -> SimplEnv -- Env for the RHS, with RHS floats in it
328 -> OutExpr -- ..and the RHS itself
329 -> SimplM (SimplEnv, OutExpr) -- New env for let, and RHS
331 chooseRhsFloats top_lvl is_rec is_strict env rhs_env rhs
332 | not (isEmptyFloats rhs_env) -- Something to float
333 , canFloat top_lvl is_rec is_strict rhs_env -- ...that can float
334 , (isTopLevel top_lvl || exprIsCheap rhs) -- ...and we want to float
335 = do { tick LetFloatFromLet -- Float
336 ; return (addFloats env rhs_env, rhs) } -- Add the floats to the main env
337 | otherwise -- Don't float
338 = return (env, wrapFloats rhs_env rhs) -- Wrap the floats around the RHS
342 %************************************************************************
344 \subsection{simplNonRec}
346 %************************************************************************
348 A specialised variant of simplNonRec used when the RHS is already simplified,
349 notably in knownCon. It uses case-binding where necessary.
352 simplNonRecX :: SimplEnv
353 -> InId -- Old binder
354 -> OutExpr -- Simplified RHS
357 simplNonRecX env bndr new_rhs
358 = do { (env, bndr') <- simplBinder env bndr
359 ; completeNonRecX env NotTopLevel NonRecursive
360 (isStrictBndr bndr) bndr bndr' new_rhs }
362 completeNonRecX :: SimplEnv
363 -> TopLevelFlag -> RecFlag -> Bool
364 -> InId -- Old binder
365 -> OutId -- New binder
366 -> OutExpr -- Simplified RHS
369 completeNonRecX env top_lvl is_rec is_strict old_bndr new_bndr new_rhs
370 = do { (env1, rhs1) <- prepareRhs (zapFloats env) new_rhs
371 ; (env2, rhs2) <- chooseRhsFloats top_lvl is_rec is_strict env env1 rhs1
372 ; completeBind env2 NotTopLevel old_bndr new_bndr rhs2 }
375 {- No, no, no! Do not try preInlineUnconditionally in completeNonRecX
376 Doing so risks exponential behaviour, because new_rhs has been simplified once already
377 In the cases described by the folowing commment, postInlineUnconditionally will
378 catch many of the relevant cases.
379 -- This happens; for example, the case_bndr during case of
380 -- known constructor: case (a,b) of x { (p,q) -> ... }
381 -- Here x isn't mentioned in the RHS, so we don't want to
382 -- create the (dead) let-binding let x = (a,b) in ...
384 -- Similarly, single occurrences can be inlined vigourously
385 -- e.g. case (f x, g y) of (a,b) -> ....
386 -- If a,b occur once we can avoid constructing the let binding for them.
388 Furthermore in the case-binding case preInlineUnconditionally risks extra thunks
389 -- Consider case I# (quotInt# x y) of
390 -- I# v -> let w = J# v in ...
391 -- If we gaily inline (quotInt# x y) for v, we end up building an
393 -- let w = J# (quotInt# x y) in ...
394 -- because quotInt# can fail.
396 | preInlineUnconditionally env NotTopLevel bndr new_rhs
397 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
400 prepareRhs takes a putative RHS, checks whether it's a PAP or
401 constructor application and, if so, converts it to ANF, so that the
402 resulting thing can be inlined more easily. Thus
410 prepareRhs :: SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
411 -- Adds new floats to the env iff that allows us to return a good RHS
413 prepareRhs env (Cast rhs co) -- Note [Float coersions]
414 = do { (env', rhs') <- makeTrivial env rhs
415 ; return (env', Cast rhs' co) }
418 | (Var fun, args) <- collectArgs rhs -- It's an application
419 , let n_args = valArgCount args
420 , n_args > 0 -- ...but not a trivial one
421 , isDataConWorkId fun || n_args < idArity fun -- ...and it's a constructor or PAP
422 = go env (Var fun) args
424 go env fun [] = return (env, fun)
425 go env fun (arg : args) = do { (env', arg') <- makeTrivial env arg
426 ; go env' (App fun arg') args }
428 prepareRhs env rhs -- The default case
432 Note [Float coercions]
433 ~~~~~~~~~~~~~~~~~~~~~~
434 When we find the binding
436 we'd like to transform it to
438 x = x `cast` co -- A trivial binding
439 There's a chance that e will be a constructor application or function, or something
440 like that, so moving the coerion to the usage site may well cancel the coersions
441 and lead to further optimisation. Example:
444 data instance T Int = T Int
446 foo :: Int -> Int -> Int
451 go n = case x of { T m -> go (n-m) }
452 -- This case should optimise
456 makeTrivial :: SimplEnv -> OutExpr -> SimplM (SimplEnv, OutExpr)
457 -- Binds the expression to a variable, if it's not trivial, returning the variable
461 | otherwise -- See Note [Take care] below
462 = do { var <- newId FSLIT("a") (exprType expr)
463 ; env <- completeNonRecX env NotTopLevel NonRecursive
465 ; return (env, substExpr env (Var var)) }
469 %************************************************************************
471 \subsection{Completing a lazy binding}
473 %************************************************************************
476 * deals only with Ids, not TyVars
477 * takes an already-simplified binder and RHS
478 * is used for both recursive and non-recursive bindings
479 * is used for both top-level and non-top-level bindings
481 It does the following:
482 - tries discarding a dead binding
483 - tries PostInlineUnconditionally
484 - add unfolding [this is the only place we add an unfolding]
487 It does *not* attempt to do let-to-case. Why? Because it is used for
488 - top-level bindings (when let-to-case is impossible)
489 - many situations where the "rhs" is known to be a WHNF
490 (so let-to-case is inappropriate).
492 Nor does it do the atomic-argument thing
495 completeBind :: SimplEnv
496 -> TopLevelFlag -- Flag stuck into unfolding
497 -> InId -- Old binder
498 -> OutId -> OutExpr -- New binder and RHS
500 -- completeBind may choose to do its work
501 -- * by extending the substitution (e.g. let x = y in ...)
502 -- * or by adding to the floats in the envt
504 completeBind env top_lvl old_bndr new_bndr new_rhs
505 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
506 -- Inline and discard the binding
507 = do { tick (PostInlineUnconditionally old_bndr)
508 ; -- pprTrace "postInlineUnconditionally" (ppr old_bndr <+> ppr new_bndr <+> ppr new_rhs) $
509 return (extendIdSubst env old_bndr (DoneEx new_rhs)) }
510 -- Use the substitution to make quite, quite sure that the
511 -- substitution will happen, since we are going to discard the binding
516 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
519 -- Add the unfolding *only* for non-loop-breakers
520 -- Making loop breakers not have an unfolding at all
521 -- means that we can avoid tests in exprIsConApp, for example.
522 -- This is important: if exprIsConApp says 'yes' for a recursive
523 -- thing, then we can get into an infinite loop
526 -- If the unfolding is a value, the demand info may
527 -- go pear-shaped, so we nuke it. Example:
529 -- case x of (p,q) -> h p q x
530 -- Here x is certainly demanded. But after we've nuked
531 -- the case, we'll get just
532 -- let x = (a,b) in h a b x
533 -- and now x is not demanded (I'm assuming h is lazy)
534 -- This really happens. Similarly
535 -- let f = \x -> e in ...f..f...
