ghc-9.2.2: GHC/CmmToAsm.hs
-- -----------------------------------------------------------------------------
--
-- (c) The University of Glasgow 1993-2004
--
--
-- -----------------------------------------------------------------------------
{-# LANGUAGE BangPatterns, CPP, GADTs, ScopedTypeVariables, PatternSynonyms,
DeriveFunctor #-}
{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE UnboxedTuples #-}
{-# LANGUAGE ScopedTypeVariables #-}
{-# OPTIONS_GHC -Wno-incomplete-record-updates #-}
-- | Native code generator
--
-- The native-code generator has machine-independent and
-- machine-dependent modules.
--
-- This module ("GHC.CmmToAsm") is the top-level machine-independent
-- module. Before entering machine-dependent land, we do some
-- machine-independent optimisations (defined below) on the
-- 'CmmStmts's.
--
-- We convert to the machine-specific 'Instr' datatype with
-- 'cmmCodeGen', assuming an infinite supply of registers. We then use
-- a machine-independent register allocator ('regAlloc') to rejoin
-- reality. Obviously, 'regAlloc' has machine-specific helper
-- functions (see about "RegAllocInfo" below).
--
-- Finally, we order the basic blocks of the function so as to minimise
-- the number of jumps between blocks, by utilising fallthrough wherever
-- possible.
--
-- The machine-dependent bits break down as follows:
--
-- * ["MachRegs"] Everything about the target platform's machine
-- registers (and immediate operands, and addresses, which tend to
-- intermingle/interact with registers).
--
-- * ["MachInstrs"] Includes the 'Instr' datatype (possibly should
-- have a module of its own), plus a miscellany of other things
-- (e.g., 'targetDoubleSize', 'smStablePtrTable', ...)
--
-- * ["MachCodeGen"] is where 'Cmm' stuff turns into
-- machine instructions.
--
-- * ["PprMach"] 'pprInstr' turns an 'Instr' into text (well, really
-- a 'SDoc').
--
-- * ["RegAllocInfo"] In the register allocator, we manipulate
-- 'MRegsState's, which are 'BitSet's, one bit per machine register.
-- When we want to say something about a specific machine register
-- (e.g., ``it gets clobbered by this instruction''), we set/unset
-- its bit. Obviously, we do this 'BitSet' thing for efficiency
-- reasons.
--
-- The 'RegAllocInfo' module collects together the machine-specific
-- info needed to do register allocation.
--
-- * ["RegisterAlloc"] The (machine-independent) register allocator.
-- -}
--
module GHC.CmmToAsm
( nativeCodeGen
-- * Test-only exports: see trac #12744
-- used by testGraphNoSpills, which needs to access
-- the register allocator intermediate data structures
-- cmmNativeGen emits
, cmmNativeGen
, NcgImpl(..)
, initNCGConfig
)
where
#include "HsVersions.h"
import GHC.Prelude
import qualified GHC.CmmToAsm.X86 as X86
import qualified GHC.CmmToAsm.PPC as PPC
import qualified GHC.CmmToAsm.SPARC as SPARC
import qualified GHC.CmmToAsm.AArch64 as AArch64
import GHC.CmmToAsm.Reg.Liveness
import qualified GHC.CmmToAsm.Reg.Linear as Linear
import qualified GHC.Data.Graph.Color as Color
import qualified GHC.CmmToAsm.Reg.Graph as Color
import qualified GHC.CmmToAsm.Reg.Graph.Stats as Color
import qualified GHC.CmmToAsm.Reg.Graph.TrivColorable as Color
import GHC.Utils.Asm
import GHC.CmmToAsm.Reg.Target
import GHC.Platform
import GHC.CmmToAsm.BlockLayout as BlockLayout
import GHC.Settings.Config
import GHC.CmmToAsm.Instr
import GHC.CmmToAsm.PIC
import GHC.Platform.Reg
import GHC.Platform.Reg.Class (RegClass)
import GHC.CmmToAsm.Monad
import GHC.CmmToAsm.CFG
import GHC.CmmToAsm.Dwarf
import GHC.CmmToAsm.Config
import GHC.CmmToAsm.Types
import GHC.Cmm.DebugBlock
import GHC.Cmm.BlockId
import GHC.StgToCmm.CgUtils ( fixStgRegisters )
import GHC.Cmm
import GHC.Cmm.Utils
import GHC.Cmm.Dataflow.Collections
import GHC.Cmm.Dataflow.Label
import GHC.Cmm.Dataflow.Block
import GHC.Cmm.Opt ( cmmMachOpFold )
import GHC.Cmm.Ppr
import GHC.Cmm.CLabel
import GHC.Types.Unique.FM
import GHC.Types.Unique.Supply
import GHC.Driver.Session
import GHC.Driver.Ppr
import GHC.Utils.Misc
import GHC.Utils.Logger
import qualified GHC.Utils.Ppr as Pretty
import GHC.Utils.BufHandle
import GHC.Utils.Outputable as Outputable
import GHC.Utils.Panic
import GHC.Data.FastString
import GHC.Types.Unique.Set
import GHC.Utils.Error
import GHC.Unit
import GHC.Data.Stream (Stream)
import qualified GHC.Data.Stream as Stream
import Data.List (sortBy, groupBy)
import Data.Maybe
import Data.Ord ( comparing )
import Control.Exception
import Control.Monad
import System.IO
--------------------
nativeCodeGen :: forall a . Logger -> DynFlags -> Module -> ModLocation -> Handle -> UniqSupply
-> Stream IO RawCmmGroup a
-> IO a
nativeCodeGen logger dflags this_mod modLoc h us cmms
= let config = initNCGConfig dflags this_mod
platform = ncgPlatform config
nCG' :: ( OutputableP Platform statics, Outputable jumpDest, Instruction instr)
=> NcgImpl statics instr jumpDest -> IO a
nCG' ncgImpl = nativeCodeGen' logger dflags config modLoc ncgImpl h us cmms
in case platformArch platform of
ArchX86 -> nCG' (X86.ncgX86 config)
ArchX86_64 -> nCG' (X86.ncgX86_64 config)
ArchPPC -> nCG' (PPC.ncgPPC config)
ArchPPC_64 _ -> nCG' (PPC.ncgPPC config)
ArchSPARC -> nCG' (SPARC.ncgSPARC config)
ArchSPARC64 -> panic "nativeCodeGen: No NCG for SPARC64"
ArchS390X -> panic "nativeCodeGen: No NCG for S390X"
ArchARM {} -> panic "nativeCodeGen: No NCG for ARM"
ArchAArch64 -> nCG' (AArch64.ncgAArch64 config)
ArchAlpha -> panic "nativeCodeGen: No NCG for Alpha"
ArchMipseb -> panic "nativeCodeGen: No NCG for mipseb"
ArchMipsel -> panic "nativeCodeGen: No NCG for mipsel"
ArchRISCV64 -> panic "nativeCodeGen: No NCG for RISCV64"
ArchUnknown -> panic "nativeCodeGen: No NCG for unknown arch"
ArchJavaScript-> panic "nativeCodeGen: No NCG for JavaScript"
-- | Data accumulated during code generation. Mostly about statistics,
-- but also collects debug data for DWARF generation.
