{-# LANGUAGE CPP, BangPatterns #-}

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

-----------------------------------------------------------------------------
--
-- Stg to C-- code generation: expressions
--
-- (c) The University of Glasgow 2004-2006
--
-----------------------------------------------------------------------------

module GHC.StgToCmm.Expr ( cgExpr, cgLit ) where

#include "HsVersions.h"

import GHC.Prelude hiding ((<*>))

import {-# SOURCE #-} GHC.StgToCmm.Bind ( cgBind )

import GHC.StgToCmm.Monad
import GHC.StgToCmm.Heap
import GHC.StgToCmm.Env
import GHC.StgToCmm.DataCon
import GHC.StgToCmm.Prof (saveCurrentCostCentre, restoreCurrentCostCentre, emitSetCCC)
import GHC.StgToCmm.Layout
import GHC.StgToCmm.Lit
import GHC.StgToCmm.Prim
import GHC.StgToCmm.Hpc
import GHC.StgToCmm.Ticky
import GHC.StgToCmm.Utils
import GHC.StgToCmm.Closure

import GHC.Stg.Syntax

import GHC.Cmm.Graph
import GHC.Cmm.BlockId
import GHC.Cmm hiding ( succ )
import GHC.Cmm.Info
import GHC.Cmm.Utils ( zeroExpr, cmmTagMask, mkWordCLit, mAX_PTR_TAG )
import GHC.Core
import GHC.Core.DataCon
import GHC.Types.ForeignCall
import GHC.Types.Id
import GHC.Builtin.PrimOps
import GHC.Core.TyCon
import GHC.Core.Type        ( isUnliftedType )
import GHC.Types.RepType    ( isVoidTy, countConRepArgs )
import GHC.Types.CostCentre ( CostCentreStack, currentCCS )
import GHC.Types.Tickish
import GHC.Data.Maybe
import GHC.Utils.Misc
import GHC.Data.FastString
import GHC.Utils.Outputable
import GHC.Utils.Panic

import Control.Monad ( unless, void )
import Control.Arrow ( first )
import Data.List     ( partition )

------------------------------------------------------------------------
--              cgExpr: the main function
------------------------------------------------------------------------

cgExpr  :: CgStgExpr -> FCode ReturnKind

cgExpr :: CgStgExpr -> FCode ReturnKind
cgExpr (StgApp Id
fun [StgArg]
args)     = Id -> [StgArg] -> FCode ReturnKind
cgIdApp Id
fun [StgArg]
args

-- seq# a s ==> a
-- See Note [seq# magic] in GHC.Core.Opt.ConstantFold
cgExpr (StgOpApp (StgPrimOp PrimOp
SeqOp) [StgVarArg Id
a, StgArg
_] Type
_res_ty) =
  Id -> [StgArg] -> FCode ReturnKind
cgIdApp Id
a []

-- dataToTag# :: a -> Int#
-- See Note [dataToTag# magic] in primops.txt.pp
cgExpr (StgOpApp (StgPrimOp PrimOp
DataToTagOp) [StgVarArg Id
a] Type
_res_ty) = do
  Platform
platform <- FCode Platform
getPlatform
  FastString -> FCode ()
emitComment (String -> FastString
mkFastString String
"dataToTag#")
  CgIdInfo
info <- Id -> FCode CgIdInfo
getCgIdInfo Id
a
  let amode :: CmmExpr
amode = CgIdInfo -> CmmExpr
idInfoToAmode CgIdInfo
info
  LocalReg
tag_reg <- CmmExpr -> FCode LocalReg
assignTemp forall a b. (a -> b) -> a -> b
$ Platform -> CmmExpr -> CmmExpr
cmmConstrTag1 Platform
platform CmmExpr
amode
  LocalReg
result_reg <- forall (m :: * -> *). MonadUnique m => CmmType -> m LocalReg
newTemp (Platform -> CmmType
bWord Platform
platform)
  let tag :: CmmExpr
tag = CmmReg -> CmmExpr
CmmReg forall a b. (a -> b) -> a -> b
$ LocalReg -> CmmReg
CmmLocal LocalReg
tag_reg
      is_tagged :: CmmExpr
is_tagged = Platform -> CmmExpr -> CmmExpr -> CmmExpr
cmmNeWord Platform
platform CmmExpr
tag (Platform -> CmmExpr
zeroExpr Platform
platform)
      is_too_big_tag :: CmmExpr
is_too_big_tag = Platform -> CmmExpr -> CmmExpr -> CmmExpr
cmmEqWord Platform
platform CmmExpr
tag (Platform -> CmmExpr
cmmTagMask Platform
platform)
  -- Here we will first check the tag bits of the pointer we were given;
  -- if this doesn't work then enter the closure and use the info table
  -- to determine the constructor. Note that all tag bits set means that
  -- the constructor index is too large to fit in the pointer and therefore
  -- we must look in the info table. See Note [Tagging big families].

  CmmAGraph
slow_path <- forall a. FCode a -> FCode CmmAGraph
getCode forall a b. (a -> b) -> a -> b
$ do
      LocalReg
tmp <- forall (m :: * -> *). MonadUnique m => CmmType -> m LocalReg
newTemp (Platform -> CmmType
bWord Platform
platform)
      ReturnKind
_ <- forall a. Sequel -> FCode a -> FCode a
withSequel ([LocalReg] -> Bool -> Sequel
AssignTo [LocalReg
tmp] Bool
False) (Id -> [StgArg] -> FCode ReturnKind
cgIdApp Id
a [])
      PtrOpts
ptr_opts <- FCode PtrOpts
getPtrOpts
      CmmReg -> CmmExpr -> FCode ()
emitAssign (LocalReg -> CmmReg
CmmLocal LocalReg
result_reg)
        forall a b. (a -> b) -> a -> b
$ PtrOpts -> CmmExpr -> CmmExpr
getConstrTag PtrOpts
ptr_opts (Platform -> CmmExpr -> CmmExpr
cmmUntag Platform
platform (CmmReg -> CmmExpr
CmmReg (LocalReg -> CmmReg
CmmLocal LocalReg
tmp)))

  CmmAGraph
fast_path <- forall a. FCode a -> FCode CmmAGraph
getCode forall a b. (a -> b) -> a -> b
$ do
      -- Return the constructor index from the pointer tag
      CmmAGraph
return_ptr_tag <- forall a. FCode a -> FCode CmmAGraph
getCode forall a b. (a -> b) -> a -> b
$ do
          CmmReg -> CmmExpr -> FCode ()
emitAssign (LocalReg -> CmmReg
CmmLocal LocalReg
result_reg)
            forall a b. (a -> b) -> a -> b
$ Platform -> CmmExpr -> CmmExpr -> CmmExpr
cmmSubWord Platform
platform CmmExpr
tag (CmmLit -> CmmExpr
CmmLit forall a b. (a -> b) -> a -> b
$ Platform -> Integer -> CmmLit
mkWordCLit Platform
platform Integer
1)
      -- Return the constructor index recorded in the info table
      CmmAGraph
return_info_tag <- forall a. FCode a -> FCode CmmAGraph
getCode forall a b. (a -> b) -> a -> b
$ do
          PtrOpts
ptr_opts <- FCode PtrOpts
getPtrOpts
          CmmReg -> CmmExpr -> FCode ()
emitAssign (LocalReg -> CmmReg
CmmLocal LocalReg
result_reg)
            forall a b. (a -> b) -> a -> b
$ PtrOpts -> CmmExpr -> CmmExpr
getConstrTag PtrOpts
ptr_opts (Platform -> CmmExpr -> CmmExpr
cmmUntag Platform
platform CmmExpr
amode)

      CmmAGraph -> FCode ()
emit forall (m :: * -> *) a b. Monad m => (a -> m b) -> m a -> m b
=<< CmmExpr -> CmmAGraph -> CmmAGraph -> Maybe Bool -> FCode CmmAGraph
mkCmmIfThenElse' CmmExpr
is_too_big_tag CmmAGraph
return_info_tag CmmAGraph
return_ptr_tag (forall a. a -> Maybe a
Just Bool
False)

  CmmAGraph -> FCode ()
emit forall (m :: * -> *) a b. Monad m => (a -> m b) -> m a -> m b
=<< CmmExpr -> CmmAGraph -> CmmAGraph -> Maybe Bool -> FCode CmmAGraph
mkCmmIfThenElse' CmmExpr
is_tagged CmmAGraph
fast_path CmmAGraph
slow_path (forall a. a -> Maybe a
Just Bool
True)
  [CmmExpr] -> FCode ReturnKind
emitReturn [CmmReg -> CmmExpr
CmmReg forall a b. (a -> b) -> a -> b
$ LocalReg -> CmmReg
CmmLocal LocalReg
result_reg]


cgExpr (StgOpApp StgOp
op [StgArg]
args Type
ty) = StgOp -> [StgArg] -> Type -> FCode ReturnKind
cgOpApp StgOp
op [StgArg]
args Type
ty
cgExpr (StgConApp DataCon
con XConApp 'CodeGen
mn [StgArg]
args [Type]
_) = DataCon -> ConstructorNumber -> [StgArg] -> FCode ReturnKind
cgConApp DataCon
con XConApp 'CodeGen
mn [StgArg]
args
cgExpr (StgTick StgTickish
t CgStgExpr
e)         = StgTickish -> FCode ()
cgTick StgTickish
t forall (m :: * -> *) a b. Monad m => m a -> m b -> m b
>> CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
e
cgExpr (StgLit Literal
lit)          = do CmmExpr
cmm_expr <- Literal -> FCode CmmExpr
cgLit Literal
lit
                                  [CmmExpr] -> FCode ReturnKind
emitReturn [CmmExpr
cmm_expr]

cgExpr (StgLet XLet 'CodeGen
_ GenStgBinding 'CodeGen
binds CgStgExpr
expr) = do { GenStgBinding 'CodeGen -> FCode ()
cgBind GenStgBinding 'CodeGen
binds;     CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
expr }
cgExpr (StgLetNoEscape XLetNoEscape 'CodeGen
_ GenStgBinding 'CodeGen
binds CgStgExpr
expr) =
  do { Unique
u <- FCode Unique
newUnique
     ; let join_id :: BlockId
join_id = Unique -> BlockId
mkBlockId Unique
u
     ; BlockId -> GenStgBinding 'CodeGen -> FCode ()
cgLneBinds BlockId
join_id GenStgBinding 'CodeGen
binds
     ; ReturnKind
r <- CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
expr
     ; BlockId -> FCode ()
emitLabel BlockId
join_id
     ; forall (m :: * -> *) a. Monad m => a -> m a
return ReturnKind
r }

cgExpr (StgCase CgStgExpr
expr BinderP 'CodeGen
bndr AltType
alt_type [GenStgAlt 'CodeGen]
alts) =
  CgStgExpr
-> Id -> AltType -> [GenStgAlt 'CodeGen] -> FCode ReturnKind
cgCase CgStgExpr
expr BinderP 'CodeGen
bndr AltType
alt_type [GenStgAlt 'CodeGen]
alts

