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 CmdLineOpts ( dopt, DynFlag(Opt_D_dump_inlinings),
15 import SimplUtils ( mkCase, mkLam, newId, prepareAlts,
16 simplBinder, simplBinders, simplLamBndrs, simplRecBndrs, simplLetBndr,
17 SimplCont(..), DupFlag(..), LetRhsFlag(..),
18 mkStop, mkBoringStop, pushContArgs,
19 contResultType, countArgs, contIsDupable, contIsRhsOrArg,
20 getContArgs, interestingCallContext, interestingArg, isStrictType
22 import Var ( mustHaveLocalBinding )
24 import Id ( Id, idType, idInfo, idArity, isDataConWorkId,
25 setIdUnfolding, isDeadBinder,
26 idNewDemandInfo, setIdInfo,
27 idSpecialisation, setIdSpecialisation,
28 setIdOccInfo, zapLamIdInfo, setOneShotLambda,
30 import OccName ( encodeFS )
31 import IdInfo ( OccInfo(..), isLoopBreaker,
36 import NewDemand ( isStrictDmd )
37 import DataCon ( dataConNumInstArgs, dataConRepStrictness )
39 import PprCore ( pprParendExpr, pprCoreExpr )
40 import CoreUnfold ( mkOtherCon, mkUnfolding, callSiteInline )
41 import CoreUtils ( exprIsDupable, exprIsTrivial, needsCaseBinding,
42 exprIsConApp_maybe, mkPiTypes, findAlt,
43 exprType, exprIsValue,
44 exprOkForSpeculation, exprArity,
45 mkCoerce, mkCoerce2, mkSCC, mkInlineMe, mkAltExpr, applyTypeToArg
47 import Rules ( lookupRule )
48 import BasicTypes ( isMarkedStrict )
49 import CostCentre ( currentCCS )
50 import Type ( isUnLiftedType, seqType, tyConAppArgs, funArgTy,
51 splitFunTy_maybe, splitFunTy, eqType
53 import Subst ( mkSubst, substTy, substExpr, substRules,
54 isInScope, lookupIdSubst, simplIdInfo
56 import TysPrim ( realWorldStatePrimTy )
57 import PrelInfo ( realWorldPrimId )
58 import BasicTypes ( TopLevelFlag(..), isTopLevel,
62 import Maybe ( Maybe )
64 import Util ( notNull )
68 The guts of the simplifier is in this module, but the driver loop for
69 the simplifier is in SimplCore.lhs.
72 -----------------------------------------
73 *** IMPORTANT NOTE ***
74 -----------------------------------------
75 The simplifier used to guarantee that the output had no shadowing, but
76 it does not do so any more. (Actually, it never did!) The reason is
77 documented with simplifyArgs.
80 -----------------------------------------
81 *** IMPORTANT NOTE ***
82 -----------------------------------------
83 Many parts of the simplifier return a bunch of "floats" as well as an
84 expression. This is wrapped as a datatype SimplUtils.FloatsWith.
86 All "floats" are let-binds, not case-binds, but some non-rec lets may
87 be unlifted (with RHS ok-for-speculation).
91 -----------------------------------------
92 ORGANISATION OF FUNCTIONS
93 -----------------------------------------
95 - simplify all top-level binders
96 - for NonRec, call simplRecOrTopPair
97 - for Rec, call simplRecBind
100 ------------------------------
101 simplExpr (applied lambda) ==> simplNonRecBind
102 simplExpr (Let (NonRec ...) ..) ==> simplNonRecBind
103 simplExpr (Let (Rec ...) ..) ==> simplify binders; simplRecBind
105 ------------------------------
106 simplRecBind [binders already simplfied]
107 - use simplRecOrTopPair on each pair in turn
109 simplRecOrTopPair [binder already simplified]
110 Used for: recursive bindings (top level and nested)
111 top-level non-recursive bindings
113 - check for PreInlineUnconditionally
117 Used for: non-top-level non-recursive bindings
118 beta reductions (which amount to the same thing)
119 Because it can deal with strict arts, it takes a
120 "thing-inside" and returns an expression
122 - check for PreInlineUnconditionally
123 - simplify binder, including its IdInfo
132 simplNonRecX: [given a *simplified* RHS, but an *unsimplified* binder]
133 Used for: binding case-binder and constr args in a known-constructor case
134 - check for PreInLineUnconditionally
138 ------------------------------
139 simplLazyBind: [binder already simplified, RHS not]
140 Used for: recursive bindings (top level and nested)
141 top-level non-recursive bindings
142 non-top-level, but *lazy* non-recursive bindings
143 [must not be strict or unboxed]
144 Returns floats + an augmented environment, not an expression
145 - substituteIdInfo and add result to in-scope
146 [so that rules are available in rec rhs]
149 - float if exposes constructor or PAP
153 completeNonRecX: [binder and rhs both simplified]
154 - if the the thing needs case binding (unlifted and not ok-for-spec)
160 completeLazyBind: [given a simplified RHS]
161 [used for both rec and non-rec bindings, top level and not]
162 - try PostInlineUnconditionally
163 - add unfolding [this is the only place we add an unfolding]
168 Right hand sides and arguments
169 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
170 In many ways we want to treat
171 (a) the right hand side of a let(rec), and
172 (b) a function argument
173 in the same way. But not always! In particular, we would
174 like to leave these arguments exactly as they are, so they
175 will match a RULE more easily.
180 It's harder to make the rule match if we ANF-ise the constructor,
181 or eta-expand the PAP:
183 f (let { a = g x; b = h x } in (a,b))
186 On the other hand if we see the let-defns
191 then we *do* want to ANF-ise and eta-expand, so that p and q
192 can be safely inlined.
194 Even floating lets out is a bit dubious. For let RHS's we float lets
195 out if that exposes a value, so that the value can be inlined more vigorously.
198 r = let x = e in (x,x)
200 Here, if we float the let out we'll expose a nice constructor. We did experiments
201 that showed this to be a generally good thing. But it was a bad thing to float
202 lets out unconditionally, because that meant they got allocated more often.
204 For function arguments, there's less reason to expose a constructor (it won't
205 get inlined). Just possibly it might make a rule match, but I'm pretty skeptical.
206 So for the moment we don't float lets out of function arguments either.
211 For eta expansion, we want to catch things like
213 case e of (a,b) -> \x -> case a of (p,q) -> \y -> r
215 If the \x was on the RHS of a let, we'd eta expand to bring the two
216 lambdas together. And in general that's a good thing to do. Perhaps
217 we should eta expand wherever we find a (value) lambda? Then the eta
218 expansion at a let RHS can concentrate solely on the PAP case.
221 %************************************************************************
223 \subsection{Bindings}
225 %************************************************************************
228 simplTopBinds :: SimplEnv -> [InBind] -> SimplM [OutBind]
230 simplTopBinds env binds
231 = -- Put all the top-level binders into scope at the start
232 -- so that if a transformation rule has unexpectedly brought
233 -- anything into scope, then we don't get a complaint about that.
234 -- It's rather as if the top-level binders were imported.
235 simplRecBndrs env (bindersOfBinds binds) `thenSmpl` \ (env, bndrs') ->
236 simpl_binds env binds bndrs' `thenSmpl` \ (floats, _) ->
237 freeTick SimplifierDone `thenSmpl_`
238 returnSmpl (floatBinds floats)
240 -- We need to track the zapped top-level binders, because
241 -- they should have their fragile IdInfo zapped (notably occurrence info)
242 -- That's why we run down binds and bndrs' simultaneously.
