2 % (c) The AQUA Project, Glasgow University, 1993-1998
4 \section[Simplify]{The main module of the simplifier}
7 module Simplify ( simplTopBinds, simplExpr ) where
9 #include "HsVersions.h"
11 import DynFlags ( dopt, DynFlag(Opt_D_dump_inlinings),
16 import SimplUtils ( mkCase, mkLam, prepareAlts,
17 SimplCont(..), DupFlag(..), LetRhsFlag(..),
18 mkRhsStop, mkBoringStop, pushContArgs,
19 contResultType, countArgs, contIsDupable, contIsRhsOrArg,
20 getContArgs, interestingCallContext, interestingArg, isStrictType,
21 preInlineUnconditionally, postInlineUnconditionally,
22 inlineMode, activeInline, activeRule
24 import Id ( Id, idType, idInfo, idArity, isDataConWorkId,
25 setIdUnfolding, isDeadBinder,
26 idNewDemandInfo, setIdInfo,
27 setIdOccInfo, zapLamIdInfo, setOneShotLambda
29 import MkId ( eRROR_ID )
30 import Literal ( mkStringLit )
31 import IdInfo ( OccInfo(..), isLoopBreaker,
32 setArityInfo, zapDemandInfo,
36 import NewDemand ( isStrictDmd )
37 import Unify ( coreRefineTys )
38 import DataCon ( dataConTyCon, dataConRepStrictness, isVanillaDataCon )
39 import TyCon ( tyConArity )
41 import PprCore ( pprParendExpr, pprCoreExpr )
42 import CoreUnfold ( mkUnfolding, callSiteInline )
43 import CoreUtils ( exprIsDupable, exprIsTrivial, needsCaseBinding,
44 exprIsConApp_maybe, mkPiTypes, findAlt,
46 exprOkForSpeculation, exprArity,
47 mkCoerce, mkCoerce2, mkSCC, mkInlineMe, applyTypeToArg
49 import Rules ( lookupRule )
50 import BasicTypes ( isMarkedStrict )
51 import CostCentre ( currentCCS )
52 import Type ( TvSubstEnv, isUnLiftedType, seqType, tyConAppArgs, funArgTy,
53 splitFunTy_maybe, splitFunTy, coreEqType
55 import VarEnv ( elemVarEnv, emptyVarEnv )
56 import TysPrim ( realWorldStatePrimTy )
57 import PrelInfo ( realWorldPrimId )
58 import BasicTypes ( TopLevelFlag(..), isTopLevel,
61 import StaticFlags ( opt_PprStyle_Debug )
63 import Maybes ( orElse )
65 import Util ( notNull )
69 The guts of the simplifier is in this module, but the driver loop for
70 the simplifier is in SimplCore.lhs.
73 -----------------------------------------
74 *** IMPORTANT NOTE ***
75 -----------------------------------------
76 The simplifier used to guarantee that the output had no shadowing, but
77 it does not do so any more. (Actually, it never did!) The reason is
78 documented with simplifyArgs.
81 -----------------------------------------
82 *** IMPORTANT NOTE ***
83 -----------------------------------------
84 Many parts of the simplifier return a bunch of "floats" as well as an
85 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
87 All "floats" are let-binds, not case-binds, but some non-rec lets may
88 be unlifted (with RHS ok-for-speculation).
92 -----------------------------------------
93 ORGANISATION OF FUNCTIONS
94 -----------------------------------------
96 - simplify all top-level binders
97 - for NonRec, call simplRecOrTopPair
98 - for Rec, call simplRecBind
101 ------------------------------
102 simplExpr (applied lambda) ==> simplNonRecBind
103 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
104 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
106 ------------------------------
107 simplRecBind [binders already simplfied]
108 - use simplRecOrTopPair on each pair in turn
110 simplRecOrTopPair [binder already simplified]
111 Used for: recursive bindings (top level and nested)
112 top-level non-recursive bindings
114 - check for PreInlineUnconditionally
118 Used for: non-top-level non-recursive bindings
119 beta reductions (which amount to the same thing)
120 Because it can deal with strict arts, it takes a
121 "thing-inside" and returns an expression
123 - check for PreInlineUnconditionally
124 - simplify binder, including its IdInfo
133 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
134 Used for: binding case-binder and constr args in a known-constructor case
135 - check for PreInLineUnconditionally
139 ------------------------------
140 simplLazyBind: [binder already simplified, RHS not]
141 Used for: recursive bindings (top level and nested)
142 top-level non-recursive bindings
143 non-top-level, but *lazy* non-recursive bindings
144 [must not be strict or unboxed]
145 Returns floats + an augmented environment, not an expression
146 - substituteIdInfo and add result to in-scope
147 [so that rules are available in rec rhs]
150 - float if exposes constructor or PAP
154 completeNonRecX: [binder and rhs both simplified]
155 - if the the thing needs case binding (unlifted and not ok-for-spec)
161 completeLazyBind: [given a simplified RHS]
162 [used for both rec and non-rec bindings, top level and not]
163 - try PostInlineUnconditionally
164 - add unfolding [this is the only place we add an unfolding]
169 Right hand sides and arguments
170 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
171 In many ways we want to treat
172 (a) the right hand side of a let(rec), and
173 (b) a function argument
174 in the same way. But not always! In particular, we would
175 like to leave these arguments exactly as they are, so they
176 will match a RULE more easily.
181 It's harder to make the rule match if we ANF-ise the constructor,
182 or eta-expand the PAP:
184 f (let { a = g x; b = h x } in (a,b))
187 On the other hand if we see the let-defns
192 then we *do* want to ANF-ise and eta-expand, so that p and q
193 can be safely inlined.
195 Even floating lets out is a bit dubious. For let RHS's we float lets
196 out if that exposes a value, so that the value can be inlined more vigorously.
199 r = let x = e in (x,x)
201 Here, if we float the let out we'll expose a nice constructor. We did experiments
202 that showed this to be a generally good thing. But it was a bad thing to float
203 lets out unconditionally, because that meant they got allocated more often.
205 For function arguments, there's less reason to expose a constructor (it won't
206 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
207 So for the moment we don't float lets out of function arguments either.
212 For eta expansion, we want to catch things like
214 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
216 If the \x was on the RHS of a let, we'd eta expand to bring the two
217 lambdas together. And in general that's a good thing to do. Perhaps
218 we should eta expand wherever we find a (value) lambda? Then the eta
219 expansion at a let RHS can concentrate solely on the PAP case.
222 %************************************************************************
224 \subsection{Bindings}
226 %************************************************************************
229 simplTopBinds :: SimplEnv -> [InBind] -> SimplM [OutBind]
231 simplTopBinds env binds
232 = -- Put all the top-level binders into scope at the start
233 -- so that if a transformation rule has unexpectedly brought
234 -- anything into scope, then we don't get a complaint about that.
235 -- It's rather as if the top-level binders were imported.
236 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 if needsCaseBinding bndr_ty rhs1
320 thing_inside env2 `thenSmpl` \ (floats, body) ->
321 returnSmpl (emptyFloats env2, Case rhs1 bndr2 (exprType body)
322 [(DEFAULT, [], wrapFloats floats body)])
324 completeNonRecX env2 True {- strict -} bndr bndr2 rhs1 thing_inside
326 | otherwise -- Normal, lazy case
327 = -- Don't use simplBinder because that doesn't keep
328 -- fragile occurrence info in the substitution
329 simplNonRecBndr env bndr `thenSmpl` \ (env, bndr') ->
330 simplLazyBind env NotTopLevel NonRecursive
331 bndr bndr' rhs rhs_se `thenSmpl` \ (floats, env) ->
332 addFloats env floats thing_inside
335 bndr_ty = idType bndr
338 A specialised variant of simplNonRec used when the RHS is already simplified, notably
339 in knownCon. It uses case-binding where necessary.
342 simplNonRecX :: SimplEnv
343 -> InId -- Old binder
344 -> OutExpr -- Simplified RHS
345 -> (SimplEnv -> SimplM FloatsWithExpr)
346 -> SimplM FloatsWithExpr
348 simplNonRecX env bndr new_rhs thing_inside
349 | needsCaseBinding (idType bndr) new_rhs
350 -- Make this test *before* the preInlineUnconditionally
351 -- Consider case I# (quotInt# x y) of
352 -- I# v -> let w = J# v in ...