data NativeGenAcc statics instr
= NGS { ngs_imports :: ![[CLabel]]
, ngs_natives :: ![[NatCmmDecl statics instr]]
-- ^ Native code generated, for statistics. This might
-- hold a lot of data, so it is important to clear this
-- field as early as possible if it isn't actually
-- required.
, ngs_colorStats :: ![[Color.RegAllocStats statics instr]]
, ngs_linearStats :: ![[Linear.RegAllocStats]]
, ngs_labels :: ![Label]
, ngs_debug :: ![DebugBlock]
, ngs_dwarfFiles :: !DwarfFiles
, ngs_unwinds :: !(LabelMap [UnwindPoint])
-- ^ see Note [Unwinding information in the NCG]
-- and Note [What is this unwinding business?] in "GHC.Cmm.DebugBlock".
}
{-
Note [Unwinding information in the NCG]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Unwind information is a type of metadata which allows a debugging tool
to reconstruct the values of machine registers at the time a procedure was
entered. For the most part, the production of unwind information is handled by
the Cmm stage, where it is represented by CmmUnwind nodes.
Unfortunately, the Cmm stage doesn't know everything necessary to produce
accurate unwinding information. For instance, the x86-64 calling convention
requires that the stack pointer be aligned to 16 bytes, which in turn means that
GHC must sometimes add padding to $sp prior to performing a foreign call. When
this happens unwind information must be updated accordingly.
For this reason, we make the NCG backends responsible for producing
unwinding tables (with the extractUnwindPoints function in NcgImpl).
We accumulate the produced unwind tables over CmmGroups in the ngs_unwinds
field of NativeGenAcc. This is a label map which contains an entry for each
procedure, containing a list of unwinding points (e.g. a label and an associated
unwinding table).
See also Note [What is this unwinding business?] in "GHC.Cmm.DebugBlock".
-}
nativeCodeGen' :: (OutputableP Platform statics, Outputable jumpDest, Instruction instr)
=> Logger
-> DynFlags
-> NCGConfig
-> ModLocation
-> NcgImpl statics instr jumpDest
-> Handle
-> UniqSupply
-> Stream IO RawCmmGroup a
-> IO a
nativeCodeGen' logger dflags config modLoc ncgImpl h us cmms
= do
-- BufHandle is a performance hack. We could hide it inside
-- Pretty if it weren't for the fact that we do lots of little
-- printDocs here (in order to do codegen in constant space).
bufh <- newBufHandle h
let ngs0 = NGS [] [] [] [] [] [] emptyUFM mapEmpty
(ngs, us', a) <- cmmNativeGenStream logger dflags config modLoc ncgImpl bufh us
cmms ngs0
_ <- finishNativeGen logger dflags config modLoc bufh us' ngs
return a
finishNativeGen :: Instruction instr
=> Logger
-> DynFlags
-> NCGConfig
-> ModLocation
-> BufHandle
-> UniqSupply
-> NativeGenAcc statics instr
-> IO UniqSupply
finishNativeGen logger dflags config modLoc bufh@(BufHandle _ _ h) us ngs
= withTimingSilent logger dflags (text "NCG") (`seq` ()) $ do
-- Write debug data and finish
us' <- if not (ncgDwarfEnabled config)
then return us
else do
(dwarf, us') <- dwarfGen config modLoc us (ngs_debug ngs)
emitNativeCode logger dflags config bufh dwarf
return us'
bFlush bufh
-- dump global NCG stats for graph coloring allocator
let stats = concat (ngs_colorStats ngs)
unless (null stats) $ do
-- build the global register conflict graph
let graphGlobal
= foldl' Color.union Color.initGraph
$ [ Color.raGraph stat
| stat@Color.RegAllocStatsStart{} <- stats]
dump_stats (Color.pprStats stats graphGlobal)
let platform = ncgPlatform config
dumpIfSet_dyn logger dflags
Opt_D_dump_asm_conflicts "Register conflict graph"
FormatText
$ Color.dotGraph
(targetRegDotColor platform)
(Color.trivColorable platform
(targetVirtualRegSqueeze platform)
(targetRealRegSqueeze platform))
$ graphGlobal
-- dump global NCG stats for linear allocator
let linearStats = concat (ngs_linearStats ngs)
unless (null linearStats) $
dump_stats (Linear.pprStats (concat (ngs_natives ngs)) linearStats)
-- write out the imports
let ctx = ncgAsmContext config
printSDocLn ctx Pretty.LeftMode h
$ makeImportsDoc config (concat (ngs_imports ngs))
return us'
where
dump_stats = putDumpMsg logger dflags (mkDumpStyle alwaysQualify)
Opt_D_dump_asm_stats "NCG stats"
FormatText
cmmNativeGenStream :: forall statics jumpDest instr a . (OutputableP Platform statics, Outputable jumpDest, Instruction instr)
=> Logger
-> DynFlags
-> NCGConfig
-> ModLocation
-> NcgImpl statics instr jumpDest
-> BufHandle
-> UniqSupply
-> Stream.Stream IO RawCmmGroup a
-> NativeGenAcc statics instr
-> IO (NativeGenAcc statics instr, UniqSupply, a)
cmmNativeGenStream logger dflags config modLoc ncgImpl h us cmm_stream ngs
= loop us (Stream.runStream cmm_stream) ngs
where
ncglabel = text "NCG"
loop :: UniqSupply
-> Stream.StreamS IO RawCmmGroup a
-> NativeGenAcc statics instr
-> IO (NativeGenAcc statics instr, UniqSupply, a)
loop us s ngs =
case s of
Stream.Done a ->
return (ngs { ngs_imports = reverse $ ngs_imports ngs
, ngs_natives = reverse $ ngs_natives ngs
, ngs_colorStats = reverse $ ngs_colorStats ngs
, ngs_linearStats = reverse $ ngs_linearStats ngs
},
us,
a)
Stream.Effect m -> m >>= \cmm_stream' -> loop us cmm_stream' ngs
Stream.Yield cmms cmm_stream' -> do
(us', ngs'') <-
withTimingSilent logger
dflags
ncglabel (\(a, b) -> a `seq` b `seq` ()) $ do
-- Generate debug information
let !ndbgs | ncgDwarfEnabled config = cmmDebugGen modLoc cmms
| otherwise = []
dbgMap = debugToMap ndbgs
-- Generate native code
(ngs',us') <- cmmNativeGens logger dflags config modLoc ncgImpl h
dbgMap us cmms ngs 0
-- Link native code information into debug blocks
-- See Note [What is this unwinding business?] in "GHC.Cmm.DebugBlock".