243 simpl_binds :: SimplEnv -> [InBind] -> [OutId] -> SimplM (FloatsWith ())
244 simpl_binds env [] bs = ASSERT( null bs ) returnSmpl (emptyFloats env, ())
245 simpl_binds env (bind:binds) bs = simpl_bind env bind bs `thenSmpl` \ (floats,env) ->
246 addFloats env floats $ \env ->
247 simpl_binds env binds (drop_bs bind bs)
249 drop_bs (NonRec _ _) (_ : bs) = bs
250 drop_bs (Rec prs) bs = drop (length prs) bs
252 simpl_bind env bind bs
253 = getDOptsSmpl `thenSmpl` \ dflags ->
254 if dopt Opt_D_dump_inlinings dflags then
255 pprTrace "SimplBind" (ppr (bindersOf bind)) $ simpl_bind1 env bind bs
257 simpl_bind1 env bind bs
259 simpl_bind1 env (NonRec b r) (b':_) = simplRecOrTopPair env TopLevel b b' r
260 simpl_bind1 env (Rec pairs) bs' = simplRecBind env TopLevel pairs bs'
264 %************************************************************************
266 \subsection{simplNonRec}
268 %************************************************************************
270 simplNonRecBind is used for
271 * non-top-level non-recursive lets in expressions
275 * An unsimplified (binder, rhs) pair
276 * The env for the RHS. It may not be the same as the
277 current env because the bind might occur via (\x.E) arg
279 It uses the CPS form because the binding might be strict, in which
280 case we might discard the continuation:
281 let x* = error "foo" in (...x...)
283 It needs to turn unlifted bindings into a @case@. They can arise
284 from, say: (\x -> e) (4# + 3#)
287 simplNonRecBind :: SimplEnv
289 -> InExpr -> SimplEnv -- Arg, with its subst-env
290 -> OutType -- Type of thing computed by the context
291 -> (SimplEnv -> SimplM FloatsWithExpr) -- The body
292 -> SimplM FloatsWithExpr
294 simplNonRecBind env bndr rhs rhs_se cont_ty thing_inside
296 = pprPanic "simplNonRecBind" (ppr bndr <+> ppr rhs)
299 simplNonRecBind env bndr rhs rhs_se cont_ty thing_inside
300 | preInlineUnconditionally env NotTopLevel bndr
301 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
302 thing_inside (extendSubst env bndr (ContEx (getSubstEnv rhs_se) rhs))
305 | isStrictDmd (idNewDemandInfo bndr) || isStrictType (idType bndr) -- A strict let
306 = -- Don't use simplBinder because that doesn't keep
307 -- fragile occurrence info in the substitution
308 simplLetBndr env bndr `thenSmpl` \ (env, bndr1) ->
309 simplStrictArg AnRhs env rhs rhs_se (idType bndr1) cont_ty $ \ env1 rhs1 ->
311 -- Now complete the binding and simplify the body
313 -- simplLetBndr doesn't deal with the IdInfo, so we must
314 -- do so here (c.f. simplLazyBind)
315 bndr2 = bndr1 `setIdInfo` simplIdInfo (getSubst env) (idInfo bndr)
316 env2 = modifyInScope env1 bndr2 bndr2
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 simplLetBndr 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 A specialised variant of simplNonRec used when the RHS is already simplified, notably
330 in knownCon. It uses case-binding where necessary.
333 simplNonRecX :: SimplEnv
334 -> InId -- Old binder
335 -> OutExpr -- Simplified RHS
336 -> (SimplEnv -> SimplM FloatsWithExpr)
337 -> SimplM FloatsWithExpr
339 simplNonRecX env bndr new_rhs thing_inside
340 | needsCaseBinding (idType bndr) new_rhs
341 -- Make this test *before* the preInlineUnconditionally
342 -- Consider case I# (quotInt# x y) of
343 -- I# v -> let w = J# v in ...
344 -- If we gaily inline (quotInt# x y) for v, we end up building an
346 -- let w = J# (quotInt# x y) in ...
347 -- because quotInt# can fail.
348 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
349 thing_inside env `thenSmpl` \ (floats, body) ->
350 returnSmpl (emptyFloats env, Case new_rhs bndr' [(DEFAULT, [], wrapFloats floats body)])
352 | preInlineUnconditionally env NotTopLevel bndr
353 -- This happens; for example, the case_bndr during case of
354 -- known constructor: case (a,b) of x { (p,q) -> ... }
355 -- Here x isn't mentioned in the RHS, so we don't want to
356 -- create the (dead) let-binding let x = (a,b) in ...
358 -- Similarly, single occurrences can be inlined vigourously
359 -- e.g. case (f x, g y) of (a,b) -> ....
360 -- If a,b occur once we can avoid constructing the let binding for them.
361 = thing_inside (extendSubst env bndr (ContEx emptySubstEnv new_rhs))
364 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
365 completeNonRecX env False {- Non-strict; pessimistic -}
366 bndr bndr' new_rhs thing_inside
368 completeNonRecX env is_strict old_bndr new_bndr new_rhs thing_inside
369 = mkAtomicArgs is_strict
370 True {- OK to float unlifted -}
371 new_rhs `thenSmpl` \ (aux_binds, rhs2) ->
373 -- Make the arguments atomic if necessary,
374 -- adding suitable bindings
375 addAtomicBindsE env (fromOL aux_binds) $ \ env ->
376 completeLazyBind env NotTopLevel
377 old_bndr new_bndr rhs2 `thenSmpl` \ (floats, env) ->
378 addFloats env floats thing_inside
382 %************************************************************************
384 \subsection{Lazy bindings}
386 %************************************************************************
388 simplRecBind is used for
389 * recursive bindings only
392 simplRecBind :: SimplEnv -> TopLevelFlag
393 -> [(InId, InExpr)] -> [OutId]
394 -> SimplM (FloatsWith SimplEnv)
395 simplRecBind env top_lvl pairs bndrs'
396 = go env pairs bndrs' `thenSmpl` \ (floats, env) ->
397 returnSmpl (flattenFloats floats, env)
399 go env [] _ = returnSmpl (emptyFloats env, env)
401 go env ((bndr, rhs) : pairs) (bndr' : bndrs')
402 = simplRecOrTopPair env top_lvl bndr bndr' rhs `thenSmpl` \ (floats, env) ->
403 addFloats env floats (\env -> go env pairs bndrs')
407 simplRecOrTopPair is used for
408 * recursive bindings (whether top level or not)
409 * top-level non-recursive bindings
411 It assumes the binder has already been simplified, but not its IdInfo.
414 simplRecOrTopPair :: SimplEnv
416 -> InId -> OutId -- Binder, both pre-and post simpl
417 -> InExpr -- The RHS and its environment
418 -> SimplM (FloatsWith SimplEnv)
420 simplRecOrTopPair env top_lvl bndr bndr' rhs
421 | preInlineUnconditionally env top_lvl bndr -- Check for unconditional inline
422 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
423 returnSmpl (emptyFloats env, extendSubst env bndr (ContEx (getSubstEnv env) rhs))
426 = simplLazyBind env top_lvl Recursive bndr bndr' rhs env
427 -- May not actually be recursive, but it doesn't matter
431 simplLazyBind is used for
432 * recursive bindings (whether top level or not)
433 * top-level non-recursive bindings
434 * non-top-level *lazy* non-recursive bindings
436 [Thus it deals with the lazy cases from simplNonRecBind, and all cases
437 from SimplRecOrTopBind]
440 1. It assumes that the binder is *already* simplified,
441 and is in scope, but not its IdInfo
443 2. It assumes that the binder type is lifted.
445 3. It does not check for pre-inline-unconditionallly;
446 that should have been done already.
449 simplLazyBind :: SimplEnv
450 -> TopLevelFlag -> RecFlag
451 -> InId -> OutId -- Binder, both pre-and post simpl
452 -> InExpr -> SimplEnv -- The RHS and its environment
453 -> SimplM (FloatsWith SimplEnv)
455 simplLazyBind env top_lvl is_rec bndr bndr1 rhs rhs_se
456 = let -- Transfer the IdInfo of the original binder to the new binder
457 -- This is crucial: we must preserve
461 -- etc. To do this we must apply the current substitution,
462 -- which incorporates earlier substitutions in this very letrec group.
464 -- NB 1. We do this *before* processing the RHS of the binder, so that
465 -- its substituted rules are visible in its own RHS.
466 -- This is important. Manuel found cases where he really, really
467 -- wanted a RULE for a recursive function to apply in that function's
468 -- own right-hand side.
470 -- NB 2: We do not transfer the arity (see Subst.substIdInfo)
471 -- The arity of an Id should not be visible
472 -- in its own RHS, else we eta-reduce
476 -- which isn't sound. And it makes the arity in f's IdInfo greater than
477 -- the manifest arity, which isn't good.
