Pipeline.hs 13.3 KB
Newer Older
1 2
{-# LANGUAGE BangPatterns #-}

3
module GHC.Cmm.Pipeline (
4 5 6 7
  -- | Converts C-- with an implicit stack and native C-- calls into
  -- optimized, CPS converted and native-call-less C--.  The latter
  -- C-- can be used to generate assembly.
  cmmPipeline
8
) where
9

10
import GHC.Prelude
11

12 13 14 15 16 17 18 19 20 21
import GHC.Cmm
import GHC.Cmm.Lint
import GHC.Cmm.Info.Build
import GHC.Cmm.CommonBlockElim
import GHC.Cmm.Switch.Implement
import GHC.Cmm.ProcPoint
import GHC.Cmm.ContFlowOpt
import GHC.Cmm.LayoutStack
import GHC.Cmm.Sink
import GHC.Cmm.Dataflow.Collections
22

Sylvain Henry's avatar
Sylvain Henry committed
23
import GHC.Types.Unique.Supply
Sylvain Henry's avatar
Sylvain Henry committed
24
import GHC.Driver.Session
Sylvain Henry's avatar
Sylvain Henry committed
25
import GHC.Driver.Backend
26
import GHC.Utils.Error
Sylvain Henry's avatar
Sylvain Henry committed
27
import GHC.Utils.Logger
Sylvain Henry's avatar
Sylvain Henry committed
28
import GHC.Driver.Env
Ian Lynagh's avatar
Ian Lynagh committed
29
import Control.Monad
30
import GHC.Utils.Outputable
John Ericson's avatar
John Ericson committed
31
import GHC.Platform
32
import Data.Either (partitionEithers)
Simon Marlow's avatar
Simon Marlow committed
33

34
-----------------------------------------------------------------------------
35
-- | Top level driver for C-- pipeline
36
-----------------------------------------------------------------------------
37

38 39 40 41 42
cmmPipeline
 :: HscEnv -- Compilation env including
           -- dynamic flags: -dcmm-lint -ddump-cmm-cps
 -> ModuleSRTInfo        -- Info about SRTs generated so far
 -> CmmGroup             -- Input C-- with Procedures
43
 -> IO (ModuleSRTInfo, CmmGroupSRTs) -- Output CPS transformed C--
44

Sylvain Henry's avatar
Sylvain Henry committed
45 46 47 48 49 50
cmmPipeline hsc_env srtInfo prog = do
  let logger = hsc_logger hsc_env
  let dflags = hsc_dflags hsc_env
  let forceRes (info, group) = info `seq` foldr (\decl r -> decl `seq` r) () group
  withTimingSilent logger dflags (text "Cmm pipeline") forceRes $ do
     tops <- {-# SCC "tops" #-} mapM (cpsTop logger dflags) prog
51

52 53
     let (procs, data_) = partitionEithers tops
     (srtInfo, cmms) <- {-# SCC "doSRTs" #-} doSRTs dflags srtInfo procs data_
Sylvain Henry's avatar
Sylvain Henry committed
54 55
     let platform = targetPlatform dflags
     dumpWith logger dflags Opt_D_dump_cmm_cps "Post CPS Cmm" FormatCMM (pdoc platform cmms)
56

57
     return (srtInfo, cmms)
58 59


Sylvain Henry's avatar
Sylvain Henry committed
60 61 62
cpsTop :: Logger -> DynFlags -> CmmDecl -> IO (Either (CAFEnv, [CmmDecl]) (CAFSet, CmmDecl))
cpsTop _logger dflags p@(CmmData _ statics) = return (Right (cafAnalData (targetPlatform dflags) statics, p))
cpsTop logger dflags proc =
63
    do
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
      ----------- Control-flow optimisations ----------------------------------

      -- The first round of control-flow optimisation speeds up the
      -- later passes by removing lots of empty blocks, so we do it
      -- even when optimisation isn't turned on.
      --
      CmmProc h l v g <- {-# SCC "cmmCfgOpts(1)" #-}
           return $ cmmCfgOptsProc splitting_proc_points proc
      dump Opt_D_dump_cmm_cfg "Post control-flow optimisations" g

      let !TopInfo {stack_info=StackInfo { arg_space = entry_off
                                         , do_layout = do_layout }} = h

