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, bndr2) ->
312 simplStrictArg AnRhs env rhs rhs_se (idType bndr2) cont_ty $ \ env2 rhs1 ->
314 -- Now complete the binding and simplify the body
315 if needsCaseBinding bndr_ty rhs1
317 thing_inside env2 `thenSmpl` \ (floats, body) ->
318 returnSmpl (emptyFloats env2, Case rhs1 bndr2 (exprType body)
319 [(DEFAULT, [], wrapFloats floats body)])
321 completeNonRecX env2 True {- strict -} bndr bndr2 rhs1 thing_inside
323 | otherwise -- Normal, lazy case
324 = -- Don't use simplBinder because that doesn't keep
325 -- fragile occurrence info in the substitution
326 simplNonRecBndr env bndr `thenSmpl` \ (env, bndr') ->
327 simplLazyBind env NotTopLevel NonRecursive
328 bndr bndr' rhs rhs_se `thenSmpl` \ (floats, env) ->
329 addFloats env floats thing_inside
332 bndr_ty = idType bndr
335 A specialised variant of simplNonRec used when the RHS is already simplified, notably
336 in knownCon. It uses case-binding where necessary.
339 simplNonRecX :: SimplEnv
340 -> InId -- Old binder
341 -> OutExpr -- Simplified RHS
342 -> (SimplEnv -> SimplM FloatsWithExpr)
343 -> SimplM FloatsWithExpr
345 simplNonRecX env bndr new_rhs thing_inside
346 | needsCaseBinding (idType bndr) new_rhs
347 -- Make this test *before* the preInlineUnconditionally
348 -- Consider case I# (quotInt# x y) of
349 -- I# v -> let w = J# v in ...
350 -- If we gaily inline (quotInt# x y) for v, we end up building an
352 -- let w = J# (quotInt# x y) in ...
353 -- because quotInt# can fail.
354 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
355 thing_inside env `thenSmpl` \ (floats, body) ->
356 let body' = wrapFloats floats body in
357 returnSmpl (emptyFloats env, Case new_rhs bndr' (exprType body') [(DEFAULT, [], body')])
359 | preInlineUnconditionally env NotTopLevel bndr new_rhs
360 -- This happens; for example, the case_bndr during case of
361 -- known constructor: case (a,b) of x { (p,q) -> ... }
362 -- Here x isn't mentioned in the RHS, so we don't want to
363 -- create the (dead) let-binding let x = (a,b) in ...
365 -- Similarly, single occurrences can be inlined vigourously
366 -- e.g. case (f x, g y) of (a,b) -> ....
367 -- If a,b occur once we can avoid constructing the let binding for them.
368 = thing_inside (extendIdSubst env bndr (DoneEx new_rhs))
371 = simplBinder env bndr `thenSmpl` \ (env, bndr') ->
372 completeNonRecX env False {- Non-strict; pessimistic -}
373 bndr bndr' new_rhs thing_inside
375 completeNonRecX env is_strict old_bndr new_bndr new_rhs thing_inside
376 = mkAtomicArgs is_strict
377 True {- OK to float unlifted -}
378 new_rhs `thenSmpl` \ (aux_binds, rhs2) ->
380 -- Make the arguments atomic if necessary,
381 -- adding suitable bindings
382 addAtomicBindsE env (fromOL aux_binds) $ \ env ->
383 completeLazyBind env NotTopLevel
384 old_bndr new_bndr rhs2 `thenSmpl` \ (floats, env) ->
385 addFloats env floats thing_inside
389 %************************************************************************
391 \subsection{Lazy bindings}
393 %************************************************************************
395 simplRecBind is used for
396 * recursive bindings only
399 simplRecBind :: SimplEnv -> TopLevelFlag
400 -> [(InId, InExpr)] -> [OutId]
401 -> SimplM (FloatsWith SimplEnv)
402 simplRecBind env top_lvl pairs bndrs'
403 = go env pairs bndrs' `thenSmpl` \ (floats, env) ->
404 returnSmpl (flattenFloats floats, env)
406 go env [] _ = returnSmpl (emptyFloats env, env)
408 go env ((bndr, rhs) : pairs) (bndr' : bndrs')
409 = simplRecOrTopPair env top_lvl bndr bndr' rhs `thenSmpl` \ (floats, env) ->
410 addFloats env floats (\env -> go env pairs bndrs')
414 simplRecOrTopPair is used for
415 * recursive bindings (whether top level or not)
416 * top-level non-recursive bindings
418 It assumes the binder has already been simplified, but not its IdInfo.
421 simplRecOrTopPair :: SimplEnv
423 -> InId -> OutId -- Binder, both pre-and post simpl
424 -> InExpr -- The RHS and its environment
425 -> SimplM (FloatsWith SimplEnv)
427 simplRecOrTopPair env top_lvl bndr bndr' rhs
428 | preInlineUnconditionally env top_lvl bndr rhs -- Check for unconditional inline
429 = tick (PreInlineUnconditionally bndr) `thenSmpl_`
430 returnSmpl (emptyFloats env, extendIdSubst env bndr (mkContEx env rhs))
433 = simplLazyBind env top_lvl Recursive bndr bndr' rhs env
434 -- May not actually be recursive, but it doesn't matter
438 simplLazyBind is used for
439 * recursive bindings (whether top level or not)
440 * top-level non-recursive bindings
441 * non-top-level *lazy* non-recursive bindings
443 [Thus it deals with the lazy cases from simplNonRecBind, and all cases
444 from SimplRecOrTopBind]
447 1. It assumes that the binder is *already* simplified,
448 and is in scope, but not its IdInfo
450 2. It assumes that the binder type is lifted.
452 3. It does not check for pre-inline-unconditionallly;
453 that should have been done already.
456 simplLazyBind :: SimplEnv
457 -> TopLevelFlag -> RecFlag
458 -> InId -> OutId -- Binder, both pre-and post simpl
459 -> InExpr -> SimplEnv -- The RHS and its environment
460 -> SimplM (FloatsWith SimplEnv)
462 simplLazyBind env top_lvl is_rec bndr bndr2 rhs rhs_se
464 rhs_env = setInScope rhs_se env
465 is_top_level = isTopLevel top_lvl
466 ok_float_unlifted = not is_top_level && isNonRec is_rec
467 rhs_cont = mkRhsStop (idType bndr2)
469 -- Simplify the RHS; note the mkRhsStop, which tells
470 -- the simplifier that this is the RHS of a let.
471 simplExprF rhs_env rhs rhs_cont `thenSmpl` \ (floats, rhs1) ->
473 -- If any of the floats can't be floated, give up now
474 -- (The allLifted predicate says True for empty floats.)
475 if (not ok_float_unlifted && not (allLifted floats)) then
476 completeLazyBind env top_lvl bndr bndr2
477 (wrapFloats floats rhs1)
480 -- ANF-ise a constructor or PAP rhs
481 mkAtomicArgs False {- Not strict -}
482 ok_float_unlifted rhs1 `thenSmpl` \ (aux_binds, rhs2) ->
484 -- If the result is a PAP, float the floats out, else wrap them
485 -- By this time it's already been ANF-ised (if necessary)
486 if isEmptyFloats floats && isNilOL aux_binds then -- Shortcut a common case
487 completeLazyBind env top_lvl bndr bndr2 rhs2
489 else if is_top_level || exprIsTrivial rhs2 || exprIsHNF rhs2 then
490 -- WARNING: long dodgy argument coming up
491 -- WANTED: a better way to do this
493 -- We can't use "exprIsCheap" instead of exprIsHNF,
494 -- because that causes a strictness bug.
495 -- x = let y* = E in case (scc y) of { T -> F; F -> T}
496 -- The case expression is 'cheap', but it's wrong to transform to
497 -- y* = E; x = case (scc y) of {...}
498 -- Either we must be careful not to float demanded non-values, or
499 -- we must use exprIsHNF for the test, which ensures that the
500 -- thing is non-strict. So exprIsHNF => bindings are non-strict
501 -- I think. The WARN below tests for this.
