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 OccName ( encodeFS )
32 import IdInfo ( OccInfo(..), isLoopBreaker,
33 setArityInfo, zapDemandInfo,
37 import NewDemand ( isStrictDmd )
38 import Unify ( coreRefineTys )
39 import DataCon ( dataConTyCon, dataConRepStrictness, isVanillaDataCon )
40 import TyCon ( tyConArity )
42 import PprCore ( pprParendExpr, pprCoreExpr )
43 import CoreUnfold ( mkUnfolding, callSiteInline )
44 import CoreUtils ( exprIsDupable, exprIsTrivial, needsCaseBinding,
45 exprIsConApp_maybe, mkPiTypes, findAlt,
47 exprOkForSpeculation, exprArity,
48 mkCoerce, mkCoerce2, mkSCC, mkInlineMe, applyTypeToArg
50 import Rules ( lookupRule )
51 import BasicTypes ( isMarkedStrict )
52 import CostCentre ( currentCCS )
53 import Type ( TvSubstEnv, isUnLiftedType, seqType, tyConAppArgs, funArgTy,
54 splitFunTy_maybe, splitFunTy, coreEqType
56 import VarEnv ( elemVarEnv, emptyVarEnv )
57 import TysPrim ( realWorldStatePrimTy )
58 import PrelInfo ( realWorldPrimId )
59 import BasicTypes ( TopLevelFlag(..), isTopLevel,
62 import StaticFlags ( opt_PprStyle_Debug )
64 import Maybes ( orElse )
66 import Util ( notNull )
70 The guts of the simplifier is in this module, but the driver loop for
71 the simplifier is in SimplCore.lhs.
74 -----------------------------------------
75 *** IMPORTANT NOTE ***
76 -----------------------------------------
77 The simplifier used to guarantee that the output had no shadowing, but
78 it does not do so any more. (Actually, it never did!) The reason is
79 documented with simplifyArgs.
82 -----------------------------------------
83 *** IMPORTANT NOTE ***
84 -----------------------------------------
85 Many parts of the simplifier return a bunch of "floats" as well as an
86 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
88 All "floats" are let-binds, not case-binds, but some non-rec lets may
89 be unlifted (with RHS ok-for-speculation).
93 -----------------------------------------
94 ORGANISATION OF FUNCTIONS
95 -----------------------------------------
97 - simplify all top-level binders
98 - for NonRec, call simplRecOrTopPair
99 - for Rec, call simplRecBind
102 ------------------------------
103 simplExpr (applied lambda) ==> simplNonRecBind
104 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
105 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
107 ------------------------------
108 simplRecBind [binders already simplfied]
109 - use simplRecOrTopPair on each pair in turn
111 simplRecOrTopPair [binder already simplified]
112 Used for: recursive bindings (top level and nested)
113 top-level non-recursive bindings
115 - check for PreInlineUnconditionally
119 Used for: non-top-level non-recursive bindings
120 beta reductions (which amount to the same thing)
121 Because it can deal with strict arts, it takes a
122 "thing-inside" and returns an expression
124 - check for PreInlineUnconditionally
125 - simplify binder, including its IdInfo
134 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
135 Used for: binding case-binder and constr args in a known-constructor case
136 - check for PreInLineUnconditionally
140 ------------------------------
141 simplLazyBind: [binder already simplified, RHS not]
142 Used for: recursive bindings (top level and nested)
143 top-level non-recursive bindings
144 non-top-level, but *lazy* non-recursive bindings
145 [must not be strict or unboxed]
146 Returns floats + an augmented environment, not an expression
147 - substituteIdInfo and add result to in-scope
148 [so that rules are available in rec rhs]
151 - float if exposes constructor or PAP
155 completeNonRecX: [binder and rhs both simplified]
156 - if the the thing needs case binding (unlifted and not ok-for-spec)
162 completeLazyBind: [given a simplified RHS]
163 [used for both rec and non-rec bindings, top level and not]
164 - try PostInlineUnconditionally
165 - add unfolding [this is the only place we add an unfolding]
170 Right hand sides and arguments
171 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
172 In many ways we want to treat
173 (a) the right hand side of a let(rec), and
174 (b) a function argument
175 in the same way. But not always! In particular, we would
176 like to leave these arguments exactly as they are, so they
177 will match a RULE more easily.
182 It's harder to make the rule match if we ANF-ise the constructor,
183 or eta-expand the PAP:
185 f (let { a = g x; b = h x } in (a,b))
188 On the other hand if we see the let-defns
193 then we *do* want to ANF-ise and eta-expand, so that p and q
194 can be safely inlined.
196 Even floating lets out is a bit dubious. For let RHS's we float lets
197 out if that exposes a value, so that the value can be inlined more vigorously.
200 r = let x = e in (x,x)
202 Here, if we float the let out we'll expose a nice constructor. We did experiments
203 that showed this to be a generally good thing. But it was a bad thing to float
204 lets out unconditionally, because that meant they got allocated more often.
206 For function arguments, there's less reason to expose a constructor (it won't
207 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
208 So for the moment we don't float lets out of function arguments either.
213 For eta expansion, we want to catch things like
215 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
217 If the \x was on the RHS of a let, we'd eta expand to bring the two
218 lambdas together. And in general that's a good thing to do. Perhaps
219 we should eta expand wherever we find a (value) lambda? Then the eta
220 expansion at a let RHS can concentrate solely on the PAP case.
223 %************************************************************************
225 \subsection{Bindings}
227 %************************************************************************
230 simplTopBinds :: SimplEnv -> [InBind] -> SimplM [OutBind]
232 simplTopBinds env binds
233 = -- Put all the top-level binders into scope at the start
234 -- so that if a transformation rule has unexpectedly brought
235 -- anything into scope, then we don't get a complaint about that.
236 -- It's rather as if the top-level binders were imported.
237 simplLetBndrs env (bindersOfBinds binds) `thenSmpl` \ (env, bndrs') ->
238 simpl_binds env binds bndrs' `thenSmpl` \ (floats, _) ->
239 freeTick SimplifierDone `thenSmpl_`
240 returnSmpl (floatBinds floats)
242 -- We need to track the zapped top-level binders, because
243 -- they should have their fragile IdInfo zapped (notably occurrence info)
244 -- That's why we run down binds and bndrs' simultaneously.
245 simpl_binds :: SimplEnv -> [InBind] -> [OutId] -> SimplM (FloatsWith ())
246 simpl_binds env [] bs = ASSERT( null bs ) returnSmpl (emptyFloats env, ())
247 simpl_binds env (bind:binds) bs = simpl_bind env bind bs `thenSmpl` \ (floats,env) ->
248 addFloats env floats $ \env ->
249 simpl_binds env binds (drop_bs bind bs)
251 drop_bs (NonRec _ _) (_ : bs) = bs
252 drop_bs (Rec prs) bs = drop (length prs) bs
254 simpl_bind env bind bs
255 = getDOptsSmpl `thenSmpl` \ dflags ->
256 if dopt Opt_D_dump_inlinings dflags then
257 pprTrace "SimplBind" (ppr (bindersOf bind)) $ simpl_bind1 env bind bs
259 simpl_bind1 env bind bs
261 simpl_bind1 env (NonRec b r) (b':_) = simplRecOrTopPair env TopLevel b b' r
262 simpl_bind1 env (Rec pairs) bs' = simplRecBind env TopLevel pairs bs'
266 %************************************************************************
268 \subsection{simplNonRec}
270 %************************************************************************
272 simplNonRecBind is used for
273 * non-top-level non-recursive lets in expressions
277 * An unsimplified (binder, rhs) pair
278 * The env for the RHS. It may not be the same as the
279 current env because the bind might occur via (\x.E) arg
281 It uses the CPS form because the binding might be strict, in which
282 case we might discard the continuation:
283 let x* = error "foo" in (...x...)