478 -- The arity will get added later.
480 -- NB 3: It's important that we *do* transer the loop-breaker OccInfo,
481 -- because that's what stops the Id getting inlined infinitely, in the body
484 -- NB 4: does no harm for non-recursive bindings
486 bndr2 = bndr1 `setIdInfo` simplIdInfo (getSubst env) (idInfo bndr)
487 env1 = modifyInScope env bndr2 bndr2
488 rhs_env = setInScope rhs_se env1
489 is_top_level = isTopLevel top_lvl
490 ok_float_unlifted = not is_top_level && isNonRec is_rec
491 rhs_cont = mkStop (idType bndr1) AnRhs
493 -- Simplify the RHS; note the mkStop, which tells
494 -- the simplifier that this is the RHS of a let.
495 simplExprF rhs_env rhs rhs_cont `thenSmpl` \ (floats, rhs1) ->
497 -- If any of the floats can't be floated, give up now
498 -- (The allLifted predicate says True for empty floats.)
499 if (not ok_float_unlifted && not (allLifted floats)) then
500 completeLazyBind env1 top_lvl bndr bndr2
501 (wrapFloats floats rhs1)
504 -- ANF-ise a constructor or PAP rhs
505 mkAtomicArgs False {- Not strict -}
506 ok_float_unlifted rhs1 `thenSmpl` \ (aux_binds, rhs2) ->
508 -- If the result is a PAP, float the floats out, else wrap them
509 -- By this time it's already been ANF-ised (if necessary)
510 if isEmptyFloats floats && isNilOL aux_binds then -- Shortcut a common case
511 completeLazyBind env1 top_lvl bndr bndr2 rhs2
513 -- We use exprIsTrivial here because we want to reveal lone variables.
514 -- E.g. let { x = letrec { y = E } in y } in ...
515 -- Here we definitely want to float the y=E defn.
516 -- exprIsValue definitely isn't right for that.
518 -- BUT we can't use "exprIsCheap", because that causes a strictness bug.
519 -- x = let y* = E in case (scc y) of { T -> F; F -> T}
520 -- The case expression is 'cheap', but it's wrong to transform to
521 -- y* = E; x = case (scc y) of {...}
522 -- Either we must be careful not to float demanded non-values, or
523 -- we must use exprIsValue for the test, which ensures that the
524 -- thing is non-strict. I think. The WARN below tests for this.
525 else if is_top_level || exprIsTrivial rhs2 || exprIsValue rhs2 then
527 -- There's a subtlety here. There may be a binding (x* = e) in the
528 -- floats, where the '*' means 'will be demanded'. So is it safe
529 -- to float it out? Answer no, but it won't matter because
530 -- we only float if (a) arg' is a WHNF, or (b) it's going to top level
531 -- and so there can't be any 'will be demanded' bindings in the floats.
533 WARN( not is_top_level && any demanded_float (floatBinds floats),
534 ppr (filter demanded_float (floatBinds floats)) )
536 tick LetFloatFromLet `thenSmpl_` (
537 addFloats env1 floats $ \ env2 ->
538 addAtomicBinds env2 (fromOL aux_binds) $ \ env3 ->
539 completeLazyBind env3 top_lvl bndr bndr2 rhs2)
542 completeLazyBind env1 top_lvl bndr bndr2 (wrapFloats floats rhs1)
545 demanded_float (NonRec b r) = isStrictDmd (idNewDemandInfo b) && not (isUnLiftedType (idType b))
546 -- Unlifted-type (cheap-eagerness) lets may well have a demanded flag on them
547 demanded_float (Rec _) = False
552 %************************************************************************
554 \subsection{Completing a lazy binding}
556 %************************************************************************
559 * deals only with Ids, not TyVars
560 * takes an already-simplified binder and RHS
561 * is used for both recursive and non-recursive bindings
562 * is used for both top-level and non-top-level bindings
564 It does the following:
565 - tries discarding a dead binding
566 - tries PostInlineUnconditionally
567 - add unfolding [this is the only place we add an unfolding]
570 It does *not* attempt to do let-to-case. Why? Because it is used for
571 - top-level bindings (when let-to-case is impossible)
572 - many situations where the "rhs" is known to be a WHNF
573 (so let-to-case is inappropriate).
576 completeLazyBind :: SimplEnv
577 -> TopLevelFlag -- Flag stuck into unfolding
578 -> InId -- Old binder
579 -> OutId -- New binder
580 -> OutExpr -- Simplified RHS
581 -> SimplM (FloatsWith SimplEnv)
582 -- We return a new SimplEnv, because completeLazyBind may choose to do its work
583 -- by extending the substitution (e.g. let x = y in ...)
584 -- The new binding (if any) is returned as part of the floats.
585 -- NB: the returned SimplEnv has the right SubstEnv, but you should
586 -- (as usual) use the in-scope-env from the floats
588 completeLazyBind env top_lvl old_bndr new_bndr new_rhs
589 | postInlineUnconditionally env new_bndr occ_info new_rhs
590 = -- Drop the binding
591 tick (PostInlineUnconditionally old_bndr) `thenSmpl_`
592 returnSmpl (emptyFloats env, extendSubst env old_bndr (DoneEx new_rhs))
593 -- Use the substitution to make quite, quite sure that the substitution
594 -- will happen, since we are going to discard the binding
599 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
601 -- Add the unfolding *only* for non-loop-breakers
602 -- Making loop breakers not have an unfolding at all
603 -- means that we can avoid tests in exprIsConApp, for example.
604 -- This is important: if exprIsConApp says 'yes' for a recursive
605 -- thing, then we can get into an infinite loop
606 info_w_unf | loop_breaker = new_bndr_info
607 | otherwise = new_bndr_info `setUnfoldingInfo` unfolding
608 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
610 final_id = new_bndr `setIdInfo` info_w_unf
612 -- These seqs forces the Id, and hence its IdInfo,
613 -- and hence any inner substitutions
615 returnSmpl (unitFloat env final_id new_rhs, env)
618 loop_breaker = isLoopBreaker occ_info
619 old_info = idInfo old_bndr
620 occ_info = occInfo old_info
625 %************************************************************************
627 \subsection[Simplify-simplExpr]{The main function: simplExpr}
629 %************************************************************************
631 The reason for this OutExprStuff stuff is that we want to float *after*
632 simplifying a RHS, not before. If we do so naively we get quadratic
633 behaviour as things float out.
635 To see why it's important to do it after, consider this (real) example:
649 a -- Can't inline a this round, cos it appears twice
653 Each of the ==> steps is a round of simplification. We'd save a
654 whole round if we float first. This can cascade. Consider
659 let f = let d1 = ..d.. in \y -> e
663 in \x -> ...(\y ->e)...
665 Only in this second round can the \y be applied, and it
666 might do the same again.
670 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
671 simplExpr env expr = simplExprC env expr (mkStop expr_ty' AnArg)
673 expr_ty' = substTy (getSubst env) (exprType expr)
674 -- The type in the Stop continuation, expr_ty', is usually not used
675 -- It's only needed when discarding continuations after finding
676 -- a function that returns bottom.
