diff options
Diffstat (limited to 'compiler/coreSyn/CoreUnfold.lhs')
| -rw-r--r-- | compiler/coreSyn/CoreUnfold.lhs | 632 |
1 files changed, 632 insertions, 0 deletions
diff --git a/compiler/coreSyn/CoreUnfold.lhs b/compiler/coreSyn/CoreUnfold.lhs new file mode 100644 index 0000000000..d57f1886fc --- /dev/null +++ b/compiler/coreSyn/CoreUnfold.lhs @@ -0,0 +1,632 @@ +% +% (c) The AQUA Project, Glasgow University, 1994-1998 +% +\section[CoreUnfold]{Core-syntax unfoldings} + +Unfoldings (which can travel across module boundaries) are in Core +syntax (namely @CoreExpr@s). + +The type @Unfolding@ sits ``above'' simply-Core-expressions +unfoldings, capturing ``higher-level'' things we know about a binding, +usually things that the simplifier found out (e.g., ``it's a +literal''). In the corner of a @CoreUnfolding@ unfolding, you will +find, unsurprisingly, a Core expression. + +\begin{code} +module CoreUnfold ( + Unfolding, UnfoldingGuidance, -- Abstract types + + noUnfolding, mkTopUnfolding, mkUnfolding, mkCompulsoryUnfolding, seqUnfolding, + evaldUnfolding, mkOtherCon, otherCons, + unfoldingTemplate, maybeUnfoldingTemplate, + isEvaldUnfolding, isValueUnfolding, isCheapUnfolding, isCompulsoryUnfolding, + hasUnfolding, hasSomeUnfolding, neverUnfold, + + couldBeSmallEnoughToInline, + certainlyWillInline, smallEnoughToInline, + + callSiteInline + ) where + +#include "HsVersions.h" + +import StaticFlags ( opt_UF_CreationThreshold, opt_UF_UseThreshold, + opt_UF_FunAppDiscount, opt_UF_KeenessFactor, + opt_UF_DearOp, + ) +import DynFlags ( DynFlags, DynFlag(..), dopt ) +import CoreSyn +import PprCore ( pprCoreExpr ) +import OccurAnal ( occurAnalyseExpr ) +import CoreUtils ( exprIsHNF, exprIsCheap, exprIsTrivial ) +import Id ( Id, idType, isId, + idUnfolding, globalIdDetails + ) +import DataCon ( isUnboxedTupleCon ) +import Literal ( litSize ) +import PrimOp ( primOpIsDupable, primOpOutOfLine ) +import IdInfo ( OccInfo(..), GlobalIdDetails(..) ) +import Type ( isUnLiftedType ) +import PrelNames ( hasKey, buildIdKey, augmentIdKey ) +import Bag +import FastTypes +import Outputable + +#if __GLASGOW_HASKELL__ >= 404 +import GLAEXTS ( Int# ) +#endif +\end{code} + + +%************************************************************************ +%* * +\subsection{Making unfoldings} +%* * +%************************************************************************ + +\begin{code} +mkTopUnfolding expr = mkUnfolding True {- Top level -} expr + +mkUnfolding top_lvl expr + = CoreUnfolding (occurAnalyseExpr expr) + top_lvl + + (exprIsHNF expr) + -- Already evaluated + + (exprIsCheap expr) + -- OK to inline inside a lambda + + (calcUnfoldingGuidance opt_UF_CreationThreshold expr) + -- Sometimes during simplification, there's a large let-bound thing + -- which has been substituted, and so is now dead; so 'expr' contains + -- two copies of the thing while the occurrence-analysed expression doesn't + -- Nevertheless, we don't occ-analyse before computing the size because the + -- size computation bales out after a while, whereas occurrence analysis does not. + -- + -- This can occasionally mean that the guidance is very pessimistic; + -- it gets fixed up next round + +mkCompulsoryUnfolding expr -- Used for things that absolutely must be unfolded + = CompulsoryUnfolding (occurAnalyseExpr expr) +\end{code} + + +%************************************************************************ +%* * +\subsection{The UnfoldingGuidance type} +%* * +%************************************************************************ + +\begin{code} +instance Outputable UnfoldingGuidance where + ppr UnfoldNever = ptext SLIT("NEVER") + ppr (UnfoldIfGoodArgs v cs size discount) + = hsep [ ptext SLIT("IF_ARGS"), int v, + brackets (hsep (map int cs)), + int size, + int discount ] +\end{code} + + +\begin{code} +calcUnfoldingGuidance + :: Int -- bomb out if size gets bigger than this + -> CoreExpr -- expression to look at + -> UnfoldingGuidance +calcUnfoldingGuidance bOMB_OUT_SIZE expr + = case collect_val_bndrs expr of { (inline, val_binders, body) -> + let + n_val_binders = length val_binders + + max_inline_size = n_val_binders+2 + -- The idea is that if there is an INLINE pragma (inline is True) + -- and there's a big body, we give a size of n_val_binders+2. This + -- This is just enough to fail the no-size-increase test in callSiteInline, + -- so that INLINE things don't get inlined into entirely boring contexts, + -- but no more. + + in + case (sizeExpr (iUnbox bOMB_OUT_SIZE) val_binders body) of + + TooBig + | not inline -> UnfoldNever + -- A big function with an INLINE pragma must + -- have an UnfoldIfGoodArgs guidance + | otherwise -> UnfoldIfGoodArgs n_val_binders + (map (const 0) val_binders) + max_inline_size 0 + + SizeIs size cased_args scrut_discount + -> UnfoldIfGoodArgs + n_val_binders + (map discount_for val_binders) + final_size + (iBox scrut_discount) + where + boxed_size = iBox size + + final_size | inline = boxed_size `min` max_inline_size + | otherwise = boxed_size + + -- Sometimes an INLINE thing is smaller than n_val_binders+2. + -- A particular case in point is a constructor, which has size 1. + -- We want to inline this regardless, hence the `min` + + discount_for b = foldlBag (\acc (b',n) -> if b==b' then acc+n else acc) + 0 cased_args + } + where + collect_val_bndrs e = go False [] e + -- We need to be a bit careful about how we collect the + -- value binders. In ptic, if we see + -- __inline_me (\x y -> e) + -- We want to say "2 value binders". Why? So that + -- we take account of information given for the arguments + + go inline rev_vbs (Note InlineMe e) = go True rev_vbs e + go inline rev_vbs (Lam b e) | isId b = go inline (b:rev_vbs) e + | otherwise = go inline rev_vbs e + go inline rev_vbs e = (inline, reverse rev_vbs, e) +\end{code} + +\begin{code} +sizeExpr :: Int# -- Bomb out if it gets bigger than this + -> [Id] -- Arguments; we're interested in which of these + -- get case'd + -> CoreExpr + -> ExprSize + +sizeExpr bOMB_OUT_SIZE top_args expr + = size_up expr + where + size_up (Type t) = sizeZero -- Types cost nothing + size_up (Var v) = sizeOne + + size_up (Note InlineMe body) = sizeOne -- Inline notes make it look very small + -- This can be important. If you have an instance decl like this: + -- instance Foo a => Foo [a] where + -- {-# INLINE op1, op2 #-} + -- op1 = ... + -- op2 = ... + -- then we'll get a dfun which is a pair of two INLINE lambdas + + size_up (Note _ body) = size_up body -- Other notes cost nothing + + size_up (App fun (Type t)) = size_up fun + size_up (App fun arg) = size_up_app fun [arg] + + size_up (Lit lit) = sizeN (litSize lit) + + size_up (Lam b e) | isId b = lamScrutDiscount (size_up e `addSizeN` 1) + | otherwise = size_up e + + size_up (Let (NonRec binder rhs) body) + = nukeScrutDiscount (size_up rhs) `addSize` + size_up body `addSizeN` + (if isUnLiftedType (idType binder) then 0 else 1) + -- For the allocation + -- If the binder has an unlifted type there is no allocation + + size_up (Let (Rec pairs) body) + = nukeScrutDiscount rhs_size `addSize` + size_up body `addSizeN` + length pairs -- For the allocation + where + rhs_size = foldr (addSize . size_up . snd) sizeZero pairs + + size_up (Case (Var v) _ _ alts) + | v `elem` top_args -- We are scrutinising an argument variable + = +{- I'm nuking this special case; BUT see the comment with case alternatives. + + (a) It's too eager. We don't want to inline a wrapper into a + context with no benefit. + E.g. \ x. f (x+x) no point in inlining (+) here! + + (b) It's ineffective. Once g's wrapper is inlined, its case-expressions + aren't scrutinising arguments any more + + case alts of + + [alt] -> size_up_alt alt `addSize` SizeIs 0# (unitBag (v, 1)) 0# + -- We want to make wrapper-style evaluation look cheap, so that + -- when we inline a wrapper it doesn't make call site (much) bigger + -- Otherwise we get nasty phase ordering stuff: + -- f x = g x x + -- h y = ...