2 % (c) The GRASP/AQUA Project, Glasgow University, 1993-1998
4 \section[WorkWrap]{Worker/wrapper-generating back-end of strictness analyser}
7 module WorkWrap ( wwTopBinds, mkWrapper ) where
10 import CoreUnfold ( certainlyWillInline, mkInlineUnfolding, mkWwInlineRule )
11 import CoreUtils ( exprType, exprIsHNF )
12 import CoreArity ( exprArity )
20 import VarEnv ( isEmptyVarEnv )
21 import Maybes ( orElse )
23 import Util ( lengthIs, notNull )
27 #include "HsVersions.h"
30 We take Core bindings whose binders have:
34 \item Strictness attached (by the front-end of the strictness
37 \item Constructed Product Result information attached by the CPR
42 and we return some ``plain'' bindings which have been
43 worker/wrapper-ified, meaning:
47 \item Functions have been split into workers and wrappers where
48 appropriate. If a function has both strictness and CPR properties
49 then only one worker/wrapper doing both transformations is produced;
51 \item Binders' @IdInfos@ have been updated to reflect the existence of
52 these workers/wrappers (this is where we get STRICTNESS and CPR pragma
53 info for exported values).
57 wwTopBinds :: UniqSupply -> [CoreBind] -> [CoreBind]
59 wwTopBinds us top_binds
61 top_binds' <- mapM wwBind top_binds
62 return (concat top_binds')
65 %************************************************************************
67 \subsection[wwBind-wwExpr]{@wwBind@ and @wwExpr@}
69 %************************************************************************
71 @wwBind@ works on a binding, trying each \tr{(binder, expr)} pair in
72 turn. Non-recursive case first, then recursive...
76 -> UniqSM [CoreBind] -- returns a WwBinding intermediate form;
77 -- the caller will convert to Expr/Binding,
80 wwBind (NonRec binder rhs) = do
82 new_pairs <- tryWW NonRecursive binder new_rhs
83 return [NonRec b e | (b,e) <- new_pairs]
84 -- Generated bindings must be non-recursive
85 -- because the original binding was.
88 = return . Rec <$> concatMapM do_one pairs
90 do_one (binder, rhs) = do new_rhs <- wwExpr rhs
91 tryWW Recursive binder new_rhs
94 @wwExpr@ basically just walks the tree, looking for appropriate
95 annotations that can be used. Remember it is @wwBind@ that does the
96 matching by looking for strict arguments of the correct type.
97 @wwExpr@ is a version that just returns the ``Plain'' Tree.
100 wwExpr :: CoreExpr -> UniqSM CoreExpr
102 wwExpr e@(Type {}) = return e
103 wwExpr e@(Lit {}) = return e
104 wwExpr e@(Var {}) = return e
106 wwExpr (Lam binder expr)
107 = Lam binder <$> wwExpr expr
110 = App <$> wwExpr f <*> wwExpr a
112 wwExpr (Note note expr)
113 = Note note <$> wwExpr expr
115 wwExpr (Cast expr co) = do
116 new_expr <- wwExpr expr
117 return (Cast new_expr co)
119 wwExpr (Let bind expr)
120 = mkLets <$> wwBind bind <*> wwExpr expr
122 wwExpr (Case expr binder ty alts) = do
123 new_expr <- wwExpr expr
124 new_alts <- mapM ww_alt alts
125 return (Case new_expr binder ty new_alts)
127 ww_alt (con, binders, rhs) = do
128 new_rhs <- wwExpr rhs
129 return (con, binders, new_rhs)
132 %************************************************************************
134 \subsection[tryWW]{@tryWW@: attempt a worker/wrapper pair}
136 %************************************************************************
138 @tryWW@ just accumulates arguments, converts strictness info from the
139 front-end into the proper form, then calls @mkWwBodies@ to do
142 The only reason this is monadised is for the unique supply.
144 Note [Don't w/w inline things (a)]
145 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
147 It's very important to refrain from w/w-ing an INLINE function (ie one
148 with an InlineRule) because the wrapper will then overwrite the
149 InlineRule unfolding.
151 Furthermore, if the programmer has marked something as INLINE,
152 we may lose by w/w'ing it.