503 -- We use exprIsTrivial here because we want to reveal lone variables.
504 -- E.g. let { x = letrec { y = E } in y } in ...
505 -- Here we definitely want to float the y=E defn.
506 -- exprIsHNF definitely isn't right for that.
508 -- Again, the floated binding can't be strict; if it's recursive it'll
509 -- be non-strict; if it's non-recursive it'd be inlined.
511 -- Note [SCC-and-exprIsTrivial]
513 -- y = let { x* = E } in scc "foo" x
514 -- then we do *not* want to float out the x binding, because
515 -- it's strict! Fortunately, exprIsTrivial replies False to
518 -- There's a subtlety here. There may be a binding (x* = e) in the
519 -- floats, where the '*' means 'will be demanded'. So is it safe
520 -- to float it out? Answer no, but it won't matter because
521 -- we only float if (a) arg' is a WHNF, or (b) it's going to top level
522 -- and so there can't be any 'will be demanded' bindings in the floats.
524 ASSERT2( is_top_level || not (any demanded_float (floatBinds floats)),
525 ppr (filter demanded_float (floatBinds floats)) )
527 tick LetFloatFromLet `thenSmpl_` (
528 addFloats env floats $ \ env2 ->
529 addAtomicBinds env2 (fromOL aux_binds) $ \ env3 ->
530 completeLazyBind env3 top_lvl bndr bndr2 rhs2)
533 completeLazyBind env top_lvl bndr bndr2 (wrapFloats floats rhs1)
536 demanded_float (NonRec b r) = isStrictDmd (idNewDemandInfo b) && not (isUnLiftedType (idType b))
537 -- Unlifted-type (cheap-eagerness) lets may well have a demanded flag on them
538 demanded_float (Rec _) = False
543 %************************************************************************
545 \subsection{Completing a lazy binding}
547 %************************************************************************
550 * deals only with Ids, not TyVars
551 * takes an already-simplified binder and RHS
552 * is used for both recursive and non-recursive bindings
553 * is used for both top-level and non-top-level bindings
555 It does the following:
556 - tries discarding a dead binding
557 - tries PostInlineUnconditionally
558 - add unfolding [this is the only place we add an unfolding]
561 It does *not* attempt to do let-to-case. Why? Because it is used for
562 - top-level bindings (when let-to-case is impossible)
563 - many situations where the "rhs" is known to be a WHNF
564 (so let-to-case is inappropriate).
567 completeLazyBind :: SimplEnv
568 -> TopLevelFlag -- Flag stuck into unfolding
569 -> InId -- Old binder
570 -> OutId -- New binder
571 -> OutExpr -- Simplified RHS
572 -> SimplM (FloatsWith SimplEnv)
573 -- We return a new SimplEnv, because completeLazyBind may choose to do its work
574 -- by extending the substitution (e.g. let x = y in ...)
575 -- The new binding (if any) is returned as part of the floats.
576 -- NB: the returned SimplEnv has the right SubstEnv, but you should
577 -- (as usual) use the in-scope-env from the floats
579 completeLazyBind env top_lvl old_bndr new_bndr new_rhs
580 | postInlineUnconditionally env top_lvl new_bndr occ_info new_rhs unfolding
581 = -- Drop the binding
582 tick (PostInlineUnconditionally old_bndr) `thenSmpl_`
583 returnSmpl (emptyFloats env, extendIdSubst env old_bndr (DoneEx new_rhs))
584 -- Use the substitution to make quite, quite sure that the substitution
585 -- will happen, since we are going to discard the binding
590 new_bndr_info = idInfo new_bndr `setArityInfo` exprArity new_rhs
592 -- Add the unfolding *only* for non-loop-breakers
593 -- Making loop breakers not have an unfolding at all
594 -- means that we can avoid tests in exprIsConApp, for example.
595 -- This is important: if exprIsConApp says 'yes' for a recursive
596 -- thing, then we can get into an infinite loop
598 -- If the unfolding is a value, the demand info may
599 -- go pear-shaped, so we nuke it. Example:
601 -- case x of (p,q) -> h p q x
602 -- Here x is certainly demanded. But after we've nuked
603 -- the case, we'll get just
604 -- let x = (a,b) in h a b x
605 -- and now x is not demanded (I'm assuming h is lazy)
606 -- This really happens. Similarly
607 -- let f = \x -> e in ...f..f...
608 -- After inling f at some of its call sites the original binding may
609 -- (for example) be no longer strictly demanded.
610 -- The solution here is a bit ad hoc...
611 info_w_unf = new_bndr_info `setUnfoldingInfo` unfolding
612 final_info | loop_breaker = new_bndr_info
613 | isEvaldUnfolding unfolding = zapDemandInfo info_w_unf `orElse` info_w_unf
614 | otherwise = info_w_unf
616 final_id = new_bndr `setIdInfo` final_info
618 -- These seqs forces the Id, and hence its IdInfo,
619 -- and hence any inner substitutions
621 returnSmpl (unitFloat env final_id new_rhs, env)
624 unfolding = mkUnfolding (isTopLevel top_lvl) new_rhs
625 loop_breaker = isLoopBreaker occ_info
626 old_info = idInfo old_bndr
627 occ_info = occInfo old_info
632 %************************************************************************
634 \subsection[Simplify-simplExpr]{The main function: simplExpr}
636 %************************************************************************
638 The reason for this OutExprStuff stuff is that we want to float *after*
639 simplifying a RHS, not before. If we do so naively we get quadratic
640 behaviour as things float out.
642 To see why it's important to do it after, consider this (real) example:
656 a -- Can't inline a this round, cos it appears twice
660 Each of the ==> steps is a round of simplification. We'd save a
661 whole round if we float first. This can cascade. Consider
666 let f = let d1 = ..d.. in \y -> e
670 in \x -> ...(\y ->e)...
672 Only in this second round can the \y be applied, and it
673 might do the same again.
677 simplExpr :: SimplEnv -> CoreExpr -> SimplM CoreExpr
678 simplExpr env expr = simplExprC env expr (mkBoringStop expr_ty')
680 expr_ty' = substTy env (exprType expr)
681 -- The type in the Stop continuation, expr_ty', is usually not used
682 -- It's only needed when discarding continuations after finding
683 -- a function that returns bottom.