677 -- Hence the lazy substitution
680 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
681 -- Simplify an expression, given a continuation
682 simplExprC env expr cont
683 = simplExprF env expr cont `thenSmpl` \ (floats, expr) ->
684 returnSmpl (wrapFloats floats expr)
686 simplExprF :: SimplEnv -> InExpr -> SimplCont -> SimplM FloatsWithExpr
687 -- Simplify an expression, returning floated binds
689 simplExprF env (Var v) cont = simplVar env v cont
690 simplExprF env (Lit lit) cont = rebuild env (Lit lit) cont
691 simplExprF env expr@(Lam _ _) cont = simplLam env expr cont
692 simplExprF env (Note note expr) cont = simplNote env note expr cont
693 simplExprF env (App fun arg) cont = simplExprF env fun (ApplyTo NoDup arg env cont)
695 simplExprF env (Type ty) cont
696 = ASSERT( contIsRhsOrArg cont )
697 simplType env ty `thenSmpl` \ ty' ->
698 rebuild env (Type ty') cont
700 simplExprF env (Case scrut bndr alts) cont
701 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
702 = -- Simplify the scrutinee with a Select continuation
703 simplExprF env scrut (Select NoDup bndr alts env cont)
706 = -- If case-of-case is off, simply simplify the case expression
707 -- in a vanilla Stop context, and rebuild the result around it
708 simplExprC env scrut case_cont `thenSmpl` \ case_expr' ->
709 rebuild env case_expr' cont
711 case_cont = Select NoDup bndr alts env (mkBoringStop (contResultType cont))
713 simplExprF env (Let (Rec pairs) body) cont
714 = simplRecBndrs env (map fst pairs) `thenSmpl` \ (env, bndrs') ->
715 -- NB: bndrs' don't have unfoldings or rules
716 -- We add them as we go down
718 simplRecBind env NotTopLevel pairs bndrs' `thenSmpl` \ (floats, env) ->
719 addFloats env floats $ \ env ->
720 simplExprF env body cont
722 -- A non-recursive let is dealt with by simplNonRecBind
723 simplExprF env (Let (NonRec bndr rhs) body) cont
724 = simplNonRecBind env bndr rhs env (contResultType cont) $ \ env ->
725 simplExprF env body cont
728 ---------------------------------
729 simplType :: SimplEnv -> InType -> SimplM OutType
730 -- Kept monadic just so we can do the seqType
732 = seqType new_ty `seq` returnSmpl new_ty
734 new_ty = substTy (getSubst env) ty
738 %************************************************************************
742 %************************************************************************
745 simplLam env fun cont
748 zap_it = mkLamBndrZapper fun (countArgs cont)
749 cont_ty = contResultType cont
751 -- Type-beta reduction
752 go env (Lam bndr body) (ApplyTo _ (Type ty_arg) arg_se body_cont)
753 = ASSERT( isTyVar bndr )
754 tick (BetaReduction bndr) `thenSmpl_`
755 simplType (setInScope arg_se env) ty_arg `thenSmpl` \ ty_arg' ->
756 go (extendSubst env bndr (DoneTy ty_arg')) body body_cont
758 -- Ordinary beta reduction
759 go env (Lam bndr body) cont@(ApplyTo _ arg arg_se body_cont)
760 = tick (BetaReduction bndr) `thenSmpl_`
761 simplNonRecBind env (zap_it bndr) arg arg_se cont_ty $ \ env ->
762 go env body body_cont
764 -- Not enough args, so there are real lambdas left to put in the result
765 go env lam@(Lam _ _) cont
766 = simplLamBndrs env bndrs `thenSmpl` \ (env, bndrs') ->
767 simplExpr env body `thenSmpl` \ body' ->
768 mkLam env bndrs' body' cont `thenSmpl` \ (floats, new_lam) ->
769 addFloats env floats $ \ env ->
770 rebuild env new_lam cont
772 (bndrs,body) = collectBinders lam
774 -- Exactly enough args
775 go env expr cont = simplExprF env expr cont
777 mkLamBndrZapper :: CoreExpr -- Function
778 -> Int -- Number of args supplied, *including* type args
779 -> Id -> Id -- Use this to zap the binders
780 mkLamBndrZapper fun n_args
781 | n_args >= n_params fun = \b -> b -- Enough args
782 | otherwise = \b -> zapLamIdInfo b
784 -- NB: we count all the args incl type args
785 -- so we must count all the binders (incl type lambdas)
786 n_params (Note _ e) = n_params e
787 n_params (Lam b e) = 1 + n_params e
788 n_params other = 0::Int
792 %************************************************************************
796 %************************************************************************
799 simplNote env (Coerce to from) body cont
801 in_scope = getInScope env
803 addCoerce s1 k1 (CoerceIt t1 cont)
804 -- coerce T1 S1 (coerce S1 K1 e)
807 -- coerce T1 K1 e, otherwise
809 -- For example, in the initial form of a worker
810 -- we may find (coerce T (coerce S (\x.e))) y
811 -- and we'd like it to simplify to e[y/x] in one round
813 | t1 `eqType` k1 = cont -- The coerces cancel out
814 | otherwise = CoerceIt t1 cont -- They don't cancel, but
815 -- the inner one is redundant
817 addCoerce t1t2 s1s2 (ApplyTo dup arg arg_se cont)
818 | not (isTypeArg arg), -- This whole case only works for value args
819 -- Could upgrade to have equiv thing for type apps too
820 Just (s1, s2) <- splitFunTy_maybe s1s2
821 -- (coerce (T1->T2) (S1->S2) F) E
823 -- coerce T2 S2 (F (coerce S1 T1 E))
825 -- t1t2 must be a function type, T1->T2, because it's applied to something
826 -- but s1s2 might conceivably not be
828 -- When we build the ApplyTo we can't mix the out-types
829 -- with the InExpr in the argument, so we simply substitute
830 -- to make it all consistent. It's a bit messy.
831 -- But it isn't a common case.
833 (t1,t2) = splitFunTy t1t2
834 new_arg = mkCoerce2 s1 t1 (substExpr (mkSubst in_scope (getSubstEnv arg_se)) arg)
836 ApplyTo dup new_arg (zapSubstEnv env) (addCoerce t2 s2 cont)
838 addCoerce to' _ cont = CoerceIt to' cont
840 simplType env to `thenSmpl` \ to' ->
841 simplType env from `thenSmpl` \ from' ->
842 simplExprF env body (addCoerce to' from' cont)
845 -- Hack: we only distinguish subsumed cost centre stacks for the purposes of
846 -- inlining. All other CCCSs are mapped to currentCCS.
847 simplNote env (SCC cc) e cont
848 = simplExpr (setEnclosingCC env currentCCS) e `thenSmpl` \ e' ->
849 rebuild env (mkSCC cc e') cont
851 simplNote env InlineCall e cont
852 = simplExprF env e (InlinePlease cont)
854 -- See notes with SimplMonad.inlineMode
855 simplNote env InlineMe e cont
856 | contIsRhsOrArg cont -- Totally boring continuation; see notes above
857 = -- Don't inline inside an INLINE expression
858 simplExpr (setMode inlineMode env ) e `thenSmpl` \ e' ->
859 rebuild env (mkInlineMe e') cont
861 | otherwise -- Dissolve the InlineMe note if there's
862 -- an interesting context of any kind to combine with
863 -- (even a type application -- anything except Stop)
864 = simplExprF env e cont
866 simplNote env (CoreNote s) e cont
867 = simplExpr env e `thenSmpl` \ e' ->
868 rebuild env (Note (CoreNote s) e') cont
872 %************************************************************************
874 \subsection{Dealing with calls}
876 %************************************************************************
879 simplVar env var cont
880 = case lookupIdSubst (getSubst env) var of
881 DoneEx e -> simplExprF (zapSubstEnv env) e cont
882 ContEx se e -> simplExprF (setSubstEnv env se) e cont
883 DoneId var1 occ -> WARN( not (isInScope var1 (getSubst env)) && mustHaveLocalBinding var1,
884 text "simplVar:" <+> ppr var )
885 completeCall (zapSubstEnv env) var1 occ cont
886 -- The template is already simplified, so don't re-substitute.
887 -- This is VITAL. Consider
889 -- let y = \z -> ...x... in
891 -- We'll clone the inner \x, adding x->x' in the id_subst
892 -- Then when we inline y, we must *not* replace x by x' in
893 -- the inlined copy!!
895 ---------------------------------------------------------
896 -- Dealing with a call site
898 completeCall env var occ_info cont
899 = -- Simplify the arguments
900 getDOptsSmpl `thenSmpl` \ dflags ->
902 chkr = getSwitchChecker env
903 (args, call_cont, inline_call) = getContArgs chkr var cont
906 simplifyArgs env fn_ty args (contResultType call_cont) $ \ env args ->
908 -- Next, look for rules or specialisations that match
910 -- It's important to simplify the args first, because the rule-matcher
911 -- doesn't do substitution as it goes. We don't want to use subst_args
912 -- (defined in the 'where') because that throws away useful occurrence info,
913 -- and perhaps-very-important specialisations.
