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,
17 SimplCont(..), DupFlag(..), LetRhsFlag(..),
18 mkRhsStop, mkBoringStop, mkLazyArgStop, pushContArgs,
19 contResultType, countArgs, contIsDupable, contIsRhsOrArg,
20 getContArgs, interestingCallContext, interestingArg, isStrictType,
21 preInlineUnconditionally, postInlineUnconditionally,
22 interestingArgContext, inlineMode, activeInline, activeRule
24 import Id ( Id, idType, idInfo, idArity, isDataConWorkId,
25 idUnfolding, setIdUnfolding, isDeadBinder,
26 idNewDemandInfo, setIdInfo,
27 setIdOccInfo, zapLamIdInfo, setOneShotLambda
29 import IdInfo ( OccInfo(..), setArityInfo, zapDemandInfo,
30 setUnfoldingInfo, occInfo
32 import NewDemand ( isStrictDmd )
33 import TcGadt ( dataConCanMatch )
34 import DataCon ( dataConTyCon, dataConRepStrictness )
35 import TyCon ( tyConArity, isAlgTyCon, isNewTyCon, tyConDataCons_maybe )
37 import PprCore ( pprParendExpr, pprCoreExpr )
38 import CoreUnfold ( mkUnfolding, callSiteInline )
39 import CoreUtils ( exprIsDupable, exprIsTrivial, needsCaseBinding,
40 exprIsConApp_maybe, mkPiTypes, findAlt,
41 exprType, exprIsHNF, findDefault, mergeAlts,
42 exprOkForSpeculation, exprArity,
43 mkCoerce, mkSCC, mkInlineMe, applyTypeToArg,
46 import Rules ( lookupRule )
47 import BasicTypes ( isMarkedStrict )
48 import CostCentre ( currentCCS )
49 import Type ( TvSubstEnv, isUnLiftedType, seqType, tyConAppArgs, funArgTy,
50 coreEqType, splitTyConApp_maybe,
51 isTyVarTy, isFunTy, tcEqType
53 import Coercion ( Coercion, coercionKind,
54 mkTransCoercion, mkSymCoercion, splitCoercionKind_maybe, decomposeCo )
55 import VarEnv ( elemVarEnv, emptyVarEnv )
56 import TysPrim ( realWorldStatePrimTy )
57 import PrelInfo ( realWorldPrimId )
58 import BasicTypes ( TopLevelFlag(..), isTopLevel,
59 RecFlag(..), isNonRec, isNonRuleLoopBreaker
63 import Maybes ( orElse )
65 import Util ( notNull, filterOut )
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 simplRecBndrs 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 simplNonRecBndr 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 (env2,bndr2) = addLetIdInfo env1 bndr bndr1
318 completeNonRecX env2 True {- strict -} bndr bndr2 rhs1 thing_inside
320 | otherwise -- Normal, lazy case
321 = -- Don't use simplBinder because that doesn't keep
322 -- fragile occurrence info in the substitution
323 simplNonRecBndr env bndr `thenSmpl` \ (env, bndr') ->
324 simplLazyBind env NotTopLevel NonRecursive
325 bndr bndr' rhs rhs_se `thenSmpl` \ (floats, env) ->
326 addFloats env floats thing_inside
329 bndr_ty = idType bndr
332 A specialised variant of simplNonRec used when the RHS is already simplified, notably
333 in knownCon. It uses case-binding where necessary.
336 simplNonRecX :: SimplEnv
337 -> InId -- Old binder
338 -> OutExpr -- Simplified RHS
339 -> (SimplEnv -> SimplM FloatsWithExpr)
340 -> SimplM FloatsWithExpr
342 simplNonRecX env bndr new_rhs thing_inside
343 = do { (env, bndr') <- simplBinder env bndr
344 ; completeNonRecX env False {- Non-strict; pessimistic -}
345 bndr bndr' new_rhs thing_inside }
348 completeNonRecX :: SimplEnv
349 -> Bool -- Strict binding
350 -> InId -- Old binder
351 -> OutId -- New binder
352 -> OutExpr -- Simplified RHS
353 -> (SimplEnv -> SimplM FloatsWithExpr)
354 -> SimplM FloatsWithExpr
356 completeNonRecX env is_strict old_bndr new_bndr new_rhs thing_inside
357 | needsCaseBinding (idType new_bndr) new_rhs
358 -- Make this test *before* the preInlineUnconditionally
359 -- Consider case I# (quotInt# x y) of
360 -- I# v -> let w = J# v in ...
361 -- If we gaily inline (quotInt# x y) for v, we end up building an
363 -- let w = J# (quotInt# x y) in ...
364 -- because quotInt# can fail.
365 = do { (floats, body) <- thing_inside env
366 ; let body' = wrapFloats floats body
367 ; return (emptyFloats env, Case new_rhs new_bndr (exprType body)
368 [(DEFAULT, [], body')]) }
371 = -- Make the arguments atomic if necessary,
372 -- adding suitable bindings
373 -- pprTrace "completeNonRecX" (ppr new_bndr <+> ppr new_rhs) $
374 mkAtomicArgsE env is_strict new_rhs $ \ env new_rhs ->
375 completeLazyBind env NotTopLevel
376 old_bndr new_bndr new_rhs `thenSmpl` \ (floats, env) ->
377 addFloats env floats thing_inside
379 {- No, no, no! Do not try preInlineUnconditionally in completeNonRecX
380 Doing so risks exponential behaviour, because new_rhs has been simplified once already
381 In the cases described by the folowing commment, postInlineUnconditionally will
382 catch many of the relevant cases.
383 -- This happens; for example, the case_bndr during case of
384 -- known constructor: case (a,b) of x { (p,q) -> ... }
385 -- Here x isn't mentioned in the RHS, so we don't want to
386 -- create the (dead) let-binding let x = (a,b) in ...
388 -- Similarly, single occurrences can be inlined vigourously
389 -- e.g. case (f x, g y) of (a,b) -> ....
390 -- If a,b occur once we can avoid constructing the let binding for them.
391 | preInlineUnconditionally env NotTopLevel bndr new_rhs
392 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
394 -- NB: completeLazyBind uses postInlineUnconditionally; no need to do that here
399 %************************************************************************
401 \subsection{Lazy bindings}
403 %************************************************************************
405 simplRecBind is used for
406 * recursive bindings only
409 simplRecBind :: SimplEnv -> TopLevelFlag
410 -> [(InId, InExpr)] -> [OutId]
411 -> SimplM (FloatsWith SimplEnv)
412 simplRecBind env top_lvl pairs bndrs'
413 = go env pairs bndrs' `thenSmpl` \ (floats, env) ->
414 returnSmpl (flattenFloats floats, env)
416 go env [] _ = returnSmpl (emptyFloats env, env)
418 go env ((bndr, rhs) : pairs) (bndr' : bndrs')
419 = simplRecOrTopPair env top_lvl bndr bndr' rhs `thenSmpl` \ (floats, env) ->
420 addFloats env floats (\env -> go env pairs bndrs')
424 simplRecOrTopPair is used for
425 * recursive bindings (whether top level or not)
426 * top-level non-recursive bindings
428 It assumes the binder has already been simplified, but not its IdInfo.
431 simplRecOrTopPair :: SimplEnv
433 -> InId -> OutId -- Binder, both pre-and post simpl
434 -> InExpr -- The RHS and its environment
435 -> SimplM (FloatsWith SimplEnv)
437 simplRecOrTopPair env top_lvl bndr bndr' rhs
438 | preInlineUnconditionally env top_lvl bndr rhs -- Check for unconditional inline
439 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
440 returnSmpl (emptyFloats env, extendIdSubst env bndr (mkContEx env rhs))
443 = simplLazyBind env top_lvl Recursive bndr bndr' rhs env
444 -- May not actually be recursive, but it doesn't matter
448 simplLazyBind is used for
449 * recursive bindings (whether top level or not)
450 * top-level non-recursive bindings
451 * non-top-level *lazy* non-recursive bindings
453 [Thus it deals with the lazy cases from simplNonRecBind, and all cases
454 from SimplRecOrTopBind]
457 1. It assumes that the binder is *already* simplified,
458 and is in scope, but not its IdInfo
460 2. It assumes that the binder type is lifted.
462 3. It does not check for pre-inline-unconditionallly;
463 that should have been done already.
466 simplLazyBind :: SimplEnv
467 -> TopLevelFlag -> RecFlag
468 -> InId -> OutId -- Binder, both pre-and post simpl
469 -> InExpr -> SimplEnv -- The RHS and its environment
470 -> SimplM (FloatsWith SimplEnv)
472 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
474 (env1,bndr2) = addLetIdInfo env bndr bndr1
475 rhs_env = setInScope rhs_se env1
476 is_top_level = isTopLevel top_lvl
477 ok_float_unlifted = not is_top_level && isNonRec is_rec
478 rhs_cont = mkRhsStop (idType bndr2)
480 -- Simplify the RHS; note the mkRhsStop, which tells
481 -- the simplifier that this is the RHS of a let.