let !ldbgs = cmmDebugLink (ngs_labels ngs') (ngs_unwinds ngs') ndbgs
platform = targetPlatform dflags
unless (null ldbgs) $
dumpIfSet_dyn logger dflags Opt_D_dump_debug "Debug Infos" FormatText
(vcat $ map (pdoc platform) ldbgs)
-- Accumulate debug information for emission in finishNativeGen.
let ngs'' = ngs' { ngs_debug = ngs_debug ngs' ++ ldbgs, ngs_labels = [] }
return (us', ngs'')
loop us' cmm_stream' ngs''
-- | Do native code generation on all these cmms.
--
cmmNativeGens :: forall statics instr jumpDest.
(OutputableP Platform statics, Outputable jumpDest, Instruction instr)
=> Logger
-> DynFlags
-> NCGConfig
-> ModLocation
-> NcgImpl statics instr jumpDest
-> BufHandle
-> LabelMap DebugBlock
-> UniqSupply
-> [RawCmmDecl]
-> NativeGenAcc statics instr
-> Int
-> IO (NativeGenAcc statics instr, UniqSupply)
cmmNativeGens logger dflags config modLoc ncgImpl h dbgMap = go
where
go :: UniqSupply -> [RawCmmDecl]
-> NativeGenAcc statics instr -> Int
-> IO (NativeGenAcc statics instr, UniqSupply)
go us [] ngs !_ =
return (ngs, us)
go us (cmm : cmms) ngs count = do
let fileIds = ngs_dwarfFiles ngs
(us', fileIds', native, imports, colorStats, linearStats, unwinds)
<- {-# SCC "cmmNativeGen" #-}
cmmNativeGen logger dflags modLoc ncgImpl us fileIds dbgMap
cmm count
-- Generate .file directives for every new file that has been
-- used. Note that it is important that we generate these in
-- ascending order, as Clang's 3.6 assembler complains.
let newFileIds = sortBy (comparing snd) $
nonDetEltsUFM $ fileIds' `minusUFM` fileIds
-- See Note [Unique Determinism and code generation]
pprDecl (f,n) = text "\t.file " <> ppr n <+>
pprFilePathString (unpackFS f)
emitNativeCode logger dflags config h $ vcat $
map pprDecl newFileIds ++
map (pprNatCmmDecl ncgImpl) native
-- force evaluation all this stuff to avoid space leaks
let platform = targetPlatform dflags
{-# SCC "seqString" #-} evaluate $ seqList (showSDoc dflags $ vcat $ map (pdoc platform) imports) ()
let !labels' = if ncgDwarfEnabled config
then cmmDebugLabels isMetaInstr native else []
!natives' = if dopt Opt_D_dump_asm_stats dflags
then native : ngs_natives ngs else []
mCon = maybe id (:)
ngs' = ngs{ ngs_imports = imports : ngs_imports ngs
, ngs_natives = natives'
, ngs_colorStats = colorStats `mCon` ngs_colorStats ngs
, ngs_linearStats = linearStats `mCon` ngs_linearStats ngs
, ngs_labels = ngs_labels ngs ++ labels'
, ngs_dwarfFiles = fileIds'
, ngs_unwinds = ngs_unwinds ngs `mapUnion` unwinds
}
go us' cmms ngs' (count + 1)
emitNativeCode :: Logger -> DynFlags -> NCGConfig -> BufHandle -> SDoc -> IO ()
emitNativeCode logger dflags config h sdoc = do
let ctx = ncgAsmContext config
{-# SCC "pprNativeCode" #-} bufLeftRenderSDoc ctx h sdoc
-- dump native code
dumpIfSet_dyn logger dflags
Opt_D_dump_asm "Asm code" FormatASM
sdoc
-- | Complete native code generation phase for a single top-level chunk of Cmm.
-- Dumping the output of each stage along the way.