353 -- If we gaily inline (quotInt# x y) for v, we end up building an
355 -- let w = J# (quotInt# x y) in ...
356 -- because quotInt# can fail.
357 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
358 thing_inside env `thenSmpl` \ (floats, body) ->
359 let body' = wrapFloats floats body in
360 returnSmpl (emptyFloats env, Case new_rhs bndr' (exprType body') [(DEFAULT, [], body')])
362 | preInlineUnconditionally env NotTopLevel bndr new_rhs
363 -- This happens; for example, the case_bndr during case of
364 -- known constructor: case (a,b) of x { (p,q) -> ... }
365 -- Here x isn't mentioned in the RHS, so we don't want to
366 -- create the (dead) let-binding let x = (a,b) in ...
368 -- Similarly, single occurrences can be inlined vigourously
369 -- e.g. case (f x, g y) of (a,b) -> ....
370 -- If a,b occur once we can avoid constructing the let binding for them.
371 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
374 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
375 completeNonRecX env False {- Non-strict; pessimistic -}
376 bndr bndr' new_rhs thing_inside
378 completeNonRecX env is_strict old_bndr new_bndr new_rhs thing_inside
379 = mkAtomicArgs is_strict
380 True {- OK to float unlifted -}
381 new_rhs `thenSmpl` \ (aux_binds, rhs2) ->
383 -- Make the arguments atomic if necessary,
384 -- adding suitable bindings
385 addAtomicBindsE env (fromOL aux_binds) $ \ env ->
386 completeLazyBind env NotTopLevel
387 old_bndr new_bndr rhs2 `thenSmpl` \ (floats, env) ->
388 addFloats env floats thing_inside
392 %************************************************************************
394 \subsection{Lazy bindings}
396 %************************************************************************
398 simplRecBind is used for
399 * recursive bindings only
402 simplRecBind :: SimplEnv -> TopLevelFlag
403 -> [(InId, InExpr)] -> [OutId]
404 -> SimplM (FloatsWith SimplEnv)
405 simplRecBind env top_lvl pairs bndrs'
406 = go env pairs bndrs' `thenSmpl` \ (floats, env) ->
407 returnSmpl (flattenFloats floats, env)
409 go env [] _ = returnSmpl (emptyFloats env, env)
411 go env ((bndr, rhs) : pairs) (bndr' : bndrs')
412 = simplRecOrTopPair env top_lvl bndr bndr' rhs `thenSmpl` \ (floats, env) ->
413 addFloats env floats (\env -> go env pairs bndrs')
417 simplRecOrTopPair is used for
418 * recursive bindings (whether top level or not)
419 * top-level non-recursive bindings
421 It assumes the binder has already been simplified, but not its IdInfo.
424 simplRecOrTopPair :: SimplEnv
426 -> InId -> OutId -- Binder, both pre-and post simpl
427 -> InExpr -- The RHS and its environment
428 -> SimplM (FloatsWith SimplEnv)
430 simplRecOrTopPair env top_lvl bndr bndr' rhs
431 | preInlineUnconditionally env top_lvl bndr rhs -- Check for unconditional inline
432 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
433 returnSmpl (emptyFloats env, extendIdSubst env bndr (mkContEx env rhs))
436 = simplLazyBind env top_lvl Recursive bndr bndr' rhs env
437 -- May not actually be recursive, but it doesn't matter
441 simplLazyBind is used for
442 * recursive bindings (whether top level or not)
443 * top-level non-recursive bindings
444 * non-top-level *lazy* non-recursive bindings
446 [Thus it deals with the lazy cases from simplNonRecBind, and all cases
447 from SimplRecOrTopBind]
450 1. It assumes that the binder is *already* simplified,
451 and is in scope, but not its IdInfo
453 2. It assumes that the binder type is lifted.
455 3. It does not check for pre-inline-unconditionallly;
456 that should have been done already.
459 simplLazyBind :: SimplEnv
460 -> TopLevelFlag -> RecFlag
461 -> InId -> OutId -- Binder, both pre-and post simpl
462 -> InExpr -> SimplEnv -- The RHS and its environment
463 -> SimplM (FloatsWith SimplEnv)
465 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
467 (env1,bndr2) = addLetIdInfo env bndr bndr1
468 rhs_env = setInScope rhs_se env1
469 is_top_level = isTopLevel top_lvl
470 ok_float_unlifted = not is_top_level && isNonRec is_rec
471 rhs_cont = mkRhsStop (idType bndr2)
473 -- Simplify the RHS; note the mkRhsStop, which tells
474 -- the simplifier that this is the RHS of a let.
475 simplExprF rhs_env rhs rhs_cont `thenSmpl` \ (floats, rhs1) ->
477 -- If any of the floats can't be floated, give up now
478 -- (The allLifted predicate says True for empty floats.)
479 if (not ok_float_unlifted && not (allLifted floats)) then
480 completeLazyBind env1 top_lvl bndr bndr2
481 (wrapFloats floats rhs1)
484 -- ANF-ise a constructor or PAP rhs
485 mkAtomicArgs False {- Not strict -}
486 ok_float_unlifted rhs1 `thenSmpl` \ (aux_binds, rhs2) ->
488 -- If the result is a PAP, float the floats out, else wrap them
489 -- By this time it's already been ANF-ised (if necessary)
490 if isEmptyFloats floats && isNilOL aux_binds then -- Shortcut a common case
491 completeLazyBind env1 top_lvl bndr bndr2 rhs2
493 else if is_top_level || exprIsTrivial rhs2 || exprIsHNF rhs2 then
494 -- WARNING: long dodgy argument coming up
495 -- WANTED: a better way to do this
497 -- We can't use "exprIsCheap" instead of exprIsHNF,
498 -- because that causes a strictness bug.
499 -- x = let y* = E in case (scc y) of { T -> F; F -> T}
500 -- The case expression is 'cheap', but it's wrong to transform to
501 -- y* = E; x = case (scc y) of {...}
502 -- Either we must be careful not to float demanded non-values, or
503 -- we must use exprIsHNF for the test, which ensures that the
504 -- thing is non-strict. So exprIsHNF => bindings are non-strict
505 -- I think. The WARN below tests for this.
507 -- We use exprIsTrivial here because we want to reveal lone variables.
508 -- E.g. let { x = letrec { y = E } in y } in ...
509 -- Here we definitely want to float the y=E defn.
510 -- exprIsHNF definitely isn't right for that.
512 -- Again, the floated binding can't be strict; if it's recursive it'll
513 -- be non-strict; if it's non-recursive it'd be inlined.
515 -- Note [SCC-and-exprIsTrivial]
517 -- y = let { x* = E } in scc "foo" x
518 -- then we do *not* want to float out the x binding, because
519 -- it's strict! Fortunately, exprIsTrivial replies False to
522 -- There's a subtlety here. There may be a binding (x* = e) in the
523 -- floats, where the '*' means 'will be demanded'. So is it safe
524 -- to float it out? Answer no, but it won't matter because
525 -- we only float if (a) arg' is a WHNF, or (b) it's going to top level
526 -- and so there can't be any 'will be demanded' bindings in the floats.
528 ASSERT2( is_top_level || not (any demanded_float (floatBinds floats)),
529 ppr (filter demanded_float (floatBinds floats)) )
531 tick LetFloatFromLet `thenSmpl_` (
532 addFloats env1 floats $ \ env2 ->
533 addAtomicBinds env2 (fromOL aux_binds) $ \ env3 ->
534 completeLazyBind env3 top_lvl bndr bndr2 rhs2)
537 completeLazyBind env1 top_lvl bndr bndr2 (wrapFloats floats rhs1)
540 demanded_float (NonRec b r) = isStrictDmd (idNewDemandInfo b) && not (isUnLiftedType (idType b))
541 -- Unlifted-type (cheap-eagerness) lets may well have a demanded flag on them
542 demanded_float (Rec _) = False
547 %************************************************************************
549 \subsection{Completing a lazy binding}
551 %************************************************************************
554 * deals only with Ids, not TyVars
555 * takes an already-simplified binder and RHS
556 * is used for both recursive and non-recursive bindings
557 * is used for both top-level and non-top-level bindings
559 It does the following:
560 - tries discarding a dead binding
561 - tries PostInlineUnconditionally
562 - add unfolding [this is the only place we add an unfolding]
565 It does *not* attempt to do let-to-case. Why? Because it is used for
566 - top-level bindings (when let-to-case is impossible)
567 - many situations where the "rhs" is known to be a WHNF
568 (so let-to-case is inappropriate).
