{-
(c) The AQUA Project, Glasgow University, 1993-1998
\section{Common subexpression}
-}
module GHC.Core.Opt.CSE (cseProgram, cseOneExpr) where
import GHC.Prelude
import GHC.Core.Subst
import GHC.Types.Var ( Var )
import GHC.Types.Var.Env ( mkInScopeSet )
import GHC.Types.Id ( Id, idType, idHasRules, zapStableUnfolding
, idInlineActivation, setInlineActivation
, zapIdOccInfo, zapIdUsageInfo, idInlinePragma
, isJoinId, isJoinId_maybe, idUnfolding )
import GHC.Core.Utils ( mkAltExpr
, exprIsTickedString
, stripTicksE, stripTicksT, mkTicks )
import GHC.Core.FVs ( exprFreeVars )
import GHC.Core.Type ( tyConAppArgs )
import GHC.Core
import GHC.Utils.Outputable
import GHC.Types.Basic
import GHC.Types.Tickish
import GHC.Core.Map.Expr
import GHC.Utils.Misc ( filterOut, equalLength )
import GHC.Utils.Panic
import Data.Functor.Identity ( Identity (..) )
import Data.List ( mapAccumL )
{-
Simple common sub-expression
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
When we see
x1 = C a b
x2 = C x1 b
we build up a reverse mapping: C a b -> x1
C x1 b -> x2
and apply that to the rest of the program.
When we then see
y1 = C a b
y2 = C y1 b
we replace the C a b with x1. But then we *dont* want to
add x1 -> y1 to the mapping. Rather, we want the reverse, y1 -> x1
so that a subsequent binding
y2 = C y1 b
will get transformed to C x1 b, and then to x2.
So we carry an extra var->var substitution which we apply *before* looking up in the
reverse mapping.
Note [Shadowing]
~~~~~~~~~~~~~~~~
We have to be careful about shadowing.
For example, consider
f = \x -> let y = x+x in
h = \x -> x+x
in ...
Here we must *not* do CSE on the inner x+x! The simplifier used to guarantee no
shadowing, but it doesn't any more (it proved too hard), so we clone as we go.
We can simply add clones to the substitution already described.
A similar tricky situation is this, with x_123 and y_123 sharing the same unique:
let x_123 = e1 in
let y_123 = e2 in
let foo = e1
Naively applying e1 = x_123 during CSE we would get:
let x_123 = e1 in
let y_123 = e2 in
let foo = x_123
But x_123 is shadowed by y_123 and things would go terribly wrong! One more reason
why we have to substitute binders as we go so we will properly get:
let x1 = e1 in
let x2 = e2 in
let foo = x1
Note [CSE for bindings]
~~~~~~~~~~~~~~~~~~~~~~~
Let-bindings have two cases, implemented by extendCSEnvWithBinding.
* SUBSTITUTE: applies when the RHS is a variable
let x = y in ...(h x)....
Here we want to extend the /substitution/ with x -> y, so that the
(h x) in the body might CSE with an enclosing (let v = h y in ...).
NB: the substitution maps InIds, so we extend the substitution with
a binding for the original InId 'x'
How can we have a variable on the RHS? Doesn't the simplifier inline them?
- First, the original RHS might have been (g z) which has CSE'd
with an enclosing (let y = g z in ...). This is super-important.
See #5996:
x1 = C a b
x2 = C x1 b
y1 = C a b
y2 = C y1 b
Here we CSE y1's rhs to 'x1', and then we must add (y1->x1) to
the substitution so that we can CSE the binding for y2.
- Second, we use extendCSEnvWithBinding for case expression scrutinees too;
see Note [CSE for case expressions]
* EXTEND THE REVERSE MAPPING: applies in all other cases
let x = h y in ...(h y)...
Here we want to extend the /reverse mapping (cs_map)/ so that
we CSE the (h y) call to x.
Note that we use EXTEND even for a trivial expression, provided it
is not a variable or literal. In particular this /includes/ type
applications. This can be important (#13156); e.g.
case f @ Int of { r1 ->
case f @ Int of { r2 -> ...
Here we want to common-up the two uses of (f @ Int) so we can
remove one of the case expressions.
See also Note [Corner case for case expressions] for another
reason not to use SUBSTITUTE for all trivial expressions.
