{-# LANGUAGE ViewPatterns #-}

{-# OPTIONS_GHC -cpp -Wno-incomplete-record-updates #-}

{-# OPTIONS_GHC -fmax-worker-args=12 #-}
-- The -fmax-worker-args=12 is there because the main functions
-- are strict in the OccEnv, and it turned out that with the default settting
-- some functions would unbox the OccEnv ad some would not, depending on how
-- many /other/ arguments the function has.  Inconsistent unboxing is very
-- bad for performance, so I increased the limit to allow it to unbox
-- consistently.

{-
(c) The GRASP/AQUA Project, Glasgow University, 1992-1998

************************************************************************
*                                                                      *
\section[OccurAnal]{Occurrence analysis pass}
*                                                                      *
************************************************************************

The occurrence analyser re-typechecks a core expression, returning a new
core expression with (hopefully) improved usage information.
-}

module GHC.Core.Opt.OccurAnal (
    occurAnalysePgm,
    occurAnalyseExpr,
    zapLambdaBndrs, BinderSwapDecision(..), scrutOkForBinderSwap
  ) where

import GHC.Prelude hiding ( head, init, last, tail )

import GHC.Core
import GHC.Core.FVs
import GHC.Core.Utils   ( exprIsTrivial, isDefaultAlt, isExpandableApp,
                          mkCastMCo, mkTicks )
import GHC.Core.Opt.Arity   ( joinRhsArity, isOneShotBndr )
import GHC.Core.Coercion
import GHC.Core.Predicate   ( isDictId )
import GHC.Core.Type
import GHC.Core.TyCo.FVs    ( tyCoVarsOfMCo )

import GHC.Data.Maybe( orElse )
import GHC.Data.Graph.Directed ( SCC(..), Node(..)
                               , stronglyConnCompFromEdgedVerticesUniq
                               , stronglyConnCompFromEdgedVerticesUniqR )
import GHC.Types.Unique
import GHC.Types.Unique.FM
import GHC.Types.Unique.Set
import GHC.Types.Id
import GHC.Types.Id.Info
import GHC.Types.Basic
import GHC.Types.Tickish
import GHC.Types.Var.Set
import GHC.Types.Var.Env
import GHC.Types.Var
import GHC.Types.Demand ( argOneShots, argsOneShots, isDeadEndSig )

import GHC.Utils.Outputable
import GHC.Utils.Panic
import GHC.Utils.Misc

import GHC.Builtin.Names( runRWKey )
import GHC.Unit.Module( Module )

import Data.List (mapAccumL)

{-
************************************************************************
*                                                                      *
    occurAnalysePgm, occurAnalyseExpr
*                                                                      *
************************************************************************

Here's the externally-callable interface:
-}

-- | Do occurrence analysis, and discard occurrence info returned
occurAnalyseExpr :: CoreExpr -> CoreExpr
occurAnalyseExpr :: CoreExpr -> CoreExpr
occurAnalyseExpr CoreExpr
expr = CoreExpr
expr'
  where
    WUD UsageDetails
_ CoreExpr
expr' = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
initOccEnv CoreExpr
expr

occurAnalysePgm :: Module         -- Used only in debug output
                -> (Id -> Bool)         -- Active unfoldings
                -> (Activation -> Bool) -- Active rules
                -> [CoreRule]           -- Local rules for imported Ids
                -> CoreProgram -> CoreProgram
occurAnalysePgm :: Module
-> (Id -> Bool)
-> (Activation -> Bool)
-> [CoreRule]
-> CoreProgram
-> CoreProgram
occurAnalysePgm Module
this_mod Id -> Bool
active_unf Activation -> Bool
active_rule [CoreRule]
imp_rules CoreProgram
binds
  | UsageDetails -> Bool
isEmptyDetails UsageDetails
final_usage
  = CoreProgram
occ_anald_binds

  | Bool
otherwise   -- See Note [Glomming]
  = Bool -> String -> SDoc -> CoreProgram -> CoreProgram
forall a. HasCallStack => Bool -> String -> SDoc -> a -> a
warnPprTrace Bool
True String
"Glomming in" (SDoc -> JoinArity -> SDoc -> SDoc
hang (Module -> SDoc
forall a. Outputable a => a -> SDoc
ppr Module
this_mod SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> SDoc
forall doc. IsLine doc => doc
colon) JoinArity
2 (UsageDetails -> SDoc
forall a. Outputable a => a -> SDoc
ppr UsageDetails
final_usage))
    CoreProgram
occ_anald_glommed_binds
  where
    init_env :: OccEnv
init_env = OccEnv
initOccEnv { occ_rule_act = active_rule
                          , occ_unf_act  = active_unf }

    WUD UsageDetails
final_usage CoreProgram
occ_anald_binds = CoreProgram -> OccEnv -> WithUsageDetails CoreProgram
go CoreProgram
binds OccEnv
init_env
    WUD UsageDetails
_ CoreProgram
occ_anald_glommed_binds = OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> [(Id, CoreExpr)]
-> UsageDetails
-> WithUsageDetails CoreProgram
occAnalRecBind OccEnv
init_env TopLevelFlag
TopLevel
                                                    ImpRuleEdges
imp_rule_edges
                                                    (CoreProgram -> [(Id, CoreExpr)]
forall b. [Bind b] -> [(b, Expr b)]
flattenBinds CoreProgram
binds)
                                                    UsageDetails
initial_uds
          -- It's crucial to re-analyse the glommed-together bindings
          -- so that we establish the right loop breakers. Otherwise
          -- we can easily create an infinite loop (#9583 is an example)
          --
          -- Also crucial to re-analyse the /original/ bindings
          -- in case the first pass accidentally discarded as dead code
          -- a binding that was actually needed (albeit before its
          -- definition site).  #17724 threw this up.

    initial_uds :: UsageDetails
initial_uds = UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
emptyDetails ([CoreRule] -> VarSet
rulesFreeVars [CoreRule]
imp_rules)
    -- The RULES declarations keep things alive!

    -- imp_rule_edges maps a top-level local binder 'f' to the
    -- RHS free vars of any IMP-RULE, a local RULE for an imported function,
    -- where 'f' appears on the LHS
    --   e.g.  RULE foldr f = blah
    --         imp_rule_edges contains f :-> fvs(blah)
    -- We treat such RULES as extra rules for 'f'
    -- See Note [Preventing loops due to imported functions rules]
    imp_rule_edges :: ImpRuleEdges
    imp_rule_edges :: ImpRuleEdges
imp_rule_edges = (ImpRuleEdges -> ImpRuleEdges -> ImpRuleEdges)
-> ImpRuleEdges -> [ImpRuleEdges] -> ImpRuleEdges
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (([(Activation, VarSet)]
 -> [(Activation, VarSet)] -> [(Activation, VarSet)])
-> ImpRuleEdges -> ImpRuleEdges -> ImpRuleEdges
forall a. (a -> a -> a) -> VarEnv a -> VarEnv a -> VarEnv a
plusVarEnv_C [(Activation, VarSet)]
-> [(Activation, VarSet)] -> [(Activation, VarSet)]
forall a. [a] -> [a] -> [a]
(++)) ImpRuleEdges
forall a. VarEnv a
emptyVarEnv
                           [ (Id -> [(Activation, VarSet)]) -> VarEnv Id -> ImpRuleEdges
forall a b. (a -> b) -> VarEnv a -> VarEnv b
mapVarEnv ([(Activation, VarSet)] -> Id -> [(Activation, VarSet)]
forall a b. a -> b -> a
const [(Activation
act,VarSet
rhs_fvs)]) (VarEnv Id -> ImpRuleEdges) -> VarEnv Id -> ImpRuleEdges
forall a b. (a -> b) -> a -> b
$ VarSet -> VarEnv Id
forall a. UniqSet a -> UniqFM a a
getUniqSet (VarSet -> VarEnv Id) -> VarSet -> VarEnv Id
forall a b. (a -> b) -> a -> b
$
                             [CoreExpr] -> VarSet
exprsFreeIds [CoreExpr]
args VarSet -> [Id] -> VarSet
`delVarSetList` [Id]
bndrs
                           | Rule { ru_act :: CoreRule -> Activation
ru_act = Activation
act, ru_bndrs :: CoreRule -> [Id]
ru_bndrs = [Id]
bndrs
                                   , ru_args :: CoreRule -> [CoreExpr]
ru_args = [CoreExpr]
args, ru_rhs :: CoreRule -> CoreExpr
ru_rhs = CoreExpr
rhs } <- [CoreRule]
imp_rules
                                   -- Not BuiltinRules; see Note [Plugin rules]
                           , let rhs_fvs :: VarSet
rhs_fvs = CoreExpr -> VarSet
exprFreeIds CoreExpr
rhs VarSet -> [Id] -> VarSet
`delVarSetList` [Id]
bndrs ]

    go :: [CoreBind] -> OccEnv -> WithUsageDetails [CoreBind]
    go :: CoreProgram -> OccEnv -> WithUsageDetails CoreProgram
go []           OccEnv
_   = UsageDetails -> CoreProgram -> WithUsageDetails CoreProgram
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
initial_uds []
    go (CoreBind
bind:CoreProgram
binds) OccEnv
env = OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> CoreBind
-> (OccEnv -> WithUsageDetails CoreProgram)
-> (CoreProgram -> CoreProgram -> CoreProgram)
-> WithUsageDetails CoreProgram
forall r.
OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> CoreBind
-> (OccEnv -> WithUsageDetails r)
-> (CoreProgram -> r -> r)
-> WithUsageDetails r
occAnalBind OccEnv
env TopLevelFlag
TopLevel
                           ImpRuleEdges
imp_rule_edges CoreBind
bind (CoreProgram -> OccEnv -> WithUsageDetails CoreProgram
go CoreProgram
binds) CoreProgram -> CoreProgram -> CoreProgram
forall a. [a] -> [a] -> [a]
(++)

{- *********************************************************************
*                                                                      *
                IMP-RULES
         Local rules for imported functions
*                                                                      *
********************************************************************* -}

type ImpRuleEdges = IdEnv [(Activation, VarSet)]
    -- Mapping from a local Id 'f' to info about its IMP-RULES,
    -- i.e. /local/ rules for an imported Id that mention 'f' on the LHS
    -- We record (a) its Activation and (b) the RHS free vars
    -- See Note [IMP-RULES: local rules for imported functions]

noImpRuleEdges :: ImpRuleEdges
noImpRuleEdges :: ImpRuleEdges
noImpRuleEdges = ImpRuleEdges
forall a. VarEnv a
emptyVarEnv

lookupImpRules :: ImpRuleEdges -> Id -> [(Activation,VarSet)]
lookupImpRules :: ImpRuleEdges -> Id -> [(Activation, VarSet)]
lookupImpRules ImpRuleEdges
imp_rule_edges Id
bndr
  = case ImpRuleEdges -> Id -> Maybe [(Activation, VarSet)]
forall a. VarEnv a -> Id -> Maybe a
lookupVarEnv ImpRuleEdges
imp_rule_edges Id
bndr of
      Maybe [(Activation, VarSet)]
Nothing -> []
      Just [(Activation, VarSet)]
vs -> [(Activation, VarSet)]
vs

impRulesScopeUsage :: [(Activation,VarSet)] -> UsageDetails
-- Variable mentioned in RHS of an IMP-RULE for the bndr,
-- whether active or not
impRulesScopeUsage :: [(Activation, VarSet)] -> UsageDetails
impRulesScopeUsage [(Activation, VarSet)]
imp_rules_info
  = ((Activation, VarSet) -> UsageDetails -> UsageDetails)
-> UsageDetails -> [(Activation, VarSet)] -> UsageDetails
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (Activation, VarSet) -> UsageDetails -> UsageDetails
forall {a}. (a, VarSet) -> UsageDetails -> UsageDetails
add UsageDetails
emptyDetails [(Activation, VarSet)]
imp_rules_info
  where
    add :: (a, VarSet) -> UsageDetails -> UsageDetails
add (a
_,VarSet
vs) UsageDetails
usage = UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
usage VarSet
vs

impRulesActiveFvs :: (Activation -> Bool) -> VarSet
                  -> [(Activation,VarSet)] -> VarSet
impRulesActiveFvs :: (Activation -> Bool) -> VarSet -> [(Activation, VarSet)] -> VarSet
impRulesActiveFvs Activation -> Bool
is_active VarSet
bndr_set [(Activation, VarSet)]
vs
  = ((Activation, VarSet) -> VarSet -> VarSet)
-> VarSet -> [(Activation, VarSet)] -> VarSet
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (Activation, VarSet) -> VarSet -> VarSet
add VarSet
emptyVarSet [(Activation, VarSet)]
vs VarSet -> VarSet -> VarSet
`intersectVarSet` VarSet
bndr_set
  where
    add :: (Activation, VarSet) -> VarSet -> VarSet
add (Activation
act,VarSet
vs) VarSet
acc | Activation -> Bool
is_active Activation
act = VarSet
vs VarSet -> VarSet -> VarSet
`unionVarSet` VarSet
acc
                     | Bool
otherwise     = VarSet
acc

{- Note [IMP-RULES: local rules for imported functions]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We quite often have
  * A /local/ rule
  * for an /imported/ function
like this:
  foo x = blah
  {-# RULE "map/foo" forall xs. map foo xs = xs #-}
We call them IMP-RULES.  They are important in practice, and occur a
lot in the libraries.

IMP-RULES are held in mg_rules of ModGuts, and passed in to
occurAnalysePgm.

Main Invariant:

* Throughout, we treat an IMP-RULE that mentions 'f' on its LHS
  just like a RULE for f.

Note [IMP-RULES: unavoidable loops]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider this
   f = /\a. B.g a
   RULE B.g Int = 1 + f Int
Note that
  * The RULE is for an imported function.
  * f is non-recursive
Now we
can get
   f Int --> B.g Int      Inlining f
         --> 1 + f Int    Firing RULE
and so the simplifier goes into an infinite loop. This
would not happen if the RULE was for a local function,
because we keep track of dependencies through rules.  But
that is pretty much impossible to do for imported Ids.  Suppose
f's definition had been
   f = /\a. C.h a
where (by some long and devious process), C.h eventually inlines to
B.g.  We could only spot such loops by exhaustively following
unfoldings of C.h etc, in case we reach B.g, and hence (via the RULE)
f.

We regard this potential infinite loop as a *programmer* error.
It's up the programmer not to write silly rules like
     RULE f x = f x
and the example above is just a more complicated version.

Note [Specialising imported functions] (referred to from Specialise)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For *automatically-generated* rules, the programmer can't be
responsible for the "programmer error" in Note [IMP-RULES: unavoidable
loops].  In particular, consider specialising a recursive function
defined in another module.  If we specialise a recursive function B.g,
we get
  g_spec = .....(B.g Int).....
  RULE B.g Int = g_spec
Here, g_spec doesn't look recursive, but when the rule fires, it
becomes so.  And if B.g was mutually recursive, the loop might not be
as obvious as it is here.

To avoid this,
 * When specialising a function that is a loop breaker,
   give a NOINLINE pragma to the specialised function

Note [Preventing loops due to imported functions rules]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider:
  import GHC.Base (foldr)

  {-# RULES "filterList" forall p. foldr (filterFB (:) p) [] = filter p #-}
  filter p xs = build (\c n -> foldr (filterFB c p) n xs)
  filterFB c p = ...

  f = filter p xs

Note that filter is not a loop-breaker, so what happens is:
  f =          filter p xs
    = {inline} build (\c n -> foldr (filterFB c p) n xs)
    = {inline} foldr (filterFB (:) p) [] xs
    = {RULE}   filter p xs

We are in an infinite loop.

A more elaborate example (that I actually saw in practice when I went to
mark GHC.List.filter as INLINABLE) is as follows. Say I have this module:
  {-# LANGUAGE RankNTypes #-}
  module GHCList where

  import Prelude hiding (filter)
  import GHC.Base (build)

  {-# INLINABLE filter #-}
  filter :: (a -> Bool) -> [a] -> [a]
  filter p [] = []
  filter p (x:xs) = if p x then x : filter p xs else filter p xs

  {-# NOINLINE [0] filterFB #-}
  filterFB :: (a -> b -> b) -> (a -> Bool) -> a -> b -> b
  filterFB c p x r | p x       = x `c` r
                   | otherwise = r

  {-# RULES
  "filter"     [~1] forall p xs.  filter p xs = build (\c n -> foldr
  (filterFB c p) n xs)
  "filterList" [1]  forall p.     foldr (filterFB (:) p) [] = filter p
   #-}

Then (because RULES are applied inside INLINABLE unfoldings, but inlinings
are not), the unfolding given to "filter" in the interface file will be:
  filter p []     = []
  filter p (x:xs) = if p x then x : build (\c n -> foldr (filterFB c p) n xs)
                           else     build (\c n -> foldr (filterFB c p) n xs

Note that because this unfolding does not mention "filter", filter is not
marked as a strong loop breaker. Therefore at a use site in another module:
  filter p xs
    = {inline}
      case xs of []     -> []
                 (x:xs) -> if p x then x : build (\c n -> foldr (filterFB c p) n xs)
                                  else     build (\c n -> foldr (filterFB c p) n xs)

  build (\c n -> foldr (filterFB c p) n xs)
    = {inline} foldr (filterFB (:) p) [] xs
    = {RULE}   filter p xs

And we are in an infinite loop again, except that this time the loop is producing an
infinitely large *term* (an unrolling of filter) and so the simplifier finally
dies with "ticks exhausted"

SOLUTION: we treat the rule "filterList" as an extra rule for 'filterFB'
because it mentions 'filterFB' on the LHS.  This is the Main Invariant
in Note [IMP-RULES: local rules for imported functions].

So, during loop-breaker analysis:

- for each active RULE for a local function 'f' we add an edge between
  'f' and the local FVs of the rule RHS

- for each active RULE for an *imported* function we add dependency
  edges between the *local* FVS of the rule LHS and the *local* FVS of
  the rule RHS.

Even with this extra hack we aren't always going to get things
right. For example, it might be that the rule LHS mentions an imported
Id, and another module has a RULE that can rewrite that imported Id to
one of our local Ids.

Note [Plugin rules]
~~~~~~~~~~~~~~~~~~~
Conal Elliott (#11651) built a GHC plugin that added some
BuiltinRules (for imported Ids) to the mg_rules field of ModGuts, to
do some domain-specific transformations that could not be expressed
with an ordinary pattern-matching CoreRule.  But then we can't extract
the dependencies (in imp_rule_edges) from ru_rhs etc, because a
BuiltinRule doesn't have any of that stuff.

So we simply assume that BuiltinRules have no dependencies, and filter
them out from the imp_rule_edges comprehension.

Note [Glomming]
~~~~~~~~~~~~~~~
RULES for imported Ids can make something at the top refer to
something at the bottom:

        foo = ...(B.f @Int)...
        $sf = blah
        RULE:  B.f @Int = $sf

Applying this rule makes foo refer to $sf, although foo doesn't appear to
depend on $sf.  (And, as in Note [IMP-RULES: local rules for imported functions], the
dependency might be more indirect. For example, foo might mention C.t
rather than B.f, where C.t eventually inlines to B.f.)

NOTICE that this cannot happen for rules whose head is a
locally-defined function, because we accurately track dependencies
through RULES.  It only happens for rules whose head is an imported
function (B.f in the example above).

Solution:
  - When simplifying, bring all top level identifiers into
    scope at the start, ignoring the Rec/NonRec structure, so
    that when 'h' pops up in f's rhs, we find it in the in-scope set
    (as the simplifier generally expects). This happens in simplTopBinds.

  - In the occurrence analyser, if there are any out-of-scope
    occurrences that pop out of the top, which will happen after
    firing the rule:      f = \x -> h x
                          h = \y -> 3
    then just glom all the bindings into a single Rec, so that
    the *next* iteration of the occurrence analyser will sort
    them all out.   This part happens in occurAnalysePgm.

This is a legitimate situation where the need for glomming doesn't
point to any problems. However, when GHC is compiled with -DDEBUG, we
produce a warning addressed to the GHC developers just in case we
require glomming due to an out-of-order reference that is caused by
some earlier transformation stage misbehaving.
-}

{-
************************************************************************
*                                                                      *
                Bindings
*                                                                      *
************************************************************************

Note [Recursive bindings: the grand plan]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Loop breaking is surprisingly subtle.  First read the section 4 of
"Secrets of the GHC inliner".  This describes our basic plan.  We
avoid infinite inlinings by choosing loop breakers, and ensuring that
a loop breaker cuts each loop.

See also Note [Inlining and hs-boot files] in GHC.Core.ToIface, which
deals with a closely related source of infinite loops.

When we come across a binding group
  Rec { x1 = r1; ...; xn = rn }
we treat it like this (occAnalRecBind):

1. Note [Forming Rec groups]
   Occurrence-analyse each right hand side, and build a
   "Details" for each binding to capture the results.
   Wrap the details in a LetrecNode, ready for SCC analysis.
   All this is done by makeNode.

   The edges of this graph are the "scope edges".

2. Do SCC-analysis on these Nodes:
   - Each CyclicSCC will become a new Rec
   - Each AcyclicSCC will become a new NonRec

   The key property is that every free variable of a binding is
   accounted for by the scope edges, so that when we are done
   everything is still in scope.

3. For each AcyclicSCC, just make a NonRec binding.

4. For each CyclicSCC of the scope-edge SCC-analysis in (2), we
   identify suitable loop-breakers to ensure that inlining terminates.
   This is done by occAnalRec.

   To do so, form the loop-breaker graph, do SCC analysis. For each
   CyclicSCC we choose a loop breaker, delete all edges to that node,
   re-analyse the SCC, and iterate. See Note [Choosing loop breakers]
   for the details


Note [Dead code]
~~~~~~~~~~~~~~~~
Dropping dead code for a cyclic Strongly Connected Component is done
in a very simple way:

        the entire SCC is dropped if none of its binders are mentioned
        in the body; otherwise the whole thing is kept.

The key observation is that dead code elimination happens after
dependency analysis: so 'occAnalBind' processes SCCs instead of the
original term's binding groups.

Thus 'occAnalBind' does indeed drop 'f' in an example like

        letrec f = ...g...
               g = ...(...g...)...
        in
           ...g...

when 'g' no longer uses 'f' at all (eg 'f' does not occur in a RULE in
'g'). 'occAnalBind' first consumes 'CyclicSCC g' and then it consumes
'AcyclicSCC f', where 'body_usage' won't contain 'f'.

Note [Forming Rec groups]
~~~~~~~~~~~~~~~~~~~~~~~~~
The key point about the "Forming Rec groups" step is that it /preserves
scoping/.  If 'x' is mentioned, it had better be bound somewhere.  So if
we start with
  Rec { f = ...h...
      ; g = ...f...
      ; h = ...f... }
we can split into SCCs
  Rec { f = ...h...
      ; h = ..f... }
  NonRec { g = ...f... }

We put bindings {f = ef; g = eg } in a Rec group if "f uses g" and "g
uses f", no matter how indirectly.  We do a SCC analysis with an edge
f -> g if "f mentions g". That is, g is free in:
  a) the rhs 'ef'
  b) or the RHS of a rule for f, whether active or inactive
       Note [Rules are extra RHSs]
  c) or the LHS or a rule for f, whether active or inactive
       Note [Rule dependency info]
  d) the RHS of an /active/ local IMP-RULE
       Note [IMP-RULES: local rules for imported functions]

(b) and (c) apply regardless of the activation of the RULE, because even if
the rule is inactive its free variables must be bound.  But (d) doesn't need
to worry about this because IMP-RULES are always notionally at the bottom
of the file.

  * Note [Rules are extra RHSs]
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~
    A RULE for 'f' is like an extra RHS for 'f'. That way the "parent"
    keeps the specialised "children" alive.  If the parent dies
    (because it isn't referenced any more), then the children will die
    too (unless they are already referenced directly).

    So in Example [eftInt], eftInt and eftIntFB will be put in the
    same Rec, even though their 'main' RHSs are both non-recursive.

    We must also include inactive rules, so that their free vars
    remain in scope.

  * Note [Rule dependency info]
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~
    The VarSet in a RuleInfo is used for dependency analysis in the
    occurrence analyser.  We must track free vars in *both* lhs and rhs.
    Hence use of idRuleVars, rather than idRuleRhsVars in occAnalBind.
    Why both? Consider
        x = y
        RULE f x = v+4
    Then if we substitute y for x, we'd better do so in the
    rule's LHS too, so we'd better ensure the RULE appears to mention 'x'
    as well as 'v'

  * Note [Rules are visible in their own rec group]
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    We want the rules for 'f' to be visible in f's right-hand side.
    And we'd like them to be visible in other functions in f's Rec
    group.  E.g. in Note [Specialisation rules] we want f' rule
    to be visible in both f's RHS, and fs's RHS.

    This means that we must simplify the RULEs first, before looking
    at any of the definitions.  This is done by Simplify.simplRecBind,
    when it calls addLetIdInfo.

Note [TailUsageDetails when forming Rec groups]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The `TailUsageDetails` stored in the `nd_uds` field of a `NodeDetails` is
computed by `occAnalLamTail` applied to the RHS, not `occAnalExpr`.
That is because the binding might still become a *non-recursive join point* in
the AcyclicSCC case of dependency analysis!
Hence we do the delayed `adjustTailUsage` in `occAnalRec`/`tagRecBinders` to get
a regular, adjusted UsageDetails.
See Note [Join points and unfoldings/rules] for more details on the contract.

Note [Stable unfoldings]
~~~~~~~~~~~~~~~~~~~~~~~~
None of the above stuff about RULES applies to a stable unfolding
stored in a CoreUnfolding.  The unfolding, if any, is simplified
at the same time as the regular RHS of the function (ie *not* like
Note [Rules are visible in their own rec group]), so it should be
treated *exactly* like an extra RHS.

Or, rather, when computing loop-breaker edges,
  * If f has an INLINE pragma, and it is active, we treat the
    INLINE rhs as f's rhs
  * If it's inactive, we treat f as having no rhs
  * If it has no INLINE pragma, we look at f's actual rhs


There is a danger that we'll be sub-optimal if we see this
     f = ...f...
     [INLINE f = ..no f...]
where f is recursive, but the INLINE is not. This can just about
happen with a sufficiently odd set of rules; eg

        foo :: Int -> Int
        {-# INLINE [1] foo #-}
        foo x = x+1

        bar :: Int -> Int
        {-# INLINE [1] bar #-}
        bar x = foo x + 1

        {-# RULES "foo" [~1] forall x. foo x = bar x #-}

Here the RULE makes bar recursive; but it's INLINE pragma remains
non-recursive. It's tempting to then say that 'bar' should not be
a loop breaker, but an attempt to do so goes wrong in two ways:
   a) We may get
         $df = ...$cfoo...
         $cfoo = ...$df....
         [INLINE $cfoo = ...no-$df...]
      But we want $cfoo to depend on $df explicitly so that we
      put the bindings in the right order to inline $df in $cfoo
      and perhaps break the loop altogether.  (Maybe this
   b)


Example [eftInt]
~~~~~~~~~~~~~~~
Example (from GHC.Enum):

  eftInt :: Int# -> Int# -> [Int]
  eftInt x y = ...(non-recursive)...

  {-# INLINE [0] eftIntFB #-}
  eftIntFB :: (Int -> r -> r) -> r -> Int# -> Int# -> r
  eftIntFB c n x y = ...(non-recursive)...

  {-# RULES
  "eftInt"  [~1] forall x y. eftInt x y = build (\ c n -> eftIntFB c n x y)
  "eftIntList"  [1] eftIntFB  (:) [] = eftInt
   #-}

Note [Specialisation rules]
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider this group, which is typical of what SpecConstr builds:

   fs a = ....f (C a)....
   f  x = ....f (C a)....
   {-# RULE f (C a) = fs a #-}

So 'f' and 'fs' are in the same Rec group (since f refers to fs via its RULE).

But watch out!  If 'fs' is not chosen as a loop breaker, we may get an infinite loop:
  - the RULE is applied in f's RHS (see Note [Rules for recursive functions] in GHC.Core.Opt.Simplify
  - fs is inlined (say it's small)
  - now there's another opportunity to apply the RULE

This showed up when compiling Control.Concurrent.Chan.getChanContents.
Hence the transitive rule_fv_env stuff described in
Note [Rules and loop breakers].

Note [Occurrence analysis for join points]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider these two somewhat artificial programs (#22404)

  Program (P1)                      Program (P2)
  ------------------------------    -------------------------------------
  let v = <small thunk> in          let v = <small thunk> in
                                    join j = case v of (a,b) -> a
  in case x of                      in case x of
        A -> case v of (a,b) -> a         A -> j
        B -> case v of (a,b) -> a         B -> j
        C -> case v of (a,b) -> b         C -> case v of (a,b) -> b
        D -> []                           D -> []

In (P1), `v` gets allocated, as a thunk, every time this code is executed.  But
notice that `v` occurs at most once in any case branch; the occurrence analyser
spots this and returns a OneOcc{ occ_n_br = 3 } for `v`.  Then the code in
GHC.Core.Opt.Simplify.Utils.postInlineUnconditionally inlines `v` at its three
use sites, and discards the let-binding.  That way, we avoid allocating `v` in
the A,B,C branches (though we still compute it of course), and branch D
doesn't involve <small thunk> at all.  This sometimes makes a Really Big
Difference.

In (P2) we have shared the common RHS of A, B, in a join point `j`.  We would
like to inline `v` in just the same way as in (P1).  But the usual strategy
for let bindings is conservative and uses `andUDs` to combine usage from j's
RHS to its body; as if `j` was called on every code path (once, albeit).  In
the case of (P2), we'll get ManyOccs for `v`.  Important optimisation lost!

Solving this problem makes the Simplifier less fragile.  For example,
the Simplifier might inline `j`, and convert (P2) into (P1)... or it might
not, depending in a perhaps-fragile way on the size of the join point.
I was motivated to implement this feature of the occurrence analyser
when trying to make optimisation join points simpler and more robust
(see e.g. #23627).

The occurrence analyser therefore has clever code that behaves just as
if you inlined `j` at all its call sites.  Here is a tricky variant
to keep in mind:

  Program (P3)
  -------------------------------
    join j = case v of (a,b) -> a
    in case f v of
          A -> j
          B -> j
          C -> []

If you mentally inline `j` you'll see that `v` is used twice on the path
through A, so it should have ManyOcc.  Bear this case in mind!

* We treat /non-recursive/ join points specially. Recursive join points are
  treated like any other letrec, as before.  Moreover, we only give this special
  treatment to /pre-existing/ non-recursive join points, not the ones that we
  discover for the first time in this sweep of the occurrence analyser.

* In occ_env, the new (occ_join_points :: IdEnv OccInfoEnv) maps
  each in-scope non-recursive join point, such as `j` above, to
  a "zeroed form" of its RHS's usage details. The "zeroed form"
    * deletes ManyOccs
    * maps a OneOcc to OneOcc{ occ_n_br = 0 }
  In our example, occ_join_points will be extended with
      [j :-> [v :-> OneOcc{occ_n_br=0}]]
  See addJoinPoint.

* At an occurrence of a join point, we do everything as normal, but add in the
  UsageDetails from the occ_join_points.  See mkOneOcc.

* Crucially, at the NonRec binding of the join point, in `occAnalBind`, we use
  `orUDs`, not `andUDs` to combine the usage from the RHS with the usage from
  the body.

Here are the consequences

* Because of the perhaps-surprising OneOcc{occ_n_br=0} idea of the zeroed
  form, the occ_n_br field of a OneOcc binder still counts the number of
  /actual lexical occurrences/ of the variable.  In Program P2, for example,
  `v` will end up with OneOcc{occ_n_br=2}, not occ_n_br=3.
  There are two lexical occurrences of `v`!
  (NB: `orUDs` adds occ_n_br together, so occ_n_br=1 is impossible, too.)

* In the tricky (P3) we'll get an `andUDs` of
    * OneOcc{occ_n_br=0} from the occurrences of `j`)
    * OneOcc{occ_n_br=1} from the (f v)
  These are `andUDs` together in `addOccInfo`, and hence
  `v` gets ManyOccs, just as it should.  Clever!

There are a couple of tricky wrinkles

(W1) Consider this example which shadows `j`:
          join j = rhs in
          in case x of { K j -> ..j..; ... }
     Clearly when we come to the pattern `K j` we must drop the `j`
     entry in occ_join_points.

     This is done by `drop_shadowed_joins` in `addInScope`.

(W2) Consider this example which shadows `v`:
          join j = ...v...
          in case x of { K v -> ..j..; ... }

     We can't make j's occurrences in the K alternative give rise to an
     occurrence of `v` (via occ_join_points), because it'll just be deleted by
     the `K v` pattern.  Yikes.  This is rare because shadowing is rare, but
     it definitely can happen.  Solution: when bringing `v` into scope at
     the `K v` pattern, chuck out of occ_join_points any elements whose
     UsageDetails mentions `v`.  Instead, just `andUDs` all that usage in
     right here.

     This requires work in two places.
     * In `preprocess_env`, we detect if the newly-bound variables intersect
       the free vars of occ_join_points.  (These free vars are conveniently
       simply the domain of the OccInfoEnv for that join point.) If so,
       we zap the entire occ_join_points.
     * In `postprcess_uds`, we add the chucked-out join points to the
       returned UsageDetails, with `andUDs`.

(W3) Consider this example, which shadows `j`, but this time in an argument
              join j = rhs
              in f (case x of { K j -> ...; ... })
     We can zap the entire occ_join_points when looking at the argument,
     because `j` can't posibly occur -- it's a join point!  And the smaller
     occ_join_points is, the better.  Smaller to look up in mkOneOcc, and
     more important, less looking-up when checking (W2).

     This is done in setNonTailCtxt.  It's important /not/ to do this for
     join-point RHS's because of course `j` can occur there!

     NB: this is just about efficiency: it is always safe /not/ to zap the
     occ_join_points.

(W4) What if the join point binding has a stable unfolding, or RULES?
     They are just alternative right-hand sides, and at each call site we
     will use only one of them. So again, we can use `orUDs` to combine
     usage info from all these alternatives RHSs.

Wrinkles (W1) and (W2) are very similar to Note [Binder swap] (BS3).

