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),
16 import SimplUtils ( mkCase, mkLam, prepareAlts,
17 SimplCont(..), DupFlag(..), LetRhsFlag(..),
18 mkRhsStop, mkBoringStop, pushContArgs,
19 contResultType, countArgs, contIsDupable, contIsRhsOrArg,
20 getContArgs, interestingCallContext, interestingArg, isStrictType,
21 preInlineUnconditionally, postInlineUnconditionally,
22 inlineMode, activeInline, activeRule
24 import Id ( Id, idType, idInfo, idArity, isDataConWorkId,
25 setIdUnfolding, isDeadBinder,
26 idNewDemandInfo, setIdInfo,
27 setIdOccInfo, zapLamIdInfo, setOneShotLambda
29 import MkId ( eRROR_ID )
30 import Literal ( mkStringLit )
31 import IdInfo ( OccInfo(..), isLoopBreaker,
32 setArityInfo, zapDemandInfo,
36 import NewDemand ( isStrictDmd )
37 import Unify ( coreRefineTys )
38 import DataCon ( dataConTyCon, dataConRepStrictness, isVanillaDataCon )
39 import TyCon ( tyConArity )
41 import PprCore ( pprParendExpr, pprCoreExpr )
42 import CoreUnfold ( mkUnfolding, callSiteInline )
43 import CoreUtils ( exprIsDupable, exprIsTrivial, needsCaseBinding,
44 exprIsConApp_maybe, mkPiTypes, findAlt,
46 exprOkForSpeculation, exprArity,
47 mkCoerce, mkCoerce2, mkSCC, mkInlineMe, applyTypeToArg
49 import Rules ( lookupRule )
50 import BasicTypes ( isMarkedStrict )
51 import CostCentre ( currentCCS )
52 import Type ( TvSubstEnv, isUnLiftedType, seqType, tyConAppArgs, funArgTy,
53 splitFunTy_maybe, splitFunTy, coreEqType
55 import VarEnv ( elemVarEnv, emptyVarEnv )
56 import TysPrim ( realWorldStatePrimTy )
57 import PrelInfo ( realWorldPrimId )
58 import BasicTypes ( TopLevelFlag(..), isTopLevel,
61 import StaticFlags ( opt_PprStyle_Debug )
63 import Maybes ( orElse )
65 import Util ( notNull )
69 The guts of the simplifier is in this module, but the driver loop for
70 the simplifier is in SimplCore.lhs.
73 -----------------------------------------
74 *** IMPORTANT NOTE ***
75 -----------------------------------------
76 The simplifier used to guarantee that the output had no shadowing, but
77 it does not do so any more. (Actually, it never did!) The reason is
78 documented with simplifyArgs.
81 -----------------------------------------
82 *** IMPORTANT NOTE ***
83 -----------------------------------------
84 Many parts of the simplifier return a bunch of "floats" as well as an
85 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
87 All "floats" are let-binds, not case-binds, but some non-rec lets may
88 be unlifted (with RHS ok-for-speculation).
92 -----------------------------------------
93 ORGANISATION OF FUNCTIONS
94 -----------------------------------------
96 - simplify all top-level binders
97 - for NonRec, call simplRecOrTopPair
98 - for Rec, call simplRecBind
101 ------------------------------
102 simplExpr (applied lambda) ==> simplNonRecBind
103 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
104 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
106 ------------------------------
107 simplRecBind [binders already simplfied]
108 - use simplRecOrTopPair on each pair in turn
110 simplRecOrTopPair [binder already simplified]
111 Used for: recursive bindings (top level and nested)
112 top-level non-recursive bindings
114 - check for PreInlineUnconditionally
118 Used for: non-top-level non-recursive bindings
119 beta reductions (which amount to the same thing)
120 Because it can deal with strict arts, it takes a
121 "thing-inside" and returns an expression
123 - check for PreInlineUnconditionally
124 - simplify binder, including its IdInfo
133 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
134 Used for: binding case-binder and constr args in a known-constructor case
135 - check for PreInLineUnconditionally
139 ------------------------------
140 simplLazyBind: [binder already simplified, RHS not]
141 Used for: recursive bindings (top level and nested)
142 top-level non-recursive bindings
143 non-top-level, but *lazy* non-recursive bindings
144 [must not be strict or unboxed]
145 Returns floats + an augmented environment, not an expression
146 - substituteIdInfo and add result to in-scope
147 [so that rules are available in rec rhs]
150 - float if exposes constructor or PAP
154 completeNonRecX: [binder and rhs both simplified]
155 - if the the thing needs case binding (unlifted and not ok-for-spec)
161 completeLazyBind: [given a simplified RHS]
162 [used for both rec and non-rec bindings, top level and not]
163 - try PostInlineUnconditionally
164 - add unfolding [this is the only place we add an unfolding]
169 Right hand sides and arguments
170 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
171 In many ways we want to treat
172 (a) the right hand side of a let(rec), and
173 (b) a function argument
174 in the same way. But not always! In particular, we would
175 like to leave these arguments exactly as they are, so they
176 will match a RULE more easily.
181 It's harder to make the rule match if we ANF-ise the constructor,
182 or eta-expand the PAP:
184 f (let { a = g x; b = h x } in (a,b))
187 On the other hand if we see the let-defns
192 then we *do* want to ANF-ise and eta-expand, so that p and q
193 can be safely inlined.
195 Even floating lets out is a bit dubious. For let RHS's we float lets
196 out if that exposes a value, so that the value can be inlined more vigorously.
199 r = let x = e in (x,x)
201 Here, if we float the let out we'll expose a nice constructor. We did experiments
202 that showed this to be a generally good thing. But it was a bad thing to float
203 lets out unconditionally, because that meant they got allocated more often.
205 For function arguments, there's less reason to expose a constructor (it won't
206 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
207 So for the moment we don't float lets out of function arguments either.
212 For eta expansion, we want to catch things like
214 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
216 If the \x was on the RHS of a let, we'd eta expand to bring the two
217 lambdas together. And in general that's a good thing to do. Perhaps
218 we should eta expand wherever we find a (value) lambda? Then the eta
219 expansion at a let RHS can concentrate solely on the PAP case.
222 %************************************************************************
224 \subsection{Bindings}
226 %************************************************************************
229 simplTopBinds :: SimplEnv -> [InBind] -> SimplM [OutBind]
231 simplTopBinds env binds
232 = -- Put all the top-level binders into scope at the start
233 -- so that if a transformation rule has unexpectedly brought
234 -- anything into scope, then we don't get a complaint about that.
235 -- It's rather as if the top-level binders were imported.
236 simplLetBndrs env (bindersOfBinds binds) `thenSmpl` \ (env, bndrs') ->
237 simpl_binds env binds bndrs' `thenSmpl` \ (floats, _) ->
238 freeTick SimplifierDone `thenSmpl_`
239 returnSmpl (floatBinds floats)
241 -- We need to track the zapped top-level binders, because
242 -- they should have their fragile IdInfo zapped (notably occurrence info)
243 -- That's why we run down binds and bndrs' simultaneously.
244 simpl_binds :: SimplEnv -> [InBind] -> [OutId] -> SimplM (FloatsWith ())
245 simpl_binds env [] bs = ASSERT( null bs ) returnSmpl (emptyFloats env, ())
246 simpl_binds env (bind:binds) bs = simpl_bind env bind bs `thenSmpl` \ (floats,env) ->
247 addFloats env floats $ \env ->
248 simpl_binds env binds (drop_bs bind bs)
250 drop_bs (NonRec _ _) (_ : bs) = bs
251 drop_bs (Rec prs) bs = drop (length prs) bs
253 simpl_bind env bind bs
254 = getDOptsSmpl `thenSmpl` \ dflags ->
255 if dopt Opt_D_dump_inlinings dflags then
256 pprTrace "SimplBind" (ppr (bindersOf bind)) $ simpl_bind1 env bind bs
258 simpl_bind1 env bind bs
260 simpl_bind1 env (NonRec b r) (b':_) = simplRecOrTopPair env TopLevel b b' r
261 simpl_bind1 env (Rec pairs) bs' = simplRecBind env TopLevel pairs bs'
265 %************************************************************************
267 \subsection{simplNonRec}
269 %************************************************************************
271 simplNonRecBind is used for
272 * non-top-level non-recursive lets in expressions
276 * An unsimplified (binder, rhs) pair
277 * The env for the RHS. It may not be the same as the
278 current env because the bind might occur via (\x.E) arg
280 It uses the CPS form because the binding might be strict, in which
281 case we might discard the continuation:
282 let x* = error "foo" in (...x...)