Notice that
- The SUBSTITUTE situation extends the substitution (cs_subst)
- The EXTEND situation extends the reverse mapping (cs_map)
Notice also that in the SUBSTITUTE case we leave behind a binding
x = y
even though we /also/ carry a substitution x -> y. Can we just drop
the binding instead? Well, not at top level! See Note [Top level and
postInlineUnconditionally] in GHC.Core.Opt.Simplify.Utils; and in any
case CSE applies only to the /bindings/ of the program, and we leave
it to the simplifier to propagate effects to the RULES. Finally, it
doesn't seem worth the effort to discard the nested bindings because
the simplifier will do it next.
Note [CSE for case expressions]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
case scrut_expr of x { ...alts... }
This is very like a strict let-binding
let !x = scrut_expr in ...
So we use (extendCSEnvWithBinding x scrut_expr) to process scrut_expr and x, and as a
result all the stuff under Note [CSE for bindings] applies directly.
For example:
* Trivial scrutinee
f = \x -> case x of wild {
(a:as) -> case a of wild1 {
(p,q) -> ...(wild1:as)...
Here, (wild1:as) is morally the same as (a:as) and hence equal to
wild. But that's not quite obvious. In the rest of the compiler we
want to keep it as (wild1:as), but for CSE purpose that's a bad
idea.
By using extendCSEnvWithBinding we add the binding (wild1 -> a) to the substitution,
which does exactly the right thing.
(Notice this is exactly backwards to what the simplifier does, which
is to try to replaces uses of 'a' with uses of 'wild1'.)
This is the main reason that extendCSEnvWithBinding is called with a trivial rhs.
* Non-trivial scrutinee
case (f x) of y { pat -> ...let z = f x in ... }
By using extendCSEnvWithBinding we'll add (f x :-> y) to the cs_map, and
thereby CSE the inner (f x) to y.
Note [CSE for INLINE and NOINLINE]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
There are some subtle interactions of CSE with functions that the user
has marked as INLINE or NOINLINE. (Examples from Roman Leshchinskiy.)
Consider
yes :: Int {-# NOINLINE yes #-}
yes = undefined
no :: Int {-# NOINLINE no #-}
no = undefined
foo :: Int -> Int -> Int {-# NOINLINE foo #-}
foo m n = n
{-# RULES "foo/no" foo no = id #-}
bar :: Int -> Int
bar = foo yes
We do not expect the rule to fire. But if we do CSE, then we risk
getting yes=no, and the rule does fire. Actually, it won't because
NOINLINE means that 'yes' will never be inlined, not even if we have
yes=no. So that's fine (now; perhaps in the olden days, yes=no would
have substituted even if 'yes' was NOINLINE).
But we do need to take care. Consider
{-# NOINLINE bar #-}
bar = <rhs> -- Same rhs as foo
foo = <rhs>
If CSE produces
foo = bar
then foo will never be inlined to <rhs> (when it should be, if <rhs>
is small). The conclusion here is this:
We should not add
<rhs> :-> bar
to the CSEnv if 'bar' has any constraints on when it can inline;
that is, if its 'activation' not always active. Otherwise we
might replace <rhs> by 'bar', and then later be unable to see that it
really was <rhs>.
An exception to the rule is when the INLINE pragma is not from the user, e.g. from
WorkWrap (see Note [Wrapper activation]). We can tell because noUserInlineSpec
is then true.
Note that we do not (currently) do CSE on the unfolding stored inside
an Id, even if it is a 'stable' unfolding. That means that when an
unfolding happens, it is always faithful to what the stable unfolding
originally was.
Note [CSE for stable unfoldings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
{-# Unf = Stable (\pq. build blah) #-}
foo = x
Here 'foo' has a stable unfolding, but its (optimised) RHS is trivial.
(Turns out that this actually happens for the enumFromTo method of
the Integer instance of Enum in GHC.Enum.) Suppose moreover that foo's
stable unfolding originates from an INLINE or INLINEABLE pragma on foo.
Then we obviously do NOT want to extend the substitution with (foo->x),
because we promised to inline foo as what the user wrote. See similar Note
[Stable unfoldings and postInlineUnconditionally] in GHC.Core.Opt.Simplify.Utils.