Note [Finding join points]
~~~~~~~~~~~~~~~~~~~~~~~~~~
It's the occurrence analyser's job to find bindings that we can turn into join
points, but it doesn't perform that transformation right away. Rather, it marks
the eligible bindings as part of their occurrence data, leaving it to the
simplifier (or to simpleOptPgm) to actually change the binder's 'IdDetails'.
The simplifier then eta-expands the RHS if needed and then updates the
occurrence sites. Dividing the work this way means that the occurrence analyser
still only takes one pass, yet one can always tell the difference between a
function call and a jump by looking at the occurrence (because the same pass
changes the 'IdDetails' and propagates the binders to their occurrence sites).

To track potential join points, we use the 'occ_tail' field of OccInfo. A value
of `AlwaysTailCalled n` indicates that every occurrence of the variable is a
tail call with `n` arguments (counting both value and type arguments). Otherwise
'occ_tail' will be 'NoTailCallInfo'. The tail call info flows bottom-up with the
rest of 'OccInfo' until it goes on the binder.

Note [Join arity prediction based on joinRhsArity]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In general, the join arity from tail occurrences of a join point (O) may be
higher or lower than the manifest join arity of the join body (M). E.g.,

  -- M > O:
  let f x y = x + y              -- M = 2
  in if b then f 1 else f 2      -- O = 1
  ==> { Contify for join arity 1 }
  join f x = \y -> x + y
  in if b then jump f 1 else jump f 2

  -- M < O
  let f = id                     -- M = 0
  in if ... then f 12 else f 13  -- O = 1
  ==> { Contify for join arity 1, eta-expand f }
  join f x = id x
  in if b then jump f 12 else jump f 13

But for *recursive* let, it is crucial that both arities match up, consider

  letrec f x y = if ... then f x else True
  in f 42

Here, M=2 but O=1. If we settled for a joinrec arity of 1, the recursive jump
would not happen in a tail context! Contification is invalid here.
So indeed it is crucial to demand that M=O.

(Side note: Actually, we could be more specific: Let O1 be the join arity of
occurrences from the letrec RHS and O2 the join arity from the let body. Then
we need M=O1 and M<=O2 and could simply eta-expand the RHS to match O2 later.
M=O is the specific case where we don't want to eta-expand. Neither the join
points paper nor GHC does this at the moment.)

We can capitalise on this observation and conclude that *if* f could become a
joinrec (without eta-expansion), it will have join arity M.
Now, M is just the result of 'joinRhsArity', a rather simple, local analysis.
It is also the join arity inside the 'TailUsageDetails' returned by
'occAnalLamTail', so we can predict join arity without doing any fixed-point
iteration or really doing any deep traversal of let body or RHS at all.
We check for M in the 'adjustTailUsage' call inside 'tagRecBinders'.

All this is quite apparent if you look at the contification transformation in
Fig. 5 of "Compiling without Continuations" (which does not account for
eta-expansion at all, mind you). The letrec case looks like this

  letrec f = /\as.\xs. L[us] in L'[es]
    ... and a bunch of conditions establishing that f only occurs
        in app heads of join arity (len as + len xs) inside us and es ...

The syntactic form `/\as.\xs. L[us]` forces M=O iff `f` occurs in `us`. However,
for non-recursive functions, this is the definition of contification from the
paper:

  let f = /\as.\xs.u in L[es]     ... conditions ...

Note that u could be a lambda itself, as we have seen. No relationship between M
and O to exploit here.

Note [Join points and unfoldings/rules]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
   let j2 y = blah
   let j x = j2 (x+x)
       {-# INLINE [2] j #-}
   in case e of { A -> j 1; B -> ...; C -> j 2 }

Before j is inlined, we'll have occurrences of j2 in
both j's RHS and in its stable unfolding.  We want to discover
j2 as a join point. So 'occAnalUnfolding' returns an unadjusted
'TailUsageDetails', like 'occAnalLamTail'. We adjust the usage details of the
unfolding to the actual join arity using the same 'adjustTailArity' as for
the RHS, see Note [Adjusting right-hand sides].

Same with rules. Suppose we have:

  let j :: Int -> Int
      j y = 2 * y
  let k :: Int -> Int -> Int
      {-# RULES "SPEC k 0" k 0 y = j y #-}
      k x y = x + 2 * y
  in case e of { A -> k 1 2; B -> k 3 5; C -> blah }

We identify k as a join point, and we want j to be a join point too.
Without the RULE it would be, and we don't want the RULE to mess it
up.  So provided the join-point arity of k matches the args of the
rule we can allow the tail-call info from the RHS of the rule to
propagate.

* Note that the join arity of the RHS and that of the unfolding or RULE might
  mismatch:

    let j x y = j2 (x+x)
        {-# INLINE[2] j = \x. g #-}
        {-# RULE forall x y z. j x y z = h 17 #-}
    in j 1 2

  So it is crucial that we adjust each TailUsageDetails individually
  with the actual join arity 2 here before we combine with `andUDs`.
  Here, that means losing tail call info on `g` and `h`.

* Wrinkle for Rec case: We store one TailUsageDetails in the node Details for
  RHS, unfolding and RULE combined. Clearly, if they don't agree on their join
  arity, we have to do some adjusting. We choose to adjust to the join arity
  of the RHS, because that is likely the join arity that the join point will
  have; see Note [Join arity prediction based on joinRhsArity].

  If the guess is correct, then tail calls in the RHS are preserved; a necessary
  condition for the whole binding becoming a joinrec.
  The guess can only be incorrect in the 'AcyclicSCC' case when the binding
  becomes a non-recursive join point with a different join arity. But then the
  eventual call to 'adjustTailUsage' in 'tagRecBinders'/'occAnalRec' will
  be with a different join arity and destroy unsound tail call info with
  'markNonTail'.

* Wrinkle for RULES.  Suppose the example was a bit different:
      let j :: Int -> Int
          j y = 2 * y
          k :: Int -> Int -> Int
          {-# RULES "SPEC k 0" k 0 = j #-}
          k x y = x + 2 * y
      in ...
  If we eta-expanded the rule all would be well, but as it stands the
  one arg of the rule don't match the join-point arity of 2.

  Conceivably we could notice that a potential join point would have
  an "undersaturated" rule and account for it. This would mean we
  could make something that's been specialised a join point, for
  instance. But local bindings are rarely specialised, and being
  overly cautious about rules only costs us anything when, for some `j`:

  * Before specialisation, `j` has non-tail calls, so it can't be a join point.
  * During specialisation, `j` gets specialised and thus acquires rules.
  * Sometime afterward, the non-tail calls to `j` disappear (as dead code, say),
    and so now `j` *could* become a join point.

  This appears to be very rare in practice. TODO Perhaps we should gather
  statistics to be sure.

------------------------------------------------------------
Note [Adjusting right-hand sides]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
There's a bit of a dance we need to do after analysing a lambda expression or
a right-hand side. In particular, we need to

  a) call 'markAllNonTail' *unless* the binding is for a join point, and
     the TailUsageDetails from the RHS has the right join arity; e.g.
        join j x y = case ... of
                       A -> j2 p
                       B -> j2 q
        in j a b
     Here we want the tail calls to j2 to be tail calls of the whole expression
  b) call 'markAllInsideLam' *unless* the binding is for a thunk, a one-shot
     lambda, or a non-recursive join point

Some examples, with how the free occurrences in e (assumed not to be a value
lambda) get marked:

                             inside lam    non-tail-called
  ------------------------------------------------------------
  let x = e                  No            Yes
  let f = \x -> e            Yes           Yes
  let f = \x{OneShot} -> e   No            Yes
  \x -> e                    Yes           Yes
  join j x = e               No            No
  joinrec j x = e            Yes           No

There are a few other caveats; most importantly, if we're marking a binding as
'AlwaysTailCalled', it's *going* to be a join point, so we treat it as one so
that the effect cascades properly. Consequently, at the time the RHS is
analysed, we won't know what adjustments to make; thus 'occAnalLamTail' must
return the unadjusted 'TailUsageDetails', to be adjusted by 'adjustTailUsage'
once join-point-hood has been decided and eventual one-shot annotations have
been added through 'markNonRecJoinOneShots'.

It is not so simple to see that 'occAnalNonRecBind' and 'occAnalRecBind' indeed
perform a similar sequence of steps. Thus, here is an interleaving of events
of both functions, serving as a specification:

  1. Call 'occAnalLamTail' to find usage information for the RHS.
     Recursive case:     'makeNode'
     Non-recursive case: 'occAnalNonRecBind'
  2. (Analyse the binding's scope. Done in 'occAnalBind'/`occAnal Let{}`.
      Same whether recursive or not.)
  3. Call 'tagNonRecBinder' or 'tagRecBinders', which decides whether to make
     the binding a join point.
     Cyclic  Recursive case:  'mkLoopBreakerNodes'
     Acyclic Recursive case:  `occAnalRec AcyclicSCC{}`
     Non-recursive case:      'occAnalNonRecBind'
  4. Non-recursive join point: Call 'markNonRecJoinOneShots' so that e.g.,
     FloatOut sees one-shot annotations on lambdas
     Acyclic Recursive case:  `occAnalRec AcyclicSCC{}`  calls 'adjustNonRecRhs'
     Non-recursive case:      'occAnalNonRecBind'        calls 'adjustNonRecRhs'
  5. Call 'adjustTailUsage' accordingly.
     Cyclic Recursive case:   'tagRecBinders'
     Acyclic Recursive case:  'adjustNonRecRhs'
     Non-recursive case:      'adjustNonRecRhs'
-}

------------------------------------------------------------------
--                 occAnalBind
------------------------------------------------------------------

occAnalBind
  :: OccEnv
  -> TopLevelFlag
  -> ImpRuleEdges
  -> CoreBind
  -> (OccEnv -> WithUsageDetails r)  -- Scope of the bind
  -> ([CoreBind] -> r -> r)          -- How to combine the scope with new binds
  -> WithUsageDetails r              -- Of the whole let(rec)

occAnalBind :: forall r.
OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> CoreBind
-> (OccEnv -> WithUsageDetails r)
-> (CoreProgram -> r -> r)
-> WithUsageDetails r
occAnalBind OccEnv
env TopLevelFlag
lvl ImpRuleEdges
ire (Rec [(Id, CoreExpr)]
pairs) OccEnv -> WithUsageDetails r
thing_inside CoreProgram -> r -> r
combine
  = OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails r) -> WithUsageDetails r
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeList OccEnv
env (((Id, CoreExpr) -> Id) -> [(Id, CoreExpr)] -> [Id]
forall a b. (a -> b) -> [a] -> [b]
map (Id, CoreExpr) -> Id
forall a b. (a, b) -> a
fst [(Id, CoreExpr)]
pairs) ((OccEnv -> WithUsageDetails r) -> WithUsageDetails r)
-> (OccEnv -> WithUsageDetails r) -> WithUsageDetails r
forall a b. (a -> b) -> a -> b
$ \OccEnv
env ->
    let WUD UsageDetails
body_uds r
body'  = OccEnv -> WithUsageDetails r
thing_inside OccEnv
env
        WUD UsageDetails
bind_uds CoreProgram
binds' = OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> [(Id, CoreExpr)]
-> UsageDetails
-> WithUsageDetails CoreProgram
occAnalRecBind OccEnv
env TopLevelFlag
lvl ImpRuleEdges
ire [(Id, CoreExpr)]
pairs UsageDetails
body_uds
    in UsageDetails -> r -> WithUsageDetails r
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
bind_uds (CoreProgram -> r -> r
combine CoreProgram
binds' r
body')

occAnalBind !OccEnv
env TopLevelFlag
lvl ImpRuleEdges
ire (NonRec Id
bndr CoreExpr
rhs) OccEnv -> WithUsageDetails r
thing_inside CoreProgram -> r -> r
combine
  | Id -> Bool
isTyVar Id
bndr      -- A type let; we don't gather usage info
  = let !(WUD UsageDetails
body_uds r
res) = OccEnv
-> Id -> (OccEnv -> WithUsageDetails r) -> WithUsageDetails r
forall a.
OccEnv
-> Id -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeOne OccEnv
env Id
bndr OccEnv -> WithUsageDetails r
thing_inside
    in UsageDetails -> r -> WithUsageDetails r
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
body_uds (CoreProgram -> r -> r
combine [Id -> CoreExpr -> CoreBind
forall b. b -> Expr b -> Bind b
NonRec Id
bndr CoreExpr
rhs] r
res)

  -- /Existing/ non-recursive join points
  -- See Note [Occurrence analysis for join points]
  | mb_join :: JoinPointHood
mb_join@(JoinPoint {}) <- Id -> JoinPointHood
idJoinPointHood Id
bndr
  = -- Analyse the RHS and /then/ the body
    let -- Analyse the rhs first, generating rhs_uds
        !([UsageDetails]
rhs_uds_s, Id
bndr', CoreExpr
rhs') = OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> JoinPointHood
-> Id
-> CoreExpr
-> ([UsageDetails], Id, CoreExpr)
occAnalNonRecRhs OccEnv
env TopLevelFlag
lvl ImpRuleEdges
ire JoinPointHood
mb_join Id
bndr CoreExpr
rhs
        rhs_uds :: UsageDetails
rhs_uds = (UsageDetails -> UsageDetails -> UsageDetails)
-> [UsageDetails] -> UsageDetails
forall a. (a -> a -> a) -> [a] -> a
forall (t :: * -> *) a. Foldable t => (a -> a -> a) -> t a -> a
foldr1 UsageDetails -> UsageDetails -> UsageDetails
orUDs [UsageDetails]
rhs_uds_s   -- NB: orUDs.  See (W4) of
                                           -- Note [Occurrence analysis for join points]

        -- Now analyse the body, adding the join point
        -- into the environment with addJoinPoint
        !(WUD UsageDetails
body_uds (OccInfo
occ, r
body)) = OccEnv
-> Id
-> (OccEnv -> WithUsageDetails r)
-> WithUsageDetails (OccInfo, r)
forall r.
OccEnv
-> Id
-> (OccEnv -> WithUsageDetails r)
-> WithUsageDetails (OccInfo, r)
occAnalNonRecBody OccEnv
env Id
bndr' ((OccEnv -> WithUsageDetails r) -> WithUsageDetails (OccInfo, r))
-> (OccEnv -> WithUsageDetails r) -> WithUsageDetails (OccInfo, r)
forall a b. (a -> b) -> a -> b
$ \OccEnv
env ->
                                      OccEnv -> WithUsageDetails r
thing_inside (OccEnv -> Id -> UsageDetails -> OccEnv
addJoinPoint OccEnv
env Id
bndr' UsageDetails
rhs_uds)
    in
    if OccInfo -> Bool
isDeadOcc OccInfo
occ     -- Drop dead code; see Note [Dead code]
    then UsageDetails -> r -> WithUsageDetails r
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
body_uds r
body
    else UsageDetails -> r -> WithUsageDetails r
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails
rhs_uds UsageDetails -> UsageDetails -> UsageDetails
`orUDs` UsageDetails
body_uds)    -- Note `orUDs`
             (CoreProgram -> r -> r
combine [Id -> CoreExpr -> CoreBind
forall b. b -> Expr b -> Bind b
NonRec ((Id, JoinPointHood) -> Id
forall a b. (a, b) -> a
fst (TopLevelFlag -> OccInfo -> Id -> (Id, JoinPointHood)
tagNonRecBinder TopLevelFlag
lvl OccInfo
occ Id
bndr')) CoreExpr
rhs']
                      r
body)

  -- The normal case, including newly-discovered join points
  -- Analyse the body and /then/ the RHS
  | WUD UsageDetails
body_uds (OccInfo
occ,r
body) <- OccEnv
-> Id
-> (OccEnv -> WithUsageDetails r)
-> WithUsageDetails (OccInfo, r)
forall r.
OccEnv
-> Id
-> (OccEnv -> WithUsageDetails r)
-> WithUsageDetails (OccInfo, r)
occAnalNonRecBody OccEnv
env Id
bndr OccEnv -> WithUsageDetails r
thing_inside
  = if OccInfo -> Bool
isDeadOcc OccInfo
occ   -- Drop dead code; see Note [Dead code]
    then UsageDetails -> r -> WithUsageDetails r
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
body_uds r
body
    else let
        -- Get the join info from the *new* decision; NB: bndr is not already a JoinId
        -- See Note [Join points and unfoldings/rules]
        -- => join arity O of Note [Join arity prediction based on joinRhsArity]
        (Id
tagged_bndr, JoinPointHood
mb_join) = TopLevelFlag -> OccInfo -> Id -> (Id, JoinPointHood)
tagNonRecBinder TopLevelFlag
lvl OccInfo
occ Id
bndr

        !([UsageDetails]
rhs_uds_s, Id
final_bndr, CoreExpr
rhs') = OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> JoinPointHood
-> Id
-> CoreExpr
-> ([UsageDetails], Id, CoreExpr)
occAnalNonRecRhs OccEnv
env TopLevelFlag
lvl ImpRuleEdges
ire JoinPointHood
mb_join Id
tagged_bndr CoreExpr
rhs
    in UsageDetails -> r -> WithUsageDetails r
forall a. UsageDetails -> a -> WithUsageDetails a
WUD ((UsageDetails -> UsageDetails -> UsageDetails)
-> UsageDetails -> [UsageDetails] -> UsageDetails
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr UsageDetails -> UsageDetails -> UsageDetails
andUDs UsageDetails
body_uds [UsageDetails]
rhs_uds_s)      -- Note `andUDs`
           (CoreProgram -> r -> r
combine [Id -> CoreExpr -> CoreBind
forall b. b -> Expr b -> Bind b
NonRec Id
final_bndr CoreExpr
rhs'] r
body)

-----------------
occAnalNonRecBody :: OccEnv -> Id
                  -> (OccEnv -> WithUsageDetails r)  -- Scope of the bind
                  -> (WithUsageDetails (OccInfo, r))
occAnalNonRecBody :: forall r.
OccEnv
-> Id
-> (OccEnv -> WithUsageDetails r)
-> WithUsageDetails (OccInfo, r)
occAnalNonRecBody OccEnv
env Id
bndr OccEnv -> WithUsageDetails r
thing_inside
  = OccEnv
-> Id
-> (OccEnv -> WithUsageDetails (OccInfo, r))
-> WithUsageDetails (OccInfo, r)
forall a.
OccEnv
-> Id -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeOne OccEnv
env Id
bndr ((OccEnv -> WithUsageDetails (OccInfo, r))
 -> WithUsageDetails (OccInfo, r))
-> (OccEnv -> WithUsageDetails (OccInfo, r))
-> WithUsageDetails (OccInfo, r)
forall a b. (a -> b) -> a -> b
$ \OccEnv
env ->
    let !(WUD UsageDetails
inner_uds r
res) = OccEnv -> WithUsageDetails r
thing_inside OccEnv
env
        !occ :: OccInfo
occ = UsageDetails -> Id -> OccInfo
lookupLetOccInfo UsageDetails
inner_uds Id
bndr
    in UsageDetails -> (OccInfo, r) -> WithUsageDetails (OccInfo, r)
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
inner_uds (OccInfo
occ, r
res)

-----------------
occAnalNonRecRhs :: OccEnv -> TopLevelFlag -> ImpRuleEdges
                -> JoinPointHood -> Id -> CoreExpr
                 -> ([UsageDetails], Id, CoreExpr)
occAnalNonRecRhs :: OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> JoinPointHood
-> Id
-> CoreExpr
-> ([UsageDetails], Id, CoreExpr)
occAnalNonRecRhs !OccEnv
env TopLevelFlag
lvl ImpRuleEdges
imp_rule_edges JoinPointHood
mb_join Id
bndr CoreExpr
rhs
  | [CoreRule] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [CoreRule]
rules, [(Activation, VarSet)] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [(Activation, VarSet)]
imp_rule_infos
  =  -- Fast path for common case of no rules. This is only worth
     -- 0.1% perf on average, but it's also only a line or two of code
    ( [UsageDetails
adj_rhs_uds, UsageDetails
adj_unf_uds],              Id
final_bndr_no_rules,   CoreExpr
final_rhs )
  | Bool
otherwise
  = (UsageDetails
adj_rhs_uds UsageDetails -> [UsageDetails] -> [UsageDetails]
forall a. a -> [a] -> [a]
: UsageDetails
adj_unf_uds UsageDetails -> [UsageDetails] -> [UsageDetails]
forall a. a -> [a] -> [a]
: [UsageDetails]
adj_rule_uds, Id
final_bndr_with_rules, CoreExpr
final_rhs )
  where
    --------- Right hand side ---------
    -- For join points, set occ_encl to OccVanilla, via setTailCtxt.  If we have
    --    join j = Just (f x) in ...
    -- we do not want to float the (f x) to
    --    let y = f x in join j = Just y in ...
    -- That's that OccRhs would do; but there's no point because
    -- j will never be scrutinised.
    rhs_env :: OccEnv
rhs_env  = OccEnv
-> RecFlag -> OccEncl -> JoinPointHood -> Id -> CoreExpr -> OccEnv
mkRhsOccEnv OccEnv
env RecFlag
NonRecursive OccEncl
rhs_ctxt JoinPointHood
mb_join Id
bndr CoreExpr
rhs
    rhs_ctxt :: OccEncl
rhs_ctxt = TopLevelFlag -> Id -> Unfolding -> OccEncl
mkNonRecRhsCtxt TopLevelFlag
lvl Id
bndr Unfolding
unf

    -- See Note [Join arity prediction based on joinRhsArity]
    -- Match join arity O from mb_join_arity with manifest join arity M as
    -- returned by of occAnalLamTail. It's totally OK for them to mismatch;
    -- hence adjust the UDs from the RHS
    WUD UsageDetails
adj_rhs_uds CoreExpr
final_rhs = JoinPointHood
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
adjustNonRecRhs JoinPointHood
mb_join (WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr)
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
forall a b. (a -> b) -> a -> b
$
                                OccEnv -> CoreExpr -> WithTailUsageDetails CoreExpr
occAnalLamTail OccEnv
rhs_env CoreExpr
rhs
    final_bndr_with_rules :: Id
final_bndr_with_rules
      | OccEnv -> Bool
noBinderSwaps OccEnv
env = Id
bndr -- See Note [Unfoldings and rules]
      | Bool
otherwise         = Id
bndr Id -> RuleInfo -> Id
`setIdSpecialisation` [CoreRule] -> RuleInfo
mkRuleInfo [CoreRule]
rules'
                                 Id -> Unfolding -> Id
`setIdUnfolding` Unfolding
unf1
    final_bndr_no_rules :: Id
final_bndr_no_rules
      | OccEnv -> Bool
noBinderSwaps OccEnv
env = Id
bndr -- See Note [Unfoldings and rules]
      | Bool
otherwise         = Id
bndr Id -> Unfolding -> Id
`setIdUnfolding` Unfolding
unf1

    --------- Unfolding ---------
    -- See Note [Join points and unfoldings/rules]
    unf :: Unfolding
unf = IdUnfoldingFun
idUnfolding Id
bndr
    WTUD TailUsageDetails
unf_tuds Unfolding
unf1 = OccEnv -> Unfolding -> WithTailUsageDetails Unfolding
occAnalUnfolding OccEnv
rhs_env Unfolding
unf
    adj_unf_uds :: UsageDetails
adj_unf_uds = JoinPointHood -> TailUsageDetails -> UsageDetails
adjustTailArity JoinPointHood
mb_join TailUsageDetails
unf_tuds

    --------- Rules ---------
    -- See Note [Rules are extra RHSs] and Note [Rule dependency info]
    -- and Note [Join points and unfoldings/rules]
    rules :: [CoreRule]
rules        = Id -> [CoreRule]
idCoreRules Id
bndr
    rules_w_uds :: [(CoreRule, UsageDetails, TailUsageDetails)]
rules_w_uds  = (CoreRule -> (CoreRule, UsageDetails, TailUsageDetails))
-> [CoreRule] -> [(CoreRule, UsageDetails, TailUsageDetails)]
forall a b. (a -> b) -> [a] -> [b]
map (OccEnv -> CoreRule -> (CoreRule, UsageDetails, TailUsageDetails)
occAnalRule OccEnv
rhs_env) [CoreRule]
rules
    rules' :: [CoreRule]
rules'       = ((CoreRule, UsageDetails, TailUsageDetails) -> CoreRule)
-> [(CoreRule, UsageDetails, TailUsageDetails)] -> [CoreRule]
forall a b. (a -> b) -> [a] -> [b]
map (CoreRule, UsageDetails, TailUsageDetails) -> CoreRule
forall a b c. (a, b, c) -> a
fstOf3 [(CoreRule, UsageDetails, TailUsageDetails)]
rules_w_uds
    imp_rule_infos :: [(Activation, VarSet)]
imp_rule_infos = ImpRuleEdges -> Id -> [(Activation, VarSet)]
lookupImpRules ImpRuleEdges
imp_rule_edges Id
bndr
    imp_rule_uds :: [UsageDetails]
imp_rule_uds   = [[(Activation, VarSet)] -> UsageDetails
impRulesScopeUsage [(Activation, VarSet)]
imp_rule_infos]
         -- imp_rule_uds: consider
         --     h = ...
         --     g = ...
         --     RULE map g = h
         -- Then we want to ensure that h is in scope everywhere
         -- that g is (since the RULE might turn g into h), so
         -- we make g mention h.

    adj_rule_uds :: [UsageDetails]
    adj_rule_uds :: [UsageDetails]
adj_rule_uds = [UsageDetails]
imp_rule_uds [UsageDetails] -> [UsageDetails] -> [UsageDetails]
forall a. [a] -> [a] -> [a]
++
                   [ UsageDetails
l UsageDetails -> UsageDetails -> UsageDetails
`andUDs` JoinPointHood -> TailUsageDetails -> UsageDetails
adjustTailArity JoinPointHood
mb_join TailUsageDetails
r
                   | (CoreRule
_,UsageDetails
l,TailUsageDetails
r) <- [(CoreRule, UsageDetails, TailUsageDetails)]
rules_w_uds ]

mkNonRecRhsCtxt :: TopLevelFlag -> Id -> Unfolding -> OccEncl
-- Precondition: Id is not a join point
mkNonRecRhsCtxt :: TopLevelFlag -> Id -> Unfolding -> OccEncl
mkNonRecRhsCtxt TopLevelFlag
lvl Id
bndr Unfolding
unf
  | Bool
certainly_inline = OccEncl
OccVanilla -- See Note [Cascading inlines]
  | Bool
otherwise        = OccEncl
OccRhs
  where
    certainly_inline :: Bool
certainly_inline -- See Note [Cascading inlines]
      = -- mkNonRecRhsCtxt is only used for non-join points, so occAnalBind
        -- has set the OccInfo for this binder before calling occAnalNonRecRhs
        case Id -> OccInfo
idOccInfo Id
bndr of
          OneOcc { occ_in_lam :: OccInfo -> InsideLam
occ_in_lam = InsideLam
NotInsideLam, occ_n_br :: OccInfo -> JoinArity
occ_n_br = JoinArity
1 }
            -> Bool
active Bool -> Bool -> Bool
&& Bool -> Bool
not Bool
stable_unf Bool -> Bool -> Bool
&& Bool -> Bool
not Bool
top_bottoming
          OccInfo
_ -> Bool
False

    active :: Bool
active     = Activation -> Bool
isAlwaysActive (Id -> Activation
idInlineActivation Id
bndr)
    stable_unf :: Bool
stable_unf = Unfolding -> Bool
isStableUnfolding Unfolding
unf
    top_bottoming :: Bool
top_bottoming = TopLevelFlag -> Bool
isTopLevel TopLevelFlag
lvl Bool -> Bool -> Bool
&& Id -> Bool
isDeadEndId Id
bndr

-----------------
occAnalRecBind :: OccEnv -> TopLevelFlag -> ImpRuleEdges -> [(Var,CoreExpr)]
               -> UsageDetails -> WithUsageDetails [CoreBind]
-- For a recursive group, we
--      * occ-analyse all the RHSs
--      * compute strongly-connected components
--      * feed those components to occAnalRec
-- See Note [Recursive bindings: the grand plan]
occAnalRecBind :: OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> [(Id, CoreExpr)]
-> UsageDetails
-> WithUsageDetails CoreProgram
occAnalRecBind !OccEnv
rhs_env TopLevelFlag
lvl ImpRuleEdges
imp_rule_edges [(Id, CoreExpr)]
pairs UsageDetails
body_usage
  = (SCC NodeDetails
 -> WithUsageDetails CoreProgram -> WithUsageDetails CoreProgram)
-> WithUsageDetails CoreProgram
-> [SCC NodeDetails]
-> WithUsageDetails CoreProgram
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (OccEnv
-> TopLevelFlag
-> SCC NodeDetails
-> WithUsageDetails CoreProgram
-> WithUsageDetails CoreProgram
occAnalRec OccEnv
rhs_env TopLevelFlag
lvl) (UsageDetails -> CoreProgram -> WithUsageDetails CoreProgram
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
body_usage []) [SCC NodeDetails]
sccs
  where
    sccs :: [SCC NodeDetails]
    sccs :: [SCC NodeDetails]
sccs = [Node Unique NodeDetails] -> [SCC NodeDetails]
forall key payload.
Uniquable key =>
[Node key payload] -> [SCC payload]
stronglyConnCompFromEdgedVerticesUniq [Node Unique NodeDetails]
nodes

    nodes :: [LetrecNode]
    nodes :: [Node Unique NodeDetails]
nodes = ((Id, CoreExpr) -> Node Unique NodeDetails)
-> [(Id, CoreExpr)] -> [Node Unique NodeDetails]
forall a b. (a -> b) -> [a] -> [b]
map (OccEnv
-> ImpRuleEdges
-> VarSet
-> (Id, CoreExpr)
-> Node Unique NodeDetails
makeNode OccEnv
rhs_env ImpRuleEdges
imp_rule_edges VarSet
bndr_set) [(Id, CoreExpr)]
pairs

    bndrs :: [Id]
bndrs    = ((Id, CoreExpr) -> Id) -> [(Id, CoreExpr)] -> [Id]
forall a b. (a -> b) -> [a] -> [b]
map (Id, CoreExpr) -> Id
forall a b. (a, b) -> a
fst [(Id, CoreExpr)]
pairs
    bndr_set :: VarSet
bndr_set = [Id] -> VarSet
mkVarSet [Id]
bndrs

-----------------------------
occAnalRec :: OccEnv -> TopLevelFlag
           -> SCC NodeDetails
           -> WithUsageDetails [CoreBind]
           -> WithUsageDetails [CoreBind]

-- The NonRec case is just like a Let (NonRec ...) above
occAnalRec :: OccEnv
-> TopLevelFlag
-> SCC NodeDetails
-> WithUsageDetails CoreProgram
-> WithUsageDetails CoreProgram
occAnalRec !OccEnv
_ TopLevelFlag
lvl
           (AcyclicSCC (ND { nd_bndr :: NodeDetails -> Id
nd_bndr = Id
bndr, nd_rhs :: NodeDetails -> WithTailUsageDetails CoreExpr
nd_rhs = WithTailUsageDetails CoreExpr
wtuds }))
           (WUD UsageDetails
body_uds CoreProgram
binds)
  | OccInfo -> Bool
isDeadOcc OccInfo
occ  -- Check for dead code: see Note [Dead code]
  = UsageDetails -> CoreProgram -> WithUsageDetails CoreProgram
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
body_uds CoreProgram
binds
  | Bool
otherwise
  = let (Id
bndr', JoinPointHood
mb_join) = TopLevelFlag -> OccInfo -> Id -> (Id, JoinPointHood)
tagNonRecBinder TopLevelFlag
lvl OccInfo
occ Id
bndr
        !(WUD UsageDetails
rhs_uds' CoreExpr
rhs') = JoinPointHood
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
adjustNonRecRhs JoinPointHood
mb_join WithTailUsageDetails CoreExpr
wtuds
    in UsageDetails -> CoreProgram -> WithUsageDetails CoreProgram
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails
body_uds UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
rhs_uds')
           (Id -> CoreExpr -> CoreBind
forall b. b -> Expr b -> Bind b
NonRec Id
bndr' CoreExpr
rhs' CoreBind -> CoreProgram -> CoreProgram
forall a. a -> [a] -> [a]
: CoreProgram
binds)
  where
    occ :: OccInfo
occ = UsageDetails -> Id -> OccInfo
lookupLetOccInfo UsageDetails
body_uds Id
bndr

-- The Rec case is the interesting one
-- See Note [Recursive bindings: the grand plan]
-- See Note [Loop breaking]
occAnalRec OccEnv
env TopLevelFlag
lvl (CyclicSCC [NodeDetails]
details_s) (WUD UsageDetails
body_uds CoreProgram
binds)
  | Bool -> Bool
not ((NodeDetails -> Bool) -> [NodeDetails] -> Bool
forall (t :: * -> *) a. Foldable t => (a -> Bool) -> t a -> Bool
any NodeDetails -> Bool
needed [NodeDetails]
details_s)
  = -- Check for dead code: see Note [Dead code]
    -- NB: Only look at body_uds, ignoring uses in the SCC
    UsageDetails -> CoreProgram -> WithUsageDetails CoreProgram
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
body_uds CoreProgram
binds

  | Bool
otherwise
  = UsageDetails -> CoreProgram -> WithUsageDetails CoreProgram
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
final_uds ([(Id, CoreExpr)] -> CoreBind
forall b. [(b, Expr b)] -> Bind b
Rec [(Id, CoreExpr)]
pairs CoreBind -> CoreProgram -> CoreProgram
forall a. a -> [a] -> [a]
: CoreProgram
binds)
  where
    all_simple :: Bool
all_simple = (NodeDetails -> Bool) -> [NodeDetails] -> Bool
forall (t :: * -> *) a. Foldable t => (a -> Bool) -> t a -> Bool
all NodeDetails -> Bool
nd_simple [NodeDetails]
details_s

    needed :: NodeDetails -> Bool
    needed :: NodeDetails -> Bool
needed (ND { nd_bndr :: NodeDetails -> Id
nd_bndr = Id
bndr }) = Id -> Bool
isExportedId Id
bndr Bool -> Bool -> Bool
|| Id
bndr Id -> VarEnv LocalOcc -> Bool
forall a. Id -> VarEnv a -> Bool
`elemVarEnv` VarEnv LocalOcc
body_env
    body_env :: VarEnv LocalOcc
body_env = UsageDetails -> VarEnv LocalOcc
ud_env UsageDetails
body_uds

    ------------------------------
    -- Make the nodes for the loop-breaker analysis
    -- See Note [Choosing loop breakers] for loop_breaker_nodes
    final_uds :: UsageDetails
    loop_breaker_nodes :: [LoopBreakerNode]
    WUD UsageDetails
final_uds [LoopBreakerNode]
loop_breaker_nodes = OccEnv
-> TopLevelFlag
-> UsageDetails
-> [NodeDetails]
-> WithUsageDetails [LoopBreakerNode]
mkLoopBreakerNodes OccEnv
env TopLevelFlag
lvl UsageDetails
body_uds [NodeDetails]
details_s

    ------------------------------
    weak_fvs :: VarSet
    weak_fvs :: VarSet
weak_fvs = (NodeDetails -> VarSet) -> [NodeDetails] -> VarSet
forall a. (a -> VarSet) -> [a] -> VarSet
mapUnionVarSet NodeDetails -> VarSet
nd_weak_fvs [NodeDetails]
details_s

    ---------------------------
    -- Now reconstruct the cycle
    pairs :: [(Id,CoreExpr)]
    pairs :: [(Id, CoreExpr)]
pairs | Bool
all_simple = JoinArity
-> VarSet
-> [LoopBreakerNode]
-> [(Id, CoreExpr)]
-> [(Id, CoreExpr)]
reOrderNodes   JoinArity
0 VarSet
weak_fvs [LoopBreakerNode]
loop_breaker_nodes []
          | Bool
otherwise  = JoinArity
-> VarSet
-> [LoopBreakerNode]
-> [(Id, CoreExpr)]
-> [(Id, CoreExpr)]
loopBreakNodes JoinArity
0 VarSet
weak_fvs [LoopBreakerNode]
loop_breaker_nodes []
          -- In the common case when all are "simple" (no rules at all)
          -- the loop_breaker_nodes will include all the scope edges
          -- so a SCC computation would yield a single CyclicSCC result;
          -- and reOrderNodes deals with exactly that case.
          -- Saves a SCC analysis in a common case


{- *********************************************************************
*                                                                      *
                Loop breaking
*                                                                      *
********************************************************************* -}

{- Note [Choosing loop breakers]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In Step 4 in Note [Recursive bindings: the grand plan]), occAnalRec does
loop-breaking on each CyclicSCC of the original program:

* mkLoopBreakerNodes: Form the loop-breaker graph for that CyclicSCC

* loopBreakNodes: Do SCC analysis on it

* reOrderNodes: For each CyclicSCC, pick a loop breaker
    * Delete edges to that loop breaker
    * Do another SCC analysis on that reduced SCC
    * Repeat

To form the loop-breaker graph, we construct a new set of Nodes, the
"loop-breaker nodes", with the same details but different edges, the
"loop-breaker edges".  The loop-breaker nodes have both more and fewer
dependencies than the scope edges:

  More edges:
     If f calls g, and g has an active rule that mentions h then
     we add an edge from f -> h.  See Note [Rules and loop breakers].

  Fewer edges: we only include dependencies
     * only on /active/ rules,
     * on rule /RHSs/ (not LHSs)

The scope edges, by contrast, must be much more inclusive.

The nd_simple flag tracks the common case when a binding has no RULES
at all, in which case the loop-breaker edges will be identical to the
scope edges.

Note that in Example [eftInt], *neither* eftInt *nor* eftIntFB is
chosen as a loop breaker, because their RHSs don't mention each other.
And indeed both can be inlined safely.

Note [inl_fvs]
~~~~~~~~~~~~~~
Note that the loop-breaker graph includes edges for occurrences in
/both/ the RHS /and/ the stable unfolding.  Consider this, which actually
occurred when compiling BooleanFormula.hs in GHC:

  Rec { lvl1 = go
      ; lvl2[StableUnf = go] = lvl1
      ; go = ...go...lvl2... }

From the point of view of infinite inlining, we need only these edges:
   lvl1 :-> go
   lvl2 :-> go       -- The RHS lvl1 will never be used for inlining
   go   :-> go, lvl2

But the danger is that, lacking any edge to lvl1, we'll put it at the
end thus
  Rec { lvl2[ StableUnf = go] = lvl1
      ; go[LoopBreaker] = ...go...lvl2... }
      ; lvl1[Occ=Once]  = go }

And now the Simplifer will try to use PreInlineUnconditionally on lvl1
(which occurs just once), but because it is last we won't actually
substitute in lvl2.  Sigh.