285 It needs to turn unlifted bindings into a @case@. They can arise
286 from, say: (\x -> e) (4# + 3#)
289 simplNonRecBind :: SimplEnv
291 -> InExpr -> SimplEnv -- Arg, with its subst-env
292 -> OutType -- Type of thing computed by the context
293 -> (SimplEnv -> SimplM FloatsWithExpr) -- The body
294 -> SimplM FloatsWithExpr
296 simplNonRecBind env bndr rhs rhs_se cont_ty thing_inside
298 = pprPanic "simplNonRecBind" (ppr bndr <+> ppr rhs)
301 simplNonRecBind env bndr rhs rhs_se cont_ty thing_inside
302 = simplNonRecBind' env bndr rhs rhs_se cont_ty thing_inside
304 simplNonRecBind' env bndr rhs rhs_se cont_ty thing_inside
305 | preInlineUnconditionally env NotTopLevel bndr rhs
306 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
307 thing_inside (extendIdSubst env bndr (mkContEx rhs_se rhs))
309 | isStrictDmd (idNewDemandInfo bndr) || isStrictType bndr_ty -- A strict let
310 = -- Don't use simplBinder because that doesn't keep
311 -- fragile occurrence info in the substitution
312 simplLetBndr env bndr `thenSmpl` \ (env, bndr1) ->
313 simplStrictArg AnRhs env rhs rhs_se (idType bndr1) cont_ty $ \ env1 rhs1 ->
315 -- Now complete the binding and simplify the body
317 -- simplLetBndr doesn't deal with the IdInfo, so we must
318 -- do so here (c.f. simplLazyBind)
319 bndr2 = bndr1 `setIdInfo` simplIdInfo env (idInfo bndr)
320 env2 = modifyInScope env1 bndr2 bndr2
322 if needsCaseBinding bndr_ty rhs1
324 thing_inside env2 `thenSmpl` \ (floats, body) ->
325 returnSmpl (emptyFloats env2, Case rhs1 bndr2 (exprType body)
326 [(DEFAULT, [], wrapFloats floats body)])
328 completeNonRecX env2 True {- strict -} bndr bndr2 rhs1 thing_inside
330 | otherwise -- Normal, lazy case
331 = -- Don't use simplBinder because that doesn't keep
332 -- fragile occurrence info in the substitution
333 simplLetBndr env bndr `thenSmpl` \ (env, bndr') ->
334 simplLazyBind env NotTopLevel NonRecursive
335 bndr bndr' rhs rhs_se `thenSmpl` \ (floats, env) ->
336 addFloats env floats thing_inside
339 bndr_ty = idType bndr
342 A specialised variant of simplNonRec used when the RHS is already simplified, notably
343 in knownCon. It uses case-binding where necessary.
346 simplNonRecX :: SimplEnv
347 -> InId -- Old binder
348 -> OutExpr -- Simplified RHS
349 -> (SimplEnv -> SimplM FloatsWithExpr)
350 -> SimplM FloatsWithExpr
352 simplNonRecX env bndr new_rhs thing_inside
353 | needsCaseBinding (idType bndr) new_rhs
354 -- Make this test *before* the preInlineUnconditionally
355 -- Consider case I# (quotInt# x y) of
356 -- I# v -> let w = J# v in ...
357 -- If we gaily inline (quotInt# x y) for v, we end up building an
359 -- let w = J# (quotInt# x y) in ...
360 -- because quotInt# can fail.
361 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
362 thing_inside env `thenSmpl` \ (floats, body) ->
363 let body' = wrapFloats floats body in
364 returnSmpl (emptyFloats env, Case new_rhs bndr' (exprType body') [(DEFAULT, [], body')])
366 | preInlineUnconditionally env NotTopLevel bndr new_rhs
367 -- This happens; for example, the case_bndr during case of
368 -- known constructor: case (a,b) of x { (p,q) -> ... }
369 -- Here x isn't mentioned in the RHS, so we don't want to
370 -- create the (dead) let-binding let x = (a,b) in ...
372 -- Similarly, single occurrences can be inlined vigourously
373 -- e.g. case (f x, g y) of (a,b) -> ....
374 -- If a,b occur once we can avoid constructing the let binding for them.
375 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
378 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
379 completeNonRecX env False {- Non-strict; pessimistic -}
380 bndr bndr' new_rhs thing_inside
382 completeNonRecX env is_strict old_bndr new_bndr new_rhs thing_inside
383 = mkAtomicArgs is_strict
384 True {- OK to float unlifted -}
385 new_rhs `thenSmpl` \ (aux_binds, rhs2) ->
387 -- Make the arguments atomic if necessary,
388 -- adding suitable bindings
389 addAtomicBindsE env (fromOL aux_binds) $ \ env ->
390 completeLazyBind env NotTopLevel
391 old_bndr new_bndr rhs2 `thenSmpl` \ (floats, env) ->
392 addFloats env floats thing_inside
396 %************************************************************************
398 \subsection{Lazy bindings}
400 %************************************************************************
402 simplRecBind is used for
403 * recursive bindings only
406 simplRecBind :: SimplEnv -> TopLevelFlag
407 -> [(InId, InExpr)] -> [OutId]
408 -> SimplM (FloatsWith SimplEnv)
409 simplRecBind env top_lvl pairs bndrs'
410 = go env pairs bndrs' `thenSmpl` \ (floats, env) ->
411 returnSmpl (flattenFloats floats, env)
413 go env [] _ = returnSmpl (emptyFloats env, env)
415 go env ((bndr, rhs) : pairs) (bndr' : bndrs')
416 = simplRecOrTopPair env top_lvl bndr bndr' rhs `thenSmpl` \ (floats, env) ->
417 addFloats env floats (\env -> go env pairs bndrs')
421 simplRecOrTopPair is used for
422 * recursive bindings (whether top level or not)
423 * top-level non-recursive bindings
425 It assumes the binder has already been simplified, but not its IdInfo.
428 simplRecOrTopPair :: SimplEnv
430 -> InId -> OutId -- Binder, both pre-and post simpl
431 -> InExpr -- The RHS and its environment
432 -> SimplM (FloatsWith SimplEnv)
434 simplRecOrTopPair env top_lvl bndr bndr' rhs
435 | preInlineUnconditionally env top_lvl bndr rhs -- Check for unconditional inline
436 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
437 returnSmpl (emptyFloats env, extendIdSubst env bndr (mkContEx env rhs))
440 = simplLazyBind env top_lvl Recursive bndr bndr' rhs env
441 -- May not actually be recursive, but it doesn't matter
445 simplLazyBind is used for
446 * recursive bindings (whether top level or not)
447 * top-level non-recursive bindings
448 * non-top-level *lazy* non-recursive bindings
450 [Thus it deals with the lazy cases from simplNonRecBind, and all cases
451 from SimplRecOrTopBind]
454 1. It assumes that the binder is *already* simplified,
455 and is in scope, but not its IdInfo
457 2. It assumes that the binder type is lifted.
459 3. It does not check for pre-inline-unconditionallly;
460 that should have been done already.
463 simplLazyBind :: SimplEnv
464 -> TopLevelFlag -> RecFlag
465 -> InId -> OutId -- Binder, both pre-and post simpl
466 -> InExpr -> SimplEnv -- The RHS and its environment
467 -> SimplM (FloatsWith SimplEnv)
469 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
470 = let -- Transfer the IdInfo of the original binder to the new binder
471 -- This is crucial: we must preserve
475 -- etc. To do this we must apply the current substitution,
476 -- which incorporates earlier substitutions in this very letrec group.
478 -- NB 1. We do this *before* processing the RHS of the binder, so that
479 -- its substituted rules are visible in its own RHS.
480 -- This is important. Manuel found cases where he really, really
481 -- wanted a RULE for a recursive function to apply in that function's
482 -- own right-hand side.
484 -- NB 2: We do not transfer the arity (see Subst.substIdInfo)
485 -- The arity of an Id should not be visible
486 -- in its own RHS, else we eta-reduce
490 -- which isn't sound. And it makes the arity in f's IdInfo greater than
491 -- the manifest arity, which isn't good.
492 -- The arity will get added later.
494 -- NB 3: It's important that we *do* transer the loop-breaker OccInfo,
495 -- because that's what stops the Id getting inlined infinitely, in the body
498 -- NB 4: does no harm for non-recursive bindings
500 bndr2 = bndr1 `setIdInfo` simplIdInfo env (idInfo bndr)
501 env1 = modifyInScope env bndr2 bndr2
502 rhs_env = setInScope rhs_se env1
503 is_top_level = isTopLevel top_lvl
504 ok_float_unlifted = not is_top_level && isNonRec is_rec
505 rhs_cont = mkRhsStop (idType bndr1)
507 -- Simplify the RHS; note the mkRhsStop, which tells
508 -- the simplifier that this is the RHS of a let.
509 simplExprF rhs_env rhs rhs_cont `thenSmpl` \ (floats, rhs1) ->
511 -- If any of the floats can't be floated, give up now
512 -- (The allLifted predicate says True for empty floats.)
513 if (not ok_float_unlifted && not (allLifted floats)) then
514 completeLazyBind env1 top_lvl bndr bndr2
515 (wrapFloats floats rhs1)
518 -- ANF-ise a constructor or PAP rhs
519 mkAtomicArgs False {- Not strict -}
520 ok_float_unlifted rhs1 `thenSmpl` \ (aux_binds, rhs2) ->
522 -- If the result is a PAP, float the floats out, else wrap them
523 -- By this time it's already been ANF-ised (if necessary)
524 if isEmptyFloats floats && isNilOL aux_binds then -- Shortcut a common case
525 completeLazyBind env1 top_lvl bndr bndr2 rhs2
527 else if is_top_level || exprIsTrivial rhs2 || exprIsHNF rhs2 then
528 -- WARNING: long dodgy argument coming up
529 -- WANTED: a better way to do this
531 -- We can't use "exprIsCheap" instead of exprIsHNF,
532 -- because that causes a strictness bug.