Nor do we want to change the reverse mapping. Suppose we have
foo {-# Unf = Stable (\pq. build blah) #-}
= <expr>
bar = <expr>
There could conceivably be merit in rewriting the RHS of bar:
bar = foo
but now bar's inlining behaviour will change, and importing
modules might see that. So it seems dodgy and we don't do it.
Wrinkles
* Stable unfoldings are also created during worker/wrapper when we
decide that a function's definition is so small that it should
always inline, or indeed for the wrapper function itself. In this
case we still want to do CSE (#13340). Hence the use of
isStableUserUnfolding/isStableSystemUnfolding rather than
isStableUnfolding.
* Consider
foo = <expr>
bar {-# Unf = Stable ... #-}
= <expr>
where the unfolding was added by strictness analysis, say. Then
CSE goes ahead, so we get
bar = foo
and probably use SUBSTITUTE that will make 'bar' dead. But just
possibly not -- see Note [Dealing with ticks]. In that case we might
be left with
bar = tick t1 (tick t2 foo)
in which case we would really like to get rid of the stable unfolding
(generated by the strictness analyser, say).
Hence the zapStableUnfolding in cse_bind. Not a big deal, and only
makes a difference when ticks get into the picture.
Note [Corner case for case expressions]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Here is another reason that we do not use SUBSTITUTE for
all trivial expressions. Consider
case x |> co of (y::Array# Int) { ... }
We do not want to extend the substitution with (y -> x |> co); since y
is of unlifted type, this would destroy the let-can-float invariant if
(x |> co) was not ok-for-speculation.
But surely (x |> co) is ok-for-speculation, because it's a trivial
expression, and x's type is also unlifted, presumably. Well, maybe
not if you are using unsafe casts. I actually found a case where we
had
(x :: HValue) |> (UnsafeCo :: HValue ~ Array# Int)
Note [CSE for join points?]
~~~~~~~~~~~~~~~~~~~~~~~~~~~
We must not be naive about join points in CSE:
join j = e in
if b then jump j else 1 + e
The expression (1 + jump j) is not good (see Note [Invariants on join points] in
GHC.Core). This seems to come up quite seldom, but it happens (first seen
compiling ppHtml in Haddock.Backends.Xhtml).
We could try and be careful by tracking which join points are still valid at
each subexpression, but since join points aren't allocated or shared, there's
less to gain by trying to CSE them. (#13219)
Note [Look inside join-point binders]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Another way how CSE for join points is tricky is
let join foo x = (x, 42)
join bar x = (x, 42)
in … jump foo 1 … jump bar 2 …
naively, CSE would turn this into
let join foo x = (x, 42)
join bar = foo
in … jump foo 1 … jump bar 2 …
but now bar is a join point that claims arity one, but its right-hand side
is not a lambda, breaking the join-point invariant (this was #15002).
So `cse_bind` must zoom past the lambdas of a join point (using
`collectNBinders`) and resume searching for CSE opportunities only in
the body of the join point.
Note [CSE for recursive bindings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
f = \x ... f....
g = \y ... g ...
where the "..." are identical. Could we CSE them? In full generality
with mutual recursion it's quite hard; but for self-recursive bindings
(which are very common) it's rather easy:
* Maintain a separate cs_rec_map, that maps
(\f. (\x. ...f...) ) -> f
Note the \f in the domain of the mapping!
* When we come across the binding for 'g', look up (\g. (\y. ...g...))
Bingo we get a hit. So we can replace the 'g' binding with
g = f
We can't use cs_map for this, because the key isn't an expression of
the program; it's a kind of synthetic key for recursive bindings.
Note [Separate envs for let rhs and body]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Substituting occurrences of the binder in the rhs with the
renamed binder is wrong for non-recursive bindings. Why?
Consider this core.
let {x_123 = e} in
let {y_123 = \eta0 -> x_123} in ...
In the second line the y_123 on the lhs and x_123 on the rhs refer to different binders
even if they share the same unique.
If we apply the substitution `123 => x2_124}` to both the lhs and rhs we will transform
`let y_123 = \eta0 -> x_123` into `let x2_124 = \eta0 -> x2_124`.
However x2_124 on the rhs is not in scope and really shouldn't have been renamed at all.
Because really this should still be x_123! In fact this exact thing happened in #21685.
To fix this we pass two different cse envs to cse_bind. One we use the cse the rhs of the binding.
And one we update with the result of cseing the rhs which we then use going forward for the
body/rest of the module.