154 If the strictness analyser is run twice, this test also prevents
155 wrappers (which are INLINEd) from being re-done. (You can end up with
156 several liked-named Ids bouncing around at the same time---absolute
159 Notice that we refrain from w/w'ing an INLINE function even if it is
160 in a recursive group. It might not be the loop breaker. (We could
161 test for loop-breaker-hood, but I'm not sure that ever matters.)
163 Note [Don't w/w inline things (b)]
164 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
165 In general, we refrain from w/w-ing *small* functions, because they'll
166 inline anyway. But we must take care: it may look small now, but get
167 to be big later after other inling has happened. So we take the
168 precaution of adding an INLINE pragma to any such functions.
170 I made this change when I observed a big function at the end of
171 compilation with a useful strictness signature but no w-w. When
172 I measured it on nofib, it didn't make much difference; just a few
173 percent improved allocation on one benchmark (bspt/Euclid.space).
174 But nothing got worse.
176 Note [Wrapper activation]
177 ~~~~~~~~~~~~~~~~~~~~~~~~~
178 When should the wrapper inlining be active? It must not be active
179 earlier than the current Activation of the Id (eg it might have a
180 NOINLINE pragma). But in fact strictness analysis happens fairly
181 late in the pipeline, and we want to prioritise specialisations over
182 strictness. Eg if we have
184 f :: Num a => a -> Int -> a
185 f n 0 = n -- Strict in the Int, hence wrapper
186 f n x = f (n+n) (x-1)
189 g x = f x x -- Provokes a specialisation for f
197 Then we want the specialisation for 'f' to kick in before the wrapper does.
199 Now in fact the 'gentle' simplification pass encourages this, by
200 having rules on, but inlinings off. But that's kind of lucky. It seems
201 more robust to give the wrapper an Activation of (ActiveAfter 0),
202 so that it becomes active in an importing module at the same time that
203 it appears in the first place in the defining module.
207 -> Id -- The fn binder
208 -> CoreExpr -- The bound rhs; its innards
210 -> UniqSM [(Id, CoreExpr)] -- either *one* or *two* pairs;
211 -- if one, then no worker (only
212 -- the orig "wrapper" lives on);
213 -- if two, then a worker and a
215 tryWW is_rec fn_id rhs
216 | isNeverActive inline_act
217 -- No point in worker/wrappering if the thing is never inlined!
218 -- Because the no-inline prag will prevent the wrapper ever
219 -- being inlined at a call site.
221 -- Furthermore, don't even expose strictness info
222 = return [ (fn_id, rhs) ]
224 | is_thunk && worthSplittingThunk maybe_fn_dmd res_info
225 -- See Note [Thunk splitting]
226 = ASSERT2( isNonRec is_rec, ppr new_fn_id ) -- The thunk must be non-recursive
227 checkSize new_fn_id rhs $
228 splitThunk new_fn_id rhs
230 | is_fun && worthSplittingFun wrap_dmds res_info
231 = checkSize new_fn_id rhs $
232 splitFun new_fn_id fn_info wrap_dmds res_info rhs
235 = return [ (new_fn_id, rhs) ]
238 fn_info = idInfo fn_id
239 maybe_fn_dmd = demandInfo fn_info
240 inline_act = inlinePragmaActivation (inlinePragInfo fn_info)
242 -- In practice it always will have a strictness
243 -- signature, even if it's a uninformative one
244 strict_sig = strictnessInfo fn_info `orElse` topSig
245 StrictSig (DmdType env wrap_dmds res_info) = strict_sig
247 -- new_fn_id has the DmdEnv zapped.