To avoid this possibility, we include edges from lvl2 to /both/ its
stable unfolding /and/ its RHS.  Hence the defn of inl_fvs in
makeNode.  Maybe we could be more clever, but it's very much a corner
case.

Note [Weak loop breakers]
~~~~~~~~~~~~~~~~~~~~~~~~~
There is a last nasty wrinkle.  Suppose we have

    Rec { f = f_rhs
          RULE f [] = g

          h = h_rhs
          g = h
          ...more... }

Remember that we simplify the RULES before any RHS (see Note
[Rules are visible in their own rec group] above).

So we must *not* postInlineUnconditionally 'g', even though
its RHS turns out to be trivial.  (I'm assuming that 'g' is
not chosen as a loop breaker.)  Why not?  Because then we
drop the binding for 'g', which leaves it out of scope in the
RULE!

Here's a somewhat different example of the same thing
    Rec { q = r
        ; r = ...p...
        ; p = p_rhs
          RULE p [] = q }
Here the RULE is "below" q, but we *still* can't postInlineUnconditionally
q, because the RULE for p is active throughout.  So the RHS of r
might rewrite to     r = ...q...
So q must remain in scope in the output program!

We "solve" this by:

    Make q a "weak" loop breaker (OccInfo = IAmLoopBreaker True)
    iff q is a mentioned in the RHS of any RULE (active on not)
    in the Rec group

Note the "active or not" comment; even if a RULE is inactive, we
want its RHS free vars to stay alive (#20820)!

A normal "strong" loop breaker has IAmLoopBreaker False.  So:

                                Inline  postInlineUnconditionally
strong   IAmLoopBreaker False    no      no
weak     IAmLoopBreaker True     yes     no
         other                   yes     yes

The **sole** reason for this kind of loop breaker is so that
postInlineUnconditionally does not fire.  Ugh.

Annoyingly, since we simplify the rules *first* we'll never inline
q into p's RULE.  That trivial binding for q will hang around until
we discard the rule.  Yuk.  But it's rare.

Note [Rules and loop breakers]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
When we form the loop-breaker graph (Step 4 in Note [Recursive
bindings: the grand plan]), we must be careful about RULEs.

For a start, we want a loop breaker to cut every cycle, so inactive
rules play no part; we need only consider /active/ rules.
See Note [Finding rule RHS free vars]

The second point is more subtle.  A RULE is like an equation for
'f' that is *always* inlined if it is applicable.  We do *not* disable
rules for loop-breakers.  It's up to whoever makes the rules to make
sure that the rules themselves always terminate.  See Note [Rules for
recursive functions] in GHC.Core.Opt.Simplify

Hence, if
    f's RHS (or its stable unfolding if it has one) mentions g, and
    g has a RULE that mentions h, and
    h has a RULE that mentions f

then we *must* choose f to be a loop breaker.  Example: see Note
[Specialisation rules]. So our plan is this:

   Take the free variables of f's RHS, and augment it with all the
   variables reachable by a transitive sequence RULES from those
   starting points.

That is the whole reason for computing rule_fv_env in mkLoopBreakerNodes.
Wrinkles:

* We only consider /active/ rules. See Note [Finding rule RHS free vars]

* We need only consider free vars that are also binders in this Rec
  group.  See also Note [Finding rule RHS free vars]

* We only consider variables free in the *RHS* of the rule, in
  contrast to the way we build the Rec group in the first place (Note
  [Rule dependency info])

* Why "transitive sequence of rules"?  Because active rules apply
  unconditionally, without checking loop-breaker-ness.
 See Note [Loop breaker dependencies].

Note [Finding rule RHS free vars]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider this real example from Data Parallel Haskell
     tagZero :: Array Int -> Array Tag
     {-# INLINE [1] tagZeroes #-}
     tagZero xs = pmap (\x -> fromBool (x==0)) xs

     {-# RULES "tagZero" [~1] forall xs n.
         pmap fromBool <blah blah> = tagZero xs #-}
So tagZero's RHS mentions pmap, and pmap's RULE mentions tagZero.
However, tagZero can only be inlined in phase 1 and later, while
the RULE is only active *before* phase 1.  So there's no problem.

To make this work, we look for the RHS free vars only for
*active* rules. That's the reason for the occ_rule_act field
of the OccEnv.

Note [loopBreakNodes]
~~~~~~~~~~~~~~~~~~~~~
loopBreakNodes is applied to the list of nodes for a cyclic strongly
connected component (there's guaranteed to be a cycle).  It returns
the same nodes, but
        a) in a better order,
        b) with some of the Ids having a IAmALoopBreaker pragma

The "loop-breaker" Ids are sufficient to break all cycles in the SCC.  This means
that the simplifier can guarantee not to loop provided it never records an inlining
for these no-inline guys.

Furthermore, the order of the binds is such that if we neglect dependencies
on the no-inline Ids then the binds are topologically sorted.  This means
that the simplifier will generally do a good job if it works from top bottom,
recording inlinings for any Ids which aren't marked as "no-inline" as it goes.
-}

type Binding = (Id,CoreExpr)

-- See Note [loopBreakNodes]
loopBreakNodes :: Int
               -> VarSet        -- Binders whose dependencies may be "missing"
                                -- See Note [Weak loop breakers]
               -> [LoopBreakerNode]
               -> [Binding]             -- Append these to the end
               -> [Binding]

-- Return the bindings sorted into a plausible order, and marked with loop breakers.
-- See Note [loopBreakNodes]
loopBreakNodes :: JoinArity
-> VarSet
-> [LoopBreakerNode]
-> [(Id, CoreExpr)]
-> [(Id, CoreExpr)]
loopBreakNodes JoinArity
depth VarSet
weak_fvs [LoopBreakerNode]
nodes [(Id, CoreExpr)]
binds
  = -- pprTrace "loopBreakNodes" (ppr nodes) $
    [SCC LoopBreakerNode] -> [(Id, CoreExpr)]
go ([LoopBreakerNode] -> [SCC LoopBreakerNode]
forall key payload.
Uniquable key =>
[Node key payload] -> [SCC (Node key payload)]
stronglyConnCompFromEdgedVerticesUniqR [LoopBreakerNode]
nodes)
  where
    go :: [SCC LoopBreakerNode] -> [(Id, CoreExpr)]
go []         = [(Id, CoreExpr)]
binds
    go (SCC LoopBreakerNode
scc:[SCC LoopBreakerNode]
sccs) = SCC LoopBreakerNode -> [(Id, CoreExpr)] -> [(Id, CoreExpr)]
loop_break_scc SCC LoopBreakerNode
scc ([SCC LoopBreakerNode] -> [(Id, CoreExpr)]
go [SCC LoopBreakerNode]
sccs)

    loop_break_scc :: SCC LoopBreakerNode -> [(Id, CoreExpr)] -> [(Id, CoreExpr)]
loop_break_scc SCC LoopBreakerNode
scc [(Id, CoreExpr)]
binds
      = case SCC LoopBreakerNode
scc of
          AcyclicSCC LoopBreakerNode
node  -> (Id -> Id) -> LoopBreakerNode -> (Id, CoreExpr)
nodeBinding (VarSet -> Id -> Id
mk_non_loop_breaker VarSet
weak_fvs) LoopBreakerNode
node (Id, CoreExpr) -> [(Id, CoreExpr)] -> [(Id, CoreExpr)]
forall a. a -> [a] -> [a]
: [(Id, CoreExpr)]
binds
          CyclicSCC [LoopBreakerNode]
nodes  -> JoinArity
-> VarSet
-> [LoopBreakerNode]
-> [(Id, CoreExpr)]
-> [(Id, CoreExpr)]
reOrderNodes JoinArity
depth VarSet
weak_fvs [LoopBreakerNode]
nodes [(Id, CoreExpr)]
binds

----------------------------------
reOrderNodes :: Int -> VarSet -> [LoopBreakerNode] -> [Binding] -> [Binding]
    -- Choose a loop breaker, mark it no-inline,
    -- and call loopBreakNodes on the rest
reOrderNodes :: JoinArity
-> VarSet
-> [LoopBreakerNode]
-> [(Id, CoreExpr)]
-> [(Id, CoreExpr)]
reOrderNodes JoinArity
_ VarSet
_ []     [(Id, CoreExpr)]
_     = String -> [(Id, CoreExpr)]
forall a. HasCallStack => String -> a
panic String
"reOrderNodes"
reOrderNodes JoinArity
_ VarSet
_ [LoopBreakerNode
node] [(Id, CoreExpr)]
binds = (Id -> Id) -> LoopBreakerNode -> (Id, CoreExpr)
nodeBinding Id -> Id
mk_loop_breaker LoopBreakerNode
node (Id, CoreExpr) -> [(Id, CoreExpr)] -> [(Id, CoreExpr)]
forall a. a -> [a] -> [a]
: [(Id, CoreExpr)]
binds
reOrderNodes JoinArity
depth VarSet
weak_fvs (LoopBreakerNode
node : [LoopBreakerNode]
nodes) [(Id, CoreExpr)]
binds
  = -- pprTrace "reOrderNodes" (vcat [ text "unchosen" <+> ppr unchosen
    --                               , text "chosen" <+> ppr chosen_nodes ]) $
    JoinArity
-> VarSet
-> [LoopBreakerNode]
-> [(Id, CoreExpr)]
-> [(Id, CoreExpr)]
loopBreakNodes JoinArity
new_depth VarSet
weak_fvs [LoopBreakerNode]
unchosen ([(Id, CoreExpr)] -> [(Id, CoreExpr)])
-> [(Id, CoreExpr)] -> [(Id, CoreExpr)]
forall a b. (a -> b) -> a -> b
$
    ((LoopBreakerNode -> (Id, CoreExpr))
-> [LoopBreakerNode] -> [(Id, CoreExpr)]
forall a b. (a -> b) -> [a] -> [b]
map ((Id -> Id) -> LoopBreakerNode -> (Id, CoreExpr)
nodeBinding Id -> Id
mk_loop_breaker) [LoopBreakerNode]
chosen_nodes [(Id, CoreExpr)] -> [(Id, CoreExpr)] -> [(Id, CoreExpr)]
forall a. [a] -> [a] -> [a]
++ [(Id, CoreExpr)]
binds)
  where
    ([LoopBreakerNode]
chosen_nodes, [LoopBreakerNode]
unchosen) = Bool
-> NodeScore
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> ([LoopBreakerNode], [LoopBreakerNode])
chooseLoopBreaker Bool
approximate_lb
                                                 (SimpleNodeDetails -> NodeScore
snd_score (LoopBreakerNode -> SimpleNodeDetails
forall key payload. Node key payload -> payload
node_payload LoopBreakerNode
node))
                                                 [LoopBreakerNode
node] [] [LoopBreakerNode]
nodes

    approximate_lb :: Bool
approximate_lb = JoinArity
depth JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
>= JoinArity
2
    new_depth :: JoinArity
new_depth | Bool
approximate_lb = JoinArity
0
              | Bool
otherwise      = JoinArity
depthJoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
+JoinArity
1
        -- After two iterations (d=0, d=1) give up
        -- and approximate, returning to d=0

nodeBinding :: (Id -> Id) -> LoopBreakerNode -> Binding
nodeBinding :: (Id -> Id) -> LoopBreakerNode -> (Id, CoreExpr)
nodeBinding Id -> Id
set_id_occ (LoopBreakerNode -> SimpleNodeDetails
forall key payload. Node key payload -> payload
node_payload -> SND { snd_bndr :: SimpleNodeDetails -> Id
snd_bndr = Id
bndr, snd_rhs :: SimpleNodeDetails -> CoreExpr
snd_rhs = CoreExpr
rhs})
  = (Id -> Id
set_id_occ Id
bndr, CoreExpr
rhs)

mk_loop_breaker :: Id -> Id
mk_loop_breaker :: Id -> Id
mk_loop_breaker Id
bndr
  = Id
bndr Id -> OccInfo -> Id
`setIdOccInfo` OccInfo
occ'
  where
    occ' :: OccInfo
occ'      = OccInfo
strongLoopBreaker { occ_tail = tail_info }
    tail_info :: TailCallInfo
tail_info = OccInfo -> TailCallInfo
tailCallInfo (Id -> OccInfo
idOccInfo Id
bndr)

mk_non_loop_breaker :: VarSet -> Id -> Id
-- See Note [Weak loop breakers]
mk_non_loop_breaker :: VarSet -> Id -> Id
mk_non_loop_breaker VarSet
weak_fvs Id
bndr
  | Id
bndr Id -> VarSet -> Bool
`elemVarSet` VarSet
weak_fvs = Id -> OccInfo -> Id
setIdOccInfo Id
bndr OccInfo
occ'
  | Bool
otherwise                  = Id
bndr
  where
    occ' :: OccInfo
occ'      = OccInfo
weakLoopBreaker { occ_tail = tail_info }
    tail_info :: TailCallInfo
tail_info = OccInfo -> TailCallInfo
tailCallInfo (Id -> OccInfo
idOccInfo Id
bndr)

----------------------------------
chooseLoopBreaker :: Bool                -- True <=> Too many iterations,
                                         --          so approximate
                  -> NodeScore           -- Best score so far
                  -> [LoopBreakerNode]   -- Nodes with this score
                  -> [LoopBreakerNode]   -- Nodes with higher scores
                  -> [LoopBreakerNode]   -- Unprocessed nodes
                  -> ([LoopBreakerNode], [LoopBreakerNode])
    -- This loop looks for the bind with the lowest score
    -- to pick as the loop  breaker.  The rest accumulate in
chooseLoopBreaker :: Bool
-> NodeScore
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> ([LoopBreakerNode], [LoopBreakerNode])
chooseLoopBreaker Bool
_ NodeScore
_ [LoopBreakerNode]
loop_nodes [LoopBreakerNode]
acc []
  = ([LoopBreakerNode]
loop_nodes, [LoopBreakerNode]
acc)        -- Done

    -- If approximate_loop_breaker is True, we pick *all*
    -- nodes with lowest score, else just one
    -- See Note [Complexity of loop breaking]
chooseLoopBreaker Bool
approx_lb NodeScore
loop_sc [LoopBreakerNode]
loop_nodes [LoopBreakerNode]
acc (LoopBreakerNode
node : [LoopBreakerNode]
nodes)
  | Bool
approx_lb
  , NodeScore -> JoinArity
rank NodeScore
sc JoinArity -> JoinArity -> Bool
forall a. Eq a => a -> a -> Bool
== NodeScore -> JoinArity
rank NodeScore
loop_sc
  = Bool
-> NodeScore
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> ([LoopBreakerNode], [LoopBreakerNode])
chooseLoopBreaker Bool
approx_lb NodeScore
loop_sc (LoopBreakerNode
node LoopBreakerNode -> [LoopBreakerNode] -> [LoopBreakerNode]
forall a. a -> [a] -> [a]
: [LoopBreakerNode]
loop_nodes) [LoopBreakerNode]
acc [LoopBreakerNode]
nodes

  | NodeScore
sc NodeScore -> NodeScore -> Bool
`betterLB` NodeScore
loop_sc  -- Better score so pick this new one
  = Bool
-> NodeScore
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> ([LoopBreakerNode], [LoopBreakerNode])
chooseLoopBreaker Bool
approx_lb NodeScore
sc [LoopBreakerNode
node] ([LoopBreakerNode]
loop_nodes [LoopBreakerNode] -> [LoopBreakerNode] -> [LoopBreakerNode]
forall a. [a] -> [a] -> [a]
++ [LoopBreakerNode]
acc) [LoopBreakerNode]
nodes

  | Bool
otherwise              -- Worse score so don't pick it
  = Bool
-> NodeScore
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> [LoopBreakerNode]
-> ([LoopBreakerNode], [LoopBreakerNode])
chooseLoopBreaker Bool
approx_lb NodeScore
loop_sc [LoopBreakerNode]
loop_nodes (LoopBreakerNode
node LoopBreakerNode -> [LoopBreakerNode] -> [LoopBreakerNode]
forall a. a -> [a] -> [a]
: [LoopBreakerNode]
acc) [LoopBreakerNode]
nodes
  where
    sc :: NodeScore
sc = SimpleNodeDetails -> NodeScore
snd_score (LoopBreakerNode -> SimpleNodeDetails
forall key payload. Node key payload -> payload
node_payload LoopBreakerNode
node)

{-
Note [Complexity of loop breaking]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The loop-breaking algorithm knocks out one binder at a time, and
performs a new SCC analysis on the remaining binders.  That can
behave very badly in tightly-coupled groups of bindings; in the
worst case it can be (N**2)*log N, because it does a full SCC
on N, then N-1, then N-2 and so on.

To avoid this, we switch plans after 2 (or whatever) attempts:
  Plan A: pick one binder with the lowest score, make it
          a loop breaker, and try again
  Plan B: pick *all* binders with the lowest score, make them
          all loop breakers, and try again
Since there are only a small finite number of scores, this will
terminate in a constant number of iterations, rather than O(N)
iterations.

You might thing that it's very unlikely, but RULES make it much
more likely.  Here's a real example from #1969:
  Rec { $dm = \d.\x. op d
        {-# RULES forall d. $dm Int d  = $s$dm1
                  forall d. $dm Bool d = $s$dm2 #-}

        dInt = MkD .... opInt ...
        dInt = MkD .... opBool ...
        opInt  = $dm dInt
        opBool = $dm dBool

        $s$dm1 = \x. op dInt
        $s$dm2 = \x. op dBool }
The RULES stuff means that we can't choose $dm as a loop breaker
(Note [Choosing loop breakers]), so we must choose at least (say)
opInt *and* opBool, and so on.  The number of loop breakers is
linear in the number of instance declarations.

Note [Loop breakers and INLINE/INLINABLE pragmas]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Avoid choosing a function with an INLINE pramga as the loop breaker!
If such a function is mutually-recursive with a non-INLINE thing,
then the latter should be the loop-breaker.

It's vital to distinguish between INLINE and INLINABLE (the
Bool returned by hasStableCoreUnfolding_maybe).  If we start with
   Rec { {-# INLINABLE f #-}
         f x = ...f... }
and then worker/wrapper it through strictness analysis, we'll get
   Rec { {-# INLINABLE $wf #-}
         $wf p q = let x = (p,q) in ...f...

         {-# INLINE f #-}
         f x = case x of (p,q) -> $wf p q }

Now it is vital that we choose $wf as the loop breaker, so we can
inline 'f' in '$wf'.

Note [DFuns should not be loop breakers]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
It's particularly bad to make a DFun into a loop breaker.  See
Note [How instance declarations are translated] in GHC.Tc.TyCl.Instance

We give DFuns a higher score than ordinary CONLIKE things because
if there's a choice we want the DFun to be the non-loop breaker. Eg

rec { sc = /\ a \$dC. $fBWrap (T a) ($fCT @ a $dC)

      $fCT :: forall a_afE. (Roman.C a_afE) => Roman.C (Roman.T a_afE)
      {-# DFUN #-}
      $fCT = /\a \$dC. MkD (T a) ((sc @ a $dC) |> blah) ($ctoF @ a $dC)
    }

Here 'sc' (the superclass) looks CONLIKE, but we'll never get to it
if we can't unravel the DFun first.

Note [Constructor applications]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
It's really really important to inline dictionaries.  Real
example (the Enum Ordering instance from GHC.Base):

     rec     f = \ x -> case d of (p,q,r) -> p x
             g = \ x -> case d of (p,q,r) -> q x
             d = (v, f, g)

Here, f and g occur just once; but we can't inline them into d.
On the other hand we *could* simplify those case expressions if
we didn't stupidly choose d as the loop breaker.
But we won't because constructor args are marked "Many".
Inlining dictionaries is really essential to unravelling
the loops in static numeric dictionaries, see GHC.Float.