684 -- Hence the lazy substitution
687 simplExprC :: SimplEnv -> CoreExpr -> SimplCont -> SimplM CoreExpr
688 -- Simplify an expression, given a continuation
689 simplExprC env expr cont
690 = simplExprF env expr cont `thenSmpl` \ (floats, expr) ->
691 returnSmpl (wrapFloats floats expr)
693 simplExprF :: SimplEnv -> InExpr -> SimplCont -> SimplM FloatsWithExpr
694 -- Simplify an expression, returning floated binds
696 simplExprF env (Var v) cont = simplVar env v cont
697 simplExprF env (Lit lit) cont = rebuild env (Lit lit) cont
698 simplExprF env expr@(Lam _ _) cont = simplLam env expr cont
699 simplExprF env (Note note expr) cont = simplNote env note expr cont
700 simplExprF env (App fun arg) cont = simplExprF env fun (ApplyTo NoDup arg env cont)
702 simplExprF env (Type ty) cont
703 = ASSERT( contIsRhsOrArg cont )
704 simplType env ty `thenSmpl` \ ty' ->
705 rebuild env (Type ty') cont
707 simplExprF env (Case scrut bndr case_ty alts) cont
708 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
709 = -- Simplify the scrutinee with a Select continuation
710 simplExprF env scrut (Select NoDup bndr alts env cont)
713 = -- If case-of-case is off, simply simplify the case expression
714 -- in a vanilla Stop context, and rebuild the result around it
715 simplExprC env scrut case_cont `thenSmpl` \ case_expr' ->
716 rebuild env case_expr' cont
718 case_cont = Select NoDup bndr alts env (mkBoringStop case_ty')
719 case_ty' = substTy env case_ty -- c.f. defn of simplExpr
721 simplExprF env (Let (Rec pairs) body) cont
722 = simplRecBndrs env (map fst pairs) `thenSmpl` \ (env, bndrs') ->
723 -- NB: bndrs' don't have unfoldings or rules
724 -- We add them as we go down
726 simplRecBind env NotTopLevel pairs bndrs' `thenSmpl` \ (floats, env) ->
727 addFloats env floats $ \ env ->
728 simplExprF env body cont
730 -- A non-recursive let is dealt with by simplNonRecBind
731 simplExprF env (Let (NonRec bndr rhs) body) cont
732 = simplNonRecBind env bndr rhs env (contResultType cont) $ \ env ->
733 simplExprF env body cont
736 ---------------------------------
737 simplType :: SimplEnv -> InType -> SimplM OutType
738 -- Kept monadic just so we can do the seqType
740 = seqType new_ty `seq` returnSmpl new_ty
742 new_ty = substTy env ty
746 %************************************************************************
750 %************************************************************************
753 simplLam env fun cont
756 zap_it = mkLamBndrZapper fun (countArgs cont)
757 cont_ty = contResultType cont
759 -- Type-beta reduction
760 go env (Lam bndr body) (ApplyTo _ (Type ty_arg) arg_se body_cont)
761 = ASSERT( isTyVar bndr )
762 tick (BetaReduction bndr) `thenSmpl_`
763 simplType (setInScope arg_se env) ty_arg `thenSmpl` \ ty_arg' ->
764 go (extendTvSubst env bndr ty_arg') body body_cont
766 -- Ordinary beta reduction
767 go env (Lam bndr body) cont@(ApplyTo _ arg arg_se body_cont)
768 = tick (BetaReduction bndr) `thenSmpl_`
769 simplNonRecBind env (zap_it bndr) arg arg_se cont_ty $ \ env ->
770 go env body body_cont
772 -- Not enough args, so there are real lambdas left to put in the result
773 go env lam@(Lam _ _) cont
774 = simplLamBndrs env bndrs `thenSmpl` \ (env, bndrs') ->
775 simplExpr env body `thenSmpl` \ body' ->
776 mkLam env bndrs' body' cont `thenSmpl` \ (floats, new_lam) ->
777 addFloats env floats $ \ env ->
778 rebuild env new_lam cont
780 (bndrs,body) = collectBinders lam
782 -- Exactly enough args
783 go env expr cont = simplExprF env expr cont
785 mkLamBndrZapper :: CoreExpr -- Function
786 -> Int -- Number of args supplied, *including* type args
787 -> Id -> Id -- Use this to zap the binders
788 mkLamBndrZapper fun n_args
789 | n_args >= n_params fun = \b -> b -- Enough args
790 | otherwise = \b -> zapLamIdInfo b
792 -- NB: we count all the args incl type args
793 -- so we must count all the binders (incl type lambdas)
794 n_params (Note _ e) = n_params e
795 n_params (Lam b e) = 1 + n_params e
796 n_params other = 0::Int
800 %************************************************************************
804 %************************************************************************
807 simplNote env (Coerce to from) body cont
809 addCoerce s1 k1 cont -- Drop redundant coerces. This can happen if a polymoprhic
810 -- (coerce a b e) is instantiated with a=ty1 b=ty2 and the
811 -- two are the same. This happens a lot in Happy-generated parsers
812 | s1 `coreEqType` k1 = cont
814 addCoerce s1 k1 (CoerceIt t1 cont)
815 -- coerce T1 S1 (coerce S1 K1 e)
818 -- coerce T1 K1 e, otherwise
820 -- For example, in the initial form of a worker
821 -- we may find (coerce T (coerce S (\x.e))) y
822 -- and we'd like it to simplify to e[y/x] in one round
824 | t1 `coreEqType` k1 = cont -- The coerces cancel out
825 | otherwise = CoerceIt t1 cont -- They don't cancel, but
826 -- the inner one is redundant
828 addCoerce t1t2 s1s2 (ApplyTo dup arg arg_se cont)
829 | not (isTypeArg arg), -- This whole case only works for value args
830 -- Could upgrade to have equiv thing for type apps too
831 Just (s1, s2) <- splitFunTy_maybe s1s2
832 -- (coerce (T1->T2) (S1->S2) F) E
834 -- coerce T2 S2 (F (coerce S1 T1 E))
836 -- t1t2 must be a function type, T1->T2, because it's applied to something
837 -- but s1s2 might conceivably not be
839 -- When we build the ApplyTo we can't mix the out-types
840 -- with the InExpr in the argument, so we simply substitute
841 -- to make it all consistent. It's a bit messy.
842 -- But it isn't a common case.
844 (t1,t2) = splitFunTy t1t2
845 new_arg = mkCoerce2 s1 t1 (substExpr arg_env arg)
846 arg_env = setInScope arg_se env
848 ApplyTo dup new_arg (zapSubstEnv env) (addCoerce t2 s2 cont)
850 addCoerce to' _ cont = CoerceIt to' cont
852 simplType env to `thenSmpl` \ to' ->
853 simplType env from `thenSmpl` \ from' ->
854 simplExprF env body (addCoerce to' from' cont)
857 -- Hack: we only distinguish subsumed cost centre stacks for the purposes of
858 -- inlining. All other CCCSs are mapped to currentCCS.
859 simplNote env (SCC cc) e cont
860 = simplExpr (setEnclosingCC env currentCCS) e `thenSmpl` \ e' ->
861 rebuild env (mkSCC cc e') cont
863 simplNote env InlineCall e cont
864 = simplExprF env e (InlinePlease cont)
866 -- See notes with SimplMonad.inlineMode
867 simplNote env InlineMe e cont
868 | contIsRhsOrArg cont -- Totally boring continuation; see notes above
869 = -- Don't inline inside an INLINE expression
870 simplExpr (setMode inlineMode env ) e `thenSmpl` \ e' ->
871 rebuild env (mkInlineMe e') cont
873 | otherwise -- Dissolve the InlineMe note if there's
874 -- an interesting context of any kind to combine with
875 -- (even a type application -- anything except Stop)
876 = simplExprF env e cont
878 simplNote env (CoreNote s) e cont
879 = simplExpr env e `thenSmpl` \ e' ->
880 rebuild env (Note (CoreNote s) e') cont
884 %************************************************************************
886 \subsection{Dealing with calls}
888 %************************************************************************
891 simplVar env var cont
892 = case substId env var of
893 DoneEx e -> simplExprF (zapSubstEnv env) e cont
894 ContEx tvs ids e -> simplExprF (setSubstEnv env tvs ids) e cont
895 DoneId var1 occ -> completeCall (zapSubstEnv env) var1 occ cont
896 -- Note [zapSubstEnv]
897 -- The template is already simplified, so don't re-substitute.
898 -- This is VITAL. Consider
900 -- let y = \z -> ...x... in
902 -- We'll clone the inner \x, adding x->x' in the id_subst
903 -- Then when we inline y, we must *not* replace x by x' in
904 -- the inlined copy!!
906 ---------------------------------------------------------
907 -- Dealing with a call site
909 completeCall env var occ_info cont
910 = -- Simplify the arguments
911 getDOptsSmpl `thenSmpl` \ dflags ->
913 chkr = getSwitchChecker env
914 (args, call_cont, inline_call) = getContArgs chkr var cont
917 simplifyArgs env fn_ty args (contResultType call_cont) $ \ env args ->
919 -- Next, look for rules or specialisations that match
921 -- It's important to simplify the args first, because the rule-matcher
922 -- doesn't do substitution as it goes. We don't want to use subst_args
923 -- (defined in the 'where') because that throws away useful occurrence info,
924 -- and perhaps-very-important specialisations.