533 -- x = let y* = E in case (scc y) of { T -> F; F -> T}
534 -- The case expression is 'cheap', but it's wrong to transform to
535 -- y* = E; x = case (scc y) of {...}
536 -- Either we must be careful not to float demanded non-values, or
537 -- we must use exprIsHNF for the test, which ensures that the
538 -- thing is non-strict. So exprIsHNF => bindings are non-strict
539 -- I think. The WARN below tests for this.
541 -- We use exprIsTrivial here because we want to reveal lone variables.
542 -- E.g. let { x = letrec { y = E } in y } in ...
543 -- Here we definitely want to float the y=E defn.
544 -- exprIsHNF definitely isn't right for that.
546 -- Again, the floated binding can't be strict; if it's recursive it'll
547 -- be non-strict; if it's non-recursive it'd be inlined.
549 -- Note [SCC-and-exprIsTrivial]
551 -- y = let { x* = E } in scc "foo" x
552 -- then we do *not* want to float out the x binding, because
553 -- it's strict! Fortunately, exprIsTrivial replies False to
556 -- There's a subtlety here. There may be a binding (x* = e) in the
557 -- floats, where the '*' means 'will be demanded'. So is it safe
558 -- to float it out? Answer no, but it won't matter because
559 -- we only float if (a) arg' is a WHNF, or (b) it's going to top level
560 -- and so there can't be any 'will be demanded' bindings in the floats.
562 ASSERT2( is_top_level || not (any demanded_float (floatBinds floats)),
563 ppr (filter demanded_float (floatBinds floats)) )
565 tick LetFloatFromLet `thenSmpl_` (
566 addFloats env1 floats $ \ env2 ->
567 addAtomicBinds env2 (fromOL aux_binds) $ \ env3 ->
568 completeLazyBind env3 top_lvl bndr bndr2 rhs2)
571 completeLazyBind env1 top_lvl bndr bndr2 (wrapFloats floats rhs1)
574 demanded_float (NonRec b r) = isStrictDmd (idNewDemandInfo b) && not (isUnLiftedType (idType b))
575 -- Unlifted-type (cheap-eagerness) lets may well have a demanded flag on them
576 demanded_float (Rec _) = False
581 %************************************************************************
583 \subsection{Completing a lazy binding}
585 %************************************************************************
588 * deals only with Ids, not TyVars
589 * takes an already-simplified binder and RHS
590 * is used for both recursive and non-recursive bindings
591 * is used for both top-level and non-top-level bindings
593 It does the following:
594 - tries discarding a dead binding
595 - tries PostInlineUnconditionally
596 - add unfolding [this is the only place we add an unfolding]
599 It does *not* attempt to do let-to-case. Why? Because it is used for
600 - top-level bindings (when let-to-case is impossible)
601 - many situations where the "rhs" is known to be a WHNF
602 (so let-to-case is inappropriate).
605 completeLazyBind :: SimplEnv
606 -> TopLevelFlag -- Flag stuck into unfolding
607 -> InId -- Old binder
608 -> OutId -- New binder
609 -> OutExpr -- Simplified RHS
610 -> SimplM (FloatsWith SimplEnv)
611 -- We return a new SimplEnv, because completeLazyBind may choose to do its work
612 -- by extending the substitution (e.g. let x = y in ...)
613 -- The new binding (if any) is returned as part of the floats.
614 -- NB: the returned SimplEnv has the right SubstEnv, but you should
615 -- (as usual) use the in-scope-env from the floats
617 completeLazyBind env top_lvl old_bndr new_bndr new_rhs
618 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
619 = -- Drop the binding
620 tick (PostInlineUnconditionally old_bndr) `thenSmpl_`
621 returnSmpl (emptyFloats env, extendIdSubst env old_bndr (DoneEx new_rhs))
622 -- Use the substitution to make quite, quite sure that the substitution
623 -- will happen, since we are going to discard the binding
628 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
630 -- Add the unfolding *only* for non-loop-breakers
631 -- Making loop breakers not have an unfolding at all
632 -- means that we can avoid tests in exprIsConApp, for example.
633 -- This is important: if exprIsConApp says 'yes' for a recursive
634 -- thing, then we can get into an infinite loop
636 -- If the unfolding is a value, the demand info may
637 -- go pear-shaped, so we nuke it. Example:
639 -- case x of (p,q) -> h p q x
640 -- Here x is certainly demanded. But after we've nuked
641 -- the case, we'll get just
642 -- let x = (a,b) in h a b x
643 -- and now x is not demanded (I'm assuming h is lazy)
644 -- This really happens. Similarly
645 -- let f = \x -> e in ...f..f...
646 -- After inling f at some of its call sites the original binding may
647 -- (for example) be no longer strictly demanded.
648 -- The solution here is a bit ad hoc...
649 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
650 final_info | loop_breaker = new_bndr_info
651 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
652 | otherwise = info_w_unf
654 final_id = new_bndr `setIdInfo` final_info
656 -- These seqs forces the Id, and hence its IdInfo,
657 -- and hence any inner substitutions
659 returnSmpl (unitFloat env final_id new_rhs, env)
662 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
663 loop_breaker = isLoopBreaker occ_info
664 old_info = idInfo old_bndr
665 occ_info = occInfo old_info
670 %************************************************************************
672 \subsection[Simplify-simplExpr]{The main function: simplExpr}
674 %************************************************************************
676 The reason for this OutExprStuff stuff is that we want to float *after*
677 simplifying a RHS, not before. If we do so naively we get quadratic
678 behaviour as things float out.
680 To see why it's important to do it after, consider this (real) example:
694 a -- Can't inline a this round, cos it appears twice
698 Each of the ==> steps is a round of simplification. We'd save a
699 whole round if we float first. This can cascade. Consider
704 let f = let d1 = ..d.. in \y -> e
708 in \x -> ...(\y ->e)...
710 Only in this second round can the \y be applied, and it
711 might do the same again.
715 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
716 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
718 expr_ty' = substTy env (exprType expr)
719 -- The type in the Stop continuation, expr_ty', is usually not used
720 -- It's only needed when discarding continuations after finding
721 -- a function that returns bottom.