Note [Closure conversion]
~~~~~~~~~~~~~~~~~~~~~~~~~
We treat (\x. C p q) as a high-score candidate in the letrec scoring algorithm.
The immediate motivation came from the result of a closure-conversion transformation
which generated code like this:

    data Clo a b = forall c. Clo (c -> a -> b) c

    ($:) :: Clo a b -> a -> b
    Clo f env $: x = f env x

    rec { plus = Clo plus1 ()

        ; plus1 _ n = Clo plus2 n

        ; plus2 Zero     n = n
        ; plus2 (Succ m) n = Succ (plus $: m $: n) }

If we inline 'plus' and 'plus1', everything unravels nicely.  But if
we choose 'plus1' as the loop breaker (which is entirely possible
otherwise), the loop does not unravel nicely.


@occAnalUnfolding@ deals with the question of bindings where the Id is marked
by an INLINE pragma.  For these we record that anything which occurs
in its RHS occurs many times.  This pessimistically assumes that this
inlined binder also occurs many times in its scope, but if it doesn't
we'll catch it next time round.  At worst this costs an extra simplifier pass.
ToDo: try using the occurrence info for the inline'd binder.

[March 97] We do the same for atomic RHSs.  Reason: see notes with loopBreakSCC.
[June 98, SLPJ]  I've undone this change; I don't understand it.  See notes with loopBreakSCC.


************************************************************************
*                                                                      *
                   Making nodes
*                                                                      *
************************************************************************
-}

-- | Digraph node as constructed by 'makeNode' and consumed by 'occAnalRec'.
-- The Unique key is gotten from the Id.
type LetrecNode = Node Unique NodeDetails

-- | Node details as consumed by 'occAnalRec'.
data NodeDetails
  = ND { NodeDetails -> Id
nd_bndr :: Id          -- Binder

       , NodeDetails -> WithTailUsageDetails CoreExpr
nd_rhs  :: !(WithTailUsageDetails CoreExpr)
         -- ^ RHS, already occ-analysed
         -- With TailUsageDetails from RHS, and RULES, and stable unfoldings,
         -- ignoring phase (ie assuming all are active).
         -- NB: Unadjusted TailUsageDetails, as if this Node becomes a
         -- non-recursive join point!
         -- See Note [TailUsageDetails when forming Rec groups]

       , NodeDetails -> VarSet
nd_inl  :: IdSet       -- Free variables of the stable unfolding and the RHS
                                -- but excluding any RULES
                                -- This is the IdSet that may be used if the Id is inlined

       , NodeDetails -> Bool
nd_simple :: Bool      -- True iff this binding has no local RULES
                                -- If all nodes are simple we don't need a loop-breaker
                                -- dep-anal before reconstructing.

       , NodeDetails -> VarSet
nd_weak_fvs :: IdSet    -- Variables bound in this Rec group that are free
                                 -- in the RHS of any rule (active or not) for this bndr
                                 -- See Note [Weak loop breakers]

       , NodeDetails -> VarSet
nd_active_rule_fvs :: IdSet    -- Variables bound in this Rec group that are free
                                        -- in the RHS of an active rule for this bndr
                                        -- See Note [Rules and loop breakers]
  }

instance Outputable NodeDetails where
   ppr :: NodeDetails -> SDoc
ppr NodeDetails
nd = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"ND" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc
braces
             ([SDoc] -> SDoc
forall doc. IsLine doc => [doc] -> doc
sep [ String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"bndr =" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> Id -> SDoc
forall a. Outputable a => a -> SDoc
ppr (NodeDetails -> Id
nd_bndr NodeDetails
nd)
                  , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"uds =" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> TailUsageDetails -> SDoc
forall a. Outputable a => a -> SDoc
ppr TailUsageDetails
uds
                  , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"inl =" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> VarSet -> SDoc
forall a. Outputable a => a -> SDoc
ppr (NodeDetails -> VarSet
nd_inl NodeDetails
nd)
                  , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"simple =" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> Bool -> SDoc
forall a. Outputable a => a -> SDoc
ppr (NodeDetails -> Bool
nd_simple NodeDetails
nd)
                  , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"active_rule_fvs =" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> VarSet -> SDoc
forall a. Outputable a => a -> SDoc
ppr (NodeDetails -> VarSet
nd_active_rule_fvs NodeDetails
nd)
             ])
            where
               WTUD TailUsageDetails
uds CoreExpr
_ = NodeDetails -> WithTailUsageDetails CoreExpr
nd_rhs NodeDetails
nd

-- | Digraph with simplified and completely occurrence analysed
-- 'SimpleNodeDetails', retaining just the info we need for breaking loops.
type LoopBreakerNode = Node Unique SimpleNodeDetails

-- | Condensed variant of 'NodeDetails' needed during loop breaking.
data SimpleNodeDetails
  = SND { SimpleNodeDetails -> Id
snd_bndr  :: IdWithOccInfo  -- OccInfo accurate
        , SimpleNodeDetails -> CoreExpr
snd_rhs   :: CoreExpr       -- properly occur-analysed
        , SimpleNodeDetails -> NodeScore
snd_score :: NodeScore
        }

instance Outputable SimpleNodeDetails where
   ppr :: SimpleNodeDetails -> SDoc
ppr SimpleNodeDetails
nd = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"SND" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc
braces
             ([SDoc] -> SDoc
forall doc. IsLine doc => [doc] -> doc
sep [ String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"bndr =" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> Id -> SDoc
forall a. Outputable a => a -> SDoc
ppr (SimpleNodeDetails -> Id
snd_bndr SimpleNodeDetails
nd)
                  , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"score =" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> NodeScore -> SDoc
forall a. Outputable a => a -> SDoc
ppr (SimpleNodeDetails -> NodeScore
snd_score SimpleNodeDetails
nd)
             ])

-- The NodeScore is compared lexicographically;
--      e.g. lower rank wins regardless of size
type NodeScore = ( Int     -- Rank: lower => more likely to be picked as loop breaker
                 , Int     -- Size of rhs: higher => more likely to be picked as LB
                           -- Maxes out at maxExprSize; we just use it to prioritise
                           -- small functions
                 , Bool )  -- Was it a loop breaker before?
                           -- True => more likely to be picked
                           -- Note [Loop breakers, node scoring, and stability]

rank :: NodeScore -> Int
rank :: NodeScore -> JoinArity
rank (JoinArity
r, JoinArity
_, Bool
_) = JoinArity
r

makeNode :: OccEnv -> ImpRuleEdges -> VarSet
         -> (Var, CoreExpr) -> LetrecNode
-- See Note [Recursive bindings: the grand plan]
makeNode :: OccEnv
-> ImpRuleEdges
-> VarSet
-> (Id, CoreExpr)
-> Node Unique NodeDetails
makeNode !OccEnv
env ImpRuleEdges
imp_rule_edges VarSet
bndr_set (Id
bndr, CoreExpr
rhs)
  = -- pprTrace "makeNode" (ppr bndr <+> ppr (sizeVarSet bndr_set)) $
    DigraphNode { node_payload :: NodeDetails
node_payload      = NodeDetails
details
                , node_key :: Unique
node_key          = Id -> Unique
varUnique Id
bndr
                , node_dependencies :: [Unique]
node_dependencies = VarSet -> [Unique]
forall elt. UniqSet elt -> [Unique]
nonDetKeysUniqSet VarSet
scope_fvs }
    -- It's OK to use nonDetKeysUniqSet here as stronglyConnCompFromEdgedVerticesR
    -- is still deterministic with edges in nondeterministic order as
    -- explained in Note [Deterministic SCC] in GHC.Data.Graph.Directed.
  where
    details :: NodeDetails
details = ND { nd_bndr :: Id
nd_bndr            = Id
bndr'
                 , nd_rhs :: WithTailUsageDetails CoreExpr
nd_rhs             = TailUsageDetails -> CoreExpr -> WithTailUsageDetails CoreExpr
forall a. TailUsageDetails -> a -> WithTailUsageDetails a
WTUD (JoinArity -> UsageDetails -> TailUsageDetails
TUD JoinArity
rhs_ja UsageDetails
unadj_scope_uds) CoreExpr
rhs'
                 , nd_inl :: VarSet
nd_inl             = VarSet
inl_fvs
                 , nd_simple :: Bool
nd_simple          = [(CoreRule, UsageDetails, UsageDetails)] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [(CoreRule, UsageDetails, UsageDetails)]
rules_w_uds Bool -> Bool -> Bool
&& [(Activation, VarSet)] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [(Activation, VarSet)]
imp_rule_info
                 , nd_weak_fvs :: VarSet
nd_weak_fvs        = VarSet
weak_fvs
                 , nd_active_rule_fvs :: VarSet
nd_active_rule_fvs = VarSet
active_rule_fvs }

    bndr' :: Id
bndr' | OccEnv -> Bool
noBinderSwaps OccEnv
env = Id
bndr  -- See Note [Unfoldings and rules]
          | Bool
otherwise         = Id
bndr Id -> Unfolding -> Id
`setIdUnfolding`      Unfolding
unf'
                                     Id -> RuleInfo -> Id
`setIdSpecialisation` [CoreRule] -> RuleInfo
mkRuleInfo [CoreRule]
rules'

    -- NB: Both adj_unf_uds and adj_rule_uds have been adjusted to match the
    --     JoinArity rhs_ja of unadj_rhs_uds.
    unadj_inl_uds :: UsageDetails
unadj_inl_uds   = UsageDetails
unadj_rhs_uds UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
adj_unf_uds
    unadj_scope_uds :: UsageDetails
unadj_scope_uds = UsageDetails
unadj_inl_uds UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
adj_rule_uds
                   -- Note [Rules are extra RHSs]
                   -- Note [Rule dependency info]
    scope_fvs :: VarSet
scope_fvs = VarSet -> UsageDetails -> VarSet
udFreeVars VarSet
bndr_set UsageDetails
unadj_scope_uds
    -- scope_fvs: all occurrences from this binder: RHS, unfolding,
    --            and RULES, both LHS and RHS thereof, active or inactive

    inl_fvs :: VarSet
inl_fvs  = VarSet -> UsageDetails -> VarSet
udFreeVars VarSet
bndr_set UsageDetails
unadj_inl_uds
    -- inl_fvs: vars that would become free if the function was inlined.
    -- We conservatively approximate that by the free vars from the RHS
    -- and the unfolding together.
    -- See Note [inl_fvs]


    --------- Right hand side ---------
    -- Constructing the edges for the main Rec computation
    -- See Note [Forming Rec groups]
    -- and Note [TailUsageDetails when forming Rec groups]
    -- Compared to occAnalNonRecBind, we can't yet adjust the RHS because
    --   (a) we don't yet know the final joinpointhood. It might not become a
    --       join point after all!
    --   (b) we don't even know whether it stays a recursive RHS after the SCC
    --       analysis we are about to seed! So we can't markAllInsideLam in
    --       advance, because if it ends up as a non-recursive join point we'll
    --       consider it as one-shot and don't need to markAllInsideLam.
    -- Instead, do the occAnalLamTail call here and postpone adjustTailUsage
    -- until occAnalRec. In effect, we pretend that the RHS becomes a
    -- non-recursive join point and fix up later with adjustTailUsage.
    rhs_env :: OccEnv
rhs_env = OccEnv
-> RecFlag -> OccEncl -> JoinPointHood -> Id -> CoreExpr -> OccEnv
mkRhsOccEnv OccEnv
env RecFlag
Recursive OccEncl
OccRhs (Id -> JoinPointHood
idJoinPointHood Id
bndr) Id
bndr CoreExpr
rhs
            -- If bndr isn't an /existing/ join point (so idJoinPointHood = NotJoinPoint),
            -- it's safe to zap the occ_join_points, because they can't occur in RHS.
    WTUD (TUD JoinArity
rhs_ja UsageDetails
unadj_rhs_uds) CoreExpr
rhs' = OccEnv -> CoreExpr -> WithTailUsageDetails CoreExpr
occAnalLamTail OccEnv
rhs_env CoreExpr
rhs
      -- The corresponding call to adjustTailUsage is in occAnalRec and tagRecBinders

    --------- Unfolding ---------
    -- See Note [Join points and unfoldings/rules]
    unf :: Unfolding
unf = IdUnfoldingFun
realIdUnfolding Id
bndr -- realIdUnfolding: Ignore loop-breaker-ness
                               -- here because that is what we are setting!
    WTUD TailUsageDetails
unf_tuds Unfolding
unf' = OccEnv -> Unfolding -> WithTailUsageDetails Unfolding
occAnalUnfolding OccEnv
rhs_env Unfolding
unf
    adj_unf_uds :: UsageDetails
adj_unf_uds = JoinPointHood -> TailUsageDetails -> UsageDetails
adjustTailArity (JoinArity -> JoinPointHood
JoinPoint JoinArity
rhs_ja) TailUsageDetails
unf_tuds
      -- `rhs_ja` is `joinRhsArity rhs` and is the prediction for source M
      -- of Note [Join arity prediction based on joinRhsArity]

    --------- IMP-RULES --------
    is_active :: Activation -> Bool
is_active     = OccEnv -> Activation -> Bool
occ_rule_act OccEnv
env :: Activation -> Bool
    imp_rule_info :: [(Activation, VarSet)]
imp_rule_info = ImpRuleEdges -> Id -> [(Activation, VarSet)]
lookupImpRules ImpRuleEdges
imp_rule_edges Id
bndr
    imp_rule_uds :: UsageDetails
imp_rule_uds  = [(Activation, VarSet)] -> UsageDetails
impRulesScopeUsage [(Activation, VarSet)]
imp_rule_info
    imp_rule_fvs :: VarSet
imp_rule_fvs  = (Activation -> Bool) -> VarSet -> [(Activation, VarSet)] -> VarSet
impRulesActiveFvs Activation -> Bool
is_active VarSet
bndr_set [(Activation, VarSet)]
imp_rule_info

    --------- All rules --------
    -- See Note [Join points and unfoldings/rules]
    -- `rhs_ja` is `joinRhsArity rhs'` and is the prediction for source M
    -- of Note [Join arity prediction based on joinRhsArity]
    rules_w_uds :: [(CoreRule, UsageDetails, UsageDetails)]
    rules_w_uds :: [(CoreRule, UsageDetails, UsageDetails)]
rules_w_uds = [ (CoreRule
r,UsageDetails
l,JoinPointHood -> TailUsageDetails -> UsageDetails
adjustTailArity (JoinArity -> JoinPointHood
JoinPoint JoinArity
rhs_ja) TailUsageDetails
rhs_wuds)
                  | CoreRule
rule <- Id -> [CoreRule]
idCoreRules Id
bndr
                  , let (CoreRule
r,UsageDetails
l,TailUsageDetails
rhs_wuds) = OccEnv -> CoreRule -> (CoreRule, UsageDetails, TailUsageDetails)
occAnalRule OccEnv
rhs_env CoreRule
rule ]
    rules' :: [CoreRule]
rules'      = ((CoreRule, UsageDetails, UsageDetails) -> CoreRule)
-> [(CoreRule, UsageDetails, UsageDetails)] -> [CoreRule]
forall a b. (a -> b) -> [a] -> [b]
map (CoreRule, UsageDetails, UsageDetails) -> CoreRule
forall a b c. (a, b, c) -> a
fstOf3 [(CoreRule, UsageDetails, UsageDetails)]
rules_w_uds

    adj_rule_uds :: UsageDetails
adj_rule_uds = ((CoreRule, UsageDetails, UsageDetails)
 -> UsageDetails -> UsageDetails)
-> UsageDetails
-> [(CoreRule, UsageDetails, UsageDetails)]
-> UsageDetails
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (CoreRule, UsageDetails, UsageDetails)
-> UsageDetails -> UsageDetails
forall {a}.
(a, UsageDetails, UsageDetails) -> UsageDetails -> UsageDetails
add_rule_uds UsageDetails
imp_rule_uds [(CoreRule, UsageDetails, UsageDetails)]
rules_w_uds
    add_rule_uds :: (a, UsageDetails, UsageDetails) -> UsageDetails -> UsageDetails
add_rule_uds (a
_, UsageDetails
l, UsageDetails
r) UsageDetails
uds = UsageDetails
l UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
r UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
uds

    -------- active_rule_fvs ------------
    active_rule_fvs :: VarSet
active_rule_fvs = ((CoreRule, UsageDetails, UsageDetails) -> VarSet -> VarSet)
-> VarSet -> [(CoreRule, UsageDetails, UsageDetails)] -> VarSet
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (CoreRule, UsageDetails, UsageDetails) -> VarSet -> VarSet
add_active_rule VarSet
imp_rule_fvs [(CoreRule, UsageDetails, UsageDetails)]
rules_w_uds
    add_active_rule :: (CoreRule, UsageDetails, UsageDetails) -> VarSet -> VarSet
add_active_rule (CoreRule
rule, UsageDetails
_, UsageDetails
rhs_uds) VarSet
fvs
      | Activation -> Bool
is_active (CoreRule -> Activation
ruleActivation CoreRule
rule)
      = VarSet -> UsageDetails -> VarSet
udFreeVars VarSet
bndr_set UsageDetails
rhs_uds VarSet -> VarSet -> VarSet
`unionVarSet` VarSet
fvs
      | Bool
otherwise
      = VarSet
fvs

    -------- weak_fvs ------------
    -- See Note [Weak loop breakers]
    weak_fvs :: VarSet
weak_fvs = ((CoreRule, UsageDetails, UsageDetails) -> VarSet -> VarSet)
-> VarSet -> [(CoreRule, UsageDetails, UsageDetails)] -> VarSet
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (CoreRule, UsageDetails, UsageDetails) -> VarSet -> VarSet
add_rule VarSet
emptyVarSet [(CoreRule, UsageDetails, UsageDetails)]
rules_w_uds
    add_rule :: (CoreRule, UsageDetails, UsageDetails) -> VarSet -> VarSet
add_rule (CoreRule
_, UsageDetails
_, UsageDetails
rhs_uds) VarSet
fvs = VarSet -> UsageDetails -> VarSet
udFreeVars VarSet
bndr_set UsageDetails
rhs_uds VarSet -> VarSet -> VarSet
`unionVarSet` VarSet
fvs

mkLoopBreakerNodes :: OccEnv -> TopLevelFlag
                   -> UsageDetails   -- for BODY of let
                   -> [NodeDetails]
                   -> WithUsageDetails [LoopBreakerNode] -- with OccInfo up-to-date
-- See Note [Choosing loop breakers]
-- This function primarily creates the Nodes for the
-- loop-breaker SCC analysis.  More specifically:
--   a) tag each binder with its occurrence info
--   b) add a NodeScore to each node
--   c) make a Node with the right dependency edges for
--      the loop-breaker SCC analysis
--   d) adjust each RHS's usage details according to
--      the binder's (new) shotness and join-point-hood
mkLoopBreakerNodes :: OccEnv
-> TopLevelFlag
-> UsageDetails
-> [NodeDetails]
-> WithUsageDetails [LoopBreakerNode]
mkLoopBreakerNodes !OccEnv
env TopLevelFlag
lvl UsageDetails
body_uds [NodeDetails]
details_s
  = UsageDetails
-> [LoopBreakerNode] -> WithUsageDetails [LoopBreakerNode]
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
final_uds (String
-> (NodeDetails -> Id -> LoopBreakerNode)
-> [NodeDetails]
-> [Id]
-> [LoopBreakerNode]
forall a b c.
HasDebugCallStack =>
String -> (a -> b -> c) -> [a] -> [b] -> [c]
zipWithEqual String
"mkLoopBreakerNodes" NodeDetails -> Id -> LoopBreakerNode
mk_lb_node [NodeDetails]
details_s [Id]
bndrs')
  where
    WUD UsageDetails
final_uds [Id]
bndrs' = TopLevelFlag
-> UsageDetails -> [NodeDetails] -> WithUsageDetails [Id]
tagRecBinders TopLevelFlag
lvl UsageDetails
body_uds [NodeDetails]
details_s

    mk_lb_node :: NodeDetails -> Id -> LoopBreakerNode
mk_lb_node nd :: NodeDetails
nd@(ND { nd_bndr :: NodeDetails -> Id
nd_bndr = Id
old_bndr, nd_inl :: NodeDetails -> VarSet
nd_inl = VarSet
inl_fvs
                      , nd_rhs :: NodeDetails -> WithTailUsageDetails CoreExpr
nd_rhs = WTUD TailUsageDetails
_ CoreExpr
rhs }) Id
new_bndr
      = DigraphNode { node_payload :: SimpleNodeDetails
node_payload      = SimpleNodeDetails
simple_nd
                    , node_key :: Unique
node_key          = Id -> Unique
varUnique Id
old_bndr
                    , node_dependencies :: [Unique]
node_dependencies = VarSet -> [Unique]
forall elt. UniqSet elt -> [Unique]
nonDetKeysUniqSet VarSet
lb_deps }
              -- It's OK to use nonDetKeysUniqSet here as
              -- stronglyConnCompFromEdgedVerticesR is still deterministic with edges
              -- in nondeterministic order as explained in
              -- Note [Deterministic SCC] in GHC.Data.Graph.Directed.
      where
        simple_nd :: SimpleNodeDetails
simple_nd = SND { snd_bndr :: Id
snd_bndr = Id
new_bndr, snd_rhs :: CoreExpr
snd_rhs = CoreExpr
rhs, snd_score :: NodeScore
snd_score = NodeScore
score }
        score :: NodeScore
score  = OccEnv -> Id -> VarSet -> NodeDetails -> NodeScore
nodeScore OccEnv
env Id
new_bndr VarSet
lb_deps NodeDetails
nd
        lb_deps :: VarSet
lb_deps = VarEnv VarSet -> VarSet -> VarSet
extendFvs_ VarEnv VarSet
rule_fv_env VarSet
inl_fvs
        -- See Note [Loop breaker dependencies]

    rule_fv_env :: IdEnv IdSet
    -- Maps a variable f to the variables from this group
    --      reachable by a sequence of RULES starting with f
    -- Domain is *subset* of bound vars (others have no rule fvs)
    -- See Note [Finding rule RHS free vars]
    -- Why transClosureFV?  See Note [Loop breaker dependencies]
    rule_fv_env :: VarEnv VarSet
rule_fv_env = VarEnv VarSet -> VarEnv VarSet
transClosureFV (VarEnv VarSet -> VarEnv VarSet) -> VarEnv VarSet -> VarEnv VarSet
forall a b. (a -> b) -> a -> b
$ [(Id, VarSet)] -> VarEnv VarSet
forall a. [(Id, a)] -> VarEnv a
mkVarEnv ([(Id, VarSet)] -> VarEnv VarSet)
-> [(Id, VarSet)] -> VarEnv VarSet
forall a b. (a -> b) -> a -> b
$
                  [ (Id
b, VarSet
rule_fvs)
                  | ND { nd_bndr :: NodeDetails -> Id
nd_bndr = Id
b, nd_active_rule_fvs :: NodeDetails -> VarSet
nd_active_rule_fvs = VarSet
rule_fvs } <- [NodeDetails]
details_s
                  , Bool -> Bool
not (VarSet -> Bool
isEmptyVarSet VarSet
rule_fvs) ]

{- Note [Loop breaker dependencies]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The loop breaker dependencies of x in a recursive
group { f1 = e1; ...; fn = en } are:

- The "inline free variables" of f: the fi free in
  f's stable unfolding and RHS; see Note [inl_fvs]

- Any fi reachable from those inline free variables by a sequence
  of RULE rewrites.  Remember, rule rewriting is not affected
  by fi being a loop breaker, so we have to take the transitive
  closure in case f is the only possible loop breaker in the loop.

  Hence rule_fv_env.  We need only account for /active/ rules.
-}

------------------------------------------
nodeScore :: OccEnv
          -> Id        -- Binder with new occ-info
          -> VarSet    -- Loop-breaker dependencies
          -> NodeDetails
          -> NodeScore
nodeScore :: OccEnv -> Id -> VarSet -> NodeDetails -> NodeScore
nodeScore !OccEnv
env Id
new_bndr VarSet
lb_deps
          (ND { nd_bndr :: NodeDetails -> Id
nd_bndr = Id
old_bndr, nd_rhs :: NodeDetails -> WithTailUsageDetails CoreExpr
nd_rhs = WTUD TailUsageDetails
_ CoreExpr
bind_rhs })

  | Bool -> Bool
not (Id -> Bool
isId Id
old_bndr)     -- A type or coercion variable is never a loop breaker
  = (JoinArity
100, JoinArity
0, Bool
False)

  | Id
old_bndr Id -> VarSet -> Bool
`elemVarSet` VarSet
lb_deps  -- Self-recursive things are great loop breakers
  = (JoinArity
0, JoinArity
0, Bool
True)                   -- See Note [Self-recursion and loop breakers]

  | Bool -> Bool
not (OccEnv -> Id -> Bool
occ_unf_act OccEnv
env Id
old_bndr) -- A binder whose inlining is inactive (e.g. has
  = (JoinArity
0, JoinArity
0, Bool
True)                   -- a NOINLINE pragma) makes a great loop breaker

  | CoreExpr -> Bool
exprIsTrivial CoreExpr
rhs
  = JoinArity -> NodeScore
mk_score JoinArity
10  -- Practically certain to be inlined
    -- Used to have also: && not (isExportedId bndr)
    -- But I found this sometimes cost an extra iteration when we have
    --      rec { d = (a,b); a = ...df...; b = ...df...; df = d }
    -- where df is the exported dictionary. Then df makes a really
    -- bad choice for loop breaker

  | DFunUnfolding { df_args :: Unfolding -> [CoreExpr]
df_args = [CoreExpr]
args } <- Unfolding
old_unf
    -- Never choose a DFun as a loop breaker
    -- Note [DFuns should not be loop breakers]
  = (JoinArity
9, [CoreExpr] -> JoinArity
forall a. [a] -> JoinArity
forall (t :: * -> *) a. Foldable t => t a -> JoinArity
length [CoreExpr]
args, Bool
is_lb)

    -- Data structures are more important than INLINE pragmas
    -- so that dictionary/method recursion unravels

  | CoreUnfolding { uf_guidance :: Unfolding -> UnfoldingGuidance
uf_guidance = UnfWhen {} } <- Unfolding
old_unf
  = JoinArity -> NodeScore
mk_score JoinArity
6

  | CoreExpr -> Bool
forall {b}. Expr b -> Bool
is_con_app CoreExpr
rhs   -- Data types help with cases:
  = JoinArity -> NodeScore
mk_score JoinArity
5       -- Note [Constructor applications]

  | Unfolding -> Bool
isStableUnfolding Unfolding
old_unf
  , Bool
can_unfold
  = JoinArity -> NodeScore
mk_score JoinArity
3

  | OccInfo -> Bool
isOneOcc (Id -> OccInfo
idOccInfo Id
new_bndr)
  = JoinArity -> NodeScore
mk_score JoinArity
2  -- Likely to be inlined

  | Bool
can_unfold  -- The Id has some kind of unfolding
  = JoinArity -> NodeScore
mk_score JoinArity
1

  | Bool
otherwise
  = (JoinArity
0, JoinArity
0, Bool
is_lb)

  where
    mk_score :: Int -> NodeScore
    mk_score :: JoinArity -> NodeScore
mk_score JoinArity
rank = (JoinArity
rank, JoinArity
rhs_size, Bool
is_lb)

    -- is_lb: see Note [Loop breakers, node scoring, and stability]
    is_lb :: Bool
is_lb = OccInfo -> Bool
isStrongLoopBreaker (Id -> OccInfo
idOccInfo Id
old_bndr)

    old_unf :: Unfolding
old_unf = IdUnfoldingFun
realIdUnfolding Id
old_bndr
    can_unfold :: Bool
can_unfold = Unfolding -> Bool
canUnfold Unfolding
old_unf
    rhs :: CoreExpr
rhs        = case Unfolding
old_unf of
                   CoreUnfolding { uf_src :: Unfolding -> UnfoldingSource
uf_src = UnfoldingSource
src, uf_tmpl :: Unfolding -> CoreExpr
uf_tmpl = CoreExpr
unf_rhs }
                     | UnfoldingSource -> Bool
isStableSource UnfoldingSource
src
                     -> CoreExpr
unf_rhs
                   Unfolding
_ -> CoreExpr
bind_rhs
       -- 'bind_rhs' is irrelevant for inlining things with a stable unfolding
    rhs_size :: JoinArity
rhs_size = case Unfolding
old_unf of
                 CoreUnfolding { uf_guidance :: Unfolding -> UnfoldingGuidance
uf_guidance = UnfoldingGuidance
guidance }
                    | UnfIfGoodArgs { ug_size :: UnfoldingGuidance -> JoinArity
ug_size = JoinArity
size } <- UnfoldingGuidance
guidance
                    -> JoinArity
size
                 Unfolding
_  -> CoreExpr -> JoinArity
cheapExprSize CoreExpr
rhs


        -- Checking for a constructor application
        -- Cheap and cheerful; the simplifier moves casts out of the way
        -- The lambda case is important to spot x = /\a. C (f a)
        -- which comes up when C is a dictionary constructor and
        -- f is a default method.
        -- Example: the instance for Show (ST s a) in GHC.ST
        --
        -- However we *also* treat (\x. C p q) as a con-app-like thing,
        --      Note [Closure conversion]
    is_con_app :: Expr b -> Bool
is_con_app (Var Id
v)    = Id -> Bool
isConLikeId Id
v
    is_con_app (App Expr b
f Expr b
_)  = Expr b -> Bool
is_con_app Expr b
f
    is_con_app (Lam b
_ Expr b
e)  = Expr b -> Bool
is_con_app Expr b
e
    is_con_app (Tick CoreTickish
_ Expr b
e) = Expr b -> Bool
is_con_app Expr b
e
    is_con_app (Let Bind b
_ Expr b
e)  = Expr b -> Bool
is_con_app Expr b
e  -- let x = let y = blah in (a,b)
    is_con_app Expr b
_          = Bool
False         -- We will float the y out, so treat
                                          -- the x-binding as a con-app (#20941)

maxExprSize :: Int
maxExprSize :: JoinArity
maxExprSize = JoinArity
20  -- Rather arbitrary

cheapExprSize :: CoreExpr -> Int
-- Maxes out at maxExprSize
cheapExprSize :: CoreExpr -> JoinArity
cheapExprSize CoreExpr
e
  = JoinArity -> CoreExpr -> JoinArity
go JoinArity
0 CoreExpr
e
  where
    go :: JoinArity -> CoreExpr -> JoinArity
go JoinArity
n CoreExpr
e | JoinArity
n JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
>= JoinArity
maxExprSize = JoinArity
n
           | Bool
otherwise        = JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
e

    go1 :: JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n (Var {})        = JoinArity
nJoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
+JoinArity
1
    go1 JoinArity
n (Lit {})        = JoinArity
nJoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
+JoinArity
1
    go1 JoinArity
n (Type {})       = JoinArity
n
    go1 JoinArity
n (Coercion {})   = JoinArity
n
    go1 JoinArity
n (Tick CoreTickish
_ CoreExpr
e)      = JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
e
    go1 JoinArity
n (Cast CoreExpr
e CoercionR
_)      = JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
e
    go1 JoinArity
n (App CoreExpr
f CoreExpr
a)       = JoinArity -> CoreExpr -> JoinArity
go (JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
f) CoreExpr
a
    go1 JoinArity
n (Lam Id
b CoreExpr
e)
      | Id -> Bool
isTyVar Id
b         = JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
e
      | Bool
otherwise         = JoinArity -> CoreExpr -> JoinArity
go (JoinArity
nJoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
+JoinArity
1) CoreExpr
e
    go1 JoinArity
n (Let CoreBind
b CoreExpr
e)       = JoinArity -> [CoreExpr] -> JoinArity
gos (JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
e) (CoreBind -> [CoreExpr]
forall b. Bind b -> [Expr b]
rhssOfBind CoreBind
b)
    go1 JoinArity
n (Case CoreExpr
e Id
_ Type
_ [Alt Id]
as) = JoinArity -> [CoreExpr] -> JoinArity
gos (JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
e) ([Alt Id] -> [CoreExpr]
forall b. [Alt b] -> [Expr b]
rhssOfAlts [Alt Id]
as)

    gos :: JoinArity -> [CoreExpr] -> JoinArity
gos JoinArity
n [] = JoinArity
n
    gos JoinArity
n (CoreExpr
e:[CoreExpr]
es) | JoinArity
n JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
>= JoinArity
maxExprSize = JoinArity
n
                 | Bool
otherwise        = JoinArity -> [CoreExpr] -> JoinArity
gos (JoinArity -> CoreExpr -> JoinArity
go1 JoinArity
n CoreExpr
e) [CoreExpr]
es

betterLB :: NodeScore -> NodeScore -> Bool
-- If  n1 `betterLB` n2  then choose n1 as the loop breaker
betterLB :: NodeScore -> NodeScore -> Bool
betterLB (JoinArity
rank1, JoinArity
size1, Bool
lb1) (JoinArity
rank2, JoinArity
size2, Bool
_)
  | JoinArity
rank1 JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
< JoinArity
rank2 = Bool
True
  | JoinArity
rank1 JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
> JoinArity
rank2 = Bool
False
  | JoinArity
size1 JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
< JoinArity
size2 = Bool
False   -- Make the bigger n2 into the loop breaker
  | JoinArity
size1 JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
> JoinArity
size2 = Bool
True
  | Bool
lb1           = Bool
True    -- Tie-break: if n1 was a loop breaker before, choose it
  | Bool
otherwise     = Bool
False   -- See Note [Loop breakers, node scoring, and stability]

{- Note [Self-recursion and loop breakers]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If we have
   rec { f = ...f...g...
       ; g = .....f...   }
then 'f' has to be a loop breaker anyway, so we may as well choose it
right away, so that g can inline freely.

This is really just a cheap hack. Consider
   rec { f = ...g...
       ; g = ..f..h...
      ;  h = ...f....}
Here f or g are better loop breakers than h; but we might accidentally
choose h.  Finding the minimal set of loop breakers is hard.

Note [Loop breakers, node scoring, and stability]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
To choose a loop breaker, we give a NodeScore to each node in the SCC,
and pick the one with the best score (according to 'betterLB').

We need to be jolly careful (#12425, #12234) about the stability
of this choice. Suppose we have

    let rec { f = ...g...g...
            ; g = ...f...f... }
    in
    case x of
      True  -> ...f..
      False -> ..f...

In each iteration of the simplifier the occurrence analyser OccAnal
chooses a loop breaker. Suppose in iteration 1 it choose g as the loop
breaker. That means it is free to inline f.

Suppose that GHC decides to inline f in the branches of the case, but
(for some reason; eg it is not saturated) in the rhs of g. So we get

    let rec { f = ...g...g...
            ; g = ...f...f... }
    in
    case x of
      True  -> ...g...g.....
      False -> ..g..g....

Now suppose that, for some reason, in the next iteration the occurrence
analyser chooses f as the loop breaker, so it can freely inline g. And
again for some reason the simplifier inlines g at its calls in the case
branches, but not in the RHS of f. Then we get

    let rec { f = ...g...g...
            ; g = ...f...f... }
    in
    case x of
      True  -> ...(...f...f...)...(...f..f..).....
      False -> ..(...f...f...)...(..f..f...)....

You can see where this is going! Each iteration of the simplifier
doubles the number of calls to f or g. No wonder GHC is slow!

(In the particular example in comment:3 of #12425, f and g are the two
mutually recursive fmap instances for CondT and Result. They are both
marked INLINE which, oddly, is why they don't inline in each other's
RHS, because the call there is not saturated.)

The root cause is that we flip-flop on our choice of loop breaker. I
always thought it didn't matter, and indeed for any single iteration
to terminate, it doesn't matter. But when we iterate, it matters a
lot!!

So The Plan is this:
   If there is a tie, choose the node that
   was a loop breaker last time round

Hence the is_lb field of NodeScore
-}

{- *********************************************************************
*                                                                      *
                  Lambda groups
*                                                                      *
********************************************************************* -}

{- Note [Occurrence analysis for lambda binders]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For value lambdas we do a special hack.  Consider
     (\x. \y. ...x...)
If we did nothing, x is used inside the \y, so would be marked
as dangerous to dup.  But in the common case where the abstraction
is applied to two arguments this is over-pessimistic, which delays
inlining x, which forces more simplifier iterations.

So the occurrence analyser collaborates with the simplifier to treat
a /lambda-group/ specially.   A lambda-group is a contiguous run of
lambda and casts, e.g.
    Lam x (Lam y (Cast (Lam z body) co))

* Occurrence analyser: we just mark each binder in the lambda-group
  (here: x,y,z) with its occurrence info in the *body* of the
  lambda-group.  See occAnalLamTail.

* Simplifier.  The simplifier is careful when partially applying
  lambda-groups. See the call to zapLambdaBndrs in
     GHC.Core.Opt.Simplify.simplExprF1
     GHC.Core.SimpleOpt.simple_app

* Why do we take care to account for intervening casts? Answer:
  currently we don't do eta-expansion and cast-swizzling in a stable
  unfolding (see Historical-note [Eta-expansion in stable unfoldings]).
  So we can get
    f = \x. ((\y. ...x...y...) |> co)
  Now, since the lambdas aren't together, the occurrence analyser will
  say that x is OnceInLam.  Now if we have a call
    (f e1 |> co) e2
  we'll end up with
    let x = e1 in ...x..e2...
  and it'll take an extra iteration of the Simplifier to substitute for x.

A thought: a lambda-group is pretty much what GHC.Core.Opt.Arity.manifestArity
recognises except that the latter looks through (some) ticks.  Maybe a lambda
group should also look through (some) ticks?
-}

isOneShotFun :: CoreExpr -> Bool
-- The top level lambdas, ignoring casts, of the expression
-- are all one-shot.  If there aren't any lambdas at all, this is True
isOneShotFun :: CoreExpr -> Bool
isOneShotFun (Lam Id
b CoreExpr
e)  = Id -> Bool
isOneShotBndr Id
b Bool -> Bool -> Bool
&& CoreExpr -> Bool
isOneShotFun CoreExpr
e
isOneShotFun (Cast CoreExpr
e CoercionR
_) = CoreExpr -> Bool
isOneShotFun CoreExpr
e
isOneShotFun CoreExpr
_          = Bool
True

zapLambdaBndrs :: CoreExpr -> FullArgCount -> CoreExpr
-- If (\xyz. t) appears under-applied to only two arguments,
-- we must zap the occ-info on x,y, because they appear under the \z
-- See Note [Occurrence analysis for lambda binders] in GHc.Core.Opt.OccurAnal
--
-- NB: `arg_count` includes both type and value args
zapLambdaBndrs :: CoreExpr -> JoinArity -> CoreExpr
zapLambdaBndrs CoreExpr
fun JoinArity
arg_count
  = -- If the lambda is fully applied, leave it alone; if not
    -- zap the OccInfo on the lambdas that do have arguments,
    -- so they beta-reduce to use-many Lets rather than used-once ones.
    JoinArity -> CoreExpr -> Maybe CoreExpr
zap JoinArity
arg_count CoreExpr
fun Maybe CoreExpr -> CoreExpr -> CoreExpr
forall a. Maybe a -> a -> a
`orElse` CoreExpr
fun
  where
    zap :: FullArgCount -> CoreExpr -> Maybe CoreExpr
    -- Nothing => No need to change the occ-info
    -- Just e  => Had to change
    zap :: JoinArity -> CoreExpr -> Maybe CoreExpr
zap JoinArity
0 CoreExpr
e | CoreExpr -> Bool
isOneShotFun CoreExpr
e = Maybe CoreExpr
forall a. Maybe a
Nothing  -- All remaining lambdas are one-shot
            | Bool
otherwise      = CoreExpr -> Maybe CoreExpr
forall a. a -> Maybe a
Just CoreExpr
e   -- in which case no need to zap
    zap JoinArity
n (Cast CoreExpr
e CoercionR
co) = do { e' <- JoinArity -> CoreExpr -> Maybe CoreExpr
zap JoinArity
n CoreExpr
e; return (Cast e' co) }
    zap JoinArity
n (Lam Id
b CoreExpr
e)   = do { e' <- JoinArity -> CoreExpr -> Maybe CoreExpr
zap (JoinArity
nJoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
-JoinArity
1) CoreExpr
e
                           ; return (Lam (zap_bndr b) e') }
    zap JoinArity
_ CoreExpr
_           = Maybe CoreExpr
forall a. Maybe a
Nothing  -- More arguments than lambdas

    zap_bndr :: Id -> Id
zap_bndr Id
b | Id -> Bool
isTyVar Id
b = Id
b
               | Bool
otherwise = Id -> Id
zapLamIdInfo Id
b

occAnalLamTail :: OccEnv -> CoreExpr -> WithTailUsageDetails CoreExpr
-- ^ See Note [Occurrence analysis for lambda binders].
-- It does the following:
--   * Sets one-shot info on the lambda binder from the OccEnv, and
--     removes that one-shot info from the OccEnv
--   * Sets the OccEnv to OccVanilla when going under a value lambda
--   * Tags each lambda with its occurrence information
--   * Walks through casts
--   * Package up the analysed lambda with its manifest join arity
--
-- This function does /not/ do
--   markAllInsideLam or
--   markAllNonTail
-- The caller does that, via adjustTailUsage (mostly calls go through
-- adjustNonRecRhs). Every call to occAnalLamTail must ultimately call
-- adjustTailUsage to discharge the assumed join arity.
--
-- In effect, the analysis result is for a non-recursive join point with
-- manifest arity and adjustTailUsage does the fixup.
-- See Note [Adjusting right-hand sides]
occAnalLamTail :: OccEnv -> CoreExpr -> WithTailUsageDetails CoreExpr
occAnalLamTail OccEnv
env CoreExpr
expr
  = let !(WUD UsageDetails
usage CoreExpr
expr') = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occ_anal_lam_tail OccEnv
env CoreExpr
expr
    in TailUsageDetails -> CoreExpr -> WithTailUsageDetails CoreExpr
forall a. TailUsageDetails -> a -> WithTailUsageDetails a
WTUD (JoinArity -> UsageDetails -> TailUsageDetails
TUD (CoreExpr -> JoinArity
joinRhsArity CoreExpr
expr) UsageDetails
usage) CoreExpr
expr'

occ_anal_lam_tail :: OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
-- Does not markInsideLam etc for the outmost batch of lambdas
occ_anal_lam_tail :: OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occ_anal_lam_tail OccEnv
env expr :: CoreExpr
expr@(Lam {})
  = OccEnv -> [Id] -> CoreExpr -> WithUsageDetails CoreExpr
go OccEnv
env [] CoreExpr
expr
  where
    go :: OccEnv -> [Var] -> CoreExpr -> WithUsageDetails CoreExpr
    go :: OccEnv -> [Id] -> CoreExpr -> WithUsageDetails CoreExpr
go OccEnv
env [Id]
rev_bndrs (Lam Id
bndr CoreExpr
body)
      | Id -> Bool
isTyVar Id
bndr
      = OccEnv -> [Id] -> CoreExpr -> WithUsageDetails CoreExpr
go OccEnv
env (Id
bndrId -> [Id] -> [Id]
forall a. a -> [a] -> [a]
:[Id]
rev_bndrs) CoreExpr
body
              -- Important: Unlike a value binder, do not modify occ_encl
              -- to OccVanilla, so that with a RHS like
              --   \(@ x) -> K @x (f @x)
              -- we'll see that (K @x (f @x)) is in a OccRhs, and hence refrain
              -- from inlining f. See the beginning of Note [Cascading inlines].

      | Bool
otherwise
      = let ([OneShotInfo]
env_one_shots', Id
bndr')
              = case OccEnv -> [OneShotInfo]
occ_one_shots OccEnv
env of
                  []         -> ([],  Id
bndr)
                  (OneShotInfo
os : [OneShotInfo]
oss) -> ([OneShotInfo]
oss, Id -> OneShotInfo -> Id
updOneShotInfo Id
bndr OneShotInfo
os)
                  -- Use updOneShotInfo, not setOneShotInfo, as pre-existing
                  -- one-shot info might be better than what we can infer, e.g.
                  -- due to explicit use of the magic 'oneShot' function.
                  -- See Note [oneShot magic]
            env' :: OccEnv
env' = OccEnv
env { occ_encl = OccVanilla, occ_one_shots = env_one_shots' }
         in OccEnv -> [Id] -> CoreExpr -> WithUsageDetails CoreExpr
go OccEnv
env' (Id
bndr'Id -> [Id] -> [Id]
forall a. a -> [a] -> [a]
:[Id]
rev_bndrs) CoreExpr
body

    go OccEnv
env [Id]
rev_bndrs CoreExpr
body
      = OccEnv
-> [Id]
-> (OccEnv -> WithUsageDetails CoreExpr)
-> WithUsageDetails CoreExpr
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScope OccEnv
env [Id]
rev_bndrs ((OccEnv -> WithUsageDetails CoreExpr)
 -> WithUsageDetails CoreExpr)
-> (OccEnv -> WithUsageDetails CoreExpr)
-> WithUsageDetails CoreExpr
forall a b. (a -> b) -> a -> b
$ \OccEnv
env ->
        let !(WUD UsageDetails
usage CoreExpr
body') = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occ_anal_lam_tail OccEnv
env CoreExpr
body
            wrap_lam :: CoreExpr -> Id -> CoreExpr
wrap_lam CoreExpr
body Id
bndr = Id -> CoreExpr -> CoreExpr
forall b. b -> Expr b -> Expr b
Lam (UsageDetails -> Id -> Id
tagLamBinder UsageDetails
usage Id
bndr) CoreExpr
body
        in UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails
usage UsageDetails -> [Id] -> UsageDetails
`addLamCoVarOccs` [Id]
rev_bndrs)
               ((CoreExpr -> Id -> CoreExpr) -> CoreExpr -> [Id] -> CoreExpr
forall b a. (b -> a -> b) -> b -> [a] -> b
forall (t :: * -> *) b a.
Foldable t =>
(b -> a -> b) -> b -> t a -> b
foldl' CoreExpr -> Id -> CoreExpr
wrap_lam CoreExpr
body' [Id]
rev_bndrs)

-- For casts, keep going in the same lambda-group
-- See Note [Occurrence analysis for lambda binders]
occ_anal_lam_tail OccEnv
env (Cast CoreExpr
expr CoercionR
co)
  = let  WUD UsageDetails
usage CoreExpr
expr' = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occ_anal_lam_tail OccEnv
env CoreExpr
expr
         -- usage1: see Note [Gather occurrences of coercion variables]
         usage1 :: UsageDetails
usage1 = UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
usage (CoercionR -> VarSet
coVarsOfCo CoercionR
co)

         -- usage2: see Note [Occ-anal and cast worker/wrapper]
         usage2 :: UsageDetails
usage2 = case CoreExpr
expr of
                    Var {} | OccEnv -> Bool
isRhsEnv OccEnv
env -> UsageDetails -> UsageDetails
markAllMany UsageDetails
usage1
                    CoreExpr
_ -> UsageDetails
usage1

         -- usage3: you might think this was not necessary, because of
         -- the markAllNonTail in adjustTailUsage; but not so!  For a
         -- join point, adjustTailUsage doesn't do this; yet if there is
         -- a cast, we must!  Also: why markAllNonTail?  See
         -- GHC.Core.Lint: Note Note [Join points and casts]
         usage3 :: UsageDetails
usage3 = UsageDetails -> UsageDetails
markAllNonTail UsageDetails
usage2

    in UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
usage3 (CoreExpr -> CoercionR -> CoreExpr
forall b. Expr b -> CoercionR -> Expr b
Cast CoreExpr
expr' CoercionR
co)

occ_anal_lam_tail OccEnv
env CoreExpr
expr  -- Not Lam, not Cast
  = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
env CoreExpr
expr

{- Note [Occ-anal and cast worker/wrapper]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider   y = e; x = y |> co
If we mark y as used-once, we'll inline y into x, and the Cast
worker/wrapper transform will float it straight back out again.  See
Note [Cast worker/wrapper] in GHC.Core.Opt.Simplify.

So in this particular case we want to mark 'y' as Many.  It's very
ad-hoc, but it's also simple.  It's also what would happen if we gave
the binding for x a stable unfolding (as we usually do for wrappers, thus
      y = e
      {-# INLINE x #-}
      x = y |> co
Now y appears twice -- once in x's stable unfolding, and once in x's
RHS. So it'll get a Many occ-info.  (Maybe Cast w/w should create a stable
unfolding, which would obviate this Note; but that seems a bit of a
heavyweight solution.)

We only need to this in occAnalLamTail, not occAnal, because the top leve
of a right hand side is handled by occAnalLamTail.
-}


{- *********************************************************************
*                                                                      *
                   Right hand sides
*                                                                      *
********************************************************************* -}

occAnalUnfolding :: OccEnv
                 -> Unfolding
                 -> WithTailUsageDetails Unfolding
-- Occurrence-analyse a stable unfolding;
-- discard a non-stable one altogether and return empty usage details.
occAnalUnfolding :: OccEnv -> Unfolding -> WithTailUsageDetails Unfolding
occAnalUnfolding !OccEnv
env Unfolding
unf
  = case Unfolding
unf of
      unf :: Unfolding
unf@(CoreUnfolding { uf_tmpl :: Unfolding -> CoreExpr
uf_tmpl = CoreExpr
rhs, uf_src :: Unfolding -> UnfoldingSource
uf_src = UnfoldingSource
src })
        | UnfoldingSource -> Bool
isStableSource UnfoldingSource
src ->
            let
              WTUD (TUD JoinArity
rhs_ja UsageDetails
uds) CoreExpr
rhs' = OccEnv -> CoreExpr -> WithTailUsageDetails CoreExpr
occAnalLamTail OccEnv
env CoreExpr
rhs
              unf' :: Unfolding
unf' = Unfolding
unf { uf_tmpl = rhs' }
            in TailUsageDetails -> Unfolding -> WithTailUsageDetails Unfolding
forall a. TailUsageDetails -> a -> WithTailUsageDetails a
WTUD (JoinArity -> UsageDetails -> TailUsageDetails
TUD JoinArity
rhs_ja (UsageDetails -> UsageDetails
markAllMany UsageDetails
uds)) Unfolding
unf'
              -- markAllMany: see Note [Occurrences in stable unfoldings]

        | Bool
otherwise -> TailUsageDetails -> Unfolding -> WithTailUsageDetails Unfolding
forall a. TailUsageDetails -> a -> WithTailUsageDetails a
WTUD (JoinArity -> UsageDetails -> TailUsageDetails
TUD JoinArity
0 UsageDetails
emptyDetails) Unfolding
unf
              -- For non-Stable unfoldings we leave them undisturbed, but
              -- don't count their usage because the simplifier will discard them.
              -- We leave them undisturbed because nodeScore uses their size info
              -- to guide its decisions.  It's ok to leave un-substituted
              -- expressions in the tree because all the variables that were in
              -- scope remain in scope; there is no cloning etc.

      unf :: Unfolding
unf@(DFunUnfolding { df_bndrs :: Unfolding -> [Id]
df_bndrs = [Id]
bndrs, df_args :: Unfolding -> [CoreExpr]
df_args = [CoreExpr]
args })
        -> let WUD UsageDetails
uds [CoreExpr]
args' = OccEnv
-> [Id]
-> (OccEnv -> WithUsageDetails [CoreExpr])
-> WithUsageDetails [CoreExpr]
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeList OccEnv
env [Id]
bndrs ((OccEnv -> WithUsageDetails [CoreExpr])
 -> WithUsageDetails [CoreExpr])
-> (OccEnv -> WithUsageDetails [CoreExpr])
-> WithUsageDetails [CoreExpr]
forall a b. (a -> b) -> a -> b
$ \ OccEnv
env ->
                               OccEnv -> [CoreExpr] -> WithUsageDetails [CoreExpr]
occAnalList OccEnv
env [CoreExpr]
args
           in TailUsageDetails -> Unfolding -> WithTailUsageDetails Unfolding
forall a. TailUsageDetails -> a -> WithTailUsageDetails a
WTUD (JoinArity -> UsageDetails -> TailUsageDetails
TUD JoinArity
0 UsageDetails
uds) (Unfolding
unf { df_args = args' })
              -- No need to use tagLamBinders because we
              -- never inline DFuns so the occ-info on binders doesn't matter

      Unfolding
unf -> TailUsageDetails -> Unfolding -> WithTailUsageDetails Unfolding
forall a. TailUsageDetails -> a -> WithTailUsageDetails a
WTUD (JoinArity -> UsageDetails -> TailUsageDetails
TUD JoinArity
0 UsageDetails
emptyDetails) Unfolding
unf

occAnalRule :: OccEnv
             -> CoreRule
             -> (CoreRule,         -- Each (non-built-in) rule
                 UsageDetails,     -- Usage details for LHS
                 TailUsageDetails) -- Usage details for RHS
occAnalRule :: OccEnv -> CoreRule -> (CoreRule, UsageDetails, TailUsageDetails)
occAnalRule OccEnv
env rule :: CoreRule
rule@(Rule { ru_bndrs :: CoreRule -> [Id]
ru_bndrs = [Id]
bndrs, ru_args :: CoreRule -> [CoreExpr]
ru_args = [CoreExpr]
args, ru_rhs :: CoreRule -> CoreExpr
ru_rhs = CoreExpr
rhs })
  = (CoreRule
rule', UsageDetails
lhs_uds', JoinArity -> UsageDetails -> TailUsageDetails
TUD JoinArity
rhs_ja UsageDetails
rhs_uds')
  where
    rule' :: CoreRule
rule' = CoreRule
rule { ru_args = args', ru_rhs = rhs' }

    WUD UsageDetails
lhs_uds [CoreExpr]
args' = OccEnv
-> [Id]
-> (OccEnv -> WithUsageDetails [CoreExpr])
-> WithUsageDetails [CoreExpr]
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeList OccEnv
env [Id]
bndrs ((OccEnv -> WithUsageDetails [CoreExpr])
 -> WithUsageDetails [CoreExpr])
-> (OccEnv -> WithUsageDetails [CoreExpr])
-> WithUsageDetails [CoreExpr]
forall a b. (a -> b) -> a -> b
$ \OccEnv
env ->
                        OccEnv -> [CoreExpr] -> WithUsageDetails [CoreExpr]
occAnalList OccEnv
env [CoreExpr]
args

    lhs_uds' :: UsageDetails
lhs_uds' = UsageDetails -> UsageDetails
markAllManyNonTail UsageDetails
lhs_uds
    WUD UsageDetails
rhs_uds CoreExpr
rhs' = OccEnv
-> [Id]
-> (OccEnv -> WithUsageDetails CoreExpr)
-> WithUsageDetails CoreExpr
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeList OccEnv
env [Id]
bndrs ((OccEnv -> WithUsageDetails CoreExpr)
 -> WithUsageDetails CoreExpr)
-> (OccEnv -> WithUsageDetails CoreExpr)
-> WithUsageDetails CoreExpr
forall a b. (a -> b) -> a -> b
$ \OccEnv
env ->
                       OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
env CoreExpr
rhs
                          -- Note [Rules are extra RHSs]
                          -- Note [Rule dependency info]
    rhs_uds' :: UsageDetails
rhs_uds' = UsageDetails -> UsageDetails
markAllMany UsageDetails
rhs_uds
    rhs_ja :: JoinArity
rhs_ja = [CoreExpr] -> JoinArity
forall a. [a] -> JoinArity
forall (t :: * -> *) a. Foldable t => t a -> JoinArity
length [CoreExpr]
args -- See Note [Join points and unfoldings/rules]

occAnalRule OccEnv
_ CoreRule
other_rule = (CoreRule
other_rule, UsageDetails
emptyDetails, JoinArity -> UsageDetails -> TailUsageDetails
TUD JoinArity
0 UsageDetails
emptyDetails)

{- Note [Occurrences in stable unfoldings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
    f p = BIG
    {-# INLINE g #-}
    g y = not (f y)
where this is the /only/ occurrence of 'f'.  So 'g' will get a stable
unfolding.  Now suppose that g's RHS gets optimised (perhaps by a rule
or inlining f) so that it doesn't mention 'f' any more.  Now the last
remaining call to f is in g's Stable unfolding. But, even though there
is only one syntactic occurrence of f, we do /not/ want to do
preinlineUnconditionally here!

The INLINE pragma says "inline exactly this RHS"; perhaps the
programmer wants to expose that 'not', say. If we inline f that will make
the Stable unfoldign big, and that wasn't what the programmer wanted.

Another way to think about it: if we inlined g as-is into multiple
call sites, now there's be multiple calls to f.

Bottom line: treat all occurrences in a stable unfolding as "Many".
We still leave tail call information intact, though, as to not spoil
potential join points.

Note [Unfoldings and rules]
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Generally unfoldings and rules are already occurrence-analysed, so we
don't want to reconstruct their trees; we just want to analyse them to
find how they use their free variables.

EXCEPT if there is a binder-swap going on, in which case we do want to
produce a new tree.

So we have a fast-path that keeps the old tree if the occ_bs_env is
empty.   This just saves a bit of allocation and reconstruction; not
a big deal.

Two tricky corners:

* Dead bindings (#22761). Supose we have
    Unfolding = \x. let y = foo in x+1
  which includes a dead binding for `y`. In occAnalUnfolding we occ-anal
  the unfolding and produce /no/ occurrences of `foo` (since `y` is
  dead).  But if we discard the occ-analysed syntax tree (which we do on
  our fast path), and use the old one, we still /have/ an occurrence of
  `foo` -- and that can lead to out-of-scope variables (#22761).

  Solution: always keep occ-analysed trees in unfoldings and rules, so they
  have no dead code.  See Note [OccInfo in unfoldings and rules] in GHC.Core.

* One-shot binders. Consider
     {- f has Stable unfolding \p q -> blah
        Demand on f is LC(L,C(1,!P(L)); that is, one-shot in its second ar -}
     f = \x y. blah

   Now we `mkRhsOccEnv` will build an OccEnv for f's RHS that has
          occ_one_shots = [NoOneShortInfo, OneShotLam]
   This will put OneShotLam on the \y.  And it'll put it on the \q.  But the
   noBinderSwap check will mean that we discard this new occ-anal'd unfolding
   and keep the old one, with no OneShotInfo.

   This looks a little inconsistent, but the Stable unfolding is just used for
   inlinings; OneShotInfo isn't a lot of use here.

Note [Cascading inlines]
~~~~~~~~~~~~~~~~~~~~~~~~
By default we use an OccRhs for the RHS of a binding.  This tells the
occ anal n that it's looking at an RHS, which has an effect in
occAnalApp.  In particular, for constructor applications, it makes
the arguments appear to have NoOccInfo, so that we don't inline into
them. Thus    x = f y
              k = Just x
we do not want to inline x.