926 -- Some functions have specialisations *and* are strict; in this case,
927 -- we don't want to inline the wrapper of the non-specialised thing; better
928 -- to call the specialised thing instead.
929 -- We used to use the black-listing mechanism to ensure that inlining of
930 -- the wrapper didn't occur for things that have specialisations till a
931 -- later phase, so but now we just try RULES first
933 -- You might think that we shouldn't apply rules for a loop breaker:
934 -- doing so might give rise to an infinite loop, because a RULE is
935 -- rather like an extra equation for the function:
936 -- RULE: f (g x) y = x+y
939 -- But it's too drastic to disable rules for loop breakers.
940 -- Even the foldr/build rule would be disabled, because foldr
941 -- is recursive, and hence a loop breaker:
942 -- foldr k z (build g) = g k z
943 -- So it's up to the programmer: rules can cause divergence
946 in_scope = getInScope env
948 maybe_rule = case activeRule env of
949 Nothing -> Nothing -- No rules apply
950 Just act_fn -> lookupRule act_fn in_scope rules var args
953 Just (rule_name, rule_rhs) ->
954 tick (RuleFired rule_name) `thenSmpl_`
955 (if dopt Opt_D_dump_inlinings dflags then
956 pprTrace "Rule fired" (vcat [
957 text "Rule:" <+> ftext rule_name,
958 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
959 text "After: " <+> pprCoreExpr rule_rhs,
960 text "Cont: " <+> ppr call_cont])
963 simplExprF env rule_rhs call_cont ;
965 Nothing -> -- No rules
967 -- Next, look for an inlining
969 arg_infos = [ interestingArg arg | arg <- args, isValArg arg]
971 interesting_cont = interestingCallContext (notNull args)
975 active_inline = activeInline env var occ_info
976 maybe_inline = callSiteInline dflags active_inline inline_call occ_info
977 var arg_infos interesting_cont
979 case maybe_inline of {
980 Just unfolding -- There is an inlining!
981 -> tick (UnfoldingDone var) `thenSmpl_`
982 (if dopt Opt_D_dump_inlinings dflags then
983 pprTrace "Inlining done" (vcat [
984 text "Before:" <+> ppr var <+> sep (map pprParendExpr args),
985 text "Inlined fn: " <+> ppr unfolding,
986 text "Cont: " <+> ppr call_cont])
989 makeThatCall env var unfolding args call_cont
992 Nothing -> -- No inlining!
995 rebuild env (mkApps (Var var) args) call_cont
998 makeThatCall :: SimplEnv
1000 -> InExpr -- Inlined function rhs
1001 -> [OutExpr] -- Arguments, already simplified
1002 -> SimplCont -- After the call
1003 -> SimplM FloatsWithExpr
1004 -- Similar to simplLam, but this time
1005 -- the arguments are already simplified
1006 makeThatCall orig_env var fun@(Lam _ _) args cont
1007 = go orig_env fun args
1009 zap_it = mkLamBndrZapper fun (length args)
1011 -- Type-beta reduction
1012 go env (Lam bndr body) (Type ty_arg : args)
1013 = ASSERT( isTyVar bndr )
1014 tick (BetaReduction bndr) `thenSmpl_`
1015 go (extendTvSubst env bndr ty_arg) body args
1017 -- Ordinary beta reduction
1018 go env (Lam bndr body) (arg : args)
1019 = tick (BetaReduction bndr) `thenSmpl_`
1020 simplNonRecX env (zap_it bndr) arg $ \ env ->
1023 -- Not enough args, so there are real lambdas left to put in the result
1025 = simplExprF env fun (pushContArgs orig_env args cont)
1026 -- NB: orig_env; the correct environment to capture with
1027 -- the arguments.... env has been augmented with substitutions
1028 -- from the beta reductions.
1030 makeThatCall env var fun args cont
1031 = simplExprF env fun (pushContArgs env args cont)
1035 %************************************************************************
1037 \subsection{Arguments}
1039 %************************************************************************
1042 ---------------------------------------------------------
1043 -- Simplifying the arguments of a call
1045 simplifyArgs :: SimplEnv
1046 -> OutType -- Type of the function
1047 -> [(InExpr, SimplEnv, Bool)] -- Details of the arguments
1048 -> OutType -- Type of the continuation
1049 -> (SimplEnv -> [OutExpr] -> SimplM FloatsWithExpr)
1050 -> SimplM FloatsWithExpr
1052 -- [CPS-like because of strict arguments]
1054 -- Simplify the arguments to a call.
1055 -- This part of the simplifier may break the no-shadowing invariant
1057 -- f (...(\a -> e)...) (case y of (a,b) -> e')
1058 -- where f is strict in its second arg
1059 -- If we simplify the innermost one first we get (...(\a -> e)...)
1060 -- Simplifying the second arg makes us float the case out, so we end up with
1061 -- case y of (a,b) -> f (...(\a -> e)...) e'
1062 -- So the output does not have the no-shadowing invariant. However, there is
1063 -- no danger of getting name-capture, because when the first arg was simplified
1064 -- we used an in-scope set that at least mentioned all the variables free in its
1065 -- static environment, and that is enough.
1067 -- We can't just do innermost first, or we'd end up with a dual problem:
1068 -- case x of (a,b) -> f e (...(\a -> e')...)
1070 -- I spent hours trying to recover the no-shadowing invariant, but I just could
1071 -- not think of an elegant way to do it. The simplifier is already knee-deep in
1072 -- continuations. We have to keep the right in-scope set around; AND we have
1073 -- to get the effect that finding (error "foo") in a strict arg position will
1074 -- discard the entire application and replace it with (error "foo"). Getting
1075 -- all this at once is TOO HARD!
1077 simplifyArgs env fn_ty args cont_ty thing_inside
1078 = go env fn_ty args thing_inside
1080 go env fn_ty [] thing_inside = thing_inside env []
1081 go env fn_ty (arg:args) thing_inside = simplifyArg env fn_ty arg cont_ty $ \ env arg' ->
1082 go env (applyTypeToArg fn_ty arg') args $ \ env args' ->
1083 thing_inside env (arg':args')
1085 simplifyArg env fn_ty (Type ty_arg, se, _) cont_ty thing_inside
1086 = simplType (setInScope se env) ty_arg `thenSmpl` \ new_ty_arg ->
1087 thing_inside env (Type new_ty_arg)
1089 simplifyArg env fn_ty (val_arg, arg_se, is_strict) cont_ty thing_inside
1091 = simplStrictArg AnArg env val_arg arg_se arg_ty cont_ty thing_inside
1093 | otherwise -- Lazy argument
1094 -- DO NOT float anything outside, hence simplExprC
1095 -- There is no benefit (unlike in a let-binding), and we'd
1096 -- have to be very careful about bogus strictness through
1097 -- floating a demanded let.