-- Global conflict graph and NGC stats
cmmNativeGen
:: forall statics instr jumpDest. (Instruction instr, OutputableP Platform statics, Outputable jumpDest)
=> Logger
-> DynFlags
-> ModLocation
-> NcgImpl statics instr jumpDest
-> UniqSupply
-> DwarfFiles
-> LabelMap DebugBlock
-> RawCmmDecl -- ^ the cmm to generate code for
-> Int -- ^ sequence number of this top thing
-> IO ( UniqSupply
, DwarfFiles
, [NatCmmDecl statics instr] -- native code
, [CLabel] -- things imported by this cmm
, Maybe [Color.RegAllocStats statics instr] -- stats for the coloring register allocator
, Maybe [Linear.RegAllocStats] -- stats for the linear register allocators
, LabelMap [UnwindPoint] -- unwinding information for blocks
)
cmmNativeGen logger dflags modLoc ncgImpl us fileIds dbgMap cmm count
= do
let config = ncgConfig ncgImpl
let platform = ncgPlatform config
let weights = ncgCfgWeights config
let proc_name = case cmm of
(CmmProc _ entry_label _ _) -> pdoc platform entry_label
_ -> text "DataChunk"
-- rewrite assignments to global regs
let fixed_cmm =
{-# SCC "fixStgRegisters" #-}
fixStgRegisters platform cmm
-- cmm to cmm optimisations
let (opt_cmm, imports) =
{-# SCC "cmmToCmm" #-}
cmmToCmm config fixed_cmm
dumpIfSet_dyn logger dflags
Opt_D_dump_opt_cmm "Optimised Cmm" FormatCMM
(pprCmmGroup platform [opt_cmm])
let cmmCfg = {-# SCC "getCFG" #-}
getCfgProc platform weights opt_cmm
-- generate native code from cmm
let ((native, lastMinuteImports, fileIds', nativeCfgWeights), usGen) =
{-# SCC "genMachCode" #-}
initUs us $ genMachCode config modLoc
(cmmTopCodeGen ncgImpl)
fileIds dbgMap opt_cmm cmmCfg
dumpIfSet_dyn logger dflags
Opt_D_dump_asm_native "Native code" FormatASM
(vcat $ map (pprNatCmmDecl ncgImpl) native)
maybeDumpCfg logger dflags (Just nativeCfgWeights) "CFG Weights - Native" proc_name
-- tag instructions with register liveness information
-- also drops dead code. We don't keep the cfg in sync on
-- some backends, so don't use it there.
let livenessCfg = if backendMaintainsCfg platform
then Just nativeCfgWeights
else Nothing
let (withLiveness, usLive) =
{-# SCC "regLiveness" #-}
initUs usGen
$ mapM (cmmTopLiveness livenessCfg platform) native
dumpIfSet_dyn logger dflags
Opt_D_dump_asm_liveness "Liveness annotations added"
FormatCMM
(vcat $ map (pprLiveCmmDecl platform) withLiveness)
-- allocate registers
(alloced, usAlloc, ppr_raStatsColor, ppr_raStatsLinear, raStats, stack_updt_blks) <-
if ( gopt Opt_RegsGraph dflags
|| gopt Opt_RegsIterative dflags )
then do
-- the regs usable for allocation
let (alloc_regs :: UniqFM RegClass (UniqSet RealReg))
= foldr (\r -> plusUFM_C unionUniqSets
$ unitUFM (targetClassOfRealReg platform r) (unitUniqSet r))
emptyUFM
$ allocatableRegs ncgImpl
-- do the graph coloring register allocation
let ((alloced, maybe_more_stack, regAllocStats), usAlloc)
= {-# SCC "RegAlloc-color" #-}
initUs usLive
$ Color.regAlloc
config
alloc_regs
(mkUniqSet [0 .. maxSpillSlots ncgImpl])
(maxSpillSlots ncgImpl)
withLiveness
livenessCfg
let ((alloced', stack_updt_blks), usAlloc')
= initUs usAlloc $
case maybe_more_stack of
Nothing -> return (alloced, [])
Just amount -> do
(alloced',stack_updt_blks) <- unzip <$>
(mapM ((ncgAllocMoreStack ncgImpl) amount) alloced)
return (alloced', concat stack_updt_blks )
-- dump out what happened during register allocation
dumpIfSet_dyn logger dflags
Opt_D_dump_asm_regalloc "Registers allocated"
FormatCMM
(vcat $ map (pprNatCmmDecl ncgImpl) alloced)
dumpIfSet_dyn logger dflags
Opt_D_dump_asm_regalloc_stages "Build/spill stages"
FormatText
(vcat $ map (\(stage, stats)
-> text "# --------------------------"
$$ text "# cmm " <> int count <> text " Stage " <> int stage
$$ ppr (fmap (pprInstr platform) stats))
$ zip [0..] regAllocStats)
let mPprStats =
if dopt Opt_D_dump_asm_stats dflags
then Just regAllocStats else Nothing
-- force evaluation of the Maybe to avoid space leak
mPprStats `seq` return ()
return ( alloced', usAlloc'
, mPprStats
, Nothing
, [], stack_updt_blks)
else do
-- do linear register allocation
let reg_alloc proc = do
(alloced, maybe_more_stack, ra_stats) <-
Linear.regAlloc config proc
case maybe_more_stack of
Nothing -> return ( alloced, ra_stats, [] )
Just amount -> do
(alloced',stack_updt_blks) <-
ncgAllocMoreStack ncgImpl amount alloced
return (alloced', ra_stats, stack_updt_blks )
let ((alloced, regAllocStats, stack_updt_blks), usAlloc)
= {-# SCC "RegAlloc-linear" #-}
initUs usLive
$ liftM unzip3
$ mapM reg_alloc withLiveness
dumpIfSet_dyn logger dflags
Opt_D_dump_asm_regalloc "Registers allocated"
FormatCMM
(vcat $ map (pprNatCmmDecl ncgImpl) alloced)
let mPprStats =
if dopt Opt_D_dump_asm_stats dflags
then Just (catMaybes regAllocStats) else Nothing
-- force evaluation of the Maybe to avoid space leak
mPprStats `seq` return ()
return ( alloced, usAlloc
, Nothing
, mPprStats, (catMaybes regAllocStats)
, concat stack_updt_blks )
-- Fixupblocks the register allocator inserted (from, regMoves, to)
let cfgRegAllocUpdates :: [(BlockId,BlockId,BlockId)]
cfgRegAllocUpdates = (concatMap Linear.ra_fixupList raStats)
let cfgWithFixupBlks =
(\cfg -> addNodesBetween weights cfg cfgRegAllocUpdates) <$> livenessCfg
-- Insert stack update blocks
let postRegCFG =
pure (foldl' (\m (from,to) -> addImmediateSuccessor weights from to m ))
<*> cfgWithFixupBlks
<*> pure stack_updt_blks
---- generate jump tables
let tabled =
{-# SCC "generateJumpTables" #-}
generateJumpTables ncgImpl alloced
when (not $ null nativeCfgWeights) $ dumpIfSet_dyn logger dflags
Opt_D_dump_cfg_weights "CFG Update information"
FormatText
( text "stack:" <+> ppr stack_updt_blks $$
text "linearAlloc:" <+> ppr cfgRegAllocUpdates )
---- shortcut branches
let (shorted, postShortCFG) =
{-# SCC "shortcutBranches" #-}
shortcutBranches dflags ncgImpl tabled postRegCFG
let optimizedCFG :: Maybe CFG
optimizedCFG =
optimizeCFG (gopt Opt_CmmStaticPred dflags) weights cmm <$!> postShortCFG
maybeDumpCfg logger dflags optimizedCFG "CFG Weights - Final" proc_name
--TODO: Partially check validity of the cfg.