536 -- After inlining f at some of its call sites the original binding may
537 -- (for example) be no longer strictly demanded.
538 -- The solution here is a bit ad hoc...
539 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
540 final_info | loop_breaker = new_bndr_info
541 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
542 | otherwise = info_w_unf
544 final_id = new_bndr `setIdInfo` final_info
546 -- These seqs forces the Id, and hence its IdInfo,
547 -- and hence any inner substitutions
549 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
550 return (addNonRec env final_id new_rhs)
552 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
553 loop_breaker = isNonRuleLoopBreaker occ_info
554 old_info = idInfo old_bndr
555 occ_info = occInfo old_info
560 %************************************************************************
562 \subsection[Simplify-simplExpr]{The main function: simplExpr}
564 %************************************************************************
566 The reason for this OutExprStuff stuff is that we want to float *after*
567 simplifying a RHS, not before. If we do so naively we get quadratic
568 behaviour as things float out.
570 To see why it's important to do it after, consider this (real) example:
584 a -- Can't inline a this round, cos it appears twice
588 Each of the ==> steps is a round of simplification. We'd save a
589 whole round if we float first. This can cascade. Consider
594 let f = let d1 = ..d.. in \y -> e
598 in \x -> ...(\y ->e)...
600 Only in this second round can the \y be applied, and it
601 might do the same again.
605 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
606 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
608 expr_ty' = substTy env (exprType expr)
609 -- The type in the Stop continuation, expr_ty', is usually not used
610 -- It's only needed when discarding continuations after finding
611 -- a function that returns bottom.
612 -- Hence the lazy substitution
615 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
616 -- Simplify an expression, given a continuation
617 simplExprC env expr cont
618 = -- pprTrace "simplExprC" (ppr expr $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
619 do { (env', expr') <- simplExprF (zapFloats env) expr cont
620 ; -- pprTrace "simplExprC ret" (ppr expr $$ ppr expr') $
621 -- pprTrace "simplExprC ret3" (ppr (seInScope env')) $
622 -- pprTrace "simplExprC ret4" (ppr (seFloats env')) $
623 return (wrapFloats env' expr') }
625 --------------------------------------------------
626 simplExprF :: SimplEnv -> InExpr -> SimplCont
627 -> SimplM (SimplEnv, OutExpr)
629 simplExprF env e cont = -- pprTrace "simplExprF" (ppr e $$ ppr cont {- $$ ppr (seIdSubst env) -} $$ ppr (seFloats env) ) $
630 simplExprF' env e cont
632 simplExprF' env (Var v) cont = simplVar env v cont
633 simplExprF' env (Lit lit) cont = rebuild env (Lit lit) cont
634 simplExprF' env (Note n expr) cont = simplNote env n expr cont
635 simplExprF' env (Cast body co) cont = simplCast env body co cont
636 simplExprF' env (App fun arg) cont = simplExprF env fun $
637 ApplyTo NoDup arg env cont
639 simplExprF' env expr@(Lam _ _) cont
640 = simplLam env (map zap bndrs) body cont
641 -- The main issue here is under-saturated lambdas
642 -- (\x1. \x2. e) arg1
643 -- Here x1 might have "occurs-once" occ-info, because occ-info
644 -- is computed assuming that a group of lambdas is applied
645 -- all at once. If there are too few args, we must zap the
648 n_args = countArgs cont
649 n_params = length bndrs
650 (bndrs, body) = collectBinders expr
651 zap | n_args >= n_params = \b -> b
652 | otherwise = \b -> if isTyVar b then b
654 -- NB: we count all the args incl type args
655 -- so we must count all the binders (incl type lambdas)
657 simplExprF' env (Type ty) cont
658 = ASSERT( contIsRhsOrArg cont )
659 do { ty' <- simplType env ty
660 ; rebuild env (Type ty') cont }
662 simplExprF' env (Case scrut bndr case_ty alts) cont
663 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
664 = -- Simplify the scrutinee with a Select continuation
665 simplExprF env scrut (Select NoDup bndr alts env cont)
668 = -- If case-of-case is off, simply simplify the case expression
669 -- in a vanilla Stop context, and rebuild the result around it
670 do { case_expr' <- simplExprC env scrut case_cont
671 ; rebuild env case_expr' cont }
673 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
674 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
676 simplExprF' env (Let (Rec pairs) body) cont
677 = do { env <- simplRecBndrs env (map fst pairs)
678 -- NB: bndrs' don't have unfoldings or rules
679 -- We add them as we go down
681 ; env <- simplRecBind env NotTopLevel pairs
682 ; simplExprF env body cont }
684 simplExprF' env (Let (NonRec bndr rhs) body) cont
685 = simplNonRecE env bndr (rhs, env) ([], body) cont
687 ---------------------------------
688 simplType :: SimplEnv -> InType -> SimplM OutType
689 -- Kept monadic just so we can do the seqType
691 = -- pprTrace "simplType" (ppr ty $$ ppr (seTvSubst env)) $
692 seqType new_ty `seq` returnSmpl new_ty
694 new_ty = substTy env ty
698 %************************************************************************
700 \subsection{The main rebuilder}
702 %************************************************************************
705 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM (SimplEnv, OutExpr)
706 -- At this point the substitution in the SimplEnv should be irrelevant
707 -- only the in-scope set and floats should matter
708 rebuild env expr cont
709 = -- pprTrace "rebuild" (ppr expr $$ ppr cont $$ ppr (seFloats env)) $
711 Stop {} -> return (env, expr)
712 CoerceIt co cont -> rebuild env (mkCoerce co expr) cont
713 Select _ bndr alts se cont -> rebuildCase (se `setFloats` env) expr bndr alts cont
714 StrictArg fun ty info cont -> rebuildCall env (fun `App` expr) (funResultTy ty) info cont
715 StrictBind b bs body se cont -> do { env' <- simplNonRecX (se `setFloats` env) b expr
716 ; simplLam env' bs body cont }
717 ApplyTo _ arg se cont -> do { arg' <- simplExpr (se `setInScope` env) arg
718 ; rebuild env (App expr arg') cont }
722 %************************************************************************
726 %************************************************************************
729 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont
730 -> SimplM (SimplEnv, OutExpr)
731 simplCast env body co cont
732 = do { co' <- simplType env co
733 ; simplExprF env body (addCoerce co' cont) }
736 | (s1, k1) <- coercionKind co
737 , s1 `coreEqType` k1 = cont
738 addCoerce co1 (CoerceIt co2 cont)
739 | (s1, k1) <- coercionKind co1
740 , (l1, t1) <- coercionKind co2
741 -- coerce T1 S1 (coerce S1 K1 e)
744 -- coerce T1 K1 e, otherwise
746 -- For example, in the initial form of a worker
747 -- we may find (coerce T (coerce S (\x.e))) y
748 -- and we'd like it to simplify to e[y/x] in one round
750 , s1 `coreEqType` t1 = cont -- The coerces cancel out
751 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
753 addCoerce co (ApplyTo dup arg arg_se cont)
754 | not (isTypeArg arg) -- This whole case only works for value args
755 -- Could upgrade to have equiv thing for type apps too
756 , Just (s1s2, t1t2) <- splitCoercionKind_maybe co
758 -- co : s1s2 :=: t1t2
759 -- (coerce (T1->T2) (S1->S2) F) E
761 -- coerce T2 S2 (F (coerce S1 T1 E))
763 -- t1t2 must be a function type, T1->T2, because it's applied
764 -- to something but s1s2 might conceivably not be
766 -- When we build the ApplyTo we can't mix the out-types
767 -- with the InExpr in the argument, so we simply substitute
768 -- to make it all consistent. It's a bit messy.