284 It needs to turn unlifted bindings into a @case@. They can arise
285 from, say: (\x -> e) (4# + 3#)
288 simplNonRecBind :: SimplEnv
290 -> InExpr -> SimplEnv -- Arg, with its subst-env
291 -> OutType -- Type of thing computed by the context
292 -> (SimplEnv -> SimplM FloatsWithExpr) -- The body
293 -> SimplM FloatsWithExpr
295 simplNonRecBind env bndr rhs rhs_se cont_ty thing_inside
297 = pprPanic "simplNonRecBind" (ppr bndr <+> ppr rhs)
300 simplNonRecBind env bndr rhs rhs_se cont_ty thing_inside
301 = simplNonRecBind' env bndr rhs rhs_se cont_ty thing_inside
303 simplNonRecBind' env bndr rhs rhs_se cont_ty thing_inside
304 | preInlineUnconditionally env NotTopLevel bndr rhs
305 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
306 thing_inside (extendIdSubst env bndr (mkContEx rhs_se rhs))
308 | isStrictDmd (idNewDemandInfo bndr) || isStrictType bndr_ty -- A strict let
309 = -- Don't use simplBinder because that doesn't keep
310 -- fragile occurrence info in the substitution
311 simplLetBndr env bndr `thenSmpl` \ (env, bndr1) ->
312 simplStrictArg AnRhs env rhs rhs_se (idType bndr1) cont_ty $ \ env1 rhs1 ->
314 -- Now complete the binding and simplify the body
316 -- simplLetBndr doesn't deal with the IdInfo, so we must
317 -- do so here (c.f. simplLazyBind)
318 bndr2 = bndr1 `setIdInfo` simplIdInfo env (idInfo bndr)
319 env2 = modifyInScope env1 bndr2 bndr2
321 if needsCaseBinding bndr_ty rhs1
323 thing_inside env2 `thenSmpl` \ (floats, body) ->
324 returnSmpl (emptyFloats env2, Case rhs1 bndr2 (exprType body)
325 [(DEFAULT, [], wrapFloats floats body)])
327 completeNonRecX env2 True {- strict -} bndr bndr2 rhs1 thing_inside
329 | otherwise -- Normal, lazy case
330 = -- Don't use simplBinder because that doesn't keep
331 -- fragile occurrence info in the substitution
332 simplLetBndr env bndr `thenSmpl` \ (env, bndr') ->
333 simplLazyBind env NotTopLevel NonRecursive
334 bndr bndr' rhs rhs_se `thenSmpl` \ (floats, env) ->
335 addFloats env floats thing_inside
338 bndr_ty = idType bndr
341 A specialised variant of simplNonRec used when the RHS is already simplified, notably
342 in knownCon. It uses case-binding where necessary.
345 simplNonRecX :: SimplEnv
346 -> InId -- Old binder
347 -> OutExpr -- Simplified RHS
348 -> (SimplEnv -> SimplM FloatsWithExpr)
349 -> SimplM FloatsWithExpr
351 simplNonRecX env bndr new_rhs thing_inside
352 | needsCaseBinding (idType bndr) new_rhs
353 -- Make this test *before* the preInlineUnconditionally
354 -- Consider case I# (quotInt# x y) of
355 -- I# v -> let w = J# v in ...
356 -- If we gaily inline (quotInt# x y) for v, we end up building an
358 -- let w = J# (quotInt# x y) in ...
359 -- because quotInt# can fail.
360 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
361 thing_inside env `thenSmpl` \ (floats, body) ->
362 let body' = wrapFloats floats body in
363 returnSmpl (emptyFloats env, Case new_rhs bndr' (exprType body') [(DEFAULT, [], body')])
365 | preInlineUnconditionally env NotTopLevel bndr new_rhs
366 -- This happens; for example, the case_bndr during case of
367 -- known constructor: case (a,b) of x { (p,q) -> ... }
368 -- Here x isn't mentioned in the RHS, so we don't want to
369 -- create the (dead) let-binding let x = (a,b) in ...
371 -- Similarly, single occurrences can be inlined vigourously
372 -- e.g. case (f x, g y) of (a,b) -> ....
373 -- If a,b occur once we can avoid constructing the let binding for them.
374 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
377 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
378 completeNonRecX env False {- Non-strict; pessimistic -}
379 bndr bndr' new_rhs thing_inside
381 completeNonRecX env is_strict old_bndr new_bndr new_rhs thing_inside
382 = mkAtomicArgs is_strict
383 True {- OK to float unlifted -}
384 new_rhs `thenSmpl` \ (aux_binds, rhs2) ->
386 -- Make the arguments atomic if necessary,
387 -- adding suitable bindings
388 addAtomicBindsE env (fromOL aux_binds) $ \ env ->
389 completeLazyBind env NotTopLevel
390 old_bndr new_bndr rhs2 `thenSmpl` \ (floats, env) ->
391 addFloats env floats thing_inside
395 %************************************************************************
397 \subsection{Lazy bindings}
399 %************************************************************************
401 simplRecBind is used for
402 * recursive bindings only
405 simplRecBind :: SimplEnv -> TopLevelFlag
406 -> [(InId, InExpr)] -> [OutId]
407 -> SimplM (FloatsWith SimplEnv)
408 simplRecBind env top_lvl pairs bndrs'
409 = go env pairs bndrs' `thenSmpl` \ (floats, env) ->
410 returnSmpl (flattenFloats floats, env)
412 go env [] _ = returnSmpl (emptyFloats env, env)
414 go env ((bndr, rhs) : pairs) (bndr' : bndrs')
415 = simplRecOrTopPair env top_lvl bndr bndr' rhs `thenSmpl` \ (floats, env) ->
416 addFloats env floats (\env -> go env pairs bndrs')
420 simplRecOrTopPair is used for
421 * recursive bindings (whether top level or not)
422 * top-level non-recursive bindings
424 It assumes the binder has already been simplified, but not its IdInfo.
427 simplRecOrTopPair :: SimplEnv
429 -> InId -> OutId -- Binder, both pre-and post simpl
430 -> InExpr -- The RHS and its environment
431 -> SimplM (FloatsWith SimplEnv)
433 simplRecOrTopPair env top_lvl bndr bndr' rhs
434 | preInlineUnconditionally env top_lvl bndr rhs -- Check for unconditional inline
435 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
436 returnSmpl (emptyFloats env, extendIdSubst env bndr (mkContEx env rhs))
439 = simplLazyBind env top_lvl Recursive bndr bndr' rhs env
440 -- May not actually be recursive, but it doesn't matter
444 simplLazyBind is used for
445 * recursive bindings (whether top level or not)
446 * top-level non-recursive bindings
447 * non-top-level *lazy* non-recursive bindings
449 [Thus it deals with the lazy cases from simplNonRecBind, and all cases
450 from SimplRecOrTopBind]
453 1. It assumes that the binder is *already* simplified,
454 and is in scope, but not its IdInfo
456 2. It assumes that the binder type is lifted.
458 3. It does not check for pre-inline-unconditionallly;
459 that should have been done already.
462 simplLazyBind :: SimplEnv
463 -> TopLevelFlag -> RecFlag
464 -> InId -> OutId -- Binder, both pre-and post simpl
465 -> InExpr -> SimplEnv -- The RHS and its environment
466 -> SimplM (FloatsWith SimplEnv)
468 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
469 = let -- Transfer the IdInfo of the original binder to the new binder
470 -- This is crucial: we must preserve
474 -- etc. To do this we must apply the current substitution,
475 -- which incorporates earlier substitutions in this very letrec group.
477 -- NB 1. We do this *before* processing the RHS of the binder, so that
478 -- its substituted rules are visible in its own RHS.
479 -- This is important. Manuel found cases where he really, really
480 -- wanted a RULE for a recursive function to apply in that function's
481 -- own right-hand side.
483 -- NB 2: We do not transfer the arity (see Subst.substIdInfo)
484 -- The arity of an Id should not be visible
485 -- in its own RHS, else we eta-reduce
489 -- which isn't sound. And it makes the arity in f's IdInfo greater than
490 -- the manifest arity, which isn't good.
491 -- The arity will get added later.
493 -- NB 3: It's important that we *do* transer the loop-breaker OccInfo,
494 -- because that's what stops the Id getting inlined infinitely, in the body
497 -- NB 4: does no harm for non-recursive bindings
499 bndr2 = bndr1 `setIdInfo` simplIdInfo env (idInfo bndr)
500 env1 = modifyInScope env bndr2 bndr2
501 rhs_env = setInScope rhs_se env1
502 is_top_level = isTopLevel top_lvl
503 ok_float_unlifted = not is_top_level && isNonRec is_rec
504 rhs_cont = mkRhsStop (idType bndr1)
506 -- Simplify the RHS; note the mkRhsStop, which tells
507 -- the simplifier that this is the RHS of a let.
508 simplExprF rhs_env rhs rhs_cont `thenSmpl` \ (floats, rhs1) ->
510 -- If any of the floats can't be floated, give up now
511 -- (The allLifted predicate says True for empty floats.)