But there's a problem.  Consider
     x1 = a0 : []
     x2 = a1 : x1
     x3 = a2 : x2
     g  = f x3
First time round, it looks as if x1 and x2 occur as an arg of a
let-bound constructor ==> give them a many-occurrence.
But then x3 is inlined (unconditionally as it happens) and
next time round, x2 will be, and the next time round x1 will be
Result: multiple simplifier iterations.  Sigh.

So, when analysing the RHS of x3 we notice that x3 will itself
definitely inline the next time round, and so we analyse x3's rhs in
an OccVanilla context, not OccRhs.  Hence the "certainly_inline" stuff.

Annoyingly, we have to approximate GHC.Core.Opt.Simplify.Utils.preInlineUnconditionally.
If (a) the RHS is expandable (see isExpandableApp in occAnalApp), and
   (b) certainly_inline says "yes" when preInlineUnconditionally says "no"
then the simplifier iterates indefinitely:
        x = f y
        k = Just x   -- We decide that k is 'certainly_inline'
        v = ...k...  -- but preInlineUnconditionally doesn't inline it
inline ==>
        k = Just (f y)
        v = ...k...
float ==>
        x1 = f y
        k = Just x1
        v = ...k...

This is worse than the slow cascade, so we only want to say "certainly_inline"
if it really is certain.  Look at the note with preInlineUnconditionally
for the various clauses.  See #24582 for an example of the two getting out of sync.


************************************************************************
*                                                                      *
                Expressions
*                                                                      *
************************************************************************
-}

occAnalList :: OccEnv -> [CoreExpr] -> WithUsageDetails [CoreExpr]
occAnalList :: OccEnv -> [CoreExpr] -> WithUsageDetails [CoreExpr]
occAnalList !OccEnv
_   []    = UsageDetails -> [CoreExpr] -> WithUsageDetails [CoreExpr]
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
emptyDetails []
occAnalList OccEnv
env (CoreExpr
e:[CoreExpr]
es) = let
                          (WUD UsageDetails
uds1 CoreExpr
e') = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
env CoreExpr
e
                          (WUD UsageDetails
uds2 [CoreExpr]
es') = OccEnv -> [CoreExpr] -> WithUsageDetails [CoreExpr]
occAnalList OccEnv
env [CoreExpr]
es
                         in UsageDetails -> [CoreExpr] -> WithUsageDetails [CoreExpr]
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails
uds1 UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
uds2) (CoreExpr
e' CoreExpr -> [CoreExpr] -> [CoreExpr]
forall a. a -> [a] -> [a]
: [CoreExpr]
es')

occAnal :: OccEnv
        -> CoreExpr
        -> WithUsageDetails CoreExpr       -- Gives info only about the "interesting" Ids

occAnal :: OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal !OccEnv
_   expr :: CoreExpr
expr@(Lit Literal
_)  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
emptyDetails CoreExpr
expr

occAnal OccEnv
env expr :: CoreExpr
expr@(Var Id
_) = OccEnv
-> (CoreExpr, [CoreExpr], [CoreTickish])
-> WithUsageDetails CoreExpr
occAnalApp OccEnv
env (CoreExpr
expr, [], [])
    -- At one stage, I gathered the idRuleVars for the variable here too,
    -- which in a way is the right thing to do.
    -- But that went wrong right after specialisation, when
    -- the *occurrences* of the overloaded function didn't have any
    -- rules in them, so the *specialised* versions looked as if they
    -- weren't used at all.

occAnal OccEnv
_ expr :: CoreExpr
expr@(Type Type
ty)
  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
emptyDetails (Type -> VarSet
coVarsOfType Type
ty)) CoreExpr
expr
occAnal OccEnv
_ expr :: CoreExpr
expr@(Coercion CoercionR
co)
  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
emptyDetails (CoercionR -> VarSet
coVarsOfCo CoercionR
co)) CoreExpr
expr
        -- See Note [Gather occurrences of coercion variables]

{- Note [Gather occurrences of coercion variables]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We need to gather info about what coercion variables appear, for two reasons:

1. So that we can sort them into the right place when doing dependency analysis.

2. So that we know when they are surely dead.

It is useful to know when they a coercion variable is surely dead,
when we want to discard a case-expression, in GHC.Core.Opt.Simplify.rebuildCase.
For example (#20143):

  case unsafeEqualityProof @blah of
     UnsafeRefl cv -> ...no use of cv...

Here we can discard the case, since unsafeEqualityProof always terminates.
But only if the coercion variable 'cv' is unused.

Another example from #15696: we had something like
  case eq_sel d of co -> ...(typeError @(...co...) "urk")...
Then 'd' was substituted by a dictionary, so the expression
simpified to
  case (Coercion <blah>) of cv -> ...(typeError @(...cv...) "urk")...

We can only  drop the case altogether if 'cv' is unused, which is not
the case here.

Conclusion: we need accurate dead-ness info for CoVars.
We gather CoVar occurrences from:

  * The (Type ty) and (Coercion co) cases of occAnal

  * The type 'ty' of a lambda-binder (\(x:ty). blah)
    See addCoVarOccs

But it is not necessary to gather CoVars from the types of other binders.

* For let-binders, if the type mentions a CoVar, so will the RHS (since
  it has the same type)

* For case-alt binders, if the type mentions a CoVar, so will the scrutinee
  (since it has the same type)
-}

occAnal OccEnv
env (Tick CoreTickish
tickish CoreExpr
body)
  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
usage' (CoreTickish -> CoreExpr -> CoreExpr
forall b. CoreTickish -> Expr b -> Expr b
Tick CoreTickish
tickish CoreExpr
body')
  where
    WUD UsageDetails
usage CoreExpr
body' = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
env CoreExpr
body

    usage' :: UsageDetails
usage'
      | CoreTickish
tickish CoreTickish -> TickishScoping -> Bool
forall (pass :: TickishPass).
GenTickish pass -> TickishScoping -> Bool
`tickishScopesLike` TickishScoping
SoftScope
      = UsageDetails
usage  -- For soft-scoped ticks (including SourceNotes) we don't want
               -- to lose join-point-hood, so we don't mess with `usage` (#24078)

      -- For a non-soft tick scope, we can inline lambdas only, so we
      -- abandon tail calls, and do markAllInsideLam too: usage_lam

      |  Breakpoint XBreakpoint 'TickishPassCore
_ JoinArity
_ [XTickishId 'TickishPassCore]
ids Module
_ <- CoreTickish
tickish
      = -- Never substitute for any of the Ids in a Breakpoint
        UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
usage_lam ([Id] -> VarSet
mkVarSet [Id]
[XTickishId 'TickishPassCore]
ids)

      | Bool
otherwise
      = UsageDetails
usage_lam

    usage_lam :: UsageDetails
usage_lam = UsageDetails -> UsageDetails
markAllNonTail (UsageDetails -> UsageDetails
markAllInsideLam UsageDetails
usage)

    -- TODO There may be ways to make ticks and join points play
    -- nicer together, but right now there are problems:
    --   let j x = ... in tick<t> (j 1)
    -- Making j a join point may cause the simplifier to drop t
    -- (if the tick is put into the continuation). So we don't
    -- count j 1 as a tail call.
    -- See #14242.

occAnal OccEnv
env (Cast CoreExpr
expr CoercionR
co)
  = let  (WUD UsageDetails
usage CoreExpr
expr') = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
env CoreExpr
expr
         usage1 :: UsageDetails
usage1 = UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
usage (CoercionR -> VarSet
coVarsOfCo CoercionR
co)
             -- usage2: see Note [Gather occurrences of coercion variables]
         usage2 :: UsageDetails
usage2 = UsageDetails -> UsageDetails
markAllNonTail UsageDetails
usage1
             -- usage3: calls inside expr aren't tail calls any more
    in UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
usage2 (CoreExpr -> CoercionR -> CoreExpr
forall b. Expr b -> CoercionR -> Expr b
Cast CoreExpr
expr' CoercionR
co)

occAnal OccEnv
env app :: CoreExpr
app@(App CoreExpr
_ CoreExpr
_)
  = OccEnv
-> (CoreExpr, [CoreExpr], [CoreTickish])
-> WithUsageDetails CoreExpr
occAnalApp OccEnv
env ((CoreTickish -> Bool)
-> CoreExpr -> (CoreExpr, [CoreExpr], [CoreTickish])
forall b.
(CoreTickish -> Bool)
-> Expr b -> (Expr b, [Expr b], [CoreTickish])
collectArgsTicks CoreTickish -> Bool
forall (pass :: TickishPass). GenTickish pass -> Bool
tickishFloatable CoreExpr
app)

occAnal OccEnv
env expr :: CoreExpr
expr@(Lam {})
  = JoinPointHood
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
adjustNonRecRhs JoinPointHood
NotJoinPoint (WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr)
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
forall a b. (a -> b) -> a -> b
$ -- NotJoinPoint <=> markAllManyNonTail
    OccEnv -> CoreExpr -> WithTailUsageDetails CoreExpr
occAnalLamTail OccEnv
env CoreExpr
expr

occAnal OccEnv
env (Case CoreExpr
scrut Id
bndr Type
ty [Alt Id]
alts)
  = let
      WUD UsageDetails
scrut_usage CoreExpr
scrut' = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal (OccEnv -> [Alt Id] -> OccEnv
setScrutCtxt OccEnv
env [Alt Id]
alts) CoreExpr
scrut

      WUD UsageDetails
alts_usage (Id
tagged_bndr, [Alt Id]
alts')
         = OccEnv
-> Id
-> (OccEnv -> WithUsageDetails (Id, [Alt Id]))
-> WithUsageDetails (Id, [Alt Id])
forall a.
OccEnv
-> Id -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeOne OccEnv
env Id
bndr ((OccEnv -> WithUsageDetails (Id, [Alt Id]))
 -> WithUsageDetails (Id, [Alt Id]))
-> (OccEnv -> WithUsageDetails (Id, [Alt Id]))
-> WithUsageDetails (Id, [Alt Id])
forall a b. (a -> b) -> a -> b
$ \OccEnv
env ->
           let alt_env :: OccEnv
alt_env = CoreExpr -> Id -> OccEnv -> OccEnv
addBndrSwap CoreExpr
scrut' Id
bndr (OccEnv -> OccEnv) -> OccEnv -> OccEnv
forall a b. (a -> b) -> a -> b
$
                         OccEnv -> OccEnv
setTailCtxt OccEnv
env  -- Kill off OccRhs
               WUD UsageDetails
alts_usage [Alt Id]
alts' = OccEnv -> [Alt Id] -> WithUsageDetails [Alt Id]
do_alts OccEnv
alt_env [Alt Id]
alts
               tagged_bndr :: Id
tagged_bndr = UsageDetails -> Id -> Id
tagLamBinder UsageDetails
alts_usage Id
bndr
           in UsageDetails -> (Id, [Alt Id]) -> WithUsageDetails (Id, [Alt Id])
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
alts_usage (Id
tagged_bndr, [Alt Id]
alts')

      total_usage :: UsageDetails
total_usage = UsageDetails -> UsageDetails
markAllNonTail UsageDetails
scrut_usage UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
alts_usage
                    -- Alts can have tail calls, but the scrutinee can't

    in UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
total_usage (CoreExpr -> Id -> Type -> [Alt Id] -> CoreExpr
forall b. Expr b -> b -> Type -> [Alt b] -> Expr b
Case CoreExpr
scrut' Id
tagged_bndr Type
ty [Alt Id]
alts')
  where
    do_alts :: OccEnv -> [CoreAlt] -> WithUsageDetails [CoreAlt]
    do_alts :: OccEnv -> [Alt Id] -> WithUsageDetails [Alt Id]
do_alts OccEnv
_   []         = UsageDetails -> [Alt Id] -> WithUsageDetails [Alt Id]
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
emptyDetails []
    do_alts OccEnv
env (Alt Id
alt:[Alt Id]
alts) = UsageDetails -> [Alt Id] -> WithUsageDetails [Alt Id]
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails
uds1 UsageDetails -> UsageDetails -> UsageDetails
`orUDs` UsageDetails
uds2) (Alt Id
alt'Alt Id -> [Alt Id] -> [Alt Id]
forall a. a -> [a] -> [a]
:[Alt Id]
alts')
      where
        WUD UsageDetails
uds1 Alt Id
alt'  = OccEnv -> Alt Id -> WithUsageDetails (Alt Id)
do_alt  OccEnv
env Alt Id
alt
        WUD UsageDetails
uds2 [Alt Id]
alts' = OccEnv -> [Alt Id] -> WithUsageDetails [Alt Id]
do_alts OccEnv
env [Alt Id]
alts

    do_alt :: OccEnv -> Alt Id -> WithUsageDetails (Alt Id)
do_alt !OccEnv
env (Alt AltCon
con [Id]
bndrs CoreExpr
rhs)
      = OccEnv
-> [Id]
-> (OccEnv -> WithUsageDetails (Alt Id))
-> WithUsageDetails (Alt Id)
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeList OccEnv
env [Id]
bndrs ((OccEnv -> WithUsageDetails (Alt Id))
 -> WithUsageDetails (Alt Id))
-> (OccEnv -> WithUsageDetails (Alt Id))
-> WithUsageDetails (Alt Id)
forall a b. (a -> b) -> a -> b
$ \ OccEnv
env ->
        let WUD UsageDetails
rhs_usage CoreExpr
rhs' = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
env CoreExpr
rhs
            tagged_bndrs :: [Id]
tagged_bndrs = UsageDetails -> [Id] -> [Id]
tagLamBinders UsageDetails
rhs_usage [Id]
bndrs
        in                 -- See Note [Binders in case alternatives]
        UsageDetails -> Alt Id -> WithUsageDetails (Alt Id)
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
rhs_usage (AltCon -> [Id] -> CoreExpr -> Alt Id
forall b. AltCon -> [b] -> Expr b -> Alt b
Alt AltCon
con [Id]
tagged_bndrs CoreExpr
rhs')

occAnal OccEnv
env (Let CoreBind
bind CoreExpr
body)
  = OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> CoreBind
-> (OccEnv -> WithUsageDetails CoreExpr)
-> (CoreProgram -> CoreExpr -> CoreExpr)
-> WithUsageDetails CoreExpr
forall r.
OccEnv
-> TopLevelFlag
-> ImpRuleEdges
-> CoreBind
-> (OccEnv -> WithUsageDetails r)
-> (CoreProgram -> r -> r)
-> WithUsageDetails r
occAnalBind OccEnv
env TopLevelFlag
NotTopLevel ImpRuleEdges
noImpRuleEdges CoreBind
bind
                (\OccEnv
env -> OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
env CoreExpr
body) CoreProgram -> CoreExpr -> CoreExpr
forall b. [Bind b] -> Expr b -> Expr b
mkLets

occAnalArgs :: OccEnv -> CoreExpr -> [CoreExpr]
            -> [OneShots]  -- Very commonly empty, notably prior to dmd anal
            -> WithUsageDetails CoreExpr
-- The `fun` argument is just an accumulating parameter,
-- the base for building the application we return
occAnalArgs :: OccEnv
-> CoreExpr
-> [CoreExpr]
-> [[OneShotInfo]]
-> WithUsageDetails CoreExpr
occAnalArgs !OccEnv
env CoreExpr
fun [CoreExpr]
args ![[OneShotInfo]]
one_shots
  = UsageDetails
-> CoreExpr
-> [CoreExpr]
-> [[OneShotInfo]]
-> WithUsageDetails CoreExpr
go UsageDetails
emptyDetails CoreExpr
fun [CoreExpr]
args [[OneShotInfo]]
one_shots
  where
    env_args :: OccEnv
env_args = OccEncl -> OccEnv -> OccEnv
setNonTailCtxt OccEncl
encl OccEnv
env

    -- Make bottoming functions interesting
    -- See Note [Bottoming function calls]
    encl :: OccEncl
encl | Var Id
f <- CoreExpr
fun, DmdSig -> Bool
isDeadEndSig (Id -> DmdSig
idDmdSig Id
f) = OccEncl
OccScrut
         | Bool
otherwise                               = OccEncl
OccVanilla

    go :: UsageDetails
-> CoreExpr
-> [CoreExpr]
-> [[OneShotInfo]]
-> WithUsageDetails CoreExpr
go UsageDetails
uds CoreExpr
fun [] [[OneShotInfo]]
_ = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
uds CoreExpr
fun
    go UsageDetails
uds CoreExpr
fun (CoreExpr
arg:[CoreExpr]
args) [[OneShotInfo]]
one_shots
      = UsageDetails
-> CoreExpr
-> [CoreExpr]
-> [[OneShotInfo]]
-> WithUsageDetails CoreExpr
go (UsageDetails
uds UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
arg_uds) (CoreExpr
fun CoreExpr -> CoreExpr -> CoreExpr
forall b. Expr b -> Expr b -> Expr b
`App` CoreExpr
arg') [CoreExpr]
args [[OneShotInfo]]
one_shots'
      where
        !(WUD UsageDetails
arg_uds CoreExpr
arg') = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal OccEnv
arg_env CoreExpr
arg
        !(OccEnv
arg_env, [[OneShotInfo]]
one_shots')
            | CoreExpr -> Bool
forall {b}. Expr b -> Bool
isTypeArg CoreExpr
arg
            = (OccEnv
env_args, [[OneShotInfo]]
one_shots)
            | Bool
otherwise
            = case [[OneShotInfo]]
one_shots of
                []                -> (OccEnv
env_args, []) -- Fast path; one_shots is often empty
                ([OneShotInfo]
os : [[OneShotInfo]]
one_shots') -> ([OneShotInfo] -> OccEnv -> OccEnv
setOneShots [OneShotInfo]
os OccEnv
env_args, [[OneShotInfo]]
one_shots')

{-
Applications are dealt with specially because we want
the "build hack" to work.

Note [Bottoming function calls]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
   let x = (a,b) in
   case p of
      A -> ...(error x)..
      B -> ...(ertor x)...

postInlineUnconditionally may duplicate x's binding, but sometimes it
does so only if the use site IsInteresting.  Pushing allocation into error
branches is good, so we try to make bottoming calls look interesting, by
setting occ_encl = OccScrut for such calls.

The slightly-artificial test T21128 is a good example.  It's probably
not a huge deal.

Note [Arguments of let-bound constructors]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
    f x = let y = expensive x in
          let z = (True,y) in
          (case z of {(p,q)->q}, case z of {(p,q)->q})
We feel free to duplicate the WHNF (True,y), but that means
that y may be duplicated thereby.

