{-
(c) The AQUA Project, Glasgow University, 1993-1998

\section{Common subexpression}
-}



{-# OPTIONS_GHC -Wno-incomplete-record-updates #-}
{-# OPTIONS_GHC -Wno-incomplete-uni-patterns   #-}

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 )
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.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 propate 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 except 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.

Stable unfoldings are also created during worker/wrapper when we decide
that a function's definition is so small that it should always inline.
In this case we still want to do CSE (#13340). Hence the use of
isAnyInlinePragma rather than isStableUnfolding.

Now 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/app 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 occurances 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 :: CoreProgram -> CoreProgram
cseProgram CoreProgram
binds = (CSEnv, CoreProgram) -> CoreProgram
forall a b. (a, b) -> b
snd ((CSEnv -> CoreBind -> (CSEnv, CoreBind))
-> CSEnv -> CoreProgram -> (CSEnv, CoreProgram)
forall (t :: * -> *) s a b.
Traversable t =>
(s -> a -> (s, b)) -> s -> t a -> (s, t b)
mapAccumL (TopLevelFlag -> CSEnv -> CoreBind -> (CSEnv, CoreBind)
cseBind TopLevelFlag
TopLevel) CSEnv
emptyCSEnv CoreProgram
binds)

cseBind :: TopLevelFlag -> CSEnv -> CoreBind -> (CSEnv, CoreBind)
cseBind :: TopLevelFlag -> CSEnv -> CoreBind -> (CSEnv, CoreBind)
cseBind TopLevelFlag
toplevel CSEnv
env (NonRec Var
b Expr Var
e)
  = (CSEnv
env2, Var -> Expr Var -> CoreBind
forall b. b -> Expr b -> Bind b
NonRec Var
b2 Expr Var
e2)
  where
    -- See Note [Separate envs for let rhs and body]
    (CSEnv
env1, Var
b1)       = CSEnv -> Var -> (CSEnv, Var)
addBinder CSEnv
env Var
b
    (CSEnv
env2, (Var
b2, Expr Var
e2)) = TopLevelFlag
-> CSEnv
-> CSEnv
-> (Var, Expr Var)
-> Var
-> (CSEnv, (Var, Expr Var))
cse_bind TopLevelFlag
toplevel CSEnv
env CSEnv
env1 (Var
b,Expr Var
e) Var
b1

cseBind TopLevelFlag
toplevel CSEnv
env (Rec [(Var
in_id, Expr Var
rhs)])
  | Var -> Bool
noCSE Var
in_id
  = (CSEnv
env1, [(Var, Expr Var)] -> CoreBind
forall b. [(b, Expr b)] -> Bind b
Rec [(Var
out_id, Expr Var
rhs')])

  -- See Note [CSE for recursive bindings]
  | Just Expr Var
previous <- CSEnv -> Var -> Expr Var -> Maybe (Expr Var)
lookupCSRecEnv CSEnv
env Var
out_id Expr Var
rhs''
  , let previous' :: Expr Var
previous' = [CoreTickish] -> Expr Var -> Expr Var
mkTicks [CoreTickish]
ticks Expr Var
previous
        out_id' :: Var
out_id'   = TopLevelFlag -> Var -> Var
delayInlining TopLevelFlag
toplevel Var
out_id
  = -- We have a hit in the recursive-binding cache
    (CSEnv -> Var -> Expr Var -> CSEnv
extendCSSubst CSEnv
env1 Var
in_id Expr Var
previous', Var -> Expr Var -> CoreBind
forall b. b -> Expr b -> Bind b
NonRec Var
out_id' Expr Var
previous')

  | Bool
otherwise
  = (CSEnv -> Var -> Expr Var -> Expr Var -> CSEnv
extendCSRecEnv CSEnv
env1 Var
out_id Expr Var
rhs'' Expr Var
id_expr', [(Var, Expr Var)] -> CoreBind
forall b. [(b, Expr b)] -> Bind b
Rec [(Var
zapped_id, Expr Var
rhs')])