482 simplExprF rhs_env rhs rhs_cont `thenSmpl` \ (floats, rhs1) ->
484 -- If any of the floats can't be floated, give up now
485 -- (The allLifted predicate says True for empty floats.)
486 if (not ok_float_unlifted && not (allLifted floats)) then
487 completeLazyBind env1 top_lvl bndr bndr2
488 (wrapFloats floats rhs1)
491 -- ANF-ise a constructor or PAP rhs
492 mkAtomicArgs False {- Not strict -}
493 ok_float_unlifted rhs1 `thenSmpl` \ (aux_binds, rhs2) ->
495 -- If the result is a PAP, float the floats out, else wrap them
496 -- By this time it's already been ANF-ised (if necessary)
497 if isEmptyFloats floats && isNilOL aux_binds then -- Shortcut a common case
498 completeLazyBind env1 top_lvl bndr bndr2 rhs2
500 else if is_top_level || exprIsTrivial rhs2 || exprIsHNF rhs2 then
501 -- WARNING: long dodgy argument coming up
502 -- WANTED: a better way to do this
504 -- We can't use "exprIsCheap" instead of exprIsHNF,
505 -- because that causes a strictness bug.
506 -- x = let y* = E in case (scc y) of { T -> F; F -> T}
507 -- The case expression is 'cheap', but it's wrong to transform to
508 -- y* = E; x = case (scc y) of {...}
509 -- Either we must be careful not to float demanded non-values, or
510 -- we must use exprIsHNF for the test, which ensures that the
511 -- thing is non-strict. So exprIsHNF => bindings are non-strict
512 -- I think. The WARN below tests for this.
514 -- We use exprIsTrivial here because we want to reveal lone variables.
515 -- E.g. let { x = letrec { y = E } in y } in ...
516 -- Here we definitely want to float the y=E defn.
517 -- exprIsHNF definitely isn't right for that.
519 -- Again, the floated binding can't be strict; if it's recursive it'll
520 -- be non-strict; if it's non-recursive it'd be inlined.
522 -- Note [SCC-and-exprIsTrivial]
524 -- y = let { x* = E } in scc "foo" x
525 -- then we do *not* want to float out the x binding, because
526 -- it's strict! Fortunately, exprIsTrivial replies False to
529 -- There's a subtlety here. There may be a binding (x* = e) in the
530 -- floats, where the '*' means 'will be demanded'. So is it safe
531 -- to float it out? Answer no, but it won't matter because
532 -- we only float if (a) arg' is a WHNF, or (b) it's going to top level
533 -- and so there can't be any 'will be demanded' bindings in the floats.
535 WARN( not (is_top_level || not (any demanded_float (floatBinds floats))),
536 ppr (filter demanded_float (floatBinds floats)) )
538 tick LetFloatFromLet `thenSmpl_` (
539 addFloats env1 floats $ \ env2 ->
540 addAtomicBinds env2 (fromOL aux_binds) $ \ env3 ->
541 completeLazyBind env3 top_lvl bndr bndr2 rhs2)
544 completeLazyBind env1 top_lvl bndr bndr2 (wrapFloats floats rhs1)
547 demanded_float (NonRec b r) = isStrictDmd (idNewDemandInfo b) && not (isUnLiftedType (idType b))
548 -- Unlifted-type (cheap-eagerness) lets may well have a demanded flag on them
549 demanded_float (Rec _) = False
554 %************************************************************************
556 \subsection{Completing a lazy binding}
558 %************************************************************************
561 * deals only with Ids, not TyVars
562 * takes an already-simplified binder and RHS
563 * is used for both recursive and non-recursive bindings
564 * is used for both top-level and non-top-level bindings
566 It does the following:
567 - tries discarding a dead binding
568 - tries PostInlineUnconditionally
569 - add unfolding [this is the only place we add an unfolding]
572 It does *not* attempt to do let-to-case. Why? Because it is used for
573 - top-level bindings (when let-to-case is impossible)
574 - many situations where the "rhs" is known to be a WHNF
575 (so let-to-case is inappropriate).
578 completeLazyBind :: SimplEnv
579 -> TopLevelFlag -- Flag stuck into unfolding
580 -> InId -- Old binder
581 -> OutId -- New binder
582 -> OutExpr -- Simplified RHS
583 -> SimplM (FloatsWith SimplEnv)
584 -- We return a new SimplEnv, because completeLazyBind may choose to do its work
585 -- by extending the substitution (e.g. let x = y in ...)
586 -- The new binding (if any) is returned as part of the floats.
587 -- NB: the returned SimplEnv has the right SubstEnv, but you should
588 -- (as usual) use the in-scope-env from the floats
590 completeLazyBind env top_lvl old_bndr new_bndr new_rhs
591 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
592 = -- Drop the binding
593 tick (PostInlineUnconditionally old_bndr) `thenSmpl_`
594 -- pprTrace "Inline unconditionally" (ppr old_bndr <+> ppr new_bndr <+> ppr new_rhs) $
595 returnSmpl (emptyFloats env, extendIdSubst env old_bndr (DoneEx new_rhs))
596 -- Use the substitution to make quite, quite sure that the substitution
597 -- will happen, since we are going to discard the binding
602 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
605 -- Add the unfolding *only* for non-loop-breakers
606 -- Making loop breakers not have an unfolding at all
607 -- means that we can avoid tests in exprIsConApp, for example.
608 -- This is important: if exprIsConApp says 'yes' for a recursive
609 -- thing, then we can get into an infinite loop
612 -- If the unfolding is a value, the demand info may
613 -- go pear-shaped, so we nuke it. Example:
615 -- case x of (p,q) -> h p q x
616 -- Here x is certainly demanded. But after we've nuked
617 -- the case, we'll get just
618 -- let x = (a,b) in h a b x
619 -- and now x is not demanded (I'm assuming h is lazy)
620 -- This really happens. Similarly
621 -- let f = \x -> e in ...f..f...
622 -- After inling f at some of its call sites the original binding may
623 -- (for example) be no longer strictly demanded.
624 -- The solution here is a bit ad hoc...
625 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
626 final_info | loop_breaker = new_bndr_info
627 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
628 | otherwise = info_w_unf
630 final_id = new_bndr `setIdInfo` final_info
632 -- These seqs forces the Id, and hence its IdInfo,
633 -- and hence any inner substitutions
635 -- pprTrace "Binding" (ppr final_id <+> ppr unfolding) $
636 returnSmpl (unitFloat env final_id new_rhs, env)
638 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
639 loop_breaker = isNonRuleLoopBreaker occ_info
640 old_info = idInfo old_bndr
641 occ_info = occInfo old_info
646 %************************************************************************
648 \subsection[Simplify-simplExpr]{The main function: simplExpr}
650 %************************************************************************
652 The reason for this OutExprStuff stuff is that we want to float *after*
653 simplifying a RHS, not before. If we do so naively we get quadratic
654 behaviour as things float out.
656 To see why it's important to do it after, consider this (real) example:
670 a -- Can't inline a this round, cos it appears twice
674 Each of the ==> steps is a round of simplification. We'd save a
675 whole round if we float first. This can cascade. Consider
680 let f = let d1 = ..d.. in \y -> e
684 in \x -> ...(\y ->e)...
686 Only in this second round can the \y be applied, and it
687 might do the same again.
691 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
692 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
694 expr_ty' = substTy env (exprType expr)
695 -- The type in the Stop continuation, expr_ty', is usually not used
696 -- It's only needed when discarding continuations after finding
697 -- a function that returns bottom.