722 -- Hence the lazy substitution
725 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
726 -- Simplify an expression, given a continuation
727 simplExprC env expr cont
728 = simplExprF env expr cont `thenSmpl` \ (floats, expr) ->
729 returnSmpl (wrapFloats floats expr)
731 simplExprF :: SimplEnv -> InExpr -> SimplCont -> SimplM FloatsWithExpr
732 -- Simplify an expression, returning floated binds
734 simplExprF env (Var v) cont = simplVar env v cont
735 simplExprF env (Lit lit) cont = rebuild env (Lit lit) cont
736 simplExprF env expr@(Lam _ _) cont = simplLam env expr cont
737 simplExprF env (Note note expr) cont = simplNote env note expr cont
738 simplExprF env (App fun arg) cont = simplExprF env fun (ApplyTo NoDup arg env cont)
740 simplExprF env (Type ty) cont
741 = ASSERT( contIsRhsOrArg cont )
742 simplType env ty `thenSmpl` \ ty' ->
743 rebuild env (Type ty') cont
745 simplExprF env (Case scrut bndr case_ty alts) cont
746 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
747 = -- Simplify the scrutinee with a Select continuation
748 simplExprF env scrut (Select NoDup bndr alts env cont)
751 = -- If case-of-case is off, simply simplify the case expression
752 -- in a vanilla Stop context, and rebuild the result around it
753 simplExprC env scrut case_cont `thenSmpl` \ case_expr' ->
754 rebuild env case_expr' cont
756 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
757 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
759 simplExprF env (Let (Rec pairs) body) cont
760 = simplLetBndrs env (map fst pairs) `thenSmpl` \ (env, bndrs') ->
761 -- NB: bndrs' don't have unfoldings or rules
762 -- We add them as we go down
764 simplRecBind env NotTopLevel pairs bndrs' `thenSmpl` \ (floats, env) ->
765 addFloats env floats $ \ env ->
766 simplExprF env body cont
768 -- A non-recursive let is dealt with by simplNonRecBind
769 simplExprF env (Let (NonRec bndr rhs) body) cont
770 = simplNonRecBind env bndr rhs env (contResultType cont) $ \ env ->
771 simplExprF env body cont
774 ---------------------------------
775 simplType :: SimplEnv -> InType -> SimplM OutType
776 -- Kept monadic just so we can do the seqType
778 = seqType new_ty `seq` returnSmpl new_ty
780 new_ty = substTy env ty
784 %************************************************************************
788 %************************************************************************
791 simplLam env fun cont
794 zap_it = mkLamBndrZapper fun (countArgs cont)
795 cont_ty = contResultType cont
797 -- Type-beta reduction
798 go env (Lam bndr body) (ApplyTo _ (Type ty_arg) arg_se body_cont)
799 = ASSERT( isTyVar bndr )
800 tick (BetaReduction bndr) `thenSmpl_`
801 simplType (setInScope arg_se env) ty_arg `thenSmpl` \ ty_arg' ->
802 go (extendTvSubst env bndr ty_arg') body body_cont
804 -- Ordinary beta reduction
805 go env (Lam bndr body) cont@(ApplyTo _ arg arg_se body_cont)
806 = tick (BetaReduction bndr) `thenSmpl_`
807 simplNonRecBind env (zap_it bndr) arg arg_se cont_ty $ \ env ->
808 go env body body_cont
810 -- Not enough args, so there are real lambdas left to put in the result
811 go env lam@(Lam _ _) cont
812 = simplLamBndrs env bndrs `thenSmpl` \ (env, bndrs') ->
813 simplExpr env body `thenSmpl` \ body' ->
814 mkLam env bndrs' body' cont `thenSmpl` \ (floats, new_lam) ->
815 addFloats env floats $ \ env ->
816 rebuild env new_lam cont
818 (bndrs,body) = collectBinders lam
820 -- Exactly enough args
821 go env expr cont = simplExprF env expr cont
823 mkLamBndrZapper :: CoreExpr -- Function
824 -> Int -- Number of args supplied, *including* type args
825 -> Id -> Id -- Use this to zap the binders
826 mkLamBndrZapper fun n_args
827 | n_args >= n_params fun = \b -> b -- Enough args
828 | otherwise = \b -> zapLamIdInfo b
830 -- NB: we count all the args incl type args
831 -- so we must count all the binders (incl type lambdas)
832 n_params (Note _ e) = n_params e
833 n_params (Lam b e) = 1 + n_params e
834 n_params other = 0::Int
838 %************************************************************************
842 %************************************************************************
845 simplNote env (Coerce to from) body cont
847 addCoerce s1 k1 cont -- Drop redundant coerces. This can happen if a polymoprhic
848 -- (coerce a b e) is instantiated with a=ty1 b=ty2 and the
849 -- two are the same. This happens a lot in Happy-generated parsers
850 | s1 `coreEqType` k1 = cont
852 addCoerce s1 k1 (CoerceIt t1 cont)
853 -- coerce T1 S1 (coerce S1 K1 e)
856 -- coerce T1 K1 e, otherwise
858 -- For example, in the initial form of a worker
859 -- we may find (coerce T (coerce S (\x.e))) y
860 -- and we'd like it to simplify to e[y/x] in one round
862 | t1 `coreEqType` k1 = cont -- The coerces cancel out
863 | otherwise = CoerceIt t1 cont -- They don't cancel, but
864 -- the inner one is redundant
866 addCoerce t1t2 s1s2 (ApplyTo dup arg arg_se cont)
867 | not (isTypeArg arg), -- This whole case only works for value args
868 -- Could upgrade to have equiv thing for type apps too
869 Just (s1, s2) <- splitFunTy_maybe s1s2
870 -- (coerce (T1->T2) (S1->S2) F) E
872 -- coerce T2 S2 (F (coerce S1 T1 E))
874 -- t1t2 must be a function type, T1->T2, because it's applied to something
875 -- but s1s2 might conceivably not be
877 -- When we build the ApplyTo we can't mix the out-types
878 -- with the InExpr in the argument, so we simply substitute
879 -- to make it all consistent. It's a bit messy.
880 -- But it isn't a common case.
882 (t1,t2) = splitFunTy t1t2
883 new_arg = mkCoerce2 s1 t1 (substExpr arg_env arg)
884 arg_env = setInScope arg_se env
886 ApplyTo dup new_arg (zapSubstEnv env) (addCoerce t2 s2 cont)
888 addCoerce to' _ cont = CoerceIt to' cont
890 simplType env to `thenSmpl` \ to' ->
891 simplType env from `thenSmpl` \ from' ->
892 simplExprF env body (addCoerce to' from' cont)
895 -- Hack: we only distinguish subsumed cost centre stacks for the purposes of
896 -- inlining. All other CCCSs are mapped to currentCCS.
897 simplNote env (SCC cc) e cont
898 = simplExpr (setEnclosingCC env currentCCS) e `thenSmpl` \ e' ->
899 rebuild env (mkSCC cc e') cont
901 simplNote env InlineCall e cont
902 = simplExprF env e (InlinePlease cont)
904 -- See notes with SimplMonad.inlineMode
905 simplNote env InlineMe e cont
906 | contIsRhsOrArg cont -- Totally boring continuation; see notes above
907 = -- Don't inline inside an INLINE expression
908 simplExpr (setMode inlineMode env ) e `thenSmpl` \ e' ->
909 rebuild env (mkInlineMe e') cont
911 | otherwise -- Dissolve the InlineMe note if there's
912 -- an interesting context of any kind to combine with
913 -- (even a type application -- anything except Stop)
914 = simplExprF env e cont
916 simplNote env (CoreNote s) e cont
917 = simplExpr env e `thenSmpl` \ e' ->
918 rebuild env (Note (CoreNote s) e') cont
922 %************************************************************************
924 \subsection{Dealing with calls}
926 %************************************************************************
929 simplVar env var cont
930 = case substId env var of
931 DoneEx e -> simplExprF (zapSubstEnv env) e cont
932 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
933 DoneId var1 occ -> completeCall (zapSubstEnv env) var1 occ cont
934 -- Note [zapSubstEnv]
935 -- The template is already simplified, so don't re-substitute.
936 -- This is VITAL. Consider
938 -- let y = \z -> ...x... in
940 -- We'll clone the inner \x, adding x->x' in the id_subst
941 -- Then when we inline y, we must *not* replace x by x' in
942 -- the inlined copy!!
944 ---------------------------------------------------------
945 -- Dealing with a call site
947 completeCall env var occ_info cont
948 = -- Simplify the arguments
949 getDOptsSmpl `thenSmpl` \ dflags ->
951 chkr = getSwitchChecker env
952 (args, call_cont, inline_call) = getContArgs chkr var cont
955 simplifyArgs env fn_ty args (contResultType call_cont) $ \ env args ->
957 -- Next, look for rules or specialisations that match
959 -- It's important to simplify the args first, because the rule-matcher
960 -- doesn't do substitution as it goes. We don't want to use subst_args
961 -- (defined in the 'where') because that throws away useful occurrence info,
962 -- and perhaps-very-important specialisations.
964 -- Some functions have specialisations *and* are strict; in this case,
965 -- we don't want to inline the wrapper of the non-specialised thing; better
966 -- to call the specialised thing instead.
967 -- We used to use the black-listing mechanism to ensure that inlining of
968 -- the wrapper didn't occur for things that have specialisations till a
969 -- later phase, so but now we just try RULES first
971 -- You might think that we shouldn't apply rules for a loop breaker:
972 -- doing so might give rise to an infinite loop, because a RULE is
973 -- rather like an extra equation for the function:
974 -- RULE: f (g x) y = x+y
977 -- But it's too drastic to disable rules for loop breakers.
978 -- Even the foldr/build rule would be disabled, because foldr
979 -- is recursive, and hence a loop breaker:
980 -- foldr k z (build g) = g k z
981 -- So it's up to the programmer: rules can cause divergence
984 in_scope = getInScope env
986 maybe_rule = case activeRule env of
987 Nothing -> Nothing -- No rules apply
988 Just act_fn -> lookupRule act_fn in_scope rules var args
991 Just (rule_name, rule_rhs) ->
992 tick (RuleFired rule_name) `thenSmpl_`
993 (if dopt Opt_D_dump_inlinings dflags then
994 pprTrace "Rule fired" (vcat [
995 text "Rule:" <+> ftext rule_name,
996 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
997 text "After: " <+> pprCoreExpr rule_rhs,
998 text "Cont: " <+> ppr call_cont])
1001 simplExprF env rule_rhs call_cont ;
1003 Nothing -> -- No rules
1005 -- Next, look for an inlining
1007 arg_infos = [ interestingArg arg | arg <- args, isValArg arg]
1009 interesting_cont = interestingCallContext (notNull args)