------------------------------------------------------------------------
--              Let no escape
------------------------------------------------------------------------

{- Generating code for a let-no-escape binding, aka join point is very
very similar to what we do for a case expression.  The duality is
between
        let-no-escape x = b
        in e
and
        case e of ... -> b

That is, the RHS of 'x' (ie 'b') will execute *later*, just like
the alternative of the case; it needs to be compiled in an environment
in which all volatile bindings are forgotten, and the free vars are
bound only to stable things like stack locations..  The 'e' part will
execute *next*, just like the scrutinee of a case. -}

-------------------------
cgLneBinds :: BlockId -> CgStgBinding -> FCode ()
cgLneBinds :: BlockId -> GenStgBinding 'CodeGen -> FCode ()
cgLneBinds BlockId
join_id (StgNonRec BinderP 'CodeGen
bndr GenStgRhs 'CodeGen
rhs)
  = do  { Maybe LocalReg
local_cc <- FCode (Maybe LocalReg)
saveCurrentCostCentre
                -- See Note [Saving the current cost centre]
        ; (CgIdInfo
info, FCode ()
fcode) <- BlockId
-> Maybe LocalReg
-> Id
-> GenStgRhs 'CodeGen
-> FCode (CgIdInfo, FCode ())
cgLetNoEscapeRhs BlockId
join_id Maybe LocalReg
local_cc BinderP 'CodeGen
bndr GenStgRhs 'CodeGen
rhs
        ; FCode ()
fcode
        ; CgIdInfo -> FCode ()
addBindC CgIdInfo
info }

cgLneBinds BlockId
join_id (StgRec [(BinderP 'CodeGen, GenStgRhs 'CodeGen)]
pairs)
  = do  { Maybe LocalReg
local_cc <- FCode (Maybe LocalReg)
saveCurrentCostCentre
        ; [(CgIdInfo, FCode ())]
r <- forall (t :: * -> *) (m :: * -> *) a.
(Traversable t, Monad m) =>
t (m a) -> m (t a)
sequence forall a b. (a -> b) -> a -> b
$ forall a b c. (a -> b -> c) -> [(a, b)] -> [c]
unzipWith (BlockId
-> Maybe LocalReg
-> Id
-> GenStgRhs 'CodeGen
-> FCode (CgIdInfo, FCode ())
cgLetNoEscapeRhs BlockId
join_id Maybe LocalReg
local_cc) [(BinderP 'CodeGen, GenStgRhs 'CodeGen)]
pairs
        ; let ([CgIdInfo]
infos, [FCode ()]
fcodes) = forall a b. [(a, b)] -> ([a], [b])
unzip [(CgIdInfo, FCode ())]
r
        ; [CgIdInfo] -> FCode ()
addBindsC [CgIdInfo]
infos
        ; forall (t :: * -> *) (m :: * -> *) a.
(Foldable t, Monad m) =>
t (m a) -> m ()
sequence_ [FCode ()]
fcodes
        }

-------------------------
cgLetNoEscapeRhs
    :: BlockId          -- join point for successor of let-no-escape
    -> Maybe LocalReg   -- Saved cost centre
    -> Id
    -> CgStgRhs
    -> FCode (CgIdInfo, FCode ())

cgLetNoEscapeRhs :: BlockId
-> Maybe LocalReg
-> Id
-> GenStgRhs 'CodeGen
-> FCode (CgIdInfo, FCode ())
cgLetNoEscapeRhs BlockId
join_id Maybe LocalReg
local_cc Id
bndr GenStgRhs 'CodeGen
rhs =
  do { (CgIdInfo
info, FCode ()
rhs_code) <- Maybe LocalReg
-> Id -> GenStgRhs 'CodeGen -> FCode (CgIdInfo, FCode ())
cgLetNoEscapeRhsBody Maybe LocalReg
local_cc Id
bndr GenStgRhs 'CodeGen
rhs
     ; let (BlockId
bid, [LocalReg]
_) = forall a. HasCallStack => String -> Maybe a -> a
expectJust String
"cgLetNoEscapeRhs" forall a b. (a -> b) -> a -> b
$ CgIdInfo -> Maybe (BlockId, [LocalReg])
maybeLetNoEscape CgIdInfo
info
     ; let code :: FCode ()
code = do { (()
_, CmmAGraphScoped
body) <- forall a. FCode a -> FCode (a, CmmAGraphScoped)
getCodeScoped FCode ()
rhs_code
                     ; BlockId -> CmmAGraphScoped -> FCode ()
emitOutOfLine BlockId
bid (forall (a :: * -> * -> *) b c d.
Arrow a =>
a b c -> a (b, d) (c, d)
first (CmmAGraph -> CmmAGraph -> CmmAGraph
<*> BlockId -> CmmAGraph
mkBranch BlockId
join_id) CmmAGraphScoped
body) }
     ; forall (m :: * -> *) a. Monad m => a -> m a
return (CgIdInfo
info, FCode ()
code)
     }

cgLetNoEscapeRhsBody
    :: Maybe LocalReg   -- Saved cost centre
    -> Id
    -> CgStgRhs
    -> FCode (CgIdInfo, FCode ())
cgLetNoEscapeRhsBody :: Maybe LocalReg
-> Id -> GenStgRhs 'CodeGen -> FCode (CgIdInfo, FCode ())
cgLetNoEscapeRhsBody Maybe LocalReg
local_cc Id
bndr (StgRhsClosure XRhsClosure 'CodeGen
_ CostCentreStack
cc UpdateFlag
_upd [BinderP 'CodeGen]
args CgStgExpr
body)
  = Id
-> Maybe LocalReg
-> CostCentreStack
-> [NonVoid Id]
-> CgStgExpr
-> FCode (CgIdInfo, FCode ())
cgLetNoEscapeClosure Id
bndr Maybe LocalReg
local_cc CostCentreStack
cc ([Id] -> [NonVoid Id]
nonVoidIds [BinderP 'CodeGen]
args) CgStgExpr
body
cgLetNoEscapeRhsBody Maybe LocalReg
local_cc Id
bndr (StgRhsCon CostCentreStack
cc DataCon
con ConstructorNumber
mn [StgTickish]
_ts [StgArg]
args)
  = Id
-> Maybe LocalReg
-> CostCentreStack
-> [NonVoid Id]
-> CgStgExpr
-> FCode (CgIdInfo, FCode ())
cgLetNoEscapeClosure Id
bndr Maybe LocalReg
local_cc CostCentreStack
cc []
      (forall (pass :: StgPass).
DataCon -> XConApp pass -> [StgArg] -> [Type] -> GenStgExpr pass
StgConApp DataCon
con ConstructorNumber
mn [StgArg]
args (forall a. HasCallStack => String -> SDoc -> a
pprPanic String
"cgLetNoEscapeRhsBody" forall a b. (a -> b) -> a -> b
$
                           String -> SDoc
text String
"StgRhsCon doesn't have type args"))
        -- For a constructor RHS we want to generate a single chunk of
        -- code which can be jumped to from many places, which will
        -- return the constructor. It's easy; just behave as if it
        -- was an StgRhsClosure with a ConApp inside!

-------------------------
cgLetNoEscapeClosure
        :: Id                   -- binder
        -> Maybe LocalReg       -- Slot for saved current cost centre
        -> CostCentreStack      -- XXX: *** NOT USED *** why not?
        -> [NonVoid Id]         -- Args (as in \ args -> body)
        -> CgStgExpr            -- Body (as in above)
        -> FCode (CgIdInfo, FCode ())

cgLetNoEscapeClosure :: Id
-> Maybe LocalReg
-> CostCentreStack
-> [NonVoid Id]
-> CgStgExpr
-> FCode (CgIdInfo, FCode ())
cgLetNoEscapeClosure Id
bndr Maybe LocalReg
cc_slot CostCentreStack
_unused_cc [NonVoid Id]
args CgStgExpr
body
  = do Platform
platform <- FCode Platform
getPlatform
       forall (m :: * -> *) a. Monad m => a -> m a
return ( Platform -> Id -> [NonVoid Id] -> CgIdInfo
lneIdInfo Platform
platform Id
bndr [NonVoid Id]
args, FCode ()
code )
  where
   code :: FCode ()
code = forall a. FCode a -> FCode a
forkLneBody forall a b. (a -> b) -> a -> b
$ forall a. Name -> [NonVoid Id] -> FCode a -> FCode a
withNewTickyCounterLNE (Id -> Name
idName Id
bndr) [NonVoid Id]
args forall a b. (a -> b) -> a -> b
$ do
            { Maybe LocalReg -> FCode ()
restoreCurrentCostCentre Maybe LocalReg
cc_slot
            ; [LocalReg]
arg_regs <- [NonVoid Id] -> FCode [LocalReg]
bindArgsToRegs [NonVoid Id]
args
            ; forall (f :: * -> *) a. Functor f => f a -> f ()
void forall a b. (a -> b) -> a -> b
$ forall a. [LocalReg] -> FCode a -> FCode a
noEscapeHeapCheck [LocalReg]
arg_regs (FCode ()
tickyEnterLNE forall (m :: * -> *) a b. Monad m => m a -> m b -> m b
>> CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
body) }


------------------------------------------------------------------------
--              Case expressions
------------------------------------------------------------------------

{- Note [Compiling case expressions]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
It is quite interesting to decide whether to put a heap-check at the
start of each alternative.  Of course we certainly have to do so if
the case forces an evaluation, or if there is a primitive op which can
trigger GC.

A more interesting situation is this (a Plan-B situation)

        !P!;
        ...P...
        case x# of
          0#      -> !Q!; ...Q...
          default -> !R!; ...R...

where !x! indicates a possible heap-check point. The heap checks
in the alternatives *can* be omitted, in which case the topmost
heapcheck will take their worst case into account.

In favour of omitting !Q!, !R!:

 - *May* save a heap overflow test,
   if ...P... allocates anything.