915 -- Some functions have specialisations *and* are strict; in this case,
916 -- we don't want to inline the wrapper of the non-specialised thing; better
917 -- to call the specialised thing instead.
918 -- We used to use the black-listing mechanism to ensure that inlining of
919 -- the wrapper didn't occur for things that have specialisations till a
920 -- later phase, so but now we just try RULES first
922 -- You might think that we shouldn't apply rules for a loop breaker:
923 -- doing so might give rise to an infinite loop, because a RULE is
924 -- rather like an extra equation for the function:
925 -- RULE: f (g x) y = x+y
928 -- But it's too drastic to disable rules for loop breakers.
929 -- Even the foldr/build rule would be disabled, because foldr
930 -- is recursive, and hence a loop breaker:
931 -- foldr k z (build g) = g k z
932 -- So it's up to the programmer: rules can cause divergence
935 in_scope = getInScope env
936 maybe_rule = case activeRule env of
937 Nothing -> Nothing -- No rules apply
938 Just act_fn -> lookupRule act_fn in_scope var args
941 Just (rule_name, rule_rhs) ->
942 tick (RuleFired rule_name) `thenSmpl_`
943 (if dopt Opt_D_dump_inlinings dflags then
944 pprTrace "Rule fired" (vcat [
945 text "Rule:" <+> ftext rule_name,
946 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
947 text "After: " <+> pprCoreExpr rule_rhs,
948 text "Cont: " <+> ppr call_cont])
951 simplExprF env rule_rhs call_cont ;
953 Nothing -> -- No rules
955 -- Next, look for an inlining
957 arg_infos = [ interestingArg arg | arg <- args, isValArg arg]
959 interesting_cont = interestingCallContext (notNull args)
963 active_inline = activeInline env var occ_info
964 maybe_inline = callSiteInline dflags active_inline inline_call occ_info
965 var arg_infos interesting_cont
967 case maybe_inline of {
968 Just unfolding -- There is an inlining!
969 -> tick (UnfoldingDone var) `thenSmpl_`
970 makeThatCall env var unfolding args call_cont
973 Nothing -> -- No inlining!
976 rebuild env (mkApps (Var var) args) call_cont
979 makeThatCall :: SimplEnv
981 -> InExpr -- Inlined function rhs
982 -> [OutExpr] -- Arguments, already simplified
983 -> SimplCont -- After the call
984 -> SimplM FloatsWithExpr
985 -- Similar to simplLam, but this time
986 -- the arguments are already simplified
987 makeThatCall orig_env var fun@(Lam _ _) args cont
988 = go orig_env fun args
990 zap_it = mkLamBndrZapper fun (length args)
992 -- Type-beta reduction
993 go env (Lam bndr body) (Type ty_arg : args)
994 = ASSERT( isTyVar bndr )
995 tick (BetaReduction bndr) `thenSmpl_`
996 go (extendSubst env bndr (DoneTy ty_arg)) body args
998 -- Ordinary beta reduction
999 go env (Lam bndr body) (arg : args)
1000 = tick (BetaReduction bndr) `thenSmpl_`
1001 simplNonRecX env (zap_it bndr) arg $ \ env ->
1004 -- Not enough args, so there are real lambdas left to put in the result
1006 = simplExprF env fun (pushContArgs orig_env args cont)
1007 -- NB: orig_env; the correct environment to capture with
1008 -- the arguments.... env has been augmented with substitutions
1009 -- from the beta reductions.
1011 makeThatCall env var fun args cont
1012 = simplExprF env fun (pushContArgs env args cont)
1016 %************************************************************************
1018 \subsection{Arguments}
1020 %************************************************************************
1023 ---------------------------------------------------------
1024 -- Simplifying the arguments of a call
1026 simplifyArgs :: SimplEnv
1027 -> OutType -- Type of the function
1028 -> [(InExpr, SimplEnv, Bool)] -- Details of the arguments
1029 -> OutType -- Type of the continuation
1030 -> (SimplEnv -> [OutExpr] -> SimplM FloatsWithExpr)
1031 -> SimplM FloatsWithExpr
1033 -- [CPS-like because of strict arguments]
1035 -- Simplify the arguments to a call.
1036 -- This part of the simplifier may break the no-shadowing invariant
1038 -- f (...(\a -> e)...) (case y of (a,b) -> e')
1039 -- where f is strict in its second arg
1040 -- If we simplify the innermost one first we get (...(\a -> e)...)
1041 -- Simplifying the second arg makes us float the case out, so we end up with
1042 -- case y of (a,b) -> f (...(\a -> e)...) e'
1043 -- So the output does not have the no-shadowing invariant. However, there is
1044 -- no danger of getting name-capture, because when the first arg was simplified
1045 -- we used an in-scope set that at least mentioned all the variables free in its
1046 -- static environment, and that is enough.
1048 -- We can't just do innermost first, or we'd end up with a dual problem:
1049 -- case x of (a,b) -> f e (...(\a -> e')...)
1051 -- I spent hours trying to recover the no-shadowing invariant, but I just could
1052 -- not think of an elegant way to do it. The simplifier is already knee-deep in
1053 -- continuations. We have to keep the right in-scope set around; AND we have
1054 -- to get the effect that finding (error "foo") in a strict arg position will
1055 -- discard the entire application and replace it with (error "foo"). Getting
1056 -- all this at once is TOO HARD!
1058 simplifyArgs env fn_ty args cont_ty thing_inside
1059 = go env fn_ty args thing_inside
1061 go env fn_ty [] thing_inside = thing_inside env []
1062 go env fn_ty (arg:args) thing_inside = simplifyArg env fn_ty arg cont_ty $ \ env arg' ->
1063 go env (applyTypeToArg fn_ty arg') args $ \ env args' ->
1064 thing_inside env (arg':args')
1066 simplifyArg env fn_ty (Type ty_arg, se, _) cont_ty thing_inside
1067 = simplType (setInScope se env) ty_arg `thenSmpl` \ new_ty_arg ->
1068 thing_inside env (Type new_ty_arg)
1070 simplifyArg env fn_ty (val_arg, arg_se, is_strict) cont_ty thing_inside
1072 = simplStrictArg AnArg env val_arg arg_se arg_ty cont_ty thing_inside
1074 | otherwise -- Lazy argument
1075 -- DO NOT float anything outside, hence simplExprC
1076 -- There is no benefit (unlike in a let-binding), and we'd
1077 -- have to be very careful about bogus strictness through
1078 -- floating a demanded let.
1079 = simplExprC (setInScope arg_se env) val_arg
1080 (mkStop arg_ty AnArg) `thenSmpl` \ arg1 ->
1081 thing_inside env arg1
1083 arg_ty = funArgTy fn_ty
1086 simplStrictArg :: LetRhsFlag
1087 -> SimplEnv -- The env of the call
1088 -> InExpr -> SimplEnv -- The arg plus its env
1089 -> OutType -- arg_ty: type of the argument
1090 -> OutType -- cont_ty: Type of thing computed by the context
1091 -> (SimplEnv -> OutExpr -> SimplM FloatsWithExpr)
1092 -- Takes an expression of type rhs_ty,
1093 -- returns an expression of type cont_ty
1094 -- The env passed to this continuation is the
1095 -- env of the call, plus any new in-scope variables
1096 -> SimplM FloatsWithExpr -- An expression of type cont_ty
1098 simplStrictArg is_rhs call_env arg arg_env arg_ty cont_ty thing_inside
1099 = simplExprF (setInScope arg_env call_env) arg
1100 (ArgOf is_rhs arg_ty cont_ty (\ new_env -> thing_inside (setInScope call_env new_env)))
1101 -- Notice the way we use arg_env (augmented with in-scope vars from call_env)
1102 -- to simplify the argument
1103 -- and call-env (augmented with in-scope vars from the arg) to pass to the continuation
1107 %************************************************************************
1109 \subsection{mkAtomicArgs}
1111 %************************************************************************
1113 mkAtomicArgs takes a putative RHS, checks whether it's a PAP or
1114 constructor application and, if so, converts it to ANF, so that the
1115 resulting thing can be inlined more easily. Thus
1122 There are three sorts of binding context, specified by the two
1128 N N Top-level or recursive Only bind args of lifted type
1130 N Y Non-top-level and non-recursive, Bind args of lifted type, or
1131 but lazy unlifted-and-ok-for-speculation
1133 Y Y Non-top-level, non-recursive, Bind all args
1134 and strict (demanded)
1141 there is no point in transforming to
1143 x = case (y div# z) of r -> MkC r
1145 because the (y div# z) can't float out of the let. But if it was
1146 a *strict* let, then it would be a good thing to do. Hence the
1147 context information.
