Pipeline.hs 13.2 KB
Newer Older
1
{-# LANGUAGE BangPatterns #-}
2 3
{-# LANGUAGE LambdaCase #-}
{-# LANGUAGE TupleSections #-}
4

5
module GHC.Cmm.Pipeline (
6 7 8 9
  -- | 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
10
) where
11

12
import GHC.Prelude
13

14 15 16 17 18 19 20 21 22 23
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
24

Sylvain Henry's avatar
Sylvain Henry committed
25
import GHC.Types.Unique.Supply
Sylvain Henry's avatar
Sylvain Henry committed
26
import GHC.Driver.Session
Sylvain Henry's avatar
Sylvain Henry committed
27
import GHC.Driver.Backend
28
import GHC.Utils.Error
Sylvain Henry's avatar
Sylvain Henry committed
29
import GHC.Driver.Types
Ian Lynagh's avatar
Ian Lynagh committed
30
import Control.Monad
31
import GHC.Utils.Outputable
John Ericson's avatar
John Ericson committed
32
import GHC.Platform
33
import Data.Either (partitionEithers)
Simon Marlow's avatar
Simon Marlow committed
34

35
-----------------------------------------------------------------------------
36
-- | Top level driver for C-- pipeline
37
-----------------------------------------------------------------------------
38

39 40 41 42 43
cmmPipeline
 :: HscEnv -- Compilation env including
           -- dynamic flags: -dcmm-lint -ddump-cmm-cps
 -> ModuleSRTInfo        -- Info about SRTs generated so far
 -> CmmGroup             -- Input C-- with Procedures
44
 -> IO (ModuleSRTInfo, CmmGroupSRTs) -- Output CPS transformed C--
45

46
cmmPipeline hsc_env srtInfo prog = withTimingSilent dflags (text "Cmm pipeline") forceRes $
47
  do let dflags = hsc_dflags hsc_env
48

49
     tops <- {-# SCC "tops" #-} mapM (cpsTop dflags) prog
50

51 52
     let (procs, data_) = partitionEithers tops
     (srtInfo, cmms) <- {-# SCC "doSRTs" #-} doSRTs dflags srtInfo procs data_
Sylvain Henry's avatar
Sylvain Henry committed
53
     dumpWith dflags Opt_D_dump_cmm_cps "Post CPS Cmm" FormatCMM (ppr cmms)
54

55
     return (srtInfo, cmms)
56

57 58 59 60
  where forceRes (info, group) =
          info `seq` foldr (\decl r -> decl `seq` r) () group

        dflags = hsc_dflags hsc_env
61

62 63 64
cpsTop :: DynFlags -> CmmDecl -> IO (Either (CAFEnv, [CmmDecl]) (CAFSet, CmmDecl))
cpsTop dflags p@(CmmData _ statics) = return (Right (cafAnalData (targetPlatform dflags) statics, p))
cpsTop dflags proc =
65
    do
Simon Marlow's avatar
Simon Marlow committed
66
       ----------- Control-flow optimisations ----------------------------------
67 68 69 70 71

       -- 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.
       --
72
       CmmProc h l v g <- {-# SCC "cmmCfgOpts(1)" #-}
73
            return $ cmmCfgOptsProc splitting_proc_points proc
Gabor Greif's avatar
Gabor Greif committed
74
       dump Opt_D_dump_cmm_cfg "Post control-flow optimisations" g
75

76 77 78
       let !TopInfo {stack_info=StackInfo { arg_space = entry_off
                                          , do_layout = do_layout }} = h

Simon Marlow's avatar
Simon Marlow committed
79
       ----------- Eliminate common blocks -------------------------------------
80 81 82
       g <- {-# SCC "elimCommonBlocks" #-}
            condPass Opt_CmmElimCommonBlocks elimCommonBlocks g
                          Opt_D_dump_cmm_cbe "Post common block elimination"
83

84 85
       -- Any work storing block Labels must be performed _after_
       -- elimCommonBlocks
86

87
       ----------- Implement switches ------------------------------------------
88
       g <- {-# SCC "createSwitchPlans" #-}
89
            runUniqSM $ cmmImplementSwitchPlans (backend dflags) platform g
90 91
       dump Opt_D_dump_cmm_switch "Post switch plan" g

Simon Marlow's avatar
Simon Marlow committed
92
       ----------- Proc points -------------------------------------------------
93 94 95
       let
         call_pps :: ProcPointSet -- LabelMap
         call_pps = {-# SCC "callProcPoints" #-} callProcPoints g
96 97
       proc_points <-
          if splitting_proc_points
98 99
             then do
               pp <- {-# SCC "minimalProcPointSet" #-} runUniqSM $
100
                  minimalProcPointSet platform call_pps g
101
               dumpWith dflags Opt_D_dump_cmm_proc "Proc points"
Sylvain Henry's avatar
Sylvain Henry committed
102
                     FormatCMM (ppr l $$ ppr pp $$ ppr g)
103
               return pp
104
             else
105
               return call_pps
106

107
       ----------- Layout the stack and manifest Sp ----------------------------
Simon Marlow's avatar
Simon Marlow committed
108 109
       (g, stackmaps) <-
            {-# SCC "layoutStack" #-}
110 111 112
            if do_layout
               then runUniqSM $ cmmLayoutStack dflags proc_points entry_off g
               else return (g, mapEmpty)
113
       dump Opt_D_dump_cmm_sp "Layout Stack" g
Simon Marlow's avatar
Simon Marlow committed
114