 - We can use relative addressing from a single Hp to
   get at all the closures so allocated.

 - No need to save volatile vars etc across heap checks
   in !Q!, !R!

Against omitting !Q!, !R!

  - May put a heap-check into the inner loop.  Suppose
        the main loop is P -> R -> P -> R...
        Q is the loop exit, and only it does allocation.
    This only hurts us if P does no allocation.  If P allocates,
    then there is a heap check in the inner loop anyway.

  - May do more allocation than reqd.  This sometimes bites us
    badly.  For example, nfib (ha!) allocates about 30\% more space if the
    worst-casing is done, because many many calls to nfib are leaf calls
    which don't need to allocate anything.

    We can un-allocate, but that costs an instruction

Neither problem hurts us if there is only one alternative.

Suppose the inner loop is P->R->P->R etc.  Then here is
how many heap checks we get in the *inner loop* under various
conditions

  Alloc   Heap check in branches (!Q!, !R!)?
  P Q R      yes     no (absorb to !P!)
--------------------------------------
  n n n      0          0
  n y n      0          1
  n . y      1          1
  y . y      2          1
  y . n      1          1

Best choices: absorb heap checks from Q and R into !P! iff
  a) P itself does some allocation
or
  b) P does allocation, or there is exactly one alternative

We adopt (b) because that is more likely to put the heap check at the
entry to a function, when not many things are live.  After a bunch of
single-branch cases, we may have lots of things live

Hence: two basic plans for

        case e of r { alts }

------ Plan A: the general case ---------

        ...save current cost centre...

        ...code for e,
           with sequel (SetLocals r)

        ...restore current cost centre...
        ...code for alts...
        ...alts do their own heap checks

------ Plan B: special case when ---------
  (i)  e does not allocate or call GC
  (ii) either upstream code performs allocation
       or there is just one alternative

  Then heap allocation in the (single) case branch
  is absorbed by the upstream check.
  Very common example: primops on unboxed values

        ...code for e,
           with sequel (SetLocals r)...

        ...code for alts...
        ...no heap check...
-}



-------------------------------------
data GcPlan
  = GcInAlts            -- Put a GC check at the start the case alternatives,
        [LocalReg]      -- which binds these registers
  | NoGcInAlts          -- The scrutinee is a primitive value, or a call to a
                        -- primitive op which does no GC.  Absorb the allocation
                        -- of the case alternative(s) into the upstream check

-------------------------------------
cgCase :: CgStgExpr -> Id -> AltType -> [CgStgAlt] -> FCode ReturnKind

{-
Note [Scrutinising VoidRep]
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Suppose we have this STG code:
   f = \[s : State# RealWorld] ->
       case s of _ -> blah
This is very odd.  Why are we scrutinising a state token?  But it
can arise with bizarre NOINLINE pragmas (#9964)
    crash :: IO ()
    crash = IO (\s -> let {-# NOINLINE s' #-}
                          s' = s
                      in (# s', () #))

Now the trouble is that 's' has VoidRep, and we do not bind void
arguments in the environment; they don't live anywhere.  See the
calls to nonVoidIds in various places.  So we must not look up
's' in the environment.  Instead, just evaluate the RHS!  Simple.

Note [Dead-binder optimisation]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
A case-binder, or data-constructor argument, may be marked as dead,
because we preserve occurrence-info on binders in GHC.Core.Tidy (see
GHC.Core.Tidy.tidyIdBndr).

If the binder is dead, we can sometimes eliminate a load.  While
CmmSink will eliminate that load, it's very easy to kill it at source
(giving CmmSink less work to do), and in any case CmmSink only runs
with -O. Since the majority of case binders are dead, this
optimisation probably still has a great benefit-cost ratio and we want
to keep it for -O0. See also Phab:D5358.

This probably also was the reason for occurrence hack in Phab:D5339 to
exist, perhaps because the occurrence information preserved by
'GHC.Core.Tidy.tidyIdBndr' was insufficient.  But now that CmmSink does the
job we deleted the hacks.
-}

cgCase :: CgStgExpr
-> Id -> AltType -> [GenStgAlt 'CodeGen] -> FCode ReturnKind
cgCase (StgApp Id
v []) Id
_ (PrimAlt PrimRep
_) [GenStgAlt 'CodeGen]
alts
  | PrimRep -> Bool
isVoidRep (Id -> PrimRep
idPrimRep Id
v)  -- See Note [Scrutinising VoidRep]
  , [(AltCon
DEFAULT, [BinderP 'CodeGen]
_, CgStgExpr
rhs)] <- [GenStgAlt 'CodeGen]
alts
  = CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
rhs

{- Note [Dodgy unsafeCoerce 1]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider
    case (x :: HValue) |> co of (y :: MutVar# Int)
        DEFAULT -> ...
We want to generate an assignment
     y := x
We want to allow this assignment to be generated in the case when the
types are compatible, because this allows some slightly-dodgy but
occasionally-useful casts to be used, such as in GHC.Runtime.Heap.Inspect
where we cast an HValue to a MutVar# so we can print out the contents
of the MutVar#.  If instead we generate code that enters the HValue,
then we'll get a runtime panic, because the HValue really is a
MutVar#.  The types are compatible though, so we can just generate an
assignment.
-}
cgCase (StgApp Id
v []) Id
bndr alt_type :: AltType
alt_type@(PrimAlt PrimRep
_) [GenStgAlt 'CodeGen]
alts
  | HasDebugCallStack => Type -> Bool
isUnliftedType (Id -> Type
idType Id
v)  -- Note [Dodgy unsafeCoerce 1]
  = -- assignment suffices for unlifted types
    do { Platform
platform <- FCode Platform
getPlatform
       ; forall (f :: * -> *). Applicative f => Bool -> f () -> f ()
unless (Platform -> Bool
reps_compatible Platform
platform) forall a b. (a -> b) -> a -> b
$
           forall a. HasCallStack => String -> SDoc -> a
pprPanic String
"cgCase: reps do not match, perhaps a dodgy unsafeCoerce?"
                    (Id -> SDoc
pp_bndr Id
v SDoc -> SDoc -> SDoc
$$ Id -> SDoc
pp_bndr Id
bndr)
       ; CgIdInfo
v_info <- Id -> FCode CgIdInfo
getCgIdInfo Id
v
       ; CmmReg -> CmmExpr -> FCode ()
emitAssign (LocalReg -> CmmReg
CmmLocal (Platform -> NonVoid Id -> LocalReg
idToReg Platform
platform (forall a. a -> NonVoid a
NonVoid Id
bndr)))
                    (CgIdInfo -> CmmExpr
idInfoToAmode CgIdInfo
v_info)
       -- Add bndr to the environment
       ; LocalReg
_ <- NonVoid Id -> FCode LocalReg
bindArgToReg (forall a. a -> NonVoid a
NonVoid Id
bndr)
       ; (GcPlan, ReturnKind)
-> NonVoid Id
-> AltType
-> [GenStgAlt 'CodeGen]
-> FCode ReturnKind
cgAlts (GcPlan
NoGcInAlts,ReturnKind
AssignedDirectly) (forall a. a -> NonVoid a
NonVoid Id
bndr) AltType
alt_type [GenStgAlt 'CodeGen]
alts }
  where
    reps_compatible :: Platform -> Bool
reps_compatible Platform
platform = Platform -> PrimRep -> PrimRep -> Bool
primRepCompatible Platform
platform (Id -> PrimRep
idPrimRep Id
v) (Id -> PrimRep
idPrimRep Id
bndr)

    pp_bndr :: Id -> SDoc
pp_bndr Id
id = forall a. Outputable a => a -> SDoc
ppr Id
id SDoc -> SDoc -> SDoc
<+> SDoc
dcolon SDoc -> SDoc -> SDoc
<+> forall a. Outputable a => a -> SDoc
ppr (Id -> Type
idType Id
id) SDoc -> SDoc -> SDoc
<+> SDoc -> SDoc
parens (forall a. Outputable a => a -> SDoc
ppr (Id -> PrimRep
idPrimRep Id
id))

{- Note [Dodgy unsafeCoerce 2, #3132]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In all other cases of a lifted Id being cast to an unlifted type, the
Id should be bound to bottom, otherwise this is an unsafe use of
unsafeCoerce.  We can generate code to enter the Id and assume that
it will never return.  Hence, we emit the usual enter/return code, and
because bottom must be untagged, it will be entered.  The Sequel is a
type-correct assignment, albeit bogus.  The (dead) continuation loops;
it would be better to invoke some kind of panic function here.
-}
cgCase scrut :: CgStgExpr
scrut@(StgApp Id
v []) Id
_ (PrimAlt PrimRep
_) [GenStgAlt 'CodeGen]
_
  = do { Platform
platform <- FCode Platform
getPlatform
       ; Maybe LocalReg
mb_cc <- Bool -> FCode (Maybe LocalReg)
maybeSaveCostCentre Bool
True
       ; ReturnKind
_ <- forall a. Sequel -> FCode a -> FCode a
withSequel
                  ([LocalReg] -> Bool -> Sequel
AssignTo [Platform -> NonVoid Id -> LocalReg
idToReg Platform
platform (forall a. a -> NonVoid a
NonVoid Id
v)] Bool
False) (CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
scrut)
       ; Maybe LocalReg -> FCode ()
restoreCurrentCostCentre Maybe LocalReg
mb_cc
       ; FastString -> FCode ()
emitComment forall a b. (a -> b) -> a -> b
$ String -> FastString
mkFastString String
"should be unreachable code"
       ; BlockId
l <- forall (m :: * -> *). MonadUnique m => m BlockId
newBlockId
       ; BlockId -> FCode ()
emitLabel BlockId
l
       ; CmmAGraph -> FCode ()
emit (BlockId -> CmmAGraph
mkBranch BlockId
l)  -- an infinite loop
       ; forall (m :: * -> *) a. Monad m => a -> m a
return ReturnKind
AssignedDirectly
       }