  where
    (CSEnv
env1, [Var
out_id]) = CSEnv -> [Var] -> (CSEnv, [Var])
addRecBinders CSEnv
env [Var
in_id]
    rhs' :: Expr Var
rhs'  = CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env1 Expr Var
rhs
    rhs'' :: Expr Var
rhs'' = (CoreTickish -> Bool) -> Expr Var -> Expr Var
forall b. (CoreTickish -> Bool) -> Expr b -> Expr b
stripTicksE CoreTickish -> Bool
forall (pass :: TickishPass). GenTickish pass -> Bool
tickishFloatable Expr Var
rhs'
    ticks :: [CoreTickish]
ticks = (CoreTickish -> Bool) -> Expr Var -> [CoreTickish]
forall b. (CoreTickish -> Bool) -> Expr b -> [CoreTickish]
stripTicksT CoreTickish -> Bool
forall (pass :: TickishPass). GenTickish pass -> Bool
tickishFloatable Expr Var
rhs'
    id_expr' :: Expr Var
id_expr'  = Var -> Expr Var
forall b. Var -> Expr b
varToCoreExpr Var
out_id
    zapped_id :: Var
zapped_id = Var -> Var
zapIdUsageInfo Var
out_id

cseBind TopLevelFlag
toplevel CSEnv
env (Rec [(Var, Expr Var)]
pairs)
  = (CSEnv
env2, [(Var, Expr Var)] -> CoreBind
forall b. [(b, Expr b)] -> Bind b
Rec [(Var, Expr Var)]
pairs')
  where
    (CSEnv
env1, [Var]
bndrs1) = CSEnv -> [Var] -> (CSEnv, [Var])
addRecBinders CSEnv
env (((Var, Expr Var) -> Var) -> [(Var, Expr Var)] -> [Var]
forall a b. (a -> b) -> [a] -> [b]
map (Var, Expr Var) -> Var
forall a b. (a, b) -> a
fst [(Var, Expr Var)]
pairs)
    (CSEnv
env2, [(Var, Expr Var)]
pairs') = (CSEnv -> ((Var, Expr Var), Var) -> (CSEnv, (Var, Expr Var)))
-> CSEnv -> [((Var, Expr Var), Var)] -> (CSEnv, [(Var, Expr Var)])
forall (t :: * -> *) s a b.
Traversable t =>
(s -> a -> (s, b)) -> s -> t a -> (s, t b)
mapAccumL CSEnv -> ((Var, Expr Var), Var) -> (CSEnv, (Var, Expr Var))
do_one CSEnv
env1 ([(Var, Expr Var)] -> [Var] -> [((Var, Expr Var), Var)]
forall a b. [a] -> [b] -> [(a, b)]
zip [(Var, Expr Var)]
pairs [Var]
bndrs1)

    do_one :: CSEnv -> ((Var, Expr Var), Var) -> (CSEnv, (Var, Expr Var))
do_one CSEnv
env ((Var, Expr Var)
pr, Var
b1) = TopLevelFlag
-> CSEnv
-> CSEnv
-> (Var, Expr Var)
-> Var
-> (CSEnv, (Var, Expr Var))
cse_bind TopLevelFlag
toplevel CSEnv
env CSEnv
env (Var, Expr Var)
pr Var
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 :: TopLevelFlag
-> CSEnv
-> CSEnv
-> (Var, Expr Var)
-> Var
-> (CSEnv, (Var, Expr Var))
cse_bind TopLevelFlag
toplevel CSEnv
env_rhs CSEnv
env_body (Var
in_id, Expr Var
in_rhs) Var
out_id
  | TopLevelFlag -> Bool
isTopLevel TopLevelFlag
toplevel, Expr Var -> Bool
exprIsTickedString Expr Var
in_rhs
      -- See Note [Take care with literal strings]
  = (CSEnv
env_body', (Var
out_id', Expr Var
in_rhs))