571 completeLazyBind :: SimplEnv
572 -> TopLevelFlag -- Flag stuck into unfolding
573 -> InId -- Old binder
574 -> OutId -- New binder
575 -> OutExpr -- Simplified RHS
576 -> SimplM (FloatsWith SimplEnv)
577 -- We return a new SimplEnv, because completeLazyBind may choose to do its work
578 -- by extending the substitution (e.g. let x = y in ...)
579 -- The new binding (if any) is returned as part of the floats.
580 -- NB: the returned SimplEnv has the right SubstEnv, but you should
581 -- (as usual) use the in-scope-env from the floats
583 completeLazyBind env top_lvl old_bndr new_bndr new_rhs
584 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
585 = -- Drop the binding
586 tick (PostInlineUnconditionally old_bndr) `thenSmpl_`
587 returnSmpl (emptyFloats env, extendIdSubst env old_bndr (DoneEx new_rhs))
588 -- Use the substitution to make quite, quite sure that the substitution
589 -- will happen, since we are going to discard the binding
594 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
596 -- Add the unfolding *only* for non-loop-breakers
597 -- Making loop breakers not have an unfolding at all
598 -- means that we can avoid tests in exprIsConApp, for example.
599 -- This is important: if exprIsConApp says 'yes' for a recursive
600 -- thing, then we can get into an infinite loop
602 -- If the unfolding is a value, the demand info may
603 -- go pear-shaped, so we nuke it. Example:
605 -- case x of (p,q) -> h p q x
606 -- Here x is certainly demanded. But after we've nuked
607 -- the case, we'll get just
608 -- let x = (a,b) in h a b x
609 -- and now x is not demanded (I'm assuming h is lazy)
610 -- This really happens. Similarly
611 -- let f = \x -> e in ...f..f...
612 -- After inling f at some of its call sites the original binding may
613 -- (for example) be no longer strictly demanded.
614 -- The solution here is a bit ad hoc...
615 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
616 final_info | loop_breaker = new_bndr_info
617 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
618 | otherwise = info_w_unf
620 final_id = new_bndr `setIdInfo` final_info
622 -- These seqs forces the Id, and hence its IdInfo,
623 -- and hence any inner substitutions
625 returnSmpl (unitFloat env final_id new_rhs, env)
628 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
629 loop_breaker = isLoopBreaker occ_info
630 old_info = idInfo old_bndr
631 occ_info = occInfo old_info
636 %************************************************************************
638 \subsection[Simplify-simplExpr]{The main function: simplExpr}
640 %************************************************************************
642 The reason for this OutExprStuff stuff is that we want to float *after*
643 simplifying a RHS, not before. If we do so naively we get quadratic
644 behaviour as things float out.
646 To see why it's important to do it after, consider this (real) example:
660 a -- Can't inline a this round, cos it appears twice
664 Each of the ==> steps is a round of simplification. We'd save a
665 whole round if we float first. This can cascade. Consider
670 let f = let d1 = ..d.. in \y -> e
674 in \x -> ...(\y ->e)...
676 Only in this second round can the \y be applied, and it
677 might do the same again.
681 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
682 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
684 expr_ty' = substTy env (exprType expr)
685 -- The type in the Stop continuation, expr_ty', is usually not used
686 -- It's only needed when discarding continuations after finding
687 -- a function that returns bottom.
688 -- Hence the lazy substitution
691 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
692 -- Simplify an expression, given a continuation
693 simplExprC env expr cont
694 = simplExprF env expr cont `thenSmpl` \ (floats, expr) ->
695 returnSmpl (wrapFloats floats expr)
697 simplExprF :: SimplEnv -> InExpr -> SimplCont -> SimplM FloatsWithExpr
698 -- Simplify an expression, returning floated binds
700 simplExprF env (Var v) cont = simplVar env v cont
701 simplExprF env (Lit lit) cont = rebuild env (Lit lit) cont
702 simplExprF env expr@(Lam _ _) cont = simplLam env expr cont
703 simplExprF env (Note note expr) cont = simplNote env note expr cont
704 simplExprF env (App fun arg) cont = simplExprF env fun (ApplyTo NoDup arg env cont)
706 simplExprF env (Type ty) cont
707 = ASSERT( contIsRhsOrArg cont )
708 simplType env ty `thenSmpl` \ ty' ->
709 rebuild env (Type ty') cont
711 simplExprF env (Case scrut bndr case_ty alts) cont
712 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
713 = -- Simplify the scrutinee with a Select continuation
714 simplExprF env scrut (Select NoDup bndr alts env cont)
717 = -- If case-of-case is off, simply simplify the case expression
718 -- in a vanilla Stop context, and rebuild the result around it
719 simplExprC env scrut case_cont `thenSmpl` \ case_expr' ->
720 rebuild env case_expr' cont
722 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
723 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
725 simplExprF env (Let (Rec pairs) body) cont
726 = simplRecBndrs env (map fst pairs) `thenSmpl` \ (env, bndrs') ->
727 -- NB: bndrs' don't have unfoldings or rules
728 -- We add them as we go down
730 simplRecBind env NotTopLevel pairs bndrs' `thenSmpl` \ (floats, env) ->
731 addFloats env floats $ \ env ->
732 simplExprF env body cont
734 -- A non-recursive let is dealt with by simplNonRecBind
735 simplExprF env (Let (NonRec bndr rhs) body) cont
736 = simplNonRecBind env bndr rhs env (contResultType cont) $ \ env ->
737 simplExprF env body cont
740 ---------------------------------
741 simplType :: SimplEnv -> InType -> SimplM OutType
742 -- Kept monadic just so we can do the seqType
744 = seqType new_ty `seq` returnSmpl new_ty
746 new_ty = substTy env ty
750 %************************************************************************
754 %************************************************************************
757 simplLam env fun cont
760 zap_it = mkLamBndrZapper fun (countArgs cont)
761 cont_ty = contResultType cont
763 -- Type-beta reduction
764 go env (Lam bndr body) (ApplyTo _ (Type ty_arg) arg_se body_cont)
765 = ASSERT( isTyVar bndr )
766 tick (BetaReduction bndr) `thenSmpl_`
767 simplType (setInScope arg_se env) ty_arg `thenSmpl` \ ty_arg' ->
768 go (extendTvSubst env bndr ty_arg') body body_cont
770 -- Ordinary beta reduction
771 go env (Lam bndr body) cont@(ApplyTo _ arg arg_se body_cont)
772 = tick (BetaReduction bndr) `thenSmpl_`
773 simplNonRecBind env (zap_it bndr) arg arg_se cont_ty $ \ env ->
774 go env body body_cont
776 -- Not enough args, so there are real lambdas left to put in the result
777 go env lam@(Lam _ _) cont
778 = simplLamBndrs env bndrs `thenSmpl` \ (env, bndrs') ->
779 simplExpr env body `thenSmpl` \ body' ->
780 mkLam env bndrs' body' cont `thenSmpl` \ (floats, new_lam) ->
781 addFloats env floats $ \ env ->
782 rebuild env new_lam cont
784 (bndrs,body) = collectBinders lam
786 -- Exactly enough args
787 go env expr cont = simplExprF env expr cont
789 mkLamBndrZapper :: CoreExpr -- Function
790 -> Int -- Number of args supplied, *including* type args
791 -> Id -> Id -- Use this to zap the binders
792 mkLamBndrZapper fun n_args
793 | n_args >= n_params fun = \b -> b -- Enough args
794 | otherwise = \b -> zapLamIdInfo b
796 -- NB: we count all the args incl type args
797 -- so we must count all the binders (incl type lambdas)
798 n_params (Note _ e) = n_params e
799 n_params (Lam b e) = 1 + n_params e
800 n_params other = 0::Int
804 %************************************************************************
808 %************************************************************************
811 simplNote env (Coerce to from) body cont
813 addCoerce s1 k1 cont -- Drop redundant coerces. This can happen if a polymoprhic
814 -- (coerce a b e) is instantiated with a=ty1 b=ty2 and the
815 -- two are the same. This happens a lot in Happy-generated parsers
816 | s1 `coreEqType` k1 = cont
818 addCoerce s1 k1 (CoerceIt t1 cont)
819 -- coerce T1 S1 (coerce S1 K1 e)
822 -- coerce T1 K1 e, otherwise
824 -- For example, in the initial form of a worker
825 -- we may find (coerce T (coerce S (\x.e))) y
826 -- and we'd like it to simplify to e[y/x] in one round
828 | t1 `coreEqType` k1 = cont -- The coerces cancel out
829 | otherwise = CoerceIt t1 cont -- They don't cancel, but
830 -- the inner one is redundant
832 addCoerce t1t2 s1s2 (ApplyTo dup arg arg_se cont)
833 | not (isTypeArg arg), -- This whole case only works for value args
834 -- Could upgrade to have equiv thing for type apps too
835 Just (s1, s2) <- splitFunTy_maybe s1s2
836 -- (coerce (T1->T2) (S1->S2) F) E
838 -- coerce T2 S2 (F (coerce S1 T1 E))
840 -- t1t2 must be a function type, T1->T2, because it's applied to something
841 -- but s1s2 might conceivably not be
843 -- When we build the ApplyTo we can't mix the out-types
844 -- with the InExpr in the argument, so we simply substitute
845 -- to make it all consistent. It's a bit messy.