{- Note [Handle seq#]
~~~~~~~~~~~~~~~~~~~~~
See Note [seq# magic] in GHC.Core.Opt.ConstantFold.
The special case for seq# in cgCase does this:

  case seq# a s of v
    (# s', a' #) -> e
==>
  case a of v
    (# s', a' #) -> e

(taking advantage of the fact that the return convention for (# State#, a #)
is the same as the return convention for just 'a')
-}

cgCase (StgOpApp (StgPrimOp PrimOp
SeqOp) [StgVarArg Id
a, StgArg
_] Type
_) Id
bndr AltType
alt_type [GenStgAlt 'CodeGen]
alts
  = -- Note [Handle seq#]
    -- And see Note [seq# magic] in GHC.Core.Opt.ConstantFold
    -- Use the same return convention as vanilla 'a'.
    CgStgExpr
-> Id -> AltType -> [GenStgAlt 'CodeGen] -> FCode ReturnKind
cgCase (forall (pass :: StgPass). Id -> [StgArg] -> GenStgExpr pass
StgApp Id
a []) Id
bndr AltType
alt_type [GenStgAlt 'CodeGen]
alts

cgCase CgStgExpr
scrut Id
bndr AltType
alt_type [GenStgAlt 'CodeGen]
alts
  = -- the general case
    do { Platform
platform <- FCode Platform
getPlatform
       ; RepArity
up_hp_usg <- FCode RepArity
getVirtHp        -- Upstream heap usage
       ; let ret_bndrs :: [NonVoid Id]
ret_bndrs = Id -> AltType -> [GenStgAlt 'CodeGen] -> [NonVoid Id]
chooseReturnBndrs Id
bndr AltType
alt_type [GenStgAlt 'CodeGen]
alts
             alt_regs :: [LocalReg]
alt_regs  = forall a b. (a -> b) -> [a] -> [b]
map (Platform -> NonVoid Id -> LocalReg
idToReg Platform
platform) [NonVoid Id]
ret_bndrs
       ; Bool
simple_scrut <- CgStgExpr -> AltType -> FCode Bool
isSimpleScrut CgStgExpr
scrut AltType
alt_type
       ; let do_gc :: Bool
do_gc  | forall {pass :: StgPass}. GenStgExpr pass -> Bool
is_cmp_op CgStgExpr
scrut  = Bool
False  -- See Note [GC for conditionals]
                    | Bool -> Bool
not Bool
simple_scrut = Bool
True
                    | forall a. [a] -> Bool
isSingleton [GenStgAlt 'CodeGen]
alts = Bool
False
                    | RepArity
up_hp_usg forall a. Ord a => a -> a -> Bool
> RepArity
0    = Bool
False
                    | Bool
otherwise        = Bool
True
               -- cf Note [Compiling case expressions]
             gc_plan :: GcPlan
gc_plan = if Bool
do_gc then [LocalReg] -> GcPlan
GcInAlts [LocalReg]
alt_regs else GcPlan
NoGcInAlts

       ; Maybe LocalReg
mb_cc <- Bool -> FCode (Maybe LocalReg)
maybeSaveCostCentre Bool
simple_scrut

       ; let sequel :: Sequel
sequel = [LocalReg] -> Bool -> Sequel
AssignTo [LocalReg]
alt_regs Bool
do_gc{- Note [scrut sequel] -}
       ; ReturnKind
ret_kind <- forall a. Sequel -> FCode a -> FCode a
withSequel Sequel
sequel (CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
scrut)
       ; Maybe LocalReg -> FCode ()
restoreCurrentCostCentre Maybe LocalReg
mb_cc
       ; [LocalReg]
_ <- [NonVoid Id] -> FCode [LocalReg]
bindArgsToRegs [NonVoid Id]
ret_bndrs
       ; (GcPlan, ReturnKind)
-> NonVoid Id
-> AltType
-> [GenStgAlt 'CodeGen]
-> FCode ReturnKind
cgAlts (GcPlan
gc_plan,ReturnKind
ret_kind) (forall a. a -> NonVoid a
NonVoid Id
bndr) AltType
alt_type [GenStgAlt 'CodeGen]
alts
       }
  where
    is_cmp_op :: GenStgExpr pass -> Bool
is_cmp_op (StgOpApp (StgPrimOp PrimOp
op) [StgArg]
_ Type
_) = PrimOp -> Bool
isComparisonPrimOp PrimOp
op
    is_cmp_op GenStgExpr pass
_                             = Bool
False

{- Note [GC for conditionals]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For boolean conditionals it seems that we have always done NoGcInAlts.
That is, we have always done the GC check before the conditional.
This is enshrined in the special case for
   case tagToEnum# (a>b) of ...
See Note [case on bool]

It's odd, and it's flagrantly inconsistent with the rules described
Note [Compiling case expressions].  However, after eliminating the
tagToEnum# (#13397) we will have:
   case (a>b) of ...
Rather than make it behave quite differently, I am testing for a
comparison operator here in the general case as well.

ToDo: figure out what the Right Rule should be.

Note [scrut sequel]
~~~~~~~~~~~~~~~~~~~
The job of the scrutinee is to assign its value(s) to alt_regs.
Additionally, if we plan to do a heap-check in the alternatives (see
Note [Compiling case expressions]), then we *must* retreat Hp to
recover any unused heap before passing control to the sequel.  If we
don't do this, then any unused heap will become slop because the heap
check will reset the heap usage. Slop in the heap breaks LDV profiling
(+RTS -hb) which needs to do a linear sweep through the nursery.


Note [Inlining out-of-line primops and heap checks]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
If shouldInlinePrimOp returns True when called from GHC.StgToCmm.Expr for the
purpose of heap check placement, we *must* inline the primop later in
GHC.StgToCmm.Prim. If we don't things will go wrong.
-}

-----------------
maybeSaveCostCentre :: Bool -> FCode (Maybe LocalReg)
maybeSaveCostCentre :: Bool -> FCode (Maybe LocalReg)
maybeSaveCostCentre Bool
simple_scrut
  | Bool
simple_scrut = forall (m :: * -> *) a. Monad m => a -> m a
return forall a. Maybe a
Nothing
  | Bool
otherwise    = FCode (Maybe LocalReg)
saveCurrentCostCentre


-----------------
isSimpleScrut :: CgStgExpr -> AltType -> FCode Bool
-- Simple scrutinee, does not block or allocate; hence safe to amalgamate
-- heap usage from alternatives into the stuff before the case
-- NB: if you get this wrong, and claim that the expression doesn't allocate
--     when it does, you'll deeply mess up allocation
isSimpleScrut :: CgStgExpr -> AltType -> FCode Bool
isSimpleScrut (StgOpApp StgOp
op [StgArg]
args Type
_) AltType
_       = StgOp -> [StgArg] -> FCode Bool
isSimpleOp StgOp
op [StgArg]
args
isSimpleScrut (StgLit Literal
_)       AltType
_           = forall (m :: * -> *) a. Monad m => a -> m a
return Bool
True       -- case 1# of { 0# -> ..; ... }
isSimpleScrut (StgApp Id
_ [])    (PrimAlt PrimRep
_) = forall (m :: * -> *) a. Monad m => a -> m a
return Bool
True       -- case x# of { 0# -> ..; ... }
isSimpleScrut CgStgExpr
_                AltType
_           = forall (m :: * -> *) a. Monad m => a -> m a
return Bool
False

isSimpleOp :: StgOp -> [StgArg] -> FCode Bool
-- True iff the op cannot block or allocate
isSimpleOp :: StgOp -> [StgArg] -> FCode Bool
isSimpleOp (StgFCallOp (CCall (CCallSpec CCallTarget
_ CCallConv
_ Safety
safe)) Type
_) [StgArg]
_ = forall (m :: * -> *) a. Monad m => a -> m a
return forall a b. (a -> b) -> a -> b
$! Bool -> Bool
not (Safety -> Bool
playSafe Safety
safe)
-- dataToTag# evaluates its argument, see Note [dataToTag#] in primops.txt.pp
isSimpleOp (StgPrimOp PrimOp
DataToTagOp) [StgArg]
_ = forall (m :: * -> *) a. Monad m => a -> m a
return Bool
False
isSimpleOp (StgPrimOp PrimOp
op) [StgArg]
stg_args                  = do
    [CmmExpr]
arg_exprs <- [StgArg] -> FCode [CmmExpr]
getNonVoidArgAmodes [StgArg]
stg_args
    DynFlags
dflags <- forall (m :: * -> *). HasDynFlags m => m DynFlags
getDynFlags
    -- See Note [Inlining out-of-line primops and heap checks]
    forall (m :: * -> *) a. Monad m => a -> m a
return forall a b. (a -> b) -> a -> b
$! DynFlags -> PrimOp -> [CmmExpr] -> Bool
shouldInlinePrimOp DynFlags
dflags PrimOp
op [CmmExpr]
arg_exprs
isSimpleOp (StgPrimCallOp PrimCall
_) [StgArg]
_                           = forall (m :: * -> *) a. Monad m => a -> m a
return Bool
False

-----------------
chooseReturnBndrs :: Id -> AltType -> [CgStgAlt] -> [NonVoid Id]
-- These are the binders of a case that are assigned by the evaluation of the
-- scrutinee.
-- They're non-void, see Note [Post-unarisation invariants] in GHC.Stg.Unarise.
chooseReturnBndrs :: Id -> AltType -> [GenStgAlt 'CodeGen] -> [NonVoid Id]
chooseReturnBndrs Id
bndr (PrimAlt PrimRep
_) [GenStgAlt 'CodeGen]
_alts
  = [Id] -> [NonVoid Id]
assertNonVoidIds [Id
bndr]

chooseReturnBndrs Id
_bndr (MultiValAlt RepArity
n) [(AltCon
_, [BinderP 'CodeGen]
ids, CgStgExpr
_)]
  = ASSERT2(ids `lengthIs` n, ppr n $$ ppr ids $$ ppr _bndr)
    [Id] -> [NonVoid Id]
assertNonVoidIds [BinderP 'CodeGen]
ids     -- 'bndr' is not assigned!

chooseReturnBndrs Id
bndr (AlgAlt TyCon
_) [GenStgAlt 'CodeGen]
_alts
  = [Id] -> [NonVoid Id]
assertNonVoidIds [Id
bndr]  -- Only 'bndr' is assigned

chooseReturnBndrs Id
bndr AltType
PolyAlt [GenStgAlt 'CodeGen]
_alts
  = [Id] -> [NonVoid Id]
assertNonVoidIds [Id
bndr]  -- Only 'bndr' is assigned

chooseReturnBndrs Id
_ AltType
_ [GenStgAlt 'CodeGen]
_ = forall a. String -> a
panic String
"chooseReturnBndrs"
                             -- MultiValAlt has only one alternative

