TwoPass.hs

{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE TypeFamilies #-}

-- | Code generation for segmented and non-segmented scans.  Uses a
-- fairly inefficient two-pass algorithm, but can handle anything.
module Futhark.CodeGen.ImpGen.GPU.SegScan.TwoPass (compileSegScan) where

import Control.Monad.Except
import Control.Monad.State
import Data.List (delete, find, foldl', zip4)
import Data.Maybe
import qualified Futhark.CodeGen.ImpCode.GPU as Imp
import Futhark.CodeGen.ImpGen
import Futhark.CodeGen.ImpGen.GPU.Base
import Futhark.IR.GPUMem
import qualified Futhark.IR.Mem.IxFun as IxFun
import Futhark.Transform.Rename
import Futhark.Util (takeLast)
import Futhark.Util.IntegralExp (divUp, quot, rem)
import Prelude hiding (quot, rem)

-- Aggressively try to reuse memory for different SegBinOps, because
-- we will run them sequentially after another.
makeLocalArrays ::
  Count GroupSize SubExp ->
  SubExp ->
  [SegBinOp GPUMem] ->
  InKernelGen [[VName]]
makeLocalArrays (Count group_size) num_threads scans = do
  (arrs, mems_and_sizes) <- runStateT (mapM onScan scans) mempty
  let maxSize sizes = Imp.bytes $ foldl' sMax64 1 $ map Imp.unCount sizes
  forM_ mems_and_sizes $ \(sizes, mem) ->
    sAlloc_ mem (maxSize sizes) (Space "local")
  return arrs
  where
    onScan (SegBinOp _ scan_op nes _) = do
      let (scan_x_params, _scan_y_params) =
            splitAt (length nes) $ lambdaParams scan_op
      (arrs, used_mems) <- fmap unzip $
        forM scan_x_params $ \p ->
          case paramDec p of
            MemArray pt shape _ (ArrayIn mem _) -> do
              let shape' = Shape [num_threads] <> shape
              arr <-
                lift $
                  sArray "scan_arr" pt shape' $
                    ArrayIn mem $ IxFun.iota $ map pe64 $ shapeDims shape'
              return (arr, [])
            _ -> do
              let pt = elemType $ paramType p
                  shape = Shape [group_size]
              (sizes, mem') <- getMem pt shape
              arr <- lift $ sArrayInMem "scan_arr" pt shape mem'
              return (arr, [(sizes, mem')])
      modify (<> concat used_mems)
      return arrs

    getMem pt shape = do
      let size = typeSize $ Array pt shape NoUniqueness
      mems <- get
      case (find ((size `elem`) . fst) mems, mems) of
        (Just mem, _) -> do
          modify $ delete mem
          return mem
        (Nothing, (size', mem) : mems') -> do
          put mems'
          return (size : size', mem)
        (Nothing, []) -> do
          mem <- lift $ sDeclareMem "scan_arr_mem" $ Space "local"
          return ([size], mem)

type CrossesSegment = Maybe (Imp.TExp Int64 -> Imp.TExp Int64 -> Imp.TExp Bool)

localArrayIndex :: KernelConstants -> Type -> Imp.TExp Int64
localArrayIndex constants t =
  if primType t
    then sExt64 (kernelLocalThreadId constants)
    else sExt64 (kernelGlobalThreadId constants)

barrierFor :: Lambda GPUMem -> (Bool, Imp.Fence, InKernelGen ())
barrierFor scan_op = (array_scan, fence, sOp $ Imp.Barrier fence)
  where
    array_scan = not $ all primType $ lambdaReturnType scan_op
    fence
      | array_scan = Imp.FenceGlobal
      | otherwise = Imp.FenceLocal

xParams, yParams :: SegBinOp GPUMem -> [LParam GPUMem]
xParams scan =
  take (length (segBinOpNeutral scan)) (lambdaParams (segBinOpLambda scan))
yParams scan =
  drop (length (segBinOpNeutral scan)) (lambdaParams (segBinOpLambda scan))

writeToScanValues ::
  [VName] ->
  ([PatElem GPUMem], SegBinOp GPUMem, [KernelResult]) ->
  InKernelGen ()
writeToScanValues gtids (pes, scan, scan_res)
  | shapeRank (segBinOpShape scan) > 0 =
    forM_ (zip pes scan_res) $ \(pe, res) ->
      copyDWIMFix
        (patElemName pe)
        (map Imp.vi64 gtids)
        (kernelResultSubExp res)
        []
  | otherwise =
    forM_ (zip (yParams scan) scan_res) $ \(p, res) ->
      copyDWIMFix (paramName p) [] (kernelResultSubExp res) []

readToScanValues ::
  [Imp.TExp Int64] ->
  [PatElem GPUMem] ->
  SegBinOp GPUMem ->
  InKernelGen ()
readToScanValues is pes scan
  | shapeRank (segBinOpShape scan) > 0 =
    forM_ (zip (yParams scan) pes) $ \(p, pe) ->
      copyDWIMFix (paramName p) [] (Var (patElemName pe)) is
  | otherwise =
    return ()