let getBlks (CmmProc _info _lbl _live (ListGraph blocks)) = blocks
getBlks _ = []
when ( backendMaintainsCfg platform &&
(gopt Opt_DoAsmLinting dflags || debugIsOn )) $ do
let blocks = concatMap getBlks shorted
let labels = setFromList $ fmap blockId blocks :: LabelSet
let cfg = fromJust optimizedCFG
return $! seq (sanityCheckCfg cfg labels $
text "cfg not in lockstep") ()
---- sequence blocks
let sequenced :: [NatCmmDecl statics instr]
sequenced =
checkLayout shorted $
{-# SCC "sequenceBlocks" #-}
map (BlockLayout.sequenceTop
ncgImpl optimizedCFG)
shorted
let branchOpt :: [NatCmmDecl statics instr]
branchOpt =
{-# SCC "invertCondBranches" #-}
map invert sequenced
where
invertConds :: LabelMap RawCmmStatics -> [NatBasicBlock instr]
-> [NatBasicBlock instr]
invertConds = invertCondBranches ncgImpl optimizedCFG
invert top@CmmData {} = top
invert (CmmProc info lbl live (ListGraph blocks)) =
CmmProc info lbl live (ListGraph $ invertConds info blocks)
---- expansion of SPARC synthetic instrs
let expanded =
{-# SCC "sparc_expand" #-}
ncgExpandTop ncgImpl branchOpt
--ncgExpandTop ncgImpl sequenced
dumpIfSet_dyn logger dflags
Opt_D_dump_asm_expanded "Synthetic instructions expanded"
FormatCMM
(vcat $ map (pprNatCmmDecl ncgImpl) expanded)
-- generate unwinding information from cmm
let unwinds :: BlockMap [UnwindPoint]
unwinds =
{-# SCC "unwindingInfo" #-}
foldl' addUnwind mapEmpty expanded
where
addUnwind acc proc =
acc `mapUnion` computeUnwinding dflags ncgImpl proc
return ( usAlloc
, fileIds'
, expanded
, lastMinuteImports ++ imports
, ppr_raStatsColor
, ppr_raStatsLinear
, unwinds )
maybeDumpCfg :: Logger -> DynFlags -> Maybe CFG -> String -> SDoc -> IO ()
maybeDumpCfg _logger _dflags Nothing _ _ = return ()
maybeDumpCfg logger dflags (Just cfg) msg proc_name
| null cfg = return ()
| otherwise
= dumpIfSet_dyn logger
dflags Opt_D_dump_cfg_weights msg
FormatText
(proc_name <> char ':' $$ pprEdgeWeights cfg)
-- | Make sure all blocks we want the layout algorithm to place have been placed.
checkLayout :: [NatCmmDecl statics instr] -> [NatCmmDecl statics instr]
-> [NatCmmDecl statics instr]
checkLayout procsUnsequenced procsSequenced =
ASSERT2(setNull diff,
ppr "Block sequencing dropped blocks:" <> ppr diff)
procsSequenced
where
blocks1 = foldl' (setUnion) setEmpty $
map getBlockIds procsUnsequenced :: LabelSet
blocks2 = foldl' (setUnion) setEmpty $
map getBlockIds procsSequenced
diff = setDifference blocks1 blocks2
getBlockIds (CmmData _ _) = setEmpty
getBlockIds (CmmProc _ _ _ (ListGraph blocks)) =
setFromList $ map blockId blocks
-- | Compute unwinding tables for the blocks of a procedure
computeUnwinding :: Instruction instr
=> DynFlags -> NcgImpl statics instr jumpDest
-> NatCmmDecl statics instr
-- ^ the native code generated for the procedure
-> LabelMap [UnwindPoint]
-- ^ unwinding tables for all points of all blocks of the
-- procedure
computeUnwinding dflags _ _
| debugLevel dflags == 0 = mapEmpty
computeUnwinding _ _ (CmmData _ _) = mapEmpty
computeUnwinding _ ncgImpl (CmmProc _ _ _ (ListGraph blks)) =
-- In general we would need to push unwinding information down the
-- block-level call-graph to ensure that we fully account for all
-- relevant register writes within a procedure.
--
-- However, the only unwinding information that we care about in GHC is for
-- Sp. The fact that GHC.Cmm.LayoutStack already ensures that we have unwind
-- information at the beginning of every block means that there is no need
-- to perform this sort of push-down.
mapFromList [ (blk_lbl, extractUnwindPoints ncgImpl instrs)
| BasicBlock blk_lbl instrs <- blks ]
-- | Build a doc for all the imports.
--
makeImportsDoc :: NCGConfig -> [CLabel] -> SDoc
makeImportsDoc config imports
= dyld_stubs imports
$$
-- On recent versions of Darwin, the linker supports
-- dead-stripping of code and data on a per-symbol basis.
-- There's a hack to make this work in PprMach.pprNatCmmDecl.