1150 mkAtomicArgs :: Bool -- A strict binding
1151 -> Bool -- OK to float unlifted args
1153 -> SimplM (OrdList (OutId,OutExpr), -- The floats (unusually) may include
1154 OutExpr) -- things that need case-binding,
1155 -- if the strict-binding flag is on
1157 mkAtomicArgs is_strict ok_float_unlifted rhs
1158 | (Var fun, args) <- collectArgs rhs, -- It's an application
1159 isDataConWorkId fun || valArgCount args < idArity fun -- And it's a constructor or PAP
1160 = go fun nilOL [] args -- Have a go
1162 | otherwise = bale_out -- Give up
1165 bale_out = returnSmpl (nilOL, rhs)
1167 go fun binds rev_args []
1168 = returnSmpl (binds, mkApps (Var fun) (reverse rev_args))
1170 go fun binds rev_args (arg : args)
1171 | exprIsTrivial arg -- Easy case
1172 = go fun binds (arg:rev_args) args
1174 | not can_float_arg -- Can't make this arg atomic
1175 = bale_out -- ... so give up
1177 | otherwise -- Don't forget to do it recursively
1178 -- E.g. x = a:b:c:[]
1179 = mkAtomicArgs is_strict ok_float_unlifted arg `thenSmpl` \ (arg_binds, arg') ->
1180 newId FSLIT("a") arg_ty `thenSmpl` \ arg_id ->
1181 go fun ((arg_binds `snocOL` (arg_id,arg')) `appOL` binds)
1182 (Var arg_id : rev_args) args
1184 arg_ty = exprType arg
1185 can_float_arg = is_strict
1186 || not (isUnLiftedType arg_ty)
1187 || (ok_float_unlifted && exprOkForSpeculation arg)
1190 addAtomicBinds :: SimplEnv -> [(OutId,OutExpr)]
1191 -> (SimplEnv -> SimplM (FloatsWith a))
1192 -> SimplM (FloatsWith a)
1193 addAtomicBinds env [] thing_inside = thing_inside env
1194 addAtomicBinds env ((v,r):bs) thing_inside = addAuxiliaryBind env (NonRec v r) $ \ env ->
1195 addAtomicBinds env bs thing_inside
1197 addAtomicBindsE :: SimplEnv -> [(OutId,OutExpr)]
1198 -> (SimplEnv -> SimplM FloatsWithExpr)
1199 -> SimplM FloatsWithExpr
1200 -- Same again, but this time we're in an expression context,
1201 -- and may need to do some case bindings
1203 addAtomicBindsE env [] thing_inside
1205 addAtomicBindsE env ((v,r):bs) thing_inside
1206 | needsCaseBinding (idType v) r
1207 = addAtomicBindsE (addNewInScopeIds env [v]) bs thing_inside `thenSmpl` \ (floats, expr) ->
1208 WARN( exprIsTrivial expr, ppr v <+> pprCoreExpr expr )
1209 returnSmpl (emptyFloats env, Case r v [(DEFAULT,[], wrapFloats floats expr)])
1212 = addAuxiliaryBind env (NonRec v r) $ \ env ->
1213 addAtomicBindsE env bs thing_inside
1217 %************************************************************************
1219 \subsection{The main rebuilder}
1221 %************************************************************************
1224 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM FloatsWithExpr
1226 rebuild env expr (Stop _ _ _) = rebuildDone env expr
1227 rebuild env expr (ArgOf _ _ _ cont_fn) = cont_fn env expr
1228 rebuild env expr (CoerceIt to_ty cont) = rebuild env (mkCoerce to_ty expr) cont
1229 rebuild env expr (InlinePlease cont) = rebuild env (Note InlineCall expr) cont
1230 rebuild env expr (Select _ bndr alts se cont) = rebuildCase (setInScope se env) expr bndr alts cont
1231 rebuild env expr (ApplyTo _ arg se cont) = rebuildApp (setInScope se env) expr arg cont
1233 rebuildApp env fun arg cont
1234 = simplExpr env arg `thenSmpl` \ arg' ->
1235 rebuild env (App fun arg') cont
1237 rebuildDone env expr = returnSmpl (emptyFloats env, expr)
1241 %************************************************************************
1243 \subsection{Functions dealing with a case}
1245 %************************************************************************
1247 Blob of helper functions for the "case-of-something-else" situation.
1250 ---------------------------------------------------------
1251 -- Eliminate the case if possible
1253 rebuildCase :: SimplEnv
1254 -> OutExpr -- Scrutinee
1255 -> InId -- Case binder
1256 -> [InAlt] -- Alternatives
1258 -> SimplM FloatsWithExpr
1260 rebuildCase env scrut case_bndr alts cont
1261 | Just (con,args) <- exprIsConApp_maybe scrut
1262 -- Works when the scrutinee is a variable with a known unfolding
1263 -- as well as when it's an explicit constructor application
1264 = knownCon env (DataAlt con) args case_bndr alts cont
1266 | Lit lit <- scrut -- No need for same treatment as constructors
1267 -- because literals are inlined more vigorously
1268 = knownCon env (LitAlt lit) [] case_bndr alts cont
1271 = prepareAlts scrut case_bndr alts `thenSmpl` \ (better_alts, handled_cons) ->
1273 -- Deal with the case binder, and prepare the continuation;
1274 -- The new subst_env is in place
1275 prepareCaseCont env better_alts cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1276 addFloats env floats $ \ env ->
1278 -- Deal with variable scrutinee
1279 simplCaseBinder env scrut case_bndr `thenSmpl` \ (alt_env, case_bndr', zap_occ_info) ->
1281 -- Deal with the case alternatives
1282 simplAlts alt_env zap_occ_info handled_cons
1283 case_bndr' better_alts dup_cont `thenSmpl` \ alts' ->
1285 -- Put the case back together
1286 mkCase scrut case_bndr' alts' `thenSmpl` \ case_expr ->
1288 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1289 -- The case binder *not* scope over the whole returned case-expression
1290 rebuild env case_expr nondup_cont
1293 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1294 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1295 way, there's a chance that v will now only be used once, and hence
1300 There is a time we *don't* want to do that, namely when
1301 -fno-case-of-case is on. This happens in the first simplifier pass,
1302 and enhances full laziness. Here's the bad case:
1303 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1304 If we eliminate the inner case, we trap it inside the I# v -> arm,
1305 which might prevent some full laziness happening. I've seen this
1306 in action in spectral/cichelli/Prog.hs:
1307 [(m,n) | m <- [1..max], n <- [1..max]]
1308 Hence the check for NoCaseOfCase.
1312 There is another situation when we don't want to do it. If we have
1314 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1315 ...other cases .... }
1317 We'll perform the binder-swap for the outer case, giving
1319 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1320 ...other cases .... }
1322 But there is no point in doing it for the inner case, because w1 can't
1323 be inlined anyway. Furthermore, doing the case-swapping involves
1324 zapping w2's occurrence info (see paragraphs that follow), and that
1325 forces us to bind w2 when doing case merging. So we get
1327 case x of w1 { A -> let w2 = w1 in e1
1328 B -> let w2 = w1 in e2
1329 ...other cases .... }
1331 This is plain silly in the common case where w2 is dead.
1333 Even so, I can't see a good way to implement this idea. I tried
1334 not doing the binder-swap if the scrutinee was already evaluated
1335 but that failed big-time:
1339 case v of w { MkT x ->
1340 case x of x1 { I# y1 ->
1341 case x of x2 { I# y2 -> ...
1343 Notice that because MkT is strict, x is marked "evaluated". But to
1344 eliminate the last case, we must either make sure that x (as well as
1345 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1346 the binder-swap. So this whole note is a no-op.