-------------------------------------
cgAlts :: (GcPlan,ReturnKind) -> NonVoid Id -> AltType -> [CgStgAlt]
       -> FCode ReturnKind
-- At this point the result of the case are in the binders
cgAlts :: (GcPlan, ReturnKind)
-> NonVoid Id
-> AltType
-> [GenStgAlt 'CodeGen]
-> FCode ReturnKind
cgAlts (GcPlan, ReturnKind)
gc_plan NonVoid Id
_bndr AltType
PolyAlt [(AltCon
_, [BinderP 'CodeGen]
_, CgStgExpr
rhs)]
  = forall a. (GcPlan, ReturnKind) -> FCode a -> FCode a
maybeAltHeapCheck (GcPlan, ReturnKind)
gc_plan (CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
rhs)

cgAlts (GcPlan, ReturnKind)
gc_plan NonVoid Id
_bndr (MultiValAlt RepArity
_) [(AltCon
_, [BinderP 'CodeGen]
_, CgStgExpr
rhs)]
  = forall a. (GcPlan, ReturnKind) -> FCode a -> FCode a
maybeAltHeapCheck (GcPlan, ReturnKind)
gc_plan (CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
rhs)
        -- Here bndrs are *already* in scope, so don't rebind them

cgAlts (GcPlan, ReturnKind)
gc_plan NonVoid Id
bndr (PrimAlt PrimRep
_) [GenStgAlt 'CodeGen]
alts
  = do  { Platform
platform <- FCode Platform
getPlatform

        ; [(AltCon, CmmAGraphScoped)]
tagged_cmms <- (GcPlan, ReturnKind)
-> NonVoid Id
-> [GenStgAlt 'CodeGen]
-> FCode [(AltCon, CmmAGraphScoped)]
cgAltRhss (GcPlan, ReturnKind)
gc_plan NonVoid Id
bndr [GenStgAlt 'CodeGen]
alts

        ; let bndr_reg :: CmmReg
bndr_reg = LocalReg -> CmmReg
CmmLocal (Platform -> NonVoid Id -> LocalReg
idToReg Platform
platform NonVoid Id
bndr)
              (AltCon
DEFAULT,CmmAGraphScoped
deflt) = forall a. [a] -> a
head [(AltCon, CmmAGraphScoped)]
tagged_cmms
                -- PrimAlts always have a DEFAULT case
                -- and it always comes first

              tagged_cmms' :: [(Literal, CmmAGraphScoped)]
tagged_cmms' = [(Literal
lit,CmmAGraphScoped
code)
                             | (LitAlt Literal
lit, CmmAGraphScoped
code) <- [(AltCon, CmmAGraphScoped)]
tagged_cmms]
        ; CmmExpr
-> [(Literal, CmmAGraphScoped)] -> CmmAGraphScoped -> FCode ()
emitCmmLitSwitch (CmmReg -> CmmExpr
CmmReg CmmReg
bndr_reg) [(Literal, CmmAGraphScoped)]
tagged_cmms' CmmAGraphScoped
deflt
        ; forall (m :: * -> *) a. Monad m => a -> m a
return ReturnKind
AssignedDirectly }

cgAlts (GcPlan, ReturnKind)
gc_plan NonVoid Id
bndr (AlgAlt TyCon
tycon) [GenStgAlt 'CodeGen]
alts
  = do  { Platform
platform <- FCode Platform
getPlatform

        ; (Maybe CmmAGraphScoped
mb_deflt, [(RepArity, CmmAGraphScoped)]
branches) <- (GcPlan, ReturnKind)
-> NonVoid Id
-> [GenStgAlt 'CodeGen]
-> FCode (Maybe CmmAGraphScoped, [(RepArity, CmmAGraphScoped)])
cgAlgAltRhss (GcPlan, ReturnKind)
gc_plan NonVoid Id
bndr [GenStgAlt 'CodeGen]
alts

        ; let !fam_sz :: RepArity
fam_sz   = TyCon -> RepArity
tyConFamilySize TyCon
tycon
              !bndr_reg :: CmmReg
bndr_reg = LocalReg -> CmmReg
CmmLocal (Platform -> NonVoid Id -> LocalReg
idToReg Platform
platform NonVoid Id
bndr)
              !ptag_expr :: CmmExpr
ptag_expr = Platform -> CmmExpr -> CmmExpr
cmmConstrTag1 Platform
platform (CmmReg -> CmmExpr
CmmReg CmmReg
bndr_reg)
              !branches' :: [(RepArity, CmmAGraphScoped)]
branches' = forall (a :: * -> * -> *) b c d.
Arrow a =>
a b c -> a (b, d) (c, d)
first forall a. Enum a => a -> a
succ forall (f :: * -> *) a b. Functor f => (a -> b) -> f a -> f b
<$> [(RepArity, CmmAGraphScoped)]
branches
              !maxpt :: RepArity
maxpt = Platform -> RepArity
mAX_PTR_TAG Platform
platform
              (![(RepArity, CmmAGraphScoped)]
via_ptr, ![(RepArity, CmmAGraphScoped)]
via_info) = forall a. (a -> Bool) -> [a] -> ([a], [a])
partition ((forall a. Ord a => a -> a -> Bool
< RepArity
maxpt) forall b c a. (b -> c) -> (a -> b) -> a -> c
. forall a b. (a, b) -> a
fst) [(RepArity, CmmAGraphScoped)]
branches'
              !small :: Bool
small = Platform -> RepArity -> Bool
isSmallFamily Platform
platform RepArity
fam_sz

                -- Is the constructor tag in the node reg?
                -- See Note [Tagging big families]
        ; if Bool
small Bool -> Bool -> Bool
|| forall (t :: * -> *) a. Foldable t => t a -> Bool
null [(RepArity, CmmAGraphScoped)]
via_info
           then -- Yes, bndr_reg has constructor tag in ls bits
               CmmExpr
-> [(RepArity, CmmAGraphScoped)]
-> Maybe CmmAGraphScoped
-> RepArity
-> RepArity
-> FCode ()
emitSwitch CmmExpr
ptag_expr [(RepArity, CmmAGraphScoped)]
branches' Maybe CmmAGraphScoped
mb_deflt RepArity
1
                 (if Bool
small then RepArity
fam_sz else RepArity
maxpt)

           else -- No, the get exact tag from info table when mAX_PTR_TAG
                -- See Note [Double switching for big families]
              do
                PtrOpts
ptr_opts <- FCode PtrOpts
getPtrOpts
                let !untagged_ptr :: CmmExpr
untagged_ptr = Platform -> CmmExpr -> CmmExpr
cmmUntag Platform
platform (CmmReg -> CmmExpr
CmmReg CmmReg
bndr_reg)
                    !itag_expr :: CmmExpr
itag_expr = PtrOpts -> CmmExpr -> CmmExpr
getConstrTag PtrOpts
ptr_opts CmmExpr
untagged_ptr
                    !info0 :: [(RepArity, CmmAGraphScoped)]
info0 = forall (a :: * -> * -> *) b c d.
Arrow a =>
a b c -> a (b, d) (c, d)
first forall a. Enum a => a -> a
pred forall (f :: * -> *) a b. Functor f => (a -> b) -> f a -> f b
<$> [(RepArity, CmmAGraphScoped)]
via_info
                if forall (t :: * -> *) a. Foldable t => t a -> Bool
null [(RepArity, CmmAGraphScoped)]
via_ptr then
                  CmmExpr
-> [(RepArity, CmmAGraphScoped)]
-> Maybe CmmAGraphScoped
-> RepArity
-> RepArity
-> FCode ()
emitSwitch CmmExpr
itag_expr [(RepArity, CmmAGraphScoped)]
info0 Maybe CmmAGraphScoped
mb_deflt RepArity
0 (RepArity
fam_sz forall a. Num a => a -> a -> a
- RepArity
1)
                else do
                  BlockId
infos_lbl <- forall (m :: * -> *). MonadUnique m => m BlockId
newBlockId
                  CmmTickScope
infos_scp <- FCode CmmTickScope
getTickScope

                  let spillover :: (RepArity, CmmAGraphScoped)
spillover = (RepArity
maxpt, (BlockId -> CmmAGraph
mkBranch BlockId
infos_lbl, CmmTickScope
infos_scp))

                  (Maybe CmmAGraphScoped
mb_shared_deflt, Maybe CmmAGraphScoped
mb_shared_branch) <- case Maybe CmmAGraphScoped
mb_deflt of
                      (Just (CmmAGraph
stmts, CmmTickScope
scp)) ->
                          do BlockId
lbl <- forall (m :: * -> *). MonadUnique m => m BlockId
newBlockId
                             forall (m :: * -> *) a. Monad m => a -> m a
return ( forall a. a -> Maybe a
Just (BlockId -> CmmTickScope -> CmmAGraph
mkLabel BlockId
lbl CmmTickScope
scp CmmAGraph -> CmmAGraph -> CmmAGraph
<*> CmmAGraph
stmts, CmmTickScope
scp)
                                    , forall a. a -> Maybe a
Just (BlockId -> CmmAGraph
mkBranch BlockId
lbl, CmmTickScope
scp))
                      Maybe CmmAGraphScoped
_ -> forall (m :: * -> *) a. Monad m => a -> m a
return (forall a. Maybe a
Nothing, forall a. Maybe a
Nothing)
                  -- Switch on pointer tag
                  CmmExpr
-> [(RepArity, CmmAGraphScoped)]
-> Maybe CmmAGraphScoped
-> RepArity
-> RepArity
-> FCode ()
emitSwitch CmmExpr
ptag_expr ((RepArity, CmmAGraphScoped)
spillover forall a. a -> [a] -> [a]
: [(RepArity, CmmAGraphScoped)]
via_ptr) Maybe CmmAGraphScoped
mb_shared_deflt RepArity
1 RepArity
maxpt
                  BlockId
join_lbl <- forall (m :: * -> *). MonadUnique m => m BlockId
newBlockId
                  CmmAGraph -> FCode ()
emit (BlockId -> CmmAGraph
mkBranch BlockId
join_lbl)
                  -- Switch on info table tag
                  BlockId -> FCode ()
emitLabel BlockId
infos_lbl
                  CmmExpr
-> [(RepArity, CmmAGraphScoped)]
-> Maybe CmmAGraphScoped
-> RepArity
-> RepArity
-> FCode ()
emitSwitch CmmExpr
itag_expr [(RepArity, CmmAGraphScoped)]
info0 Maybe CmmAGraphScoped
mb_shared_branch
                    (RepArity
maxpt forall a. Num a => a -> a -> a
- RepArity
1) (RepArity
fam_sz forall a. Num a => a -> a -> a
- RepArity
1)
                  BlockId -> FCode ()
emitLabel BlockId
join_lbl