(if platformHasSubsectionsViaSymbols platform
then text ".subsections_via_symbols"
else Outputable.empty)
$$
-- On recent GNU ELF systems one can mark an object file
-- as not requiring an executable stack. If all objects
-- linked into a program have this note then the program
-- will not use an executable stack, which is good for
-- security. GHC generated code does not need an executable
-- stack so add the note in:
(if platformHasGnuNonexecStack platform
then text ".section .note.GNU-stack,\"\"," <> sectionType platform "progbits"
else Outputable.empty)
$$
-- And just because every other compiler does, let's stick in
-- an identifier directive: .ident "GHC x.y.z"
(if platformHasIdentDirective platform
then let compilerIdent = text "GHC" <+> text cProjectVersion
in text ".ident" <+> doubleQuotes compilerIdent
else Outputable.empty)
where
platform = ncgPlatform config
-- Generate "symbol stubs" for all external symbols that might
-- come from a dynamic library.
dyld_stubs :: [CLabel] -> SDoc
{- dyld_stubs imps = vcat $ map pprDyldSymbolStub $
map head $ group $ sort imps-}
-- (Hack) sometimes two Labels pretty-print the same, but have
-- different uniques; so we compare their text versions...
dyld_stubs imps
| needImportedSymbols config
= vcat $
(pprGotDeclaration config :) $
map ( pprImportedSymbol config . fst . head) $
groupBy (\(_,a) (_,b) -> a == b) $
sortBy (\(_,a) (_,b) -> compare a b) $
map doPpr $
imps
| otherwise
= Outputable.empty
doPpr lbl = (lbl, renderWithContext
(ncgAsmContext config)
(pprCLabel platform AsmStyle lbl))
-- -----------------------------------------------------------------------------
-- Generate jump tables
-- Analyzes all native code and generates data sections for all jump
-- table instructions.
generateJumpTables
:: NcgImpl statics instr jumpDest
-> [NatCmmDecl statics instr] -> [NatCmmDecl statics instr]
generateJumpTables ncgImpl xs = concatMap f xs
where f p@(CmmProc _ _ _ (ListGraph xs)) = p : concatMap g xs
f p = [p]
g (BasicBlock _ xs) = catMaybes (map (generateJumpTableForInstr ncgImpl) xs)
-- -----------------------------------------------------------------------------
-- Shortcut branches
shortcutBranches
:: forall statics instr jumpDest. (Outputable jumpDest) => DynFlags
-> NcgImpl statics instr jumpDest
-> [NatCmmDecl statics instr]
-> Maybe CFG
-> ([NatCmmDecl statics instr],Maybe CFG)
shortcutBranches dflags ncgImpl tops weights
| gopt Opt_AsmShortcutting dflags
= ( map (apply_mapping ncgImpl mapping) tops'
, shortcutWeightMap mappingBid <$!> weights )
| otherwise
= (tops, weights)
where
(tops', mappings) = mapAndUnzip (build_mapping ncgImpl) tops
mapping = mapUnions mappings :: LabelMap jumpDest
mappingBid = fmap (getJumpDestBlockId ncgImpl) mapping
build_mapping :: forall instr t d statics jumpDest.
NcgImpl statics instr jumpDest
-> GenCmmDecl d (LabelMap t) (ListGraph instr)
-> (GenCmmDecl d (LabelMap t) (ListGraph instr)
,LabelMap jumpDest)
build_mapping _ top@(CmmData _ _) = (top, mapEmpty)
build_mapping _ (CmmProc info lbl live (ListGraph []))
= (CmmProc info lbl live (ListGraph []), mapEmpty)
build_mapping ncgImpl (CmmProc info lbl live (ListGraph (head:blocks)))
= (CmmProc info lbl live (ListGraph (head:others)), mapping)
-- drop the shorted blocks, but don't ever drop the first one,
-- because it is pointed to by a global label.
where
-- find all the blocks that just consist of a jump that can be
-- shorted.
-- Don't completely eliminate loops here -- that can leave a dangling jump!
shortcut_blocks :: [(BlockId, jumpDest)]
(_, shortcut_blocks, others) =
foldl' split (setEmpty :: LabelSet, [], []) blocks
split (s, shortcut_blocks, others) b@(BasicBlock id [insn])
| Just jd <- canShortcut ncgImpl insn
, Just dest <- getJumpDestBlockId ncgImpl jd
, not (has_info id)
, (setMember dest s) || dest == id -- loop checks
= (s, shortcut_blocks, b : others)
split (s, shortcut_blocks, others) (BasicBlock id [insn])
| Just dest <- canShortcut ncgImpl insn
, not (has_info id)
= (setInsert id s, (id,dest) : shortcut_blocks, others)
split (s, shortcut_blocks, others) other = (s, shortcut_blocks, other : others)
-- do not eliminate blocks that have an info table
has_info l = mapMember l info
-- build a mapping from BlockId to JumpDest for shorting branches
mapping = mapFromList shortcut_blocks
apply_mapping :: NcgImpl statics instr jumpDest
-> LabelMap jumpDest
-> GenCmmDecl statics h (ListGraph instr)
-> GenCmmDecl statics h (ListGraph instr)
apply_mapping ncgImpl ufm (CmmData sec statics)
= CmmData sec (shortcutStatics ncgImpl (\bid -> mapLookup bid ufm) statics)
apply_mapping ncgImpl ufm (CmmProc info lbl live (ListGraph blocks))
= CmmProc info lbl live (ListGraph $ map short_bb blocks)
where
short_bb (BasicBlock id insns) = BasicBlock id $! map short_insn insns
short_insn i = shortcutJump ncgImpl (\bid -> mapLookup bid ufm) i
-- shortcutJump should apply the mapping repeatedly,
-- just in case we can short multiple branches.
-- -----------------------------------------------------------------------------
-- Instruction selection
-- Native code instruction selection for a chunk of stix code. For
-- this part of the computation, we switch from the UniqSM monad to
-- the NatM monad. The latter carries not only a Unique, but also an
-- Int denoting the current C stack pointer offset in the generated
-- code; this is needed for creating correct spill offsets on
-- architectures which don't offer, or for which it would be
-- prohibitively expensive to employ, a frame pointer register. Viz,
-- x86.
-- The offset is measured in bytes, and indicates the difference
-- between the current (simulated) C stack-ptr and the value it was at
-- the beginning of the block. For stacks which grow down, this value
-- should be either zero or negative.