698 -- Hence the lazy substitution
701 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
702 -- Simplify an expression, given a continuation
703 simplExprC env expr cont
704 = simplExprF env expr cont `thenSmpl` \ (floats, expr) ->
705 returnSmpl (wrapFloats floats expr)
707 simplExprF :: SimplEnv -> InExpr -> SimplCont -> SimplM FloatsWithExpr
708 -- Simplify an expression, returning floated binds
710 simplExprF env (Var v) cont = simplVar env v cont
711 simplExprF env (Lit lit) cont = rebuild env (Lit lit) cont
712 simplExprF env expr@(Lam _ _) cont = simplLam env expr cont
713 simplExprF env (Note note expr) cont = simplNote env note expr cont
714 simplExprF env (Cast body co) cont = simplCast env body co cont
715 simplExprF env (App fun arg) cont = simplExprF env fun
716 (ApplyTo NoDup arg (Just env) cont)
718 simplExprF env (Type ty) cont
719 = ASSERT( contIsRhsOrArg cont )
720 simplType env ty `thenSmpl` \ ty' ->
721 rebuild env (Type ty') cont
723 simplExprF env (Case scrut bndr case_ty alts) cont
724 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
725 = -- Simplify the scrutinee with a Select continuation
726 simplExprF env scrut (Select NoDup bndr alts env cont)
729 = -- If case-of-case is off, simply simplify the case expression
730 -- in a vanilla Stop context, and rebuild the result around it
731 simplExprC env scrut case_cont `thenSmpl` \ case_expr' ->
732 rebuild env case_expr' cont
734 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
735 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
737 simplExprF env (Let (Rec pairs) body) cont
738 = simplRecBndrs env (map fst pairs) `thenSmpl` \ (env, bndrs') ->
739 -- NB: bndrs' don't have unfoldings or rules
740 -- We add them as we go down
742 simplRecBind env NotTopLevel pairs bndrs' `thenSmpl` \ (floats, env) ->
743 addFloats env floats $ \ env ->
744 simplExprF env body cont
746 -- A non-recursive let is dealt with by simplNonRecBind
747 simplExprF env (Let (NonRec bndr rhs) body) cont
748 = simplNonRecBind env bndr rhs env (contResultType cont) $ \ env ->
749 simplExprF env body cont
752 ---------------------------------
753 simplType :: SimplEnv -> InType -> SimplM OutType
754 -- Kept monadic just so we can do the seqType
756 = seqType new_ty `seq` returnSmpl new_ty
758 new_ty = substTy env ty
762 %************************************************************************
766 %************************************************************************
769 simplCast :: SimplEnv -> InExpr -> Coercion -> SimplCont -> SimplM FloatsWithExpr
770 simplCast env body co cont
773 | (s1, k1) <- coercionKind co
774 , s1 `tcEqType` k1 = cont
775 addCoerce co1 (CoerceIt co2 cont)
776 | (s1, k1) <- coercionKind co1
777 , (l1, t1) <- coercionKind co2
778 -- coerce T1 S1 (coerce S1 K1 e)
781 -- coerce T1 K1 e, otherwise
783 -- For example, in the initial form of a worker
784 -- we may find (coerce T (coerce S (\x.e))) y
785 -- and we'd like it to simplify to e[y/x] in one round
787 , s1 `coreEqType` t1 = cont -- The coerces cancel out
788 | otherwise = CoerceIt (mkTransCoercion co1 co2) cont
790 addCoerce co (ApplyTo dup arg arg_se cont)
791 | not (isTypeArg arg) -- This whole case only works for value args
792 -- Could upgrade to have equiv thing for type apps too
793 , Just (s1s2, t1t2) <- splitCoercionKind_maybe co
795 -- co : s1s2 :=: t1t2
796 -- (coerce (T1->T2) (S1->S2) F) E
798 -- coerce T2 S2 (F (coerce S1 T1 E))
800 -- t1t2 must be a function type, T1->T2, because it's applied
801 -- to something but s1s2 might conceivably not be
803 -- When we build the ApplyTo we can't mix the out-types
804 -- with the InExpr in the argument, so we simply substitute
805 -- to make it all consistent. It's a bit messy.
806 -- But it isn't a common case.
809 -- we split coercion t1->t2 :=: s1->s2 into t1 :=: s1 and
810 -- t2 :=: s2 with left and right on the curried form:
811 -- (->) t1 t2 :=: (->) s1 s2
812 [co1, co2] = decomposeCo 2 co
813 new_arg = mkCoerce (mkSymCoercion co1) arg'
814 arg' = case arg_se of
816 Just arg_se -> substExpr (setInScope arg_se env) arg
817 result = ApplyTo dup new_arg (Just $ zapSubstEnv env)
819 addCoerce co cont = CoerceIt co cont
821 simplType env co `thenSmpl` \ co' ->
822 simplExprF env body (addCoerce co' cont)
825 %************************************************************************
829 %************************************************************************
832 simplLam env fun cont
835 zap_it = mkLamBndrZapper fun (countArgs cont)
836 cont_ty = contResultType cont
838 -- Type-beta reduction
839 go env (Lam bndr body) (ApplyTo _ (Type ty_arg) mb_arg_se body_cont)
840 = ASSERT( isTyVar bndr )
841 do { tick (BetaReduction bndr)
842 ; ty_arg' <- case mb_arg_se of
843 Just arg_se -> simplType (setInScope arg_se env) ty_arg
844 Nothing -> return ty_arg
845 ; go (extendTvSubst env bndr ty_arg') body body_cont }
847 -- Ordinary beta reduction
848 go env (Lam bndr body) cont@(ApplyTo _ arg (Just arg_se) body_cont)
849 = do { tick (BetaReduction bndr)
850 ; simplNonRecBind env (zap_it bndr) arg arg_se cont_ty $ \ env ->
851 go env body body_cont }
853 go env (Lam bndr body) cont@(ApplyTo _ arg Nothing body_cont)
854 = do { tick (BetaReduction bndr)
855 ; simplNonRecX env (zap_it bndr) arg $ \ env ->
856 go env body body_cont }
858 -- Not enough args, so there are real lambdas left to put in the result
859 go env lam@(Lam _ _) cont
860 = do { (env, bndrs') <- simplLamBndrs env bndrs
861 ; body' <- simplExpr env body
862 ; (floats, new_lam) <- mkLam env bndrs' body' cont
863 ; addFloats env floats $ \ env ->
864 rebuild env new_lam cont }
866 (bndrs,body) = collectBinders lam
868 -- Exactly enough args
869 go env expr cont = simplExprF env expr cont
871 mkLamBndrZapper :: CoreExpr -- Function
872 -> Int -- Number of args supplied, *including* type args
873 -> Id -> Id -- Use this to zap the binders
874 mkLamBndrZapper fun n_args
875 | n_args >= n_params fun = \b -> b -- Enough args
876 | otherwise = \b -> zapLamIdInfo b
878 -- NB: we count all the args incl type args
879 -- so we must count all the binders (incl type lambdas)
880 n_params (Note _ e) = n_params e
881 n_params (Lam b e) = 1 + n_params e
882 n_params other = 0::Int
886 %************************************************************************
890 %************************************************************************
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 -- See notes with SimplMonad.inlineMode
902 simplNote env InlineMe e cont
903 | contIsRhsOrArg cont -- Totally boring continuation; see notes above
904 = -- Don't inline inside an INLINE expression
905 simplExpr (setMode inlineMode env ) e `thenSmpl` \ e' ->
906 rebuild env (mkInlineMe e') cont
908 | otherwise -- Dissolve the InlineMe note if there's
909 -- an interesting context of any kind to combine with
910 -- (even a type application -- anything except Stop)
911 = simplExprF env e cont
913 simplNote env (CoreNote s) e cont
914 = simplExpr env e `thenSmpl` \ e' ->
915 rebuild env (Note (CoreNote s) e') cont
919 %************************************************************************
921 \subsection{Dealing with calls}
923 %************************************************************************
926 simplVar env var cont
927 = case substId env var of
928 DoneEx e -> simplExprF (zapSubstEnv env) e cont
929 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
930 DoneId var1 occ -> completeCall (zapSubstEnv env) var1 occ cont
931 -- Note [zapSubstEnv]
932 -- The template is already simplified, so don't re-substitute.
933 -- This is VITAL. Consider
935 -- let y = \z -> ...x... in
937 -- We'll clone the inner \x, adding x->x' in the id_subst
938 -- Then when we inline y, we must *not* replace x by x' in
939 -- the inlined copy!!
941 ---------------------------------------------------------
942 -- Dealing with a call site
944 completeCall env var occ_info cont
945 = -- Simplify the arguments
946 getDOptsSmpl `thenSmpl` \ dflags ->
948 chkr = getSwitchChecker env
949 (args, call_cont) = getContArgs chkr var cont
952 simplifyArgs env fn_ty (interestingArgContext var call_cont) args
953 (contResultType call_cont) $ \ env args ->
955 -- Next, look for rules or specialisations that match
957 -- It's important to simplify the args first, because the rule-matcher
958 -- doesn't do substitution as it goes. We don't want to use subst_args
959 -- (defined in the 'where') because that throws away useful occurrence info,
960 -- and perhaps-very-important specialisations.
962 -- Some functions have specialisations *and* are strict; in this case,
963 -- we don't want to inline the wrapper of the non-specialised thing; better
964 -- to call the specialised thing instead.
965 -- We used to use the black-listing mechanism to ensure that inlining of
966 -- the wrapper didn't occur for things that have specialisations till a
967 -- later phase, so but now we just try RULES first
969 -- You might think that we shouldn't apply rules for a loop breaker:
970 -- doing so might give rise to an infinite loop, because a RULE is
971 -- rather like an extra equation for the function:
972 -- RULE: f (g x) y = x+y
975 -- But it's too drastic to disable rules for loop breakers.