      ----------- Eliminate common blocks -------------------------------------
      g <- {-# SCC "elimCommonBlocks" #-}
           condPass Opt_CmmElimCommonBlocks elimCommonBlocks g
                         Opt_D_dump_cmm_cbe "Post common block elimination"

      -- Any work storing block Labels must be performed _after_
      -- elimCommonBlocks

      ----------- Implement switches ------------------------------------------
      g <- {-# SCC "createSwitchPlans" #-}
           runUniqSM $ cmmImplementSwitchPlans (backend dflags) platform g
      dump Opt_D_dump_cmm_switch "Post switch plan" g

      ----------- Proc points -------------------------------------------------
      let
        call_pps :: ProcPointSet -- LabelMap
        call_pps = {-# SCC "callProcPoints" #-} callProcPoints g
      proc_points <-
         if splitting_proc_points
96
            then do
97 98
              pp <- {-# SCC "minimalProcPointSet" #-} runUniqSM $
                 minimalProcPointSet platform call_pps g
Sylvain Henry's avatar
Sylvain Henry committed
99
              dumpWith logger dflags Opt_D_dump_cmm_proc "Proc points"
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119
                    FormatCMM (pdoc platform l $$ ppr pp $$ pdoc platform g)
              return pp
            else
              return call_pps

      ----------- Layout the stack and manifest Sp ----------------------------
      (g, stackmaps) <-
           {-# SCC "layoutStack" #-}
           if do_layout
              then runUniqSM $ cmmLayoutStack dflags proc_points entry_off g
              else return (g, mapEmpty)
      dump Opt_D_dump_cmm_sp "Layout Stack" g

      ----------- Sink and inline assignments  --------------------------------
      g <- {-# SCC "sink" #-} -- See Note [Sinking after stack layout]
           condPass Opt_CmmSink (cmmSink platform) g
                    Opt_D_dump_cmm_sink "Sink assignments"

      ------------- CAF analysis ----------------------------------------------
      let cafEnv = {-# SCC "cafAnal" #-} cafAnal platform call_pps l g
Sylvain Henry's avatar
Sylvain Henry committed
120
      dumpWith logger dflags Opt_D_dump_cmm_caf "CAFEnv" FormatText (pdoc platform cafEnv)
121 122 123 124 125 126

      g <- if splitting_proc_points
           then do
             ------------- Split into separate procedures -----------------------
             let pp_map = {-# SCC "procPointAnalysis" #-}
                          procPointAnalysis proc_points g
Sylvain Henry's avatar
Sylvain Henry committed
127
             dumpWith logger dflags Opt_D_dump_cmm_procmap "procpoint map"
128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152
                FormatCMM (ppr pp_map)
             g <- {-# SCC "splitAtProcPoints" #-} runUniqSM $
                  splitAtProcPoints platform l call_pps proc_points pp_map
                                    (CmmProc h l v g)
             dumps Opt_D_dump_cmm_split "Post splitting" g
             return g
           else
             -- attach info tables to return points
             return $ [attachContInfoTables call_pps (CmmProc h l v g)]

      ------------- Populate info tables with stack info -----------------
      g <- {-# SCC "setInfoTableStackMap" #-}
           return $ map (setInfoTableStackMap platform stackmaps) g
      dumps Opt_D_dump_cmm_info "after setInfoTableStackMap" g

      ----------- Control-flow optimisations -----------------------------
      g <- {-# SCC "cmmCfgOpts(2)" #-}
           return $ if optLevel dflags >= 1
                    then map (cmmCfgOptsProc splitting_proc_points) g
                    else g
      g <- return (map removeUnreachableBlocksProc g)
           -- See Note [unreachable blocks]
      dumps Opt_D_dump_cmm_cfg "Post control-flow optimisations" g

      return (Left (cafEnv, g))
153

154
  where platform = targetPlatform dflags
Sylvain Henry's avatar
Sylvain Henry committed
155
        dump = dumpGraph logger dflags
156

Simon Marlow's avatar
Simon Marlow committed
157
        dumps flag name
Sylvain Henry's avatar
Sylvain Henry committed
158
           = mapM_ (dumpWith logger dflags flag name FormatCMM . pdoc platform)
Simon Marlow's avatar
Simon Marlow committed
159

Simon Marlow's avatar
Simon Marlow committed
160
        condPass flag pass g dumpflag dumpname =
ian@well-typed.com's avatar
ian@well-typed.com committed
161
            if gopt flag dflags
Simon Marlow's avatar
Simon Marlow committed
162 163 164 165 166 167
               then do
                    g <- return $ pass g
                    dump dumpflag dumpname g
                    return g
               else return g