1013 active_inline = activeInline env var occ_info
1014 maybe_inline = callSiteInline dflags active_inline inline_call occ_info
1015 var arg_infos interesting_cont
1017 case maybe_inline of {
1018 Just unfolding -- There is an inlining!
1019 -> tick (UnfoldingDone var) `thenSmpl_`
1020 (if dopt Opt_D_dump_inlinings dflags then
1021 pprTrace "Inlining done" (vcat [
1022 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1023 text "Inlined fn: " <+> ppr unfolding,
1024 text "Cont: " <+> ppr call_cont])
1027 makeThatCall env var unfolding args call_cont
1030 Nothing -> -- No inlining!
1033 rebuild env (mkApps (Var var) args) call_cont
1036 makeThatCall :: SimplEnv
1038 -> InExpr -- Inlined function rhs
1039 -> [OutExpr] -- Arguments, already simplified
1040 -> SimplCont -- After the call
1041 -> SimplM FloatsWithExpr
1042 -- Similar to simplLam, but this time
1043 -- the arguments are already simplified
1044 makeThatCall orig_env var fun@(Lam _ _) args cont
1045 = go orig_env fun args
1047 zap_it = mkLamBndrZapper fun (length args)
1049 -- Type-beta reduction
1050 go env (Lam bndr body) (Type ty_arg : args)
1051 = ASSERT( isTyVar bndr )
1052 tick (BetaReduction bndr) `thenSmpl_`
1053 go (extendTvSubst env bndr ty_arg) body args
1055 -- Ordinary beta reduction
1056 go env (Lam bndr body) (arg : args)
1057 = tick (BetaReduction bndr) `thenSmpl_`
1058 simplNonRecX env (zap_it bndr) arg $ \ env ->
1061 -- Not enough args, so there are real lambdas left to put in the result
1063 = simplExprF env fun (pushContArgs orig_env args cont)
1064 -- NB: orig_env; the correct environment to capture with
1065 -- the arguments.... env has been augmented with substitutions
1066 -- from the beta reductions.
1068 makeThatCall env var fun args cont
1069 = simplExprF env fun (pushContArgs env args cont)
1073 %************************************************************************
1075 \subsection{Arguments}
1077 %************************************************************************
1080 ---------------------------------------------------------
1081 -- Simplifying the arguments of a call
1083 simplifyArgs :: SimplEnv
1084 -> OutType -- Type of the function
1085 -> [(InExpr, SimplEnv, Bool)] -- Details of the arguments
1086 -> OutType -- Type of the continuation
1087 -> (SimplEnv -> [OutExpr] -> SimplM FloatsWithExpr)
1088 -> SimplM FloatsWithExpr
1090 -- [CPS-like because of strict arguments]
1092 -- Simplify the arguments to a call.
1093 -- This part of the simplifier may break the no-shadowing invariant
1095 -- f (...(\a -> e)...) (case y of (a,b) -> e')
1096 -- where f is strict in its second arg
1097 -- If we simplify the innermost one first we get (...(\a -> e)...)
1098 -- Simplifying the second arg makes us float the case out, so we end up with
1099 -- case y of (a,b) -> f (...(\a -> e)...) e'
1100 -- So the output does not have the no-shadowing invariant. However, there is
1101 -- no danger of getting name-capture, because when the first arg was simplified
1102 -- we used an in-scope set that at least mentioned all the variables free in its
1103 -- static environment, and that is enough.
1105 -- We can't just do innermost first, or we'd end up with a dual problem:
1106 -- case x of (a,b) -> f e (...(\a -> e')...)
1108 -- I spent hours trying to recover the no-shadowing invariant, but I just could
1109 -- not think of an elegant way to do it. The simplifier is already knee-deep in
1110 -- continuations. We have to keep the right in-scope set around; AND we have
1111 -- to get the effect that finding (error "foo") in a strict arg position will
1112 -- discard the entire application and replace it with (error "foo"). Getting
1113 -- all this at once is TOO HARD!
1115 simplifyArgs env fn_ty args cont_ty thing_inside
1116 = go env fn_ty args thing_inside
1118 go env fn_ty [] thing_inside = thing_inside env []
1119 go env fn_ty (arg:args) thing_inside = simplifyArg env fn_ty arg cont_ty $ \ env arg' ->
1120 go env (applyTypeToArg fn_ty arg') args $ \ env args' ->
1121 thing_inside env (arg':args')
1123 simplifyArg env fn_ty (Type ty_arg, se, _) cont_ty thing_inside
1124 = simplType (setInScope se env) ty_arg `thenSmpl` \ new_ty_arg ->
1125 thing_inside env (Type new_ty_arg)
1127 simplifyArg env fn_ty (val_arg, arg_se, is_strict) cont_ty thing_inside
1129 = simplStrictArg AnArg env val_arg arg_se arg_ty cont_ty thing_inside
1131 | otherwise -- Lazy argument
1132 -- DO NOT float anything outside, hence simplExprC
1133 -- There is no benefit (unlike in a let-binding), and we'd
1134 -- have to be very careful about bogus strictness through
1135 -- floating a demanded let.
1136 = simplExprC (setInScope arg_se env) val_arg
1137 (mkBoringStop arg_ty) `thenSmpl` \ arg1 ->
1138 thing_inside env arg1
1140 arg_ty = funArgTy fn_ty
1143 simplStrictArg :: LetRhsFlag
1144 -> SimplEnv -- The env of the call
1145 -> InExpr -> SimplEnv -- The arg plus its env
1146 -> OutType -- arg_ty: type of the argument
1147 -> OutType -- cont_ty: Type of thing computed by the context
1148 -> (SimplEnv -> OutExpr -> SimplM FloatsWithExpr)
1149 -- Takes an expression of type rhs_ty,
1150 -- returns an expression of type cont_ty
1151 -- The env passed to this continuation is the
1152 -- env of the call, plus any new in-scope variables
1153 -> SimplM FloatsWithExpr -- An expression of type cont_ty
1155 simplStrictArg is_rhs call_env arg arg_env arg_ty cont_ty thing_inside
1156 = simplExprF (setInScope arg_env call_env) arg
1157 (ArgOf is_rhs arg_ty cont_ty (\ new_env -> thing_inside (setInScope call_env new_env)))
1158 -- Notice the way we use arg_env (augmented with in-scope vars from call_env)
1159 -- to simplify the argument
1160 -- and call-env (augmented with in-scope vars from the arg) to pass to the continuation
1164 %************************************************************************
1166 \subsection{mkAtomicArgs}
1168 %************************************************************************
1170 mkAtomicArgs takes a putative RHS, checks whether it's a PAP or
1171 constructor application and, if so, converts it to ANF, so that the
1172 resulting thing can be inlined more easily. Thus
1179 There are three sorts of binding context, specified by the two
1185 N N Top-level or recursive Only bind args of lifted type
1187 N Y Non-top-level and non-recursive, Bind args of lifted type, or
1188 but lazy unlifted-and-ok-for-speculation
1190 Y Y Non-top-level, non-recursive, Bind all args
1191 and strict (demanded)
1198 there is no point in transforming to
1200 x = case (y div# z) of r -> MkC r
1202 because the (y div# z) can't float out of the let. But if it was
1203 a *strict* let, then it would be a good thing to do. Hence the
1204 context information.