-- Along with the stack pointer offset, we also carry along a LabelMap of
-- DebugBlocks, which we read to generate .location directives.
--
-- Switching between the two monads whilst carrying along the same
-- Unique supply breaks abstraction. Is that bad?
genMachCode
:: NCGConfig
-> ModLocation
-> (RawCmmDecl -> NatM [NatCmmDecl statics instr])
-> DwarfFiles
-> LabelMap DebugBlock
-> RawCmmDecl
-> CFG
-> UniqSM
( [NatCmmDecl statics instr]
, [CLabel]
, DwarfFiles
, CFG
)
genMachCode config modLoc cmmTopCodeGen fileIds dbgMap cmm_top cmm_cfg
= do { initial_us <- getUniqueSupplyM
; let initial_st = mkNatM_State initial_us 0 config
modLoc fileIds dbgMap cmm_cfg
(new_tops, final_st) = initNat initial_st (cmmTopCodeGen cmm_top)
final_delta = natm_delta final_st
final_imports = natm_imports final_st
final_cfg = natm_cfg final_st
; if final_delta == 0
then return (new_tops, final_imports
, natm_fileid final_st, final_cfg)
else pprPanic "genMachCode: nonzero final delta" (int final_delta)
}
-- -----------------------------------------------------------------------------
-- Generic Cmm optimiser
{-
Here we do:
(a) Constant folding
(c) Position independent code and dynamic linking
(i) introduce the appropriate indirections
and position independent refs
(ii) compile a list of imported symbols
(d) Some arch-specific optimizations
(a) will be moving to the new Hoopl pipeline, however, (c) and
(d) are only needed by the native backend and will continue to live
here.
Ideas for other things we could do (put these in Hoopl please!):
- shortcut jumps-to-jumps
- simple CSE: if an expr is assigned to a temp, then replace later occs of
that expr with the temp, until the expr is no longer valid (can push through
temp assignments, and certain assigns to mem...)
-}
cmmToCmm :: NCGConfig -> RawCmmDecl -> (RawCmmDecl, [CLabel])
cmmToCmm _ top@(CmmData _ _) = (top, [])
cmmToCmm config (CmmProc info lbl live graph)
= runCmmOpt config $
do blocks' <- mapM cmmBlockConFold (toBlockList graph)
return $ CmmProc info lbl live (ofBlockList (g_entry graph) blocks')
type OptMResult a = (# a, [CLabel] #)
pattern OptMResult :: a -> b -> (# a, b #)
pattern OptMResult x y = (# x, y #)
{-# COMPLETE OptMResult #-}
newtype CmmOptM a = CmmOptM (NCGConfig -> [CLabel] -> OptMResult a)
deriving (Functor)
instance Applicative CmmOptM where
pure x = CmmOptM $ \_ imports -> OptMResult x imports
(<*>) = ap
instance Monad CmmOptM where
(CmmOptM f) >>= g =
CmmOptM $ \config imports0 ->
case f config imports0 of
OptMResult x imports1 ->
case g x of
CmmOptM g' -> g' config imports1
instance CmmMakeDynamicReferenceM CmmOptM where
addImport = addImportCmmOpt
addImportCmmOpt :: CLabel -> CmmOptM ()
addImportCmmOpt lbl = CmmOptM $ \_ imports -> OptMResult () (lbl:imports)
getCmmOptConfig :: CmmOptM NCGConfig
getCmmOptConfig = CmmOptM $ \config imports -> OptMResult config imports
runCmmOpt :: NCGConfig -> CmmOptM a -> (a, [CLabel])
runCmmOpt config (CmmOptM f) =
case f config [] of
OptMResult result imports -> (result, imports)
cmmBlockConFold :: CmmBlock -> CmmOptM CmmBlock
cmmBlockConFold block = do
let (entry, middle, last) = blockSplit block
stmts = blockToList middle
stmts' <- mapM cmmStmtConFold stmts
last' <- cmmStmtConFold last
return $ blockJoin entry (blockFromList stmts') last'
-- This does three optimizations, but they're very quick to check, so we don't
-- bother turning them off even when the Hoopl code is active. Since
-- this is on the old Cmm representation, we can't reuse the code either:
-- * reg = reg --> nop
-- * if 0 then jump --> nop
-- * if 1 then jump --> jump
-- We might be tempted to skip this step entirely of not Opt_PIC, but
-- there is some PowerPC code for the non-PIC case, which would also
-- have to be separated.
cmmStmtConFold :: CmmNode e x -> CmmOptM (CmmNode e x)
cmmStmtConFold stmt
= case stmt of
CmmAssign reg src
-> do src' <- cmmExprConFold DataReference src
return $ case src' of
CmmReg reg' | reg == reg' -> CmmComment (fsLit "nop")
new_src -> CmmAssign reg new_src
CmmStore addr src align
-> do addr' <- cmmExprConFold DataReference addr
src' <- cmmExprConFold DataReference src
return $ CmmStore addr' src' align
CmmCall { cml_target = addr }
-> do addr' <- cmmExprConFold JumpReference addr
return $ stmt { cml_target = addr' }
CmmUnsafeForeignCall target regs args
-> do target' <- case target of
ForeignTarget e conv -> do
e' <- cmmExprConFold CallReference e
return $ ForeignTarget e' conv
PrimTarget _ ->
return target
args' <- mapM (cmmExprConFold DataReference) args
return $ CmmUnsafeForeignCall target' regs args'
CmmCondBranch test true false likely
-> do test' <- cmmExprConFold DataReference test
return $ case test' of
CmmLit (CmmInt 0 _) -> CmmBranch false
CmmLit (CmmInt _ _) -> CmmBranch true
_other -> CmmCondBranch test' true false likely
CmmSwitch expr ids
-> do expr' <- cmmExprConFold DataReference expr
return $ CmmSwitch expr' ids
other
-> return other
cmmExprConFold :: ReferenceKind -> CmmExpr -> CmmOptM CmmExpr
cmmExprConFold referenceKind expr = do
config <- getCmmOptConfig
let expr' = if not (ncgDoConstantFolding config)
then expr
else cmmExprCon config expr
cmmExprNative referenceKind expr'
cmmExprCon :: NCGConfig -> CmmExpr -> CmmExpr
cmmExprCon config (CmmLoad addr rep align) = CmmLoad (cmmExprCon config addr) rep align
cmmExprCon config (CmmMachOp mop args)
= cmmMachOpFold (ncgPlatform config) mop (map (cmmExprCon config) args)
cmmExprCon _ other = other
-- handles both PIC and non-PIC cases... a very strange mixture
-- of things to do.