512 if (not ok_float_unlifted && not (allLifted floats)) then
513 completeLazyBind env1 top_lvl bndr bndr2
514 (wrapFloats floats rhs1)
517 -- ANF-ise a constructor or PAP rhs
518 mkAtomicArgs False {- Not strict -}
519 ok_float_unlifted rhs1 `thenSmpl` \ (aux_binds, rhs2) ->
521 -- If the result is a PAP, float the floats out, else wrap them
522 -- By this time it's already been ANF-ised (if necessary)
523 if isEmptyFloats floats && isNilOL aux_binds then -- Shortcut a common case
524 completeLazyBind env1 top_lvl bndr bndr2 rhs2
526 else if is_top_level || exprIsTrivial rhs2 || exprIsHNF rhs2 then
527 -- WARNING: long dodgy argument coming up
528 -- WANTED: a better way to do this
530 -- We can't use "exprIsCheap" instead of exprIsHNF,
531 -- because that causes a strictness bug.
532 -- x = let y* = E in case (scc y) of { T -> F; F -> T}
533 -- The case expression is 'cheap', but it's wrong to transform to
534 -- y* = E; x = case (scc y) of {...}
535 -- Either we must be careful not to float demanded non-values, or
536 -- we must use exprIsHNF for the test, which ensures that the
537 -- thing is non-strict. So exprIsHNF => bindings are non-strict
538 -- I think. The WARN below tests for this.
540 -- We use exprIsTrivial here because we want to reveal lone variables.
541 -- E.g. let { x = letrec { y = E } in y } in ...
542 -- Here we definitely want to float the y=E defn.
543 -- exprIsHNF definitely isn't right for that.
545 -- Again, the floated binding can't be strict; if it's recursive it'll
546 -- be non-strict; if it's non-recursive it'd be inlined.
548 -- Note [SCC-and-exprIsTrivial]
550 -- y = let { x* = E } in scc "foo" x
551 -- then we do *not* want to float out the x binding, because
552 -- it's strict! Fortunately, exprIsTrivial replies False to
555 -- There's a subtlety here. There may be a binding (x* = e) in the
556 -- floats, where the '*' means 'will be demanded'. So is it safe
557 -- to float it out? Answer no, but it won't matter because
558 -- we only float if (a) arg' is a WHNF, or (b) it's going to top level
559 -- and so there can't be any 'will be demanded' bindings in the floats.
561 ASSERT2( is_top_level || not (any demanded_float (floatBinds floats)),
562 ppr (filter demanded_float (floatBinds floats)) )
564 tick LetFloatFromLet `thenSmpl_` (
565 addFloats env1 floats $ \ env2 ->
566 addAtomicBinds env2 (fromOL aux_binds) $ \ env3 ->
567 completeLazyBind env3 top_lvl bndr bndr2 rhs2)
570 completeLazyBind env1 top_lvl bndr bndr2 (wrapFloats floats rhs1)
573 demanded_float (NonRec b r) = isStrictDmd (idNewDemandInfo b) && not (isUnLiftedType (idType b))
574 -- Unlifted-type (cheap-eagerness) lets may well have a demanded flag on them
575 demanded_float (Rec _) = False
580 %************************************************************************
582 \subsection{Completing a lazy binding}
584 %************************************************************************
587 * deals only with Ids, not TyVars
588 * takes an already-simplified binder and RHS
589 * is used for both recursive and non-recursive bindings
590 * is used for both top-level and non-top-level bindings
592 It does the following:
593 - tries discarding a dead binding
594 - tries PostInlineUnconditionally
595 - add unfolding [this is the only place we add an unfolding]
598 It does *not* attempt to do let-to-case. Why? Because it is used for
599 - top-level bindings (when let-to-case is impossible)
600 - many situations where the "rhs" is known to be a WHNF
601 (so let-to-case is inappropriate).
604 completeLazyBind :: SimplEnv
605 -> TopLevelFlag -- Flag stuck into unfolding
606 -> InId -- Old binder
607 -> OutId -- New binder
608 -> OutExpr -- Simplified RHS
609 -> SimplM (FloatsWith SimplEnv)
610 -- We return a new SimplEnv, because completeLazyBind may choose to do its work
611 -- by extending the substitution (e.g. let x = y in ...)
612 -- The new binding (if any) is returned as part of the floats.
613 -- NB: the returned SimplEnv has the right SubstEnv, but you should
614 -- (as usual) use the in-scope-env from the floats
616 completeLazyBind env top_lvl old_bndr new_bndr new_rhs
617 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
618 = -- Drop the binding
619 tick (PostInlineUnconditionally old_bndr) `thenSmpl_`
620 returnSmpl (emptyFloats env, extendIdSubst env old_bndr (DoneEx new_rhs))
621 -- Use the substitution to make quite, quite sure that the substitution
622 -- will happen, since we are going to discard the binding
627 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
629 -- Add the unfolding *only* for non-loop-breakers
630 -- Making loop breakers not have an unfolding at all
631 -- means that we can avoid tests in exprIsConApp, for example.
632 -- This is important: if exprIsConApp says 'yes' for a recursive
633 -- thing, then we can get into an infinite loop
635 -- If the unfolding is a value, the demand info may
636 -- go pear-shaped, so we nuke it. Example:
638 -- case x of (p,q) -> h p q x
639 -- Here x is certainly demanded. But after we've nuked
640 -- the case, we'll get just
641 -- let x = (a,b) in h a b x
642 -- and now x is not demanded (I'm assuming h is lazy)
643 -- This really happens. Similarly
644 -- let f = \x -> e in ...f..f...
645 -- After inling f at some of its call sites the original binding may
646 -- (for example) be no longer strictly demanded.
647 -- The solution here is a bit ad hoc...
648 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
649 final_info | loop_breaker = new_bndr_info
650 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
651 | otherwise = info_w_unf
653 final_id = new_bndr `setIdInfo` final_info
655 -- These seqs forces the Id, and hence its IdInfo,
656 -- and hence any inner substitutions
658 returnSmpl (unitFloat env final_id new_rhs, env)
661 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
662 loop_breaker = isLoopBreaker occ_info
663 old_info = idInfo old_bndr
664 occ_info = occInfo old_info
669 %************************************************************************
671 \subsection[Simplify-simplExpr]{The main function: simplExpr}
673 %************************************************************************
675 The reason for this OutExprStuff stuff is that we want to float *after*
676 simplifying a RHS, not before. If we do so naively we get quadratic
677 behaviour as things float out.
679 To see why it's important to do it after, consider this (real) example:
693 a -- Can't inline a this round, cos it appears twice
697 Each of the ==> steps is a round of simplification. We'd save a
698 whole round if we float first. This can cascade. Consider
703 let f = let d1 = ..d.. in \y -> e
707 in \x -> ...(\y ->e)...
709 Only in this second round can the \y be applied, and it
710 might do the same again.
714 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
715 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
717 expr_ty' = substTy env (exprType expr)
718 -- The type in the Stop continuation, expr_ty', is usually not used
719 -- It's only needed when discarding continuations after finding
720 -- a function that returns bottom.