1098 = simplExprC (setInScope arg_se env) val_arg
1099 (mkBoringStop arg_ty) `thenSmpl` \ arg1 ->
1100 thing_inside env arg1
1102 arg_ty = funArgTy fn_ty
1105 simplStrictArg :: LetRhsFlag
1106 -> SimplEnv -- The env of the call
1107 -> InExpr -> SimplEnv -- The arg plus its env
1108 -> OutType -- arg_ty: type of the argument
1109 -> OutType -- cont_ty: Type of thing computed by the context
1110 -> (SimplEnv -> OutExpr -> SimplM FloatsWithExpr)
1111 -- Takes an expression of type rhs_ty,
1112 -- returns an expression of type cont_ty
1113 -- The env passed to this continuation is the
1114 -- env of the call, plus any new in-scope variables
1115 -> SimplM FloatsWithExpr -- An expression of type cont_ty
1117 simplStrictArg is_rhs call_env arg arg_env arg_ty cont_ty thing_inside
1118 = simplExprF (setInScope arg_env call_env) arg
1119 (ArgOf is_rhs arg_ty cont_ty (\ new_env -> thing_inside (setInScope call_env new_env)))
1120 -- Notice the way we use arg_env (augmented with in-scope vars from call_env)
1121 -- to simplify the argument
1122 -- and call-env (augmented with in-scope vars from the arg) to pass to the continuation
1126 %************************************************************************
1128 \subsection{mkAtomicArgs}
1130 %************************************************************************
1132 mkAtomicArgs takes a putative RHS, checks whether it's a PAP or
1133 constructor application and, if so, converts it to ANF, so that the
1134 resulting thing can be inlined more easily. Thus
1141 There are three sorts of binding context, specified by the two
1147 N N Top-level or recursive Only bind args of lifted type
1149 N Y Non-top-level and non-recursive, Bind args of lifted type, or
1150 but lazy unlifted-and-ok-for-speculation
1152 Y Y Non-top-level, non-recursive, Bind all args
1153 and strict (demanded)
1160 there is no point in transforming to
1162 x = case (y div# z) of r -> MkC r
1164 because the (y div# z) can't float out of the let. But if it was
1165 a *strict* let, then it would be a good thing to do. Hence the
1166 context information.
1169 mkAtomicArgs :: Bool -- A strict binding
1170 -> Bool -- OK to float unlifted args
1172 -> SimplM (OrdList (OutId,OutExpr), -- The floats (unusually) may include
1173 OutExpr) -- things that need case-binding,
1174 -- if the strict-binding flag is on
1176 mkAtomicArgs is_strict ok_float_unlifted rhs
1177 | (Var fun, args) <- collectArgs rhs, -- It's an application
1178 isDataConWorkId fun || valArgCount args < idArity fun -- And it's a constructor or PAP
1179 = go fun nilOL [] args -- Have a go
1181 | otherwise = bale_out -- Give up
1184 bale_out = returnSmpl (nilOL, rhs)
1186 go fun binds rev_args []
1187 = returnSmpl (binds, mkApps (Var fun) (reverse rev_args))
1189 go fun binds rev_args (arg : args)
1190 | exprIsTrivial arg -- Easy case
1191 = go fun binds (arg:rev_args) args
1193 | not can_float_arg -- Can't make this arg atomic
1194 = bale_out -- ... so give up
1196 | otherwise -- Don't forget to do it recursively
1197 -- E.g. x = a:b:c:[]
1198 = mkAtomicArgs is_strict ok_float_unlifted arg `thenSmpl` \ (arg_binds, arg') ->
1199 newId FSLIT("a") arg_ty `thenSmpl` \ arg_id ->
1200 go fun ((arg_binds `snocOL` (arg_id,arg')) `appOL` binds)
1201 (Var arg_id : rev_args) args
1203 arg_ty = exprType arg
1204 can_float_arg = is_strict
1205 || not (isUnLiftedType arg_ty)
1206 || (ok_float_unlifted && exprOkForSpeculation arg)
1209 addAtomicBinds :: SimplEnv -> [(OutId,OutExpr)]
1210 -> (SimplEnv -> SimplM (FloatsWith a))
1211 -> SimplM (FloatsWith a)
1212 addAtomicBinds env [] thing_inside = thing_inside env
1213 addAtomicBinds env ((v,r):bs) thing_inside = addAuxiliaryBind env (NonRec v r) $ \ env ->
1214 addAtomicBinds env bs thing_inside
1216 addAtomicBindsE :: SimplEnv -> [(OutId,OutExpr)]
1217 -> (SimplEnv -> SimplM FloatsWithExpr)
1218 -> SimplM FloatsWithExpr
1219 -- Same again, but this time we're in an expression context,
1220 -- and may need to do some case bindings
1222 addAtomicBindsE env [] thing_inside
1224 addAtomicBindsE env ((v,r):bs) thing_inside
1225 | needsCaseBinding (idType v) r
1226 = addAtomicBindsE (addNewInScopeIds env [v]) bs thing_inside `thenSmpl` \ (floats, expr) ->
1227 WARN( exprIsTrivial expr, ppr v <+> pprCoreExpr expr )
1228 (let body = wrapFloats floats expr in
1229 returnSmpl (emptyFloats env, Case r v (exprType body) [(DEFAULT,[],body)]))
1232 = addAuxiliaryBind env (NonRec v r) $ \ env ->
1233 addAtomicBindsE env bs thing_inside
1237 %************************************************************************
1239 \subsection{The main rebuilder}
1241 %************************************************************************
1244 rebuild :: SimplEnv -> OutExpr -> SimplCont -> SimplM FloatsWithExpr
1246 rebuild env expr (Stop _ _ _) = rebuildDone env expr
1247 rebuild env expr (ArgOf _ _ _ cont_fn) = cont_fn env expr
1248 rebuild env expr (CoerceIt to_ty cont) = rebuild env (mkCoerce to_ty expr) cont
1249 rebuild env expr (InlinePlease cont) = rebuild env (Note InlineCall expr) cont
1250 rebuild env expr (Select _ bndr alts se cont) = rebuildCase (setInScope se env) expr bndr alts cont
1251 rebuild env expr (ApplyTo _ arg se cont) = rebuildApp (setInScope se env) expr arg cont
1253 rebuildApp env fun arg cont
1254 = simplExpr env arg `thenSmpl` \ arg' ->
1255 rebuild env (App fun arg') cont
1257 rebuildDone env expr = returnSmpl (emptyFloats env, expr)
1261 %************************************************************************
1263 \subsection{Functions dealing with a case}
1265 %************************************************************************
1267 Blob of helper functions for the "case-of-something-else" situation.
1270 ---------------------------------------------------------
1271 -- Eliminate the case if possible
1273 rebuildCase :: SimplEnv
1274 -> OutExpr -- Scrutinee
1275 -> InId -- Case binder
1276 -> [InAlt] -- Alternatives (inceasing order)
1278 -> SimplM FloatsWithExpr
1280 rebuildCase env scrut case_bndr alts cont
1281 | Just (con,args) <- exprIsConApp_maybe scrut
1282 -- Works when the scrutinee is a variable with a known unfolding
1283 -- as well as when it's an explicit constructor application
1284 = knownCon env (DataAlt con) args case_bndr alts cont
1286 | Lit lit <- scrut -- No need for same treatment as constructors
1287 -- because literals are inlined more vigorously
1288 = knownCon env (LitAlt lit) [] case_bndr alts cont
1291 = -- Prepare the alternatives.