769 -- But it isn't a common case.
770 = ApplyTo dup new_arg (zapSubstEnv env) (addCoerce co2 cont)
772 -- we split coercion t1->t2 :=: s1->s2 into t1 :=: s1 and
773 -- t2 :=: s2 with left and right on the curried form:
774 -- (->) t1 t2 :=: (->) s1 s2
775 [co1, co2] = decomposeCo 2 co
776 new_arg = mkCoerce (mkSymCoercion co1) arg'
777 arg' = substExpr arg_se arg
779 addCoerce co cont = CoerceIt co cont
783 %************************************************************************
787 %************************************************************************
790 simplLam :: SimplEnv -> [InId] -> InExpr -> SimplCont
791 -> SimplM (SimplEnv, OutExpr)
793 simplLam env [] body cont = simplExprF env body cont
795 -- Type-beta reduction
796 simplLam env (bndr:bndrs) body (ApplyTo _ (Type ty_arg) arg_se cont)
797 = ASSERT( isTyVar bndr )
798 do { tick (BetaReduction bndr)
799 ; ty_arg' <- simplType (arg_se `setInScope` env) ty_arg
800 ; simplLam (extendTvSubst env bndr ty_arg') bndrs body cont }
802 -- Ordinary beta reduction
803 simplLam env (bndr:bndrs) body (ApplyTo _ arg arg_se cont)
804 = do { tick (BetaReduction bndr)
805 ; simplNonRecE env bndr (arg, arg_se) (bndrs, body) cont }
807 -- Not enough args, so there are real lambdas left to put in the result
808 simplLam env bndrs body cont
809 = do { (env, bndrs') <- simplLamBndrs env bndrs
810 ; body' <- simplExpr env body
811 ; new_lam <- mkLam bndrs' body'
812 ; rebuild env new_lam cont }
815 simplNonRecE :: SimplEnv
816 -> InId -- The binder
817 -> (InExpr, SimplEnv) -- Rhs of binding (or arg of lambda)
818 -> ([InId], InExpr) -- Body of the let/lambda
821 -> SimplM (SimplEnv, OutExpr)
823 -- simplNonRecE is used for
824 -- * non-top-level non-recursive lets in expressions
827 -- It deals with strict bindings, via the StrictBind continuation,
828 -- which may abort the whole process
830 -- The "body" of the binding comes as a pair of ([InId],InExpr)
831 -- representing a lambda; so we recurse back to simplLam
832 -- Why? Because of the binder-occ-info-zapping done before
833 -- the call to simplLam in simplExprF (Lam ...)
835 simplNonRecE env bndr (rhs, rhs_se) (bndrs, body) cont
836 | preInlineUnconditionally env NotTopLevel bndr rhs
837 = do { tick (PreInlineUnconditionally bndr)
838 ; simplLam (extendIdSubst env bndr (mkContEx rhs_se rhs)) bndrs body cont }
841 = do { simplExprF (rhs_se `setFloats` env) rhs
842 (StrictBind bndr bndrs body env cont) }
845 = do { (env, bndr') <- simplBinder env bndr
846 ; env <- simplLazyBind env NotTopLevel NonRecursive bndr bndr' rhs rhs_se
847 ; simplLam env bndrs body cont }
851 %************************************************************************
855 %************************************************************************
858 -- Hack alert: we only distinguish subsumed cost centre stacks for the
859 -- purposes of inlining. All other CCCSs are mapped to currentCCS.
860 simplNote env (SCC cc) e cont
861 = do { e' <- simplExpr (setEnclosingCC env currentCCS) e
862 ; rebuild env (mkSCC cc e') cont }
864 -- See notes with SimplMonad.inlineMode
865 simplNote env InlineMe e cont
866 | contIsRhsOrArg cont -- Totally boring continuation; see notes above
867 = do { -- Don't inline inside an INLINE expression
868 e' <- simplExpr (setMode inlineMode env) e
869 ; rebuild env (mkInlineMe e') cont }
871 | otherwise -- Dissolve the InlineMe note if there's
872 -- an interesting context of any kind to combine with
873 -- (even a type application -- anything except Stop)
874 = simplExprF env e cont
876 simplNote env (CoreNote s) e cont
877 = do { e' <- simplExpr env e
878 ; rebuild env (Note (CoreNote s) e') cont }
880 simplNote env note@(TickBox {}) e cont
881 = do { e' <- simplExpr env e
882 ; rebuild env (Note note e') cont }
884 simplNote env note@(BinaryTickBox {}) e cont
885 = do { e' <- simplExpr env e
886 ; rebuild env (Note note e') cont }
890 %************************************************************************
892 \subsection{Dealing with calls}
894 %************************************************************************
897 simplVar env var cont
898 = case substId env var of
899 DoneEx e -> simplExprF (zapSubstEnv env) e cont
900 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
901 DoneId var1 -> completeCall (zapSubstEnv env) var1 cont
902 -- Note [zapSubstEnv]
903 -- The template is already simplified, so don't re-substitute.
904 -- This is VITAL. Consider
906 -- let y = \z -> ...x... in
908 -- We'll clone the inner \x, adding x->x' in the id_subst
909 -- Then when we inline y, we must *not* replace x by x' in
910 -- the inlined copy!!
912 ---------------------------------------------------------
913 -- Dealing with a call site
915 completeCall env var cont
916 = do { dflags <- getDOptsSmpl
917 ; let (args,call_cont) = contArgs cont
918 -- The args are OutExprs, obtained by *lazily* substituting
919 -- in the args found in cont. These args are only examined
920 -- to limited depth (unless a rule fires). But we must do
921 -- the substitution; rule matching on un-simplified args would
924 ------------- First try rules ----------------
925 -- Do this before trying inlining. Some functions have
926 -- rules *and* are strict; in this case, we don't want to
927 -- inline the wrapper of the non-specialised thing; better
928 -- to call the specialised thing instead.
930 -- We used to use the black-listing mechanism to ensure that inlining of
931 -- the wrapper didn't occur for things that have specialisations till a
932 -- later phase, so but now we just try RULES first
934 -- You might think that we shouldn't apply rules for a loop breaker:
935 -- doing so might give rise to an infinite loop, because a RULE is
936 -- rather like an extra equation for the function:
937 -- RULE: f (g x) y = x+y
940 -- But it's too drastic to disable rules for loop breakers.