  | Just JoinArity
arity <- Var -> Maybe JoinArity
isJoinId_maybe Var
out_id
      -- See Note [Look inside join-point binders]
  = let ([Var]
params, Expr Var
in_body) = JoinArity -> Expr Var -> ([Var], Expr Var)
forall b. JoinArity -> Expr b -> ([b], Expr b)
collectNBinders JoinArity
arity Expr Var
in_rhs
        (CSEnv
env', [Var]
params') = CSEnv -> [Var] -> (CSEnv, [Var])
addBinders CSEnv
env_rhs [Var]
params
        out_body :: Expr Var
out_body = CSEnv -> Expr Var -> Expr Var
tryForCSE CSEnv
env' Expr Var
in_body
    in (CSEnv
env_body , (Var
out_id, [Var] -> Expr Var -> Expr Var
forall b. [b] -> Expr b -> Expr b
mkLams [Var]
params' Expr Var
out_body))

  | Bool
otherwise
  = (CSEnv
env_body', (Var
out_id'', Expr Var
out_rhs))
  where
    (CSEnv
env_body', Var
out_id') = CSEnv -> Var -> Var -> Expr Var -> Bool -> (CSEnv, Var)
extendCSEnvWithBinding CSEnv
env_body  Var
in_id Var
out_id Expr Var
out_rhs Bool
cse_done
    (Bool
cse_done, Expr Var
out_rhs)  = CSEnv -> Expr Var -> (Bool, Expr Var)
try_for_cse CSEnv
env_rhs Expr Var
in_rhs
    out_id'' :: Var
out_id'' | Bool
cse_done  = Var -> Var
zapStableUnfolding (Var -> Var) -> Var -> Var
forall a b. (a -> b) -> a -> b
$
                           TopLevelFlag -> Var -> Var
delayInlining TopLevelFlag
toplevel Var
out_id'
             | Bool
otherwise = Var
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 :: TopLevelFlag -> Var -> Var
delayInlining TopLevelFlag
top_lvl Var
bndr
  | TopLevelFlag -> Bool
isTopLevel TopLevelFlag
top_lvl
  , Activation -> Bool
isAlwaysActive (Var -> Activation
idInlineActivation Var
bndr)
  , Var -> Bool
idHasRules Var
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
  = Var
bndr Var -> Activation -> Var
`setInlineActivation` Activation
activateAfterInitial
  | Bool
otherwise
  = Var
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 :: CSEnv -> Var -> Var -> Expr Var -> Bool -> (CSEnv, Var)
extendCSEnvWithBinding CSEnv
env Var
in_id Var
out_id Expr Var
rhs' Bool
cse_done
  | Bool -> Bool
not (Var -> Bool
isId Var
out_id) = (CSEnv -> Var -> Expr Var -> CSEnv
extendCSSubst CSEnv
env Var
in_id Expr Var
rhs',     Var
out_id)
  | Var -> Bool
noCSE Var
out_id      = (CSEnv
env,                              Var
out_id)
  | Bool
use_subst         = (CSEnv -> Var -> Expr Var -> CSEnv
extendCSSubst CSEnv
env Var
in_id Expr Var
rhs',     Var
out_id)
  | Bool
cse_done          = (CSEnv
env,                              Var
out_id)
                       -- See Note [Dealing with ticks]
  | Bool
otherwise         = (CSEnv -> Expr Var -> Expr Var -> CSEnv
extendCSEnv CSEnv
env Expr Var
rhs' Expr Var
id_expr', Var
zapped_id)
  where
    id_expr' :: Expr Var
id_expr'  = Var -> Expr Var
forall b. Var -> Expr b
varToCoreExpr Var
out_id
    zapped_id :: Var
zapped_id = Var -> Var
zapIdUsageInfo Var
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 :: Bool
use_subst | Var {} <- Expr Var
rhs' = Bool
True
              | Bool
otherwise      = Bool
False