1207 mkAtomicArgs :: Bool -- A strict binding
1208 -> Bool -- OK to float unlifted args
1210 -> SimplM (OrdList (OutId,OutExpr), -- The floats (unusually) may include
1211 OutExpr) -- things that need case-binding,
1212 -- if the strict-binding flag is on
1214 mkAtomicArgs is_strict ok_float_unlifted rhs
1215 | (Var fun, args) <- collectArgs rhs, -- It's an application
1216 isDataConWorkId fun || valArgCount args < idArity fun -- And it's a constructor or PAP
1217 = go fun nilOL [] args -- Have a go
1219 | otherwise = bale_out -- Give up
1222 bale_out = returnSmpl (nilOL, rhs)
1224 go fun binds rev_args []
1225 = returnSmpl (binds, mkApps (Var fun) (reverse rev_args))
1227 go fun binds rev_args (arg : args)
1228 | exprIsTrivial arg -- Easy case
1229 = go fun binds (arg:rev_args) args
1231 | not can_float_arg -- Can't make this arg atomic
1232 = bale_out -- ... so give up
1234 | otherwise -- Don't forget to do it recursively
1235 -- E.g. x = a:b:c:[]
1236 = mkAtomicArgs is_strict ok_float_unlifted arg `thenSmpl` \ (arg_binds, arg') ->
1237 newId FSLIT("a") arg_ty `thenSmpl` \ arg_id ->
1238 go fun ((arg_binds `snocOL` (arg_id,arg')) `appOL` binds)
1239 (Var arg_id : rev_args) args
1241 arg_ty = exprType arg
1242 can_float_arg = is_strict
1243 || not (isUnLiftedType arg_ty)
1244 || (ok_float_unlifted && exprOkForSpeculation arg)
1247 addAtomicBinds :: SimplEnv -> [(OutId,OutExpr)]
1248 -> (SimplEnv -> SimplM (FloatsWith a))
1249 -> SimplM (FloatsWith a)
1250 addAtomicBinds env [] thing_inside = thing_inside env
1251 addAtomicBinds env ((v,r):bs) thing_inside = addAuxiliaryBind env (NonRec v r) $ \ env ->
1252 addAtomicBinds env bs thing_inside
1254 addAtomicBindsE :: SimplEnv -> [(OutId,OutExpr)]
1255 -> (SimplEnv -> SimplM FloatsWithExpr)
1256 -> SimplM FloatsWithExpr
1257 -- Same again, but this time we're in an expression context,
1258 -- and may need to do some case bindings
1260 addAtomicBindsE env [] thing_inside
1262 addAtomicBindsE env ((v,r):bs) thing_inside
1263 | needsCaseBinding (idType v) r
1264 = addAtomicBindsE (addNewInScopeIds env [v]) bs thing_inside `thenSmpl` \ (floats, expr) ->
1265 WARN( exprIsTrivial expr, ppr v <+> pprCoreExpr expr )
1266 (let body = wrapFloats floats expr in
1267 returnSmpl (emptyFloats env, Case r v (exprType body) [(DEFAULT,[],body)]))
1270 = addAuxiliaryBind env (NonRec v r) $ \ env ->
1271 addAtomicBindsE env bs thing_inside
1275 %************************************************************************
1277 \subsection{The main rebuilder}
1279 %************************************************************************
1282 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM FloatsWithExpr
1284 rebuild env expr (Stop _ _ _) = rebuildDone env expr
1285 rebuild env expr (ArgOf _ _ _ cont_fn) = cont_fn env expr
1286 rebuild env expr (CoerceIt to_ty cont) = rebuild env (mkCoerce to_ty expr) cont
1287 rebuild env expr (InlinePlease cont) = rebuild env (Note InlineCall expr) cont
1288 rebuild env expr (Select _ bndr alts se cont) = rebuildCase (setInScope se env) expr bndr alts cont
1289 rebuild env expr (ApplyTo _ arg se cont) = rebuildApp (setInScope se env) expr arg cont
1291 rebuildApp env fun arg cont
1292 = simplExpr env arg `thenSmpl` \ arg' ->
1293 rebuild env (App fun arg') cont
1295 rebuildDone env expr = returnSmpl (emptyFloats env, expr)
1299 %************************************************************************
1301 \subsection{Functions dealing with a case}
1303 %************************************************************************
1305 Blob of helper functions for the "case-of-something-else" situation.
1308 ---------------------------------------------------------
1309 -- Eliminate the case if possible
1311 rebuildCase :: SimplEnv
1312 -> OutExpr -- Scrutinee
1313 -> InId -- Case binder
1314 -> [InAlt] -- Alternatives (inceasing order)
1316 -> SimplM FloatsWithExpr
1318 rebuildCase env scrut case_bndr alts cont
1319 | Just (con,args) <- exprIsConApp_maybe scrut
1320 -- Works when the scrutinee is a variable with a known unfolding
1321 -- as well as when it's an explicit constructor application
1322 = knownCon env (DataAlt con) args case_bndr alts cont
1324 | Lit lit <- scrut -- No need for same treatment as constructors
1325 -- because literals are inlined more vigorously
1326 = knownCon env (LitAlt lit) [] case_bndr alts cont
1329 = -- Prepare the alternatives.
1330 prepareAlts scrut case_bndr alts `thenSmpl` \ (better_alts, handled_cons) ->
1332 -- Prepare the continuation;
1333 -- The new subst_env is in place
1334 prepareCaseCont env better_alts cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1335 addFloats env floats $ \ env ->
1338 -- The case expression is annotated with the result type of the continuation
1339 -- This may differ from the type originally on the case. For example
1340 -- case(T) (case(Int#) a of { True -> 1#; False -> 0# }) of
1343 -- let j a# = <blob>
1344 -- in case(T) a of { True -> j 1#; False -> j 0# }
1345 -- Note that the case that scrutinises a now returns a T not an Int#
1346 res_ty' = contResultType dup_cont
1349 -- Deal with case binder
1350 simplCaseBinder env scrut case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1352 -- Deal with the case alternatives
1353 simplAlts alt_env handled_cons
1354 case_bndr' better_alts dup_cont `thenSmpl` \ alts' ->
1356 -- Put the case back together
1357 mkCase scrut case_bndr' res_ty' alts' `thenSmpl` \ case_expr ->
1359 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1360 -- The case binder *not* scope over the whole returned case-expression
1361 rebuild env case_expr nondup_cont
1364 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1365 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1366 way, there's a chance that v will now only be used once, and hence
1371 There is a time we *don't* want to do that, namely when
1372 -fno-case-of-case is on. This happens in the first simplifier pass,
1373 and enhances full laziness. Here's the bad case:
1374 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1375 If we eliminate the inner case, we trap it inside the I# v -> arm,
1376 which might prevent some full laziness happening. I've seen this
1377 in action in spectral/cichelli/Prog.hs:
1378 [(m,n) | m <- [1..max], n <- [1..max]]
1379 Hence the check for NoCaseOfCase.
1383 There is another situation when we don't want to do it. If we have
1385 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1386 ...other cases .... }
1388 We'll perform the binder-swap for the outer case, giving
1390 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1391 ...other cases .... }
1393 But there is no point in doing it for the inner case, because w1 can't
1394 be inlined anyway. Furthermore, doing the case-swapping involves
1395 zapping w2's occurrence info (see paragraphs that follow), and that
1396 forces us to bind w2 when doing case merging. So we get
1398 case x of w1 { A -> let w2 = w1 in e1
1399 B -> let w2 = w1 in e2
1400 ...other cases .... }
1402 This is plain silly in the common case where w2 is dead.
1404 Even so, I can't see a good way to implement this idea. I tried
1405 not doing the binder-swap if the scrutinee was already evaluated
1406 but that failed big-time:
1410 case v of w { MkT x ->
1411 case x of x1 { I# y1 ->
1412 case x of x2 { I# y2 -> ...
1414 Notice that because MkT is strict, x is marked "evaluated". But to
1415 eliminate the last case, we must either make sure that x (as well as
1416 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1417 the binder-swap. So this whole note is a no-op.
1421 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1422 any occurrence info (eg IAmDead) in the case binder, because the
1423 case-binder now effectively occurs whenever v does. AND we have to do
1424 the same for the pattern-bound variables! Example:
1426 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1428 Here, b and p are dead. But when we move the argment inside the first
1429 case RHS, and eliminate the second case, we get
1431 case x of { (a,b) -> a b }
1433 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1436 Indeed, this can happen anytime the case binder isn't dead:
1437 case <any> of x { (a,b) ->
1438 case x of { (p,q) -> p } }
1439 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1440 The point is that we bring into the envt a binding
1442 after the outer case, and that makes (a,b) alive. At least we do unless
1443 the case binder is guaranteed dead.
1446 simplCaseBinder env (Var v) case_bndr
1447 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
1449 -- Failed try [see Note 2 above]
1450 -- not (isEvaldUnfolding (idUnfolding v))
1452 = simplBinder env (zap case_bndr) `thenSmpl` \ (env, case_bndr') ->
1453 returnSmpl (modifyInScope env v case_bndr', case_bndr')
1454 -- We could extend the substitution instead, but it would be
1455 -- a hack because then the substitution wouldn't be idempotent
1456 -- any more (v is an OutId). And this does just as well.
1458 zap b = b `setIdOccInfo` NoOccInfo
1460 simplCaseBinder env other_scrut case_bndr
1461 = simplBinder env case_bndr `thenSmpl` \ (env, case_bndr') ->
1462 returnSmpl (env, case_bndr')
1468 simplAlts :: SimplEnv
1469 -> [AltCon] -- Alternatives the scrutinee can't be
1470 -- in the default case
1471 -> OutId -- Case binder
1472 -> [InAlt] -> SimplCont
1473 -> SimplM [OutAlt] -- Includes the continuation
1475 simplAlts env handled_cons case_bndr' alts cont'
1476 = do { mb_alts <- mapSmpl simpl_alt alts
1477 ; return [alt' | Just (_, alt') <- mb_alts] }
1478 -- Filter out the alternatives that are inaccessible
1480 simpl_alt alt = simplAlt env handled_cons case_bndr' alt cont'
1482 simplAlt :: SimplEnv -> [AltCon] -> OutId -> InAlt -> SimplCont
1483 -> SimplM (Maybe (TvSubstEnv, OutAlt))
1484 -- Simplify an alternative, returning the type refinement for the
1485 -- alternative, if the alternative does any refinement at all
1486 -- Nothing => the alternative is inaccessible
1488 simplAlt env handled_cons case_bndr' (DEFAULT, bndrs, rhs) cont'
1489 = ASSERT( null bndrs )
1490 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1491 returnSmpl (Just (emptyVarEnv, (DEFAULT, [], rhs')))
1493 env' = mk_rhs_env env case_bndr' (mkOtherCon handled_cons)
1494 -- Record the constructors that the case-binder *can't* be.