************************************************************************
* *
\section{Common subexpression}
* *
************************************************************************
-}
cseProgram :: CoreProgram -> CoreProgram
cseProgram binds = snd (mapAccumL (cseBind TopLevel) emptyCSEnv binds)
cseBind :: TopLevelFlag -> CSEnv -> CoreBind -> (CSEnv, CoreBind)
cseBind toplevel env (NonRec b e)
= (env2, NonRec b2 e2)
where
-- See Note [Separate envs for let rhs and body]
(env1, b1) = addBinder env b
(env2, (b2, e2)) = cse_bind toplevel env env1 (b,e) b1
cseBind toplevel env (Rec [(in_id, rhs)])
| noCSE in_id
= (env1, Rec [(out_id, rhs')])
-- See Note [CSE for recursive bindings]
| Just previous <- lookupCSRecEnv env out_id rhs''
, let previous' = mkTicks ticks previous
out_id' = delayInlining toplevel out_id
= -- We have a hit in the recursive-binding cache
(extendCSSubst env1 in_id previous', NonRec out_id' previous')
| otherwise
= (extendCSRecEnv env1 out_id rhs'' id_expr', Rec [(zapped_id, rhs')])
where
(env1, Identity out_id) = addRecBinders env (Identity in_id)
rhs' = cseExpr env1 rhs
rhs'' = stripTicksE tickishFloatable rhs'
ticks = stripTicksT tickishFloatable rhs'
id_expr' = varToCoreExpr out_id
zapped_id = zapIdUsageInfo out_id
cseBind toplevel env (Rec pairs)
= (env2, Rec pairs')
where
(env1, bndrs1) = addRecBinders env (map fst pairs)
(env2, pairs') = mapAccumL do_one env1 (zip pairs bndrs1)
do_one env (pr, b1) = cse_bind toplevel env env pr b1
-- | Given a binding of @in_id@ to @in_rhs@, and a fresh name to refer
-- to @in_id@ (@out_id@, created from addBinder or addRecBinders),
-- first try to CSE @in_rhs@, and then add the resulting (possibly CSE'd)
-- binding to the 'CSEnv', so that we attempt to CSE any expressions
-- which are equal to @out_rhs@.
-- We use a different env for cse on the rhs and for extendCSEnvWithBinding
-- for reasons explain in See Note [Separate envs for let rhs and body]
cse_bind :: TopLevelFlag -> CSEnv -> CSEnv -> (InId, InExpr) -> OutId -> (CSEnv, (OutId, OutExpr))
cse_bind toplevel env_rhs env_body (in_id, in_rhs) out_id
| isTopLevel toplevel, exprIsTickedString in_rhs
-- See Note [Take care with literal strings]
= (env_body', (out_id', in_rhs))
| Just arity <- isJoinId_maybe out_id
-- See Note [Look inside join-point binders]
= let (params, in_body) = collectNBinders arity in_rhs
(env', params') = addBinders env_rhs params
out_body = tryForCSE env' in_body
in (env_body , (out_id, mkLams params' out_body))
| otherwise
= (env_body', (out_id'', out_rhs))
where
(env_body', out_id') = extendCSEnvWithBinding env_body in_id out_id out_rhs cse_done
(cse_done, out_rhs) = try_for_cse env_rhs in_rhs
out_id'' | cse_done = zapStableUnfolding $
delayInlining toplevel out_id'
| otherwise = out_id'
delayInlining :: TopLevelFlag -> Id -> Id
-- Add a NOINLINE[2] if the Id doesn't have an INLNE pragma already
-- See Note [Delay inlining after CSE]
delayInlining top_lvl bndr
| isTopLevel top_lvl
, isAlwaysActive (idInlineActivation bndr)
, idHasRules bndr -- Only if the Id has some RULES,
-- which might otherwise get lost
-- These rules are probably auto-generated specialisations,
-- since Ids with manual rules usually have manually-inserted
-- delayed inlining anyway
= bndr `setInlineActivation` activateAfterInitial
| otherwise
= bndr
extendCSEnvWithBinding
:: CSEnv -- Includes InId->OutId cloning
-> InVar -- Could be a let-bound type
-> OutId -> OutExpr -- Processed binding
-> Bool -- True <=> RHS was CSE'd and is a variable
-- or maybe (Tick t variable)
-> (CSEnv, OutId) -- Final env, final bndr
-- Extend the CSE env with a mapping [rhs -> out-id]
-- unless we can instead just substitute [in-id -> rhs]
--
-- It's possible for the binder to be a type variable,
-- in which case we can just substitute.