721 -- Hence the lazy substitution
724 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
725 -- Simplify an expression, given a continuation
726 simplExprC env expr cont
727 = simplExprF env expr cont `thenSmpl` \ (floats, expr) ->
728 returnSmpl (wrapFloats floats expr)
730 simplExprF :: SimplEnv -> InExpr -> SimplCont -> SimplM FloatsWithExpr
731 -- Simplify an expression, returning floated binds
733 simplExprF env (Var v) cont = simplVar env v cont
734 simplExprF env (Lit lit) cont = rebuild env (Lit lit) cont
735 simplExprF env expr@(Lam _ _) cont = simplLam env expr cont
736 simplExprF env (Note note expr) cont = simplNote env note expr cont
737 simplExprF env (App fun arg) cont = simplExprF env fun (ApplyTo NoDup arg env cont)
739 simplExprF env (Type ty) cont
740 = ASSERT( contIsRhsOrArg cont )
741 simplType env ty `thenSmpl` \ ty' ->
742 rebuild env (Type ty') cont
744 simplExprF env (Case scrut bndr case_ty alts) cont
745 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
746 = -- Simplify the scrutinee with a Select continuation
747 simplExprF env scrut (Select NoDup bndr alts env cont)
750 = -- If case-of-case is off, simply simplify the case expression
751 -- in a vanilla Stop context, and rebuild the result around it
752 simplExprC env scrut case_cont `thenSmpl` \ case_expr' ->
753 rebuild env case_expr' cont
755 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
756 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
758 simplExprF env (Let (Rec pairs) body) cont
759 = simplLetBndrs env (map fst pairs) `thenSmpl` \ (env, bndrs') ->
760 -- NB: bndrs' don't have unfoldings or rules
761 -- We add them as we go down
763 simplRecBind env NotTopLevel pairs bndrs' `thenSmpl` \ (floats, env) ->
764 addFloats env floats $ \ env ->
765 simplExprF env body cont
767 -- A non-recursive let is dealt with by simplNonRecBind
768 simplExprF env (Let (NonRec bndr rhs) body) cont
769 = simplNonRecBind env bndr rhs env (contResultType cont) $ \ env ->
770 simplExprF env body cont
773 ---------------------------------
774 simplType :: SimplEnv -> InType -> SimplM OutType
775 -- Kept monadic just so we can do the seqType
777 = seqType new_ty `seq` returnSmpl new_ty
779 new_ty = substTy env ty
783 %************************************************************************
787 %************************************************************************
790 simplLam env fun cont
793 zap_it = mkLamBndrZapper fun (countArgs cont)
794 cont_ty = contResultType cont
796 -- Type-beta reduction
797 go env (Lam bndr body) (ApplyTo _ (Type ty_arg) arg_se body_cont)
798 = ASSERT( isTyVar bndr )
799 tick (BetaReduction bndr) `thenSmpl_`
800 simplType (setInScope arg_se env) ty_arg `thenSmpl` \ ty_arg' ->
801 go (extendTvSubst env bndr ty_arg') body body_cont
803 -- Ordinary beta reduction
804 go env (Lam bndr body) cont@(ApplyTo _ arg arg_se body_cont)
805 = tick (BetaReduction bndr) `thenSmpl_`
806 simplNonRecBind env (zap_it bndr) arg arg_se cont_ty $ \ env ->
807 go env body body_cont
809 -- Not enough args, so there are real lambdas left to put in the result
810 go env lam@(Lam _ _) cont
811 = simplLamBndrs env bndrs `thenSmpl` \ (env, bndrs') ->
812 simplExpr env body `thenSmpl` \ body' ->
813 mkLam env bndrs' body' cont `thenSmpl` \ (floats, new_lam) ->
814 addFloats env floats $ \ env ->
815 rebuild env new_lam cont
817 (bndrs,body) = collectBinders lam
819 -- Exactly enough args
820 go env expr cont = simplExprF env expr cont
822 mkLamBndrZapper :: CoreExpr -- Function
823 -> Int -- Number of args supplied, *including* type args
824 -> Id -> Id -- Use this to zap the binders
825 mkLamBndrZapper fun n_args
826 | n_args >= n_params fun = \b -> b -- Enough args
827 | otherwise = \b -> zapLamIdInfo b
829 -- NB: we count all the args incl type args
830 -- so we must count all the binders (incl type lambdas)
831 n_params (Note _ e) = n_params e
832 n_params (Lam b e) = 1 + n_params e
833 n_params other = 0::Int
837 %************************************************************************
841 %************************************************************************
844 simplNote env (Coerce to from) body cont
846 addCoerce s1 k1 cont -- Drop redundant coerces. This can happen if a polymoprhic
847 -- (coerce a b e) is instantiated with a=ty1 b=ty2 and the
848 -- two are the same. This happens a lot in Happy-generated parsers
849 | s1 `coreEqType` k1 = cont
851 addCoerce s1 k1 (CoerceIt t1 cont)
852 -- coerce T1 S1 (coerce S1 K1 e)
855 -- coerce T1 K1 e, otherwise
857 -- For example, in the initial form of a worker
858 -- we may find (coerce T (coerce S (\x.e))) y
859 -- and we'd like it to simplify to e[y/x] in one round
861 | t1 `coreEqType` k1 = cont -- The coerces cancel out
862 | otherwise = CoerceIt t1 cont -- They don't cancel, but
863 -- the inner one is redundant
865 addCoerce t1t2 s1s2 (ApplyTo dup arg arg_se cont)
866 | not (isTypeArg arg), -- This whole case only works for value args
867 -- Could upgrade to have equiv thing for type apps too
868 Just (s1, s2) <- splitFunTy_maybe s1s2
869 -- (coerce (T1->T2) (S1->S2) F) E
871 -- coerce T2 S2 (F (coerce S1 T1 E))
873 -- t1t2 must be a function type, T1->T2, because it's applied to something
874 -- but s1s2 might conceivably not be
876 -- When we build the ApplyTo we can't mix the out-types
877 -- with the InExpr in the argument, so we simply substitute
878 -- to make it all consistent. It's a bit messy.
879 -- But it isn't a common case.
881 (t1,t2) = splitFunTy t1t2
882 new_arg = mkCoerce2 s1 t1 (substExpr arg_env arg)
883 arg_env = setInScope arg_se env
885 ApplyTo dup new_arg (zapSubstEnv env) (addCoerce t2 s2 cont)
887 addCoerce to' _ cont = CoerceIt to' cont
889 simplType env to `thenSmpl` \ to' ->
890 simplType env from `thenSmpl` \ from' ->
891 simplExprF env body (addCoerce to' from' cont)
894 -- Hack: we only distinguish subsumed cost centre stacks for the purposes of
895 -- inlining. All other CCCSs are mapped to currentCCS.
896 simplNote env (SCC cc) e cont
897 = simplExpr (setEnclosingCC env currentCCS) e `thenSmpl` \ e' ->
898 rebuild env (mkSCC cc e') cont
900 simplNote env InlineCall e cont
901 = simplExprF env e (InlinePlease cont)
903 -- See notes with SimplMonad.inlineMode
904 simplNote env InlineMe e cont
905 | contIsRhsOrArg cont -- Totally boring continuation; see notes above
906 = -- Don't inline inside an INLINE expression
907 simplExpr (setMode inlineMode env ) e `thenSmpl` \ e' ->
908 rebuild env (mkInlineMe e') cont
910 | otherwise -- Dissolve the InlineMe note if there's
911 -- an interesting context of any kind to combine with
912 -- (even a type application -- anything except Stop)
913 = simplExprF env e cont
915 simplNote env (CoreNote s) e cont
916 = simplExpr env e `thenSmpl` \ e' ->
917 rebuild env (Note (CoreNote s) e') cont
921 %************************************************************************
923 \subsection{Dealing with calls}
925 %************************************************************************
928 simplVar env var cont
929 = case substId env var of
930 DoneEx e -> simplExprF (zapSubstEnv env) e cont
931 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
932 DoneId var1 occ -> completeCall (zapSubstEnv env) var1 occ cont
933 -- Note [zapSubstEnv]
934 -- The template is already simplified, so don't re-substitute.
935 -- This is VITAL. Consider
937 -- let y = \z -> ...x... in
939 -- We'll clone the inner \x, adding x->x' in the id_subst
940 -- Then when we inline y, we must *not* replace x by x' in
941 -- the inlined copy!!
943 ---------------------------------------------------------
944 -- Dealing with a call site
946 completeCall env var occ_info cont
947 = -- Simplify the arguments
948 getDOptsSmpl `thenSmpl` \ dflags ->
950 chkr = getSwitchChecker env
951 (args, call_cont, inline_call) = getContArgs chkr var cont
954 simplifyArgs env fn_ty args (contResultType call_cont) $ \ env args ->
956 -- Next, look for rules or specialisations that match
958 -- It's important to simplify the args first, because the rule-matcher
959 -- doesn't do substitution as it goes. We don't want to use subst_args
960 -- (defined in the 'where') because that throws away useful occurrence info,
961 -- and perhaps-very-important specialisations.
963 -- Some functions have specialisations *and* are strict; in this case,
964 -- we don't want to inline the wrapper of the non-specialised thing; better
965 -- to call the specialised thing instead.
966 -- We used to use the black-listing mechanism to ensure that inlining of
967 -- the wrapper didn't occur for things that have specialisations till a
968 -- later phase, so but now we just try RULES first
970 -- You might think that we shouldn't apply rules for a loop breaker:
971 -- doing so might give rise to an infinite loop, because a RULE is
972 -- rather like an extra equation for the function:
973 -- RULE: f (g x) y = x+y
976 -- But it's too drastic to disable rules for loop breakers.