1496 simplAlt env handled_cons case_bndr' (LitAlt lit, bndrs, rhs) cont'
1497 = ASSERT( null bndrs )
1498 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1499 returnSmpl (Just (emptyVarEnv, (LitAlt lit, [], rhs')))
1501 env' = mk_rhs_env env case_bndr' (mkUnfolding False (Lit lit))
1503 simplAlt env handled_cons case_bndr' (DataAlt con, vs, rhs) cont'
1504 | isVanillaDataCon con
1505 = -- Deal with the pattern-bound variables
1506 -- Mark the ones that are in ! positions in the data constructor
1507 -- as certainly-evaluated.
1508 -- NB: it happens that simplBinders does *not* erase the OtherCon
1509 -- form of unfolding, so it's ok to add this info before
1510 -- doing simplBinders
1511 simplBinders env (add_evals con vs) `thenSmpl` \ (env, vs') ->
1513 -- Bind the case-binder to (con args)
1514 let unf = mkUnfolding False (mkConApp con con_args)
1515 inst_tys' = tyConAppArgs (idType case_bndr')
1516 con_args = map Type inst_tys' ++ map varToCoreExpr vs'
1517 env' = mk_rhs_env env case_bndr' unf
1519 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1520 returnSmpl (Just (emptyVarEnv, (DataAlt con, vs', rhs')))
1522 | otherwise -- GADT case
1524 (tvs,ids) = span isTyVar vs
1526 simplBinders env tvs `thenSmpl` \ (env1, tvs') ->
1527 case coreRefineTys (getInScope env1) con tvs' (idType case_bndr') of {
1528 Nothing -- Inaccessible
1529 | opt_PprStyle_Debug -- Hack: if debugging is on, generate an error case
1531 -> let rhs' = mkApps (Var eRROR_ID)
1532 [Type (substTy env (exprType rhs)),
1533 Lit (mkStringLit "Impossible alternative (GADT)")]
1535 simplBinders env1 ids `thenSmpl` \ (env2, ids') ->
1536 returnSmpl (Just (emptyVarEnv, (DataAlt con, tvs' ++ ids', rhs')))
1538 | otherwise -- Filter out the inaccessible branch
1541 Just refine@(tv_subst_env, _) -> -- The normal case
1544 env2 = refineSimplEnv env1 refine
1545 -- Simplify the Ids in the refined environment, so their types
1546 -- reflect the refinement. Usually this doesn't matter, but it helps
1547 -- in mkDupableAlt, when we want to float a lambda that uses these binders
1548 -- Furthermore, it means the binders contain maximal type information
1550 simplBinders env2 (add_evals con ids) `thenSmpl` \ (env3, ids') ->
1551 let unf = mkUnfolding False con_app
1552 con_app = mkConApp con con_args
1553 con_args = map varToCoreExpr vs' -- NB: no inst_tys'
1554 env_w_unf = mk_rhs_env env3 case_bndr' unf
1557 simplExprC env_w_unf rhs cont' `thenSmpl` \ rhs' ->
1558 returnSmpl (Just (tv_subst_env, (DataAlt con, vs', rhs'))) }
1561 -- add_evals records the evaluated-ness of the bound variables of
1562 -- a case pattern. This is *important*. Consider
1563 -- data T = T !Int !Int
1565 -- case x of { T a b -> T (a+1) b }
1567 -- We really must record that b is already evaluated so that we don't
1568 -- go and re-evaluate it when constructing the result.
1569 add_evals dc vs = cat_evals dc vs (dataConRepStrictness dc)
1571 cat_evals dc vs strs
1575 go (v:vs) strs | isTyVar v = v : go vs strs
1576 go (v:vs) (str:strs)
1577 | isMarkedStrict str = evald_v : go vs strs
1578 | otherwise = zapped_v : go vs strs
1580 zapped_v = zap_occ_info v
1581 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1582 go _ _ = pprPanic "cat_evals" (ppr dc $$ ppr vs $$ ppr strs)
1584 -- If the case binder is alive, then we add the unfolding
1586 -- to the envt; so vs are now very much alive
1587 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1588 | otherwise = \id -> id `setIdOccInfo` NoOccInfo
1590 mk_rhs_env env case_bndr' case_bndr_unf
1591 = modifyInScope env case_bndr' (case_bndr' `setIdUnfolding` case_bndr_unf)
1595 %************************************************************************
1597 \subsection{Known constructor}
1599 %************************************************************************
1601 We are a bit careful with occurrence info. Here's an example
1603 (\x* -> case x of (a*, b) -> f a) (h v, e)
1605 where the * means "occurs once". This effectively becomes
1606 case (h v, e) of (a*, b) -> f a)
1608 let a* = h v; b = e in f a
1612 All this should happen in one sweep.
1615 knownCon :: SimplEnv -> AltCon -> [OutExpr]
1616 -> InId -> [InAlt] -> SimplCont
1617 -> SimplM FloatsWithExpr
1619 knownCon env con args bndr alts cont
1620 = tick (KnownBranch bndr) `thenSmpl_`
1621 case findAlt con alts of
1622 (DEFAULT, bs, rhs) -> ASSERT( null bs )
1623 simplNonRecX env bndr scrut $ \ env ->
1624 -- This might give rise to a binding with non-atomic args
1625 -- like x = Node (f x) (g x)
1626 -- but no harm will be done
1627 simplExprF env rhs cont
1630 LitAlt lit -> Lit lit
1631 DataAlt dc -> mkConApp dc args
1633 (LitAlt lit, bs, rhs) -> ASSERT( null bs )
1634 simplNonRecX env bndr (Lit lit) $ \ env ->
1635 simplExprF env rhs cont
1637 (DataAlt dc, bs, rhs)
1638 -> ASSERT( n_drop_tys + length bs == length args )
1639 bind_args env bs (drop n_drop_tys args) $ \ env ->
1641 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1642 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1643 -- args are aready OutExprs, but bs are InIds
1645 simplNonRecX env bndr con_app $ \ env ->
1646 simplExprF env rhs cont
1648 n_drop_tys | isVanillaDataCon dc = tyConArity (dataConTyCon dc)
1650 -- Vanilla data constructors lack type arguments in the pattern
1653 bind_args env [] _ thing_inside = thing_inside env
1655 bind_args env (b:bs) (Type ty : args) thing_inside
1656 = ASSERT( isTyVar b )
1657 bind_args (extendTvSubst env b ty) bs args thing_inside
1659 bind_args env (b:bs) (arg : args) thing_inside
1661 simplNonRecX env b arg $ \ env ->
1662 bind_args env bs args thing_inside
1666 %************************************************************************
1668 \subsection{Duplicating continuations}
1670 %************************************************************************
1673 prepareCaseCont :: SimplEnv
1674 -> [InAlt] -> SimplCont
1675 -> SimplM (FloatsWith (SimplCont,SimplCont))
1676 -- Return a duplicatable continuation, a non-duplicable part
1677 -- plus some extra bindings
1679 -- No need to make it duplicatable if there's only one alternative
1680 prepareCaseCont env [alt] cont = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1681 prepareCaseCont env alts cont = mkDupableCont env cont
1685 mkDupableCont :: SimplEnv -> SimplCont
1686 -> SimplM (FloatsWith (SimplCont, SimplCont))
1688 mkDupableCont env cont
1689 | contIsDupable cont
1690 = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1692 mkDupableCont env (CoerceIt ty cont)
1693 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1694 returnSmpl (floats, (CoerceIt ty dup_cont, nondup_cont))
1696 mkDupableCont env (InlinePlease cont)
1697 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1698 returnSmpl (floats, (InlinePlease dup_cont, nondup_cont))
1700 mkDupableCont env cont@(ArgOf _ arg_ty _ _)
1701 = returnSmpl (emptyFloats env, (mkBoringStop arg_ty, cont))
1702 -- Do *not* duplicate an ArgOf continuation
1703 -- Because ArgOf continuations are opaque, we gain nothing by
1704 -- propagating them into the expressions, and we do lose a lot.
1705 -- Here's an example:
1706 -- && (case x of { T -> F; F -> T }) E
1707 -- Now, && is strict so we end up simplifying the case with
1708 -- an ArgOf continuation. If we let-bind it, we get
1710 -- let $j = \v -> && v E
1711 -- in simplExpr (case x of { T -> F; F -> T })
1712 -- (ArgOf (\r -> $j r)
1713 -- And after simplifying more we get
1715 -- let $j = \v -> && v E
1716 -- in case of { T -> $j F; F -> $j T }
1717 -- Which is a Very Bad Thing
1719 -- The desire not to duplicate is the entire reason that
1720 -- mkDupableCont returns a pair of continuations.