(f e)... + -- If we inline g's wrapper, f looks big, and doesn't get inlined + -- into h; if we inline f first, while it looks small, then g's + -- wrapper will get inlined later anyway. To avoid this nasty + -- ordering difference, we make (case a of (x,y) -> ...), + -- *where a is one of the arguments* look free. + + other -> +-} + alts_size (foldr addSize sizeOne alt_sizes) -- The 1 is for the scrutinee + (foldr1 maxSize alt_sizes) + + -- Good to inline if an arg is scrutinised, because + -- that may eliminate allocation in the caller + -- And it eliminates the case itself + + where + alt_sizes = map size_up_alt alts + + -- alts_size tries to compute a good discount for + -- the case when we are scrutinising an argument variable + alts_size (SizeIs tot tot_disc tot_scrut) -- Size of all alternatives + (SizeIs max max_disc max_scrut) -- Size of biggest alternative + = SizeIs tot (unitBag (v, iBox (_ILIT 1 +# tot -# max)) `unionBags` max_disc) max_scrut + -- If the variable is known, we produce a discount that + -- will take us back to 'max', the size of rh largest alternative + -- The 1+ is a little discount for reduced allocation in the caller + alts_size tot_size _ = tot_size + +-- gaw 2004 + size_up (Case e _ _ alts) = nukeScrutDiscount (size_up e) `addSize` + foldr (addSize . size_up_alt) sizeZero alts + -- We don't charge for the case itself + -- It's a strict thing, and the price of the call + -- is paid by scrut. Also consider + -- case f x of DEFAULT -> e + -- This is just ';'! Don't charge for it. + + ------------ + size_up_app (App fun arg) args + | isTypeArg arg = size_up_app fun args + | otherwise = size_up_app fun (arg:args) + size_up_app fun args = foldr (addSize . nukeScrutDiscount . size_up) + (size_up_fun fun args) + args + + -- A function application with at least one value argument + -- so if the function is an argument give it an arg-discount + -- + -- Also behave specially if the function is a build + -- + -- Also if the function is a constant Id (constr or primop) + -- compute discounts specially + size_up_fun (Var fun) args + | fun `hasKey` buildIdKey = buildSize + | fun `hasKey` augmentIdKey = augmentSize + | otherwise + = case globalIdDetails fun of + DataConWorkId dc -> conSizeN dc (valArgCount args) + + FCallId fc -> sizeN opt_UF_DearOp + PrimOpId op -> primOpSize op (valArgCount args) + -- foldr addSize (primOpSize op) (map arg_discount args) + -- At one time I tried giving an arg-discount if a primop + -- is applied to one of the function's arguments, but it's + -- not good. At the moment, any unlifted-type arg gets a + -- 'True' for 'yes I'm evald', so we collect the discount even + -- if we know nothing about it. And just having it in a primop + -- doesn't help at all if we don't know something more. + + other -> fun_discount fun `addSizeN` + (1 + length (filter (not . exprIsTrivial) args)) + -- The 1+ is for the function itself + -- Add 1 for each non-trivial arg; + -- the allocation cost, as in let(rec) + -- Slight hack here: for constructors the args are almost always + -- trivial; and for primops they are almost always prim typed + -- We should really only count for non-prim-typed args in the + -- general case, but that seems too much like hard work + + size_up_fun other args = size_up other + + ------------ + size_up_alt (con, bndrs, rhs) = size_up rhs + -- Don't charge for args, so that wrappers look cheap + -- (See comments about wrappers with Case) + + ------------ + -- We want to record if we're case'ing, or applying, an argument + fun_discount v | v `elem` top_args = SizeIs 0# (unitBag (v, opt_UF_FunAppDiscount)) 0# + fun_discount other = sizeZero + + ------------ + -- These addSize things have to be here because + -- I don't want to give them bOMB_OUT_SIZE as an argument + + addSizeN TooBig _ = TooBig + addSizeN (SizeIs n xs d) m = mkSizeIs