977 -- Even the foldr/build rule would be disabled, because foldr
978 -- is recursive, and hence a loop breaker:
979 -- foldr k z (build g) = g k z
980 -- So it's up to the programmer: rules can cause divergence
983 in_scope = getInScope env
985 maybe_rule = case activeRule env of
986 Nothing -> Nothing -- No rules apply
987 Just act_fn -> lookupRule act_fn in_scope rules var args
990 Just (rule_name, rule_rhs) ->
991 tick (RuleFired rule_name) `thenSmpl_`
992 (if dopt Opt_D_dump_inlinings dflags then
993 pprTrace "Rule fired" (vcat [
994 text "Rule:" <+> ftext rule_name,
995 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
996 text "After: " <+> pprCoreExpr rule_rhs,
997 text "Cont: " <+> ppr call_cont])
1000 simplExprF env rule_rhs call_cont ;
1002 Nothing -> -- No rules
1004 -- Next, look for an inlining
1006 arg_infos = [ interestingArg arg | arg <- args, isValArg arg]
1008 interesting_cont = interestingCallContext (notNull args)
1012 active_inline = activeInline env var occ_info
1013 maybe_inline = callSiteInline dflags active_inline inline_call occ_info
1014 var arg_infos interesting_cont
1016 case maybe_inline of {
1017 Just unfolding -- There is an inlining!
1018 -> tick (UnfoldingDone var) `thenSmpl_`
1019 (if dopt Opt_D_dump_inlinings dflags then
1020 pprTrace "Inlining done" (vcat [
1021 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1022 text "Inlined fn: " <+> ppr unfolding,
1023 text "Cont: " <+> ppr call_cont])
1026 makeThatCall env var unfolding args call_cont
1029 Nothing -> -- No inlining!
1032 rebuild env (mkApps (Var var) args) call_cont
1035 makeThatCall :: SimplEnv
1037 -> InExpr -- Inlined function rhs
1038 -> [OutExpr] -- Arguments, already simplified
1039 -> SimplCont -- After the call
1040 -> SimplM FloatsWithExpr
1041 -- Similar to simplLam, but this time
1042 -- the arguments are already simplified
1043 makeThatCall orig_env var fun@(Lam _ _) args cont
1044 = go orig_env fun args
1046 zap_it = mkLamBndrZapper fun (length args)
1048 -- Type-beta reduction
1049 go env (Lam bndr body) (Type ty_arg : args)
1050 = ASSERT( isTyVar bndr )
1051 tick (BetaReduction bndr) `thenSmpl_`
1052 go (extendTvSubst env bndr ty_arg) body args
1054 -- Ordinary beta reduction
1055 go env (Lam bndr body) (arg : args)
1056 = tick (BetaReduction bndr) `thenSmpl_`
1057 simplNonRecX env (zap_it bndr) arg $ \ env ->
1060 -- Not enough args, so there are real lambdas left to put in the result
1062 = simplExprF env fun (pushContArgs orig_env args cont)
1063 -- NB: orig_env; the correct environment to capture with
1064 -- the arguments.... env has been augmented with substitutions
1065 -- from the beta reductions.
1067 makeThatCall env var fun args cont
1068 = simplExprF env fun (pushContArgs env args cont)
1072 %************************************************************************
1074 \subsection{Arguments}
1076 %************************************************************************
1079 ---------------------------------------------------------
1080 -- Simplifying the arguments of a call
1082 simplifyArgs :: SimplEnv
1083 -> OutType -- Type of the function
1084 -> [(InExpr, SimplEnv, Bool)] -- Details of the arguments
1085 -> OutType -- Type of the continuation
1086 -> (SimplEnv -> [OutExpr] -> SimplM FloatsWithExpr)
1087 -> SimplM FloatsWithExpr
1089 -- [CPS-like because of strict arguments]
1091 -- Simplify the arguments to a call.
1092 -- This part of the simplifier may break the no-shadowing invariant
1094 -- f (...(\a -> e)...) (case y of (a,b) -> e')
1095 -- where f is strict in its second arg
1096 -- If we simplify the innermost one first we get (...(\a -> e)...)
1097 -- Simplifying the second arg makes us float the case out, so we end up with
1098 -- case y of (a,b) -> f (...(\a -> e)...) e'
1099 -- So the output does not have the no-shadowing invariant. However, there is
1100 -- no danger of getting name-capture, because when the first arg was simplified
1101 -- we used an in-scope set that at least mentioned all the variables free in its
1102 -- static environment, and that is enough.
1104 -- We can't just do innermost first, or we'd end up with a dual problem:
1105 -- case x of (a,b) -> f e (...(\a -> e')...)
1107 -- I spent hours trying to recover the no-shadowing invariant, but I just could
1108 -- not think of an elegant way to do it. The simplifier is already knee-deep in
1109 -- continuations. We have to keep the right in-scope set around; AND we have
1110 -- to get the effect that finding (error "foo") in a strict arg position will
1111 -- discard the entire application and replace it with (error "foo"). Getting
1112 -- all this at once is TOO HARD!
1114 simplifyArgs env fn_ty args cont_ty thing_inside
1115 = go env fn_ty args thing_inside
1117 go env fn_ty [] thing_inside = thing_inside env []
1118 go env fn_ty (arg:args) thing_inside = simplifyArg env fn_ty arg cont_ty $ \ env arg' ->
1119 go env (applyTypeToArg fn_ty arg') args $ \ env args' ->
1120 thing_inside env (arg':args')
1122 simplifyArg env fn_ty (Type ty_arg, se, _) cont_ty thing_inside
1123 = simplType (setInScope se env) ty_arg `thenSmpl` \ new_ty_arg ->
1124 thing_inside env (Type new_ty_arg)
1126 simplifyArg env fn_ty (val_arg, arg_se, is_strict) cont_ty thing_inside
1128 = simplStrictArg AnArg env val_arg arg_se arg_ty cont_ty thing_inside
1130 | otherwise -- Lazy argument
1131 -- DO NOT float anything outside, hence simplExprC
1132 -- There is no benefit (unlike in a let-binding), and we'd
1133 -- have to be very careful about bogus strictness through
1134 -- floating a demanded let.
1135 = simplExprC (setInScope arg_se env) val_arg
1136 (mkBoringStop arg_ty) `thenSmpl` \ arg1 ->
1137 thing_inside env arg1
1139 arg_ty = funArgTy fn_ty
1142 simplStrictArg :: LetRhsFlag
1143 -> SimplEnv -- The env of the call
1144 -> InExpr -> SimplEnv -- The arg plus its env
1145 -> OutType -- arg_ty: type of the argument
1146 -> OutType -- cont_ty: Type of thing computed by the context
1147 -> (SimplEnv -> OutExpr -> SimplM FloatsWithExpr)
1148 -- Takes an expression of type rhs_ty,
1149 -- returns an expression of type cont_ty
1150 -- The env passed to this continuation is the
1151 -- env of the call, plus any new in-scope variables
1152 -> SimplM FloatsWithExpr -- An expression of type cont_ty
1154 simplStrictArg is_rhs call_env arg arg_env arg_ty cont_ty thing_inside
1155 = simplExprF (setInScope arg_env call_env) arg
1156 (ArgOf is_rhs arg_ty cont_ty (\ new_env -> thing_inside (setInScope call_env new_env)))
1157 -- Notice the way we use arg_env (augmented with in-scope vars from call_env)
1158 -- to simplify the argument
1159 -- and call-env (augmented with in-scope vars from the arg) to pass to the continuation
1163 %************************************************************************
1165 \subsection{mkAtomicArgs}
1167 %************************************************************************
1169 mkAtomicArgs takes a putative RHS, checks whether it's a PAP or
1170 constructor application and, if so, converts it to ANF, so that the
1171 resulting thing can be inlined more easily. Thus
1178 There are three sorts of binding context, specified by the two
1184 N N Top-level or recursive Only bind args of lifted type
1186 N Y Non-top-level and non-recursive, Bind args of lifted type, or
1187 but lazy unlifted-and-ok-for-speculation
1189 Y Y Non-top-level, non-recursive, Bind all args
1190 and strict (demanded)
1197 there is no point in transforming to
1199 x = case (y div# z) of r -> MkC r
1201 because the (y div# z) can't float out of the let. But if it was
1202 a *strict* let, then it would be a good thing to do. Hence the
1203 context information.