-- See Note [CSE for bindings]
extendCSEnvWithBinding env in_id out_id rhs' cse_done
| not (isId out_id) = (extendCSSubst env in_id rhs', out_id)
| noCSE out_id = (env, out_id)
| use_subst = (extendCSSubst env in_id rhs', out_id)
| cse_done = (env, out_id)
-- See Note [Dealing with ticks]
| otherwise = (extendCSEnv env rhs' id_expr', zapped_id)
where
id_expr' = varToCoreExpr out_id
zapped_id = zapIdUsageInfo out_id
-- Putting the Id into the cs_map makes it possible that
-- it'll become shared more than it is now, which would
-- invalidate (the usage part of) its demand info.
-- This caused #100218.
-- Easiest thing is to zap the usage info; subsequently
-- performing late demand-analysis will restore it. Don't zap
-- the strictness info; it's not necessary to do so, and losing
-- it is bad for performance if you don't do late demand
-- analysis
-- Should we use SUBSTITUTE or EXTEND?
-- See Note [CSE for bindings]
use_subst | Var {} <- rhs' = True
| otherwise = False
-- | Given a binder `let x = e`, this function
-- determines whether we should add `e -> x` to the cs_map
noCSE :: InId -> Bool
noCSE id
| isJoinId id = no_cse -- See Note [CSE for join points?]
| isStableUserUnfolding unf = no_cse -- See Note [CSE for stable unfoldings]
| user_activation_control = no_cse -- See Note [CSE for INLINE and NOINLINE]
| otherwise = yes_cse
where
unf = idUnfolding id
user_activation_control = not (isAlwaysActive (idInlineActivation id))
&& not (noUserInlineSpec (inlinePragmaSpec (idInlinePragma id)))
yes_cse = False
no_cse = True
{- Note [Take care with literal strings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider this example:
x = "foo"#
y = "foo"#
...x...y...x...y....
We would normally turn this into:
x = "foo"#
y = x
...x...x...x...x....
But this breaks an invariant of Core, namely that the RHS of a top-level binding
of type Addr# must be a string literal, not another variable. See Note
[Core top-level string literals] in GHC.Core.
For this reason, we special case top-level bindings to literal strings and leave
the original RHS unmodified. This produces:
x = "foo"#
y = "foo"#
...x...x...x...x....
Now 'y' will be discarded as dead code, and we are done.
The net effect is that for the y-binding we want to
- Use SUBSTITUTE, by extending the substitution with y :-> x
- but leave the original binding for y undisturbed
This is done by cse_bind. I got it wrong the first time (#13367).
Note [Dealing with ticks]
~~~~~~~~~~~~~~~~~~~~~~~~~
Ticks complicate CSE a bit, as I discovered in the fallout from
fixing #19360.
* To get more CSE-ing, we strip all the tickishFloatable ticks from
an expression
- when inserting into the cs_map (see extendCSEnv)
- when looking up in the cs_map (see call to lookupCSEnv in try_for_cse)
Quite why only the tickishFloatable ticks, I'm not quite sure.
AK: I think we only do this for floatable ticks since generally we don't mind them
being less accurate as much. E.g. consider
case e of
C1 -> f (<tick1> e1)
C2 -> f (<tick2> e1)
If the ticks are (floatable) source notes nothing too bad happens if the debug info for
both branches says the code comes from the same source location. Even if it will be inaccurate
for one of the branches. We should probably still consider this worthwhile.
However if the ticks are cost centres we really don't want the cost of both branches to be
attributed to the same cost centre. Because a user might explicitly have inserted different
cost centres in order to distinguish between evaluations resulting from the two different branches.
e.g. something like this:
case e of
C1 -> f ({ SCC "evalAlt1"} e1)
C1 -> f ({ SCC "evalAlt2"} e1)
But it's still a bit suspicious.