bOMB_OUT_SIZE (n +# iUnbox m) xs d + + addSize TooBig _ = TooBig + addSize _ TooBig = TooBig + addSize (SizeIs n1 xs d1) (SizeIs n2 ys d2) + = mkSizeIs bOMB_OUT_SIZE (n1 +# n2) (xs `unionBags` ys) (d1 +# d2) +\end{code} + +Code for manipulating sizes + +\begin{code} +data ExprSize = TooBig + | SizeIs FastInt -- Size found + (Bag (Id,Int)) -- Arguments cased herein, and discount for each such + FastInt -- Size to subtract if result is scrutinised + -- by a case expression + +-- subtract the discount before deciding whether to bale out. eg. we +-- want to inline a large constructor application into a selector: +-- tup = (a_1, ..., a_99) +-- x = case tup of ... +-- +mkSizeIs max n xs d | (n -# d) ># max = TooBig + | otherwise = SizeIs n xs d + +maxSize TooBig _ = TooBig +maxSize _ TooBig = TooBig +maxSize s1@(SizeIs n1 _ _) s2@(SizeIs n2 _ _) | n1 ># n2 = s1 + | otherwise = s2 + +sizeZero = SizeIs (_ILIT 0) emptyBag (_ILIT 0) +sizeOne = SizeIs (_ILIT 1) emptyBag (_ILIT 0) +sizeN n = SizeIs (iUnbox n) emptyBag (_ILIT 0) +conSizeN dc n + | isUnboxedTupleCon dc = SizeIs (_ILIT 0) emptyBag (iUnbox n +# _ILIT 1) + | otherwise = SizeIs (_ILIT 1) emptyBag (iUnbox n +# _ILIT 1) + -- Treat constructors as size 1; we are keen to expose them + -- (and we charge separately for their args). We can't treat + -- them as size zero, else we find that (iBox x) has size 1, + -- which is the same as a lone variable; and hence 'v' will + -- always be replaced by (iBox x), where v is bound to iBox x. + -- + -- However, unboxed tuples count as size zero + -- I found occasions where we had + -- f x y z = case op# x y z of { s -> (# s, () #) } + -- and f wasn't getting inlined + +primOpSize op n_args + | not (primOpIsDupable op) = sizeN opt_UF_DearOp + | not (primOpOutOfLine op) = sizeN (2 - n_args) + -- Be very keen to inline simple primops. + -- We give a discount of 1 for each arg so that (op# x y z) costs 2. + -- We can't make it cost 1, else we'll inline let v = (op# x y z) + -- at every use of v, which is excessive. + -- + -- A good example is: + -- let x = +# p q in C {x} + -- Even though x get's an occurrence of 'many', its RHS looks cheap, + -- and there's a good chance it'll get inlined back into C's RHS. Urgh! + | otherwise = sizeOne + +buildSize = SizeIs (-2#) emptyBag 4# + -- We really want to inline applications of build + -- build t (\cn -> e) should cost only the cost of e (because build will be inlined later) + -- Indeed, we should add a result_discount becuause build is + -- very like a constructor. We don't bother to check that the + -- build is saturated (it usually is). The "-2" discounts for the \c n, + -- The "4" is rather arbitrary. + +augmentSize = SizeIs (-2#) emptyBag 4# + -- Ditto (augment t (\cn -> e) ys) should cost only the cost of + -- e plus ys. The -2 accounts for the \cn + +nukeScrutDiscount (SizeIs n vs d) = SizeIs n vs 0# +nukeScrutDiscount TooBig = TooBig + +-- When we return a lambda, give a discount if it's used (applied) +lamScrutDiscount (SizeIs n vs d) = case opt_UF_FunAppDiscount of { d -> SizeIs n vs (iUnbox d) } +lamScrutDiscount TooBig = TooBig +\end{code} + + +%************************************************************************ +%* * +\subsection[considerUnfolding]{Given all the info, do (not) do the unfolding} +%* * +%************************************************************************ + +We have very limited information about an unfolding expression: (1)~so +many type arguments and so many value arguments expected---for our +purposes here, we assume we've got those. (2)~A ``size'' or ``cost,'' +a single integer. (3)~An ``argument info'' vector. For this, what we +have at the moment is a Boolean per argument position that says, ``I +will look with great favour on an explicit constructor in this +position.'' (4)~The ``discount'' to subtract if the expression +is being scrutinised. + +Assuming we have enough type- and value arguments (if not, we