1292 prepareAlts scrut case_bndr alts `thenSmpl` \ (better_alts, handled_cons) ->
1294 -- Prepare the continuation;
1295 -- The new subst_env is in place
1296 prepareCaseCont env better_alts cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1297 addFloats env floats $ \ env ->
1300 -- The case expression is annotated with the result type of the continuation
1301 -- This may differ from the type originally on the case. For example
1302 -- case(T) (case(Int#) a of { True -> 1#; False -> 0# }) of
1305 -- let j a# = <blob>
1306 -- in case(T) a of { True -> j 1#; False -> j 0# }
1307 -- Note that the case that scrutinises a now returns a T not an Int#
1308 res_ty' = contResultType dup_cont
1311 -- Deal with case binder
1312 simplCaseBinder env scrut case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1314 -- Deal with the case alternatives
1315 simplAlts alt_env handled_cons
1316 case_bndr' better_alts dup_cont `thenSmpl` \ alts' ->
1318 -- Put the case back together
1319 mkCase scrut case_bndr' res_ty' alts' `thenSmpl` \ case_expr ->
1321 -- Notice that rebuildDone returns the in-scope set from env, not alt_env
1322 -- The case binder *not* scope over the whole returned case-expression
1323 rebuild env case_expr nondup_cont
1326 simplCaseBinder checks whether the scrutinee is a variable, v. If so,
1327 try to eliminate uses of v in the RHSs in favour of case_bndr; that
1328 way, there's a chance that v will now only be used once, and hence
1333 There is a time we *don't* want to do that, namely when
1334 -fno-case-of-case is on. This happens in the first simplifier pass,
1335 and enhances full laziness. Here's the bad case:
1336 f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
1337 If we eliminate the inner case, we trap it inside the I# v -> arm,
1338 which might prevent some full laziness happening. I've seen this
1339 in action in spectral/cichelli/Prog.hs:
1340 [(m,n) | m <- [1..max], n <- [1..max]]
1341 Hence the check for NoCaseOfCase.
1345 There is another situation when we don't want to do it. If we have
1347 case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
1348 ...other cases .... }
1350 We'll perform the binder-swap for the outer case, giving
1352 case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
1353 ...other cases .... }
1355 But there is no point in doing it for the inner case, because w1 can't
1356 be inlined anyway. Furthermore, doing the case-swapping involves
1357 zapping w2's occurrence info (see paragraphs that follow), and that
1358 forces us to bind w2 when doing case merging. So we get
1360 case x of w1 { A -> let w2 = w1 in e1
1361 B -> let w2 = w1 in e2
1362 ...other cases .... }
1364 This is plain silly in the common case where w2 is dead.
1366 Even so, I can't see a good way to implement this idea. I tried
1367 not doing the binder-swap if the scrutinee was already evaluated
1368 but that failed big-time:
1372 case v of w { MkT x ->
1373 case x of x1 { I# y1 ->
1374 case x of x2 { I# y2 -> ...
1376 Notice that because MkT is strict, x is marked "evaluated". But to
1377 eliminate the last case, we must either make sure that x (as well as
1378 x1) has unfolding MkT y1. THe straightforward thing to do is to do
1379 the binder-swap. So this whole note is a no-op.
1383 If we replace the scrutinee, v, by tbe case binder, then we have to nuke
1384 any occurrence info (eg IAmDead) in the case binder, because the
1385 case-binder now effectively occurs whenever v does. AND we have to do
1386 the same for the pattern-bound variables! Example:
1388 (case x of { (a,b) -> a }) (case x of { (p,q) -> q })
1390 Here, b and p are dead. But when we move the argment inside the first
1391 case RHS, and eliminate the second case, we get
1393 case x of { (a,b) -> a b }
1395 Urk! b is alive! Reason: the scrutinee was a variable, and case elimination
1398 Indeed, this can happen anytime the case binder isn't dead:
1399 case <any> of x { (a,b) ->
1400 case x of { (p,q) -> p } }
1401 Here (a,b) both look dead, but come alive after the inner case is eliminated.
1402 The point is that we bring into the envt a binding
1404 after the outer case, and that makes (a,b) alive. At least we do unless
1405 the case binder is guaranteed dead.
1408 simplCaseBinder env (Var v) case_bndr
1409 | not (switchIsOn (getSwitchChecker env) NoCaseOfCase)
1411 -- Failed try [see Note 2 above]
1412 -- not (isEvaldUnfolding (idUnfolding v))
1414 = simplBinder env (zap case_bndr) `thenSmpl` \ (env, case_bndr') ->
1415 returnSmpl (modifyInScope env v case_bndr', case_bndr')
1416 -- We could extend the substitution instead, but it would be
1417 -- a hack because then the substitution wouldn't be idempotent
1418 -- any more (v is an OutId). And this does just as well.
1420 zap b = b `setIdOccInfo` NoOccInfo
1422 simplCaseBinder env other_scrut case_bndr
1423 = simplBinder env case_bndr `thenSmpl` \ (env, case_bndr') ->
1424 returnSmpl (env, case_bndr')
1430 simplAlts :: SimplEnv
1431 -> [AltCon] -- Alternatives the scrutinee can't be
1432 -- in the default case
1433 -> OutId -- Case binder
1434 -> [InAlt] -> SimplCont
1435 -> SimplM [OutAlt] -- Includes the continuation
1437 simplAlts env handled_cons case_bndr' alts cont'
1438 = do { mb_alts <- mapSmpl simpl_alt alts
1439 ; return [alt' | Just (_, alt') <- mb_alts] }
1440 -- Filter out the alternatives that are inaccessible
1442 simpl_alt alt = simplAlt env handled_cons case_bndr' alt cont'
1444 simplAlt :: SimplEnv -> [AltCon] -> OutId -> InAlt -> SimplCont
1445 -> SimplM (Maybe (TvSubstEnv, OutAlt))
1446 -- Simplify an alternative, returning the type refinement for the
1447 -- alternative, if the alternative does any refinement at all
1448 -- Nothing => the alternative is inaccessible
1450 simplAlt env handled_cons case_bndr' (DEFAULT, bndrs, rhs) cont'
1451 = ASSERT( null bndrs )
1452 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1453 returnSmpl (Just (emptyVarEnv, (DEFAULT, [], rhs')))
1455 env' = mk_rhs_env env case_bndr' (mkOtherCon handled_cons)
1456 -- Record the constructors that the case-binder *can't* be.
1458 simplAlt env handled_cons case_bndr' (LitAlt lit, bndrs, rhs) cont'
1459 = ASSERT( null bndrs )
1460 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1461 returnSmpl (Just (emptyVarEnv, (LitAlt lit, [], rhs')))
1463 env' = mk_rhs_env env case_bndr' (mkUnfolding False (Lit lit))
1465 simplAlt env handled_cons case_bndr' (DataAlt con, vs, rhs) cont'
1466 | isVanillaDataCon con
1467 = -- Deal with the pattern-bound variables
1468 -- Mark the ones that are in ! positions in the data constructor
1469 -- as certainly-evaluated.
1470 -- NB: it happens that simplBinders does *not* erase the OtherCon
1471 -- form of unfolding, so it's ok to add this info before
1472 -- doing simplBinders
1473 simplBinders env (add_evals con vs) `thenSmpl` \ (env, vs') ->
1475 -- Bind the case-binder to (con args)
1476 let unf = mkUnfolding False (mkConApp con con_args)
1477 inst_tys' = tyConAppArgs (idType case_bndr')
1478 con_args = map Type inst_tys' ++ map varToCoreExpr vs'
1479 env' = mk_rhs_env env case_bndr' unf
1481 simplExprC env' rhs cont' `thenSmpl` \ rhs' ->
1482 returnSmpl (Just (emptyVarEnv, (DataAlt con, vs', rhs')))
1484 | otherwise -- GADT case
1486 (tvs,ids) = span isTyVar vs
1488 simplBinders env tvs `thenSmpl` \ (env1, tvs') ->
1489 case coreRefineTys con tvs' (idType case_bndr') of {
1490 Nothing -- Inaccessible
1491 | opt_PprStyle_Debug -- Hack: if debugging is on, generate an error case
1493 -> let rhs' = mkApps (Var eRROR_ID)
1494 [Type (substTy env (exprType rhs)),
1495 Lit (mkStringLit "Impossible alternative (GADT)")]
1497 simplBinders env1 ids `thenSmpl` \ (env2, ids') ->
1498 returnSmpl (Just (emptyVarEnv, (DataAlt con, tvs' ++ ids', rhs')))
1500 | otherwise -- Filter out the inaccessible branch
1503 Just refine@(tv_subst_env, _) -> -- The normal case
1506 env2 = refineSimplEnv env1 refine
1507 -- Simplify the Ids in the refined environment, so their types
1508 -- reflect the refinement. Usually this doesn't matter, but it helps
1509 -- in mkDupableAlt, when we want to float a lambda that uses these binders
1510 -- Furthermore, it means the binders contain maximal type information
1512 simplBinders env2 (add_evals con ids) `thenSmpl` \ (env3, ids') ->
1513 let unf = mkUnfolding False con_app
1514 con_app = mkConApp con con_args
1515 con_args = map varToCoreExpr vs' -- NB: no inst_tys'
1516 env_w_unf = mk_rhs_env env3 case_bndr' unf
1519 simplExprC env_w_unf rhs cont' `thenSmpl` \ rhs' ->
1520 returnSmpl (Just (tv_subst_env, (DataAlt con, vs', rhs'))) }
1523 -- add_evals records the evaluated-ness of the bound variables of
1524 -- a case pattern. This is *important*. Consider
1525 -- data T = T !Int !Int
1527 -- case x of { T a b -> T (a+1) b }
1529 -- We really must record that b is already evaluated so that we don't
1530 -- go and re-evaluate it when constructing the result.