846 -- But it isn't a common case.
848 (t1,t2) = splitFunTy t1t2
849 new_arg = mkCoerce2 s1 t1 (substExpr arg_env arg)
850 arg_env = setInScope arg_se env
852 ApplyTo dup new_arg (zapSubstEnv env) (addCoerce t2 s2 cont)
854 addCoerce to' _ cont = CoerceIt to' cont
856 simplType env to `thenSmpl` \ to' ->
857 simplType env from `thenSmpl` \ from' ->
858 simplExprF env body (addCoerce to' from' cont)
861 -- Hack: we only distinguish subsumed cost centre stacks for the purposes of
862 -- inlining. All other CCCSs are mapped to currentCCS.
863 simplNote env (SCC cc) e cont
864 = simplExpr (setEnclosingCC env currentCCS) e `thenSmpl` \ e' ->
865 rebuild env (mkSCC cc e') cont
867 simplNote env InlineCall e cont
868 = simplExprF env e (InlinePlease cont)
870 -- See notes with SimplMonad.inlineMode
871 simplNote env InlineMe e cont
872 | contIsRhsOrArg cont -- Totally boring continuation; see notes above
873 = -- Don't inline inside an INLINE expression
874 simplExpr (setMode inlineMode env ) e `thenSmpl` \ e' ->
875 rebuild env (mkInlineMe e') cont
877 | otherwise -- Dissolve the InlineMe note if there's
878 -- an interesting context of any kind to combine with
879 -- (even a type application -- anything except Stop)
880 = simplExprF env e cont
882 simplNote env (CoreNote s) e cont
883 = simplExpr env e `thenSmpl` \ e' ->
884 rebuild env (Note (CoreNote s) e') cont
888 %************************************************************************
890 \subsection{Dealing with calls}
892 %************************************************************************
895 simplVar env var cont
896 = case substId env var of
897 DoneEx e -> simplExprF (zapSubstEnv env) e cont
898 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
899 DoneId var1 occ -> completeCall (zapSubstEnv env) var1 occ cont
900 -- Note [zapSubstEnv]
901 -- The template is already simplified, so don't re-substitute.
902 -- This is VITAL. Consider
904 -- let y = \z -> ...x... in
906 -- We'll clone the inner \x, adding x->x' in the id_subst
907 -- Then when we inline y, we must *not* replace x by x' in
908 -- the inlined copy!!
910 ---------------------------------------------------------
911 -- Dealing with a call site
913 completeCall env var occ_info cont
914 = -- Simplify the arguments
915 getDOptsSmpl `thenSmpl` \ dflags ->
917 chkr = getSwitchChecker env
918 (args, call_cont, inline_call) = getContArgs chkr var cont
921 simplifyArgs env fn_ty args (contResultType call_cont) $ \ env args ->
923 -- Next, look for rules or specialisations that match
925 -- It's important to simplify the args first, because the rule-matcher
926 -- doesn't do substitution as it goes. We don't want to use subst_args
927 -- (defined in the 'where') because that throws away useful occurrence info,
928 -- and perhaps-very-important specialisations.
930 -- Some functions have specialisations *and* are strict; in this case,
931 -- we don't want to inline the wrapper of the non-specialised thing; better
932 -- to call the specialised thing instead.
933 -- We used to use the black-listing mechanism to ensure that inlining of
934 -- the wrapper didn't occur for things that have specialisations till a
935 -- later phase, so but now we just try RULES first
937 -- You might think that we shouldn't apply rules for a loop breaker:
938 -- doing so might give rise to an infinite loop, because a RULE is
939 -- rather like an extra equation for the function:
940 -- RULE: f (g x) y = x+y
943 -- But it's too drastic to disable rules for loop breakers.
944 -- Even the foldr/build rule would be disabled, because foldr
945 -- is recursive, and hence a loop breaker:
946 -- foldr k z (build g) = g k z
947 -- So it's up to the programmer: rules can cause divergence
950 in_scope = getInScope env
952 maybe_rule = case activeRule env of
953 Nothing -> Nothing -- No rules apply
954 Just act_fn -> lookupRule act_fn in_scope rules var args
957 Just (rule_name, rule_rhs) ->
958 tick (RuleFired rule_name) `thenSmpl_`
959 (if dopt Opt_D_dump_inlinings dflags then
960 pprTrace "Rule fired" (vcat [
961 text "Rule:" <+> ftext rule_name,
962 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
963 text "After: " <+> pprCoreExpr rule_rhs,
964 text "Cont: " <+> ppr call_cont])
967 simplExprF env rule_rhs call_cont ;
969 Nothing -> -- No rules
971 -- Next, look for an inlining
973 arg_infos = [ interestingArg arg | arg <- args, isValArg arg]
975 interesting_cont = interestingCallContext (notNull args)
979 active_inline = activeInline env var occ_info
980 maybe_inline = callSiteInline dflags active_inline inline_call occ_info
981 var arg_infos interesting_cont
983 case maybe_inline of {
984 Just unfolding -- There is an inlining!
985 -> tick (UnfoldingDone var) `thenSmpl_`
986 (if dopt Opt_D_dump_inlinings dflags then
987 pprTrace "Inlining done" (vcat [
988 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
989 text "Inlined fn: " <+> ppr unfolding,
990 text "Cont: " <+> ppr call_cont])
993 makeThatCall env var unfolding args call_cont
996 Nothing -> -- No inlining!
999 rebuild env (mkApps (Var var) args) call_cont
1002 makeThatCall :: SimplEnv
1004 -> InExpr -- Inlined function rhs
1005 -> [OutExpr] -- Arguments, already simplified
1006 -> SimplCont -- After the call
1007 -> SimplM FloatsWithExpr
1008 -- Similar to simplLam, but this time
1009 -- the arguments are already simplified
1010 makeThatCall orig_env var fun@(Lam _ _) args cont
1011 = go orig_env fun args
1013 zap_it = mkLamBndrZapper fun (length args)
1015 -- Type-beta reduction
1016 go env (Lam bndr body) (Type ty_arg : args)
1017 = ASSERT( isTyVar bndr )
1018 tick (BetaReduction bndr) `thenSmpl_`
1019 go (extendTvSubst env bndr ty_arg) body args
1021 -- Ordinary beta reduction
1022 go env (Lam bndr body) (arg : args)
1023 = tick (BetaReduction bndr) `thenSmpl_`
1024 simplNonRecX env (zap_it bndr) arg $ \ env ->
1027 -- Not enough args, so there are real lambdas left to put in the result
1029 = simplExprF env fun (pushContArgs orig_env args cont)
1030 -- NB: orig_env; the correct environment to capture with
1031 -- the arguments.... env has been augmented with substitutions
1032 -- from the beta reductions.