If we aren't careful we duplicate the (expensive x) call!
Constructors are rather like lambdas in this way.
-}

occAnalApp :: OccEnv
           -> (Expr CoreBndr, [Arg CoreBndr], [CoreTickish])
           -> WithUsageDetails (Expr CoreBndr)
-- Naked variables (not applied) end up here too
occAnalApp :: OccEnv
-> (CoreExpr, [CoreExpr], [CoreTickish])
-> WithUsageDetails CoreExpr
occAnalApp !OccEnv
env (Var Id
fun, [CoreExpr]
args, [CoreTickish]
ticks)
  -- Account for join arity of runRW# continuation
  -- See Note [Simplification of runRW#]
  --
  -- NB: Do not be tempted to make the next (Var fun, args, tick)
  --     equation into an 'otherwise' clause for this equation
  --     The former has a bang-pattern to occ-anal the args, and
  --     we don't want to occ-anal them twice in the runRW# case!
  --     This caused #18296
  | Id
fun Id -> Unique -> Bool
forall a. Uniquable a => a -> Unique -> Bool
`hasKey` Unique
runRWKey
  , [CoreExpr
t1, CoreExpr
t2, CoreExpr
arg]  <- [CoreExpr]
args
  , WUD UsageDetails
usage CoreExpr
arg' <- JoinPointHood
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
adjustNonRecRhs (JoinArity -> JoinPointHood
JoinPoint JoinArity
1) (WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr)
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
forall a b. (a -> b) -> a -> b
$ OccEnv -> CoreExpr -> WithTailUsageDetails CoreExpr
occAnalLamTail OccEnv
env CoreExpr
arg
  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
usage ([CoreTickish] -> CoreExpr -> CoreExpr
mkTicks [CoreTickish]
ticks (CoreExpr -> CoreExpr) -> CoreExpr -> CoreExpr
forall a b. (a -> b) -> a -> b
$ CoreExpr -> [CoreExpr] -> CoreExpr
forall b. Expr b -> [Expr b] -> Expr b
mkApps (Id -> CoreExpr
forall b. Id -> Expr b
Var Id
fun) [CoreExpr
t1, CoreExpr
t2, CoreExpr
arg'])

occAnalApp OccEnv
env (Var Id
fun_id, [CoreExpr]
args, [CoreTickish]
ticks)
  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
all_uds ([CoreTickish] -> CoreExpr -> CoreExpr
mkTicks [CoreTickish]
ticks CoreExpr
app')
  where
    -- Lots of banged bindings: this is a very heavily bit of code,
    -- so it pays not to make lots of thunks here, all of which
    -- will ultimately be forced.
    !(CoreExpr
fun', Id
fun_id')  = OccEnv -> Id -> (CoreExpr, Id)
lookupBndrSwap OccEnv
env Id
fun_id
    !(WUD UsageDetails
args_uds CoreExpr
app') = OccEnv
-> CoreExpr
-> [CoreExpr]
-> [[OneShotInfo]]
-> WithUsageDetails CoreExpr
occAnalArgs OccEnv
env CoreExpr
fun' [CoreExpr]
args [[OneShotInfo]]
one_shots

    fun_uds :: UsageDetails
fun_uds = OccEnv -> Id -> InterestingCxt -> JoinArity -> UsageDetails
mkOneOcc OccEnv
env Id
fun_id' InterestingCxt
int_cxt JoinArity
n_args
       -- NB: fun_uds is computed for fun_id', not fun_id
       -- See (BS1) in Note [The binder-swap substitution]

    all_uds :: UsageDetails
all_uds = UsageDetails
fun_uds UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
final_args_uds

    !final_args_uds :: UsageDetails
final_args_uds = UsageDetails -> UsageDetails
markAllNonTail                              (UsageDetails -> UsageDetails) -> UsageDetails -> UsageDetails
forall a b. (a -> b) -> a -> b
$
                      Bool -> UsageDetails -> UsageDetails
markAllInsideLamIf (OccEnv -> Bool
isRhsEnv OccEnv
env Bool -> Bool -> Bool
&& Bool
is_exp) (UsageDetails -> UsageDetails) -> UsageDetails -> UsageDetails
forall a b. (a -> b) -> a -> b
$
                        -- isRhsEnv: see Note [OccEncl]
                      UsageDetails
args_uds
       -- We mark the free vars of the argument of a constructor or PAP
       -- as "inside-lambda", if it is the RHS of a let(rec).
       -- This means that nothing gets inlined into a constructor or PAP
       -- argument position, which is what we want.  Typically those
       -- constructor arguments are just variables, or trivial expressions.
       -- We use inside-lam because it's like eta-expanding the PAP.
       --
       -- This is the *whole point* of the isRhsEnv predicate
       -- See Note [Arguments of let-bound constructors]

    !n_val_args :: JoinArity
n_val_args = [CoreExpr] -> JoinArity
forall b. [Arg b] -> JoinArity
valArgCount [CoreExpr]
args
    !n_args :: JoinArity
n_args     = [CoreExpr] -> JoinArity
forall a. [a] -> JoinArity
forall (t :: * -> *) a. Foldable t => t a -> JoinArity
length [CoreExpr]
args
    !int_cxt :: InterestingCxt
int_cxt    = case OccEnv -> OccEncl
occ_encl OccEnv
env of
                   OccEncl
OccScrut -> InterestingCxt
IsInteresting
                   OccEncl
_other   | JoinArity
n_val_args JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
> JoinArity
0 -> InterestingCxt
IsInteresting
                            | Bool
otherwise      -> InterestingCxt
NotInteresting

    !is_exp :: Bool
is_exp     = CheapAppFun
isExpandableApp Id
fun_id JoinArity
n_val_args
        -- See Note [CONLIKE pragma] in GHC.Types.Basic
        -- The definition of is_exp should match that in GHC.Core.Opt.Simplify.prepareRhs

    one_shots :: [[OneShotInfo]]
one_shots  = DmdSig -> JoinArity -> [[OneShotInfo]]
argsOneShots (Id -> DmdSig
idDmdSig Id
fun_id) JoinArity
guaranteed_val_args
    guaranteed_val_args :: JoinArity
guaranteed_val_args = JoinArity
n_val_args JoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
+ [OneShotInfo] -> JoinArity
forall a. [a] -> JoinArity
forall (t :: * -> *) a. Foldable t => t a -> JoinArity
length ((OneShotInfo -> Bool) -> [OneShotInfo] -> [OneShotInfo]
forall a. (a -> Bool) -> [a] -> [a]
takeWhile OneShotInfo -> Bool
isOneShotInfo
                                                         (OccEnv -> [OneShotInfo]
occ_one_shots OccEnv
env))
        -- See Note [Sources of one-shot information], bullet point A']

occAnalApp OccEnv
env (CoreExpr
fun, [CoreExpr]
args, [CoreTickish]
ticks)
  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (UsageDetails -> UsageDetails
markAllNonTail (UsageDetails
fun_uds UsageDetails -> UsageDetails -> UsageDetails
`andUDs` UsageDetails
args_uds))
                     ([CoreTickish] -> CoreExpr -> CoreExpr
mkTicks [CoreTickish]
ticks CoreExpr
app')
  where
    !(WUD UsageDetails
args_uds CoreExpr
app') = OccEnv
-> CoreExpr
-> [CoreExpr]
-> [[OneShotInfo]]
-> WithUsageDetails CoreExpr
occAnalArgs OccEnv
env CoreExpr
fun' [CoreExpr]
args []
    !(WUD UsageDetails
fun_uds CoreExpr
fun')  = OccEnv -> CoreExpr -> WithUsageDetails CoreExpr
occAnal (OccEnv -> [CoreExpr] -> OccEnv
addAppCtxt OccEnv
env [CoreExpr]
args) CoreExpr
fun
        -- The addAppCtxt is a bit cunning.  One iteration of the simplifier
        -- often leaves behind beta redexes like
        --      (\x y -> e) a1 a2
        -- Here we would like to mark x,y as one-shot, and treat the whole
        -- thing much like a let.  We do this by pushing some OneShotLam items
        -- onto the context stack.

addAppCtxt :: OccEnv -> [Arg CoreBndr] -> OccEnv
addAppCtxt :: OccEnv -> [CoreExpr] -> OccEnv
addAppCtxt env :: OccEnv
env@(OccEnv { occ_one_shots :: OccEnv -> [OneShotInfo]
occ_one_shots = [OneShotInfo]
ctxt }) [CoreExpr]
args
  | JoinArity
n_val_args JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
> JoinArity
0
  = OccEnv
env { occ_one_shots = replicate n_val_args OneShotLam ++ ctxt
        , occ_encl      = OccVanilla }
          -- OccVanilla: the function part of the application
          -- is no longer on OccRhs or OccScrut
  | Bool
otherwise
  = OccEnv
env
  where
    n_val_args :: JoinArity
n_val_args = [CoreExpr] -> JoinArity
forall b. [Arg b] -> JoinArity
valArgCount [CoreExpr]
args


{-
Note [Sources of one-shot information]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The occurrence analyser obtains one-shot-lambda information from two sources:

A:  Saturated applications:  eg   f e1 .. en

    In general, given a call (f e1 .. en) we can propagate one-shot info from
    f's strictness signature into e1 .. en, but /only/ if n is enough to
    saturate the strictness signature. A strictness signature like

          f :: C(1,C(1,L))LS

    means that *if f is applied to three arguments* then it will guarantee to
    call its first argument at most once, and to call the result of that at
    most once. But if f has fewer than three arguments, all bets are off; e.g.

          map (f (\x y. expensive) e2) xs

    Here the \x y abstraction may be called many times (once for each element of
    xs) so we should not mark x and y as one-shot. But if it was

          map (f (\x y. expensive) 3 2) xs

    then the first argument of f will be called at most once.

    The one-shot info, derived from f's strictness signature, is
    computed by 'argsOneShots', called in occAnalApp.

A': Non-obviously saturated applications: eg    build (f (\x y -> expensive))
    where f is as above.

    In this case, f is only manifestly applied to one argument, so it does not
    look saturated. So by the previous point, we should not use its strictness
    signature to learn about the one-shotness of \x y. But in this case we can:
    build is fully applied, so we may use its strictness signature; and from
    that we learn that build calls its argument with two arguments *at most once*.

    So there is really only one call to f, and it will have three arguments. In
    that sense, f is saturated, and we may proceed as described above.

    Hence the computation of 'guaranteed_val_args' in occAnalApp, using
    '(occ_one_shots env)'.  See also #13227, comment:9

B:  Let-bindings:  eg   let f = \c. let ... in \n -> blah
                        in (build f, build f)

    Propagate one-shot info from the demand-info on 'f' to the
    lambdas in its RHS (which may not be syntactically at the top)

    This information must have come from a previous run of the demand
    analyser.

Previously, the demand analyser would *also* set the one-shot information, but
that code was buggy (see #11770), so doing it only in on place, namely here, is
saner.

Note [OneShots]
~~~~~~~~~~~~~~~
When analysing an expression, the occ_one_shots argument contains information
about how the function is being used. The length of the list indicates
how many arguments will eventually be passed to the analysed expression,
and the OneShotInfo indicates whether this application is once or multiple times.

Example:

 Context of f                occ_one_shots when analysing f

 f 1 2                       [OneShot, OneShot]
 map (f 1)                   [OneShot, NoOneShotInfo]
 build f                     [OneShot, OneShot]
 f 1 2 `seq` f 2 1           [NoOneShotInfo, OneShot]

Note [Binders in case alternatives]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
    case x of y { (a,b) -> f y }
We treat 'a', 'b' as dead, because they don't physically occur in the
case alternative.  (Indeed, a variable is dead iff it doesn't occur in
its scope in the output of OccAnal.)  It really helps to know when
binders are unused.  See esp the call to isDeadBinder in
Simplify.mkDupableAlt

In this example, though, the Simplifier will bring 'a' and 'b' back to
life, because it binds 'y' to (a,b) (imagine got inlined and
scrutinised y).
-}

{-
************************************************************************
*                                                                      *
                    OccEnv
*                                                                      *
************************************************************************
-}

data OccEnv
  = OccEnv { OccEnv -> OccEncl
occ_encl       :: !OccEncl      -- Enclosing context information
           , OccEnv -> [OneShotInfo]
occ_one_shots  :: !OneShots     -- See Note [OneShots]
           , OccEnv -> Id -> Bool
occ_unf_act    :: Id -> Bool          -- Which Id unfoldings are active
           , OccEnv -> Activation -> Bool
occ_rule_act   :: Activation -> Bool  -- Which rules are active
             -- See Note [Finding rule RHS free vars]

           -- See Note [The binder-swap substitution]
           -- If  x :-> (y, co)  is in the env,
           -- then please replace x by (y |> mco)
           -- Invariant of course: idType x = exprType (y |> mco)
           , OccEnv -> IdEnv (Id, MCoercion)
occ_bs_env  :: !(IdEnv (OutId, MCoercion))
              -- Domain is Global and Local Ids
              -- Range is just Local Ids
           , OccEnv -> VarSet
occ_bs_rng  :: !VarSet
               -- Vars (TyVars and Ids) free in the range of occ_bs_env

             -- Usage details of the RHS of in-scope non-recursive join points
             -- Invariant: no Id maps to an empty OccInfoEnv
             -- See Note [Occurrence analysis for join points]
           , OccEnv -> JoinPointInfo
occ_join_points :: !JoinPointInfo
    }

type JoinPointInfo = IdEnv OccInfoEnv

-----------------------------
{- Note [OccEncl]
~~~~~~~~~~~~~~~~~
OccEncl is used to control whether to inline into constructor arguments.

* OccRhs: consider
     let p = <blah> in
     let x = Just p
     in ...case p of ...

  Here `p` occurs syntactically once, but we want to mark it as InsideLam
  to stop `p` inlining.  We want to leave the x-binding as a constructor
  applied to variables, so that the Simplifier can simplify that inner `case`.

  The OccRhs just tells occAnalApp to mark occurrences in constructor args

* OccScrut: consider (case x of ...).  Here we want to give `x` OneOcc
  with "interesting context" field int_cxt = True.  The OccScrut tells
  occAnalApp (which deals with lone variables too) when to set this field
  to True.
-}

data OccEncl -- See Note [OccEncl]
  = OccRhs         -- RHS of let(rec), albeit perhaps inside a type lambda
  | OccScrut       -- Scrutintee of a case
  | OccVanilla     -- Everything else

instance Outputable OccEncl where
  ppr :: OccEncl -> SDoc
ppr OccEncl
OccRhs     = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"occRhs"
  ppr OccEncl
OccScrut   = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"occScrut"
  ppr OccEncl
OccVanilla = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"occVanilla"

-- See Note [OneShots]
type OneShots = [OneShotInfo]

initOccEnv :: OccEnv
initOccEnv :: OccEnv
initOccEnv
  = OccEnv { occ_encl :: OccEncl
occ_encl      = OccEncl
OccVanilla
           , occ_one_shots :: [OneShotInfo]
occ_one_shots = []

                 -- To be conservative, we say that all
                 -- inlines and rules are active
           , occ_unf_act :: Id -> Bool
occ_unf_act   = \Id
_ -> Bool
True
           , occ_rule_act :: Activation -> Bool
occ_rule_act  = \Activation
_ -> Bool
True

           , occ_join_points :: JoinPointInfo
occ_join_points = JoinPointInfo
forall a. VarEnv a
emptyVarEnv
           , occ_bs_env :: IdEnv (Id, MCoercion)
occ_bs_env = IdEnv (Id, MCoercion)
forall a. VarEnv a
emptyVarEnv
           , occ_bs_rng :: VarSet
occ_bs_rng = VarSet
emptyVarSet }

noBinderSwaps :: OccEnv -> Bool
noBinderSwaps :: OccEnv -> Bool
noBinderSwaps (OccEnv { occ_bs_env :: OccEnv -> IdEnv (Id, MCoercion)
occ_bs_env = IdEnv (Id, MCoercion)
bs_env }) = IdEnv (Id, MCoercion) -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv IdEnv (Id, MCoercion)
bs_env

setScrutCtxt :: OccEnv -> [CoreAlt] -> OccEnv
setScrutCtxt :: OccEnv -> [Alt Id] -> OccEnv
setScrutCtxt !OccEnv
env [Alt Id]
alts
  = OccEncl -> OccEnv -> OccEnv
setNonTailCtxt OccEncl
encl OccEnv
env
  where
    encl :: OccEncl
encl | Bool
interesting_alts = OccEncl
OccScrut
         | Bool
otherwise        = OccEncl
OccVanilla

    interesting_alts :: Bool
interesting_alts = case [Alt Id]
alts of
                         []    -> Bool
False
                         [Alt Id
alt] -> Bool -> Bool
not (Alt Id -> Bool
forall b. Alt b -> Bool
isDefaultAlt Alt Id
alt)
                         [Alt Id]
_     -> Bool
True
     -- 'interesting_alts' is True if the case has at least one
     -- non-default alternative.  That in turn influences
     -- pre/postInlineUnconditionally.  Grep for "occ_int_cxt"!

{- Note [The OccEnv for a right hand side]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
How do we create the OccEnv for a RHS (in mkRhsOccEnv)?

For a non-join point binding, x = rhs

  * occ_encl: set to OccRhs; but see `mkNonRecRhsCtxt` for wrinkles

  * occ_join_points: zap them!

  * occ_one_shots: initialise from the idDemandInfo;
    see Note [Sources of one-shot information]

For a join point binding,  j x = rhs

  * occ_encl: Consider
       x = e
       join j = Just x
    We want to inline x into j right away, so we don't want to give the join point
    a OccRhs (#14137); we want OccVanilla.  It's not a huge deal, because the
    FloatIn pass knows to float into join point RHSs; and the simplifier does not
    float things out of join point RHSs.  But it's a simple, cheap thing to do.

  * occ_join_points: no need to zap.

  * occ_one_shots: we start with one-shot-info from the context, which indeed
    applies to the /body/ of the join point, after walking past the binders.
    So we add to the front a OneShotInfo for each value-binder of the join
    point: see `extendOneShotsForJoinPoint`. (Failing to account for the join-point
    binders caused #25096.)

    For the join point binders themselves, of a /non-recursive/ join point,
    we make the binder a OneShotLam.  Again see `extendOneShotsForJoinPoint`.

    These one-shot infos then get attached to the binder by `occAnalLamTail`.
-}

setNonTailCtxt :: OccEncl -> OccEnv -> OccEnv
setNonTailCtxt :: OccEncl -> OccEnv -> OccEnv
setNonTailCtxt OccEncl
ctxt !OccEnv
env
  = OccEnv
env { occ_encl        = ctxt
        , occ_one_shots   = []
        , occ_join_points = zapJoinPointInfo (occ_join_points env) }

setTailCtxt :: OccEnv -> OccEnv
setTailCtxt :: OccEnv -> OccEnv
setTailCtxt !OccEnv
env = OccEnv
env { occ_encl = OccVanilla }
    -- Preserve occ_one_shots, occ_join points
    -- Do not use OccRhs for the RHS of a join point (which is a tail ctxt):

mkRhsOccEnv :: OccEnv -> RecFlag -> OccEncl -> JoinPointHood -> Id -> CoreExpr -> OccEnv
-- See Note [The OccEnv for a right hand side]
-- For a join point:
--   - Keep occ_one_shots, occ_joinPoints from the context
--   - But push enough OneShotInfo onto occ_one_shots to account
--     for the join-point value binders
--   - Set occ_encl to OccVanilla
-- For non-join points
--   - Zap occ_one_shots and occ_join_points
--   - Set occ_encl to specified OccEncl
mkRhsOccEnv :: OccEnv
-> RecFlag -> OccEncl -> JoinPointHood -> Id -> CoreExpr -> OccEnv
mkRhsOccEnv env :: OccEnv
env@(OccEnv { occ_one_shots :: OccEnv -> [OneShotInfo]
occ_one_shots = [OneShotInfo]
ctxt_one_shots, occ_join_points :: OccEnv -> JoinPointInfo
occ_join_points = JoinPointInfo
ctxt_join_points })
            RecFlag
is_rec OccEncl
encl JoinPointHood
jp_hood Id
bndr CoreExpr
rhs
  | JoinPoint JoinArity
join_arity <- JoinPointHood
jp_hood
  = OccEnv
env { occ_encl        = OccVanilla
        , occ_one_shots   = extendOneShotsForJoinPoint is_rec join_arity rhs ctxt_one_shots
        , occ_join_points = ctxt_join_points }

  | Bool
otherwise
  = OccEnv
env { occ_encl        = encl
        , occ_one_shots   = argOneShots (idDemandInfo bndr)
                            -- argOneShots: see Note [Sources of one-shot information]
        , occ_join_points = zapJoinPointInfo ctxt_join_points }

zapJoinPointInfo :: JoinPointInfo -> JoinPointInfo
-- (zapJoinPointInfo jp_info) basically just returns emptyVarEnv (hence zapped).
-- See (W3) of Note [Occurrence analysis for join points]
--
-- Zapping improves efficiency, slightly, if you accidentally introduce a bug,
-- in which you zap [jx :-> uds] and then find an occurrence of jx anyway, you
-- might lose those uds, and that might mean we don't record all occurrencs, and
-- that means we duplicate a redex....  a very nasty bug (which I encountered!).
-- Hence this DEBUG code which doesn't remove jx from the envt; it just gives it
-- emptyDetails, which in turn causes a panic in mkOneOcc. That will catch this
-- bug before it does any damage.
#ifdef DEBUG
zapJoinPointInfo jp_info = mapVarEnv (\ _ -> emptyVarEnv) jp_info
#else
zapJoinPointInfo :: JoinPointInfo -> JoinPointInfo
zapJoinPointInfo JoinPointInfo
_       = JoinPointInfo
forall a. VarEnv a
emptyVarEnv
#endif

extendOneShotsForJoinPoint
  :: RecFlag -> JoinArity -> CoreExpr
  -> [OneShotInfo] -> [OneShotInfo]
-- Push enough OneShortInfos on the front of ctxt_one_shots
-- to account for the value lambdas of the join point
extendOneShotsForJoinPoint :: RecFlag -> JoinArity -> CoreExpr -> [OneShotInfo] -> [OneShotInfo]
extendOneShotsForJoinPoint RecFlag
is_rec JoinArity
join_arity CoreExpr
rhs [OneShotInfo]
ctxt_one_shots
  = JoinArity -> CoreExpr -> [OneShotInfo]
go JoinArity
join_arity CoreExpr
rhs
  where
    -- For a /non-recursive/ join point we can mark all
    -- its join-lambda as one-shot; and it's a good idea to do so
    -- But not so for recursive ones
    os :: OneShotInfo
os = case RecFlag
is_rec of
           RecFlag
NonRecursive -> OneShotInfo
OneShotLam
           RecFlag
Recursive    -> OneShotInfo
NoOneShotInfo

    go :: JoinArity -> CoreExpr -> [OneShotInfo]
go JoinArity
0 CoreExpr
_        = [OneShotInfo]
ctxt_one_shots
    go JoinArity
n (Lam Id
b CoreExpr
rhs)
      | Id -> Bool
isId Id
b    = OneShotInfo
os OneShotInfo -> [OneShotInfo] -> [OneShotInfo]
forall a. a -> [a] -> [a]
: JoinArity -> CoreExpr -> [OneShotInfo]
go (JoinArity
nJoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
-JoinArity
1) CoreExpr
rhs
      | Bool
otherwise =      JoinArity -> CoreExpr -> [OneShotInfo]
go (JoinArity
nJoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
-JoinArity
1) CoreExpr
rhs
    go JoinArity
_ CoreExpr
_        = []  -- Not enough lambdas.  This can legitimately happen.
                        -- e.g.    let j = case ... in j True
                        -- This will become an arity-1 join point after the
                        -- simplifier has eta-expanded it; but it may not have
                        -- enough lambdas /yet/. (Lint checks that JoinIds do
                        -- have enough lambdas.)

setOneShots :: OneShots -> OccEnv -> OccEnv
setOneShots :: [OneShotInfo] -> OccEnv -> OccEnv
setOneShots [OneShotInfo]
os !OccEnv
env
  | [OneShotInfo] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [OneShotInfo]
os   = OccEnv
env  -- Fast path for common case
  | Bool
otherwise = OccEnv
env { occ_one_shots = os }

isRhsEnv :: OccEnv -> Bool
isRhsEnv :: OccEnv -> Bool
isRhsEnv (OccEnv { occ_encl :: OccEnv -> OccEncl
occ_encl = OccEncl
cxt }) = case OccEncl
cxt of
                                          OccEncl
OccRhs -> Bool
True
                                          OccEncl
_      -> Bool
False

addInScopeList :: OccEnv -> [Var]
               -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
{-# INLINE addInScopeList #-}
addInScopeList :: forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeList OccEnv
env [Id]
bndrs OccEnv -> WithUsageDetails a
thing_inside
 | [Id] -> Bool
forall a. [a] -> Bool
forall (t :: * -> *) a. Foldable t => t a -> Bool
null [Id]
bndrs = OccEnv -> WithUsageDetails a
thing_inside OccEnv
env  -- E.g. nullary constructors in a `case`
 | Bool
otherwise  = OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScope OccEnv
env [Id]
bndrs OccEnv -> WithUsageDetails a
thing_inside

addInScopeOne :: OccEnv -> Id
               -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
{-# INLINE addInScopeOne #-}
addInScopeOne :: forall a.
OccEnv
-> Id -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScopeOne OccEnv
env Id
bndr = OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScope OccEnv
env [Id
bndr]

addInScope :: OccEnv -> [Var]
           -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
{-# INLINE addInScope #-}
-- This function is called a lot, so we want to inline the fast path
-- so we don't have to allocate thing_inside and call it
-- The bndrs must include TyVars as well as Ids, because of
--     (BS3) in Note [Binder swap]
-- We do not assume that the bndrs are in scope order; in fact the
-- call in occ_anal_lam_tail gives them to addInScope in /reverse/ order

-- Fast path when the is no environment-munging to do
-- This is rather common: notably at top level, but nested too
addInScope :: forall a.
OccEnv
-> [Id] -> (OccEnv -> WithUsageDetails a) -> WithUsageDetails a
addInScope OccEnv
env [Id]
bndrs OccEnv -> WithUsageDetails a
thing_inside
  | IdEnv (Id, MCoercion) -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv (OccEnv -> IdEnv (Id, MCoercion)
occ_bs_env OccEnv
env)
  , JoinPointInfo -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv (OccEnv -> JoinPointInfo
occ_join_points OccEnv
env)
  , WUD UsageDetails
uds a
res <- OccEnv -> WithUsageDetails a
thing_inside OccEnv
env
  = UsageDetails -> a -> WithUsageDetails a
forall a. UsageDetails -> a -> WithUsageDetails a
WUD ([Id] -> UsageDetails -> UsageDetails
delBndrsFromUDs [Id]
bndrs UsageDetails
uds) a
res

addInScope OccEnv
env [Id]
bndrs OccEnv -> WithUsageDetails a
thing_inside
  = UsageDetails -> a -> WithUsageDetails a
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
uds' a
res
  where
    bndr_set :: VarSet
bndr_set           = [Id] -> VarSet
mkVarSet [Id]
bndrs
    !(OccEnv
env', JoinPointInfo
bad_joins) = OccEnv -> VarSet -> (OccEnv, JoinPointInfo)
preprocess_env OccEnv
env VarSet
bndr_set
    !(WUD UsageDetails
uds a
res)     = OccEnv -> WithUsageDetails a
thing_inside OccEnv
env'
    uds' :: UsageDetails
uds'               = [Id] -> JoinPointInfo -> UsageDetails -> UsageDetails
postprocess_uds [Id]
bndrs JoinPointInfo
bad_joins UsageDetails
uds

preprocess_env :: OccEnv -> VarSet -> (OccEnv, JoinPointInfo)
preprocess_env :: OccEnv -> VarSet -> (OccEnv, JoinPointInfo)
preprocess_env env :: OccEnv
env@(OccEnv { occ_join_points :: OccEnv -> JoinPointInfo
occ_join_points = JoinPointInfo
join_points
                           , occ_bs_rng :: OccEnv -> VarSet
occ_bs_rng = VarSet
bs_rng_vars })
               VarSet
bndr_set
  | Bool
bad_joins = (OccEnv -> OccEnv
drop_shadowed_swaps (OccEnv -> OccEnv
drop_shadowed_joins OccEnv
env), JoinPointInfo
join_points)
  | Bool
otherwise = (OccEnv -> OccEnv
drop_shadowed_swaps OccEnv
env,                       JoinPointInfo
forall a. VarEnv a
emptyVarEnv)
  where
    drop_shadowed_swaps :: OccEnv -> OccEnv
    -- See Note [The binder-swap substitution] (BS3)
    drop_shadowed_swaps :: OccEnv -> OccEnv
drop_shadowed_swaps env :: OccEnv
env@(OccEnv { occ_bs_env :: OccEnv -> IdEnv (Id, MCoercion)
occ_bs_env = IdEnv (Id, MCoercion)
swap_env })
      | IdEnv (Id, MCoercion) -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv IdEnv (Id, MCoercion)
swap_env
      = OccEnv
env
      | VarSet
bs_rng_vars VarSet -> VarSet -> Bool
`intersectsVarSet` VarSet
bndr_set
      = OccEnv
env { occ_bs_env = emptyVarEnv, occ_bs_rng = emptyVarSet }
      | Bool
otherwise
      = OccEnv
env { occ_bs_env = swap_env `minusUFM` bndr_fm }

    drop_shadowed_joins :: OccEnv -> OccEnv
    -- See Note [Occurrence analysis for join points] wrinkle2 (W1) and (W2)
    drop_shadowed_joins :: OccEnv -> OccEnv
drop_shadowed_joins OccEnv
env = OccEnv
env { occ_join_points = emptyVarEnv }

    -- bad_joins is true if it would be wrong to push occ_join_points inwards
    --  (a) `bndrs` includes any of the occ_join_points
    --  (b) `bndrs` includes any variables free in the RHSs of occ_join_points
    bad_joins :: Bool
    bad_joins :: Bool
bad_joins = (Unique -> VarEnv LocalOcc -> Bool -> Bool)
-> Bool -> JoinPointInfo -> Bool
forall a r. (Unique -> a -> r -> r) -> r -> VarEnv a -> r
nonDetStrictFoldVarEnv_Directly Unique -> VarEnv LocalOcc -> Bool -> Bool
is_bad Bool
False JoinPointInfo
join_points

    bndr_fm :: UniqFM Var Var
    bndr_fm :: VarEnv Id
bndr_fm = VarSet -> VarEnv Id
forall a. UniqSet a -> UniqFM a a
getUniqSet VarSet
bndr_set

    is_bad :: Unique -> OccInfoEnv -> Bool -> Bool
    is_bad :: Unique -> VarEnv LocalOcc -> Bool -> Bool
is_bad Unique
uniq VarEnv LocalOcc
join_uds Bool
rest
      = Unique
uniq Unique -> VarSet -> Bool
forall a. Unique -> UniqSet a -> Bool
`elemUniqSet_Directly` VarSet
bndr_set Bool -> Bool -> Bool
||
        Bool -> Bool
not (VarEnv Id
bndr_fm VarEnv Id -> VarEnv LocalOcc -> Bool
forall {k} (key :: k) elt1 elt2.
UniqFM key elt1 -> UniqFM key elt2 -> Bool
`disjointUFM` VarEnv LocalOcc
join_uds) Bool -> Bool -> Bool
||
        Bool
rest

postprocess_uds :: [Var] -> JoinPointInfo -> UsageDetails -> UsageDetails
postprocess_uds :: [Id] -> JoinPointInfo -> UsageDetails -> UsageDetails
postprocess_uds [Id]
bndrs JoinPointInfo
bad_joins UsageDetails
uds
  = UsageDetails -> UsageDetails
add_bad_joins ([Id] -> UsageDetails -> UsageDetails
delBndrsFromUDs [Id]
bndrs UsageDetails
uds)
  where
    add_bad_joins :: UsageDetails -> UsageDetails
    -- Add usage info for occ_join_points that we cannot push inwards
    -- because of shadowing
    -- See Note [Occurrence analysis for join points] wrinkle (W2)
    add_bad_joins :: UsageDetails -> UsageDetails
add_bad_joins UsageDetails
uds
       | JoinPointInfo -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv JoinPointInfo
bad_joins = UsageDetails
uds
       | Bool
otherwise               = (VarEnv LocalOcc -> VarEnv LocalOcc)
-> UsageDetails -> UsageDetails
modifyUDEnv VarEnv LocalOcc -> VarEnv LocalOcc
extend_with_bad_joins UsageDetails
uds

    extend_with_bad_joins :: OccInfoEnv -> OccInfoEnv
    extend_with_bad_joins :: VarEnv LocalOcc -> VarEnv LocalOcc
extend_with_bad_joins VarEnv LocalOcc
env
       = (Unique -> VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc)
-> VarEnv LocalOcc -> JoinPointInfo -> VarEnv LocalOcc
forall {k} elt a (key :: k).
(Unique -> elt -> a -> a) -> a -> UniqFM key elt -> a
nonDetStrictFoldUFM_Directly Unique -> VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc
add_bad_join VarEnv LocalOcc
env JoinPointInfo
bad_joins

    add_bad_join :: Unique -> OccInfoEnv -> OccInfoEnv -> OccInfoEnv
    -- Behave like `andUDs` when adding in the bad_joins
    add_bad_join :: Unique -> VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc
add_bad_join Unique
uniq VarEnv LocalOcc
join_env VarEnv LocalOcc
env
      | Unique
uniq Unique -> VarEnv LocalOcc -> Bool
forall a. Unique -> VarEnv a -> Bool
`elemVarEnvByKey` VarEnv LocalOcc
env = (LocalOcc -> LocalOcc -> LocalOcc)
-> VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc
forall a. (a -> a -> a) -> VarEnv a -> VarEnv a -> VarEnv a
plusVarEnv_C LocalOcc -> LocalOcc -> LocalOcc
andLocalOcc VarEnv LocalOcc
env VarEnv LocalOcc
join_env
      | Bool
otherwise                  = VarEnv LocalOcc
env

addJoinPoint :: OccEnv -> Id -> UsageDetails -> OccEnv
addJoinPoint :: OccEnv -> Id -> UsageDetails -> OccEnv
addJoinPoint OccEnv
env Id
bndr UsageDetails
rhs_uds
  | VarEnv LocalOcc -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv VarEnv LocalOcc
zeroed_form
  = OccEnv
env
  | Bool
otherwise
  = OccEnv
env { occ_join_points = extendVarEnv (occ_join_points env) bndr zeroed_form }
  where
    zeroed_form :: VarEnv LocalOcc
zeroed_form = UsageDetails -> VarEnv LocalOcc
mkZeroedForm UsageDetails
rhs_uds

mkZeroedForm :: UsageDetails -> OccInfoEnv
-- See Note [Occurrence analysis for join points] for "zeroed form"
mkZeroedForm :: UsageDetails -> VarEnv LocalOcc
mkZeroedForm (UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
rhs_occs })
  = (LocalOcc -> Maybe LocalOcc) -> VarEnv LocalOcc -> VarEnv LocalOcc
forall {k} elt1 elt2 (key :: k).
(elt1 -> Maybe elt2) -> UniqFM key elt1 -> UniqFM key elt2
mapMaybeUFM LocalOcc -> Maybe LocalOcc
do_one VarEnv LocalOcc
rhs_occs
  where
    do_one :: LocalOcc -> Maybe LocalOcc
    do_one :: LocalOcc -> Maybe LocalOcc
do_one (ManyOccL {})    = Maybe LocalOcc
forall a. Maybe a
Nothing
    do_one occ :: LocalOcc
occ@(OneOccL {}) = LocalOcc -> Maybe LocalOcc
forall a. a -> Maybe a
Just (LocalOcc
occ { lo_n_br = 0 })

--------------------
transClosureFV :: VarEnv VarSet -> VarEnv VarSet
-- If (f,g), (g,h) are in the input, then (f,h) is in the output
--                                   as well as (f,g), (g,h)
transClosureFV :: VarEnv VarSet -> VarEnv VarSet
transClosureFV VarEnv VarSet
env
  | Bool
no_change = VarEnv VarSet
env
  | Bool
otherwise = VarEnv VarSet -> VarEnv VarSet
transClosureFV ([(Unique, VarSet)] -> VarEnv VarSet
forall {k} elt (key :: k). [(Unique, elt)] -> UniqFM key elt
listToUFM_Directly [(Unique, VarSet)]
new_fv_list)
  where
    (Bool
no_change, [(Unique, VarSet)]
new_fv_list) = (Bool -> (Unique, VarSet) -> (Bool, (Unique, VarSet)))
-> Bool -> [(Unique, VarSet)] -> (Bool, [(Unique, VarSet)])
forall (t :: * -> *) s a b.
Traversable t =>
(s -> a -> (s, b)) -> s -> t a -> (s, t b)
mapAccumL Bool -> (Unique, VarSet) -> (Bool, (Unique, VarSet))
bump Bool
True (VarEnv VarSet -> [(Unique, VarSet)]
forall {k} (key :: k) elt. UniqFM key elt -> [(Unique, elt)]
nonDetUFMToList VarEnv VarSet
env)
      -- It's OK to use nonDetUFMToList here because we'll forget the
      -- ordering by creating a new set with listToUFM
    bump :: Bool -> (Unique, VarSet) -> (Bool, (Unique, VarSet))
bump Bool
no_change (Unique
b,VarSet
fvs)
      | Bool
no_change_here = (Bool
no_change, (Unique
b,VarSet
fvs))
      | Bool
otherwise      = (Bool
False,     (Unique
b,VarSet
new_fvs))
      where
        (VarSet
new_fvs, Bool
no_change_here) = VarEnv VarSet -> VarSet -> (VarSet, Bool)
extendFvs VarEnv VarSet
env VarSet
fvs