1531 add_evals dc vs = cat_evals dc vs (dataConRepStrictness dc)
1533 cat_evals dc vs strs
1537 go (v:vs) strs | isTyVar v = v : go vs strs
1538 go (v:vs) (str:strs)
1539 | isMarkedStrict str = evald_v : go vs strs
1540 | otherwise = zapped_v : go vs strs
1542 zapped_v = zap_occ_info v
1543 evald_v = zapped_v `setIdUnfolding` evaldUnfolding
1544 go _ _ = pprPanic "cat_evals" (ppr dc $$ ppr vs $$ ppr strs)
1546 -- If the case binder is alive, then we add the unfolding
1548 -- to the envt; so vs are now very much alive
1549 zap_occ_info | isDeadBinder case_bndr' = \id -> id
1550 | otherwise = \id -> id `setIdOccInfo` NoOccInfo
1552 mk_rhs_env env case_bndr' case_bndr_unf
1553 = modifyInScope env case_bndr' (case_bndr' `setIdUnfolding` case_bndr_unf)
1557 %************************************************************************
1559 \subsection{Known constructor}
1561 %************************************************************************
1563 We are a bit careful with occurrence info. Here's an example
1565 (\x* -> case x of (a*, b) -> f a) (h v, e)
1567 where the * means "occurs once". This effectively becomes
1568 case (h v, e) of (a*, b) -> f a)
1570 let a* = h v; b = e in f a
1574 All this should happen in one sweep.
1577 knownCon :: SimplEnv -> AltCon -> [OutExpr]
1578 -> InId -> [InAlt] -> SimplCont
1579 -> SimplM FloatsWithExpr
1581 knownCon env con args bndr alts cont
1582 = tick (KnownBranch bndr) `thenSmpl_`
1583 case findAlt con alts of
1584 (DEFAULT, bs, rhs) -> ASSERT( null bs )
1585 simplNonRecX env bndr scrut $ \ env ->
1586 -- This might give rise to a binding with non-atomic args
1587 -- like x = Node (f x) (g x)
1588 -- but no harm will be done
1589 simplExprF env rhs cont
1592 LitAlt lit -> Lit lit
1593 DataAlt dc -> mkConApp dc args
1595 (LitAlt lit, bs, rhs) -> ASSERT( null bs )
1596 simplNonRecX env bndr (Lit lit) $ \ env ->
1597 simplExprF env rhs cont
1599 (DataAlt dc, bs, rhs)
1600 -> ASSERT( n_drop_tys + length bs == length args )
1601 bind_args env bs (drop n_drop_tys args) $ \ env ->
1603 con_app = mkConApp dc (take n_drop_tys args ++ con_args)
1604 con_args = [substExpr env (varToCoreExpr b) | b <- bs]
1605 -- args are aready OutExprs, but bs are InIds
1607 simplNonRecX env bndr con_app $ \ env ->
1608 simplExprF env rhs cont
1610 n_drop_tys | isVanillaDataCon dc = tyConArity (dataConTyCon dc)
1612 -- Vanilla data constructors lack type arguments in the pattern
1615 bind_args env [] _ thing_inside = thing_inside env
1617 bind_args env (b:bs) (Type ty : args) thing_inside
1618 = ASSERT( isTyVar b )
1619 bind_args (extendTvSubst env b ty) bs args thing_inside
1621 bind_args env (b:bs) (arg : args) thing_inside
1623 simplNonRecX env b arg $ \ env ->
1624 bind_args env bs args thing_inside
1628 %************************************************************************
1630 \subsection{Duplicating continuations}
1632 %************************************************************************
1635 prepareCaseCont :: SimplEnv
1636 -> [InAlt] -> SimplCont
1637 -> SimplM (FloatsWith (SimplCont,SimplCont))
1638 -- Return a duplicatable continuation, a non-duplicable part
1639 -- plus some extra bindings
1641 -- No need to make it duplicatable if there's only one alternative
1642 prepareCaseCont env [alt] cont = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1643 prepareCaseCont env alts cont = mkDupableCont env cont
1647 mkDupableCont :: SimplEnv -> SimplCont
1648 -> SimplM (FloatsWith (SimplCont, SimplCont))
1650 mkDupableCont env cont
1651 | contIsDupable cont
1652 = returnSmpl (emptyFloats env, (cont, mkBoringStop (contResultType cont)))
1654 mkDupableCont env (CoerceIt ty cont)
1655 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1656 returnSmpl (floats, (CoerceIt ty dup_cont, nondup_cont))
1658 mkDupableCont env (InlinePlease cont)
1659 = mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1660 returnSmpl (floats, (InlinePlease dup_cont, nondup_cont))
1662 mkDupableCont env cont@(ArgOf _ arg_ty _ _)
1663 = returnSmpl (emptyFloats env, (mkBoringStop arg_ty, cont))
1664 -- Do *not* duplicate an ArgOf continuation
1665 -- Because ArgOf continuations are opaque, we gain nothing by
1666 -- propagating them into the expressions, and we do lose a lot.
1667 -- Here's an example:
1668 -- && (case x of { T -> F; F -> T }) E
1669 -- Now, && is strict so we end up simplifying the case with
1670 -- an ArgOf continuation. If we let-bind it, we get
1672 -- let $j = \v -> && v E
1673 -- in simplExpr (case x of { T -> F; F -> T })
1674 -- (ArgOf (\r -> $j r)
1675 -- And after simplifying more we get
1677 -- let $j = \v -> && v E
1678 -- in case of { T -> $j F; F -> $j T }
1679 -- Which is a Very Bad Thing
1681 -- The desire not to duplicate is the entire reason that
1682 -- mkDupableCont returns a pair of continuations.
1684 -- The original plan had:
1685 -- e.g. (...strict-fn...) [...hole...]
1687 -- let $j = \a -> ...strict-fn...
1688 -- in $j [...hole...]
1690 mkDupableCont env (ApplyTo _ arg se cont)
1691 = -- e.g. [...hole...] (...arg...)
1693 -- let a = ...arg...
1694 -- in [...hole...] a
1695 simplExpr (setInScope se env) arg `thenSmpl` \ arg' ->
1697 mkDupableCont env cont `thenSmpl` \ (floats, (dup_cont, nondup_cont)) ->
1698 addFloats env floats $ \ env ->
1700 if exprIsDupable arg' then
1701 returnSmpl (emptyFloats env, (ApplyTo OkToDup arg' (zapSubstEnv se) dup_cont, nondup_cont))
1703 newId FSLIT("a") (exprType arg') `thenSmpl` \ arg_id ->
1705 tick (CaseOfCase arg_id) `thenSmpl_`
1706 -- Want to tick here so that we go round again,
1707 -- and maybe copy or inline the code.
1708 -- Not strictly CaseOfCase, but never mind
1710 returnSmpl (unitFloat env arg_id arg',
1711 (ApplyTo OkToDup (Var arg_id) (zapSubstEnv se) dup_cont,
1713 -- But what if the arg should be case-bound?