941 -- Even the foldr/build rule would be disabled, because foldr
942 -- is recursive, and hence a loop breaker:
943 -- foldr k z (build g) = g k z
944 -- So it's up to the programmer: rules can cause divergence
945 ; let in_scope = getInScope env
947 maybe_rule = case activeRule env of
948 Nothing -> Nothing -- No rules apply
949 Just act_fn -> lookupRule act_fn in_scope
951 ; case maybe_rule of {
952 Just (rule, rule_rhs) ->
953 tick (RuleFired (ru_name rule)) `thenSmpl_`
954 (if dopt Opt_D_dump_inlinings dflags then
955 pprTrace "Rule fired" (vcat [
956 text "Rule:" <+> ftext (ru_name rule),
957 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
958 text "After: " <+> pprCoreExpr rule_rhs,
959 text "Cont: " <+> ppr call_cont])
962 simplExprF env rule_rhs (dropArgs (ruleArity rule) cont)
963 -- The ruleArity says how many args the rule consumed
965 ; Nothing -> do -- No rules
967 ------------- Next try inlining ----------------
968 { let arg_infos = [interestingArg arg | arg <- args, isValArg arg]
969 n_val_args = length arg_infos
970 interesting_cont = interestingCallContext (notNull args)
973 active_inline = activeInline env var
974 maybe_inline = callSiteInline dflags active_inline
975 var arg_infos interesting_cont
976 ; case maybe_inline of {
977 Just unfolding -- There is an inlining!
978 -> do { tick (UnfoldingDone var)
979 ; (if dopt Opt_D_dump_inlinings dflags then
980 pprTrace "Inlining done" (vcat [
981 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
982 text "Inlined fn: " <+> nest 2 (ppr unfolding),
983 text "Cont: " <+> ppr call_cont])
986 simplExprF env unfolding cont }
988 ; Nothing -> -- No inlining!
990 ------------- No inlining! ----------------
991 -- Next, look for rules or specialisations that match
993 rebuildCall env (Var var) (idType var)
994 (mkArgInfo var n_val_args call_cont) cont
997 rebuildCall :: SimplEnv
998 -> OutExpr -> OutType -- Function and its type
999 -> (Bool, [Bool]) -- See SimplUtils.mkArgInfo
1001 -> SimplM (SimplEnv, OutExpr)
1002 rebuildCall env fun fun_ty (has_rules, []) cont
1003 -- When we run out of strictness args, it means
1004 -- that the call is definitely bottom; see SimplUtils.mkArgInfo
1005 -- Then we want to discard the entire strict continuation. E.g.
1006 -- * case (error "hello") of { ... }
1007 -- * (error "Hello") arg
1008 -- * f (error "Hello") where f is strict
1010 -- Then, especially in the first of these cases, we'd like to discard
1011 -- the continuation, leaving just the bottoming expression. But the
1012 -- type might not be right, so we may have to add a coerce.
1013 | not (contIsTrivial cont) -- Only do thia if there is a non-trivial
1014 = return (env, mk_coerce fun) -- contination to discard, else we do it
1015 where -- again and again!
1016 cont_ty = contResultType cont
1017 co = mkUnsafeCoercion fun_ty cont_ty
1018 mk_coerce expr | cont_ty `coreEqType` fun_ty = fun
1019 | otherwise = mkCoerce co fun
1021 rebuildCall env fun fun_ty info (ApplyTo _ (Type arg_ty) se cont)
1022 = do { ty' <- simplType (se `setInScope` env) arg_ty
1023 ; rebuildCall env (fun `App` Type ty') (applyTy fun_ty ty') info cont }
1025 rebuildCall env fun fun_ty (has_rules, str:strs) (ApplyTo _ arg arg_se cont)
1026 | str || isStrictType arg_ty -- Strict argument
1027 = -- pprTrace "Strict Arg" (ppr arg $$ ppr (seIdSubst env) $$ ppr (seInScope env)) $
1028 simplExprF (arg_se `setFloats` env) arg
1029 (StrictArg fun fun_ty (has_rules, strs) cont)
1032 | otherwise -- Lazy argument
1033 -- DO NOT float anything outside, hence simplExprC
1034 -- There is no benefit (unlike in a let-binding), and we'd
1035 -- have to be very careful about bogus strictness through
1036 -- floating a demanded let.
1037 = do { arg' <- simplExprC (arg_se `setInScope` env) arg
1038 (mkLazyArgStop arg_ty has_rules)
1039 ; rebuildCall env (fun `App` arg') res_ty (has_rules, strs) cont }
1041 (arg_ty, res_ty) = splitFunTy fun_ty
1043 rebuildCall env fun fun_ty info cont
1044 = rebuild env fun cont
1049 This part of the simplifier may break the no-shadowing invariant
1051 f (...(\a -> e)...) (case y of (a,b) -> e')
1052 where f is strict in its second arg
1053 If we simplify the innermost one first we get (...(\a -> e)...)
1054 Simplifying the second arg makes us float the case out, so we end up with
1055 case y of (a,b) -> f (...(\a -> e)...) e'
1056 So the output does not have the no-shadowing invariant. However, there is
1057 no danger of getting name-capture, because when the first arg was simplified
1058 we used an in-scope set that at least mentioned all the variables free in its
1059 static environment, and that is enough.
1061 We can't just do innermost first, or we'd end up with a dual problem:
1062 case x of (a,b) -> f e (...(\a -> e')...)
1064 I spent hours trying to recover the no-shadowing invariant, but I just could
1065 not think of an elegant way to do it. The simplifier is already knee-deep in
1066 continuations. We have to keep the right in-scope set around; AND we have
1067 to get the effect that finding (error "foo") in a strict arg position will
1068 discard the entire application and replace it with (error "foo"). Getting
1069 all this at once is TOO HARD!
1071 %************************************************************************
1073 Rebuilding a cse expression
1075 %************************************************************************
1077 Blob of helper functions for the "case-of-something-else" situation.
1080 ---------------------------------------------------------
1081 -- Eliminate the case if possible
1083 rebuildCase :: SimplEnv
1084 -> OutExpr -- Scrutinee
1085 -> InId -- Case binder
1086 -> [InAlt] -- Alternatives (inceasing order)
1088 -> SimplM (SimplEnv, OutExpr)
1090 rebuildCase env scrut case_bndr alts cont
1091 | Just (con,args) <- exprIsConApp_maybe scrut
1092 -- Works when the scrutinee is a variable with a known unfolding
1093 -- as well as when it's an explicit constructor application
1094 = knownCon env scrut (DataAlt con) args case_bndr alts cont
1096 | Lit lit <- scrut -- No need for same treatment as constructors
1097 -- because literals are inlined more vigorously
1098 = knownCon env scrut (LitAlt lit) [] case_bndr alts cont
1101 = do { -- Prepare the continuation;
1102 -- The new subst_env is in place
1103 (env, dup_cont, nodup_cont) <- prepareCaseCont env alts cont
1105 -- Simplify the alternatives
1106 ; (case_bndr', alts') <- simplAlts env scrut case_bndr alts dup_cont
1107 ; let res_ty' = contResultType dup_cont
1108 ; case_expr <- mkCase scrut case_bndr' res_ty' alts'
1110 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1111 -- The case binder *not* scope over the whole returned case-expression
1112 ; rebuild env case_expr nodup_cont }
1115 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1116 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1117 way, there's a chance that v will now only be used once, and hence
1120 Note [no-case-of-case]
1121 ~~~~~~~~~~~~~~~~~~~~~~
1122 There is a time we *don't* want to do that, namely when
1123 -fno-case-of-case is on. This happens in the first simplifier pass,
1124 and enhances full laziness. Here's the bad case:
1125 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1126 If we eliminate the inner case, we trap it inside the I# v -> arm,
1127 which might prevent some full laziness happening. I've seen this
1128 in action in spectral/cichelli/Prog.hs:
1129 [(m,n) | m <- [1..max], n <- [1..max]]
1130 Hence the check for NoCaseOfCase.
1134 Consider case (v `cast` co) of x { I# ->
1135 ... (case (v `cast` co) of {...}) ...