-- | Given a binder `let x = e`, this function
-- determines whether we should add `e -> x` to the cs_map
noCSE :: InId -> Bool
noCSE :: Var -> Bool
noCSE Var
id =  Bool -> Bool
not (Activation -> Bool
isAlwaysActive (Var -> Activation
idInlineActivation Var
id)) Bool -> Bool -> Bool
&&
            Bool -> Bool
not (InlineSpec -> Bool
noUserInlineSpec (InlinePragma -> InlineSpec
inlinePragmaSpec (Var -> InlinePragma
idInlinePragma Var
id)))
             -- See Note [CSE for INLINE and NOINLINE]
         Bool -> Bool -> Bool
|| InlinePragma -> Bool
isAnyInlinePragma (Var -> InlinePragma
idInlinePragma Var
id)
             -- See Note [CSE for stable unfoldings]
         Bool -> Bool -> Bool
|| Var -> Bool
isJoinId Var
id
             -- See Note [CSE for join points?]


{- 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 :: CSEnv -> Expr Var -> Expr Var
tryForCSE CSEnv
env Expr Var
expr = (Bool, Expr Var) -> Expr Var
forall a b. (a, b) -> b
snd (CSEnv -> Expr Var -> (Bool, Expr Var)
try_for_cse CSEnv
env Expr Var
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 :: CSEnv -> Expr Var -> (Bool, Expr Var)
try_for_cse CSEnv
env Expr Var
expr
  | Just Expr Var
e <- CSEnv -> Expr Var -> Maybe (Expr Var)
lookupCSEnv CSEnv
env Expr Var
expr'' = (Bool
True,  [CoreTickish] -> Expr Var -> Expr Var
mkTicks [CoreTickish]
ticks Expr Var
e)
  | Bool
otherwise                        = (Bool
False, Expr Var
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' :: Expr Var
expr'  = CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env Expr Var
expr
    expr'' :: Expr Var
expr'' = (CoreTickish -> Bool) -> Expr Var -> Expr Var
forall b. (CoreTickish -> Bool) -> Expr b -> Expr b
stripTicksE CoreTickish -> Bool
forall (pass :: TickishPass). GenTickish pass -> Bool
tickishFloatable Expr Var
expr'
    ticks :: [CoreTickish]
ticks  = (CoreTickish -> Bool) -> Expr Var -> [CoreTickish]
forall b. (CoreTickish -> Bool) -> Expr b -> [CoreTickish]
stripTicksT CoreTickish -> Bool
forall (pass :: TickishPass). GenTickish pass -> Bool
tickishFloatable Expr Var
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 :: Expr Var -> Expr Var
cseOneExpr Expr Var
e = CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env Expr Var
e
  where env :: CSEnv
env = CSEnv
emptyCSEnv {cs_subst :: Subst
cs_subst = InScopeSet -> Subst
mkEmptySubst (VarSet -> InScopeSet
mkInScopeSet (Expr Var -> VarSet
exprFreeVars Expr Var
e)) }