976 -- Even the foldr/build rule would be disabled, because foldr
977 -- is recursive, and hence a loop breaker:
978 -- foldr k z (build g) = g k z
979 -- So it's up to the programmer: rules can cause divergence
982 in_scope = getInScope env
984 maybe_rule = case activeRule env of
985 Nothing -> Nothing -- No rules apply
986 Just act_fn -> lookupRule act_fn in_scope rules var args
989 Just (rule_name, rule_rhs) ->
990 tick (RuleFired rule_name) `thenSmpl_`
991 (if dopt Opt_D_dump_inlinings dflags then
992 pprTrace "Rule fired" (vcat [
993 text "Rule:" <+> ftext rule_name,
994 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
995 text "After: " <+> pprCoreExpr rule_rhs,
996 text "Cont: " <+> ppr call_cont])
999 simplExprF env rule_rhs call_cont ;
1001 Nothing -> -- No rules
1003 -- Next, look for an inlining
1005 arg_infos = [ interestingArg arg | arg <- args, isValArg arg]
1006 interesting_cont = interestingCallContext (notNull args)
1009 active_inline = activeInline env var occ_info
1010 maybe_inline = callSiteInline dflags active_inline occ_info
1011 var arg_infos interesting_cont
1013 case maybe_inline of {
1014 Just unfolding -- There is an inlining!
1015 -> tick (UnfoldingDone var) `thenSmpl_`
1016 (if dopt Opt_D_dump_inlinings dflags then
1017 pprTrace "Inlining done" (vcat [
1018 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
1019 text "Inlined fn: " <+> ppr unfolding,
1020 text "Cont: " <+> ppr call_cont])
1023 simplExprF env unfolding (pushContArgs args call_cont)
1026 Nothing -> -- No inlining!
1029 rebuild env (mkApps (Var var) args) call_cont
1033 %************************************************************************
1035 \subsection{Arguments}
1037 %************************************************************************
1040 ---------------------------------------------------------
1041 -- Simplifying the arguments of a call
1043 simplifyArgs :: SimplEnv
1044 -> OutType -- Type of the function
1045 -> Bool -- True if the fn has RULES
1046 -> [(InExpr, Maybe SimplEnv, Bool)] -- Details of the arguments
1047 -> OutType -- Type of the continuation
1048 -> (SimplEnv -> [OutExpr] -> SimplM FloatsWithExpr)
1049 -> SimplM FloatsWithExpr
1051 -- [CPS-like because of strict arguments]
1053 -- Simplify the arguments to a call.
1054 -- This part of the simplifier may break the no-shadowing invariant
1056 -- f (...(\a -> e)...) (case y of (a,b) -> e')
1057 -- where f is strict in its second arg
1058 -- If we simplify the innermost one first we get (...(\a -> e)...)
1059 -- Simplifying the second arg makes us float the case out, so we end up with
1060 -- case y of (a,b) -> f (...(\a -> e)...) e'
1061 -- So the output does not have the no-shadowing invariant. However, there is
1062 -- no danger of getting name-capture, because when the first arg was simplified
1063 -- we used an in-scope set that at least mentioned all the variables free in its
1064 -- static environment, and that is enough.
1066 -- We can't just do innermost first, or we'd end up with a dual problem:
1067 -- case x of (a,b) -> f e (...(\a -> e')...)
1069 -- I spent hours trying to recover the no-shadowing invariant, but I just could
1070 -- not think of an elegant way to do it. The simplifier is already knee-deep in
1071 -- continuations. We have to keep the right in-scope set around; AND we have
1072 -- to get the effect that finding (error "foo") in a strict arg position will
1073 -- discard the entire application and replace it with (error "foo"). Getting
1074 -- all this at once is TOO HARD!
1076 simplifyArgs env fn_ty has_rules args cont_ty thing_inside
1077 = go env fn_ty args thing_inside
1079 go env fn_ty [] thing_inside = thing_inside env []
1080 go env fn_ty (arg:args) thing_inside = simplifyArg env fn_ty has_rules arg cont_ty $ \ env arg' ->
1081 go env (applyTypeToArg fn_ty arg') args $ \ env args' ->
1082 thing_inside env (arg':args')
1084 simplifyArg env fn_ty has_rules (arg, Nothing, _) cont_ty thing_inside
1085 = thing_inside env arg -- Already simplified
1087 simplifyArg env fn_ty has_rules (Type ty_arg, Just se, _) cont_ty thing_inside
1088 = simplType (setInScope se env) ty_arg `thenSmpl` \ new_ty_arg ->
1089 thing_inside env (Type new_ty_arg)
1091 simplifyArg env fn_ty has_rules (val_arg, Just arg_se, is_strict) cont_ty thing_inside
1093 = simplStrictArg AnArg env val_arg arg_se arg_ty cont_ty thing_inside
1095 | otherwise -- Lazy argument
1096 -- DO NOT float anything outside, hence simplExprC
1097 -- There is no benefit (unlike in a let-binding), and we'd
1098 -- have to be very careful about bogus strictness through
1099 -- floating a demanded let.
1100 = simplExprC (setInScope arg_se env) val_arg
1101 (mkLazyArgStop arg_ty has_rules) `thenSmpl` \ arg1 ->
1102 thing_inside env arg1
1104 arg_ty = funArgTy fn_ty
1107 simplStrictArg :: LetRhsFlag
1108 -> SimplEnv -- The env of the call
1109 -> InExpr -> SimplEnv -- The arg plus its env
1110 -> OutType -- arg_ty: type of the argument
1111 -> OutType -- cont_ty: Type of thing computed by the context
1112 -> (SimplEnv -> OutExpr -> SimplM FloatsWithExpr)
1113 -- Takes an expression of type rhs_ty,
1114 -- returns an expression of type cont_ty
1115 -- The env passed to this continuation is the
1116 -- env of the call, plus any new in-scope variables
1117 -> SimplM FloatsWithExpr -- An expression of type cont_ty
1119 simplStrictArg is_rhs call_env arg arg_env arg_ty cont_ty thing_inside
1120 = simplExprF (setInScope arg_env call_env) arg
1121 (ArgOf is_rhs arg_ty cont_ty (\ new_env -> thing_inside (setInScope call_env new_env)))
1122 -- Notice the way we use arg_env (augmented with in-scope vars from call_env)
1123 -- to simplify the argument
1124 -- and call-env (augmented with in-scope vars from the arg) to pass to the continuation
1128 %************************************************************************
1130 \subsection{mkAtomicArgs}
1132 %************************************************************************
1134 mkAtomicArgs takes a putative RHS, checks whether it's a PAP or
1135 constructor application and, if so, converts it to ANF, so that the
1136 resulting thing can be inlined more easily. Thus
1143 There are three sorts of binding context, specified by the two
1149 N N Top-level or recursive Only bind args of lifted type
1151 N Y Non-top-level and non-recursive, Bind args of lifted type, or
1152 but lazy unlifted-and-ok-for-speculation
1154 Y Y Non-top-level, non-recursive, Bind all args
1155 and strict (demanded)
1162 there is no point in transforming to
1164 x = case (y div# z) of r -> MkC r
1166 because the (y div# z) can't float out of the let. But if it was
1167 a *strict* let, then it would be a good thing to do. Hence the
1168 context information.