1034 makeThatCall env var fun args cont
1035 = simplExprF env fun (pushContArgs env args cont)
1039 %************************************************************************
1041 \subsection{Arguments}
1043 %************************************************************************
1046 ---------------------------------------------------------
1047 -- Simplifying the arguments of a call
1049 simplifyArgs :: SimplEnv
1050 -> OutType -- Type of the function
1051 -> [(InExpr, SimplEnv, Bool)] -- Details of the arguments
1052 -> OutType -- Type of the continuation
1053 -> (SimplEnv -> [OutExpr] -> SimplM FloatsWithExpr)
1054 -> SimplM FloatsWithExpr
1056 -- [CPS-like because of strict arguments]
1058 -- Simplify the arguments to a call.
1059 -- This part of the simplifier may break the no-shadowing invariant
1061 -- f (...(\a -> e)...) (case y of (a,b) -> e')
1062 -- where f is strict in its second arg
1063 -- If we simplify the innermost one first we get (...(\a -> e)...)
1064 -- Simplifying the second arg makes us float the case out, so we end up with
1065 -- case y of (a,b) -> f (...(\a -> e)...) e'
1066 -- So the output does not have the no-shadowing invariant. However, there is
1067 -- no danger of getting name-capture, because when the first arg was simplified
1068 -- we used an in-scope set that at least mentioned all the variables free in its
1069 -- static environment, and that is enough.
1071 -- We can't just do innermost first, or we'd end up with a dual problem:
1072 -- case x of (a,b) -> f e (...(\a -> e')...)
1074 -- I spent hours trying to recover the no-shadowing invariant, but I just could
1075 -- not think of an elegant way to do it. The simplifier is already knee-deep in
1076 -- continuations. We have to keep the right in-scope set around; AND we have
1077 -- to get the effect that finding (error "foo") in a strict arg position will
1078 -- discard the entire application and replace it with (error "foo"). Getting
1079 -- all this at once is TOO HARD!
1081 simplifyArgs env fn_ty args cont_ty thing_inside
1082 = go env fn_ty args thing_inside
1084 go env fn_ty [] thing_inside = thing_inside env []
1085 go env fn_ty (arg:args) thing_inside = simplifyArg env fn_ty arg cont_ty $ \ env arg' ->
1086 go env (applyTypeToArg fn_ty arg') args $ \ env args' ->
1087 thing_inside env (arg':args')
1089 simplifyArg env fn_ty (Type ty_arg, se, _) cont_ty thing_inside
1090 = simplType (setInScope se env) ty_arg `thenSmpl` \ new_ty_arg ->
1091 thing_inside env (Type new_ty_arg)
1093 simplifyArg env fn_ty (val_arg, arg_se, is_strict) cont_ty thing_inside
1095 = simplStrictArg AnArg env val_arg arg_se arg_ty cont_ty thing_inside
1097 | otherwise -- Lazy argument
1098 -- DO NOT float anything outside, hence simplExprC
1099 -- There is no benefit (unlike in a let-binding), and we'd
1100 -- have to be very careful about bogus strictness through
1101 -- floating a demanded let.
1102 = simplExprC (setInScope arg_se env) val_arg
1103 (mkBoringStop arg_ty) `thenSmpl` \ arg1 ->
1104 thing_inside env arg1
1106 arg_ty = funArgTy fn_ty
1109 simplStrictArg :: LetRhsFlag
1110 -> SimplEnv -- The env of the call
1111 -> InExpr -> SimplEnv -- The arg plus its env
1112 -> OutType -- arg_ty: type of the argument
1113 -> OutType -- cont_ty: Type of thing computed by the context
1114 -> (SimplEnv -> OutExpr -> SimplM FloatsWithExpr)
1115 -- Takes an expression of type rhs_ty,
1116 -- returns an expression of type cont_ty
1117 -- The env passed to this continuation is the
1118 -- env of the call, plus any new in-scope variables
1119 -> SimplM FloatsWithExpr -- An expression of type cont_ty
1121 simplStrictArg is_rhs call_env arg arg_env arg_ty cont_ty thing_inside
1122 = simplExprF (setInScope arg_env call_env) arg
1123 (ArgOf is_rhs arg_ty cont_ty (\ new_env -> thing_inside (setInScope call_env new_env)))
1124 -- Notice the way we use arg_env (augmented with in-scope vars from call_env)
1125 -- to simplify the argument
1126 -- and call-env (augmented with in-scope vars from the arg) to pass to the continuation
1130 %************************************************************************
1132 \subsection{mkAtomicArgs}
1134 %************************************************************************
1136 mkAtomicArgs takes a putative RHS, checks whether it's a PAP or
1137 constructor application and, if so, converts it to ANF, so that the
1138 resulting thing can be inlined more easily. Thus
1145 There are three sorts of binding context, specified by the two
1151 N N Top-level or recursive Only bind args of lifted type
1153 N Y Non-top-level and non-recursive, Bind args of lifted type, or
1154 but lazy unlifted-and-ok-for-speculation
1156 Y Y Non-top-level, non-recursive, Bind all args
1157 and strict (demanded)
1164 there is no point in transforming to
1166 x = case (y div# z) of r -> MkC r
1168 because the (y div# z) can't float out of the let. But if it was
1169 a *strict* let, then it would be a good thing to do. Hence the
1170 context information.
1173 mkAtomicArgs :: Bool -- A strict binding
1174 -> Bool -- OK to float unlifted args
1176 -> SimplM (OrdList (OutId,OutExpr), -- The floats (unusually) may include
1177 OutExpr) -- things that need case-binding,
1178 -- if the strict-binding flag is on
1180 mkAtomicArgs is_strict ok_float_unlifted rhs
1181 | (Var fun, args) <- collectArgs rhs, -- It's an application
1182 isDataConWorkId fun || valArgCount args < idArity fun -- And it's a constructor or PAP
1183 = go fun nilOL [] args -- Have a go
1185 | otherwise = bale_out -- Give up
1188 bale_out = returnSmpl (nilOL, rhs)
1190 go fun binds rev_args []
1191 = returnSmpl (binds, mkApps (Var fun) (reverse rev_args))
1193 go fun binds rev_args (arg : args)
1194 | exprIsTrivial arg -- Easy case
1195 = go fun binds (arg:rev_args) args
1197 | not can_float_arg -- Can't make this arg atomic
1198 = bale_out -- ... so give up
1200 | otherwise -- Don't forget to do it recursively
1201 -- E.g. x = a:b:c:[]
1202 = mkAtomicArgs is_strict ok_float_unlifted arg `thenSmpl` \ (arg_binds, arg') ->
1203 newId FSLIT("a") arg_ty `thenSmpl` \ arg_id ->
1204 go fun ((arg_binds `snocOL` (arg_id,arg')) `appOL` binds)
1205 (Var arg_id : rev_args) args
1207 arg_ty = exprType arg
1208 can_float_arg = is_strict
1209 || not (isUnLiftedType arg_ty)
1210 || (ok_float_unlifted && exprOkForSpeculation arg)
1213 addAtomicBinds :: SimplEnv -> [(OutId,OutExpr)]
1214 -> (SimplEnv -> SimplM (FloatsWith a))
1215 -> SimplM (FloatsWith a)
1216 addAtomicBinds env [] thing_inside = thing_inside env
1217 addAtomicBinds env ((v,r):bs) thing_inside = addAuxiliaryBind env (NonRec v r) $ \ env ->
1218 addAtomicBinds env bs thing_inside
1220 addAtomicBindsE :: SimplEnv -> [(OutId,OutExpr)]
1221 -> (SimplEnv -> SimplM FloatsWithExpr)
1222 -> SimplM FloatsWithExpr
1223 -- Same again, but this time we're in an expression context,
1224 -- and may need to do some case bindings
1226 addAtomicBindsE env [] thing_inside
1228 addAtomicBindsE env ((v,r):bs) thing_inside
1229 | needsCaseBinding (idType v) r
1230 = addAtomicBindsE (addNewInScopeIds env [v]) bs thing_inside `thenSmpl` \ (floats, expr) ->
1231 WARN( exprIsTrivial expr, ppr v <+> pprCoreExpr expr )
1232 (let body = wrapFloats floats expr in
1233 returnSmpl (emptyFloats env, Case r v (exprType body) [(DEFAULT,[],body)]))
1236 = addAuxiliaryBind env (NonRec v r) $ \ env ->
1237 addAtomicBindsE env bs thing_inside
1241 %************************************************************************
1243 \subsection{The main rebuilder}
1245 %************************************************************************
1248 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM FloatsWithExpr
1250 rebuild env expr (Stop _ _ _) = rebuildDone env expr
1251 rebuild env expr (ArgOf _ _ _ cont_fn) = cont_fn env expr
1252 rebuild env expr (CoerceIt to_ty cont) = rebuild env (mkCoerce to_ty expr) cont
1253 rebuild env expr (InlinePlease cont) = rebuild env (Note InlineCall expr) cont
1254 rebuild env expr (Select _ bndr alts se cont) = rebuildCase (setInScope se env) expr bndr alts cont
1255 rebuild env expr (ApplyTo _ arg se cont) = rebuildApp (setInScope se env) expr arg cont
1257 rebuildApp env fun arg cont
1258 = simplExpr env arg `thenSmpl` \ arg' ->
1259 rebuild env (App fun arg') cont
1261 rebuildDone env expr = returnSmpl (emptyFloats env, expr)