1136 We'd like to eliminate the inner case. We can get this neatly by
1137 arranging that inside the outer case we add the unfolding
1138 v |-> x `cast` (sym co)
1139 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1143 There is another situation when we don't want to do it. If we have
1145 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1146 ...other cases .... }
1148 We'll perform the binder-swap for the outer case, giving
1150 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1151 ...other cases .... }
1153 But there is no point in doing it for the inner case, because w1 can't
1154 be inlined anyway. Furthermore, doing the case-swapping involves
1155 zapping w2's occurrence info (see paragraphs that follow), and that
1156 forces us to bind w2 when doing case merging. So we get
1158 case x of w1 { A -> let w2 = w1 in e1
1159 B -> let w2 = w1 in e2
1160 ...other cases .... }
1162 This is plain silly in the common case where w2 is dead.
1164 Even so, I can't see a good way to implement this idea. I tried
1165 not doing the binder-swap if the scrutinee was already evaluated
1166 but that failed big-time:
1170 case v of w { MkT x ->
1171 case x of x1 { I# y1 ->
1172 case x of x2 { I# y2 -> ...
1174 Notice that because MkT is strict, x is marked "evaluated". But to
1175 eliminate the last case, we must either make sure that x (as well as
1176 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1177 the binder-swap. So this whole note is a no-op.
1181 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1182 any occurrence info (eg IAmDead) in the case binder, because the
1183 case-binder now effectively occurs whenever v does. AND we have to do
1184 the same for the pattern-bound variables! Example:
1186 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1188 Here, b and p are dead. But when we move the argment inside the first
1189 case RHS, and eliminate the second case, we get
1191 case x of { (a,b) -> a b }
1193 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1196 Indeed, this can happen anytime the case binder isn't dead:
1197 case <any> of x { (a,b) ->
1198 case x of { (p,q) -> p } }
1199 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1200 The point is that we bring into the envt a binding
1202 after the outer case, and that makes (a,b) alive. At least we do unless
1203 the case binder is guaranteed dead.
1206 simplCaseBinder :: SimplEnv -> OutExpr -> InId -> SimplM (SimplEnv, OutId)
1207 simplCaseBinder env scrut case_bndr
1208 | switchIsOn (getSwitchChecker env) NoCaseOfCase
1209 -- See Note [no-case-of-case]
1210 = do { (env, case_bndr') <- simplBinder env case_bndr
1211 ; return (env, case_bndr') }
1213 simplCaseBinder env (Var v) case_bndr
1214 -- Failed try [see Note 2 above]
1215 -- not (isEvaldUnfolding (idUnfolding v))
1216 = do { (env, case_bndr') <- simplBinder env (zapOccInfo case_bndr)
1217 ; return (modifyInScope env v case_bndr', case_bndr') }
1218 -- We could extend the substitution instead, but it would be
1219 -- a hack because then the substitution wouldn't be idempotent
1220 -- any more (v is an OutId). And this does just as well.
1222 simplCaseBinder env (Cast (Var v) co) case_bndr -- Note [Case of cast]
1223 = do { (env, case_bndr') <- simplBinder env (zapOccInfo case_bndr)
1224 ; let rhs = Cast (Var case_bndr') (mkSymCoercion co)
1225 ; return (addBinderUnfolding env v rhs, case_bndr') }
1227 simplCaseBinder env other_scrut case_bndr
1228 = do { (env, case_bndr') <- simplBinder env case_bndr
1229 ; return (env, case_bndr') }
1231 zapOccInfo :: InId -> InId -- See Note [zapOccInfo]
1232 zapOccInfo b = b `setIdOccInfo` NoOccInfo
1236 simplAlts does two things:
1238 1. Eliminate alternatives that cannot match, including the
1239 DEFAULT alternative.
1241 2. If the DEFAULT alternative can match only one possible constructor,
1242 then make that constructor explicit.
1244 case e of x { DEFAULT -> rhs }
1246 case e of x { (a,b) -> rhs }
1247 where the type is a single constructor type. This gives better code
1248 when rhs also scrutinises x or e.
1250 Here "cannot match" includes knowledge from GADTs
1252 It's a good idea do do this stuff before simplifying the alternatives, to
1253 avoid simplifying alternatives we know can't happen, and to come up with
1254 the list of constructors that are handled, to put into the IdInfo of the
1255 case binder, for use when simplifying the alternatives.
1257 Eliminating the default alternative in (1) isn't so obvious, but it can
1260 data Colour = Red | Green | Blue
1269 DEFAULT -> [ case y of ... ]
1271 If we inline h into f, the default case of the inlined h can't happen.
1272 If we don't notice this, we may end up filtering out *all* the cases
1273 of the inner case y, which give us nowhere to go!
1277 simplAlts :: SimplEnv
1279 -> InId -- Case binder
1280 -> [InAlt] -> SimplCont
1281 -> SimplM (OutId, [OutAlt]) -- Includes the continuation
1282 -- Like simplExpr, this just returns the simplified alternatives;
1283 -- it not return an environment
1285 simplAlts env scrut case_bndr alts cont'
1286 = -- pprTrace "simplAlts" (ppr alts $$ ppr (seIdSubst env)) $
1287 do { let alt_env = zapFloats env
1288 ; (alt_env, case_bndr') <- simplCaseBinder alt_env scrut case_bndr
1290 ; default_alts <- prepareDefault alt_env case_bndr' imposs_deflt_cons cont' maybe_deflt
1292 ; let inst_tys = tyConAppArgs (idType case_bndr')
1293 trimmed_alts = filter (is_possible inst_tys) alts_wo_default
1294 in_alts = mergeAlts default_alts trimmed_alts
1295 -- We need the mergeAlts in case the new default_alt
1296 -- has turned into a constructor alternative.
1298 ; alts' <- mapM (simplAlt alt_env imposs_cons case_bndr' cont') in_alts
1299 ; return (case_bndr', alts') }
1301 (alts_wo_default, maybe_deflt) = findDefault alts
1302 imposs_cons = case scrut of
1303 Var v -> otherCons (idUnfolding v)
1306 -- "imposs_deflt_cons" are handled either by the context,
1307 -- OR by a branch in this case expression. (Don't include DEFAULT!!)
1308 imposs_deflt_cons = nub (imposs_cons ++ [con | (con,_,_) <- alts_wo_default])
1310 is_possible :: [Type] -> CoreAlt -> Bool
1311 is_possible tys (con, _, _) | con `elem` imposs_cons = False
1312 is_possible tys (DataAlt con, _, _) = dataConCanMatch tys con
1313 is_possible tys alt = True
1315 ------------------------------------
1316 prepareDefault :: SimplEnv
1317 -> OutId -- Case binder; need just for its type. Note that as an
1318 -- OutId, it has maximum information; this is important.