1171 mkAtomicArgsE :: SimplEnv
1172 -> Bool -- A strict binding
1173 -> OutExpr -- The rhs
1174 -> (SimplEnv -> OutExpr -> SimplM FloatsWithExpr)
1175 -> SimplM FloatsWithExpr
1177 mkAtomicArgsE env is_strict rhs thing_inside
1178 | (Var fun, args) <- collectArgs rhs, -- It's an application
1179 isDataConWorkId fun || valArgCount args < idArity fun -- And it's a constructor or PAP
1180 = go env (Var fun) args
1182 | otherwise = thing_inside env rhs
1185 go env fun [] = thing_inside env fun
1187 go env fun (arg : args)
1188 | exprIsTrivial arg -- Easy case
1189 || no_float_arg -- Can't make it atomic
1190 = go env (App fun arg) args
1193 = do { arg_id <- newId FSLIT("a") arg_ty
1194 ; completeNonRecX env False {- pessimistic -} arg_id arg_id arg $ \env ->
1195 go env (App fun (Var arg_id)) args }
1197 arg_ty = exprType arg
1198 no_float_arg = not is_strict && (isUnLiftedType arg_ty) && not (exprOkForSpeculation arg)
1201 -- Old code: consider rewriting to be more like mkAtomicArgsE
1203 mkAtomicArgs :: Bool -- A strict binding
1204 -> Bool -- OK to float unlifted args
1206 -> SimplM (OrdList (OutId,OutExpr), -- The floats (unusually) may include
1207 OutExpr) -- things that need case-binding,
1208 -- if the strict-binding flag is on
1210 mkAtomicArgs is_strict ok_float_unlifted rhs
1211 | (Var fun, args) <- collectArgs rhs, -- It's an application
1212 isDataConWorkId fun || valArgCount args < idArity fun -- And it's a constructor or PAP
1213 = go fun nilOL [] args -- Have a go
1215 | otherwise = bale_out -- Give up
1218 bale_out = returnSmpl (nilOL, rhs)
1220 go fun binds rev_args []
1221 = returnSmpl (binds, mkApps (Var fun) (reverse rev_args))
1223 go fun binds rev_args (arg : args)
1224 | exprIsTrivial arg -- Easy case
1225 = go fun binds (arg:rev_args) args
1227 | not can_float_arg -- Can't make this arg atomic
1228 = bale_out -- ... so give up
1230 | otherwise -- Don't forget to do it recursively
1231 -- E.g. x = a:b:c:[]
1232 = mkAtomicArgs is_strict ok_float_unlifted arg `thenSmpl` \ (arg_binds, arg') ->
1233 newId FSLIT("a") arg_ty `thenSmpl` \ arg_id ->
1234 go fun ((arg_binds `snocOL` (arg_id,arg')) `appOL` binds)
1235 (Var arg_id : rev_args) args
1237 arg_ty = exprType arg
1238 can_float_arg = is_strict
1239 || not (isUnLiftedType arg_ty)
1240 || (ok_float_unlifted && exprOkForSpeculation arg)
1243 addAtomicBinds :: SimplEnv -> [(OutId,OutExpr)]
1244 -> (SimplEnv -> SimplM (FloatsWith a))
1245 -> SimplM (FloatsWith a)
1246 addAtomicBinds env [] thing_inside = thing_inside env
1247 addAtomicBinds env ((v,r):bs) thing_inside = addAuxiliaryBind env (NonRec v r) $ \ env ->
1248 addAtomicBinds env bs thing_inside
1252 %************************************************************************
1254 \subsection{The main rebuilder}
1256 %************************************************************************
1259 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM FloatsWithExpr
1261 rebuild env expr (Stop _ _ _) = rebuildDone env expr
1262 rebuild env expr (ArgOf _ _ _ cont_fn) = cont_fn env expr
1263 rebuild env expr (CoerceIt co cont) = rebuild env (mkCoerce co expr) cont
1264 rebuild env expr (Select _ bndr alts se cont) = rebuildCase (setInScope se env) expr bndr alts cont
1265 rebuild env expr (ApplyTo _ arg mb_se cont) = rebuildApp env expr arg mb_se cont
1267 rebuildApp env fun arg mb_se cont
1268 = do { arg' <- simplArg env arg mb_se
1269 ; rebuild env (App fun arg') cont }
1271 simplArg :: SimplEnv -> CoreExpr -> Maybe SimplEnv -> SimplM CoreExpr
1272 simplArg env arg Nothing = return arg -- The arg is already simplified
1273 simplArg env arg (Just arg_env) = simplExpr (setInScope arg_env env) arg
1275 rebuildDone env expr = returnSmpl (emptyFloats env, expr)
1279 %************************************************************************
1281 \subsection{Functions dealing with a case}
1283 %************************************************************************
1285 Blob of helper functions for the "case-of-something-else" situation.
1288 ---------------------------------------------------------
1289 -- Eliminate the case if possible
1291 rebuildCase :: SimplEnv
1292 -> OutExpr -- Scrutinee
1293 -> InId -- Case binder
1294 -> [InAlt] -- Alternatives (inceasing order)
1296 -> SimplM FloatsWithExpr
1298 rebuildCase env scrut case_bndr alts cont
1299 | Just (con,args) <- exprIsConApp_maybe scrut
1300 -- Works when the scrutinee is a variable with a known unfolding
1301 -- as well as when it's an explicit constructor application
1302 = knownCon env scrut (DataAlt con) args case_bndr alts cont
1304 | Lit lit <- scrut -- No need for same treatment as constructors
1305 -- because literals are inlined more vigorously
1306 = knownCon env scrut (LitAlt lit) [] case_bndr alts cont
1309 = -- Prepare the continuation;
1310 -- The new subst_env is in place
1311 prepareCaseCont env alts cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1312 addFloats env floats $ \ env ->
1315 -- The case expression is annotated with the result type of the continuation
1316 -- This may differ from the type originally on the case. For example
1317 -- case(T) (case(Int#) a of { True -> 1#; False -> 0# }) of
1320 -- let j a# = <blob>
1321 -- in case(T) a of { True -> j 1#; False -> j 0# }
1322 -- Note that the case that scrutinises a now returns a T not an Int#
1323 res_ty' = contResultType dup_cont
1326 -- Deal with case binder
1327 simplCaseBinder env scrut case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1329 -- Deal with the case alternatives
1330 simplAlts alt_env scrut case_bndr' alts dup_cont `thenSmpl` \ alts' ->
1332 -- Put the case back together
1333 mkCase scrut case_bndr' res_ty' alts' `thenSmpl` \ case_expr ->
1335 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1336 -- The case binder *not* scope over the whole returned case-expression
1337 rebuild env case_expr nondup_cont
1340 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1341 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1342 way, there's a chance that v will now only be used once, and hence
1345 Note [no-case-of-case]
1346 ~~~~~~~~~~~~~~~~~~~~~~
1347 There is a time we *don't* want to do that, namely when
1348 -fno-case-of-case is on. This happens in the first simplifier pass,
1349 and enhances full laziness. Here's the bad case:
1350 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1351 If we eliminate the inner case, we trap it inside the I# v -> arm,
1352 which might prevent some full laziness happening. I've seen this
1353 in action in spectral/cichelli/Prog.hs:
1354 [(m,n) | m <- [1..max], n <- [1..max]]
1355 Hence the check for NoCaseOfCase.
1359 Consider case (v `cast` co) of x { I# ->
1360 ... (case (v `cast` co) of {...}) ...
1361 We'd like to eliminate the inner case. We can get this neatly by
1362 arranging that inside the outer case we add the unfolding
1363 v |-> x `cast` (sym co)
1364 to v. Then we should inline v at the inner case, cancel the casts, and away we go
1368 There is another situation when we don't want to do it. If we have
1370 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1371 ...other cases .... }
1373 We'll perform the binder-swap for the outer case, giving
1375 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1376 ...other cases .... }
1378 But there is no point in doing it for the inner case, because w1 can't
1379 be inlined anyway. Furthermore, doing the case-swapping involves
1380 zapping w2's occurrence info (see paragraphs that follow), and that
1381 forces us to bind w2 when doing case merging. So we get
1383 case x of w1 { A -> let w2 = w1 in e1
1384 B -> let w2 = w1 in e2
1385 ...other cases .... }
1387 This is plain silly in the common case where w2 is dead.
1389 Even so, I can't see a good way to implement this idea. I tried
1390 not doing the binder-swap if the scrutinee was already evaluated
1391 but that failed big-time:
1395 case v of w { MkT x ->
1396 case x of x1 { I# y1 ->
1397 case x of x2 { I# y2 -> ...
1399 Notice that because MkT is strict, x is marked "evaluated". But to
1400 eliminate the last case, we must either make sure that x (as well as
1401 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1402 the binder-swap. So this whole note is a no-op.
1406 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1407 any occurrence info (eg IAmDead) in the case binder, because the
1408 case-binder now effectively occurs whenever v does. AND we have to do
1409 the same for the pattern-bound variables! Example:
1411 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1413 Here, b and p are dead. But when we move the argment inside the first
1414 case RHS, and eliminate the second case, we get
1416 case x of { (a,b) -> a b }
1418 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1421 Indeed, this can happen anytime the case binder isn't dead:
1422 case <any> of x { (a,b) ->
1423 case x of { (p,q) -> p } }
1424 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1425 The point is that we bring into the envt a binding
1427 after the outer case, and that makes (a,b) alive. At least we do unless
1428 the case binder is guaranteed dead.
1431 simplCaseBinder env scrut case_bndr
1432 | switchIsOn (getSwitchChecker env) NoCaseOfCase
1433 -- See Note [no-case-of-case]
1434 = do { (env, case_bndr') <- simplBinder env case_bndr
1435 ; return (env, case_bndr') }
1437 simplCaseBinder env (Var v) case_bndr
1438 -- Failed try [see Note 2 above]
1439 -- not (isEvaldUnfolding (idUnfolding v))
1440 = do { (env, case_bndr') <- simplBinder env (zapOccInfo case_bndr)
1441 ; return (modifyInScope env v case_bndr', case_bndr') }
1442 -- We could extend the substitution instead, but it would be
1443 -- a hack because then the substitution wouldn't be idempotent
1444 -- any more (v is an OutId). And this does just as well.