1206 mkAtomicArgs :: Bool -- A strict binding
1207 -> Bool -- OK to float unlifted args
1209 -> SimplM (OrdList (OutId,OutExpr), -- The floats (unusually) may include
1210 OutExpr) -- things that need case-binding,
1211 -- if the strict-binding flag is on
1213 mkAtomicArgs is_strict ok_float_unlifted rhs
1214 | (Var fun, args) <- collectArgs rhs, -- It's an application
1215 isDataConWorkId fun || valArgCount args < idArity fun -- And it's a constructor or PAP
1216 = go fun nilOL [] args -- Have a go
1218 | otherwise = bale_out -- Give up
1221 bale_out = returnSmpl (nilOL, rhs)
1223 go fun binds rev_args []
1224 = returnSmpl (binds, mkApps (Var fun) (reverse rev_args))
1226 go fun binds rev_args (arg : args)
1227 | exprIsTrivial arg -- Easy case
1228 = go fun binds (arg:rev_args) args
1230 | not can_float_arg -- Can't make this arg atomic
1231 = bale_out -- ... so give up
1233 | otherwise -- Don't forget to do it recursively
1234 -- E.g. x = a:b:c:[]
1235 = mkAtomicArgs is_strict ok_float_unlifted arg `thenSmpl` \ (arg_binds, arg') ->
1236 newId FSLIT("a") arg_ty `thenSmpl` \ arg_id ->
1237 go fun ((arg_binds `snocOL` (arg_id,arg')) `appOL` binds)
1238 (Var arg_id : rev_args) args
1240 arg_ty = exprType arg
1241 can_float_arg = is_strict
1242 || not (isUnLiftedType arg_ty)
1243 || (ok_float_unlifted && exprOkForSpeculation arg)
1246 addAtomicBinds :: SimplEnv -> [(OutId,OutExpr)]
1247 -> (SimplEnv -> SimplM (FloatsWith a))
1248 -> SimplM (FloatsWith a)
1249 addAtomicBinds env [] thing_inside = thing_inside env
1250 addAtomicBinds env ((v,r):bs) thing_inside = addAuxiliaryBind env (NonRec v r) $ \ env ->
1251 addAtomicBinds env bs thing_inside
1253 addAtomicBindsE :: SimplEnv -> [(OutId,OutExpr)]
1254 -> (SimplEnv -> SimplM FloatsWithExpr)
1255 -> SimplM FloatsWithExpr
1256 -- Same again, but this time we're in an expression context,
1257 -- and may need to do some case bindings
1259 addAtomicBindsE env [] thing_inside
1261 addAtomicBindsE env ((v,r):bs) thing_inside
1262 | needsCaseBinding (idType v) r
1263 = addAtomicBindsE (addNewInScopeIds env [v]) bs thing_inside `thenSmpl` \ (floats, expr) ->
1264 WARN( exprIsTrivial expr, ppr v <+> pprCoreExpr expr )
1265 (let body = wrapFloats floats expr in
1266 returnSmpl (emptyFloats env, Case r v (exprType body) [(DEFAULT,[],body)]))
1269 = addAuxiliaryBind env (NonRec v r) $ \ env ->
1270 addAtomicBindsE env bs thing_inside
1274 %************************************************************************
1276 \subsection{The main rebuilder}
1278 %************************************************************************
1281 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM FloatsWithExpr
1283 rebuild env expr (Stop _ _ _) = rebuildDone env expr
1284 rebuild env expr (ArgOf _ _ _ cont_fn) = cont_fn env expr
1285 rebuild env expr (CoerceIt to_ty cont) = rebuild env (mkCoerce to_ty expr) cont
1286 rebuild env expr (InlinePlease cont) = rebuild env (Note InlineCall expr) cont
1287 rebuild env expr (Select _ bndr alts se cont) = rebuildCase (setInScope se env) expr bndr alts cont
1288 rebuild env expr (ApplyTo _ arg se cont) = rebuildApp (setInScope se env) expr arg cont
1290 rebuildApp env fun arg cont
1291 = simplExpr env arg `thenSmpl` \ arg' ->
1292 rebuild env (App fun arg') cont
1294 rebuildDone env expr = returnSmpl (emptyFloats env, expr)
1298 %************************************************************************
1300 \subsection{Functions dealing with a case}
1302 %************************************************************************
1304 Blob of helper functions for the "case-of-something-else" situation.
1307 ---------------------------------------------------------
1308 -- Eliminate the case if possible
1310 rebuildCase :: SimplEnv
1311 -> OutExpr -- Scrutinee
1312 -> InId -- Case binder
1313 -> [InAlt] -- Alternatives (inceasing order)
1315 -> SimplM FloatsWithExpr
1317 rebuildCase env scrut case_bndr alts cont
1318 | Just (con,args) <- exprIsConApp_maybe scrut
1319 -- Works when the scrutinee is a variable with a known unfolding
1320 -- as well as when it's an explicit constructor application
1321 = knownCon env (DataAlt con) args case_bndr alts cont
1323 | Lit lit <- scrut -- No need for same treatment as constructors
1324 -- because literals are inlined more vigorously
1325 = knownCon env (LitAlt lit) [] case_bndr alts cont
1328 = -- Prepare the alternatives.
1329 prepareAlts scrut case_bndr alts `thenSmpl` \ (better_alts, handled_cons) ->
1331 -- Prepare the continuation;
1332 -- The new subst_env is in place
1333 prepareCaseCont env better_alts cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1334 addFloats env floats $ \ env ->
1337 -- The case expression is annotated with the result type of the continuation
1338 -- This may differ from the type originally on the case. For example
1339 -- case(T) (case(Int#) a of { True -> 1#; False -> 0# }) of
1342 -- let j a# = <blob>
1343 -- in case(T) a of { True -> j 1#; False -> j 0# }
1344 -- Note that the case that scrutinises a now returns a T not an Int#
1345 res_ty' = contResultType dup_cont
1348 -- Deal with case binder
1349 simplCaseBinder env scrut case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1351 -- Deal with the case alternatives
1352 simplAlts alt_env handled_cons
1353 case_bndr' better_alts dup_cont `thenSmpl` \ alts' ->
1355 -- Put the case back together
1356 mkCase scrut case_bndr' res_ty' alts' `thenSmpl` \ case_expr ->
1358 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1359 -- The case binder *not* scope over the whole returned case-expression
1360 rebuild env case_expr nondup_cont
1363 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1364 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1365 way, there's a chance that v will now only be used once, and hence
1370 There is a time we *don't* want to do that, namely when
1371 -fno-case-of-case is on. This happens in the first simplifier pass,
1372 and enhances full laziness. Here's the bad case:
1373 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1374 If we eliminate the inner case, we trap it inside the I# v -> arm,
1375 which might prevent some full laziness happening. I've seen this
1376 in action in spectral/cichelli/Prog.hs:
1377 [(m,n) | m <- [1..max], n <- [1..max]]
1378 Hence the check for NoCaseOfCase.
1382 There is another situation when we don't want to do it. If we have
1384 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1385 ...other cases .... }
1387 We'll perform the binder-swap for the outer case, giving
1389 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1390 ...other cases .... }
1392 But there is no point in doing it for the inner case, because w1 can't
1393 be inlined anyway. Furthermore, doing the case-swapping involves
1394 zapping w2's occurrence info (see paragraphs that follow), and that
1395 forces us to bind w2 when doing case merging. So we get
1397 case x of w1 { A -> let w2 = w1 in e1
1398 B -> let w2 = w1 in e2
1399 ...other cases .... }
1401 This is plain silly in the common case where w2 is dead.
1403 Even so, I can't see a good way to implement this idea. I tried
1404 not doing the binder-swap if the scrutinee was already evaluated
1405 but that failed big-time:
1409 case v of w { MkT x ->
1410 case x of x1 { I# y1 ->
1411 case x of x2 { I# y2 -> ...
1413 Notice that because MkT is strict, x is marked "evaluated". But to
1414 eliminate the last case, we must either make sure that x (as well as
1415 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1416 the binder-swap. So this whole note is a no-op.
1420 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1421 any occurrence info (eg IAmDead) in the case binder, because the
1422 case-binder now effectively occurs whenever v does. AND we have to do
1423 the same for the pattern-bound variables! Example:
1425 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1427 Here, b and p are dead. But when we move the argment inside the first
1428 case RHS, and eliminate the second case, we get
1430 case x of { (a,b) -> a b }
1432 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1435 Indeed, this can happen anytime the case binder isn't dead:
1436 case <any> of x { (a,b) ->
1437 case x of { (p,q) -> p } }
1438 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1439 The point is that we bring into the envt a binding
1441 after the outer case, and that makes (a,b) alive. At least we do unless
1442 the case binder is guaranteed dead.
1445 simplCaseBinder env (Var v) case_bndr
1446 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
1448 -- Failed try [see Note 2 above]
1449 -- not (isEvaldUnfolding (idUnfolding v))
1451 = simplBinder env (zap case_bndr) `thenSmpl` \ (env, case_bndr') ->
1452 returnSmpl (modifyInScope env v case_bndr', case_bndr')
1453 -- We could extend the substitution instead, but it would be
1454 -- a hack because then the substitution wouldn't be idempotent
1455 -- any more (v is an OutId). And this does just as well.
1457 zap b = b `setIdOccInfo` NoOccInfo
1459 simplCaseBinder env other_scrut case_bndr
1460 = simplBinder env case_bndr `thenSmpl` \ (env, case_bndr') ->
1461 returnSmpl (env, case_bndr')
1467 simplAlts :: SimplEnv
1468 -> [AltCon] -- Alternatives the scrutinee can't be
1469 -- in the default case
1470 -> OutId -- Case binder
1471 -> [InAlt] -> SimplCont
1472 -> SimplM [OutAlt] -- Includes the continuation
1474 simplAlts env handled_cons case_bndr' alts cont'
1475 = do { mb_alts <- mapSmpl simpl_alt alts
1476 ; return [alt' | Just (_, alt') <- mb_alts] }
1477 -- Filter out the alternatives that are inaccessible
1479 simpl_alt alt = simplAlt env handled_cons case_bndr' alt cont'
1481 simplAlt :: SimplEnv -> [AltCon] -> OutId -> InAlt -> SimplCont
1482 -> SimplM (Maybe (TvSubstEnv, OutAlt))
1483 -- Simplify an alternative, returning the type refinement for the
1484 -- alternative, if the alternative does any refinement at all
1485 -- Nothing => the alternative is inaccessible
1487 simplAlt env handled_cons case_bndr' (DEFAULT, bndrs, rhs) cont'
1488 = ASSERT( null bndrs )
1489 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1490 returnSmpl (Just (emptyVarEnv, (DEFAULT, [], rhs')))
1492 env' = mk_rhs_env env case_bndr' (mkOtherCon handled_cons)
1493 -- Record the constructors that the case-binder *can't* be.