        ; forall (m :: * -> *) a. Monad m => a -> m a
return ReturnKind
AssignedDirectly }

cgAlts (GcPlan, ReturnKind)
_ NonVoid Id
_ AltType
_ [GenStgAlt 'CodeGen]
_ = forall a. String -> a
panic String
"cgAlts"
        -- UbxTupAlt and PolyAlt have only one alternative

-- Note [Double switching for big families]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- An algebraic data type can have a n >= 0 summands
-- (or alternatives), which are identified (labeled) by
-- constructors. In memory they are kept apart by tags
-- (see Note [Data constructor dynamic tags] in GHC.StgToCmm.Closure).
-- Due to the characteristics of the platform that
-- contribute to the alignment of memory objects, there
-- is a natural limit of information about constructors
-- that can be encoded in the pointer tag. When the mapping
-- of constructors to the pointer tag range 1..mAX_PTR_TAG
-- is not injective, then we have a "big data type", also
-- called a "big (constructor) family" in the literature.
-- Constructor tags residing in the info table are injective,
-- but considerably more expensive to obtain, due to additional
-- memory access(es).
--
-- When doing case analysis on a value of a "big data type"
-- we need two nested switch statements to make up for the lack
-- of injectivity of pointer tagging, also taking the info
-- table tag into account. The exact mechanism is described next.
--
-- In the general case, switching on big family alternatives
-- is done by two nested switch statements. According to
-- Note [Tagging big families], the outer switch
-- looks at the pointer tag and the inner dereferences the
-- pointer and switches on the info table tag.
--
-- We can handle a simple case first, namely when none
-- of the case alternatives mention a constructor having
-- a pointer tag of 1..mAX_PTR_TAG-1. In this case we
-- simply emit a switch on the info table tag.
-- Note that the other simple case is when all mentioned
-- alternatives lie in 1..mAX_PTR_TAG-1, in which case we can
-- switch on the ptr tag only, just like in the small family case.
--
-- There is a single intricacy with a nested switch:
-- Both should branch to the same default alternative, and as such
-- avoid duplicate codegen of potentially heavy code. The outer
-- switch generates the actual code with a prepended fresh label,
-- while the inner one only generates a jump to that label.
--
-- For example, let's assume a 64-bit architecture, so that all
-- heap objects are 8-byte aligned, and hence the address of a
-- heap object ends in `000` (three zero bits).
--
-- Then consider the following data type
--
--   > data Big = T0 | T1 | T2 | T3 | T4 | T5 | T6 | T7 | T8
--   Ptr tag:      1    2    3    4    5    6    7    7    7
--   As bits:    001  010  011  100  101  110  111  111  111
--   Info pointer tag (zero based):
--                 0    1    2    3    4    5    6    7    8
--
-- Then     \case T2 -> True; T8 -> True; _ -> False
-- will result in following code (slightly cleaned-up and
-- commented -ddump-cmm-from-stg):
{-
           R1 = _sqI::P64;  -- scrutinee
           if (R1 & 7 != 0) goto cqO; else goto cqP;
       cqP: // global       -- enter
           call (I64[R1])(R1) returns to cqO, args: 8, res: 8, upd: 8;
       cqO: // global       -- already WHNF
           _sqJ::P64 = R1;
           _cqX::P64 = _sqJ::P64 & 7;  -- extract pointer tag
           switch [1 .. 7] _cqX::P64 {
               case 3 : goto cqW;
               case 7 : goto cqR;
               default: {goto cqS;}
           }
       cqR: // global
           _cr2 = I32[I64[_sqJ::P64 & (-8)] - 4]; -- tag from info pointer
           switch [6 .. 8] _cr2::I64 {
               case 8 : goto cr1;
               default: {goto cr0;}
           }
       cr1: // global
           R1 = GHC.Types.True_closure+2;
           call (P64[(old + 8)])(R1) args: 8, res: 0, upd: 8;
       cr0: // global     -- technically necessary label
           goto cqS;
       cqW: // global
           R1 = GHC.Types.True_closure+2;
           call (P64[(old + 8)])(R1) args: 8, res: 0, upd: 8;
       cqS: // global
           R1 = GHC.Types.False_closure+1;
           call (P64[(old + 8)])(R1) args: 8, res: 0, upd: 8;
-}
--
-- For 32-bit systems we only have 2 tag bits in the pointers at our disposal,
-- so the performance win is dubious, especially in face of the increased code
-- size due to double switching. But we can take the viewpoint that 32-bit
-- architectures are not relevant for performance any more, so this can be
-- considered as moot.


-- Note [alg-alt heap check]
--
-- In an algebraic case with more than one alternative, we will have
-- code like
--
-- L0:
--   x = R1
--   goto L1
-- L1:
--   if (x & 7 >= 2) then goto L2 else goto L3
-- L2:
--   Hp = Hp + 16
--   if (Hp > HpLim) then goto L4
--   ...
-- L4:
--   call gc() returns to L5
-- L5:
--   x = R1
--   goto L1


-- Note [Tagging big families]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- Both the big and the small constructor families are tagged,
-- that is, greater unions which overflow the tag space of TAG_BITS
-- (i.e. 3 on 32 resp. 7 constructors on 64 bit archs).
--
-- For example, let's assume a 64-bit architecture, so that all
-- heap objects are 8-byte aligned, and hence the address of a
-- heap object ends in `000` (three zero bits).  Then consider
-- > data Maybe a = Nothing | Just a
-- > data Day a = Mon | Tue | Wed | Thu | Fri | Sat | Sun
-- > data Grade = G1 | G2 | G3 | G4 | G5 | G6 | G7 | G8 | G9 | G10
--
-- Since `Grade` has more than 7 constructors, it counts as a
-- "big data type" (also referred to as "big constructor family" in papers).
-- On the other hand, `Maybe` and `Day` have 7 constructors or fewer, so they
-- are "small data types".
--
-- Then
--   * A pointer to an unevaluated thunk of type `Maybe Int`, `Day` or `Grade` will end in `000`
--   * A tagged pointer to a `Nothing`, `Mon` or `G1` will end in `001`
--   * A tagged pointer to a `Just x`, `Tue` or `G2`  will end in `010`
--   * A tagged pointer to `Wed` or `G3` will end in `011`
--       ...
--   * A tagged pointer to `Sat` or `G6` will end in `110`
--   * A tagged pointer to `Sun` or `G7` or `G8` or `G9` or `G10` will end in `111`
--
-- For big families we employ a mildly clever way of combining pointer and
-- info-table tagging. We use 1..MAX_PTR_TAG-1 as pointer-resident tags where
-- the tags in the pointer and the info table are in a one-to-one
-- relation, whereas tag MAX_PTR_TAG is used as "spill over", signifying
-- we have to fall back and get the precise constructor tag from the
-- info-table.
--
-- Consequently we now cascade switches, because we have to check
-- the pointer tag first, and when it is MAX_PTR_TAG, fetch the precise
-- tag from the info table, and switch on that. The only technically
-- tricky part is that the default case needs (logical) duplication.
-- To do this we emit an extra label for it and branch to that from
-- the second switch. This avoids duplicated codegen. See Trac #14373.
-- See note [Double switching for big families] for the mechanics
-- involved.
--
-- Also see note [Data constructor dynamic tags]
-- and the wiki https://gitlab.haskell.org/ghc/ghc/wikis/commentary/rts/haskell-execution/pointer-tagging
--

-------------------
cgAlgAltRhss :: (GcPlan,ReturnKind) -> NonVoid Id -> [CgStgAlt]
             -> FCode ( Maybe CmmAGraphScoped
                      , [(ConTagZ, CmmAGraphScoped)] )
cgAlgAltRhss :: (GcPlan, ReturnKind)
-> NonVoid Id
-> [GenStgAlt 'CodeGen]
-> FCode (Maybe CmmAGraphScoped, [(RepArity, CmmAGraphScoped)])
cgAlgAltRhss (GcPlan, ReturnKind)
gc_plan NonVoid Id
bndr [GenStgAlt 'CodeGen]
alts
  = do { [(AltCon, CmmAGraphScoped)]
tagged_cmms <- (GcPlan, ReturnKind)
-> NonVoid Id
-> [GenStgAlt 'CodeGen]
-> FCode [(AltCon, CmmAGraphScoped)]
cgAltRhss (GcPlan, ReturnKind)
gc_plan NonVoid Id
bndr [GenStgAlt 'CodeGen]
alts

       ; let { mb_deflt :: Maybe CmmAGraphScoped
mb_deflt = case [(AltCon, CmmAGraphScoped)]
tagged_cmms of
                           ((AltCon
DEFAULT,CmmAGraphScoped
rhs) : [(AltCon, CmmAGraphScoped)]
_) -> forall a. a -> Maybe a
Just CmmAGraphScoped
rhs
                           [(AltCon, CmmAGraphScoped)]
_other              -> forall a. Maybe a
Nothing
                            -- DEFAULT is always first, if present

              ; branches :: [(RepArity, CmmAGraphScoped)]
branches = [ (DataCon -> RepArity
dataConTagZ DataCon
con, CmmAGraphScoped
cmm)
                           | (DataAlt DataCon
con, CmmAGraphScoped
cmm) <- [(AltCon, CmmAGraphScoped)]
tagged_cmms ]
              }

       ; forall (m :: * -> *) a. Monad m => a -> m a
return (Maybe CmmAGraphScoped
mb_deflt, [(RepArity, CmmAGraphScoped)]
branches)
       }