1446 simplCaseBinder env (Cast (Var v) co) case_bndr -- Note [Case of cast]
1447 = do { (env, case_bndr') <- simplBinder env (zapOccInfo case_bndr)
1448 ; let rhs = Cast (Var case_bndr') (mkSymCoercion co)
1449 ; return (addBinderUnfolding env v rhs, case_bndr') }
1451 simplCaseBinder env other_scrut case_bndr
1452 = do { (env, case_bndr') <- simplBinder env case_bndr
1453 ; return (env, case_bndr') }
1455 zapOccInfo :: InId -> InId -- See Note [zapOccInfo]
1456 zapOccInfo b = b `setIdOccInfo` NoOccInfo
1460 simplAlts does two things:
1462 1. Eliminate alternatives that cannot match, including the
1463 DEFAULT alternative.
1465 2. If the DEFAULT alternative can match only one possible constructor,
1466 then make that constructor explicit.
1468 case e of x { DEFAULT -> rhs }
1470 case e of x { (a,b) -> rhs }
1471 where the type is a single constructor type. This gives better code
1472 when rhs also scrutinises x or e.
1474 Here "cannot match" includes knowledge from GADTs
1476 It's a good idea do do this stuff before simplifying the alternatives, to
1477 avoid simplifying alternatives we know can't happen, and to come up with
1478 the list of constructors that are handled, to put into the IdInfo of the
1479 case binder, for use when simplifying the alternatives.
1481 Eliminating the default alternative in (1) isn't so obvious, but it can
1484 data Colour = Red | Green | Blue
1493 DEFAULT -> [ case y of ... ]
1495 If we inline h into f, the default case of the inlined h can't happen.
1496 If we don't notice this, we may end up filtering out *all* the cases
1497 of the inner case y, which give us nowhere to go!
1501 simplAlts :: SimplEnv
1503 -> OutId -- Case binder
1504 -> [InAlt] -> SimplCont
1505 -> SimplM [OutAlt] -- Includes the continuation
1507 simplAlts env scrut case_bndr' alts cont'
1508 = do { mb_alts <- mapSmpl (simplAlt env imposs_cons case_bndr' cont') alts_wo_default
1509 ; default_alts <- simplDefault env case_bndr' imposs_deflt_cons cont' maybe_deflt
1510 ; return (mergeAlts default_alts [alt' | Just (_, alt') <- mb_alts]) }
1511 -- We need the mergeAlts in case the new default_alt
1512 -- has turned into a constructor alternative.
1514 (alts_wo_default, maybe_deflt) = findDefault alts
1515 imposs_cons = case scrut of
1516 Var v -> otherCons (idUnfolding v)
1519 -- "imposs_deflt_cons" are handled either by the context,
1520 -- OR by a branch in this case expression. (Don't include DEFAULT!!)
1521 imposs_deflt_cons = nub (imposs_cons ++ [con | (con,_,_) <- alts_wo_default])
1523 simplDefault :: SimplEnv
1524 -> OutId -- Case binder; need just for its type. Note that as an
1525 -- OutId, it has maximum information; this is important.
1526 -- Test simpl013 is an example
1527 -> [AltCon] -- These cons can't happen when matching the default
1530 -> SimplM [OutAlt] -- One branch or none; we use a list because it's what
1531 -- mergeAlts expects
1534 simplDefault env case_bndr' imposs_cons cont Nothing
1535 = return [] -- No default branch
1537 simplDefault env case_bndr' imposs_cons cont (Just rhs)
1538 | -- This branch handles the case where we are
1539 -- scrutinisng an algebraic data type
1540 Just (tycon, inst_tys) <- splitTyConApp_maybe (idType case_bndr'),
1541 isAlgTyCon tycon, -- It's a data type, tuple, or unboxed tuples.
1542 not (isNewTyCon tycon), -- We can have a newtype, if we are just doing an eval:
1543 -- case x of { DEFAULT -> e }
1544 -- and we don't want to fill in a default for them!
1545 Just all_cons <- tyConDataCons_maybe tycon,
1546 not (null all_cons), -- This is a tricky corner case. If the data type has no constructors,
1547 -- which GHC allows, then the case expression will have at most a default
1548 -- alternative. We don't want to eliminate that alternative, because the
1549 -- invariant is that there's always one alternative. It's more convenient
1551 -- case x of { DEFAULT -> e }
1552 -- as it is, rather than transform it to
1553 -- error "case cant match"
1554 -- which would be quite legitmate. But it's a really obscure corner, and
1555 -- not worth wasting code on.
1557 let imposs_data_cons = [con | DataAlt con <- imposs_cons] -- We now know it's a data type
1558 poss_data_cons = filterOut (`elem` imposs_data_cons) all_cons
1559 gadt_imposs | all isTyVarTy inst_tys = []
1560 | otherwise = filter (cant_match inst_tys) poss_data_cons
1561 final_poss = filterOut (`elem` gadt_imposs) poss_data_cons
1563 = case final_poss of
1564 [] -> returnSmpl [] -- Eliminate the default alternative
1565 -- altogether if it can't match
1567 [con] -> -- It matches exactly one constructor, so fill it in
1568 do { tick (FillInCaseDefault case_bndr')
1569 ; us <- getUniquesSmpl
1570 ; let (ex_tvs, co_tvs, arg_ids) =
1571 dataConRepInstPat us con inst_tys
1572 ; let con_alt = (DataAlt con, ex_tvs ++ co_tvs ++ arg_ids, rhs)
1573 ; Just (_, alt') <- simplAlt env [] case_bndr' cont con_alt
1574 -- The simplAlt must succeed with Just because we have
1575 -- already filtered out construtors that can't match
1578 two_or_more -> simplify_default (map DataAlt gadt_imposs ++ imposs_cons)
1581 = simplify_default imposs_cons
1583 cant_match tys data_con = not (dataConCanMatch data_con tys)
1585 simplify_default imposs_cons
1586 = do { let env' = addBinderOtherCon env case_bndr' imposs_cons
1587 -- Record the constructors that the case-binder *can't* be.
1588 ; rhs' <- simplExprC env' rhs cont
1589 ; return [(DEFAULT, [], rhs')] }
1591 simplAlt :: SimplEnv
1592 -> [AltCon] -- These constructors can't be present when
1593 -- matching this alternative
1594 -> OutId -- The case binder
1597 -> SimplM (Maybe (TvSubstEnv, OutAlt))
1599 -- Simplify an alternative, returning the type refinement for the
1600 -- alternative, if the alternative does any refinement at all
1601 -- Nothing => the alternative is inaccessible
1603 simplAlt env imposs_cons case_bndr' cont' (con, bndrs, rhs)
1604 | con `elem` imposs_cons -- This case can't match
1607 simplAlt env handled_cons case_bndr' cont' (DEFAULT, bndrs, rhs)
1608 -- TURGID DUPLICATION, needed only for the simplAlt call
1609 -- in mkDupableAlt. Clean this up when moving to FC
1610 = ASSERT( null bndrs )
1611 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1612 returnSmpl (Just (emptyVarEnv, (DEFAULT, [], rhs')))
1614 env' = addBinderOtherCon env case_bndr' handled_cons
1615 -- Record the constructors that the case-binder *can't* be.
1617 simplAlt env handled_cons case_bndr' cont' (LitAlt lit, bndrs, rhs)
1618 = ASSERT( null bndrs )
1619 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1620 returnSmpl (Just (emptyVarEnv, (LitAlt lit, [], rhs')))
1622 env' = addBinderUnfolding env case_bndr' (Lit lit)
1624 simplAlt env handled_cons case_bndr' cont' (DataAlt con, vs, rhs)
1625 = -- Deal with the pattern-bound variables
1626 -- Mark the ones that are in ! positions in the data constructor
1627 -- as certainly-evaluated.
1628 -- NB: it happens that simplBinders does *not* erase the OtherCon
1629 -- form of unfolding, so it's ok to add this info before
1630 -- doing simplBinders
1631 simplBinders env (add_evals con vs) `thenSmpl` \ (env, vs') ->
1633 -- Bind the case-binder to (con args)
1634 let inst_tys' = tyConAppArgs (idType case_bndr')
1635 con_args = map Type inst_tys' ++ varsToCoreExprs vs'
1636 env' = addBinderUnfolding env case_bndr' (mkConApp con con_args)
1638 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1639 returnSmpl (Just (emptyVarEnv, (DataAlt con, vs', rhs')))
1641 -- add_evals records the evaluated-ness of the bound variables of
1642 -- a case pattern. This is *important*. Consider
1643 -- data T = T !Int !Int
1645 -- case x of { T a b -> T (a+1) b }
1647 -- We really must record that b is already evaluated so that we don't
1648 -- go and re-evaluate it when constructing the result.