give up +immediately), then we see if the ``discounted size'' is below some +(semi-arbitrary) threshold. It works like this: for every argument +position where we're looking for a constructor AND WE HAVE ONE in our +hands, we get a (again, semi-arbitrary) discount [proportion to the +number of constructors in the type being scrutinized]. + +If we're in the context of a scrutinee ( \tr{(case <expr > of A .. -> ...;.. )}) +and the expression in question will evaluate to a constructor, we use +the computed discount size *for the result only* rather than +computing the argument discounts. Since we know the result of +the expression is going to be taken apart, discounting its size +is more accurate (see @sizeExpr@ above for how this discount size +is computed). + +We use this one to avoid exporting inlinings that we ``couldn't possibly +use'' on the other side. Can be overridden w/ flaggery. +Just the same as smallEnoughToInline, except that it has no actual arguments. + +\begin{code} +couldBeSmallEnoughToInline :: Int -> CoreExpr -> Bool +couldBeSmallEnoughToInline threshold rhs = case calcUnfoldingGuidance threshold rhs of + UnfoldNever -> False + other -> True + +certainlyWillInline :: Unfolding -> Bool + -- Sees if the unfolding is pretty certain to inline +certainlyWillInline (CoreUnfolding _ _ _ is_cheap (UnfoldIfGoodArgs n_vals _ size _)) + = is_cheap && size - (n_vals +1) <= opt_UF_UseThreshold +certainlyWillInline other + = False + +smallEnoughToInline :: Unfolding -> Bool +smallEnoughToInline (CoreUnfolding _ _ _ _ (UnfoldIfGoodArgs _ _ size _)) + = size <= opt_UF_UseThreshold +smallEnoughToInline other + = False +\end{code} + +%************************************************************************ +%* * +\subsection{callSiteInline} +%* * +%************************************************************************ + +This is the key function. It decides whether to inline a variable at a call site + +callSiteInline is used at call sites, so it is a bit more generous. +It's a very important function that embodies lots of heuristics. +A non-WHNF can be inlined if it doesn't occur inside a lambda, +and occurs exactly once or + occurs once in each branch of a case and is small + +If the thing is in WHNF, there's no danger of duplicating work, +so we can inline if it occurs once, or is small + +NOTE: we don't want to inline top-level functions that always diverge. +It just makes the code bigger. Tt turns out that the convenient way to prevent +them inlining is to give them a NOINLINE pragma, which we do in +StrictAnal.addStrictnessInfoToTopId + +\begin{code} +callSiteInline :: DynFlags + -> Bool -- True <=> the Id can be inlined + -> Bool -- 'inline' note at call site + -> OccInfo + -> Id -- The Id + -> [Bool] -- One for each value arg; True if it is interesting + -> Bool -- True <=> continuation is interesting + -> Maybe CoreExpr -- Unfolding, if any + + +callSiteInline dflags active_inline inline_call occ id arg_infos interesting_cont + = case idUnfolding id of { + NoUnfolding -> Nothing ; + OtherCon cs -> Nothing ; + + CompulsoryUnfolding unf_template -> Just unf_template ; + -- CompulsoryUnfolding => there is no top-level binding + -- for these things, so we must inline it. + -- Only a couple of primop-like things have + -- compulsory unfoldings (see MkId.lhs). + -- We don't allow them to be inactive + + CoreUnfolding unf_template is_top is_value is_cheap guidance -> + + let + result | yes_or_no = Just unf_template + | otherwise = Nothing + + n_val_args = length arg_infos + + yes_or_no + | not active_inline = False + | otherwise = case occ of + IAmDead -> pprTrace "callSiteInline: dead" (ppr id) False + IAmALoopBreaker -> False + --OneOcc in_lam _ _ -> (not in_lam || is_cheap) && consider_safe True + other -> is_cheap && consider_safe False + -- we consider even the once-in-one-branch + -- occurrences, because they won't all have been + -- caught by preInlineUnconditionally. In