168 169 170 171
        -- we don't need to split proc points for the NCG, unless
        -- tablesNextToCode is off.  The latter is because we have no
        -- label to put on info tables for basic blocks that are not
        -- the entry point.
Sylvain Henry's avatar
Sylvain Henry committed
172
        splitting_proc_points = backend dflags /= NCG
173
                             || not (platformTablesNextToCode platform)
ian@well-typed.com's avatar
ian@well-typed.com committed
174 175
                             || -- Note [inconsistent-pic-reg]
                                usingInconsistentPicReg
176
        usingInconsistentPicReg
177
           = case (platformArch platform, platformOS platform, positionIndependent dflags)
178 179
             of   (ArchX86, OSDarwin, pic) -> pic
                  _                        -> False
180

Jan Stolarek's avatar
Jan Stolarek committed
181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
-- Note [Sinking after stack layout]
-- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
--
-- In the past we considered running sinking pass also before stack
-- layout, but after making some measurements we realized that:
--
--   a) running sinking only before stack layout produces slower
--      code than running sinking only before stack layout
--
--   b) running sinking both before and after stack layout produces
--      code that has the same performance as when running sinking
--      only after stack layout.
--
-- In other words sinking before stack layout doesn't buy as anything.
--
-- An interesting question is "why is it better to run sinking after
-- stack layout"? It seems that the major reason are stores and loads
-- generated by stack layout. Consider this code before stack layout:
--
--  c1E:
--      _c1C::P64 = R3;
--      _c1B::P64 = R2;
--      _c1A::P64 = R1;
--      I64[(young<c1D> + 8)] = c1D;
--      call stg_gc_noregs() returns to c1D, args: 8, res: 8, upd: 8;
--  c1D:
--      R3 = _c1C::P64;
--      R2 = _c1B::P64;
--      R1 = _c1A::P64;
--      call (P64[(old + 8)])(R3, R2, R1) args: 8, res: 0, upd: 8;
--
-- Stack layout pass will save all local variables live across a call
-- (_c1C, _c1B and _c1A in this example) on the stack just before
-- making a call and reload them from the stack after returning from a
-- call:
--
--  c1E:
--      _c1C::P64 = R3;
--      _c1B::P64 = R2;
--      _c1A::P64 = R1;
--      I64[Sp - 32] = c1D;
--      P64[Sp - 24] = _c1A::P64;
--      P64[Sp - 16] = _c1B::P64;
--      P64[Sp - 8] = _c1C::P64;
--      Sp = Sp - 32;
--      call stg_gc_noregs() returns to c1D, args: 8, res: 8, upd: 8;
--  c1D:
--      _c1A::P64 = P64[Sp + 8];
--      _c1B::P64 = P64[Sp + 16];
--      _c1C::P64 = P64[Sp + 24];
--      R3 = _c1C::P64;
--      R2 = _c1B::P64;
--      R1 = _c1A::P64;
--      Sp = Sp + 32;
--      call (P64[Sp])(R3, R2, R1) args: 8, res: 0, upd: 8;
--
-- If we don't run sinking pass after stack layout we are basically
-- left with such code. However, running sinking on this code can lead
-- to significant improvements:
--
--  c1E:
--      I64[Sp - 32] = c1D;
--      P64[Sp - 24] = R1;
--      P64[Sp - 16] = R2;
--      P64[Sp - 8] = R3;
--      Sp = Sp - 32;
--      call stg_gc_noregs() returns to c1D, args: 8, res: 8, upd: 8;
--  c1D:
--      R3 = P64[Sp + 24];
--      R2 = P64[Sp + 16];
--      R1 = P64[Sp + 8];
--      Sp = Sp + 32;
--      call (P64[Sp])(R3, R2, R1) args: 8, res: 0, upd: 8;
--
-- Now we only have 9 assignments instead of 15.
--
-- There is one case when running sinking before stack layout could
-- be beneficial. Consider this:
--
--   L1:
--      x = y
--      call f() returns L2
--   L2: ...x...y...
--
-- Since both x and y are live across a call to f, they will be stored
-- on the stack during stack layout and restored after the call:
--
--   L1:
--      x = y
--      P64[Sp - 24] = L2
--      P64[Sp - 16] = x
--      P64[Sp - 8]  = y
--      Sp = Sp - 24
--      call f() returns L2
--   L2:
--      y = P64[Sp + 16]
--      x = P64[Sp + 8]
--      Sp = Sp + 24
--      ...x...y...
--
-- However, if we run sinking before stack layout we would propagate x
-- to its usage place (both x and y must be local register for this to
-- be possible - global registers cannot be floated past a call):
--
--   L1:
--      x = y
--      call f() returns L2
--   L2: ...y...y...
--
-- Thus making x dead at the call to f(). If we ran stack layout now
-- we would generate less stores and loads:
--
--   L1:
--      x = y
--      P64[Sp - 16] = L2
--      P64[Sp - 8]  = y
--      Sp = Sp - 16
--      call f() returns L2
--   L2:
--      y = P64[Sp + 8]
--      Sp = Sp + 16
--      ...y...y...
--
304
-- But since we don't see any benefits from running sinking before stack
Jan Stolarek's avatar
Jan Stolarek committed
305 306 307
-- layout, this situation probably doesn't arise too often in practice.
--