1350 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1351 any occurrence info (eg IAmDead) in the case binder, because the
1352 case-binder now effectively occurs whenever v does. AND we have to do
1353 the same for the pattern-bound variables! Example:
1355 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1357 Here, b and p are dead. But when we move the argment inside the first
1358 case RHS, and eliminate the second case, we get
1360 case x or { (a,b) -> a b }
1362 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1363 happened. Hence the zap_occ_info function returned by simplCaseBinder
1366 simplCaseBinder env (Var v) case_bndr
1367 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
1369 -- Failed try [see Note 2 above]
1370 -- not (isEvaldUnfolding (idUnfolding v))
1372 = simplBinder env (zap case_bndr) `thenSmpl` \ (env, case_bndr') ->
1373 returnSmpl (modifyInScope env v case_bndr', case_bndr', zap)
1374 -- We could extend the substitution instead, but it would be
1375 -- a hack because then the substitution wouldn't be idempotent
1376 -- any more (v is an OutId). And this just just as well.
1378 zap b = b `setIdOccInfo` NoOccInfo
1380 simplCaseBinder env other_scrut case_bndr
1381 = simplBinder env case_bndr `thenSmpl` \ (env, case_bndr') ->
1382 returnSmpl (env, case_bndr', \ bndr -> bndr) -- NoOp on bndr
1388 simplAlts :: SimplEnv
1389 -> (InId -> InId) -- Occ-info zapper
1390 -> [AltCon] -- Alternatives the scrutinee can't be
1391 -- in the default case
1392 -> OutId -- Case binder
1393 -> [InAlt] -> SimplCont
1394 -> SimplM [OutAlt] -- Includes the continuation
1396 simplAlts env zap_occ_info handled_cons case_bndr' alts cont'
1397 = mapSmpl simpl_alt alts
1399 inst_tys' = tyConAppArgs (idType case_bndr')
1401 simpl_alt (DEFAULT, _, rhs)
1403 -- In the default case we record the constructors that the
1404 -- case-binder *can't* be.
1405 -- We take advantage of any OtherCon info in the case scrutinee
1406 case_bndr_w_unf = case_bndr' `setIdUnfolding` mkOtherCon handled_cons
1407 env_with_unf = modifyInScope env case_bndr' case_bndr_w_unf
1409 simplExprC env_with_unf rhs cont' `thenSmpl` \ rhs' ->
1410 returnSmpl (DEFAULT, [], rhs')
1412 simpl_alt (con, vs, rhs)
1413 = -- Deal with the pattern-bound variables
1414 -- Mark the ones that are in ! positions in the data constructor
1415 -- as certainly-evaluated.
1416 -- NB: it happens that simplBinders does *not* erase the OtherCon
1417 -- form of unfolding, so it's ok to add this info before
1418 -- doing simplBinders
1419 simplBinders env (add_evals con vs) `thenSmpl` \ (env, vs') ->
1421 -- Bind the case-binder to (con args)
1423 unfolding = mkUnfolding False (mkAltExpr con vs' inst_tys')
1424 env_with_unf = modifyInScope env case_bndr' (case_bndr' `setIdUnfolding` unfolding)
1426 simplExprC env_with_unf rhs cont' `thenSmpl` \ rhs' ->
1427 returnSmpl (con, vs', rhs')
1430 -- add_evals records the evaluated-ness of the bound variables of
1431 -- a case pattern. This is *important*. Consider
1432 -- data T = T !Int !Int
1434 -- case x of { T a b -> T (a+1) b }
1436 -- We really must record that b is already evaluated so that we don't
1437 -- go and re-evaluate it when constructing the result.
1439 add_evals (DataAlt dc) vs = cat_evals vs (dataConRepStrictness dc)
1440 add_evals other_con vs = vs
1442 cat_evals [] [] = []
1443 cat_evals (v:vs) (str:strs)
1444 | isTyVar v = v : cat_evals vs (str:strs)
1445 | isMarkedStrict str = evald_v : cat_evals vs strs
1446 | otherwise = zapped_v : cat_evals vs strs
1448 zapped_v = zap_occ_info v
1449 evald_v = zapped_v `setIdUnfolding` mkOtherCon []
1453 %************************************************************************
1455 \subsection{Known constructor}
1457 %************************************************************************
1459 We are a bit careful with occurrence info. Here's an example
1461 (\x* -> case x of (a*, b) -> f a) (h v, e)
1463 where the * means "occurs once". This effectively becomes
1464 case (h v, e) of (a*, b) -> f a)
1466 let a* = h v; b = e in f a
1470 All this should happen in one sweep.
1473 knownCon :: SimplEnv -> AltCon -> [OutExpr]
1474 -> InId -> [InAlt] -> SimplCont
1475 -> SimplM FloatsWithExpr
1477 knownCon env con args bndr alts cont
1478 = tick (KnownBranch bndr) `thenSmpl_`
1479 case findAlt con alts of
1480 (DEFAULT, bs, rhs) -> ASSERT( null bs )
1481 simplNonRecX env bndr scrut $ \ env ->
1482 -- This might give rise to a binding with non-atomic args
1483 -- like x = Node (f x) (g x)
1484 -- but no harm will be done
1485 simplExprF env rhs cont
1488 LitAlt lit -> Lit lit
1489 DataAlt dc -> mkConApp dc args
1491 (LitAlt lit, bs, rhs) -> ASSERT( null bs )
1492 simplNonRecX env bndr (Lit lit) $ \ env ->
1493 simplExprF env rhs cont
1495 (DataAlt dc, bs, rhs) -> ASSERT( length bs + n_tys == length args )
1496 bind_args env bs (drop n_tys args) $ \ env ->
1498 con_app = mkConApp dc (take n_tys args ++ con_args)
1499 con_args = [substExpr (getSubst env) (varToCoreExpr b) | b <- bs]
1500 -- args are aready OutExprs, but bs are InIds
1502 simplNonRecX env bndr con_app $ \ env ->
1503 simplExprF env rhs cont
1505 n_tys = dataConNumInstArgs dc -- Non-existential type args
1507 bind_args env [] _ thing_inside = thing_inside env
1509 bind_args env (b:bs) (Type ty : args) thing_inside
1510 = bind_args (extendSubst env b (DoneTy ty)) bs args thing_inside
1512 bind_args env (b:bs) (arg : args) thing_inside
1513 = simplNonRecX env b arg $ \ env ->
1514 bind_args env bs args thing_inside
1518 %************************************************************************
1520 \subsection{Duplicating continuations}
1522 %************************************************************************
1525 prepareCaseCont :: SimplEnv
1526 -> [InAlt] -> SimplCont
1527 -> SimplM (FloatsWith (SimplCont,SimplCont))
1528 -- Return a duplicatable continuation, a non-duplicable part
1529 -- plus some extra bindings
1531 -- No need to make it duplicatable if there's only one alternative
1532 prepareCaseCont env [alt] cont = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1533 prepareCaseCont env alts cont = mkDupableCont env cont
1537 mkDupableCont :: SimplEnv -> SimplCont
1538 -> SimplM (FloatsWith (SimplCont, SimplCont))
1540 mkDupableCont env cont
1541 | contIsDupable cont
1542 = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1544 mkDupableCont env (CoerceIt ty cont)
1545 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1546 returnSmpl (floats, (CoerceIt ty dup_cont, nondup_cont))
1548 mkDupableCont env (InlinePlease cont)
1549 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1550 returnSmpl (floats, (InlinePlease dup_cont, nondup_cont))
1552 mkDupableCont env cont@(ArgOf _ arg_ty _ _)
1553 = returnSmpl (emptyFloats env, (mkBoringStop arg_ty, cont))
1554 -- Do *not* duplicate an ArgOf continuation
1555 -- Because ArgOf continuations are opaque, we gain nothing by
1556 -- propagating them into the expressions, and we do lose a lot.