248 -- (a) it is never used again
249 -- (b) it wastes space
250 -- (c) it becomes incorrect as things are cloned, because
251 -- we don't push the substitution into it
252 new_fn_id | isEmptyVarEnv env = fn_id
253 | otherwise = fn_id `setIdStrictness`
254 StrictSig (mkTopDmdType wrap_dmds res_info)
256 is_fun = notNull wrap_dmds
257 is_thunk = not is_fun && not (exprIsHNF rhs)
259 ---------------------
260 checkSize :: Id -> CoreExpr
261 -> UniqSM [(Id,CoreExpr)] -> UniqSM [(Id,CoreExpr)]
262 -- See Note [Don't w/w inline things (a) and (b)]
263 checkSize fn_id rhs thing_inside
264 | isStableUnfolding unfolding -- For DFuns and INLINE things, leave their
265 = return [ (fn_id, rhs) ] -- unfolding unchanged; but still attach
266 -- strictness info to the Id
268 | certainlyWillInline unfolding
269 = return [ (fn_id `setIdUnfolding` inline_rule, rhs) ]
270 -- Note [Don't w/w inline things (b)]
272 | otherwise = thing_inside
274 unfolding = idUnfolding fn_id
275 inline_rule = mkInlineUnfolding Nothing rhs
277 ---------------------
278 splitFun :: Id -> IdInfo -> [Demand] -> DmdResult -> Expr Var
279 -> UniqSM [(Id, CoreExpr)]
280 splitFun fn_id fn_info wrap_dmds res_info rhs
281 = WARN( not (wrap_dmds `lengthIs` arity), ppr fn_id <+> (ppr arity $$ ppr wrap_dmds $$ ppr res_info) )
283 -- The arity should match the signature
284 (work_demands, wrap_fn, work_fn) <- mkWwBodies fun_ty wrap_dmds res_info one_shots
285 ; work_uniq <- getUniqueM
287 work_rhs = work_fn rhs
288 work_id = mkWorkerId work_uniq fn_id (exprType work_rhs)
289 `setIdOccInfo` occInfo fn_info
290 -- Copy over occurrence info from parent
291 -- Notably whether it's a loop breaker
292 -- Doesn't matter much, since we will simplify next, but
293 -- seems right-er to do so
295 `setInlineActivation` (inlinePragmaActivation inl_prag)
296 -- Any inline activation (which sets when inlining is active)
297 -- on the original function is duplicated on the worker
298 -- It *matters* that the pragma stays on the wrapper
299 -- It seems sensible to have it on the worker too, although we
300 -- can't think of a compelling reason. (In ptic, INLINE things are
301 -- not w/wd). However, the RuleMatchInfo is not transferred since
302 -- it does not make sense for workers to be constructorlike.
304 `setIdStrictness` StrictSig (mkTopDmdType work_demands work_res_info)
305 -- Even though we may not be at top level,
306 -- it's ok to give it an empty DmdEnv
308 `setIdArity` (exprArity work_rhs)
309 -- Set the arity so that the Core Lint check that the
310 -- arity is consistent with the demand type goes through
312 wrap_rhs = wrap_fn work_id
313 wrap_prag = InlinePragma { inl_inline = Inline
315 , inl_act = ActiveAfter 0
316 , inl_rule = rule_match_info }
317 -- See Note [Wrapper activation]
318 -- The RuleMatchInfo is (and must be) unaffected
319 -- The inl_inline is bound to be False, else we would not be
322 wrap_id = fn_id `setIdUnfolding` mkWwInlineRule work_id wrap_rhs arity
323 `setInlinePragma` wrap_prag
324 `setIdOccInfo` NoOccInfo
325 -- Zap any loop-breaker-ness, to avoid bleating from Lint
326 -- about a loop breaker with an INLINE rule
328 ; return ([(work_id, work_rhs), (wrap_id, wrap_rhs)]) })
329 -- Worker first, because wrapper mentions it
330 -- mkWwBodies has already built a wrap_rhs with an INLINE pragma wrapped around it
332 fun_ty = idType fn_id
333 inl_prag = inlinePragInfo fn_info
334 rule_match_info = inlinePragmaRuleMatchInfo inl_prag
335 arity = arityInfo fn_info
336 -- The arity is set by the simplifier using exprEtaExpandArity
337 -- So it may be more than the number of top-level-visible lambdas
339 work_res_info | isBotRes res_info = BotRes -- Cpr stuff done by wrapper
342 one_shots = get_one_shots rhs
344 -- If the original function has one-shot arguments, it is important to
345 -- make the wrapper and worker have corresponding one-shot arguments too.