1319 -- Test simpl013 is an example
1320 -> [AltCon] -- These cons can't happen when matching the default
1323 -> SimplM [InAlt] -- One branch or none; still unsimplified
1324 -- We use a list because it's what mergeAlts expects
1326 prepareDefault env case_bndr' imposs_cons cont Nothing
1327 = return [] -- No default branch
1329 prepareDefault env case_bndr' imposs_cons cont (Just rhs)
1330 | -- This branch handles the case where we are
1331 -- scrutinisng an algebraic data type
1332 Just (tycon, inst_tys) <- splitTyConApp_maybe (idType case_bndr'),
1333 isAlgTyCon tycon, -- It's a data type, tuple, or unboxed tuples.
1334 not (isNewTyCon tycon), -- We can have a newtype, if we are just doing an eval:
1335 -- case x of { DEFAULT -> e }
1336 -- and we don't want to fill in a default for them!
1337 Just all_cons <- tyConDataCons_maybe tycon,
1338 not (null all_cons), -- This is a tricky corner case. If the data type has no constructors,
1339 -- which GHC allows, then the case expression will have at most a default
1340 -- alternative. We don't want to eliminate that alternative, because the
1341 -- invariant is that there's always one alternative. It's more convenient
1343 -- case x of { DEFAULT -> e }
1344 -- as it is, rather than transform it to
1345 -- error "case cant match"
1346 -- which would be quite legitmate. But it's a really obscure corner, and
1347 -- not worth wasting code on.
1349 let imposs_data_cons = [con | DataAlt con <- imposs_cons] -- We now know it's a data type
1350 is_possible con = not (con `elem` imposs_data_cons)
1351 && dataConCanMatch inst_tys con
1352 = case filter is_possible all_cons of
1353 [] -> return [] -- Eliminate the default alternative
1354 -- altogether if it can't match
1356 [con] -> -- It matches exactly one constructor, so fill it in
1357 do { tick (FillInCaseDefault case_bndr')
1358 ; us <- getUniquesSmpl
1359 ; let (ex_tvs, co_tvs, arg_ids) =
1360 dataConRepInstPat us con inst_tys
1361 ; return [(DataAlt con, ex_tvs ++ co_tvs ++ arg_ids, rhs)] }
1363 two_or_more -> return [(DEFAULT, [], rhs)]
1366 = return [(DEFAULT, [], rhs)]
1368 ------------------------------------
1369 simplAlt :: SimplEnv
1370 -> [AltCon] -- These constructors can't be present when
1371 -- matching this alternative
1372 -> OutId -- The case binder
1377 -- Simplify an alternative, returning the type refinement for the
1378 -- alternative, if the alternative does any refinement at all
1380 simplAlt env handled_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1381 = ASSERT( null bndrs )
1382 do { let env' = addBinderOtherCon env case_bndr' handled_cons
1383 -- Record the constructors that the case-binder *can't* be.
1384 ; rhs' <- simplExprC env' rhs cont'
1385 ; return (DEFAULT, [], rhs') }
1387 simplAlt env handled_cons case_bndr' cont' (LitAlt lit, bndrs, rhs)
1388 = ASSERT( null bndrs )
1389 do { let env' = addBinderUnfolding env case_bndr' (Lit lit)
1390 ; rhs' <- simplExprC env' rhs cont'
1391 ; return (LitAlt lit, [], rhs') }
1393 simplAlt env handled_cons case_bndr' cont' (DataAlt con, vs, rhs)
1394 = do { -- Deal with the pattern-bound variables
1395 -- Mark the ones that are in ! positions in the data constructor
1396 -- as certainly-evaluated.
1397 -- NB: it happens that simplBinders does *not* erase the OtherCon
1398 -- form of unfolding, so it's ok to add this info before
1399 -- doing simplBinders
1400 (env, vs') <- simplBinders env (add_evals con vs)
1402 -- Bind the case-binder to (con args)
1403 ; let inst_tys' = tyConAppArgs (idType case_bndr')
1404 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1405 env' = addBinderUnfolding env case_bndr' (mkConApp con con_args)
1407 ; rhs' <- simplExprC env' rhs cont'
1408 ; return (DataAlt con, vs', rhs') }
1410 -- add_evals records the evaluated-ness of the bound variables of
1411 -- a case pattern. This is *important*. Consider
1412 -- data T = T !Int !Int
1414 -- case x of { T a b -> T (a+1) b }
1416 -- We really must record that b is already evaluated so that we don't
1417 -- go and re-evaluate it when constructing the result.
1418 -- See Note [Data-con worker strictness] in MkId.lhs
1419 add_evals dc vs = cat_evals dc vs (dataConRepStrictness dc)
1421 cat_evals dc vs strs
1425 go (v:vs) strs | isTyVar v = v : go vs strs
1426 go (v:vs) (str:strs)
1427 | isMarkedStrict str = evald_v : go vs strs
1428 | otherwise = zapped_v : go vs strs
1430 zapped_v = zap_occ_info v
1431 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1432 go _ _ = pprPanic "cat_evals" (ppr dc $$ ppr vs $$ ppr strs)
1434 -- If the case binder is alive, then we add the unfolding
1436 -- to the envt; so vs are now very much alive
1437 -- Note [Aug06] I can't see why this actually matters
1438 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1439 | otherwise = zapOccInfo
1441 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1442 addBinderUnfolding env bndr rhs
1443 = modifyInScope env bndr (bndr `setIdUnfolding` mkUnfolding False rhs)
1445 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1446 addBinderOtherCon env bndr cons
1447 = modifyInScope env bndr (bndr `setIdUnfolding` mkOtherCon cons)
1451 %************************************************************************
1453 \subsection{Known constructor}
1455 %************************************************************************
1457 We are a bit careful with occurrence info. Here's an example
1459 (\x* -> case x of (a*, b) -> f a) (h v, e)
1461 where the * means "occurs once". This effectively becomes
1462 case (h v, e) of (a*, b) -> f a)