1649 add_evals dc vs = cat_evals dc vs (dataConRepStrictness dc)
1651 cat_evals dc vs strs
1655 go (v:vs) strs | isTyVar v = v : go vs strs
1656 go (v:vs) (str:strs)
1657 | isMarkedStrict str = evald_v : go vs strs
1658 | otherwise = zapped_v : go vs strs
1660 zapped_v = zap_occ_info v
1661 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1662 go _ _ = pprPanic "cat_evals" (ppr dc $$ ppr vs $$ ppr strs)
1664 -- If the case binder is alive, then we add the unfolding
1666 -- to the envt; so vs are now very much alive
1667 -- Note [Aug06] I can't see why this actually matters
1668 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1669 | otherwise = zapOccInfo
1671 addBinderUnfolding :: SimplEnv -> Id -> CoreExpr -> SimplEnv
1672 addBinderUnfolding env bndr rhs
1673 = modifyInScope env bndr (bndr `setIdUnfolding` mkUnfolding False rhs)
1675 addBinderOtherCon :: SimplEnv -> Id -> [AltCon] -> SimplEnv
1676 addBinderOtherCon env bndr cons
1677 = modifyInScope env bndr (bndr `setIdUnfolding` mkOtherCon cons)
1681 %************************************************************************
1683 \subsection{Known constructor}
1685 %************************************************************************
1687 We are a bit careful with occurrence info. Here's an example
1689 (\x* -> case x of (a*, b) -> f a) (h v, e)
1691 where the * means "occurs once". This effectively becomes
1692 case (h v, e) of (a*, b) -> f a)
1694 let a* = h v; b = e in f a
1698 All this should happen in one sweep.
1701 knownCon :: SimplEnv -> OutExpr -> AltCon -> [OutExpr]
1702 -> InId -> [InAlt] -> SimplCont
1703 -> SimplM FloatsWithExpr
1705 knownCon env scrut con args bndr alts cont
1706 = tick (KnownBranch bndr) `thenSmpl_`
1707 case findAlt con alts of
1708 (DEFAULT, bs, rhs) -> ASSERT( null bs )
1709 simplNonRecX env bndr scrut $ \ env ->
1710 -- This might give rise to a binding with non-atomic args
1711 -- like x = Node (f x) (g x)
1712 -- but simplNonRecX will atomic-ify it
1713 simplExprF env rhs cont
1715 (LitAlt lit, bs, rhs) -> ASSERT( null bs )
1716 simplNonRecX env bndr scrut $ \ env ->
1717 simplExprF env rhs cont
1719 (DataAlt dc, bs, rhs)
1720 -> -- ASSERT( n_drop_tys + length bs == length args )
1721 bind_args env dead_bndr bs (drop n_drop_tys args) $ \ env ->
1723 -- It's useful to bind bndr to scrut, rather than to a fresh
1724 -- binding x = Con arg1 .. argn
1725 -- because very often the scrut is a variable, so we avoid
1726 -- creating, and then subsequently eliminating, a let-binding
1727 -- BUT, if scrut is a not a variable, we must be careful
1728 -- about duplicating the arg redexes; in that case, make
1729 -- a new con-app from the args
1730 bndr_rhs = case scrut of
1733 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1734 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1735 -- args are aready OutExprs, but bs are InIds
1737 simplNonRecX env bndr bndr_rhs $ \ env ->
1738 simplExprF env rhs cont
1740 dead_bndr = isDeadBinder bndr
1741 n_drop_tys = tyConArity (dataConTyCon dc)
1744 bind_args env dead_bndr [] _ thing_inside = thing_inside env
1746 bind_args env dead_bndr (b:bs) (Type ty : args) thing_inside
1747 = ASSERT( isTyVar b )
1748 bind_args (extendTvSubst env b ty) dead_bndr bs args thing_inside
1750 bind_args env dead_bndr (b:bs) (arg : args) thing_inside
1753 b' = if dead_bndr then b else zapOccInfo b
1754 -- Note that the binder might be "dead", because it doesn't occur
1755 -- in the RHS; and simplNonRecX may therefore discard it via postInlineUnconditionally
1756 -- Nevertheless we must keep it if the case-binder is alive, because it may
1757 -- be used in the con_app. See Note [zapOccInfo]
1759 simplNonRecX env b' arg $ \ env ->
1760 bind_args env dead_bndr bs args thing_inside
1764 %************************************************************************
1766 \subsection{Duplicating continuations}
1768 %************************************************************************
1771 prepareCaseCont :: SimplEnv
1772 -> [InAlt] -> SimplCont
1773 -> SimplM (FloatsWith (SimplCont,SimplCont))
1774 -- Return a duplicatable continuation, a non-duplicable part
1775 -- plus some extra bindings (that scope over the entire
1778 -- No need to make it duplicatable if there's only one alternative
1779 prepareCaseCont env [alt] cont = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1780 prepareCaseCont env alts cont = mkDupableCont env cont
1784 mkDupableCont :: SimplEnv -> SimplCont
1785 -> SimplM (FloatsWith (SimplCont, SimplCont))
1787 mkDupableCont env cont
1788 | contIsDupable cont
1789 = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1791 mkDupableCont env (CoerceIt ty cont)
1792 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1793 returnSmpl (floats, (CoerceIt ty dup_cont, nondup_cont))
1795 mkDupableCont env cont@(ArgOf _ arg_ty _ _)
1796 = returnSmpl (emptyFloats env, (mkBoringStop arg_ty, cont))
1797 -- Do *not* duplicate an ArgOf continuation
1798 -- Because ArgOf continuations are opaque, we gain nothing by
1799 -- propagating them into the expressions, and we do lose a lot.
1800 -- Here's an example:
1801 -- && (case x of { T -> F; F -> T }) E
1802 -- Now, && is strict so we end up simplifying the case with
1803 -- an ArgOf continuation. If we let-bind it, we get
1805 -- let $j = \v -> && v E
1806 -- in simplExpr (case x of { T -> F; F -> T })
1807 -- (ArgOf (\r -> $j r)
1808 -- And after simplifying more we get
1810 -- let $j = \v -> && v E
1811 -- in case of { T -> $j F; F -> $j T }
1812 -- Which is a Very Bad Thing
1814 -- The desire not to duplicate is the entire reason that
1815 -- mkDupableCont returns a pair of continuations.
1817 -- The original plan had:
1818 -- e.g. (...strict-fn...) [...hole...]
1820 -- let $j = \a -> ...strict-fn...
1821 -- in $j [...hole...]
1823 mkDupableCont env (ApplyTo _ arg mb_se cont)
1824 = -- e.g. [...hole...] (...arg...)
1826 -- let a = ...arg...
1827 -- in [...hole...] a
1828 do { (floats, (dup_cont, nondup_cont)) <- mkDupableCont env cont
1829 ; addFloats env floats $ \ env -> do
1830 { arg1 <- simplArg env arg mb_se
1831 ; (floats2, arg2) <- mkDupableArg env arg1
1832 ; return (floats2, (ApplyTo OkToDup arg2 Nothing dup_cont, nondup_cont)) }}
1834 mkDupableCont env cont@(Select _ case_bndr [(_,bs,rhs)] se case_cont)
1835 -- | not (exprIsDupable rhs && contIsDupable case_cont) -- See notes below
1836 -- | not (isDeadBinder case_bndr)
1837 | all isDeadBinder bs
1838 = returnSmpl (emptyFloats env, (mkBoringStop scrut_ty, cont))
1840 scrut_ty = substTy se (idType case_bndr)
1842 {- Note [Single-alternative cases]
1843 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1844 This case is just like the ArgOf case. Here's an example:
1848 case (case x of I# x' ->
1850 True -> I# (negate# x')
1851 False -> I# x') of y {
1853 Because the (case x) has only one alternative, we'll transform to
1855 case (case x' <# 0# of
1856 True -> I# (negate# x')
1857 False -> I# x') of y {
1859 But now we do *NOT* want to make a join point etc, giving
1861 let $j = \y -> MkT y
1863 True -> $j (I# (negate# x'))
1865 In this case the $j will inline again, but suppose there was a big
1866 strict computation enclosing the orginal call to MkT. Then, it won't
1867 "see" the MkT any more, because it's big and won't get duplicated.
1868 And, what is worse, nothing was gained by the case-of-case transform.
1870 When should use this case of mkDupableCont?
1871 However, matching on *any* single-alternative case is a *disaster*;
1872 e.g. case (case ....) of (a,b) -> (# a,b #)
1873 We must push the outer case into the inner one!
1876 * Match [(DEFAULT,_,_)], but in the common case of Int,
1877 the alternative-filling-in code turned the outer case into
1878 case (...) of y { I# _ -> MkT y }
1880 * Match on single alternative plus (not (isDeadBinder case_bndr))
1881 Rationale: pushing the case inwards won't eliminate the construction.
1882 But there's a risk of
1883 case (...) of y { (a,b) -> let z=(a,b) in ... }
1884 Now y looks dead, but it'll come alive again. Still, this
1885 seems like the best option at the moment.