* If we get a hit in cs_map, we wrap the result in the ticks from the
thing we are looking up (see try_for_cse)
Net result: if we get a hit, we might replace
let x = tick t1 (tick t2 e)
with
let x = tick t1 (tick t2 y)
where 'y' is the variable that 'e' maps to. Now consider extendCSEnvWithBinding for
the binding for 'x':
* We can't use SUBSTITUTE because those ticks might not be trivial (we
use tickishIsCode in exprIsTrivial)
* We should not use EXTEND, because we definitely don't want to
add (tick t1 (tick t2 y)) :-> x
to the cs_map. Remember we strip off the ticks, so that would amount
to adding y :-> x, very silly.
TL;DR: we do neither; hence the cse_done case in extendCSEnvWithBinding.
Note [Delay inlining after CSE]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Suppose (#15445) we have
f,g :: Num a => a -> a
f x = ...f (x-1).....
g y = ...g (y-1) ....
and we make some specialisations of 'g', either automatically, or via
a SPECIALISE pragma. Then CSE kicks in and notices that the RHSs of
'f' and 'g' are identical, so we get
f x = ...f (x-1)...
g = f
{-# RULES g @Int _ = $sg #-}
Now there is terrible danger that, in an importing module, we'll inline
'g' before we have a chance to run its specialisation!
Solution: during CSE, after a "hit" in the CSE cache
* when adding a binding
g = f
* for a top-level function g
* and g has specialisation RULES
add a NOINLINE[2] activation to it, to ensure it's not inlined
right away.
Notes:
* Why top level only? Because for nested bindings we are already past
phase 2 and will never return there.
* Why "only if g has RULES"? Because there is no point in
doing this if there are no RULES; and other things being
equal it delays optimisation to delay inlining (#17409)
---- Historical note ---
This patch is simpler and more direct than an earlier
version:
commit 2110738b280543698407924a16ac92b6d804dc36
Author: Simon Peyton Jones <simonpj@microsoft.com>
Date: Mon Jul 30 13:43:56 2018 +0100
Don't inline functions with RULES too early
We had to revert this patch because it made GHC itself slower.
Why? It delayed inlining of /all/ functions with RULES, and that was
very bad in GHC.Tc.Solver.Flatten.flatten_ty_con_app
* It delayed inlining of liftM
* That delayed the unravelling of the recursion in some dictionary
bindings.
* That delayed some eta expansion, leaving
flatten_ty_con_app = \x y. let <stuff> in \z. blah
* That allowed the float-out pass to put sguff between
the \y and \z.
* And that permanently stopped eta expansion of the function,
even once <stuff> was simplified.
-}
tryForCSE :: CSEnv -> InExpr -> OutExpr
tryForCSE env expr = snd (try_for_cse env expr)
try_for_cse :: CSEnv -> InExpr -> (Bool, OutExpr)
-- (False, e') => We did not CSE the entire expression,
-- but we might have CSE'd some sub-expressions,
-- yielding e'
--
-- (True, te') => We CSE'd the entire expression,
-- yielding the trivial expression te'
try_for_cse env expr
| Just e <- lookupCSEnv env expr'' = (True, mkTicks ticks e)
| otherwise = (False, expr')
-- The varToCoreExpr is needed if we have
-- case e of xco { ...case e of yco { ... } ... }
-- Then CSE will substitute yco -> xco;
-- but these are /coercion/ variables
where
expr' = cseExpr env expr
expr'' = stripTicksE tickishFloatable expr'
ticks = stripTicksT tickishFloatable expr'
-- We don't want to lose the source notes when a common sub
-- expression gets eliminated. Hence we push all (!) of them on
-- top of the replaced sub-expression. This is probably not too
-- useful in practice, but upholds our semantics.
-- | Runs CSE on a single expression.
--
-- This entry point is not used in the compiler itself, but is provided
-- as a convenient entry point for users of the GHC API.