particular, + -- if the occurrence is once inside a lambda, and the + -- rhs is cheap but not a manifest lambda, then + -- pre-inline will not have inlined it for fear of + -- invalidating the occurrence info in the rhs. + + consider_safe once + -- consider_safe decides whether it's a good idea to + -- inline something, given that there's no + -- work-duplication issue (the caller checks that). + | inline_call = True + + | otherwise + = case guidance of + UnfoldNever -> False + UnfoldIfGoodArgs n_vals_wanted arg_discounts size res_discount + + | enough_args && size <= (n_vals_wanted + 1) + -- Inline unconditionally if there no size increase + -- Size of call is n_vals_wanted (+1 for the function) + -> True + + | otherwise + -> some_benefit && small_enough + + where + some_benefit = or arg_infos || really_interesting_cont || + (not is_top && ({- once || -} (n_vals_wanted > 0 && enough_args))) + -- [was (once && not in_lam)] + -- If it occurs more than once, there must be + -- something interesting about some argument, or the + -- result context, to make it worth inlining + -- + -- If a function has a nested defn we also record + -- some-benefit, on the grounds that we are often able + -- to eliminate the binding, and hence the allocation, + -- for the function altogether; this is good for join + -- points. But this only makes sense for *functions*; + -- inlining a constructor doesn't help allocation + -- unless the result is scrutinised. UNLESS the + -- constructor occurs just once, albeit possibly in + -- multiple case branches. Then inlining it doesn't + -- increase allocation, but it does increase the + -- chance that the constructor won't be allocated at + -- all in the branches that don't use it. + + enough_args = n_val_args >= n_vals_wanted + really_interesting_cont | n_val_args < n_vals_wanted = False -- Too few args + | n_val_args == n_vals_wanted = interesting_cont + | otherwise = True -- Extra args + -- really_interesting_cont tells if the result of the + -- call is in an interesting context. + + small_enough = (size - discount) <= opt_UF_UseThreshold + discount = computeDiscount n_vals_wanted arg_discounts res_discount + arg_infos really_interesting_cont + + in + if dopt Opt_D_dump_inlinings dflags then + pprTrace "Considering inlining" + (ppr id <+> vcat [text "active:" <+> ppr active_inline, + text "occ info:" <+> ppr occ, + text "arg infos" <+> ppr arg_infos, + text "interesting continuation" <+> ppr interesting_cont, + text "is value:" <+> ppr is_value, + text "is cheap:" <+> ppr is_cheap, + text "guidance" <+> ppr guidance, + text "ANSWER =" <+> if yes_or_no then text "YES" else text "NO"]) + result + else + result + } + +computeDiscount :: Int -> [Int] -> Int -> [Bool] -> Bool -> Int +computeDiscount n_vals_wanted arg_discounts res_discount arg_infos result_used + -- We multiple the raw discounts (args_discount and result_discount) + -- ty opt_UnfoldingKeenessFactor because the former have to do with + -- *size* whereas the discounts imply that there's some extra + -- *efficiency* to be gained (e.g. beta reductions, case reductions) + -- by inlining. + + -- we also discount 1 for each argument passed, because these will + -- reduce with the lambdas in the function (we count 1 for a lambda + -- in size_up). + = 1 + -- Discount of 1 because the result replaces the call + -- so we count 1 for the function itself + length (take n_vals_wanted arg_infos) + + -- Discount of 1 for each arg supplied, because the + -- result replaces the call + round (opt_UF_KeenessFactor * + fromIntegral (arg_discount + result_discount)) + where + arg_discount = sum (zipWith mk_arg_discount arg_discounts arg_infos) + + mk_arg_discount discount is_evald | is_evald = discount + | otherwise = 0 + + -- Don't give a result discount unless there are enough args + result_discount | result_used = res_discount -- Over-applied, or case scrut + | otherwise = 0 +\end{code} |