-------------------
cgAltRhss :: (GcPlan,ReturnKind) -> NonVoid Id -> [CgStgAlt]
          -> FCode [(AltCon, CmmAGraphScoped)]
cgAltRhss :: (GcPlan, ReturnKind)
-> NonVoid Id
-> [GenStgAlt 'CodeGen]
-> FCode [(AltCon, CmmAGraphScoped)]
cgAltRhss (GcPlan, ReturnKind)
gc_plan NonVoid Id
bndr [GenStgAlt 'CodeGen]
alts = do
  Platform
platform <- FCode Platform
getPlatform
  let
    base_reg :: LocalReg
base_reg = Platform -> NonVoid Id -> LocalReg
idToReg Platform
platform NonVoid Id
bndr
    cg_alt :: CgStgAlt -> FCode (AltCon, CmmAGraphScoped)
    cg_alt :: GenStgAlt 'CodeGen -> FCode (AltCon, CmmAGraphScoped)
cg_alt (AltCon
con, [BinderP 'CodeGen]
bndrs, CgStgExpr
rhs)
      = forall a. FCode a -> FCode (a, CmmAGraphScoped)
getCodeScoped             forall a b. (a -> b) -> a -> b
$
        forall a. (GcPlan, ReturnKind) -> FCode a -> FCode a
maybeAltHeapCheck (GcPlan, ReturnKind)
gc_plan forall a b. (a -> b) -> a -> b
$
        do { [LocalReg]
_ <- AltCon -> LocalReg -> [NonVoid Id] -> FCode [LocalReg]
bindConArgs AltCon
con LocalReg
base_reg ([Id] -> [NonVoid Id]
assertNonVoidIds [BinderP 'CodeGen]
bndrs)
                    -- alt binders are always non-void,
                    -- see Note [Post-unarisation invariants] in GHC.Stg.Unarise
           ; ReturnKind
_ <- CgStgExpr -> FCode ReturnKind
cgExpr CgStgExpr
rhs
           ; forall (m :: * -> *) a. Monad m => a -> m a
return AltCon
con }
  forall a. [FCode a] -> FCode [a]
forkAlts (forall a b. (a -> b) -> [a] -> [b]
map GenStgAlt 'CodeGen -> FCode (AltCon, CmmAGraphScoped)
cg_alt [GenStgAlt 'CodeGen]
alts)

maybeAltHeapCheck :: (GcPlan,ReturnKind) -> FCode a -> FCode a
maybeAltHeapCheck :: forall a. (GcPlan, ReturnKind) -> FCode a -> FCode a
maybeAltHeapCheck (GcPlan
NoGcInAlts,ReturnKind
_)  FCode a
code = FCode a
code
maybeAltHeapCheck (GcInAlts [LocalReg]
regs, ReturnKind
AssignedDirectly) FCode a
code =
  forall a. [LocalReg] -> FCode a -> FCode a
altHeapCheck [LocalReg]
regs FCode a
code
maybeAltHeapCheck (GcInAlts [LocalReg]
regs, ReturnedTo BlockId
lret RepArity
off) FCode a
code =
  forall a. [LocalReg] -> BlockId -> RepArity -> FCode a -> FCode a
altHeapCheckReturnsTo [LocalReg]
regs BlockId
lret RepArity
off FCode a
code

-----------------------------------------------------------------------------
--      Tail calls
-----------------------------------------------------------------------------

cgConApp :: DataCon -> ConstructorNumber -> [StgArg] -> FCode ReturnKind
cgConApp :: DataCon -> ConstructorNumber -> [StgArg] -> FCode ReturnKind
cgConApp DataCon
con ConstructorNumber
mn [StgArg]
stg_args
  | DataCon -> Bool
isUnboxedTupleDataCon DataCon
con       -- Unboxed tuple: assign and return
  = do { [CmmExpr]
arg_exprs <- [StgArg] -> FCode [CmmExpr]
getNonVoidArgAmodes [StgArg]
stg_args
       ; RepArity -> FCode ()
tickyUnboxedTupleReturn (forall (t :: * -> *) a. Foldable t => t a -> RepArity
length [CmmExpr]
arg_exprs)
       ; [CmmExpr] -> FCode ReturnKind
emitReturn [CmmExpr]
arg_exprs }

  | Bool
otherwise   --  Boxed constructors; allocate and return
  = ASSERT2( stg_args `lengthIs` countConRepArgs con, ppr con <> parens (ppr (countConRepArgs con)) <+> ppr stg_args )
    do  { (CgIdInfo
idinfo, FCode CmmAGraph
fcode_init) <- Id
-> ConstructorNumber
-> Bool
-> CostCentreStack
-> DataCon
-> [NonVoid StgArg]
-> FCode (CgIdInfo, FCode CmmAGraph)
buildDynCon (DataCon -> Id
dataConWorkId DataCon
con) ConstructorNumber
mn Bool
False
                                     CostCentreStack
currentCCS DataCon
con ([StgArg] -> [NonVoid StgArg]
assertNonVoidStgArgs [StgArg]
stg_args)
                                     -- con args are always non-void,
                                     -- see Note [Post-unarisation invariants] in GHC.Stg.Unarise
                -- The first "con" says that the name bound to this
                -- closure is "con", which is a bit of a fudge, but
                -- it only affects profiling (hence the False)

        ; CmmAGraph -> FCode ()
emit forall (m :: * -> *) a b. Monad m => (a -> m b) -> m a -> m b
=<< FCode CmmAGraph
fcode_init
        ; RepArity -> FCode ()
tickyReturnNewCon (forall (t :: * -> *) a. Foldable t => t a -> RepArity
length [StgArg]
stg_args)
        ; [CmmExpr] -> FCode ReturnKind
emitReturn [CgIdInfo -> CmmExpr
idInfoToAmode CgIdInfo
idinfo] }

cgIdApp :: Id -> [StgArg] -> FCode ReturnKind
cgIdApp :: Id -> [StgArg] -> FCode ReturnKind
cgIdApp Id
fun_id [StgArg]
args = do
    CgIdInfo
fun_info       <- Id -> FCode CgIdInfo
getCgIdInfo Id
fun_id
    Maybe SelfLoopInfo
self_loop_info <- FCode (Maybe SelfLoopInfo)
getSelfLoop
    CallOpts
call_opts      <- FCode CallOpts
getCallOpts
    Profile
profile        <- FCode Profile
getProfile
    let fun_arg :: StgArg
fun_arg     = Id -> StgArg
StgVarArg Id
fun_id
        fun_name :: Name
fun_name    = Id -> Name
idName    Id
fun_id
        fun :: CmmExpr
fun         = CgIdInfo -> CmmExpr
idInfoToAmode CgIdInfo
fun_info
        lf_info :: LambdaFormInfo
lf_info     = CgIdInfo -> LambdaFormInfo
cg_lf         CgIdInfo
fun_info
        n_args :: RepArity
n_args      = forall (t :: * -> *) a. Foldable t => t a -> RepArity
length [StgArg]
args
        v_args :: RepArity
v_args      = forall (t :: * -> *) a. Foldable t => t a -> RepArity
length forall a b. (a -> b) -> a -> b
$ forall a. (a -> Bool) -> [a] -> [a]
filter (Type -> Bool
isVoidTy forall b c a. (b -> c) -> (a -> b) -> a -> c
. StgArg -> Type
stgArgType) [StgArg]
args
    case CallOpts
-> Name
-> Id
-> LambdaFormInfo
-> RepArity
-> RepArity
-> CgLoc
-> Maybe SelfLoopInfo
-> CallMethod
getCallMethod CallOpts
call_opts Name
fun_name Id
fun_id LambdaFormInfo
lf_info RepArity
n_args RepArity
v_args (CgIdInfo -> CgLoc
cg_loc CgIdInfo
fun_info) Maybe SelfLoopInfo
self_loop_info of
            -- A value in WHNF, so we can just return it.
        CallMethod
ReturnIt
          | Type -> Bool
isVoidTy (Id -> Type
idType Id
fun_id) -> [CmmExpr] -> FCode ReturnKind
emitReturn []
          | Bool
otherwise                -> [CmmExpr] -> FCode ReturnKind
emitReturn [CmmExpr
fun]
          -- ToDo: does ReturnIt guarantee tagged?

        CallMethod
EnterIt -> ASSERT( null args )  -- Discarding arguments
                   CmmExpr -> FCode ReturnKind
emitEnter CmmExpr
fun

        CallMethod
SlowCall -> do      -- A slow function call via the RTS apply routines
                { LambdaFormInfo -> [StgArg] -> FCode ()
tickySlowCall LambdaFormInfo
lf_info [StgArg]
args
                ; FastString -> FCode ()
emitComment forall a b. (a -> b) -> a -> b
$ String -> FastString
mkFastString String
"slowCall"
                ; CmmExpr -> [StgArg] -> FCode ReturnKind
slowCall CmmExpr
fun [StgArg]
args }

        -- A direct function call (possibly with some left-over arguments)
        DirectEntry CLabel
lbl RepArity
arity -> do
                { RepArity -> [StgArg] -> FCode ()
tickyDirectCall RepArity
arity [StgArg]
args
                ; if Profile -> LambdaFormInfo -> Bool
nodeMustPointToIt Profile
profile LambdaFormInfo
lf_info
                     then Convention -> CLabel -> RepArity -> [StgArg] -> FCode ReturnKind
directCall Convention
NativeNodeCall   CLabel
lbl RepArity
arity (StgArg
fun_argforall a. a -> [a] -> [a]
:[StgArg]
args)
                     else Convention -> CLabel -> RepArity -> [StgArg] -> FCode ReturnKind
directCall Convention
NativeDirectCall CLabel
lbl RepArity
arity [StgArg]
args }

        -- Let-no-escape call or self-recursive tail-call
        JumpToIt BlockId
blk_id [LocalReg]
lne_regs -> do
          { FCode ()
adjustHpBackwards -- always do this before a tail-call
          ; [CmmExpr]
cmm_args <- [StgArg] -> FCode [CmmExpr]
getNonVoidArgAmodes [StgArg]
args
          ; [LocalReg] -> [CmmExpr] -> FCode ()
emitMultiAssign [LocalReg]
lne_regs [CmmExpr]
cmm_args
          ; CmmAGraph -> FCode ()
emit (BlockId -> CmmAGraph
mkBranch BlockId
blk_id)
          ; forall (m :: * -> *) a. Monad m => a -> m a
return ReturnKind
AssignedDirectly }