cseOneExpr :: InExpr -> OutExpr
cseOneExpr e = cseExpr env e
where env = emptyCSEnv {cs_subst = mkEmptySubst (mkInScopeSet (exprFreeVars e)) }
cseExpr :: CSEnv -> InExpr -> OutExpr
cseExpr env (Type t) = Type (substTyUnchecked (csEnvSubst env) t)
cseExpr env (Coercion c) = Coercion (substCo (csEnvSubst env) c)
cseExpr _ (Lit lit) = Lit lit
cseExpr env (Var v) = lookupSubst env v
cseExpr env (App f a) = App (cseExpr env f) (tryForCSE env a)
cseExpr env (Tick t e) = Tick t (cseExpr env e)
cseExpr env (Cast e co) = Cast (tryForCSE env e) (substCo (csEnvSubst env) co)
cseExpr env (Lam b e) = let (env', b') = addBinder env b
in Lam b' (cseExpr env' e)
cseExpr env (Let bind e) = let (env', bind') = cseBind NotTopLevel env bind
in Let bind' (cseExpr env' e)
cseExpr env (Case e bndr ty alts) = cseCase env e bndr ty alts
cseCase :: CSEnv -> InExpr -> InId -> InType -> [InAlt] -> OutExpr
cseCase env scrut bndr ty alts
= Case scrut1 bndr3 ty' $
combineAlts (map cse_alt alts)
where
ty' = substTyUnchecked (csEnvSubst env) ty
(cse_done, scrut1) = try_for_cse env scrut
bndr1 = zapIdOccInfo bndr
-- Zapping the OccInfo is needed because the extendCSEnv
-- in cse_alt may mean that a dead case binder
-- becomes alive, and Lint rejects that
(env1, bndr2) = addBinder env bndr1
(alt_env, bndr3) = extendCSEnvWithBinding env1 bndr bndr2 scrut1 cse_done
-- extendCSEnvWithBinding: see Note [CSE for case expressions]
con_target :: OutExpr
con_target = lookupSubst alt_env bndr
arg_tys :: [OutType]
arg_tys = tyConAppArgs (idType bndr3)
-- See Note [CSE for case alternatives]
cse_alt (Alt (DataAlt con) args rhs)
= Alt (DataAlt con) args' (tryForCSE new_env rhs)
where
(env', args') = addBinders alt_env args
new_env = extendCSEnv env' con_expr con_target
con_expr = mkAltExpr (DataAlt con) args' arg_tys
cse_alt (Alt con args rhs)
= Alt con args' (tryForCSE env' rhs)
where
(env', args') = addBinders alt_env args
combineAlts :: [OutAlt] -> [OutAlt]
-- See Note [Combine case alternatives]
combineAlts alts
| (Just alt1, rest_alts) <- find_bndr_free_alt alts
, Alt _ bndrs1 rhs1 <- alt1
, let filtered_alts = filterOut (identical_alt rhs1) rest_alts
, not (equalLength rest_alts filtered_alts)
= assertPpr (null bndrs1) (ppr alts) $
Alt DEFAULT [] rhs1 : filtered_alts
| otherwise
= alts
where
find_bndr_free_alt :: [CoreAlt] -> (Maybe CoreAlt, [CoreAlt])
-- The (Just alt) is a binder-free alt
-- See Note [Combine case alts: awkward corner]
find_bndr_free_alt []
= (Nothing, [])
find_bndr_free_alt (alt@(Alt _ bndrs _) : alts)
| null bndrs = (Just alt, alts)
| otherwise = case find_bndr_free_alt alts of
(mb_bf, alts) -> (mb_bf, alt:alts)
identical_alt rhs1 (Alt _ _ rhs) = eqCoreExpr rhs1 rhs
-- Even if this alt has binders, they will have been cloned
-- If any of these binders are mentioned in 'rhs', then
-- 'rhs' won't compare equal to 'rhs1' (which is from an
-- alt with no binders).
{- Note [CSE for case alternatives]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider case e of x
K1 y -> ....(K1 y)...
K2 -> ....K2....
We definitely want to CSE that (K1 y) into just x.
But what about the lone K2? At first you would think "no" because
turning K2 into 'x' increases the number of live variables. But
* Turning K2 into x increases the chance of combining identical alts.
Example case xs of
(_:_) -> f xs
[] -> f []
See #17901 and simplCore/should_compile/T17901 for more examples
of this kind.
* The next run of the simplifier will turn 'x' back into K2, so we won't
permanently bloat the free-var count.
Note [Combine case alternatives]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
combineAlts is just a more heavyweight version of the use of
combineIdenticalAlts in GHC.Core.Opt.Simplify.Utils.prepareAlts. The basic idea is
to transform
DEFAULT -> e1
K x -> e1
W y z -> e2
===>
DEFAULT -> e1
W y z -> e2
In the simplifier we use cheapEqExpr, because it is called a lot.