-------------
extendFvs_ :: VarEnv VarSet -> VarSet -> VarSet
extendFvs_ :: VarEnv VarSet -> VarSet -> VarSet
extendFvs_ VarEnv VarSet
env VarSet
s = (VarSet, Bool) -> VarSet
forall a b. (a, b) -> a
fst (VarEnv VarSet -> VarSet -> (VarSet, Bool)
extendFvs VarEnv VarSet
env VarSet
s)   -- Discard the Bool flag

extendFvs :: VarEnv VarSet -> VarSet -> (VarSet, Bool)
-- (extendFVs env s) returns
--     (s `union` env(s), env(s) `subset` s)
extendFvs :: VarEnv VarSet -> VarSet -> (VarSet, Bool)
extendFvs VarEnv VarSet
env VarSet
s
  | VarEnv VarSet -> Bool
forall {k} (key :: k) elt. UniqFM key elt -> Bool
isNullUFM VarEnv VarSet
env
  = (VarSet
s, Bool
True)
  | Bool
otherwise
  = (VarSet
s VarSet -> VarSet -> VarSet
`unionVarSet` VarSet
extras, VarSet
extras VarSet -> VarSet -> Bool
`subVarSet` VarSet
s)
  where
    extras :: VarSet    -- env(s)
    extras :: VarSet
extras = (VarSet -> VarSet -> VarSet) -> VarSet -> VarEnv VarSet -> VarSet
forall {k} elt a (key :: k).
(elt -> a -> a) -> a -> UniqFM key elt -> a
nonDetStrictFoldUFM VarSet -> VarSet -> VarSet
unionVarSet VarSet
emptyVarSet (VarEnv VarSet -> VarSet) -> VarEnv VarSet -> VarSet
forall a b. (a -> b) -> a -> b
$
      -- It's OK to use nonDetStrictFoldUFM here because unionVarSet commutes
             (VarSet -> Id -> VarSet)
-> VarEnv VarSet -> VarEnv Id -> VarEnv VarSet
forall {k} elt1 elt2 elt3 (key :: k).
(elt1 -> elt2 -> elt3)
-> UniqFM key elt1 -> UniqFM key elt2 -> UniqFM key elt3
intersectUFM_C (\VarSet
x Id
_ -> VarSet
x) VarEnv VarSet
env (VarSet -> VarEnv Id
forall a. UniqSet a -> UniqFM a a
getUniqSet VarSet
s)

{-
************************************************************************
*                                                                      *
                    Binder swap
*                                                                      *
************************************************************************

Note [Binder swap]
~~~~~~~~~~~~~~~~~~
The "binder swap" transformation swaps occurrence of the
scrutinee of a case for occurrences of the case-binder:

 (1)  case x of b { pi -> ri }
         ==>
      case x of b { pi -> ri[b/x] }

 (2)  case (x |> co) of b { pi -> ri }
        ==>
      case (x |> co) of b { pi -> ri[b |> sym co/x] }

The substitution ri[b/x] etc is done by the occurrence analyser.
See Note [The binder-swap substitution].

There are two reasons for making this swap:

(A) It reduces the number of occurrences of the scrutinee, x.
    That in turn might reduce its occurrences to one, so we
    can inline it and save an allocation.  E.g.
      let x = factorial y in case x of b { I# v -> ...x... }
    If we replace 'x' by 'b' in the alternative we get
      let x = factorial y in case x of b { I# v -> ...b... }
    and now we can inline 'x', thus
      case (factorial y) of b { I# v -> ...b... }

(B) The case-binder b has unfolding information; in the
    example above we know that b = I# v. That in turn allows
    nested cases to simplify.  Consider
       case x of b { I# v ->
       ...(case x of b2 { I# v2 -> rhs })...
    If we replace 'x' by 'b' in the alternative we get
       case x of b { I# v ->
       ...(case b of b2 { I# v2 -> rhs })...
    and now it is trivial to simplify the inner case:
       case x of b { I# v ->
       ...(let b2 = b in rhs)...

    The same can happen even if the scrutinee is a variable
    with a cast: see Note [Case of cast]

The reason for doing these transformations /here in the occurrence
analyser/ is because it allows us to adjust the OccInfo for 'x' and
'b' as we go.

  * Suppose the only occurrences of 'x' are the scrutinee and in the
    ri; then this transformation makes it occur just once, and hence
    get inlined right away.

  * If instead the Simplifier replaces occurrences of x with
    occurrences of b, that will mess up b's occurrence info. That in
    turn might have consequences.

There is a danger though.  Consider
      let v = x +# y
      in case (f v) of w -> ...v...v...
And suppose that (f v) expands to just v.  Then we'd like to
use 'w' instead of 'v' in the alternative.  But it may be too
late; we may have substituted the (cheap) x+#y for v in the
same simplifier pass that reduced (f v) to v.

I think this is just too bad.  CSE will recover some of it.

Note [The binder-swap substitution]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The binder-swap is implemented by the occ_bs_env field of OccEnv.
There are two main pieces:

* Given    case x |> co of b { alts }
  we add [x :-> (b, sym co)] to the occ_bs_env environment; this is
  done by addBndrSwap.

* Then, at an occurrence of a variable, we look up in the occ_bs_env
  to perform the swap. This is done by lookupBndrSwap.

Some tricky corners:

(BS1) We do the substitution before gathering occurrence info. So in
      the above example, an occurrence of x turns into an occurrence
      of b, and that's what we gather in the UsageDetails.  It's as
      if the binder-swap occurred before occurrence analysis. See
      the computation of fun_uds in occAnalApp.

(BS2) When doing a lookup in occ_bs_env, we may need to iterate,
      as you can see implemented in lookupBndrSwap.  Why?
      Consider   case x of a { 1# -> e1; DEFAULT ->
                 case x of b { 2# -> e2; DEFAULT ->
                 case x of c { 3# -> e3; DEFAULT -> ..x..a..b.. }}}
      At the first case addBndrSwap will extend occ_bs_env with
          [x :-> a]
      At the second case we occ-anal the scrutinee 'x', which looks up
        'x in occ_bs_env, returning 'a', as it should.
      Then addBndrSwap will add [a :-> b] to occ_bs_env, yielding
         occ_bs_env = [x :-> a, a :-> b]
      At the third case we'll again look up 'x' which returns 'a'.
      But we don't want to stop the lookup there, else we'll end up with
                 case x of a { 1# -> e1; DEFAULT ->
                 case a of b { 2# -> e2; DEFAULT ->
                 case a of c { 3# -> e3; DEFAULT -> ..a..b..c.. }}}
      Instead, we want iterate the lookup in addBndrSwap, to give
                 case x of a { 1# -> e1; DEFAULT ->
                 case a of b { 2# -> e2; DEFAULT ->
                 case b of c { 3# -> e3; DEFAULT -> ..c..c..c.. }}}
      This makes a particular difference for case-merge, which works
      only if the scrutinee is the case-binder of the immediately enclosing
      case (Note [Merge Nested Cases] in GHC.Core.Opt.Simplify.Utils
      See #19581 for the bug report that showed this up.

(BS3) We need care when shadowing.  Suppose [x :-> b] is in occ_bs_env,
      and we encounter:
         (i) \x. blah
             Here we want to delete the x-binding from occ_bs_env

         (ii) \b. blah
              This is harder: we really want to delete all bindings that
              have 'b' free in the range.  That is a bit tiresome to implement,
              so we compromise.  We keep occ_bs_rng, which is the set of
              free vars of rng(occc_bs_env).  If a binder shadows any of these
              variables, we discard all of occ_bs_env.  Safe, if a bit
              brutal.  NB, however: the simplifer de-shadows the code, so the
              next time around this won't happen.

      These checks are implemented in addInScope.
      (i) is needed only for Ids, but (ii) is needed for tyvars too (#22623)
      because if occ_bs_env has [x :-> ...a...] where `a` is a tyvar, we
      must not replace `x` by `...a...` under /\a. ...x..., or similarly
      under a case pattern match that binds `a`.

      An alternative would be for the occurrence analyser to do cloning as
      it goes.  In principle it could do so, but it'd make it a bit more
      complicated and there is no great benefit. The simplifer uses
      cloning to get a no-shadowing situation, the care-when-shadowing
      behaviour above isn't needed for long.

(BS4) The domain of occ_bs_env can include GlobaIds.  Eg
         case M.foo of b { alts }
      We extend occ_bs_env with [M.foo :-> b].  That's fine.

(BS5) We have to apply the occ_bs_env substitution uniformly,
      including to (local) rules and unfoldings.

(BS6) We must be very careful with dictionaries.
      See Note [Care with binder-swap on dictionaries]

Note [Case of cast]
~~~~~~~~~~~~~~~~~~~
Consider        case (x `cast` co) of b { I# ->
                ... (case (x `cast` co) of {...}) ...
We'd like to eliminate the inner case.  That is the motivation for
equation (2) in Note [Binder swap].  When we get to the inner case, we
inline x, cancel the casts, and away we go.

Note [Care with binder-swap on dictionaries]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This Note explains why we need isDictId in scrutOkForBinderSwap.
Consider this tricky example (#21229, #21470):

  class Sing (b :: Bool) where sing :: Bool
  instance Sing 'True  where sing = True
  instance Sing 'False where sing = False

  f :: forall a. Sing a => blah

  h = \ @(a :: Bool) ($dSing :: Sing a)
      let the_co =  Main.N:Sing[0] <a> :: Sing a ~R# Bool
      case ($dSing |> the_co) of wild
        True  -> f @'True (True |> sym the_co)
        False -> f @a     dSing

Now do a binder-swap on the case-expression:

  h = \ @(a :: Bool) ($dSing :: Sing a)
      let the_co =  Main.N:Sing[0] <a> :: Sing a ~R# Bool
      case ($dSing |> the_co) of wild
        True  -> f @'True (True |> sym the_co)
        False -> f @a     (wild |> sym the_co)

And now substitute `False` for `wild` (since wild=False in the False branch):

  h = \ @(a :: Bool) ($dSing :: Sing a)
      let the_co =  Main.N:Sing[0] <a> :: Sing a ~R# Bool
      case ($dSing |> the_co) of wild
        True  -> f @'True (True  |> sym the_co)
        False -> f @a     (False |> sym the_co)

And now we have a problem.  The specialiser will specialise (f @a d)a (for all
vtypes a and dictionaries d!!) with the dictionary (False |> sym the_co), using
Note [Specialising polymorphic dictionaries] in GHC.Core.Opt.Specialise.

The real problem is the binder-swap.  It swaps a dictionary variable $dSing
(of kind Constraint) for a term variable wild (of kind Type).  And that is
dangerous: a dictionary is a /singleton/ type whereas a general term variable is
not.  In this particular example, Bool is most certainly not a singleton type!

Conclusion:
  for a /dictionary variable/ do not perform
  the clever cast version of the binder-swap

Hence the subtle isDictId in scrutOkForBinderSwap.

Note [Zap case binders in proxy bindings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From the original
     case x of cb(dead) { p -> ...x... }
we will get
     case x of cb(live) { p -> ...cb... }

Core Lint never expects to find an *occurrence* of an Id marked
as Dead, so we must zap the OccInfo on cb before making the
binding x = cb.  See #5028.

NB: the OccInfo on /occurrences/ really doesn't matter much; the simplifier
doesn't use it. So this is only to satisfy the perhaps-over-picky Lint.

-}

addBndrSwap :: OutExpr -> Id -> OccEnv -> OccEnv
-- See Note [The binder-swap substitution]
addBndrSwap :: CoreExpr -> Id -> OccEnv -> OccEnv
addBndrSwap CoreExpr
scrut Id
case_bndr
            env :: OccEnv
env@(OccEnv { occ_bs_env :: OccEnv -> IdEnv (Id, MCoercion)
occ_bs_env = IdEnv (Id, MCoercion)
swap_env, occ_bs_rng :: OccEnv -> VarSet
occ_bs_rng = VarSet
rng_vars })
  | DoBinderSwap Id
scrut_var MCoercion
mco <- CoreExpr -> BinderSwapDecision
scrutOkForBinderSwap CoreExpr
scrut
  , Id
scrut_var Id -> Id -> Bool
forall a. Eq a => a -> a -> Bool
/= Id
case_bndr
      -- Consider: case x of x { ... }
      -- Do not add [x :-> x] to occ_bs_env, else lookupBndrSwap will loop
  = OccEnv
env { occ_bs_env = extendVarEnv swap_env scrut_var (case_bndr', mco)
        , occ_bs_rng = rng_vars `extendVarSet` case_bndr'
                       `unionVarSet` tyCoVarsOfMCo mco }

  | Bool
otherwise
  = OccEnv
env
  where
    case_bndr' :: Id
case_bndr' = Id -> Id
zapIdOccInfo Id
case_bndr
                 -- See Note [Zap case binders in proxy bindings]

-- | See bBinderSwaOk.
data BinderSwapDecision
  = NoBinderSwap
  | DoBinderSwap OutVar MCoercion

scrutOkForBinderSwap :: OutExpr -> BinderSwapDecision
-- If (scrutOkForBinderSwap e = DoBinderSwap v mco, then
--    v = e |> mco
-- See Note [Case of cast]
-- See Note [Care with binder-swap on dictionaries]
--
-- We use this same function in SpecConstr, and Simplify.Iteration,
-- when something binder-swap-like is happening
scrutOkForBinderSwap :: CoreExpr -> BinderSwapDecision
scrutOkForBinderSwap (Var Id
v)    = Id -> MCoercion -> BinderSwapDecision
DoBinderSwap Id
v MCoercion
MRefl
scrutOkForBinderSwap (Cast (Var Id
v) CoercionR
co)
  | Bool -> Bool
not (Id -> Bool
isDictId Id
v)             = Id -> MCoercion -> BinderSwapDecision
DoBinderSwap Id
v (CoercionR -> MCoercion
MCo (CoercionR -> CoercionR
mkSymCo CoercionR
co))
        -- Cast: see Note [Case of cast]
        -- isDictId: see Note [Care with binder-swap on dictionaries]
        -- The isDictId rejects a Constraint/Constraint binder-swap, perhaps
        -- over-conservatively. But I have never seen one, so I'm leaving
        -- the code as simple as possible. Losing the binder-swap in a
        -- rare case probably has very low impact.
scrutOkForBinderSwap (Tick CoreTickish
_ CoreExpr
e) = CoreExpr -> BinderSwapDecision
scrutOkForBinderSwap CoreExpr
e  -- Drop ticks
scrutOkForBinderSwap CoreExpr
_          = BinderSwapDecision
NoBinderSwap

lookupBndrSwap :: OccEnv -> Id -> (CoreExpr, Id)
-- See Note [The binder-swap substitution]
-- Returns an expression of the same type as Id
lookupBndrSwap :: OccEnv -> Id -> (CoreExpr, Id)
lookupBndrSwap env :: OccEnv
env@(OccEnv { occ_bs_env :: OccEnv -> IdEnv (Id, MCoercion)
occ_bs_env = IdEnv (Id, MCoercion)
bs_env })  Id
bndr
  = case IdEnv (Id, MCoercion) -> Id -> Maybe (Id, MCoercion)
forall a. VarEnv a -> Id -> Maybe a
lookupVarEnv IdEnv (Id, MCoercion)
bs_env Id
bndr of {
       Maybe (Id, MCoercion)
Nothing           -> (Id -> CoreExpr
forall b. Id -> Expr b
Var Id
bndr, Id
bndr) ;
       Just (Id
bndr1, MCoercion
mco) ->

    -- Why do we iterate here?
    -- See (BS2) in Note [The binder-swap substitution]
    case OccEnv -> Id -> (CoreExpr, Id)
lookupBndrSwap OccEnv
env Id
bndr1 of
      (CoreExpr
fun, Id
fun_id) -> (CoreExpr -> MCoercion -> CoreExpr
mkCastMCo CoreExpr
fun MCoercion
mco, Id
fun_id) }

{- Historical note [Proxy let-bindings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We used to do the binder-swap transformation by introducing
a proxy let-binding, thus;

   case x of b { pi -> ri }
      ==>
   case x of b { pi -> let x = b in ri }

But that had two problems:

1. If 'x' is an imported GlobalId, we'd end up with a GlobalId
   on the LHS of a let-binding which isn't allowed.  We worked
   around this for a while by "localising" x, but it turned
   out to be very painful #16296,

2. In CorePrep we use the occurrence analyser to do dead-code
   elimination (see Note [Dead code in CorePrep]).  But that
   occasionally led to an unlifted let-binding
       case x of b { DEFAULT -> let x::Int# = b in ... }
   which disobeys one of CorePrep's output invariants (no unlifted
   let-bindings) -- see #5433.

Doing a substitution (via occ_bs_env) is much better.

Historical Note [no-case-of-case]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
We *used* to suppress the binder-swap in case expressions when
-fno-case-of-case is on.  Old remarks:
    "This happens in the first simplifier pass,
    and enhances full laziness.  Here's the bad case:
            f = \ y -> ...(case x of I# v -> ...(case x of ...) ... )
    If we eliminate the inner case, we trap it inside the I# v -> arm,
    which might prevent some full laziness happening.  I've seen this
    in action in spectral/cichelli/Prog.hs:
             [(m,n) | m <- [1..max], n <- [1..max]]
    Hence the check for NoCaseOfCase."
However, now the full-laziness pass itself reverses the binder-swap, so this
check is no longer necessary.

Historical Note [Suppressing the case binder-swap]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This old note describes a problem that is also fixed by doing the
binder-swap in OccAnal:

    There is another situation when it might make sense to suppress the
    case-expression binde-swap. If we have

        case x of w1 { DEFAULT -> case x of w2 { A -> e1; B -> e2 }
                       ...other cases .... }

    We'll perform the binder-swap for the outer case, giving

        case x of w1 { DEFAULT -> case w1 of w2 { A -> e1; B -> e2 }
                       ...other cases .... }

    But there is no point in doing it for the inner case, because w1 can't
    be inlined anyway.  Furthermore, doing the case-swapping involves
    zapping w2's occurrence info (see paragraphs that follow), and that
    forces us to bind w2 when doing case merging.  So we get

        case x of w1 { A -> let w2 = w1 in e1
                       B -> let w2 = w1 in e2
                       ...other cases .... }

    This is plain silly in the common case where w2 is dead.

    Even so, I can't see a good way to implement this idea.  I tried
    not doing the binder-swap if the scrutinee was already evaluated
    but that failed big-time:

            data T = MkT !Int

            case v of w  { MkT x ->
            case x of x1 { I# y1 ->
            case x of x2 { I# y2 -> ...

    Notice that because MkT is strict, x is marked "evaluated".  But to
    eliminate the last case, we must either make sure that x (as well as
    x1) has unfolding MkT y1.  The straightforward thing to do is to do
    the binder-swap.  So this whole note is a no-op.

It's fixed by doing the binder-swap in OccAnal because we can do the
binder-swap unconditionally and still get occurrence analysis
information right.


************************************************************************
*                                                                      *
\subsection[OccurAnal-types]{OccEnv}
*                                                                      *
************************************************************************

Note [UsageDetails and zapping]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
On many occasions, we must modify all gathered occurrence data at once. For
instance, all occurrences underneath a (non-one-shot) lambda set the
'occ_in_lam' flag to become 'True'. We could use 'mapVarEnv' to do this, but
that takes O(n) time and we will do this often---in particular, there are many
places where tail calls are not allowed, and each of these causes all variables
to get marked with 'NoTailCallInfo'.

Instead of relying on `mapVarEnv`, then, we carry three 'IdEnv's around along
with the 'OccInfoEnv'. Each of these extra environments is a "zapped set"
recording which variables have been zapped in some way. Zapping all occurrence
info then simply means setting the corresponding zapped set to the whole
'OccInfoEnv', a fast O(1) operation.