1265 %************************************************************************
1267 \subsection{Functions dealing with a case}
1269 %************************************************************************
1271 Blob of helper functions for the "case-of-something-else" situation.
1274 ---------------------------------------------------------
1275 -- Eliminate the case if possible
1277 rebuildCase :: SimplEnv
1278 -> OutExpr -- Scrutinee
1279 -> InId -- Case binder
1280 -> [InAlt] -- Alternatives (inceasing order)
1282 -> SimplM FloatsWithExpr
1284 rebuildCase env scrut case_bndr alts cont
1285 | Just (con,args) <- exprIsConApp_maybe scrut
1286 -- Works when the scrutinee is a variable with a known unfolding
1287 -- as well as when it's an explicit constructor application
1288 = knownCon env (DataAlt con) args case_bndr alts cont
1290 | Lit lit <- scrut -- No need for same treatment as constructors
1291 -- because literals are inlined more vigorously
1292 = knownCon env (LitAlt lit) [] case_bndr alts cont
1295 = -- Prepare the alternatives.
1296 prepareAlts scrut case_bndr alts `thenSmpl` \ (better_alts, handled_cons) ->
1298 -- Prepare the continuation;
1299 -- The new subst_env is in place
1300 prepareCaseCont env better_alts cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1301 addFloats env floats $ \ env ->
1304 -- The case expression is annotated with the result type of the continuation
1305 -- This may differ from the type originally on the case. For example
1306 -- case(T) (case(Int#) a of { True -> 1#; False -> 0# }) of
1309 -- let j a# = <blob>
1310 -- in case(T) a of { True -> j 1#; False -> j 0# }
1311 -- Note that the case that scrutinises a now returns a T not an Int#
1312 res_ty' = contResultType dup_cont
1315 -- Deal with case binder
1316 simplCaseBinder env scrut case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1318 -- Deal with the case alternatives
1319 simplAlts alt_env handled_cons
1320 case_bndr' better_alts dup_cont `thenSmpl` \ alts' ->
1322 -- Put the case back together
1323 mkCase scrut case_bndr' res_ty' alts' `thenSmpl` \ case_expr ->
1325 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1326 -- The case binder *not* scope over the whole returned case-expression
1327 rebuild env case_expr nondup_cont
1330 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1331 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1332 way, there's a chance that v will now only be used once, and hence
1337 There is a time we *don't* want to do that, namely when
1338 -fno-case-of-case is on. This happens in the first simplifier pass,
1339 and enhances full laziness. Here's the bad case:
1340 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1341 If we eliminate the inner case, we trap it inside the I# v -> arm,
1342 which might prevent some full laziness happening. I've seen this
1343 in action in spectral/cichelli/Prog.hs:
1344 [(m,n) | m <- [1..max], n <- [1..max]]
1345 Hence the check for NoCaseOfCase.
1349 There is another situation when we don't want to do it. If we have
1351 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1352 ...other cases .... }
1354 We'll perform the binder-swap for the outer case, giving
1356 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1357 ...other cases .... }
1359 But there is no point in doing it for the inner case, because w1 can't
1360 be inlined anyway. Furthermore, doing the case-swapping involves
1361 zapping w2's occurrence info (see paragraphs that follow), and that
1362 forces us to bind w2 when doing case merging. So we get
1364 case x of w1 { A -> let w2 = w1 in e1
1365 B -> let w2 = w1 in e2
1366 ...other cases .... }
1368 This is plain silly in the common case where w2 is dead.
1370 Even so, I can't see a good way to implement this idea. I tried
1371 not doing the binder-swap if the scrutinee was already evaluated
1372 but that failed big-time:
1376 case v of w { MkT x ->
1377 case x of x1 { I# y1 ->
1378 case x of x2 { I# y2 -> ...
1380 Notice that because MkT is strict, x is marked "evaluated". But to
1381 eliminate the last case, we must either make sure that x (as well as
1382 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1383 the binder-swap. So this whole note is a no-op.
1387 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1388 any occurrence info (eg IAmDead) in the case binder, because the
1389 case-binder now effectively occurs whenever v does. AND we have to do
1390 the same for the pattern-bound variables! Example:
1392 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1394 Here, b and p are dead. But when we move the argment inside the first
1395 case RHS, and eliminate the second case, we get
1397 case x of { (a,b) -> a b }
1399 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1402 Indeed, this can happen anytime the case binder isn't dead:
1403 case <any> of x { (a,b) ->
1404 case x of { (p,q) -> p } }
1405 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1406 The point is that we bring into the envt a binding
1408 after the outer case, and that makes (a,b) alive. At least we do unless
1409 the case binder is guaranteed dead.
1412 simplCaseBinder env (Var v) case_bndr
1413 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
1415 -- Failed try [see Note 2 above]
1416 -- not (isEvaldUnfolding (idUnfolding v))
1418 = simplBinder env (zap case_bndr) `thenSmpl` \ (env, case_bndr') ->
1419 returnSmpl (modifyInScope env v case_bndr', case_bndr')
1420 -- We could extend the substitution instead, but it would be
1421 -- a hack because then the substitution wouldn't be idempotent
1422 -- any more (v is an OutId). And this does just as well.
1424 zap b = b `setIdOccInfo` NoOccInfo
1426 simplCaseBinder env other_scrut case_bndr
1427 = simplBinder env case_bndr `thenSmpl` \ (env, case_bndr') ->
1428 returnSmpl (env, case_bndr')
1434 simplAlts :: SimplEnv
1435 -> [AltCon] -- Alternatives the scrutinee can't be
1436 -- in the default case
1437 -> OutId -- Case binder
1438 -> [InAlt] -> SimplCont
1439 -> SimplM [OutAlt] -- Includes the continuation
1441 simplAlts env handled_cons case_bndr' alts cont'
1442 = do { mb_alts <- mapSmpl simpl_alt alts
1443 ; return [alt' | Just (_, alt') <- mb_alts] }
1444 -- Filter out the alternatives that are inaccessible
1446 simpl_alt alt = simplAlt env handled_cons case_bndr' alt cont'
1448 simplAlt :: SimplEnv -> [AltCon] -> OutId -> InAlt -> SimplCont
1449 -> SimplM (Maybe (TvSubstEnv, OutAlt))
1450 -- Simplify an alternative, returning the type refinement for the
1451 -- alternative, if the alternative does any refinement at all
1452 -- Nothing => the alternative is inaccessible
1454 simplAlt env handled_cons case_bndr' (DEFAULT, bndrs, rhs) cont'
1455 = ASSERT( null bndrs )
1456 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1457 returnSmpl (Just (emptyVarEnv, (DEFAULT, [], rhs')))
1459 env' = mk_rhs_env env case_bndr' (mkOtherCon handled_cons)
1460 -- Record the constructors that the case-binder *can't* be.
1462 simplAlt env handled_cons case_bndr' (LitAlt lit, bndrs, rhs) cont'
1463 = ASSERT( null bndrs )
1464 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1465 returnSmpl (Just (emptyVarEnv, (LitAlt lit, [], rhs')))
1467 env' = mk_rhs_env env case_bndr' (mkUnfolding False (Lit lit))
1469 simplAlt env handled_cons case_bndr' (DataAlt con, vs, rhs) cont'
1470 | isVanillaDataCon con
1471 = -- Deal with the pattern-bound variables
1472 -- Mark the ones that are in ! positions in the data constructor
1473 -- as certainly-evaluated.