cseExpr :: CSEnv -> InExpr -> OutExpr
cseExpr :: CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env (Type Type
t)              = Type -> Expr Var
forall b. Type -> Expr b
Type (Subst -> Type -> Type
substTy (CSEnv -> Subst
csEnvSubst CSEnv
env) Type
t)
cseExpr CSEnv
env (Coercion Coercion
c)          = Coercion -> Expr Var
forall b. Coercion -> Expr b
Coercion (HasCallStack => Subst -> Coercion -> Coercion
Subst -> Coercion -> Coercion
substCo (CSEnv -> Subst
csEnvSubst CSEnv
env) Coercion
c)
cseExpr CSEnv
_   (Lit Literal
lit)             = Literal -> Expr Var
forall b. Literal -> Expr b
Lit Literal
lit
cseExpr CSEnv
env (Var Var
v)               = CSEnv -> Var -> Expr Var
lookupSubst CSEnv
env Var
v
cseExpr CSEnv
env (App Expr Var
f Expr Var
a)             = Expr Var -> Expr Var -> Expr Var
forall b. Expr b -> Expr b -> Expr b
App (CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env Expr Var
f) (CSEnv -> Expr Var -> Expr Var
tryForCSE CSEnv
env Expr Var
a)
cseExpr CSEnv
env (Tick CoreTickish
t Expr Var
e)            = CoreTickish -> Expr Var -> Expr Var
forall b. CoreTickish -> Expr b -> Expr b
Tick CoreTickish
t (CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env Expr Var
e)
cseExpr CSEnv
env (Cast Expr Var
e Coercion
co)           = Expr Var -> Coercion -> Expr Var
forall b. Expr b -> Coercion -> Expr b
Cast (CSEnv -> Expr Var -> Expr Var
tryForCSE CSEnv
env Expr Var
e) (HasCallStack => Subst -> Coercion -> Coercion
Subst -> Coercion -> Coercion
substCo (CSEnv -> Subst
csEnvSubst CSEnv
env) Coercion
co)
cseExpr CSEnv
env (Lam Var
b Expr Var
e)             = let (CSEnv
env', Var
b') = CSEnv -> Var -> (CSEnv, Var)
addBinder CSEnv
env Var
b
                                    in Var -> Expr Var -> Expr Var
forall b. b -> Expr b -> Expr b
Lam Var
b' (CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env' Expr Var
e)
cseExpr CSEnv
env (Let CoreBind
bind Expr Var
e)          = let (CSEnv
env', CoreBind
bind') = TopLevelFlag -> CSEnv -> CoreBind -> (CSEnv, CoreBind)
cseBind TopLevelFlag
NotTopLevel CSEnv
env CoreBind
bind
                                    in CoreBind -> Expr Var -> Expr Var
forall b. Bind b -> Expr b -> Expr b
Let CoreBind
bind' (CSEnv -> Expr Var -> Expr Var
cseExpr CSEnv
env' Expr Var
e)
cseExpr CSEnv
env (Case Expr Var
e Var
bndr Type
ty [OutAlt]
alts) = CSEnv -> Expr Var -> Var -> Type -> [OutAlt] -> Expr Var
cseCase CSEnv
env Expr Var
e Var
bndr Type
ty [OutAlt]
alts

cseCase :: CSEnv -> InExpr -> InId -> InType -> [InAlt] -> OutExpr
cseCase :: CSEnv -> Expr Var -> Var -> Type -> [OutAlt] -> Expr Var
cseCase CSEnv
env Expr Var
scrut Var
bndr Type
ty [OutAlt]
alts
  = Expr Var -> Var -> Type -> [OutAlt] -> Expr Var
forall b. Expr b -> b -> Type -> [Alt b] -> Expr b
Case Expr Var
scrut1 Var
bndr3 Type
ty' ([OutAlt] -> Expr Var) -> [OutAlt] -> Expr Var
forall a b. (a -> b) -> a -> b
$
    [OutAlt] -> [OutAlt]
combineAlts ((OutAlt -> OutAlt) -> [OutAlt] -> [OutAlt]
forall a b. (a -> b) -> [a] -> [b]
map OutAlt -> OutAlt
cse_alt [OutAlt]
alts)
  where
    ty' :: Type
ty' = Subst -> Type -> Type
substTy (CSEnv -> Subst
csEnvSubst CSEnv
env) Type
ty
    (Bool
cse_done, Expr Var
scrut1) = CSEnv -> Expr Var -> (Bool, Expr Var)
try_for_cse CSEnv
env Expr Var
scrut

    bndr1 :: Var
bndr1 = Var -> Var
zapIdOccInfo Var
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
    (CSEnv
env1, Var
bndr2)    = CSEnv -> Var -> (CSEnv, Var)
addBinder CSEnv
env Var
bndr1
    (CSEnv
alt_env, Var
bndr3) = CSEnv -> Var -> Var -> Expr Var -> Bool -> (CSEnv, Var)
extendCSEnvWithBinding CSEnv
env1 Var
bndr Var
bndr2 Expr Var
scrut1 Bool
cse_done
         -- extendCSEnvWithBinding: see Note [CSE for case expressions]

    con_target :: OutExpr
    con_target :: Expr Var
con_target = CSEnv -> Var -> Expr Var
lookupSubst CSEnv
alt_env Var
bndr

    arg_tys :: [OutType]
    arg_tys :: [Type]
arg_tys = Type -> [Type]
tyConAppArgs (Var -> Type
idType Var
bndr3)