1557 -- Here's an example:
1558 -- && (case x of { T -> F; F -> T }) E
1559 -- Now, && is strict so we end up simplifying the case with
1560 -- an ArgOf continuation. If we let-bind it, we get
1562 -- let $j = \v -> && v E
1563 -- in simplExpr (case x of { T -> F; F -> T })
1564 -- (ArgOf (\r -> $j r)
1565 -- And after simplifying more we get
1567 -- let $j = \v -> && v E
1568 -- in case of { T -> $j F; F -> $j T }
1569 -- Which is a Very Bad Thing
1571 -- The desire not to duplicate is the entire reason that
1572 -- mkDupableCont returns a pair of continuations.
1574 -- The original plan had:
1575 -- e.g. (...strict-fn...) [...hole...]
1577 -- let $j = \a -> ...strict-fn...
1578 -- in $j [...hole...]
1580 mkDupableCont env (ApplyTo _ arg se cont)
1581 = -- e.g. [...hole...] (...arg...)
1583 -- let a = ...arg...
1584 -- in [...hole...] a
1585 simplExpr (setInScope se env) arg `thenSmpl` \ arg' ->
1587 mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1588 addFloats env floats $ \ env ->
1590 if exprIsDupable arg' then
1591 returnSmpl (emptyFloats env, (ApplyTo OkToDup arg' (zapSubstEnv se) dup_cont, nondup_cont))
1593 newId FSLIT("a") (exprType arg') `thenSmpl` \ arg_id ->
1595 tick (CaseOfCase arg_id) `thenSmpl_`
1596 -- Want to tick here so that we go round again,
1597 -- and maybe copy or inline the code.
1598 -- Not strictly CaseOfCase, but never mind
1600 returnSmpl (unitFloat env arg_id arg',
1601 (ApplyTo OkToDup (Var arg_id) (zapSubstEnv se) dup_cont,
1603 -- But what if the arg should be case-bound?
1604 -- This has been this way for a long time, so I'll leave it,
1605 -- but I can't convince myself that it's right.
1608 mkDupableCont env (Select _ case_bndr alts se cont)
1609 = -- e.g. (case [...hole...] of { pi -> ei })
1611 -- let ji = \xij -> ei
1612 -- in case [...hole...] of { pi -> ji xij }
1613 tick (CaseOfCase case_bndr) `thenSmpl_`
1615 alt_env = setInScope se env
1617 prepareCaseCont alt_env alts cont `thenSmpl` \ (floats1, (dup_cont, nondup_cont)) ->
1618 addFloats alt_env floats1 $ \ alt_env ->
1620 simplBinder alt_env case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1621 -- NB: simplBinder does not zap deadness occ-info, so
1622 -- a dead case_bndr' will still advertise its deadness
1623 -- This is really important because in
1624 -- case e of b { (# a,b #) -> ... }
1625 -- b is always dead, and indeed we are not allowed to bind b to (# a,b #),
1626 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1627 -- In the new alts we build, we have the new case binder, so it must retain
1630 mkDupableAlts alt_env case_bndr' alts dup_cont `thenSmpl` \ (floats2, alts') ->
1631 addFloats alt_env floats2 $ \ alt_env ->
1632 returnSmpl (emptyFloats alt_env,
1633 (Select OkToDup case_bndr' alts' (zapSubstEnv se)
1634 (mkBoringStop (contResultType dup_cont)),
1637 mkDupableAlts :: SimplEnv -> OutId -> [InAlt] -> SimplCont
1638 -> SimplM (FloatsWith [InAlt])
1639 -- Absorbs the continuation into the new alternatives
1641 mkDupableAlts env case_bndr' alts dupable_cont
1644 go env [] = returnSmpl (emptyFloats env, [])
1646 = mkDupableAlt env case_bndr' dupable_cont alt `thenSmpl` \ (floats1, alt') ->
1647 addFloats env floats1 $ \ env ->
1648 go env alts `thenSmpl` \ (floats2, alts') ->
1649 returnSmpl (floats2, alt' : alts')
1651 mkDupableAlt env case_bndr' cont alt@(con, bndrs, rhs)
1652 = simplBinders env bndrs `thenSmpl` \ (env, bndrs') ->
1653 simplExprC env rhs cont `thenSmpl` \ rhs' ->
1655 if exprIsDupable rhs' then
1656 returnSmpl (emptyFloats env, (con, bndrs', rhs'))
1657 -- It is worth checking for a small RHS because otherwise we
1658 -- get extra let bindings that may cause an extra iteration of the simplifier to
1659 -- inline back in place. Quite often the rhs is just a variable or constructor.
1660 -- The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1661 -- iterations because the version with the let bindings looked big, and so wasn't
1662 -- inlined, but after the join points had been inlined it looked smaller, and so
1665 -- NB: we have to check the size of rhs', not rhs.
1666 -- Duplicating a small InAlt might invalidate occurrence information
1667 -- However, if it *is* dupable, we return the *un* simplified alternative,
1668 -- because otherwise we'd need to pair it up with an empty subst-env....
1669 -- but we only have one env shared between all the alts.
1670 -- (Remember we must zap the subst-env before re-simplifying something).
1671 -- Rather than do this we simply agree to re-simplify the original (small) thing later.
1675 rhs_ty' = exprType rhs'
1676 used_bndrs' = filter (not . isDeadBinder) (case_bndr' : bndrs')
1677 -- The deadness info on the new binders is unscathed
1679 -- If we try to lift a primitive-typed something out
1680 -- for let-binding-purposes, we will *caseify* it (!),
1681 -- with potentially-disastrous strictness results. So
1682 -- instead we turn it into a function: \v -> e
1683 -- where v::State# RealWorld#. The value passed to this function
1684 -- is realworld#, which generates (almost) no code.
1686 -- There's a slight infelicity here: we pass the overall
1687 -- case_bndr to all the join points if it's used in *any* RHS,
1688 -- because we don't know its usage in each RHS separately
1690 -- We used to say "&& isUnLiftedType rhs_ty'" here, but now
1691 -- we make the join point into a function whenever used_bndrs'
1692 -- is empty. This makes the join-point more CPR friendly.
1693 -- Consider: let j = if .. then I# 3 else I# 4
1694 -- in case .. of { A -> j; B -> j; C -> ... }
1696 -- Now CPR doesn't w/w j because it's a thunk, so
1697 -- that means that the enclosing function can't w/w either,
1698 -- which is a lose. Here's the example that happened in practice:
1699 -- kgmod :: Int -> Int -> Int
1700 -- kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1704 -- I have seen a case alternative like this:
1705 -- True -> \v -> ...
1706 -- It's a bit silly to add the realWorld dummy arg in this case, making
1709 -- (the \v alone is enough to make CPR happy) but I think it's rare
1711 ( if null used_bndrs'
1712 then newId FSLIT("w") realWorldStatePrimTy `thenSmpl` \ rw_id ->
1713 returnSmpl ([rw_id], [Var realWorldPrimId])
1715 returnSmpl (used_bndrs', map varToCoreExpr used_bndrs')
1716 ) `thenSmpl` \ (final_bndrs', final_args) ->
1718 -- See comment about "$j" name above
1719 newId (encodeFS FSLIT("$j")) (mkPiTypes final_bndrs' rhs_ty') `thenSmpl` \ join_bndr ->
1720 -- Notice the funky mkPiTypes. If the contructor has existentials
1721 -- it's possible that the join point will be abstracted over
1722 -- type varaibles as well as term variables.
1723 -- Example: Suppose we have
1724 -- data T = forall t. C [t]
1726 -- case (case e of ...) of
1727 -- C t xs::[t] -> rhs
1728 -- We get the join point
1729 -- let j :: forall t. [t] -> ...
1730 -- j = /\t \xs::[t] -> rhs
1732 -- case (case e of ...) of
1733 -- C t xs::[t] -> j t xs
1735 -- We make the lambdas into one-shot-lambdas. The
1736 -- join point is sure to be applied at most once, and doing so
1737 -- prevents the body of the join point being floated out by
1738 -- the full laziness pass
1739 really_final_bndrs = map one_shot final_bndrs'
1740 one_shot v | isId v = setOneShotLambda v
1742 join_rhs = mkLams really_final_bndrs rhs'
1743 join_call = mkApps (Var join_bndr) final_args
1745 returnSmpl (unitFloat env join_bndr join_rhs, (con, bndrs', join_call))