1714 -- This has been this way for a long time, so I'll leave it,
1715 -- but I can't convince myself that it's right.
1717 mkDupableCont env (Select _ case_bndr alts se cont)
1718 = -- e.g. (case [...hole...] of { pi -> ei })
1720 -- let ji = \xij -> ei
1721 -- in case [...hole...] of { pi -> ji xij }
1722 tick (CaseOfCase case_bndr) `thenSmpl_`
1724 alt_env = setInScope se env
1726 prepareCaseCont alt_env alts cont `thenSmpl` \ (floats1, (dup_cont, nondup_cont)) ->
1727 addFloats alt_env floats1 $ \ alt_env ->
1729 simplBinder alt_env case_bndr `thenSmpl` \ (alt_env, case_bndr') ->
1730 -- NB: simplBinder does not zap deadness occ-info, so
1731 -- a dead case_bndr' will still advertise its deadness
1732 -- This is really important because in
1733 -- case e of b { (# a,b #) -> ... }
1734 -- b is always dead, and indeed we are not allowed to bind b to (# a,b #),
1735 -- which might happen if e was an explicit unboxed pair and b wasn't marked dead.
1736 -- In the new alts we build, we have the new case binder, so it must retain
1739 mkDupableAlts alt_env case_bndr' alts dup_cont `thenSmpl` \ (floats2, alts') ->
1740 addFloats alt_env floats2 $ \ alt_env ->
1741 returnSmpl (emptyFloats alt_env,
1742 (Select OkToDup case_bndr' alts' (zapSubstEnv se)
1743 (mkBoringStop (contResultType dup_cont)),
1746 mkDupableAlts :: SimplEnv -> OutId -> [InAlt] -> SimplCont
1747 -> SimplM (FloatsWith [InAlt])
1748 -- Absorbs the continuation into the new alternatives
1750 mkDupableAlts env case_bndr' alts dupable_cont
1753 go env [] = returnSmpl (emptyFloats env, [])
1755 = do { (floats1, mb_alt') <- mkDupableAlt env case_bndr' dupable_cont alt
1756 ; addFloats env floats1 $ \ env -> do
1757 { (floats2, alts') <- go env alts
1758 ; returnSmpl (floats2, case mb_alt' of
1759 Just alt' -> alt' : alts'
1763 mkDupableAlt env case_bndr' cont alt
1764 = simplAlt env [] case_bndr' alt cont `thenSmpl` \ mb_stuff ->
1766 Nothing -> returnSmpl (emptyFloats env, Nothing) ;
1768 Just (reft, (con, bndrs', rhs')) ->
1769 -- Safe to say that there are no handled-cons for the DEFAULT case
1771 if exprIsDupable rhs' then
1772 returnSmpl (emptyFloats env, Just (con, bndrs', rhs'))
1773 -- It is worth checking for a small RHS because otherwise we
1774 -- get extra let bindings that may cause an extra iteration of the simplifier to
1775 -- inline back in place. Quite often the rhs is just a variable or constructor.
1776 -- The Ord instance of Maybe in PrelMaybe.lhs, for example, took several extra
1777 -- iterations because the version with the let bindings looked big, and so wasn't
1778 -- inlined, but after the join points had been inlined it looked smaller, and so
1781 -- NB: we have to check the size of rhs', not rhs.
1782 -- Duplicating a small InAlt might invalidate occurrence information
1783 -- However, if it *is* dupable, we return the *un* simplified alternative,
1784 -- because otherwise we'd need to pair it up with an empty subst-env....
1785 -- but we only have one env shared between all the alts.
1786 -- (Remember we must zap the subst-env before re-simplifying something).
1787 -- Rather than do this we simply agree to re-simplify the original (small) thing later.
1791 rhs_ty' = exprType rhs'
1792 used_bndrs' = filter abstract_over (case_bndr' : bndrs')
1794 | isTyVar bndr = not (bndr `elemVarEnv` reft)
1795 -- Don't abstract over tyvar binders which are refined away
1796 -- See Note [Refinement] below
1797 | otherwise = not (isDeadBinder bndr)
1798 -- The deadness info on the new Ids is preserved by simplBinders
1800 -- If we try to lift a primitive-typed something out
1801 -- for let-binding-purposes, we will *caseify* it (!),
1802 -- with potentially-disastrous strictness results. So
1803 -- instead we turn it into a function: \v -> e
1804 -- where v::State# RealWorld#. The value passed to this function
1805 -- is realworld#, which generates (almost) no code.
1807 -- There's a slight infelicity here: we pass the overall
1808 -- case_bndr to all the join points if it's used in *any* RHS,
1809 -- because we don't know its usage in each RHS separately
1811 -- We used to say "&& isUnLiftedType rhs_ty'" here, but now
1812 -- we make the join point into a function whenever used_bndrs'
1813 -- is empty. This makes the join-point more CPR friendly.
1814 -- Consider: let j = if .. then I# 3 else I# 4
1815 -- in case .. of { A -> j; B -> j; C -> ... }
1817 -- Now CPR doesn't w/w j because it's a thunk, so
1818 -- that means that the enclosing function can't w/w either,
1819 -- which is a lose. Here's the example that happened in practice:
1820 -- kgmod :: Int -> Int -> Int
1821 -- kgmod x y = if x > 0 && y < 0 || x < 0 && y > 0
1825 -- I have seen a case alternative like this:
1826 -- True -> \v -> ...
1827 -- It's a bit silly to add the realWorld dummy arg in this case, making
1830 -- (the \v alone is enough to make CPR happy) but I think it's rare
1832 ( if not (any isId used_bndrs')
1833 then newId FSLIT("w") realWorldStatePrimTy `thenSmpl` \ rw_id ->
1834 returnSmpl ([rw_id], [Var realWorldPrimId])
1836 returnSmpl (used_bndrs', map varToCoreExpr used_bndrs')
1837 ) `thenSmpl` \ (final_bndrs', final_args) ->
1839 -- See comment about "$j" name above
1840 newId FSLIT("$j") (mkPiTypes final_bndrs' rhs_ty') `thenSmpl` \ join_bndr ->
1841 -- Notice the funky mkPiTypes. If the contructor has existentials
1842 -- it's possible that the join point will be abstracted over
1843 -- type varaibles as well as term variables.
1844 -- Example: Suppose we have
1845 -- data T = forall t. C [t]
1847 -- case (case e of ...) of
1848 -- C t xs::[t] -> rhs
1849 -- We get the join point
1850 -- let j :: forall t. [t] -> ...
1851 -- j = /\t \xs::[t] -> rhs
1853 -- case (case e of ...) of
1854 -- C t xs::[t] -> j t xs
1856 -- We make the lambdas into one-shot-lambdas. The
1857 -- join point is sure to be applied at most once, and doing so
1858 -- prevents the body of the join point being floated out by
1859 -- the full laziness pass
1860 really_final_bndrs = map one_shot final_bndrs'
1861 one_shot v | isId v = setOneShotLambda v
1863 join_rhs = mkLams really_final_bndrs rhs'
1864 join_call = mkApps (Var join_bndr) final_args
1866 returnSmpl (unitFloat env join_bndr join_rhs, Just (con, bndrs', join_call)) }
1873 MkT :: a -> b -> T a
1877 MkT a' b (p::a') (q::b) -> [p,w]
1879 The danger is that we'll make a join point
1883 and that's ill-typed, because (p::a') but (w::a).
1885 Solution so far: don't abstract over a', because the type refinement
1886 maps [a' -> a] . Ultimately that won't work when real refinement goes on.
1888 Then we must abstract over any refined free variables. Hmm. Maybe we
1889 could just abstract over *all* free variables, thereby lambda-lifting
1890 the join point? We should try this.