    -- See Note [CSE for case alternatives]
    cse_alt :: OutAlt -> OutAlt
cse_alt (Alt (DataAlt DataCon
con) [Var]
args Expr Var
rhs)
        = AltCon -> [Var] -> Expr Var -> OutAlt
forall b. AltCon -> [b] -> Expr b -> Alt b
Alt (DataCon -> AltCon
DataAlt DataCon
con) [Var]
args' (CSEnv -> Expr Var -> Expr Var
tryForCSE CSEnv
new_env Expr Var
rhs)
        where
          (CSEnv
env', [Var]
args') = CSEnv -> [Var] -> (CSEnv, [Var])
addBinders CSEnv
alt_env [Var]
args
          new_env :: CSEnv
new_env       = CSEnv -> Expr Var -> Expr Var -> CSEnv
extendCSEnv CSEnv
env' Expr Var
con_expr Expr Var
con_target
          con_expr :: Expr Var
con_expr      = AltCon -> [Var] -> [Type] -> Expr Var
mkAltExpr (DataCon -> AltCon
DataAlt DataCon
con) [Var]
args' [Type]
arg_tys

    cse_alt (Alt AltCon
con [Var]
args Expr Var
rhs)
        = AltCon -> [Var] -> Expr Var -> OutAlt
forall b. AltCon -> [b] -> Expr b -> Alt b
Alt AltCon
con [Var]
args' (CSEnv -> Expr Var -> Expr Var
tryForCSE CSEnv
env' Expr Var
rhs)
        where
          (CSEnv
env', [Var]
args') = CSEnv -> [Var] -> (CSEnv, [Var])
addBinders CSEnv
alt_env [Var]
args

combineAlts :: [OutAlt] -> [OutAlt]
-- See Note [Combine case alternatives]
combineAlts :: [OutAlt] -> [OutAlt]
combineAlts [OutAlt]
alts
  | (Just OutAlt
alt1, [OutAlt]
rest_alts) <- [OutAlt] -> (Maybe OutAlt, [OutAlt])
find_bndr_free_alt [OutAlt]
alts
  , Alt AltCon
_ [Var]
bndrs1 Expr Var
rhs1 <- OutAlt
alt1
  , let filtered_alts :: [OutAlt]
filtered_alts = (OutAlt -> Bool) -> [OutAlt] -> [OutAlt]
forall a. (a -> Bool) -> [a] -> [a]
filterOut (Expr Var -> OutAlt -> Bool
identical_alt Expr Var
rhs1) [OutAlt]
rest_alts
  , Bool -> Bool
not ([OutAlt] -> [OutAlt] -> Bool
forall a b. [a] -> [b] -> Bool
equalLength [OutAlt]
rest_alts [OutAlt]
filtered_alts)
  = Bool -> SDoc -> [OutAlt] -> [OutAlt]
forall a. HasCallStack => Bool -> SDoc -> a -> a
assertPpr ([Var] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [Var]
bndrs1) ([OutAlt] -> SDoc
forall a. Outputable a => a -> SDoc
ppr [OutAlt]
alts) ([OutAlt] -> [OutAlt]) -> [OutAlt] -> [OutAlt]
forall a b. (a -> b) -> a -> b
$
    AltCon -> [Var] -> Expr Var -> OutAlt
forall b. AltCon -> [b] -> Expr b -> Alt b
Alt AltCon
DEFAULT [] Expr Var
rhs1 OutAlt -> [OutAlt] -> [OutAlt]
forall a. a -> [a] -> [a]
: [OutAlt]
filtered_alts