-- Note [Self-recursive tail calls]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- Self-recursive tail calls can be optimized into a local jump in the same
-- way as let-no-escape bindings (see Note [What is a non-escaping let] in
-- "GHC.CoreToStg"). Consider this:
--
-- foo.info:
--     a = R1  // calling convention
--     b = R2
--     goto L1
-- L1: ...
--     ...
-- ...
-- L2: R1 = x
--     R2 = y
--     call foo(R1,R2)
--
-- Instead of putting x and y into registers (or other locations required by the
-- calling convention) and performing a call we can put them into local
-- variables a and b and perform jump to L1:
--
-- foo.info:
--     a = R1
--     b = R2
--     goto L1
-- L1: ...
--     ...
-- ...
-- L2: a = x
--     b = y
--     goto L1
--
-- This can be done only when function is calling itself in a tail position
-- and only if the call passes number of parameters equal to function's arity.
-- Note that this cannot be performed if a function calls itself with a
-- continuation.
--
-- This in fact implements optimization known as "loopification". It was
-- described in "Low-level code optimizations in the Glasgow Haskell Compiler"
-- by Krzysztof Woś, though we use different approach. Krzysztof performed his
-- optimization at the Cmm level, whereas we perform ours during code generation
-- (Stg-to-Cmm pass) essentially making sure that optimized Cmm code is
-- generated in the first place.
--
-- Implementation is spread across a couple of places in the code:
--
--   * FCode monad stores additional information in its reader environment
--     (cgd_self_loop field). This information tells us which function can
--     tail call itself in an optimized way (it is the function currently
--     being compiled), what is the label of a loop header (L1 in example above)
--     and information about local registers in which we should arguments
--     before making a call (this would be a and b in example above).
--
--   * Whenever we are compiling a function, we set that information to reflect
--     the fact that function currently being compiled can be jumped to, instead
--     of called. This is done in closureCodyBody in GHC.StgToCmm.Bind.
--
--   * We also have to emit a label to which we will be jumping. We make sure
--     that the label is placed after a stack check but before the heap
--     check. The reason is that making a recursive tail-call does not increase
--     the stack so we only need to check once. But it may grow the heap, so we
--     have to repeat the heap check in every self-call. This is done in
--     do_checks in GHC.StgToCmm.Heap.
--
--   * When we begin compilation of another closure we remove the additional
--     information from the environment. This is done by forkClosureBody
--     in GHC.StgToCmm.Monad. Other functions that duplicate the environment -
--     forkLneBody, forkAlts, codeOnly - duplicate that information. In other
--     words, we only need to clean the environment of the self-loop information
--     when compiling right hand side of a closure (binding).
--
--   * When compiling a call (cgIdApp) we use getCallMethod to decide what kind
--     of call will be generated. getCallMethod decides to generate a self
--     recursive tail call when (a) environment stores information about
--     possible self tail-call; (b) that tail call is to a function currently
--     being compiled; (c) number of passed non-void arguments is equal to
--     function's arity. (d) loopification is turned on via -floopification
--     command-line option.
--
--   * Command line option to turn loopification on and off is implemented in
--     DynFlags.
--
--
-- Note [Void arguments in self-recursive tail calls]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- State# tokens can get in the way of the loopification optimization as seen in
-- #11372. Consider this:
--
-- foo :: [a]
--     -> (a -> State# s -> (# State s, Bool #))
--     -> State# s
--     -> (# State# s, Maybe a #)
-- foo [] f s = (# s, Nothing #)
-- foo (x:xs) f s = case f x s of
--      (# s', b #) -> case b of
--          True -> (# s', Just x #)
--          False -> foo xs f s'
--
-- We would like to compile the call to foo as a local jump instead of a call
-- (see Note [Self-recursive tail calls]). However, the generated function has
-- an arity of 2 while we apply it to 3 arguments, one of them being of void
-- type. Thus, we mustn't count arguments of void type when checking whether
-- we can turn a call into a self-recursive jump.
--

emitEnter :: CmmExpr -> FCode ReturnKind
emitEnter :: CmmExpr -> FCode ReturnKind
emitEnter CmmExpr
fun = do
  { PtrOpts
ptr_opts <- FCode PtrOpts
getPtrOpts
  ; Platform
platform <- FCode Platform
getPlatform
  ; Profile
profile <- FCode Profile
getProfile
  ; FCode ()
adjustHpBackwards
  ; Sequel
sequel <- FCode Sequel
getSequel
  ; RepArity
updfr_off <- FCode RepArity
getUpdFrameOff
  ; case Sequel
sequel of
      -- For a return, we have the option of generating a tag-test or
      -- not.  If the value is tagged, we can return directly, which
      -- is quicker than entering the value.  This is a code
      -- size/speed trade-off: when optimising for speed rather than
      -- size we could generate the tag test.
      --
      -- Right now, we do what the old codegen did, and omit the tag
      -- test, just generating an enter.
      Sequel
Return -> do
        { let entry :: CmmExpr
entry = Platform -> CmmExpr -> CmmExpr
entryCode Platform
platform forall a b. (a -> b) -> a -> b
$ PtrOpts -> CmmExpr -> CmmExpr
closureInfoPtr PtrOpts
ptr_opts forall a b. (a -> b) -> a -> b
$ CmmReg -> CmmExpr
CmmReg CmmReg
nodeReg
        ; CmmAGraph -> FCode ()
emit forall a b. (a -> b) -> a -> b
$ Profile
-> Convention -> CmmExpr -> [CmmExpr] -> RepArity -> CmmAGraph
mkJump Profile
profile Convention
NativeNodeCall CmmExpr
entry
                        [Platform -> CmmExpr -> CmmExpr
cmmUntag Platform
platform CmmExpr
fun] RepArity
updfr_off
        ; forall (m :: * -> *) a. Monad m => a -> m a
return ReturnKind
AssignedDirectly
        }

      -- The result will be scrutinised in the sequel.  This is where
      -- we generate a tag-test to avoid entering the closure if
      -- possible.
      --
      -- The generated code will be something like this:
      --
      --    R1 = fun  -- copyout
      --    if (fun & 7 != 0) goto Lret else goto Lcall
      --  Lcall:
      --    call [fun] returns to Lret
      --  Lret:
      --    fun' = R1  -- copyin
      --    ...
      --
      -- Note in particular that the label Lret is used as a
      -- destination by both the tag-test and the call.  This is
      -- because Lret will necessarily be a proc-point, and we want to
      -- ensure that we generate only one proc-point for this
      -- sequence.
      --
      -- Furthermore, we tell the caller that we generated a native
      -- return continuation by returning (ReturnedTo Lret off), so
      -- that the continuation can be reused by the heap-check failure
      -- code in the enclosing case expression.
      --
      AssignTo [LocalReg]
res_regs Bool
_ -> do
       { BlockId
lret <- forall (m :: * -> *). MonadUnique m => m BlockId
newBlockId
       ; let (RepArity
off, [GlobalReg]
_, CmmAGraph
copyin) = Profile
-> Convention
-> Area
-> [LocalReg]
-> [LocalReg]
-> (RepArity, [GlobalReg], CmmAGraph)
copyInOflow Profile
profile Convention
NativeReturn (BlockId -> Area
Young BlockId
lret) [LocalReg]
res_regs []
       ; BlockId
lcall <- forall (m :: * -> *). MonadUnique m => m BlockId
newBlockId
       ; RepArity
updfr_off <- FCode RepArity
getUpdFrameOff
       ; let area :: Area
area = BlockId -> Area
Young BlockId
lret
       ; let (RepArity
outArgs, [GlobalReg]
regs, CmmAGraph
copyout) = Profile
-> Convention
-> Transfer
-> Area
-> [CmmExpr]
-> RepArity
-> [CmmExpr]
-> (RepArity, [GlobalReg], CmmAGraph)
copyOutOflow Profile
profile Convention
NativeNodeCall Transfer
Call Area
area
                                          [CmmExpr
fun] RepArity
updfr_off []
         -- refer to fun via nodeReg after the copyout, to avoid having
         -- both live simultaneously; this sometimes enables fun to be
         -- inlined in the RHS of the R1 assignment.
       ; let entry :: CmmExpr
entry = Platform -> CmmExpr -> CmmExpr
entryCode Platform
platform (PtrOpts -> CmmExpr -> CmmExpr
closureInfoPtr PtrOpts
ptr_opts (CmmReg -> CmmExpr
CmmReg CmmReg
nodeReg))
             the_call :: CmmAGraph
the_call = CmmExpr
-> Maybe BlockId
-> RepArity
-> RepArity
-> RepArity
-> [GlobalReg]
-> CmmAGraph
toCall CmmExpr
entry (forall a. a -> Maybe a
Just BlockId
lret) RepArity
updfr_off RepArity
off RepArity
outArgs [GlobalReg]
regs
       ; CmmTickScope
tscope <- FCode CmmTickScope
getTickScope
       ; CmmAGraph -> FCode ()
emit forall a b. (a -> b) -> a -> b
$
           CmmAGraph
copyout CmmAGraph -> CmmAGraph -> CmmAGraph
<*>
           CmmExpr -> BlockId -> BlockId -> Maybe Bool -> CmmAGraph
mkCbranch (Platform -> CmmExpr -> CmmExpr
cmmIsTagged Platform
platform (CmmReg -> CmmExpr
CmmReg CmmReg
nodeReg))
                     BlockId
lret BlockId
lcall forall a. Maybe a
Nothing CmmAGraph -> CmmAGraph -> CmmAGraph
<*>
           BlockId -> CmmAGraphScoped -> CmmAGraph
outOfLine BlockId
lcall (CmmAGraph
the_call,CmmTickScope
tscope) CmmAGraph -> CmmAGraph -> CmmAGraph
<*>
           BlockId -> CmmTickScope -> CmmAGraph
mkLabel BlockId
lret CmmTickScope
tscope CmmAGraph -> CmmAGraph -> CmmAGraph
<*>
           CmmAGraph
copyin
       ; forall (m :: * -> *) a. Monad m => a -> m a
return (BlockId -> RepArity -> ReturnKind
ReturnedTo BlockId
lret RepArity
off)
       }
  }

------------------------------------------------------------------------
--              Ticks
------------------------------------------------------------------------

-- | Generate Cmm code for a tick. Depending on the type of Tickish,
-- this will either generate actual Cmm instrumentation code, or
-- simply pass on the annotation as a @CmmTickish@.
cgTick :: StgTickish -> FCode ()
cgTick :: StgTickish -> FCode ()
cgTick StgTickish
tick
  = do { Platform
platform <- FCode Platform
getPlatform
       ; case StgTickish
tick of
           ProfNote   CostCentre
cc Bool
t Bool
p -> CostCentre -> Bool -> Bool -> FCode ()
emitSetCCC CostCentre
cc Bool
t Bool
p
           HpcTick    Module
m RepArity
n    -> CmmAGraph -> FCode ()
emit (Platform -> Module -> RepArity -> CmmAGraph
mkTickBox Platform
platform Module
m RepArity
n)
           SourceNote RealSrcSpan
s String
n    -> CmmTickish -> FCode ()
emitTick forall a b. (a -> b) -> a -> b
$ forall (pass :: TickishPass).
RealSrcSpan -> String -> GenTickish pass
SourceNote RealSrcSpan
s String
n
           StgTickish
_other            -> forall (m :: * -> *) a. Monad m => a -> m a
return () -- ignore
       }