But here in CSE we use the full eqCoreExpr. After all, two alternatives usually
differ near the root, so it probably isn't expensive to compare the full
alternative. It seems like the same kind of thing that CSE is supposed
to be doing, which is why I put it here.
I actually saw some examples in the wild, where some inlining made e1 too
big for cheapEqExpr to catch it.
Note [Combine case alts: awkward corner]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We would really like to check isDeadBinder on the binders in the
alternative. But alas, the simplifer zaps occ-info on binders in case
alternatives; see Note [Case alternative occ info] in GHC.Core.Opt.Simplify.
* One alternative (perhaps a good one) would be to do OccAnal
just before CSE. Then perhaps we could get rid of combineIdenticalAlts
in the Simplifier, which might save work.
* Another would be for CSE to return free vars as it goes.
* But the current solution is to find a nullary alternative (including
the DEFAULT alt, if any). This will not catch
case x of
A y -> blah
B z p -> blah
where no alternative is nullary or DEFAULT. But the current
solution is at least cheap.
************************************************************************
* *
\section{The CSE envt}
* *
************************************************************************
-}
data CSEnv
= CS { cs_subst :: Subst -- Maps InBndrs to OutExprs
-- The substitution variables to
-- /trivial/ OutExprs, not arbitrary expressions
, cs_map :: CoreMap OutExpr
-- The "reverse" mapping.
-- Maps a OutExpr to a /trivial/ OutExpr
-- The key of cs_map is stripped of all Ticks
-- It maps arbitrary expressions to trivial expressions
-- representing the same value. E.g @C a b@ to @x1@.
, cs_rec_map :: CoreMap OutExpr
-- See Note [CSE for recursive bindings]
}
emptyCSEnv :: CSEnv
emptyCSEnv = CS { cs_map = emptyCoreMap, cs_rec_map = emptyCoreMap
, cs_subst = emptySubst }
lookupCSEnv :: CSEnv -> OutExpr -> Maybe OutExpr
lookupCSEnv (CS { cs_map = csmap }) expr
= lookupCoreMap csmap expr
-- | @extendCSEnv env e triv_expr@ will replace any occurrence of @e@ with @triv_expr@ going forward.
extendCSEnv :: CSEnv -> OutExpr -> OutExpr -> CSEnv
extendCSEnv cse expr triv_expr
= cse { cs_map = extendCoreMap (cs_map cse) sexpr triv_expr }
where
sexpr = stripTicksE tickishFloatable expr
extendCSRecEnv :: CSEnv -> OutId -> OutExpr -> OutExpr -> CSEnv
-- See Note [CSE for recursive bindings]
extendCSRecEnv cse bndr expr triv_expr
= cse { cs_rec_map = extendCoreMap (cs_rec_map cse) (Lam bndr expr) triv_expr }
lookupCSRecEnv :: CSEnv -> OutId -> OutExpr -> Maybe OutExpr
-- See Note [CSE for recursive bindings]
lookupCSRecEnv (CS { cs_rec_map = csmap }) bndr expr
= lookupCoreMap csmap (Lam bndr expr)
csEnvSubst :: CSEnv -> Subst
csEnvSubst = cs_subst
lookupSubst :: CSEnv -> Id -> OutExpr
lookupSubst (CS { cs_subst = sub}) x = lookupIdSubst sub x
extendCSSubst :: CSEnv -> Id -> CoreExpr -> CSEnv
extendCSSubst cse x rhs = cse { cs_subst = extendSubst (cs_subst cse) x rhs }
-- | Add clones to the substitution to deal with shadowing. See
-- Note [Shadowing] for more details. You should call this whenever
-- you go under a binder.
addBinder :: CSEnv -> Var -> (CSEnv, Var)
addBinder cse v = (cse { cs_subst = sub' }, v')
where
(sub', v') = substBndr (cs_subst cse) v
addBinders :: CSEnv -> [Var] -> (CSEnv, [Var])
addBinders cse vs = (cse { cs_subst = sub' }, vs')
where
(sub', vs') = substBndrs (cs_subst cse) vs
addRecBinders :: Traversable f => CSEnv -> f Id -> (CSEnv, f Id)
addRecBinders = \ cse vs ->
let (sub', vs') = substRecBndrs (cs_subst cse) vs
in (cse { cs_subst = sub' }, vs')
{-# INLINE addRecBinders #-}