1474 -- NB: it happens that simplBinders does *not* erase the OtherCon
1475 -- form of unfolding, so it's ok to add this info before
1476 -- doing simplBinders
1477 simplBinders env (add_evals con vs) `thenSmpl` \ (env, vs') ->
1479 -- Bind the case-binder to (con args)
1480 let unf = mkUnfolding False (mkConApp con con_args)
1481 inst_tys' = tyConAppArgs (idType case_bndr')
1482 con_args = map Type inst_tys' ++ map varToCoreExpr vs'
1483 env' = mk_rhs_env env case_bndr' unf
1485 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1486 returnSmpl (Just (emptyVarEnv, (DataAlt con, vs', rhs')))
1488 | otherwise -- GADT case
1490 (tvs,ids) = span isTyVar vs
1492 simplBinders env tvs `thenSmpl` \ (env1, tvs') ->
1493 case coreRefineTys con tvs' (idType case_bndr') of {
1494 Nothing -- Inaccessible
1495 | opt_PprStyle_Debug -- Hack: if debugging is on, generate an error case
1497 -> let rhs' = mkApps (Var eRROR_ID)
1498 [Type (substTy env (exprType rhs)),
1499 Lit (mkStringLit "Impossible alternative (GADT)")]
1501 simplBinders env1 ids `thenSmpl` \ (env2, ids') ->
1502 returnSmpl (Just (emptyVarEnv, (DataAlt con, tvs' ++ ids', rhs')))
1504 | otherwise -- Filter out the inaccessible branch
1507 Just refine@(tv_subst_env, _) -> -- The normal case
1510 env2 = refineSimplEnv env1 refine
1511 -- Simplify the Ids in the refined environment, so their types
1512 -- reflect the refinement. Usually this doesn't matter, but it helps
1513 -- in mkDupableAlt, when we want to float a lambda that uses these binders
1514 -- Furthermore, it means the binders contain maximal type information
1516 simplBinders env2 (add_evals con ids) `thenSmpl` \ (env3, ids') ->
1517 let unf = mkUnfolding False con_app
1518 con_app = mkConApp con con_args
1519 con_args = map varToCoreExpr vs' -- NB: no inst_tys'
1520 env_w_unf = mk_rhs_env env3 case_bndr' unf
1523 simplExprC env_w_unf rhs cont' `thenSmpl` \ rhs' ->
1524 returnSmpl (Just (tv_subst_env, (DataAlt con, vs', rhs'))) }
1527 -- add_evals records the evaluated-ness of the bound variables of
1528 -- a case pattern. This is *important*. Consider
1529 -- data T = T !Int !Int
1531 -- case x of { T a b -> T (a+1) b }
1533 -- We really must record that b is already evaluated so that we don't
1534 -- go and re-evaluate it when constructing the result.
1535 add_evals dc vs = cat_evals dc vs (dataConRepStrictness dc)
1537 cat_evals dc vs strs
1541 go (v:vs) strs | isTyVar v = v : go vs strs
1542 go (v:vs) (str:strs)
1543 | isMarkedStrict str = evald_v : go vs strs
1544 | otherwise = zapped_v : go vs strs
1546 zapped_v = zap_occ_info v
1547 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1548 go _ _ = pprPanic "cat_evals" (ppr dc $$ ppr vs $$ ppr strs)
1550 -- If the case binder is alive, then we add the unfolding
1552 -- to the envt; so vs are now very much alive
1553 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1554 | otherwise = \id -> id `setIdOccInfo` NoOccInfo
1556 mk_rhs_env env case_bndr' case_bndr_unf
1557 = modifyInScope env case_bndr' (case_bndr' `setIdUnfolding` case_bndr_unf)
1561 %************************************************************************
1563 \subsection{Known constructor}
1565 %************************************************************************
1567 We are a bit careful with occurrence info. Here's an example
1569 (\x* -> case x of (a*, b) -> f a) (h v, e)
1571 where the * means "occurs once". This effectively becomes
1572 case (h v, e) of (a*, b) -> f a)
1574 let a* = h v; b = e in f a
1578 All this should happen in one sweep.
1581 knownCon :: SimplEnv -> AltCon -> [OutExpr]
1582 -> InId -> [InAlt] -> SimplCont
1583 -> SimplM FloatsWithExpr
1585 knownCon env con args bndr alts cont
1586 = tick (KnownBranch bndr) `thenSmpl_`
1587 case findAlt con alts of
1588 (DEFAULT, bs, rhs) -> ASSERT( null bs )
1589 simplNonRecX env bndr scrut $ \ env ->
1590 -- This might give rise to a binding with non-atomic args
1591 -- like x = Node (f x) (g x)
1592 -- but no harm will be done
1593 simplExprF env rhs cont
1596 LitAlt lit -> Lit lit
1597 DataAlt dc -> mkConApp dc args
1599 (LitAlt lit, bs, rhs) -> ASSERT( null bs )
1600 simplNonRecX env bndr (Lit lit) $ \ env ->
1601 simplExprF env rhs cont
1603 (DataAlt dc, bs, rhs)
1604 -> ASSERT( n_drop_tys + length bs == length args )
1605 bind_args env bs (drop n_drop_tys args) $ \ env ->
1607 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1608 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1609 -- args are aready OutExprs, but bs are InIds
1611 simplNonRecX env bndr con_app $ \ env ->
1612 simplExprF env rhs cont
1614 n_drop_tys | isVanillaDataCon dc = tyConArity (dataConTyCon dc)
1616 -- Vanilla data constructors lack type arguments in the pattern
1619 bind_args env [] _ thing_inside = thing_inside env
1621 bind_args env (b:bs) (Type ty : args) thing_inside
1622 = ASSERT( isTyVar b )
1623 bind_args (extendTvSubst env b ty) bs args thing_inside
1625 bind_args env (b:bs) (arg : args) thing_inside
1627 simplNonRecX env b arg $ \ env ->
1628 bind_args env bs args thing_inside
1632 %************************************************************************
1634 \subsection{Duplicating continuations}
1636 %************************************************************************
1639 prepareCaseCont :: SimplEnv
1640 -> [InAlt] -> SimplCont
1641 -> SimplM (FloatsWith (SimplCont,SimplCont))
1642 -- Return a duplicatable continuation, a non-duplicable part
1643 -- plus some extra bindings
1645 -- No need to make it duplicatable if there's only one alternative
1646 prepareCaseCont env [alt] cont = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1647 prepareCaseCont env alts cont = mkDupableCont env cont
1651 mkDupableCont :: SimplEnv -> SimplCont
1652 -> SimplM (FloatsWith (SimplCont, SimplCont))
1654 mkDupableCont env cont
1655 | contIsDupable cont
1656 = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1658 mkDupableCont env (CoerceIt ty cont)
1659 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1660 returnSmpl (floats, (CoerceIt ty dup_cont, nondup_cont))
1662 mkDupableCont env (InlinePlease cont)
1663 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1664 returnSmpl (floats, (InlinePlease dup_cont, nondup_cont))
1666 mkDupableCont env cont@(ArgOf _ arg_ty _ _)
1667 = returnSmpl (emptyFloats env, (mkBoringStop arg_ty, cont))
1668 -- Do *not* duplicate an ArgOf continuation
1669 -- Because ArgOf continuations are opaque, we gain nothing by
1670 -- propagating them into the expressions, and we do lose a lot.
1671 -- Here's an example:
1672 -- && (case x of { T -> F; F -> T }) E
1673 -- Now, && is strict so we end up simplifying the case with
1674 -- an ArgOf continuation. If we let-bind it, we get
1676 -- let $j = \v -> && v E
1677 -- in simplExpr (case x of { T -> F; F -> T })
1678 -- (ArgOf (\r -> $j r)
1679 -- And after simplifying more we get
1681 -- let $j = \v -> && v E
1682 -- in case of { T -> $j F; F -> $j T }
1683 -- Which is a Very Bad Thing
1685 -- The desire not to duplicate is the entire reason that
1686 -- mkDupableCont returns a pair of continuations.
1688 -- The original plan had:
1689 -- e.g. (...strict-fn...) [...hole...]
1691 -- let $j = \a -> ...strict-fn...
1692 -- in $j [...hole...]
1694 mkDupableCont env (ApplyTo _ arg se cont)
1695 = -- e.g. [...hole...] (...arg...)