  | Bool
otherwise
  = [OutAlt]
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 :: [OutAlt] -> (Maybe OutAlt, [OutAlt])
find_bndr_free_alt []
      = (Maybe OutAlt
forall a. Maybe a
Nothing, [])
    find_bndr_free_alt (alt :: OutAlt
alt@(Alt AltCon
_ [Var]
bndrs Expr Var
_) : [OutAlt]
alts)
      | [Var] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [Var]
bndrs = (OutAlt -> Maybe OutAlt
forall a. a -> Maybe a
Just OutAlt
alt, [OutAlt]
alts)
      | Bool
otherwise  = case [OutAlt] -> (Maybe OutAlt, [OutAlt])
find_bndr_free_alt [OutAlt]
alts of
                       (Maybe OutAlt
mb_bf, [OutAlt]
alts) -> (Maybe OutAlt
mb_bf, OutAlt
altOutAlt -> [OutAlt] -> [OutAlt]
forall a. a -> [a] -> [a]
:[OutAlt]
alts)

    identical_alt :: Expr Var -> OutAlt -> Bool
identical_alt Expr Var
rhs1 (Alt AltCon
_ [Var]
_ Expr Var
rhs) = Expr Var -> Expr Var -> Bool
eqCoreExpr Expr Var
rhs1 Expr Var
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 eqExpr.  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 { CSEnv -> Subst
cs_subst :: Subst  -- Maps InBndrs to OutExprs
            -- The substitution variables to
            -- /trivial/ OutExprs, not arbitrary expressions

       , CSEnv -> CoreMap (Expr Var)
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@.

       , CSEnv -> CoreMap (Expr Var)
cs_rec_map :: CoreMap OutExpr
            -- See Note [CSE for recursive bindings]
       }

emptyCSEnv :: CSEnv
emptyCSEnv :: CSEnv
emptyCSEnv = CS { cs_map :: CoreMap (Expr Var)
cs_map = CoreMap (Expr Var)
forall a. CoreMap a
emptyCoreMap, cs_rec_map :: CoreMap (Expr Var)
cs_rec_map = CoreMap (Expr Var)
forall a. CoreMap a
emptyCoreMap
                , cs_subst :: Subst
cs_subst = Subst
emptySubst }

lookupCSEnv :: CSEnv -> OutExpr -> Maybe OutExpr
lookupCSEnv :: CSEnv -> Expr Var -> Maybe (Expr Var)
lookupCSEnv (CS { cs_map :: CSEnv -> CoreMap (Expr Var)
cs_map = CoreMap (Expr Var)
csmap }) Expr Var
expr
  = CoreMap (Expr Var) -> Expr Var -> Maybe (Expr Var)
forall a. CoreMap a -> Expr Var -> Maybe a
lookupCoreMap CoreMap (Expr Var)
csmap Expr Var
expr

-- | @extendCSEnv env e triv_expr@ will replace any occurrence of @e@ with @triv_expr@ going forward.
extendCSEnv :: CSEnv -> OutExpr -> OutExpr -> CSEnv
extendCSEnv :: CSEnv -> Expr Var -> Expr Var -> CSEnv
extendCSEnv CSEnv
cse Expr Var
expr Expr Var
triv_expr
  = CSEnv
cse { cs_map :: CoreMap (Expr Var)
cs_map = CoreMap (Expr Var) -> Expr Var -> Expr Var -> CoreMap (Expr Var)
forall a. CoreMap a -> Expr Var -> a -> CoreMap a
extendCoreMap (CSEnv -> CoreMap (Expr Var)
cs_map CSEnv
cse) Expr Var
sexpr Expr Var
triv_expr }
  where
    sexpr :: Expr Var
sexpr = (CoreTickish -> Bool) -> Expr Var -> Expr Var
forall b. (CoreTickish -> Bool) -> Expr b -> Expr b
stripTicksE CoreTickish -> Bool
forall (pass :: TickishPass). GenTickish pass -> Bool
tickishFloatable Expr Var
expr