1464 let a* = h v; b = e in f a
1468 All this should happen in one sweep.
1471 knownCon :: SimplEnv -> OutExpr -> AltCon -> [OutExpr]
1472 -> InId -> [InAlt] -> SimplCont
1473 -> SimplM (SimplEnv, OutExpr)
1475 knownCon env scrut con args bndr alts cont
1476 = do { tick (KnownBranch bndr)
1477 ; knownAlt env scrut args bndr (findAlt con alts) cont }
1479 knownAlt env scrut args bndr (DEFAULT, bs, rhs) cont
1481 do { env <- simplNonRecX env bndr scrut
1482 -- This might give rise to a binding with non-atomic args
1483 -- like x = Node (f x) (g x)
1484 -- but simplNonRecX will atomic-ify it
1485 ; simplExprF env rhs cont }
1487 knownAlt env scrut args bndr (LitAlt lit, bs, rhs) cont
1489 do { env <- simplNonRecX env bndr scrut
1490 ; simplExprF env rhs cont }
1492 knownAlt env scrut args bndr (DataAlt dc, bs, rhs) cont
1493 = do { let dead_bndr = isDeadBinder bndr
1494 n_drop_tys = tyConArity (dataConTyCon dc)
1495 ; env <- bind_args env dead_bndr bs (drop n_drop_tys args)
1497 -- It's useful to bind bndr to scrut, rather than to a fresh
1498 -- binding x = Con arg1 .. argn
1499 -- because very often the scrut is a variable, so we avoid
1500 -- creating, and then subsequently eliminating, a let-binding
1501 -- BUT, if scrut is a not a variable, we must be careful
1502 -- about duplicating the arg redexes; in that case, make
1503 -- a new con-app from the args
1504 bndr_rhs = case scrut of
1507 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1508 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1509 -- args are aready OutExprs, but bs are InIds
1511 ; env <- simplNonRecX env bndr bndr_rhs
1512 ; -- pprTrace "knownCon2" (ppr bs $$ ppr rhs $$ ppr (seIdSubst env)) $
1513 simplExprF env rhs cont }
1516 bind_args env dead_bndr [] _ = return env
1518 bind_args env dead_bndr (b:bs) (Type ty : args)
1519 = ASSERT( isTyVar b )
1520 bind_args (extendTvSubst env b ty) dead_bndr bs args
1522 bind_args env dead_bndr (b:bs) (arg : args)
1524 do { let b' = if dead_bndr then b else zapOccInfo b
1525 -- Note that the binder might be "dead", because it doesn't occur
1526 -- in the RHS; and simplNonRecX may therefore discard it via postInlineUnconditionally
1527 -- Nevertheless we must keep it if the case-binder is alive, because it may
1528 -- be used in the con_app. See Note [zapOccInfo]
1529 ; env <- simplNonRecX env b' arg
1530 ; bind_args env dead_bndr bs args }
1532 bind_args _ _ _ _ = panic "bind_args"
1536 %************************************************************************
1538 \subsection{Duplicating continuations}
1540 %************************************************************************
1543 prepareCaseCont :: SimplEnv
1544 -> [InAlt] -> SimplCont
1545 -> SimplM (SimplEnv, SimplCont,SimplCont)
1546 -- Return a duplicatable continuation, a non-duplicable part
1547 -- plus some extra bindings (that scope over the entire
1550 -- No need to make it duplicatable if there's only one alternative
1551 prepareCaseCont env [alt] cont = return (env, cont, mkBoringStop (contResultType cont))
1552 prepareCaseCont env alts cont = mkDupableCont env cont
1556 mkDupableCont :: SimplEnv -> SimplCont
1557 -> SimplM (SimplEnv, SimplCont, SimplCont)
1559 mkDupableCont env cont
1560 | contIsDupable cont
1561 = returnSmpl (env, cont, mkBoringStop (contResultType cont))
1563 mkDupableCont env (Stop {}) = panic "mkDupableCont" -- Handled by previous eqn
1565 mkDupableCont env (CoerceIt ty cont)
1566 = do { (env, dup, nodup) <- mkDupableCont env cont
1567 ; return (env, CoerceIt ty dup, nodup) }
1569 mkDupableCont env cont@(StrictBind bndr _ _ se _)
1570 = return (env, mkBoringStop (substTy se (idType bndr)), cont)
1571 -- See Note [Duplicating strict continuations]
1573 mkDupableCont env cont@(StrictArg _ fun_ty _ _)
1574 = return (env, mkBoringStop (funArgTy fun_ty), cont)
1575 -- See Note [Duplicating strict continuations]
1577 mkDupableCont env (ApplyTo _ arg se cont)
1578 = -- e.g. [...hole...] (...arg...)
1580 -- let a = ...arg...
1581 -- in [...hole...] a
1582 do { (env, dup_cont, nodup_cont) <- mkDupableCont env cont
1583 ; arg <- simplExpr (se `setInScope` env) arg
1584 ; (env, arg) <- makeTrivial env arg
1585 ; let app_cont = ApplyTo OkToDup arg (zapSubstEnv env) dup_cont
1586 ; return (env, app_cont, nodup_cont) }
1588 mkDupableCont env cont@(Select _ case_bndr [(_,bs,rhs)] se case_cont)
1589 -- See Note [Single-alternative case]
1590 -- | not (exprIsDupable rhs && contIsDupable case_cont)
1591 -- | not (isDeadBinder case_bndr)
1592 | all isDeadBinder bs
1593 = return (env, mkBoringStop scrut_ty, cont)
1595 scrut_ty = substTy se (idType case_bndr)
1597 mkDupableCont env (Select _ case_bndr alts se cont)
1598 = -- e.g. (case [...hole...] of { pi -> ei })
1600 -- let ji = \xij -> ei
1601 -- in case [...hole...] of { pi -> ji xij }
1602 do { tick (CaseOfCase case_bndr)
1603 ; (env, dup_cont, nodup_cont) <- mkDupableCont env cont
1604 -- NB: call mkDupableCont here, *not* prepareCaseCont
1605 -- We must make a duplicable continuation, whereas prepareCaseCont
1606 -- doesn't when there is a single case branch
1608 ; let alt_env = se `setInScope` env
1609 ; (alt_env, case_bndr') <- simplBinder alt_env case_bndr
1610 ; alts' <- mapM (simplAlt alt_env [] case_bndr' dup_cont) alts
1611 -- Safe to say that there are no handled-cons for the DEFAULT case
1612 -- NB: simplBinder does not zap deadness occ-info, so
1613 -- a dead case_bndr' will still advertise its deadness
1614 -- This is really important because in
1615 -- case e of b { (# a,b #) -> ... }
1616 -- b is always dead, and indeed we are not allowed to bind b to (# a,b #),
1617 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1618 -- In the new alts we build, we have the new case binder, so it must retain
1620 -- NB: we don't use alt_env further; it has the substEnv for
1621 -- the alternatives, and we don't want that
1623 ; (env, alts') <- mkDupableAlts env case_bndr' alts'
1624 ; return (env, -- Note [Duplicated env]
1625 Select OkToDup case_bndr' alts' (zapSubstEnv env)
1626 (mkBoringStop (contResultType dup_cont)),
1630 mkDupableAlts :: SimplEnv -> OutId -> [InAlt]
1631 -> SimplM (SimplEnv, [InAlt])
1632 -- Absorbs the continuation into the new alternatives
1634 mkDupableAlts env case_bndr' alts
1637 go env [] = return (env, [])
1639 = do { (env, alt') <- mkDupableAlt env case_bndr' alt
1640 ; (env, alts') <- go env alts
1641 ; return (env, alt' : alts' ) }
1643 mkDupableAlt env case_bndr' (con, bndrs', rhs')
1644 | exprIsDupable rhs' -- Note [Small alternative rhs]
1645 = return (env, (con, bndrs', rhs'))
1647 = do { let rhs_ty' = exprType rhs'
1648 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1650 | isTyVar bndr = True -- Abstract over all type variables just in case
1651 | otherwise = not (isDeadBinder bndr)
1652 -- The deadness info on the new Ids is preserved by simplBinders
1654 ; (final_bndrs', final_args) -- Note [Join point abstraction]
1655 <- if (any isId used_bndrs')
1656 then return (used_bndrs', varsToCoreExprs used_bndrs')
1657 else do { rw_id <- newId FSLIT("w") realWorldStatePrimTy
1658 ; return ([rw_id], [Var realWorldPrimId]) }
1660 ; join_bndr <- newId FSLIT("$j") (mkPiTypes final_bndrs' rhs_ty')
1661 -- Note [Funky mkPiTypes]
1663 ; let -- We make the lambdas into one-shot-lambdas. The
1664 -- join point is sure to be applied at most once, and doing so
1665 -- prevents the body of the join point being floated out by
1666 -- the full laziness pass
1667 really_final_bndrs = map one_shot final_bndrs'
1668 one_shot v | isId v = setOneShotLambda v
1670 join_rhs = mkLams really_final_bndrs rhs'
1671 join_call = mkApps (Var join_bndr) final_args
1673 ; return (addNonRec env join_bndr join_rhs, (con, bndrs', join_call)) }
1674 -- See Note [Duplicated env]
1677 Note [Duplicated env]
1678 ~~~~~~~~~~~~~~~~~~~~~
1679 Some of the alternatives are simplified, but have not been turned into a join point
1680 So they *must* have an zapped subst-env. So we can't use completeNonRecX to
1681 bind the join point, because it might to do PostInlineUnconditionally, and
1682 we'd lose that when zapping the subst-env. We could have a per-alt subst-env,
1683 but zapping it (as we do in mkDupableCont, the Select case) is safe, and
1684 at worst delays the join-point inlining.