Note [LocalOcc]
~~~~~~~~~~~~~~~
LocalOcc is used purely internally, in the occurrence analyser.  It differs from
GHC.Types.Basic.OccInfo because it has only OneOcc and ManyOcc; it does not need
IAmDead or IAmALoopBreaker.

Note that `OneOccL` doesn't meant that it occurs /syntactially/ only once; it
means that it is /used/ only once. It might occur syntactically many times.
For example, in (case x of A -> y; B -> y; C -> True),
* `y` is used only once
* but it occurs syntactically twice

-}

type OccInfoEnv = IdEnv LocalOcc  -- A finite map from an expression's
                                  -- free variables to their usage

data LocalOcc  -- See Note [LocalOcc]
     = OneOccL { LocalOcc -> JoinArity
lo_n_br  :: {-# UNPACK #-} !BranchCount  -- Number of syntactic occurrences
               , LocalOcc -> TailCallInfo
lo_tail  :: !TailCallInfo
                   -- Combining (AlwaysTailCalled 2) and (AlwaysTailCalled 3)
                   -- gives NoTailCallInfo
              , LocalOcc -> InterestingCxt
lo_int_cxt :: !InterestingCxt }
    | ManyOccL !TailCallInfo

instance Outputable LocalOcc where
  ppr :: LocalOcc -> SDoc
ppr (OneOccL { lo_n_br :: LocalOcc -> JoinArity
lo_n_br = JoinArity
n, lo_tail :: LocalOcc -> TailCallInfo
lo_tail = TailCallInfo
tci })
    = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"OneOccL" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc
braces (JoinArity -> SDoc
forall a. Outputable a => a -> SDoc
ppr JoinArity
n SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> SDoc
forall doc. IsLine doc => doc
comma SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> TailCallInfo -> SDoc
forall a. Outputable a => a -> SDoc
ppr TailCallInfo
tci)
  ppr (ManyOccL TailCallInfo
tci) = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"ManyOccL" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc
braces (TailCallInfo -> SDoc
forall a. Outputable a => a -> SDoc
ppr TailCallInfo
tci)

localTailCallInfo :: LocalOcc -> TailCallInfo
localTailCallInfo :: LocalOcc -> TailCallInfo
localTailCallInfo (OneOccL  { lo_tail :: LocalOcc -> TailCallInfo
lo_tail = TailCallInfo
tci }) = TailCallInfo
tci
localTailCallInfo (ManyOccL TailCallInfo
tci)               = TailCallInfo
tci

type ZappedSet = OccInfoEnv -- Values are ignored

data UsageDetails
  = UD { UsageDetails -> VarEnv LocalOcc
ud_env       :: !OccInfoEnv
       , UsageDetails -> VarEnv LocalOcc
ud_z_many    :: !ZappedSet   -- apply 'markMany' to these
       , UsageDetails -> VarEnv LocalOcc
ud_z_in_lam  :: !ZappedSet   -- apply 'markInsideLam' to these
       , UsageDetails -> VarEnv LocalOcc
ud_z_tail    :: !ZappedSet   -- zap tail-call info for these
       }
  -- INVARIANT: All three zapped sets are subsets of ud_env

instance Outputable UsageDetails where
  ppr :: UsageDetails -> SDoc
ppr ud :: UsageDetails
ud@(UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env, ud_z_tail :: UsageDetails -> VarEnv LocalOcc
ud_z_tail = VarEnv LocalOcc
z_tail })
    = String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"UD" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> (SDoc -> SDoc
forall doc. IsLine doc => doc -> doc
braces (SDoc -> SDoc) -> SDoc -> SDoc
forall a b. (a -> b) -> a -> b
$ [SDoc] -> SDoc
forall doc. IsLine doc => [doc] -> doc
fsep ([SDoc] -> SDoc) -> [SDoc] -> SDoc
forall a b. (a -> b) -> a -> b
$ SDoc -> [SDoc] -> [SDoc]
forall doc. IsLine doc => doc -> [doc] -> [doc]
punctuate SDoc
forall doc. IsLine doc => doc
comma ([SDoc] -> [SDoc]) -> [SDoc] -> [SDoc]
forall a b. (a -> b) -> a -> b
$
      [ Unique -> SDoc
forall a. Outputable a => a -> SDoc
ppr Unique
uq SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> String -> SDoc
forall doc. IsLine doc => String -> doc
text String
":->" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> OccInfo -> SDoc
forall a. Outputable a => a -> SDoc
ppr (UsageDetails -> Unique -> OccInfo
lookupOccInfoByUnique UsageDetails
ud Unique
uq)
      | (Unique
uq, LocalOcc
_) <- (Unique
 -> LocalOcc -> [(Unique, LocalOcc)] -> [(Unique, LocalOcc)])
-> [(Unique, LocalOcc)] -> VarEnv LocalOcc -> [(Unique, LocalOcc)]
forall a r. (Unique -> a -> r -> r) -> r -> VarEnv a -> r
nonDetStrictFoldVarEnv_Directly Unique -> LocalOcc -> [(Unique, LocalOcc)] -> [(Unique, LocalOcc)]
do_one [] VarEnv LocalOcc
env ])
      SDoc -> SDoc -> SDoc
forall doc. IsDoc doc => doc -> doc -> doc
$$ JoinArity -> SDoc -> SDoc
nest JoinArity
2 (String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"ud_z_tail" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> VarEnv LocalOcc -> SDoc
forall a. Outputable a => a -> SDoc
ppr VarEnv LocalOcc
z_tail)
    where
      do_one :: Unique -> LocalOcc -> [(Unique,LocalOcc)] -> [(Unique,LocalOcc)]
      do_one :: Unique -> LocalOcc -> [(Unique, LocalOcc)] -> [(Unique, LocalOcc)]
do_one Unique
uniq LocalOcc
occ [(Unique, LocalOcc)]
occs = (Unique
uniq, LocalOcc
occ) (Unique, LocalOcc) -> [(Unique, LocalOcc)] -> [(Unique, LocalOcc)]
forall a. a -> [a] -> [a]
: [(Unique, LocalOcc)]
occs

---------------------
-- | TailUsageDetails captures the result of applying 'occAnalLamTail'
--   to a function `\xyz.body`. The TailUsageDetails pairs together
--   * the number of lambdas (including type lambdas: a JoinArity)
--   * UsageDetails for the `body` of the lambda, unadjusted by `adjustTailUsage`.
-- If the binding turns out to be a join point with the indicated join
-- arity, this unadjusted usage details is just what we need; otherwise we
-- need to discard tail calls. That's what `adjustTailUsage` does.
data TailUsageDetails = TUD !JoinArity !UsageDetails

instance Outputable TailUsageDetails where
  ppr :: TailUsageDetails -> SDoc
ppr (TUD JoinArity
ja UsageDetails
uds) = SDoc
lambda SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> JoinArity -> SDoc
forall a. Outputable a => a -> SDoc
ppr JoinArity
ja SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<> UsageDetails -> SDoc
forall a. Outputable a => a -> SDoc
ppr UsageDetails
uds

---------------------
data WithUsageDetails     a = WUD  !UsageDetails     !a
data WithTailUsageDetails a = WTUD !TailUsageDetails !a

-------------------
-- UsageDetails API

andUDs, orUDs
        :: UsageDetails -> UsageDetails -> UsageDetails
andUDs :: UsageDetails -> UsageDetails -> UsageDetails
andUDs = (LocalOcc -> LocalOcc -> LocalOcc)
-> UsageDetails -> UsageDetails -> UsageDetails
combineUsageDetailsWith LocalOcc -> LocalOcc -> LocalOcc
andLocalOcc
orUDs :: UsageDetails -> UsageDetails -> UsageDetails
orUDs  = (LocalOcc -> LocalOcc -> LocalOcc)
-> UsageDetails -> UsageDetails -> UsageDetails
combineUsageDetailsWith LocalOcc -> LocalOcc -> LocalOcc
orLocalOcc

mkOneOcc :: OccEnv -> Id -> InterestingCxt -> JoinArity -> UsageDetails
mkOneOcc :: OccEnv -> Id -> InterestingCxt -> JoinArity -> UsageDetails
mkOneOcc !OccEnv
env Id
id InterestingCxt
int_cxt JoinArity
arity
  | Bool -> Bool
not (Id -> Bool
isLocalId Id
id)
  = UsageDetails
emptyDetails

  | Just VarEnv LocalOcc
join_uds <- JoinPointInfo -> Id -> Maybe (VarEnv LocalOcc)
forall a. VarEnv a -> Id -> Maybe a
lookupVarEnv (OccEnv -> JoinPointInfo
occ_join_points OccEnv
env) Id
id
  = -- See Note [Occurrence analysis for join points]
    Bool -> SDoc -> UsageDetails -> UsageDetails
forall a. HasCallStack => Bool -> SDoc -> a -> a
assertPpr (Bool -> Bool
not (VarEnv LocalOcc -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv VarEnv LocalOcc
join_uds)) (Id -> SDoc
forall a. Outputable a => a -> SDoc
ppr Id
id) (UsageDetails -> UsageDetails) -> UsageDetails -> UsageDetails
forall a b. (a -> b) -> a -> b
$
       -- We only put non-empty join-points into occ_join_points
    VarEnv LocalOcc -> UsageDetails
mkSimpleDetails (VarEnv LocalOcc -> Id -> LocalOcc -> VarEnv LocalOcc
forall a. VarEnv a -> Id -> a -> VarEnv a
extendVarEnv VarEnv LocalOcc
join_uds Id
id LocalOcc
occ)

  | Bool
otherwise
  = VarEnv LocalOcc -> UsageDetails
mkSimpleDetails (Id -> LocalOcc -> VarEnv LocalOcc
forall a. Id -> a -> VarEnv a
unitVarEnv Id
id LocalOcc
occ)

  where
    occ :: LocalOcc
occ = OneOccL { lo_n_br :: JoinArity
lo_n_br = JoinArity
1, lo_int_cxt :: InterestingCxt
lo_int_cxt = InterestingCxt
int_cxt
                  , lo_tail :: TailCallInfo
lo_tail = JoinArity -> TailCallInfo
AlwaysTailCalled JoinArity
arity }

-- Add several occurrences, assumed not to be tail calls
add_many_occ :: Var -> OccInfoEnv -> OccInfoEnv
add_many_occ :: Id -> VarEnv LocalOcc -> VarEnv LocalOcc
add_many_occ Id
v VarEnv LocalOcc
env | Id -> Bool
isId Id
v    = VarEnv LocalOcc -> Id -> LocalOcc -> VarEnv LocalOcc
forall a. VarEnv a -> Id -> a -> VarEnv a
extendVarEnv VarEnv LocalOcc
env Id
v (TailCallInfo -> LocalOcc
ManyOccL TailCallInfo
NoTailCallInfo)
                   | Bool
otherwise = VarEnv LocalOcc
env
        -- Give a non-committal binder info (i.e noOccInfo) because
        --   a) Many copies of the specialised thing can appear
        --   b) We don't want to substitute a BIG expression inside a RULE
        --      even if that's the only occurrence of the thing
        --      (Same goes for INLINE.)

addManyOccs :: UsageDetails -> VarSet -> UsageDetails
addManyOccs :: UsageDetails -> VarSet -> UsageDetails
addManyOccs UsageDetails
uds VarSet
var_set
  | VarSet -> Bool
isEmptyVarSet VarSet
var_set = UsageDetails
uds
  | Bool
otherwise             = UsageDetails
uds { ud_env = add_to (ud_env uds) }
  where
    add_to :: VarEnv LocalOcc -> VarEnv LocalOcc
add_to VarEnv LocalOcc
env = (Id -> VarEnv LocalOcc -> VarEnv LocalOcc)
-> VarEnv LocalOcc -> VarSet -> VarEnv LocalOcc
forall elt a. (elt -> a -> a) -> a -> UniqSet elt -> a
nonDetStrictFoldUniqSet Id -> VarEnv LocalOcc -> VarEnv LocalOcc
add_many_occ VarEnv LocalOcc
env VarSet
var_set
    -- It's OK to use nonDetStrictFoldUniqSet here because add_many_occ commutes

addLamCoVarOccs :: UsageDetails -> [Var] -> UsageDetails
-- Add any CoVars free in the type of a lambda-binder
-- See Note [Gather occurrences of coercion variables]
addLamCoVarOccs :: UsageDetails -> [Id] -> UsageDetails
addLamCoVarOccs UsageDetails
uds [Id]
bndrs
  = (Id -> UsageDetails -> UsageDetails)
-> UsageDetails -> [Id] -> UsageDetails
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr Id -> UsageDetails -> UsageDetails
add UsageDetails
uds [Id]
bndrs
  where
    add :: Id -> UsageDetails -> UsageDetails
add Id
bndr UsageDetails
uds = UsageDetails
uds UsageDetails -> VarSet -> UsageDetails
`addManyOccs` Type -> VarSet
coVarsOfType (Id -> Type
varType Id
bndr)

emptyDetails :: UsageDetails
emptyDetails :: UsageDetails
emptyDetails = VarEnv LocalOcc -> UsageDetails
mkSimpleDetails VarEnv LocalOcc
forall a. VarEnv a
emptyVarEnv

isEmptyDetails :: UsageDetails -> Bool
isEmptyDetails :: UsageDetails -> Bool
isEmptyDetails (UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env }) = VarEnv LocalOcc -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv VarEnv LocalOcc
env

mkSimpleDetails :: OccInfoEnv -> UsageDetails
mkSimpleDetails :: VarEnv LocalOcc -> UsageDetails
mkSimpleDetails VarEnv LocalOcc
env = UD { ud_env :: VarEnv LocalOcc
ud_env       = VarEnv LocalOcc
env
                         , ud_z_many :: VarEnv LocalOcc
ud_z_many    = VarEnv LocalOcc
forall a. VarEnv a
emptyVarEnv
                         , ud_z_in_lam :: VarEnv LocalOcc
ud_z_in_lam  = VarEnv LocalOcc
forall a. VarEnv a
emptyVarEnv
                         , ud_z_tail :: VarEnv LocalOcc
ud_z_tail    = VarEnv LocalOcc
forall a. VarEnv a
emptyVarEnv }

modifyUDEnv :: (OccInfoEnv -> OccInfoEnv) -> UsageDetails -> UsageDetails
modifyUDEnv :: (VarEnv LocalOcc -> VarEnv LocalOcc)
-> UsageDetails -> UsageDetails
modifyUDEnv VarEnv LocalOcc -> VarEnv LocalOcc
f uds :: UsageDetails
uds@(UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env }) = UsageDetails
uds { ud_env = f env }

delBndrsFromUDs :: [Var] -> UsageDetails -> UsageDetails
-- Delete these binders from the UsageDetails
delBndrsFromUDs :: [Id] -> UsageDetails -> UsageDetails
delBndrsFromUDs [Id]
bndrs (UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env, ud_z_many :: UsageDetails -> VarEnv LocalOcc
ud_z_many = VarEnv LocalOcc
z_many
                          , ud_z_in_lam :: UsageDetails -> VarEnv LocalOcc
ud_z_in_lam  = VarEnv LocalOcc
z_in_lam, ud_z_tail :: UsageDetails -> VarEnv LocalOcc
ud_z_tail = VarEnv LocalOcc
z_tail })
  = UD { ud_env :: VarEnv LocalOcc
ud_env       = VarEnv LocalOcc
env      VarEnv LocalOcc -> [Id] -> VarEnv LocalOcc
forall a. VarEnv a -> [Id] -> VarEnv a
`delVarEnvList` [Id]
bndrs
       , ud_z_many :: VarEnv LocalOcc
ud_z_many    = VarEnv LocalOcc
z_many   VarEnv LocalOcc -> [Id] -> VarEnv LocalOcc
forall a. VarEnv a -> [Id] -> VarEnv a
`delVarEnvList` [Id]
bndrs
       , ud_z_in_lam :: VarEnv LocalOcc
ud_z_in_lam  = VarEnv LocalOcc
z_in_lam VarEnv LocalOcc -> [Id] -> VarEnv LocalOcc
forall a. VarEnv a -> [Id] -> VarEnv a
`delVarEnvList` [Id]
bndrs
       , ud_z_tail :: VarEnv LocalOcc
ud_z_tail    = VarEnv LocalOcc
z_tail   VarEnv LocalOcc -> [Id] -> VarEnv LocalOcc
forall a. VarEnv a -> [Id] -> VarEnv a
`delVarEnvList` [Id]
bndrs }

markAllMany, markAllInsideLam, markAllNonTail, markAllManyNonTail
  :: UsageDetails -> UsageDetails
markAllMany :: UsageDetails -> UsageDetails
markAllMany      ud :: UsageDetails
ud@(UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env }) = UsageDetails
ud { ud_z_many   = env }
markAllInsideLam :: UsageDetails -> UsageDetails
markAllInsideLam ud :: UsageDetails
ud@(UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env }) = UsageDetails
ud { ud_z_in_lam = env }
markAllNonTail :: UsageDetails -> UsageDetails
markAllNonTail   ud :: UsageDetails
ud@(UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env }) = UsageDetails
ud { ud_z_tail   = env }
markAllManyNonTail :: UsageDetails -> UsageDetails
markAllManyNonTail = UsageDetails -> UsageDetails
markAllMany (UsageDetails -> UsageDetails)
-> (UsageDetails -> UsageDetails) -> UsageDetails -> UsageDetails
forall b c a. (b -> c) -> (a -> b) -> a -> c
. UsageDetails -> UsageDetails
markAllNonTail -- effectively sets to noOccInfo

markAllInsideLamIf, markAllNonTailIf :: Bool -> UsageDetails -> UsageDetails

markAllInsideLamIf :: Bool -> UsageDetails -> UsageDetails
markAllInsideLamIf  Bool
True  UsageDetails
ud = UsageDetails -> UsageDetails
markAllInsideLam UsageDetails
ud
markAllInsideLamIf  Bool
False UsageDetails
ud = UsageDetails
ud

markAllNonTailIf :: Bool -> UsageDetails -> UsageDetails
markAllNonTailIf Bool
True  UsageDetails
ud = UsageDetails -> UsageDetails
markAllNonTail UsageDetails
ud
markAllNonTailIf Bool
False UsageDetails
ud = UsageDetails
ud

lookupTailCallInfo :: UsageDetails -> Id -> TailCallInfo
lookupTailCallInfo :: UsageDetails -> Id -> TailCallInfo
lookupTailCallInfo UsageDetails
uds Id
id
  | UD { ud_z_tail :: UsageDetails -> VarEnv LocalOcc
ud_z_tail = VarEnv LocalOcc
z_tail, ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env } <- UsageDetails
uds
  , Bool -> Bool
not (Id
id Id -> VarEnv LocalOcc -> Bool
forall a. Id -> VarEnv a -> Bool
`elemVarEnv` VarEnv LocalOcc
z_tail)
  , Just LocalOcc
occ <- VarEnv LocalOcc -> Id -> Maybe LocalOcc
forall a. VarEnv a -> Id -> Maybe a
lookupVarEnv VarEnv LocalOcc
env Id
id
  = LocalOcc -> TailCallInfo
localTailCallInfo LocalOcc
occ
  | Bool
otherwise
  = TailCallInfo
NoTailCallInfo

udFreeVars :: VarSet -> UsageDetails -> VarSet
-- Find the subset of bndrs that are mentioned in uds
udFreeVars :: VarSet -> UsageDetails -> VarSet
udFreeVars VarSet
bndrs (UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env }) = VarSet -> VarEnv LocalOcc -> VarSet
restrictFreeVars VarSet
bndrs VarEnv LocalOcc
env

restrictFreeVars :: VarSet -> OccInfoEnv -> VarSet
restrictFreeVars :: VarSet -> VarEnv LocalOcc -> VarSet
restrictFreeVars VarSet
bndrs VarEnv LocalOcc
fvs = VarSet -> VarEnv LocalOcc -> VarSet
forall key b. UniqSet key -> UniqFM key b -> UniqSet key
restrictUniqSetToUFM VarSet
bndrs VarEnv LocalOcc
fvs

-------------------
-- Auxiliary functions for UsageDetails implementation

combineUsageDetailsWith :: (LocalOcc -> LocalOcc -> LocalOcc)
                        -> UsageDetails -> UsageDetails -> UsageDetails
{-# INLINE combineUsageDetailsWith #-}
combineUsageDetailsWith :: (LocalOcc -> LocalOcc -> LocalOcc)
-> UsageDetails -> UsageDetails -> UsageDetails
combineUsageDetailsWith LocalOcc -> LocalOcc -> LocalOcc
plus_occ_info
    uds1 :: UsageDetails
uds1@(UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env1, ud_z_many :: UsageDetails -> VarEnv LocalOcc
ud_z_many = VarEnv LocalOcc
z_many1, ud_z_in_lam :: UsageDetails -> VarEnv LocalOcc
ud_z_in_lam = VarEnv LocalOcc
z_in_lam1, ud_z_tail :: UsageDetails -> VarEnv LocalOcc
ud_z_tail = VarEnv LocalOcc
z_tail1 })
    uds2 :: UsageDetails
uds2@(UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env = VarEnv LocalOcc
env2, ud_z_many :: UsageDetails -> VarEnv LocalOcc
ud_z_many = VarEnv LocalOcc
z_many2, ud_z_in_lam :: UsageDetails -> VarEnv LocalOcc
ud_z_in_lam = VarEnv LocalOcc
z_in_lam2, ud_z_tail :: UsageDetails -> VarEnv LocalOcc
ud_z_tail = VarEnv LocalOcc
z_tail2 })
  | VarEnv LocalOcc -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv VarEnv LocalOcc
env1 = UsageDetails
uds2
  | VarEnv LocalOcc -> Bool
forall a. VarEnv a -> Bool
isEmptyVarEnv VarEnv LocalOcc
env2 = UsageDetails
uds1
  | Bool
otherwise
  = UD { ud_env :: VarEnv LocalOcc
ud_env       = (LocalOcc -> LocalOcc -> LocalOcc)
-> VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc
forall a. (a -> a -> a) -> VarEnv a -> VarEnv a -> VarEnv a
plusVarEnv_C LocalOcc -> LocalOcc -> LocalOcc
plus_occ_info VarEnv LocalOcc
env1 VarEnv LocalOcc
env2
       , ud_z_many :: VarEnv LocalOcc
ud_z_many    = VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc
forall a. VarEnv a -> VarEnv a -> VarEnv a
plusVarEnv VarEnv LocalOcc
z_many1   VarEnv LocalOcc
z_many2
       , ud_z_in_lam :: VarEnv LocalOcc
ud_z_in_lam  = VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc
forall a. VarEnv a -> VarEnv a -> VarEnv a
plusVarEnv VarEnv LocalOcc
z_in_lam1 VarEnv LocalOcc
z_in_lam2
       , ud_z_tail :: VarEnv LocalOcc
ud_z_tail    = VarEnv LocalOcc -> VarEnv LocalOcc -> VarEnv LocalOcc
forall a. VarEnv a -> VarEnv a -> VarEnv a
plusVarEnv VarEnv LocalOcc
z_tail1   VarEnv LocalOcc
z_tail2 }

lookupLetOccInfo :: UsageDetails -> Id -> OccInfo
-- Don't use locally-generated occ_info for exported (visible-elsewhere)
-- things.  Instead just give noOccInfo.
-- NB: setBinderOcc will (rightly) erase any LoopBreaker info;
--     we are about to re-generate it and it shouldn't be "sticky"
lookupLetOccInfo :: UsageDetails -> Id -> OccInfo
lookupLetOccInfo UsageDetails
ud Id
id
 | Id -> Bool
isExportedId Id
id = OccInfo
noOccInfo
 | Bool
otherwise       = UsageDetails -> Unique -> OccInfo
lookupOccInfoByUnique UsageDetails
ud (Id -> Unique
idUnique Id
id)

lookupOccInfo :: UsageDetails -> Id -> OccInfo
lookupOccInfo :: UsageDetails -> Id -> OccInfo
lookupOccInfo UsageDetails
ud Id
id = UsageDetails -> Unique -> OccInfo
lookupOccInfoByUnique UsageDetails
ud (Id -> Unique
idUnique Id
id)

lookupOccInfoByUnique :: UsageDetails -> Unique -> OccInfo
lookupOccInfoByUnique :: UsageDetails -> Unique -> OccInfo
lookupOccInfoByUnique (UD { ud_env :: UsageDetails -> VarEnv LocalOcc
ud_env       = VarEnv LocalOcc
env
                          , ud_z_many :: UsageDetails -> VarEnv LocalOcc
ud_z_many    = VarEnv LocalOcc
z_many
                          , ud_z_in_lam :: UsageDetails -> VarEnv LocalOcc
ud_z_in_lam  = VarEnv LocalOcc
z_in_lam
                          , ud_z_tail :: UsageDetails -> VarEnv LocalOcc
ud_z_tail    = VarEnv LocalOcc
z_tail })
                  Unique
uniq
  = case VarEnv LocalOcc -> Unique -> Maybe LocalOcc
forall a. VarEnv a -> Unique -> Maybe a
lookupVarEnv_Directly VarEnv LocalOcc
env Unique
uniq of
      Maybe LocalOcc
Nothing -> OccInfo
IAmDead
      Just (OneOccL { lo_n_br :: LocalOcc -> JoinArity
lo_n_br = JoinArity
n_br, lo_int_cxt :: LocalOcc -> InterestingCxt
lo_int_cxt = InterestingCxt
int_cxt
                    , lo_tail :: LocalOcc -> TailCallInfo
lo_tail = TailCallInfo
tail_info })
          | Unique
uniq Unique -> VarEnv LocalOcc -> Bool
forall a. Unique -> VarEnv a -> Bool
`elemVarEnvByKey`VarEnv LocalOcc
z_many
          -> ManyOccs { occ_tail :: TailCallInfo
occ_tail = TailCallInfo -> TailCallInfo
mk_tail_info TailCallInfo
tail_info }
          | Bool
otherwise
          -> OneOcc { occ_in_lam :: InsideLam
occ_in_lam  = InsideLam
in_lam
                    , occ_n_br :: JoinArity
occ_n_br    = JoinArity
n_br
                    , occ_int_cxt :: InterestingCxt
occ_int_cxt = InterestingCxt
int_cxt
                    , occ_tail :: TailCallInfo
occ_tail    = TailCallInfo -> TailCallInfo
mk_tail_info TailCallInfo
tail_info }
         where
           in_lam :: InsideLam
in_lam | Unique
uniq Unique -> VarEnv LocalOcc -> Bool
forall a. Unique -> VarEnv a -> Bool
`elemVarEnvByKey` VarEnv LocalOcc
z_in_lam = InsideLam
IsInsideLam
                  | Bool
otherwise                       = InsideLam
NotInsideLam

      Just (ManyOccL TailCallInfo
tail_info) -> ManyOccs { occ_tail :: TailCallInfo
occ_tail = TailCallInfo -> TailCallInfo
mk_tail_info TailCallInfo
tail_info }
  where
    mk_tail_info :: TailCallInfo -> TailCallInfo
mk_tail_info TailCallInfo
ti
        | Unique
uniq Unique -> VarEnv LocalOcc -> Bool
forall a. Unique -> VarEnv a -> Bool
`elemVarEnvByKey` VarEnv LocalOcc
z_tail = TailCallInfo
NoTailCallInfo
        | Bool
otherwise                     = TailCallInfo
ti



-------------------
-- See Note [Adjusting right-hand sides]

adjustNonRecRhs :: JoinPointHood
                -> WithTailUsageDetails CoreExpr
                -> WithUsageDetails CoreExpr
-- ^ This function concentrates shared logic between occAnalNonRecBind and the
-- AcyclicSCC case of occAnalRec.
-- It returns the adjusted rhs UsageDetails combined with the body usage
adjustNonRecRhs :: JoinPointHood
-> WithTailUsageDetails CoreExpr -> WithUsageDetails CoreExpr
adjustNonRecRhs JoinPointHood
mb_join_arity rhs_wuds :: WithTailUsageDetails CoreExpr
rhs_wuds@(WTUD TailUsageDetails
_ CoreExpr
rhs)
  = UsageDetails -> CoreExpr -> WithUsageDetails CoreExpr
forall a. UsageDetails -> a -> WithUsageDetails a
WUD (JoinPointHood -> WithTailUsageDetails CoreExpr -> UsageDetails
adjustTailUsage JoinPointHood
mb_join_arity WithTailUsageDetails CoreExpr
rhs_wuds) CoreExpr
rhs


adjustTailUsage :: JoinPointHood
                -> WithTailUsageDetails CoreExpr    -- Rhs usage, AFTER occAnalLamTail
                -> UsageDetails
adjustTailUsage :: JoinPointHood -> WithTailUsageDetails CoreExpr -> UsageDetails
adjustTailUsage JoinPointHood
mb_join_arity (WTUD (TUD JoinArity
rhs_ja UsageDetails
uds) CoreExpr
rhs)
  = -- c.f. occAnal (Lam {})
    Bool -> UsageDetails -> UsageDetails
markAllInsideLamIf (Bool -> Bool
not Bool
one_shot) (UsageDetails -> UsageDetails) -> UsageDetails -> UsageDetails
forall a b. (a -> b) -> a -> b
$
    Bool -> UsageDetails -> UsageDetails
markAllNonTailIf (Bool -> Bool
not Bool
exact_join) (UsageDetails -> UsageDetails) -> UsageDetails -> UsageDetails
forall a b. (a -> b) -> a -> b
$
    UsageDetails
uds
  where
    one_shot :: Bool
one_shot   = CoreExpr -> Bool
isOneShotFun CoreExpr
rhs
    exact_join :: Bool
exact_join = JoinPointHood
mb_join_arity JoinPointHood -> JoinPointHood -> Bool
forall a. Eq a => a -> a -> Bool
== JoinArity -> JoinPointHood
JoinPoint JoinArity
rhs_ja

adjustTailArity :: JoinPointHood -> TailUsageDetails -> UsageDetails
adjustTailArity :: JoinPointHood -> TailUsageDetails -> UsageDetails
adjustTailArity JoinPointHood
mb_rhs_ja (TUD JoinArity
ja UsageDetails
usage)
  = Bool -> UsageDetails -> UsageDetails
markAllNonTailIf (JoinPointHood
mb_rhs_ja JoinPointHood -> JoinPointHood -> Bool
forall a. Eq a => a -> a -> Bool
/= JoinArity -> JoinPointHood
JoinPoint JoinArity
ja) UsageDetails
usage

type IdWithOccInfo = Id

tagLamBinders :: UsageDetails        -- Of scope
              -> [Id]                -- Binders
              -> [IdWithOccInfo]     -- Tagged binders
tagLamBinders :: UsageDetails -> [Id] -> [Id]
tagLamBinders UsageDetails
usage [Id]
binders
  = (Id -> Id) -> [Id] -> [Id]
forall a b. (a -> b) -> [a] -> [b]
map (UsageDetails -> Id -> Id
tagLamBinder UsageDetails
usage) [Id]
binders

tagLamBinder :: UsageDetails       -- Of scope
             -> Id                 -- Binder
             -> IdWithOccInfo      -- Tagged binders
-- Used for lambda and case binders
-- No-op on TyVars
-- A lambda binder never has an unfolding, so no need to look for that
tagLamBinder :: UsageDetails -> Id -> Id
tagLamBinder UsageDetails
usage Id
bndr
  = OccInfo -> Id -> Id
setBinderOcc (OccInfo -> OccInfo
markNonTail OccInfo
occ) Id
bndr
      -- markNonTail: don't try to make an argument into a join point
  where
    occ :: OccInfo
occ = UsageDetails -> Id -> OccInfo
lookupOccInfo UsageDetails
usage Id
bndr

tagNonRecBinder :: TopLevelFlag           -- At top level?
                -> OccInfo                -- Of scope
                -> CoreBndr               -- Binder
                -> (IdWithOccInfo, JoinPointHood)  -- Tagged binder
-- No-op on TyVars
-- Precondition: OccInfo is not IAmDead
tagNonRecBinder :: TopLevelFlag -> OccInfo -> Id -> (Id, JoinPointHood)
tagNonRecBinder TopLevelFlag
lvl OccInfo
occ Id
bndr
  | TopLevelFlag -> Id -> TailCallInfo -> Bool
okForJoinPoint TopLevelFlag
lvl Id
bndr TailCallInfo
tail_call_info
  , AlwaysTailCalled JoinArity
ar <- TailCallInfo
tail_call_info
  = (OccInfo -> Id -> Id
setBinderOcc OccInfo
occ Id
bndr,        JoinArity -> JoinPointHood
JoinPoint JoinArity
ar)
  | Bool
otherwise
  = (OccInfo -> Id -> Id
setBinderOcc OccInfo
zapped_occ Id
bndr, JoinPointHood
NotJoinPoint)
 where
    tail_call_info :: TailCallInfo
tail_call_info = OccInfo -> TailCallInfo
tailCallInfo OccInfo
occ
    zapped_occ :: OccInfo
zapped_occ     = OccInfo -> OccInfo
markNonTail OccInfo
occ

tagRecBinders :: TopLevelFlag           -- At top level?
              -> UsageDetails           -- Of body of let ONLY
              -> [NodeDetails]
              -> WithUsageDetails       -- Adjusted details for whole scope,
                                        -- with binders removed
                  [IdWithOccInfo]       -- Tagged binders
-- Substantially more complicated than non-recursive case. Need to adjust RHS
-- details *before* tagging binders (because the tags depend on the RHSes).
tagRecBinders :: TopLevelFlag
-> UsageDetails -> [NodeDetails] -> WithUsageDetails [Id]
tagRecBinders TopLevelFlag
lvl UsageDetails
body_uds [NodeDetails]
details_s
 = let
     bndrs :: [Id]
bndrs = (NodeDetails -> Id) -> [NodeDetails] -> [Id]
forall a b. (a -> b) -> [a] -> [b]
map NodeDetails -> Id
nd_bndr [NodeDetails]
details_s

     -- 1. See Note [Join arity prediction based on joinRhsArity]
     --    Determine possible join-point-hood of whole group, by testing for
     --    manifest join arity M.
     --    This (re-)asserts that makeNode had made tuds for that same arity M!
     unadj_uds :: UsageDetails
unadj_uds = (NodeDetails -> UsageDetails -> UsageDetails)
-> UsageDetails -> [NodeDetails] -> UsageDetails
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr (UsageDetails -> UsageDetails -> UsageDetails
andUDs (UsageDetails -> UsageDetails -> UsageDetails)
-> (NodeDetails -> UsageDetails)
-> NodeDetails
-> UsageDetails
-> UsageDetails
forall b c a. (b -> c) -> (a -> b) -> a -> c
. NodeDetails -> UsageDetails
test_manifest_arity) UsageDetails
body_uds [NodeDetails]
details_s
     test_manifest_arity :: NodeDetails -> UsageDetails
test_manifest_arity ND{nd_rhs :: NodeDetails -> WithTailUsageDetails CoreExpr
nd_rhs = WTUD TailUsageDetails
tuds CoreExpr
rhs}
       = JoinPointHood -> TailUsageDetails -> UsageDetails
adjustTailArity (JoinArity -> JoinPointHood
JoinPoint (CoreExpr -> JoinArity
joinRhsArity CoreExpr
rhs)) TailUsageDetails
tuds

     will_be_joins :: Bool
will_be_joins = TopLevelFlag -> UsageDetails -> [Id] -> Bool
decideRecJoinPointHood TopLevelFlag
lvl UsageDetails
unadj_uds [Id]
bndrs

     mb_join_arity :: Id -> JoinPointHood
     -- mb_join_arity: See Note [Join arity prediction based on joinRhsArity]
     -- This is the source O
     mb_join_arity :: Id -> JoinPointHood
mb_join_arity Id
bndr
         -- Can't use willBeJoinId_maybe here because we haven't tagged
         -- the binder yet (the tag depends on these adjustments!)
       | Bool
will_be_joins
       , AlwaysTailCalled JoinArity
arity <- UsageDetails -> Id -> TailCallInfo
lookupTailCallInfo UsageDetails
unadj_uds Id
bndr
       = JoinArity -> JoinPointHood
JoinPoint JoinArity
arity
       | Bool
otherwise
       = Bool -> JoinPointHood -> JoinPointHood
forall a. HasCallStack => Bool -> a -> a
assert (Bool -> Bool
not Bool
will_be_joins) -- Should be AlwaysTailCalled if
         JoinPointHood
NotJoinPoint               -- we are making join points!

     -- 2. Adjust usage details of each RHS, taking into account the
     --    join-point-hood decision
     rhs_udss' :: [UsageDetails]
rhs_udss' = [ JoinPointHood -> WithTailUsageDetails CoreExpr -> UsageDetails
adjustTailUsage (Id -> JoinPointHood
mb_join_arity Id
bndr) WithTailUsageDetails CoreExpr
rhs_wuds
                     -- Matching occAnalLamTail in makeNode
                 | ND { nd_bndr :: NodeDetails -> Id
nd_bndr = Id
bndr, nd_rhs :: NodeDetails -> WithTailUsageDetails CoreExpr
nd_rhs = WithTailUsageDetails CoreExpr
rhs_wuds } <- [NodeDetails]
details_s ]

     -- 3. Compute final usage details from adjusted RHS details
     adj_uds :: UsageDetails
adj_uds = (UsageDetails -> UsageDetails -> UsageDetails)
-> UsageDetails -> [UsageDetails] -> UsageDetails
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr UsageDetails -> UsageDetails -> UsageDetails
andUDs UsageDetails
body_uds [UsageDetails]
rhs_udss'

     -- 4. Tag each binder with its adjusted details
     bndrs' :: [Id]
bndrs'    = [ OccInfo -> Id -> Id
setBinderOcc (UsageDetails -> Id -> OccInfo
lookupLetOccInfo UsageDetails
adj_uds Id
bndr) Id
bndr
                 | Id
bndr <- [Id]
bndrs ]

   in
   UsageDetails -> [Id] -> WithUsageDetails [Id]
forall a. UsageDetails -> a -> WithUsageDetails a
WUD UsageDetails
adj_uds [Id]
bndrs'

setBinderOcc :: OccInfo -> CoreBndr -> CoreBndr
setBinderOcc :: OccInfo -> Id -> Id
setBinderOcc OccInfo
occ_info Id
bndr
  | Id -> Bool
isTyVar Id
bndr               = Id
bndr
  | OccInfo
occ_info OccInfo -> OccInfo -> Bool
forall a. Eq a => a -> a -> Bool
== Id -> OccInfo
idOccInfo Id
bndr = Id
bndr
  | Bool
otherwise                  = Id -> OccInfo -> Id
setIdOccInfo Id
bndr OccInfo
occ_info

-- | Decide whether some bindings should be made into join points or not, based
-- on its occurrences. This is
-- Returns `False` if they can't be join points. Note that it's an
-- all-or-nothing decision, as if multiple binders are given, they're
-- assumed to be mutually recursive.
--
-- It must, however, be a final decision. If we say `True` for 'f',
-- and then subsequently decide /not/ make 'f' into a join point, then
-- the decision about another binding 'g' might be invalidated if (say)
-- 'f' tail-calls 'g'.
--
-- See Note [Invariants on join points] in "GHC.Core".
decideRecJoinPointHood :: TopLevelFlag -> UsageDetails
                       -> [CoreBndr] -> Bool
decideRecJoinPointHood :: TopLevelFlag -> UsageDetails -> [Id] -> Bool
decideRecJoinPointHood TopLevelFlag
lvl UsageDetails
usage [Id]
bndrs
  = (Id -> Bool) -> [Id] -> Bool
forall (t :: * -> *) a. Foldable t => (a -> Bool) -> t a -> Bool
all Id -> Bool
ok [Id]
bndrs  -- Invariant 3: Either all are join points or none are
  where
    ok :: Id -> Bool
ok Id
bndr = TopLevelFlag -> Id -> TailCallInfo -> Bool
okForJoinPoint TopLevelFlag
lvl Id
bndr (UsageDetails -> Id -> TailCallInfo
lookupTailCallInfo UsageDetails
usage Id
bndr)

okForJoinPoint :: TopLevelFlag -> Id -> TailCallInfo -> Bool
    -- See Note [Invariants on join points]; invariants cited by number below.
    -- Invariant 2 is always satisfiable by the simplifier by eta expansion.
okForJoinPoint :: TopLevelFlag -> Id -> TailCallInfo -> Bool
okForJoinPoint TopLevelFlag
lvl Id
bndr TailCallInfo
tail_call_info
  | Id -> Bool
isJoinId Id
bndr        -- A current join point should still be one!
  = Bool -> String -> SDoc -> Bool -> Bool
forall a. HasCallStack => Bool -> String -> SDoc -> a -> a
warnPprTrace Bool
lost_join String
"Lost join point" SDoc
lost_join_doc (Bool -> Bool) -> Bool -> Bool
forall a b. (a -> b) -> a -> b
$
    Bool
True
  | Bool
valid_join
  = Bool
True
  | Bool
otherwise
  = Bool
False
  where
    valid_join :: Bool
valid_join | TopLevelFlag
NotTopLevel <- TopLevelFlag
lvl
               , AlwaysTailCalled JoinArity
arity <- TailCallInfo
tail_call_info

               , -- Invariant 1 as applied to LHSes of rules
                 (CoreRule -> Bool) -> [CoreRule] -> Bool
forall (t :: * -> *) a. Foldable t => (a -> Bool) -> t a -> Bool
all (JoinArity -> CoreRule -> Bool
ok_rule JoinArity
arity) (Id -> [CoreRule]
idCoreRules Id
bndr)

                 -- Invariant 2a: stable unfoldings
                  -- See Note [Join points and INLINE pragmas]
               , JoinArity -> Unfolding -> Bool
ok_unfolding JoinArity
arity (IdUnfoldingFun
realIdUnfolding Id
bndr)

                 -- Invariant 4: Satisfies polymorphism rule
               , JoinArity -> Type -> Bool
isValidJoinPointType JoinArity
arity (Id -> Type
idType Id
bndr)
               = Bool
True
               | Bool
otherwise
               = Bool
False

    lost_join :: Bool
lost_join | JoinPoint JoinArity
ja <- Id -> JoinPointHood
idJoinPointHood Id
bndr
              = Bool -> Bool
not Bool
valid_join Bool -> Bool -> Bool
||
                (case TailCallInfo
tail_call_info of  -- Valid join but arity differs
                   AlwaysTailCalled JoinArity
ja' -> JoinArity
ja JoinArity -> JoinArity -> Bool
forall a. Eq a => a -> a -> Bool
/= JoinArity
ja'
                   TailCallInfo
_                    -> Bool
False)
              | Bool
otherwise = Bool
False

    ok_rule :: JoinArity -> CoreRule -> Bool
ok_rule JoinArity
_ BuiltinRule{} = Bool
False -- only possible with plugin shenanigans
    ok_rule JoinArity
join_arity (Rule { ru_args :: CoreRule -> [CoreExpr]
ru_args = [CoreExpr]
args })
      = [CoreExpr]
args [CoreExpr] -> JoinArity -> Bool
forall a. [a] -> JoinArity -> Bool
`lengthIs` JoinArity
join_arity
        -- Invariant 1 as applied to LHSes of rules

    -- ok_unfolding returns False if we should /not/ convert a non-join-id
    -- into a join-id, even though it is AlwaysTailCalled
    ok_unfolding :: JoinArity -> Unfolding -> Bool
ok_unfolding JoinArity
join_arity (CoreUnfolding { uf_src :: Unfolding -> UnfoldingSource
uf_src = UnfoldingSource
src, uf_tmpl :: Unfolding -> CoreExpr
uf_tmpl = CoreExpr
rhs })
      = Bool -> Bool
not (UnfoldingSource -> Bool
isStableSource UnfoldingSource
src Bool -> Bool -> Bool
&& JoinArity
join_arity JoinArity -> JoinArity -> Bool
forall a. Ord a => a -> a -> Bool
> CoreExpr -> JoinArity
joinRhsArity CoreExpr
rhs)
    ok_unfolding JoinArity
_ (DFunUnfolding {})
      = Bool
False
    ok_unfolding JoinArity
_ Unfolding
_
      = Bool
True

    lost_join_doc :: SDoc
lost_join_doc
      = [SDoc] -> SDoc
forall doc. IsDoc doc => [doc] -> doc
vcat [ String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"bndr:" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> Id -> SDoc
forall a. Outputable a => a -> SDoc
ppr Id
bndr
             , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"tc:" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> TailCallInfo -> SDoc
forall a. Outputable a => a -> SDoc
ppr TailCallInfo
tail_call_info
             , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"rules:" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> [CoreRule] -> SDoc
forall a. Outputable a => a -> SDoc
ppr (Id -> [CoreRule]
idCoreRules Id
bndr)
             , case TailCallInfo
tail_call_info of
                 AlwaysTailCalled JoinArity
arity ->
                    [SDoc] -> SDoc
forall doc. IsDoc doc => [doc] -> doc
vcat [ String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"ok_unf:" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> Bool -> SDoc
forall a. Outputable a => a -> SDoc
ppr (JoinArity -> Unfolding -> Bool
ok_unfolding JoinArity
arity (IdUnfoldingFun
realIdUnfolding Id
bndr))
                         , String -> SDoc
forall doc. IsLine doc => String -> doc
text String
"ok_type:" SDoc -> SDoc -> SDoc
forall doc. IsLine doc => doc -> doc -> doc
<+> Bool -> SDoc
forall a. Outputable a => a -> SDoc
ppr (JoinArity -> Type -> Bool
isValidJoinPointType JoinArity
arity (Id -> Type
idType Id
bndr)) ]
                 TailCallInfo
_ -> SDoc
forall doc. IsOutput doc => doc
empty ]

{- Note [Join points and INLINE pragmas]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
   f x = let g = \x. not  -- Arity 1
             {-# INLINE g #-}
         in case x of
              A -> g True True
              B -> g True False
              C -> blah2

Here 'g' is always tail-called applied to 2 args, but the stable
unfolding captured by the INLINE pragma has arity 1.  If we try to
convert g to be a join point, its unfolding will still have arity 1
(since it is stable, and we don't meddle with stable unfoldings), and
Lint will complain (see Note [Invariants on join points], (2a), in
GHC.Core.  #13413.

Moreover, since g is going to be inlined anyway, there is no benefit
from making it a join point.

If it is recursive, and uselessly marked INLINE, this will stop us
making it a join point, which is annoying.  But occasionally
(notably in class methods; see Note [Instances and loop breakers] in
GHC.Tc.TyCl.Instance) we mark recursive things as INLINE but the recursion
unravels; so ignoring INLINE pragmas on recursive things isn't good
either.

See Invariant 2a of Note [Invariants on join points] in GHC.Core


************************************************************************
*                                                                      *
\subsection{Operations over OccInfo}
*                                                                      *
************************************************************************
-}

markNonTail :: OccInfo -> OccInfo
markNonTail :: OccInfo -> OccInfo
markNonTail OccInfo
IAmDead = OccInfo
IAmDead
markNonTail OccInfo
occ     = OccInfo
occ { occ_tail = NoTailCallInfo }

andLocalOcc :: LocalOcc -> LocalOcc -> LocalOcc
andLocalOcc :: LocalOcc -> LocalOcc -> LocalOcc
andLocalOcc LocalOcc
occ1 LocalOcc
occ2 = TailCallInfo -> LocalOcc
ManyOccL (TailCallInfo
tci1 TailCallInfo -> TailCallInfo -> TailCallInfo
`andTailCallInfo` TailCallInfo
tci2)
  where
    !tci1 :: TailCallInfo
tci1 = LocalOcc -> TailCallInfo
localTailCallInfo LocalOcc
occ1
    !tci2 :: TailCallInfo
tci2 = LocalOcc -> TailCallInfo
localTailCallInfo LocalOcc
occ2

orLocalOcc :: LocalOcc -> LocalOcc -> LocalOcc
-- (orLocalOcc occ1 occ2) is used
-- when combining occurrence info from branches of a case
orLocalOcc :: LocalOcc -> LocalOcc -> LocalOcc
orLocalOcc (OneOccL { lo_n_br :: LocalOcc -> JoinArity
lo_n_br = JoinArity
nbr1, lo_int_cxt :: LocalOcc -> InterestingCxt
lo_int_cxt = InterestingCxt
int_cxt1, lo_tail :: LocalOcc -> TailCallInfo
lo_tail = TailCallInfo
tci1 })
           (OneOccL { lo_n_br :: LocalOcc -> JoinArity
lo_n_br = JoinArity
nbr2, lo_int_cxt :: LocalOcc -> InterestingCxt
lo_int_cxt = InterestingCxt
int_cxt2, lo_tail :: LocalOcc -> TailCallInfo
lo_tail = TailCallInfo
tci2 })
  = OneOccL { lo_n_br :: JoinArity
lo_n_br    = JoinArity
nbr1 JoinArity -> JoinArity -> JoinArity
forall a. Num a => a -> a -> a
+ JoinArity
nbr2
            , lo_int_cxt :: InterestingCxt
lo_int_cxt = InterestingCxt
int_cxt1 InterestingCxt -> InterestingCxt -> InterestingCxt
forall a. Monoid a => a -> a -> a
`mappend` InterestingCxt
int_cxt2
            , lo_tail :: TailCallInfo
lo_tail    = TailCallInfo
tci1 TailCallInfo -> TailCallInfo -> TailCallInfo
`andTailCallInfo` TailCallInfo
tci2 }
orLocalOcc LocalOcc
occ1 LocalOcc
occ2 = LocalOcc -> LocalOcc -> LocalOcc
andLocalOcc LocalOcc
occ1 LocalOcc
occ2

andTailCallInfo :: TailCallInfo -> TailCallInfo -> TailCallInfo
andTailCallInfo :: TailCallInfo -> TailCallInfo -> TailCallInfo
andTailCallInfo info :: TailCallInfo
info@(AlwaysTailCalled JoinArity
arity1) (AlwaysTailCalled JoinArity
arity2)
  | JoinArity
arity1 JoinArity -> JoinArity -> Bool
forall a. Eq a => a -> a -> Bool
== JoinArity
arity2 = TailCallInfo
info
andTailCallInfo TailCallInfo
_ TailCallInfo
_  = TailCallInfo
NoTailCallInfo