extendCSRecEnv :: CSEnv -> OutId -> OutExpr -> OutExpr -> CSEnv
-- See Note [CSE for recursive bindings]
extendCSRecEnv :: CSEnv -> Var -> Expr Var -> Expr Var -> CSEnv
extendCSRecEnv CSEnv
cse Var
bndr Expr Var
expr Expr Var
triv_expr
  = CSEnv
cse { cs_rec_map :: CoreMap (Expr Var)
cs_rec_map = CoreMap (Expr Var) -> Expr Var -> Expr Var -> CoreMap (Expr Var)
forall a. CoreMap a -> Expr Var -> a -> CoreMap a
extendCoreMap (CSEnv -> CoreMap (Expr Var)
cs_rec_map CSEnv
cse) (Var -> Expr Var -> Expr Var
forall b. b -> Expr b -> Expr b
Lam Var
bndr Expr Var
expr) Expr Var
triv_expr }

lookupCSRecEnv :: CSEnv -> OutId -> OutExpr -> Maybe OutExpr
-- See Note [CSE for recursive bindings]
lookupCSRecEnv :: CSEnv -> Var -> Expr Var -> Maybe (Expr Var)
lookupCSRecEnv (CS { cs_rec_map :: CSEnv -> CoreMap (Expr Var)
cs_rec_map = CoreMap (Expr Var)
csmap }) Var
bndr Expr Var
expr
  = CoreMap (Expr Var) -> Expr Var -> Maybe (Expr Var)
forall a. CoreMap a -> Expr Var -> Maybe a
lookupCoreMap CoreMap (Expr Var)
csmap (Var -> Expr Var -> Expr Var
forall b. b -> Expr b -> Expr b
Lam Var
bndr Expr Var
expr)

csEnvSubst :: CSEnv -> Subst
csEnvSubst :: CSEnv -> Subst
csEnvSubst = CSEnv -> Subst
cs_subst

lookupSubst :: CSEnv -> Id -> OutExpr
lookupSubst :: CSEnv -> Var -> Expr Var
lookupSubst (CS { cs_subst :: CSEnv -> Subst
cs_subst = Subst
sub}) Var
x = (() :: Constraint) => Subst -> Var -> Expr Var
Subst -> Var -> Expr Var
lookupIdSubst Subst
sub Var
x

extendCSSubst :: CSEnv -> Id  -> CoreExpr -> CSEnv
extendCSSubst :: CSEnv -> Var -> Expr Var -> CSEnv
extendCSSubst CSEnv
cse Var
x Expr Var
rhs = CSEnv
cse { cs_subst :: Subst
cs_subst = Subst -> Var -> Expr Var -> Subst
extendSubst (CSEnv -> Subst
cs_subst CSEnv
cse) Var
x Expr Var
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 :: CSEnv -> Var -> (CSEnv, Var)
addBinder CSEnv
cse Var
v = (CSEnv
cse { cs_subst :: Subst
cs_subst = Subst
sub' }, Var
v')
                where
                  (Subst
sub', Var
v') = Subst -> Var -> (Subst, Var)
substBndr (CSEnv -> Subst
cs_subst CSEnv
cse) Var
v

addBinders :: CSEnv -> [Var] -> (CSEnv, [Var])
addBinders :: CSEnv -> [Var] -> (CSEnv, [Var])
addBinders CSEnv
cse [Var]
vs = (CSEnv
cse { cs_subst :: Subst
cs_subst = Subst
sub' }, [Var]
vs')
                where
                  (Subst
sub', [Var]
vs') = Subst -> [Var] -> (Subst, [Var])
substBndrs (CSEnv -> Subst
cs_subst CSEnv
cse) [Var]
vs

addRecBinders :: CSEnv -> [Id] -> (CSEnv, [Id])
addRecBinders :: CSEnv -> [Var] -> (CSEnv, [Var])
addRecBinders CSEnv
cse [Var]
vs = (CSEnv
cse { cs_subst :: Subst
cs_subst = Subst
sub' }, [Var]
vs')
                where
                  (Subst
sub', [Var]
vs') = Subst -> [Var] -> (Subst, [Var])
substRecBndrs (CSEnv -> Subst
cs_subst CSEnv
cse) [Var]
vs