ian@well-typed.com's avatar
ian@well-typed.com committed
308
{- Note [inconsistent-pic-reg]
309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335

On x86/Darwin, PIC is implemented by inserting a sequence like

    call 1f
 1: popl %reg

at the proc entry point, and then referring to labels as offsets from
%reg.  If we don't split proc points, then we could have many entry
points in a proc that would need this sequence, and each entry point
would then get a different value for %reg.  If there are any join
points, then at the join point we don't have a consistent value for
%reg, so we don't know how to refer to labels.

Hence, on x86/Darwin, we have to split proc points, and then each proc
point will get its own PIC initialisation sequence.

This isn't an issue on x86/ELF, where the sequence is

    call 1f
 1: popl %reg
    addl $_GLOBAL_OFFSET_TABLE_+(.-1b), %reg

so %reg always has a consistent value: the address of
_GLOBAL_OFFSET_TABLE_, regardless of which entry point we arrived via.

-}

336
{- Note [unreachable blocks]
337

338
The control-flow optimiser sometimes leaves unreachable blocks behind
Simon Marlow's avatar
Simon Marlow committed
339 340 341
containing junk code.  These aren't necessarily a problem, but
removing them is good because it might save time in the native code
generator later.
342 343

-}
344

345 346 347 348
runUniqSM :: UniqSM a -> IO a
runUniqSM m = do
  us <- mkSplitUniqSupply 'u'
  return (initUs_ us m)
349

350

Sylvain Henry's avatar
Sylvain Henry committed
351 352
dumpGraph :: Logger -> DynFlags -> DumpFlag -> String -> CmmGraph -> IO ()
dumpGraph logger dflags flag name g = do
ian@well-typed.com's avatar
ian@well-typed.com committed
353
  when (gopt Opt_DoCmmLinting dflags) $ do_lint g
Sylvain Henry's avatar
Sylvain Henry committed
354
  dumpWith logger dflags flag name FormatCMM (pdoc platform g)
355
 where
Sylvain Henry's avatar
Sylvain Henry committed
356 357
  platform = targetPlatform dflags
  do_lint g = case cmmLintGraph platform g of
Sylvain Henry's avatar
Sylvain Henry committed
358 359
                 Just err -> do { fatalErrorMsg logger dflags err
                                ; ghcExit logger dflags 1
360 361
                                }
                 Nothing  -> return ()
Simon Marlow's avatar
Simon Marlow committed
362

Sylvain Henry's avatar
Sylvain Henry committed
363 364 365
dumpWith :: Logger -> DynFlags -> DumpFlag -> String -> DumpFormat -> SDoc -> IO ()
dumpWith logger dflags flag txt fmt sdoc = do
  dumpIfSet_dyn logger dflags flag txt fmt sdoc
366 367 368 369
  when (not (dopt flag dflags)) $
    -- If `-ddump-cmm-verbose -ddump-to-file` is specified,
    -- dump each Cmm pipeline stage output to a separate file.  #16930
    when (dopt Opt_D_dump_cmm_verbose dflags)
Sylvain Henry's avatar
Sylvain Henry committed
370 371
      $ putDumpMsg logger dflags (mkDumpStyle alwaysQualify) flag txt fmt sdoc
  dumpIfSet_dyn logger dflags Opt_D_dump_cmm_verbose_by_proc txt fmt sdoc