1697 -- let a = ...arg...
1698 -- in [...hole...] a
1699 simplExpr (setInScope se env) arg `thenSmpl` \ arg' ->
1701 mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1702 addFloats env floats $ \ env ->
1704 if exprIsDupable arg' then
1705 returnSmpl (emptyFloats env, (ApplyTo OkToDup arg' (zapSubstEnv se) dup_cont, nondup_cont))
1707 newId FSLIT("a") (exprType arg') `thenSmpl` \ arg_id ->
1709 tick (CaseOfCase arg_id) `thenSmpl_`
1710 -- Want to tick here so that we go round again,
1711 -- and maybe copy or inline the code.
1712 -- Not strictly CaseOfCase, but never mind
1714 returnSmpl (unitFloat env arg_id arg',
1715 (ApplyTo OkToDup (Var arg_id) (zapSubstEnv se) dup_cont,
1717 -- But what if the arg should be case-bound?
1718 -- This has been this way for a long time, so I'll leave it,
1719 -- but I can't convince myself that it's right.
1721 mkDupableCont env (Select _ case_bndr alts se cont)
1722 = -- e.g. (case [...hole...] of { pi -> ei })
1724 -- let ji = \xij -> ei
1725 -- in case [...hole...] of { pi -> ji xij }
1726 tick (CaseOfCase case_bndr) `thenSmpl_`
1728 alt_env = setInScope se env
1730 prepareCaseCont alt_env alts cont `thenSmpl` \ (floats1, (dup_cont, nondup_cont)) ->
1731 addFloats alt_env floats1 $ \ alt_env ->
1733 simplBinder alt_env case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1734 -- NB: simplBinder does not zap deadness occ-info, so
1735 -- a dead case_bndr' will still advertise its deadness
1736 -- This is really important because in
1737 -- case e of b { (# a,b #) -> ... }
1738 -- b is always dead, and indeed we are not allowed to bind b to (# a,b #),
1739 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1740 -- In the new alts we build, we have the new case binder, so it must retain
1743 mkDupableAlts alt_env case_bndr' alts dup_cont `thenSmpl` \ (floats2, alts') ->
1744 addFloats alt_env floats2 $ \ alt_env ->
1745 returnSmpl (emptyFloats alt_env,
1746 (Select OkToDup case_bndr' alts' (zapSubstEnv se)
1747 (mkBoringStop (contResultType dup_cont)),
1750 mkDupableAlts :: SimplEnv -> OutId -> [InAlt] -> SimplCont
1751 -> SimplM (FloatsWith [InAlt])
1752 -- Absorbs the continuation into the new alternatives
1754 mkDupableAlts env case_bndr' alts dupable_cont
1757 go env [] = returnSmpl (emptyFloats env, [])
1759 = do { (floats1, mb_alt') <- mkDupableAlt env case_bndr' dupable_cont alt
1760 ; addFloats env floats1 $ \ env -> do
1761 { (floats2, alts') <- go env alts
1762 ; returnSmpl (floats2, case mb_alt' of
1763 Just alt' -> alt' : alts'
1767 mkDupableAlt env case_bndr' cont alt
1768 = simplAlt env [] case_bndr' alt cont `thenSmpl` \ mb_stuff ->
1770 Nothing -> returnSmpl (emptyFloats env, Nothing) ;
1772 Just (reft, (con, bndrs', rhs')) ->
1773 -- Safe to say that there are no handled-cons for the DEFAULT case
1775 if exprIsDupable rhs' then
1776 returnSmpl (emptyFloats env, Just (con, bndrs', rhs'))
1777 -- It is worth checking for a small RHS because otherwise we
1778 -- get extra let bindings that may cause an extra iteration of the simplifier to
1779 -- inline back in place. Quite often the rhs is just a variable or constructor.
1780 -- The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1781 -- iterations because the version with the let bindings looked big, and so wasn't
1782 -- inlined, but after the join points had been inlined it looked smaller, and so
1785 -- NB: we have to check the size of rhs', not rhs.
1786 -- Duplicating a small InAlt might invalidate occurrence information
1787 -- However, if it *is* dupable, we return the *un* simplified alternative,
1788 -- because otherwise we'd need to pair it up with an empty subst-env....
1789 -- but we only have one env shared between all the alts.
1790 -- (Remember we must zap the subst-env before re-simplifying something).
1791 -- Rather than do this we simply agree to re-simplify the original (small) thing later.
1795 rhs_ty' = exprType rhs'
1796 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1798 | isTyVar bndr = not (bndr `elemVarEnv` reft)
1799 -- Don't abstract over tyvar binders which are refined away
1800 -- See Note [Refinement] below
1801 | otherwise = not (isDeadBinder bndr)
1802 -- The deadness info on the new Ids is preserved by simplBinders
1804 -- If we try to lift a primitive-typed something out
1805 -- for let-binding-purposes, we will *caseify* it (!),
1806 -- with potentially-disastrous strictness results. So
1807 -- instead we turn it into a function: \v -> e
1808 -- where v::State# RealWorld#. The value passed to this function
1809 -- is realworld#, which generates (almost) no code.
1811 -- There's a slight infelicity here: we pass the overall
1812 -- case_bndr to all the join points if it's used in *any* RHS,
1813 -- because we don't know its usage in each RHS separately
1815 -- We used to say "&& isUnLiftedType rhs_ty'" here, but now
1816 -- we make the join point into a function whenever used_bndrs'
1817 -- is empty. This makes the join-point more CPR friendly.
1818 -- Consider: let j = if .. then I# 3 else I# 4
1819 -- in case .. of { A -> j; B -> j; C -> ... }
1821 -- Now CPR doesn't w/w j because it's a thunk, so
1822 -- that means that the enclosing function can't w/w either,
1823 -- which is a lose. Here's the example that happened in practice:
1824 -- kgmod :: Int -> Int -> Int
1825 -- kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1829 -- I have seen a case alternative like this:
1830 -- True -> \v -> ...
1831 -- It's a bit silly to add the realWorld dummy arg in this case, making
1834 -- (the \v alone is enough to make CPR happy) but I think it's rare
1836 ( if not (any isId used_bndrs')
1837 then newId FSLIT("w") realWorldStatePrimTy `thenSmpl` \ rw_id ->
1838 returnSmpl ([rw_id], [Var realWorldPrimId])
1840 returnSmpl (used_bndrs', map varToCoreExpr used_bndrs')
1841 ) `thenSmpl` \ (final_bndrs', final_args) ->
1843 -- See comment about "$j" name above
1844 newId FSLIT("$j") (mkPiTypes final_bndrs' rhs_ty') `thenSmpl` \ join_bndr ->
1845 -- Notice the funky mkPiTypes. If the contructor has existentials
1846 -- it's possible that the join point will be abstracted over
1847 -- type varaibles as well as term variables.
1848 -- Example: Suppose we have
1849 -- data T = forall t. C [t]
1851 -- case (case e of ...) of
1852 -- C t xs::[t] -> rhs
1853 -- We get the join point
1854 -- let j :: forall t. [t] -> ...
1855 -- j = /\t \xs::[t] -> rhs
1857 -- case (case e of ...) of
1858 -- C t xs::[t] -> j t xs
1860 -- We make the lambdas into one-shot-lambdas. The
1861 -- join point is sure to be applied at most once, and doing so
1862 -- prevents the body of the join point being floated out by
1863 -- the full laziness pass
1864 really_final_bndrs = map one_shot final_bndrs'
1865 one_shot v | isId v = setOneShotLambda v
1867 join_rhs = mkLams really_final_bndrs rhs'
1868 join_call = mkApps (Var join_bndr) final_args
1870 returnSmpl (unitFloat env join_bndr join_rhs, Just (con, bndrs', join_call)) }
1877 MkT :: a -> b -> T a
1881 MkT a' b (p::a') (q::b) -> [p,w]
1883 The danger is that we'll make a join point
1887 and that's ill-typed, because (p::a') but (w::a).
1889 Solution so far: don't abstract over a', because the type refinement
1890 maps [a' -> a] . Ultimately that won't work when real refinement goes on.
1892 Then we must abstract over any refined free variables. Hmm. Maybe we
1893 could just abstract over *all* free variables, thereby lambda-lifting
1894 the join point? We should try this.