1887 * Match on single alternative plus (all (isDeadBinder bndrs))
1888 Rationale: this is essentially seq.
1890 * Match when the rhs is *not* duplicable, and hence would lead to a
1891 join point. This catches the disaster-case above. We can test
1892 the *un-simplified* rhs, which is fine. It might get bigger or
1893 smaller after simplification; if it gets smaller, this case might
1894 fire next time round. NB also that we must test contIsDupable
1895 case_cont *btoo, because case_cont might be big!
1897 HOWEVER: I found that this version doesn't work well, because
1898 we can get let x = case (...) of { small } in ...case x...
1899 When x is inlined into its full context, we find that it was a bad
1900 idea to have pushed the outer case inside the (...) case.
1903 mkDupableCont env (Select _ case_bndr alts se cont)
1904 = -- e.g. (case [...hole...] of { pi -> ei })
1906 -- let ji = \xij -> ei
1907 -- in case [...hole...] of { pi -> ji xij }
1908 do { tick (CaseOfCase case_bndr)
1909 ; let alt_env = setInScope se env
1910 ; (floats1, (dup_cont, nondup_cont)) <- mkDupableCont alt_env cont
1911 -- NB: call mkDupableCont here, *not* prepareCaseCont
1912 -- We must make a duplicable continuation, whereas prepareCaseCont
1913 -- doesn't when there is a single case branch
1914 ; addFloats alt_env floats1 $ \ alt_env -> do
1916 { (alt_env, case_bndr') <- simplBinder alt_env case_bndr
1917 -- NB: simplBinder does not zap deadness occ-info, so
1918 -- a dead case_bndr' will still advertise its deadness
1919 -- This is really important because in
1920 -- case e of b { (# a,b #) -> ... }
1921 -- b is always dead, and indeed we are not allowed to bind b to (# a,b #),
1922 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1923 -- In the new alts we build, we have the new case binder, so it must retain
1926 ; (floats2, alts') <- mkDupableAlts alt_env case_bndr' alts dup_cont
1927 ; return (floats2, (Select OkToDup case_bndr' alts' (zapSubstEnv se)
1928 (mkBoringStop (contResultType dup_cont)),
1932 mkDupableArg :: SimplEnv -> OutExpr -> SimplM (FloatsWith OutExpr)
1933 -- Let-bind the thing if necessary
1934 mkDupableArg env arg
1936 = return (emptyFloats env, arg)
1938 = do { arg_id <- newId FSLIT("a") (exprType arg)
1939 ; tick (CaseOfCase arg_id)
1940 -- Want to tick here so that we go round again,
1941 -- and maybe copy or inline the code.
1942 -- Not strictly CaseOfCase, but never mind
1943 ; return (unitFloat env arg_id arg, Var arg_id) }
1944 -- What if the arg should be case-bound?
1945 -- This has been this way for a long time, so I'll leave it,
1946 -- but I can't convince myself that it's right.
1948 mkDupableAlts :: SimplEnv -> OutId -> [InAlt] -> SimplCont
1949 -> SimplM (FloatsWith [InAlt])
1950 -- Absorbs the continuation into the new alternatives
1952 mkDupableAlts env case_bndr' alts dupable_cont
1955 go env [] = returnSmpl (emptyFloats env, [])
1957 = do { (floats1, mb_alt') <- mkDupableAlt env case_bndr' dupable_cont alt
1958 ; addFloats env floats1 $ \ env -> do
1959 { (floats2, alts') <- go env alts
1960 ; returnSmpl (floats2, case mb_alt' of
1961 Just alt' -> alt' : alts'
1965 mkDupableAlt env case_bndr' cont alt
1966 = simplAlt env [] case_bndr' cont alt `thenSmpl` \ mb_stuff ->
1968 Nothing -> returnSmpl (emptyFloats env, Nothing) ;
1970 Just (reft, (con, bndrs', rhs')) ->
1971 -- Safe to say that there are no handled-cons for the DEFAULT case
1973 if exprIsDupable rhs' then
1974 returnSmpl (emptyFloats env, Just (con, bndrs', rhs'))
1975 -- It is worth checking for a small RHS because otherwise we
1976 -- get extra let bindings that may cause an extra iteration of the simplifier to
1977 -- inline back in place. Quite often the rhs is just a variable or constructor.
1978 -- The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1979 -- iterations because the version with the let bindings looked big, and so wasn't
1980 -- inlined, but after the join points had been inlined it looked smaller, and so
1983 -- NB: we have to check the size of rhs', not rhs.
1984 -- Duplicating a small InAlt might invalidate occurrence information
1985 -- However, if it *is* dupable, we return the *un* simplified alternative,
1986 -- because otherwise we'd need to pair it up with an empty subst-env....
1987 -- but we only have one env shared between all the alts.
1988 -- (Remember we must zap the subst-env before re-simplifying something).
1989 -- Rather than do this we simply agree to re-simplify the original (small) thing later.
1993 rhs_ty' = exprType rhs'
1994 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1996 | isTyVar bndr = not (bndr `elemVarEnv` reft)
1997 -- Don't abstract over tyvar binders which are refined away
1998 -- See Note [Refinement] below
1999 | otherwise = not (isDeadBinder bndr)
2000 -- The deadness info on the new Ids is preserved by simplBinders
2002 -- If we try to lift a primitive-typed something out
2003 -- for let-binding-purposes, we will *caseify* it (!),
2004 -- with potentially-disastrous strictness results. So
2005 -- instead we turn it into a function: \v -> e
2006 -- where v::State# RealWorld#. The value passed to this function
2007 -- is realworld#, which generates (almost) no code.
2009 -- There's a slight infelicity here: we pass the overall
2010 -- case_bndr to all the join points if it's used in *any* RHS,
2011 -- because we don't know its usage in each RHS separately
2013 -- We used to say "&& isUnLiftedType rhs_ty'" here, but now
2014 -- we make the join point into a function whenever used_bndrs'
2015 -- is empty. This makes the join-point more CPR friendly.
2016 -- Consider: let j = if .. then I# 3 else I# 4
2017 -- in case .. of { A -> j; B -> j; C -> ... }
2019 -- Now CPR doesn't w/w j because it's a thunk, so
2020 -- that means that the enclosing function can't w/w either,
2021 -- which is a lose. Here's the example that happened in practice:
2022 -- kgmod :: Int -> Int -> Int
2023 -- kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
2027 -- I have seen a case alternative like this:
2028 -- True -> \v -> ...
2029 -- It's a bit silly to add the realWorld dummy arg in this case, making
2032 -- (the \v alone is enough to make CPR happy) but I think it's rare
2034 ( if not (any isId used_bndrs')
2035 then newId FSLIT("w") realWorldStatePrimTy `thenSmpl` \ rw_id ->
2036 returnSmpl ([rw_id], [Var realWorldPrimId])
2038 returnSmpl (used_bndrs', varsToCoreExprs used_bndrs')
2039 ) `thenSmpl` \ (final_bndrs', final_args) ->
2041 -- See comment about "$j" name above
2042 newId FSLIT("$j") (mkPiTypes final_bndrs' rhs_ty') `thenSmpl` \ join_bndr ->
2043 -- Notice the funky mkPiTypes. If the contructor has existentials
2044 -- it's possible that the join point will be abstracted over
2045 -- type varaibles as well as term variables.
2046 -- Example: Suppose we have
2047 -- data T = forall t. C [t]
2049 -- case (case e of ...) of
2050 -- C t xs::[t] -> rhs
2051 -- We get the join point
2052 -- let j :: forall t. [t] -> ...
2053 -- j = /\t \xs::[t] -> rhs
2055 -- case (case e of ...) of
2056 -- C t xs::[t] -> j t xs
2058 -- We make the lambdas into one-shot-lambdas. The
2059 -- join point is sure to be applied at most once, and doing so
2060 -- prevents the body of the join point being floated out by
2061 -- the full laziness pass
2062 really_final_bndrs = map one_shot final_bndrs'
2063 one_shot v | isId v = setOneShotLambda v
2065 join_rhs = mkLams really_final_bndrs rhs'
2066 join_call = mkApps (Var join_bndr) final_args
2068 returnSmpl (unitFloat env join_bndr join_rhs, Just (con, bndrs', join_call)) }
2075 MkT :: a -> b -> T a
2079 MkT a' b (p::a') (q::b) -> [p,w]
2081 The danger is that we'll make a join point
2085 and that's ill-typed, because (p::a') but (w::a).
2087 Solution so far: don't abstract over a', because the type refinement
2088 maps [a' -> a] . Ultimately that won't work when real refinement goes on.
2090 Then we must abstract over any refined free variables. Hmm. Maybe we
2091 could just abstract over *all* free variables, thereby lambda-lifting
2092 the join point? We should try this.