346 -- Otherwise we spuriously float stuff out of case-expression join points,
347 -- which is very annoying.
348 get_one_shots :: Expr Var -> [Bool]
349 get_one_shots (Lam b e)
350 | isId b = isOneShotLambda b : get_one_shots e
351 | otherwise = get_one_shots e
352 get_one_shots (Note _ e) = get_one_shots e
353 get_one_shots _ = noOneShotInfo
356 Note [Thunk splitting]
357 ~~~~~~~~~~~~~~~~~~~~~~
358 Suppose x is used strictly (never mind whether it has the CPR
365 splitThunk transforms like this:
368 x* = case x-rhs of { I# a -> I# a }
371 Now simplifier will transform to
374 I# a -> let x* = I# a
377 which is what we want. Now suppose x-rhs is itself a case:
379 x-rhs = case e of { T -> I# a; F -> I# b }
381 The join point will abstract over a, rather than over (which is
382 what would have happened before) which is fine.
384 Notice that x certainly has the CPR property now!
386 In fact, splitThunk uses the function argument w/w splitting
387 function, so that if x's demand is deeper (say U(U(L,L),L))
388 then the splitting will go deeper too.
391 -- See Note [Thunk splitting]
392 -- splitThunk converts the *non-recursive* binding
397 -- I# y -> let x = I# y in x }
398 -- See comments above. Is it not beautifully short?
400 splitThunk :: Var -> Expr Var -> UniqSM [(Var, Expr Var)]
401 splitThunk fn_id rhs = do
402 (_, wrap_fn, work_fn) <- mkWWstr [fn_id]
403 return [ (fn_id, Let (NonRec fn_id rhs) (wrap_fn (work_fn (Var fn_id)))) ]
407 %************************************************************************
409 \subsection{Functions over Demands}
411 %************************************************************************
414 worthSplittingFun :: [Demand] -> DmdResult -> Bool
415 -- True <=> the wrapper would not be an identity function
416 worthSplittingFun ds res
417 = any worth_it ds || returnsCPR res
418 -- worthSplitting returns False for an empty list of demands,
419 -- and hence do_strict_ww is False if arity is zero and there is no CPR
420 -- See Note [Worker-wrapper for bottoming functions]
422 worth_it Abs = True -- Absent arg
423 worth_it (Eval (Prod _)) = True -- Product arg to evaluate
426 worthSplittingThunk :: Maybe Demand -- Demand on the thunk
427 -> DmdResult -- CPR info for the thunk
429 worthSplittingThunk maybe_dmd res
430 = worth_it maybe_dmd || returnsCPR res
432 -- Split if the thing is unpacked
433 worth_it (Just (Eval (Prod ds))) = not (all isAbsent ds)
437 Note [Worker-wrapper for bottoming functions]
438 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
439 We used not to split if the result is bottom.
440 [Justification: there's no efficiency to be gained.]
442 But it's sometimes bad not to make a wrapper. Consider
443 fw = \x# -> let x = I# x# in case e of
447 The re-boxing code won't go away unless error_fn gets a wrapper too.
448 [We don't do reboxing now, but in general it's better to pass an
449 unboxed thing to f, and have it reboxed in the error cases....]
452 %************************************************************************
454 \subsection{The worker wrapper core}
456 %************************************************************************
458 @mkWrapper@ is called when importing a function. We have the type of
459 the function and the name of its worker, and we want to make its body (the wrapper).
462 mkWrapper :: Type -- Wrapper type
463 -> StrictSig -- Wrapper strictness info
464 -> UniqSM (Id -> CoreExpr) -- Wrapper body, missing worker Id
466 mkWrapper fun_ty (StrictSig (DmdType _ demands res_info)) = do
467 (_, wrap_fn, _) <- mkWwBodies fun_ty demands res_info noOneShotInfo
470 noOneShotInfo :: [Bool]
471 noOneShotInfo = repeat False