1495 simplAlt env handled_cons case_bndr' (LitAlt lit, bndrs, rhs) cont'
1496 = ASSERT( null bndrs )
1497 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1498 returnSmpl (Just (emptyVarEnv, (LitAlt lit, [], rhs')))
1500 env' = mk_rhs_env env case_bndr' (mkUnfolding False (Lit lit))
1502 simplAlt env handled_cons case_bndr' (DataAlt con, vs, rhs) cont'
1503 | isVanillaDataCon con
1504 = -- Deal with the pattern-bound variables
1505 -- Mark the ones that are in ! positions in the data constructor
1506 -- as certainly-evaluated.
1507 -- NB: it happens that simplBinders does *not* erase the OtherCon
1508 -- form of unfolding, so it's ok to add this info before
1509 -- doing simplBinders
1510 simplBinders env (add_evals con vs) `thenSmpl` \ (env, vs') ->
1512 -- Bind the case-binder to (con args)
1513 let unf = mkUnfolding False (mkConApp con con_args)
1514 inst_tys' = tyConAppArgs (idType case_bndr')
1515 con_args = map Type inst_tys' ++ map varToCoreExpr vs'
1516 env' = mk_rhs_env env case_bndr' unf
1518 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1519 returnSmpl (Just (emptyVarEnv, (DataAlt con, vs', rhs')))
1521 | otherwise -- GADT case
1523 (tvs,ids) = span isTyVar vs
1525 simplBinders env tvs `thenSmpl` \ (env1, tvs') ->
1526 case coreRefineTys con tvs' (idType case_bndr') of {
1527 Nothing -- Inaccessible
1528 | opt_PprStyle_Debug -- Hack: if debugging is on, generate an error case
1530 -> let rhs' = mkApps (Var eRROR_ID)
1531 [Type (substTy env (exprType rhs)),
1532 Lit (mkStringLit "Impossible alternative (GADT)")]
1534 simplBinders env1 ids `thenSmpl` \ (env2, ids') ->
1535 returnSmpl (Just (emptyVarEnv, (DataAlt con, tvs' ++ ids', rhs')))
1537 | otherwise -- Filter out the inaccessible branch
1540 Just refine@(tv_subst_env, _) -> -- The normal case
1543 env2 = refineSimplEnv env1 refine
1544 -- Simplify the Ids in the refined environment, so their types
1545 -- reflect the refinement. Usually this doesn't matter, but it helps
1546 -- in mkDupableAlt, when we want to float a lambda that uses these binders
1547 -- Furthermore, it means the binders contain maximal type information
1549 simplBinders env2 (add_evals con ids) `thenSmpl` \ (env3, ids') ->
1550 let unf = mkUnfolding False con_app
1551 con_app = mkConApp con con_args
1552 con_args = map varToCoreExpr vs' -- NB: no inst_tys'
1553 env_w_unf = mk_rhs_env env3 case_bndr' unf
1556 simplExprC env_w_unf rhs cont' `thenSmpl` \ rhs' ->
1557 returnSmpl (Just (tv_subst_env, (DataAlt con, vs', rhs'))) }
1560 -- add_evals records the evaluated-ness of the bound variables of
1561 -- a case pattern. This is *important*. Consider
1562 -- data T = T !Int !Int
1564 -- case x of { T a b -> T (a+1) b }
1566 -- We really must record that b is already evaluated so that we don't
1567 -- go and re-evaluate it when constructing the result.
1568 add_evals dc vs = cat_evals dc vs (dataConRepStrictness dc)
1570 cat_evals dc vs strs
1574 go (v:vs) strs | isTyVar v = v : go vs strs
1575 go (v:vs) (str:strs)
1576 | isMarkedStrict str = evald_v : go vs strs
1577 | otherwise = zapped_v : go vs strs
1579 zapped_v = zap_occ_info v
1580 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1581 go _ _ = pprPanic "cat_evals" (ppr dc $$ ppr vs $$ ppr strs)
1583 -- If the case binder is alive, then we add the unfolding
1585 -- to the envt; so vs are now very much alive
1586 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1587 | otherwise = \id -> id `setIdOccInfo` NoOccInfo
1589 mk_rhs_env env case_bndr' case_bndr_unf
1590 = modifyInScope env case_bndr' (case_bndr' `setIdUnfolding` case_bndr_unf)
1594 %************************************************************************
1596 \subsection{Known constructor}
1598 %************************************************************************
1600 We are a bit careful with occurrence info. Here's an example
1602 (\x* -> case x of (a*, b) -> f a) (h v, e)
1604 where the * means "occurs once". This effectively becomes
1605 case (h v, e) of (a*, b) -> f a)
1607 let a* = h v; b = e in f a
1611 All this should happen in one sweep.
1614 knownCon :: SimplEnv -> AltCon -> [OutExpr]
1615 -> InId -> [InAlt] -> SimplCont
1616 -> SimplM FloatsWithExpr
1618 knownCon env con args bndr alts cont
1619 = tick (KnownBranch bndr) `thenSmpl_`
1620 case findAlt con alts of
1621 (DEFAULT, bs, rhs) -> ASSERT( null bs )
1622 simplNonRecX env bndr scrut $ \ env ->
1623 -- This might give rise to a binding with non-atomic args
1624 -- like x = Node (f x) (g x)
1625 -- but no harm will be done
1626 simplExprF env rhs cont
1629 LitAlt lit -> Lit lit
1630 DataAlt dc -> mkConApp dc args
1632 (LitAlt lit, bs, rhs) -> ASSERT( null bs )
1633 simplNonRecX env bndr (Lit lit) $ \ env ->
1634 simplExprF env rhs cont
1636 (DataAlt dc, bs, rhs)
1637 -> ASSERT( n_drop_tys + length bs == length args )
1638 bind_args env bs (drop n_drop_tys args) $ \ env ->
1640 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1641 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1642 -- args are aready OutExprs, but bs are InIds
1644 simplNonRecX env bndr con_app $ \ env ->
1645 simplExprF env rhs cont
1647 n_drop_tys | isVanillaDataCon dc = tyConArity (dataConTyCon dc)
1649 -- Vanilla data constructors lack type arguments in the pattern
1652 bind_args env [] _ thing_inside = thing_inside env
1654 bind_args env (b:bs) (Type ty : args) thing_inside
1655 = ASSERT( isTyVar b )
1656 bind_args (extendTvSubst env b ty) bs args thing_inside
1658 bind_args env (b:bs) (arg : args) thing_inside
1660 simplNonRecX env b arg $ \ env ->
1661 bind_args env bs args thing_inside
1665 %************************************************************************
1667 \subsection{Duplicating continuations}
1669 %************************************************************************
1672 prepareCaseCont :: SimplEnv
1673 -> [InAlt] -> SimplCont
1674 -> SimplM (FloatsWith (SimplCont,SimplCont))
1675 -- Return a duplicatable continuation, a non-duplicable part
1676 -- plus some extra bindings
1678 -- No need to make it duplicatable if there's only one alternative
1679 prepareCaseCont env [alt] cont = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1680 prepareCaseCont env alts cont = mkDupableCont env cont
1684 mkDupableCont :: SimplEnv -> SimplCont
1685 -> SimplM (FloatsWith (SimplCont, SimplCont))
1687 mkDupableCont env cont
1688 | contIsDupable cont
1689 = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1691 mkDupableCont env (CoerceIt ty cont)
1692 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1693 returnSmpl (floats, (CoerceIt ty dup_cont, nondup_cont))
1695 mkDupableCont env (InlinePlease cont)
1696 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1697 returnSmpl (floats, (InlinePlease dup_cont, nondup_cont))
1699 mkDupableCont env cont@(ArgOf _ arg_ty _ _)
1700 = returnSmpl (emptyFloats env, (mkBoringStop arg_ty, cont))
1701 -- Do *not* duplicate an ArgOf continuation
1702 -- Because ArgOf continuations are opaque, we gain nothing by
1703 -- propagating them into the expressions, and we do lose a lot.
1704 -- Here's an example:
1705 -- && (case x of { T -> F; F -> T }) E
1706 -- Now, && is strict so we end up simplifying the case with
1707 -- an ArgOf continuation. If we let-bind it, we get
1709 -- let $j = \v -> && v E
1710 -- in simplExpr (case x of { T -> F; F -> T })
1711 -- (ArgOf (\r -> $j r)
1712 -- And after simplifying more we get
1714 -- let $j = \v -> && v E
1715 -- in case of { T -> $j F; F -> $j T }
1716 -- Which is a Very Bad Thing
1718 -- The desire not to duplicate is the entire reason that
1719 -- mkDupableCont returns a pair of continuations.