1722 -- The original plan had:
1723 -- e.g. (...strict-fn...) [...hole...]
1725 -- let $j = \a -> ...strict-fn...
1726 -- in $j [...hole...]
1728 mkDupableCont env (ApplyTo _ arg se cont)
1729 = -- e.g. [...hole...] (...arg...)
1731 -- let a = ...arg...
1732 -- in [...hole...] a
1733 simplExpr (setInScope se env) arg `thenSmpl` \ arg' ->
1735 mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1736 addFloats env floats $ \ env ->
1738 if exprIsDupable arg' then
1739 returnSmpl (emptyFloats env, (ApplyTo OkToDup arg' (zapSubstEnv se) dup_cont, nondup_cont))
1741 newId FSLIT("a") (exprType arg') `thenSmpl` \ arg_id ->
1743 tick (CaseOfCase arg_id) `thenSmpl_`
1744 -- Want to tick here so that we go round again,
1745 -- and maybe copy or inline the code.
1746 -- Not strictly CaseOfCase, but never mind
1748 returnSmpl (unitFloat env arg_id arg',
1749 (ApplyTo OkToDup (Var arg_id) (zapSubstEnv se) dup_cont,
1751 -- But what if the arg should be case-bound?
1752 -- This has been this way for a long time, so I'll leave it,
1753 -- but I can't convince myself that it's right.
1755 mkDupableCont env (Select _ case_bndr alts se cont)
1756 = -- e.g. (case [...hole...] of { pi -> ei })
1758 -- let ji = \xij -> ei
1759 -- in case [...hole...] of { pi -> ji xij }
1760 tick (CaseOfCase case_bndr) `thenSmpl_`
1762 alt_env = setInScope se env
1764 prepareCaseCont alt_env alts cont `thenSmpl` \ (floats1, (dup_cont, nondup_cont)) ->
1765 addFloats alt_env floats1 $ \ alt_env ->
1767 simplBinder alt_env case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1768 -- NB: simplBinder does not zap deadness occ-info, so
1769 -- a dead case_bndr' will still advertise its deadness
1770 -- This is really important because in
1771 -- case e of b { (# a,b #) -> ... }
1772 -- b is always dead, and indeed we are not allowed to bind b to (# a,b #),
1773 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1774 -- In the new alts we build, we have the new case binder, so it must retain
1777 mkDupableAlts alt_env case_bndr' alts dup_cont `thenSmpl` \ (floats2, alts') ->
1778 addFloats alt_env floats2 $ \ alt_env ->
1779 returnSmpl (emptyFloats alt_env,
1780 (Select OkToDup case_bndr' alts' (zapSubstEnv se)
1781 (mkBoringStop (contResultType dup_cont)),
1784 mkDupableAlts :: SimplEnv -> OutId -> [InAlt] -> SimplCont
1785 -> SimplM (FloatsWith [InAlt])
1786 -- Absorbs the continuation into the new alternatives
1788 mkDupableAlts env case_bndr' alts dupable_cont
1791 go env [] = returnSmpl (emptyFloats env, [])
1793 = do { (floats1, mb_alt') <- mkDupableAlt env case_bndr' dupable_cont alt
1794 ; addFloats env floats1 $ \ env -> do
1795 { (floats2, alts') <- go env alts
1796 ; returnSmpl (floats2, case mb_alt' of
1797 Just alt' -> alt' : alts'
1801 mkDupableAlt env case_bndr' cont alt
1802 = simplAlt env [] case_bndr' alt cont `thenSmpl` \ mb_stuff ->
1804 Nothing -> returnSmpl (emptyFloats env, Nothing) ;
1806 Just (reft, (con, bndrs', rhs')) ->
1807 -- Safe to say that there are no handled-cons for the DEFAULT case
1809 if exprIsDupable rhs' then
1810 returnSmpl (emptyFloats env, Just (con, bndrs', rhs'))
1811 -- It is worth checking for a small RHS because otherwise we
1812 -- get extra let bindings that may cause an extra iteration of the simplifier to
1813 -- inline back in place. Quite often the rhs is just a variable or constructor.
1814 -- The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1815 -- iterations because the version with the let bindings looked big, and so wasn't
1816 -- inlined, but after the join points had been inlined it looked smaller, and so
1819 -- NB: we have to check the size of rhs', not rhs.
1820 -- Duplicating a small InAlt might invalidate occurrence information
1821 -- However, if it *is* dupable, we return the *un* simplified alternative,
1822 -- because otherwise we'd need to pair it up with an empty subst-env....
1823 -- but we only have one env shared between all the alts.
1824 -- (Remember we must zap the subst-env before re-simplifying something).
1825 -- Rather than do this we simply agree to re-simplify the original (small) thing later.
1829 rhs_ty' = exprType rhs'
1830 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1832 | isTyVar bndr = not (bndr `elemVarEnv` reft)
1833 -- Don't abstract over tyvar binders which are refined away
1834 -- See Note [Refinement] below
1835 | otherwise = not (isDeadBinder bndr)
1836 -- The deadness info on the new Ids is preserved by simplBinders
1838 -- If we try to lift a primitive-typed something out
1839 -- for let-binding-purposes, we will *caseify* it (!),
1840 -- with potentially-disastrous strictness results. So
1841 -- instead we turn it into a function: \v -> e
1842 -- where v::State# RealWorld#. The value passed to this function
1843 -- is realworld#, which generates (almost) no code.
1845 -- There's a slight infelicity here: we pass the overall
1846 -- case_bndr to all the join points if it's used in *any* RHS,
1847 -- because we don't know its usage in each RHS separately
1849 -- We used to say "&& isUnLiftedType rhs_ty'" here, but now
1850 -- we make the join point into a function whenever used_bndrs'
1851 -- is empty. This makes the join-point more CPR friendly.
1852 -- Consider: let j = if .. then I# 3 else I# 4
1853 -- in case .. of { A -> j; B -> j; C -> ... }
1855 -- Now CPR doesn't w/w j because it's a thunk, so
1856 -- that means that the enclosing function can't w/w either,
1857 -- which is a lose. Here's the example that happened in practice:
1858 -- kgmod :: Int -> Int -> Int
1859 -- kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1863 -- I have seen a case alternative like this:
1864 -- True -> \v -> ...
1865 -- It's a bit silly to add the realWorld dummy arg in this case, making
1868 -- (the \v alone is enough to make CPR happy) but I think it's rare
1870 ( if not (any isId used_bndrs')
1871 then newId FSLIT("w") realWorldStatePrimTy `thenSmpl` \ rw_id ->
1872 returnSmpl ([rw_id], [Var realWorldPrimId])
1874 returnSmpl (used_bndrs', map varToCoreExpr used_bndrs')
1875 ) `thenSmpl` \ (final_bndrs', final_args) ->
1877 -- See comment about "$j" name above
1878 newId (encodeFS FSLIT("$j")) (mkPiTypes final_bndrs' rhs_ty') `thenSmpl` \ join_bndr ->
1879 -- Notice the funky mkPiTypes. If the contructor has existentials
1880 -- it's possible that the join point will be abstracted over
1881 -- type varaibles as well as term variables.
1882 -- Example: Suppose we have
1883 -- data T = forall t. C [t]
1885 -- case (case e of ...) of
1886 -- C t xs::[t] -> rhs
1887 -- We get the join point
1888 -- let j :: forall t. [t] -> ...
1889 -- j = /\t \xs::[t] -> rhs
1891 -- case (case e of ...) of
1892 -- C t xs::[t] -> j t xs
1894 -- We make the lambdas into one-shot-lambdas. The
1895 -- join point is sure to be applied at most once, and doing so
1896 -- prevents the body of the join point being floated out by
1897 -- the full laziness pass
1898 really_final_bndrs = map one_shot final_bndrs'
1899 one_shot v | isId v = setOneShotLambda v
1901 join_rhs = mkLams really_final_bndrs rhs'
1902 join_call = mkApps (Var join_bndr) final_args
1904 returnSmpl (unitFloat env join_bndr join_rhs, Just (con, bndrs', join_call)) }
1911 MkT :: a -> b -> T a
1915 MkT a' b (p::a') (q::b) -> [p,w]
1917 The danger is that we'll make a join point
1921 and that's ill-typed, because (p::a') but (w::a).
1923 Solution so far: don't abstract over a', because the type refinement
1924 maps [a' -> a] . Ultimately that won't work when real refinement goes on.
1926 Then we must abstract over any refined free variables. Hmm. Maybe we
1927 could just abstract over *all* free variables, thereby lambda-lifting
1928 the join point? We should try this.