Jan Stolarek's avatar
Jan Stolarek committed
115 116
       ----------- Sink and inline assignments  --------------------------------
       g <- {-# SCC "sink" #-} -- See Note [Sinking after stack layout]
117
            condPass Opt_CmmSink (cmmSink platform) g
118
                     Opt_D_dump_cmm_sink "Sink assignments"
119

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

124 125 126
       g <- if splitting_proc_points
            then do
               ------------- Split into separate procedures -----------------------
127 128
               let pp_map = {-# SCC "procPointAnalysis" #-}
                            procPointAnalysis proc_points g
Sylvain Henry's avatar
Sylvain Henry committed
129 130
               dumpWith dflags Opt_D_dump_cmm_procmap "procpoint map"
                  FormatCMM (ppr pp_map)
131
               g <- {-# SCC "splitAtProcPoints" #-} runUniqSM $
132
                    splitAtProcPoints platform l call_pps proc_points pp_map
133 134 135 136 137 138 139 140 141
                                      (CmmProc h l v g)
               dumps Opt_D_dump_cmm_split "Post splitting" g
               return g
             else do
               -- attach info tables to return points
               return $ [attachContInfoTables call_pps (CmmProc h l v g)]

       ------------- Populate info tables with stack info -----------------
       g <- {-# SCC "setInfoTableStackMap" #-}
142
            return $ map (setInfoTableStackMap platform stackmaps) g
143 144 145 146 147 148 149 150 151 152 153
       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

154
       return (Left (cafEnv, g))
155

156
  where platform = targetPlatform dflags
157 158
        dump = dumpGraph dflags

Simon Marlow's avatar
Simon Marlow committed
159
        dumps flag name
Sylvain Henry's avatar
Sylvain Henry committed
160
           = mapM_ (dumpWith dflags flag name FormatCMM . ppr)
Simon Marlow's avatar
Simon Marlow committed
161

Simon Marlow's avatar
Simon Marlow committed
162
        condPass flag pass g dumpflag dumpname =
ian@well-typed.com's avatar
ian@well-typed.com committed
163
            if gopt flag dflags
Simon Marlow's avatar
Simon Marlow committed
164 165 166 167 168 169
               then do
                    g <- return $ pass g
                    dump dumpflag dumpname g
                    return g
               else return g

170 171 172 173
        -- 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
174
        splitting_proc_points = backend dflags /= NCG
175
                             || not (platformTablesNextToCode platform)
ian@well-typed.com's avatar
ian@well-typed.com committed
176 177
                             || -- Note [inconsistent-pic-reg]
                                usingInconsistentPicReg
178
        usingInconsistentPicReg
179
           = case (platformArch platform, platformOS platform, positionIndependent dflags)
180 181
             of   (ArchX86, OSDarwin, pic) -> pic
                  _                        -> False
182

Jan Stolarek's avatar
Jan Stolarek committed
183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305
-- 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...
--
306
-- But since we don't see any benefits from running sinking before stack
Jan Stolarek's avatar
Jan Stolarek committed
307 308 309
-- layout, this situation probably doesn't arise too often in practice.
--

ian@well-typed.com's avatar
ian@well-typed.com committed
310
{- Note [inconsistent-pic-reg]
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337

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.

-}

338
{- Note [unreachable blocks]
339

340
The control-flow optimiser sometimes leaves unreachable blocks behind
Simon Marlow's avatar
Simon Marlow committed
341 342 343
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.
344 345

-}
346

347 348 349 350
runUniqSM :: UniqSM a -> IO a
runUniqSM m = do
  us <- mkSplitUniqSupply 'u'
  return (initUs_ us m)
351

352

353
dumpGraph :: DynFlags -> DumpFlag -> String -> CmmGraph -> IO ()
Simon Marlow's avatar
Simon Marlow committed
354
dumpGraph dflags flag name g = do
ian@well-typed.com's avatar
ian@well-typed.com committed
355
  when (gopt Opt_DoCmmLinting dflags) $ do_lint g
Sylvain Henry's avatar
Sylvain Henry committed
356
  dumpWith dflags flag name FormatCMM (ppr g)
357
 where
358
  do_lint g = case cmmLintGraph (targetPlatform dflags) g of
Simon Marlow's avatar
Simon Marlow committed
359
                 Just err -> do { fatalErrorMsg dflags err
360 361 362
                                ; ghcExit dflags 1
                                }
                 Nothing  -> return ()
Simon Marlow's avatar
Simon Marlow committed
363

Sylvain Henry's avatar
Sylvain Henry committed
364 365 366
dumpWith :: DynFlags -> DumpFlag -> String -> DumpFormat -> SDoc -> IO ()
dumpWith dflags flag txt fmt sdoc = do
  dumpIfSet_dyn dflags flag txt fmt sdoc
367 368 369 370
  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
371
      $ dumpAction dflags (mkDumpStyle alwaysQualify)
Sylvain Henry's avatar
Sylvain Henry committed
372 373
                   (dumpOptionsFromFlag flag) txt fmt sdoc
  dumpIfSet_dyn dflags Opt_D_dump_cmm_verbose_by_proc txt fmt sdoc