1721 -- The original plan had:
1722 -- e.g. (...strict-fn...) [...hole...]
1724 -- let $j = \a -> ...strict-fn...
1725 -- in $j [...hole...]
1727 mkDupableCont env (ApplyTo _ arg se cont)
1728 = -- e.g. [...hole...] (...arg...)
1730 -- let a = ...arg...
1731 -- in [...hole...] a
1732 simplExpr (setInScope se env) arg `thenSmpl` \ arg' ->
1734 mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1735 addFloats env floats $ \ env ->
1737 if exprIsDupable arg' then
1738 returnSmpl (emptyFloats env, (ApplyTo OkToDup arg' (zapSubstEnv se) dup_cont, nondup_cont))
1740 newId FSLIT("a") (exprType arg') `thenSmpl` \ arg_id ->
1742 tick (CaseOfCase arg_id) `thenSmpl_`
1743 -- Want to tick here so that we go round again,
1744 -- and maybe copy or inline the code.
1745 -- Not strictly CaseOfCase, but never mind
1747 returnSmpl (unitFloat env arg_id arg',
1748 (ApplyTo OkToDup (Var arg_id) (zapSubstEnv se) dup_cont,
1750 -- But what if the arg should be case-bound?
1751 -- This has been this way for a long time, so I'll leave it,
1752 -- but I can't convince myself that it's right.
1754 mkDupableCont env (Select _ case_bndr alts se cont)
1755 = -- e.g. (case [...hole...] of { pi -> ei })
1757 -- let ji = \xij -> ei
1758 -- in case [...hole...] of { pi -> ji xij }
1759 tick (CaseOfCase case_bndr) `thenSmpl_`
1761 alt_env = setInScope se env
1763 prepareCaseCont alt_env alts cont `thenSmpl` \ (floats1, (dup_cont, nondup_cont)) ->
1764 addFloats alt_env floats1 $ \ alt_env ->
1766 simplBinder alt_env case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1767 -- NB: simplBinder does not zap deadness occ-info, so
1768 -- a dead case_bndr' will still advertise its deadness
1769 -- This is really important because in
1770 -- case e of b { (# a,b #) -> ... }
1771 -- b is always dead, and indeed we are not allowed to bind b to (# a,b #),
1772 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1773 -- In the new alts we build, we have the new case binder, so it must retain
1776 mkDupableAlts alt_env case_bndr' alts dup_cont `thenSmpl` \ (floats2, alts') ->
1777 addFloats alt_env floats2 $ \ alt_env ->
1778 returnSmpl (emptyFloats alt_env,
1779 (Select OkToDup case_bndr' alts' (zapSubstEnv se)
1780 (mkBoringStop (contResultType dup_cont)),
1783 mkDupableAlts :: SimplEnv -> OutId -> [InAlt] -> SimplCont
1784 -> SimplM (FloatsWith [InAlt])
1785 -- Absorbs the continuation into the new alternatives
1787 mkDupableAlts env case_bndr' alts dupable_cont
1790 go env [] = returnSmpl (emptyFloats env, [])
1792 = do { (floats1, mb_alt') <- mkDupableAlt env case_bndr' dupable_cont alt
1793 ; addFloats env floats1 $ \ env -> do
1794 { (floats2, alts') <- go env alts
1795 ; returnSmpl (floats2, case mb_alt' of
1796 Just alt' -> alt' : alts'
1800 mkDupableAlt env case_bndr' cont alt
1801 = simplAlt env [] case_bndr' alt cont `thenSmpl` \ mb_stuff ->
1803 Nothing -> returnSmpl (emptyFloats env, Nothing) ;
1805 Just (reft, (con, bndrs', rhs')) ->
1806 -- Safe to say that there are no handled-cons for the DEFAULT case
1808 if exprIsDupable rhs' then
1809 returnSmpl (emptyFloats env, Just (con, bndrs', rhs'))
1810 -- It is worth checking for a small RHS because otherwise we
1811 -- get extra let bindings that may cause an extra iteration of the simplifier to
1812 -- inline back in place. Quite often the rhs is just a variable or constructor.
1813 -- The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1814 -- iterations because the version with the let bindings looked big, and so wasn't
1815 -- inlined, but after the join points had been inlined it looked smaller, and so
1818 -- NB: we have to check the size of rhs', not rhs.
1819 -- Duplicating a small InAlt might invalidate occurrence information
1820 -- However, if it *is* dupable, we return the *un* simplified alternative,
1821 -- because otherwise we'd need to pair it up with an empty subst-env....
1822 -- but we only have one env shared between all the alts.
1823 -- (Remember we must zap the subst-env before re-simplifying something).
1824 -- Rather than do this we simply agree to re-simplify the original (small) thing later.
1828 rhs_ty' = exprType rhs'
1829 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1831 | isTyVar bndr = not (bndr `elemVarEnv` reft)
1832 -- Don't abstract over tyvar binders which are refined away
1833 -- See Note [Refinement] below
1834 | otherwise = not (isDeadBinder bndr)
1835 -- The deadness info on the new Ids is preserved by simplBinders
1837 -- If we try to lift a primitive-typed something out
1838 -- for let-binding-purposes, we will *caseify* it (!),
1839 -- with potentially-disastrous strictness results. So
1840 -- instead we turn it into a function: \v -> e
1841 -- where v::State# RealWorld#. The value passed to this function
1842 -- is realworld#, which generates (almost) no code.
1844 -- There's a slight infelicity here: we pass the overall
1845 -- case_bndr to all the join points if it's used in *any* RHS,
1846 -- because we don't know its usage in each RHS separately
1848 -- We used to say "&& isUnLiftedType rhs_ty'" here, but now
1849 -- we make the join point into a function whenever used_bndrs'
1850 -- is empty. This makes the join-point more CPR friendly.
1851 -- Consider: let j = if .. then I# 3 else I# 4
1852 -- in case .. of { A -> j; B -> j; C -> ... }
1854 -- Now CPR doesn't w/w j because it's a thunk, so
1855 -- that means that the enclosing function can't w/w either,
1856 -- which is a lose. Here's the example that happened in practice:
1857 -- kgmod :: Int -> Int -> Int
1858 -- kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1862 -- I have seen a case alternative like this:
1863 -- True -> \v -> ...
1864 -- It's a bit silly to add the realWorld dummy arg in this case, making
1867 -- (the \v alone is enough to make CPR happy) but I think it's rare
1869 ( if not (any isId used_bndrs')
1870 then newId FSLIT("w") realWorldStatePrimTy `thenSmpl` \ rw_id ->
1871 returnSmpl ([rw_id], [Var realWorldPrimId])
1873 returnSmpl (used_bndrs', map varToCoreExpr used_bndrs')
1874 ) `thenSmpl` \ (final_bndrs', final_args) ->
1876 -- See comment about "$j" name above
1877 newId FSLIT("$j") (mkPiTypes final_bndrs' rhs_ty') `thenSmpl` \ join_bndr ->
1878 -- Notice the funky mkPiTypes. If the contructor has existentials
1879 -- it's possible that the join point will be abstracted over
1880 -- type varaibles as well as term variables.
1881 -- Example: Suppose we have
1882 -- data T = forall t. C [t]
1884 -- case (case e of ...) of
1885 -- C t xs::[t] -> rhs
1886 -- We get the join point
1887 -- let j :: forall t. [t] -> ...
1888 -- j = /\t \xs::[t] -> rhs
1890 -- case (case e of ...) of
1891 -- C t xs::[t] -> j t xs
1893 -- We make the lambdas into one-shot-lambdas. The
1894 -- join point is sure to be applied at most once, and doing so
1895 -- prevents the body of the join point being floated out by
1896 -- the full laziness pass
1897 really_final_bndrs = map one_shot final_bndrs'
1898 one_shot v | isId v = setOneShotLambda v
1900 join_rhs = mkLams really_final_bndrs rhs'
1901 join_call = mkApps (Var join_bndr) final_args
1903 returnSmpl (unitFloat env join_bndr join_rhs, Just (con, bndrs', join_call)) }
1910 MkT :: a -> b -> T a
1914 MkT a' b (p::a') (q::b) -> [p,w]
1916 The danger is that we'll make a join point
1920 and that's ill-typed, because (p::a') but (w::a).
1922 Solution so far: don't abstract over a', because the type refinement
1923 maps [a' -> a] . Ultimately that won't work when real refinement goes on.
1925 Then we must abstract over any refined free variables. Hmm. Maybe we
1926 could just abstract over *all* free variables, thereby lambda-lifting
1927 the join point? We should try this.