1686 Note [Small alterantive rhs]
1687 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1688 It is worth checking for a small RHS because otherwise we
1689 get extra let bindings that may cause an extra iteration of the simplifier to
1690 inline back in place. Quite often the rhs is just a variable or constructor.
1691 The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1692 iterations because the version with the let bindings looked big, and so wasn't
1693 inlined, but after the join points had been inlined it looked smaller, and so
1696 NB: we have to check the size of rhs', not rhs.
1697 Duplicating a small InAlt might invalidate occurrence information
1698 However, if it *is* dupable, we return the *un* simplified alternative,
1699 because otherwise we'd need to pair it up with an empty subst-env....
1700 but we only have one env shared between all the alts.
1701 (Remember we must zap the subst-env before re-simplifying something).
1702 Rather than do this we simply agree to re-simplify the original (small) thing later.
1704 Note [Funky mkPiTypes]
1705 ~~~~~~~~~~~~~~~~~~~~~~
1706 Notice the funky mkPiTypes. If the contructor has existentials
1707 it's possible that the join point will be abstracted over
1708 type varaibles as well as term variables.
1709 Example: Suppose we have
1710 data T = forall t. C [t]
1712 case (case e of ...) of
1714 We get the join point
1715 let j :: forall t. [t] -> ...
1716 j = /\t \xs::[t] -> rhs
1718 case (case e of ...) of
1719 C t xs::[t] -> j t xs
1721 Note [Join point abstaction]
1722 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1723 If we try to lift a primitive-typed something out
1724 for let-binding-purposes, we will *caseify* it (!),
1725 with potentially-disastrous strictness results. So
1726 instead we turn it into a function: \v -> e
1727 where v::State# RealWorld#. The value passed to this function
1728 is realworld#, which generates (almost) no code.
1730 There's a slight infelicity here: we pass the overall
1731 case_bndr to all the join points if it's used in *any* RHS,
1732 because we don't know its usage in each RHS separately
1734 We used to say "&& isUnLiftedType rhs_ty'" here, but now
1735 we make the join point into a function whenever used_bndrs'
1736 is empty. This makes the join-point more CPR friendly.
1737 Consider: let j = if .. then I# 3 else I# 4
1738 in case .. of { A -> j; B -> j; C -> ... }
1740 Now CPR doesn't w/w j because it's a thunk, so
1741 that means that the enclosing function can't w/w either,
1742 which is a lose. Here's the example that happened in practice:
1743 kgmod :: Int -> Int -> Int
1744 kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1748 I have seen a case alternative like this:
1750 It's a bit silly to add the realWorld dummy arg in this case, making
1753 (the \v alone is enough to make CPR happy) but I think it's rare
1755 Note [Duplicating strict continuations]
1756 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1757 Do *not* duplicate StrictBind and StritArg continuations. We gain
1758 nothing by propagating them into the expressions, and we do lose a
1759 lot. Here's an example:
1760 && (case x of { T -> F; F -> T }) E
1761 Now, && is strict so we end up simplifying the case with
1762 an ArgOf continuation. If we let-bind it, we get
1764 let $j = \v -> && v E
1765 in simplExpr (case x of { T -> F; F -> T })
1767 And after simplifying more we get
1769 let $j = \v -> && v E
1770 in case x of { T -> $j F; F -> $j T }
1771 Which is a Very Bad Thing
1773 The desire not to duplicate is the entire reason that
1774 mkDupableCont returns a pair of continuations.
1776 The original plan had:
1777 e.g. (...strict-fn...) [...hole...]
1779 let $j = \a -> ...strict-fn...
1782 Note [Single-alternative cases]
1783 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1784 This case is just like the ArgOf case. Here's an example:
1788 case (case x of I# x' ->
1790 True -> I# (negate# x')
1791 False -> I# x') of y {
1793 Because the (case x) has only one alternative, we'll transform to
1795 case (case x' <# 0# of
1796 True -> I# (negate# x')
1797 False -> I# x') of y {
1799 But now we do *NOT* want to make a join point etc, giving
1801 let $j = \y -> MkT y
1803 True -> $j (I# (negate# x'))
1805 In this case the $j will inline again, but suppose there was a big
1806 strict computation enclosing the orginal call to MkT. Then, it won't
1807 "see" the MkT any more, because it's big and won't get duplicated.
1808 And, what is worse, nothing was gained by the case-of-case transform.
1810 When should use this case of mkDupableCont?
1811 However, matching on *any* single-alternative case is a *disaster*;
1812 e.g. case (case ....) of (a,b) -> (# a,b #)
1813 We must push the outer case into the inner one!
1816 * Match [(DEFAULT,_,_)], but in the common case of Int,
1817 the alternative-filling-in code turned the outer case into
1818 case (...) of y { I# _ -> MkT y }
1820 * Match on single alternative plus (not (isDeadBinder case_bndr))
1821 Rationale: pushing the case inwards won't eliminate the construction.
1822 But there's a risk of
1823 case (...) of y { (a,b) -> let z=(a,b) in ... }
1824 Now y looks dead, but it'll come alive again. Still, this
1825 seems like the best option at the moment.
1827 * Match on single alternative plus (all (isDeadBinder bndrs))
1828 Rationale: this is essentially seq.
1830 * Match when the rhs is *not* duplicable, and hence would lead to a
1831 join point. This catches the disaster-case above. We can test
1832 the *un-simplified* rhs, which is fine. It might get bigger or
1833 smaller after simplification; if it gets smaller, this case might
1834 fire next time round. NB also that we must test contIsDupable
1835 case_cont *btoo, because case_cont might be big!
1837 HOWEVER: I found that this version doesn't work well, because
1838 we can get let x = case (...) of { small } in ...case x...
1839 When x is inlined into its full context, we find that it was a bad
1840 idea to have pushed the outer case inside the (...) case.