cmmExprNative :: ReferenceKind -> CmmExpr -> CmmOptM CmmExpr
cmmExprNative referenceKind expr = do
config <- getCmmOptConfig
let platform = ncgPlatform config
arch = platformArch platform
case expr of
CmmLoad addr rep align
-> do addr' <- cmmExprNative DataReference addr
return $ CmmLoad addr' rep align
CmmMachOp mop args
-> do args' <- mapM (cmmExprNative DataReference) args
return $ CmmMachOp mop args'
CmmLit (CmmBlock id)
-> cmmExprNative referenceKind (CmmLit (CmmLabel (infoTblLbl id)))
-- we must convert block Ids to CLabels here, because we
-- might have to do the PIC transformation. Hence we must
-- not modify BlockIds beyond this point.
CmmLit (CmmLabel lbl)
-> cmmMakeDynamicReference config referenceKind lbl
CmmLit (CmmLabelOff lbl off)
-> do dynRef <- cmmMakeDynamicReference config referenceKind lbl
-- need to optimize here, since it's late
return $ cmmMachOpFold platform (MO_Add (wordWidth platform)) [
dynRef,
(CmmLit $ CmmInt (fromIntegral off) (wordWidth platform))
]
-- On powerpc (non-PIC), it's easier to jump directly to a label than
-- to use the register table, so we replace these registers
-- with the corresponding labels:
CmmReg (CmmGlobal EagerBlackholeInfo)
| arch == ArchPPC && not (ncgPIC config)
-> cmmExprNative referenceKind $
CmmLit (CmmLabel (mkCmmCodeLabel rtsUnitId (fsLit "__stg_EAGER_BLACKHOLE_info")))
CmmReg (CmmGlobal GCEnter1)
| arch == ArchPPC && not (ncgPIC config)
-> cmmExprNative referenceKind $
CmmLit (CmmLabel (mkCmmCodeLabel rtsUnitId (fsLit "__stg_gc_enter_1")))
CmmReg (CmmGlobal GCFun)
| arch == ArchPPC && not (ncgPIC config)
-> cmmExprNative referenceKind $
CmmLit (CmmLabel (mkCmmCodeLabel rtsUnitId (fsLit "__stg_gc_fun")))
other
-> return other
-- | Initialize the native code generator configuration from the DynFlags
initNCGConfig :: DynFlags -> Module -> NCGConfig
initNCGConfig dflags this_mod = NCGConfig
{ ncgPlatform = targetPlatform dflags
, ncgThisModule = this_mod
, ncgAsmContext = initSDocContext dflags (PprCode AsmStyle)
, ncgProcAlignment = cmmProcAlignment dflags
, ncgExternalDynamicRefs = gopt Opt_ExternalDynamicRefs dflags
, ncgPIC = positionIndependent dflags
, ncgInlineThresholdMemcpy = fromIntegral $ maxInlineMemcpyInsns dflags
, ncgInlineThresholdMemset = fromIntegral $ maxInlineMemsetInsns dflags
, ncgSplitSections = gopt Opt_SplitSections dflags
, ncgRegsIterative = gopt Opt_RegsIterative dflags
, ncgAsmLinting = gopt Opt_DoAsmLinting dflags
, ncgCfgWeights = cfgWeights dflags
, ncgCfgBlockLayout = gopt Opt_CfgBlocklayout dflags
, ncgCfgWeightlessLayout = gopt Opt_WeightlessBlocklayout dflags
-- With -O1 and greater, the cmmSink pass does constant-folding, so
-- we don't need to do it again in the native code generator.
, ncgDoConstantFolding = optLevel dflags < 1
, ncgDumpRegAllocStages = dopt Opt_D_dump_asm_regalloc_stages dflags
, ncgDumpAsmStats = dopt Opt_D_dump_asm_stats dflags
, ncgDumpAsmConflicts = dopt Opt_D_dump_asm_conflicts dflags
, ncgBmiVersion = case platformArch (targetPlatform dflags) of
ArchX86_64 -> bmiVersion dflags
ArchX86 -> bmiVersion dflags
_ -> Nothing
-- We assume SSE1 and SSE2 operations are available on both
-- x86 and x86_64. Historically we didn't default to SSE2 and
-- SSE1 on x86, which results in defacto nondeterminism for how
-- rounding behaves in the associated x87 floating point instructions
-- because variations in the spill/fpu stack placement of arguments for
-- operations would change the precision and final result of what
-- would otherwise be the same expressions with respect to single or
-- double precision IEEE floating point computations.
, ncgSseVersion =
let v | sseVersion dflags < Just SSE2 = Just SSE2
| otherwise = sseVersion dflags
in case platformArch (targetPlatform dflags) of
ArchX86_64 -> v
ArchX86 -> v
_ -> Nothing
, ncgDwarfEnabled = osElfTarget (platformOS (targetPlatform dflags)) && debugLevel dflags > 0 && platformArch (targetPlatform dflags) /= ArchAArch64
, ncgDwarfUnwindings = osElfTarget (platformOS (targetPlatform dflags)) && debugLevel dflags > 0
, ncgDwarfStripBlockInfo = osElfTarget (platformOS (targetPlatform dflags)) && debugLevel dflags < 2 -- We strip out block information when running with -g0 or -g1.
, ncgDwarfSourceNotes = osElfTarget (platformOS (targetPlatform dflags)) && debugLevel dflags > 2 -- We produce GHC-specific source-note DIEs only with -g3
, ncgExposeInternalSymbols = gopt Opt_ExposeInternalSymbols dflags
}