readCarries ::
  Imp.TExp Int64 ->
  Imp.TExp Int64 ->
  [Imp.TExp Int64] ->
  [Imp.TExp Int64] ->
  [PatElem GPUMem] ->
  SegBinOp GPUMem ->
  InKernelGen ()
readCarries chunk_id chunk_offset dims' vec_is pes scan
  | shapeRank (segBinOpShape scan) > 0 = do
    ltid <- kernelLocalThreadId . kernelConstants <$> askEnv
    -- We may have to reload the carries from the output of the
    -- previous chunk.
    sIf
      (chunk_id .>. 0 .&&. ltid .==. 0)
      ( do
          let is = unflattenIndex dims' $ chunk_offset - 1
          forM_ (zip (xParams scan) pes) $ \(p, pe) ->
            copyDWIMFix (paramName p) [] (Var (patElemName pe)) (is ++ vec_is)
      )
      ( forM_ (zip (xParams scan) (segBinOpNeutral scan)) $ \(p, ne) ->
          copyDWIMFix (paramName p) [] ne []
      )
  | otherwise =
    return ()

-- | Produce partially scanned intervals; one per workgroup.
scanStage1 ::
  Pat GPUMem ->
  Count NumGroups SubExp ->
  Count GroupSize SubExp ->
  SegSpace ->
  [SegBinOp GPUMem] ->
  KernelBody GPUMem ->
  CallKernelGen (TV Int32, Imp.TExp Int64, CrossesSegment)
scanStage1 (Pat all_pes) num_groups group_size space scans kbody = do
  let num_groups' = fmap toInt64Exp num_groups
      group_size' = fmap toInt64Exp group_size
  num_threads <- dPrimV "num_threads" $ sExt32 $ unCount num_groups' * unCount group_size'

  let (gtids, dims) = unzip $ unSegSpace space
      dims' = map toInt64Exp dims
  let num_elements = product dims'
      elems_per_thread = num_elements `divUp` sExt64 (tvExp num_threads)
      elems_per_group = unCount group_size' * elems_per_thread

  let crossesSegment =
        case reverse dims' of
          segment_size : _ : _ -> Just $ \from to ->
            (to - from) .>. (to `rem` segment_size)
          _ -> Nothing

  sKernelThread "scan_stage1" num_groups' group_size' (segFlat space) $ do
    constants <- kernelConstants <$> askEnv
    all_local_arrs <- makeLocalArrays group_size (tvSize num_threads) scans

    -- The variables from scan_op will be used for the carry and such
    -- in the big chunking loop.
    forM_ scans $ \scan -> do
      dScope Nothing $ scopeOfLParams $ lambdaParams $ segBinOpLambda scan
      forM_ (zip (xParams scan) (segBinOpNeutral scan)) $ \(p, ne) ->
        copyDWIMFix (paramName p) [] ne []

    sFor "j" elems_per_thread $ \j -> do
      chunk_offset <-
        dPrimV "chunk_offset" $
          sExt64 (kernelGroupSize constants) * j
            + sExt64 (kernelGroupId constants) * elems_per_group
      flat_idx <-
        dPrimV "flat_idx" $
          tvExp chunk_offset + sExt64 (kernelLocalThreadId constants)
      -- Construct segment indices.
      zipWithM_ dPrimV_ gtids $ unflattenIndex dims' $ tvExp flat_idx

      let per_scan_pes = segBinOpChunks scans all_pes

          in_bounds =
            foldl1 (.&&.) $ zipWith (.<.) (map Imp.vi64 gtids) dims'

          when_in_bounds = compileStms mempty (kernelBodyStms kbody) $ do
            let (all_scan_res, map_res) =
                  splitAt (segBinOpResults scans) $ kernelBodyResult kbody
                per_scan_res =
                  segBinOpChunks scans all_scan_res

            sComment "write to-scan values to parameters" $
              mapM_ (writeToScanValues gtids) $
                zip3 per_scan_pes scans per_scan_res

            sComment "write mapped values results to global memory" $
              forM_ (zip (takeLast (length map_res) all_pes) map_res) $ \(pe, se) ->
                copyDWIMFix
                  (patElemName pe)
                  (map Imp.vi64 gtids)
                  (kernelResultSubExp se)
                  []

      sComment "threads in bounds read input" $
        sWhen in_bounds when_in_bounds

      unless (all (null . segBinOpShape) scans) $
        sOp $ Imp.Barrier Imp.FenceGlobal

      forM_ (zip3 per_scan_pes scans all_local_arrs) $
        \(pes, scan@(SegBinOp _ scan_op nes vec_shape), local_arrs) ->
          sComment "do one intra-group scan operation" $ do
            let rets = lambdaReturnType scan_op
                scan_x_params = xParams scan
                (array_scan, fence, barrier) = barrierFor scan_op

            when array_scan barrier

            sLoopNest vec_shape $ \vec_is -> do
              sComment "maybe restore some to-scan values to parameters, or read neutral" $
                sIf
                  in_bounds
                  ( do
                      readToScanValues (map Imp.vi64 gtids ++ vec_is) pes scan
                      readCarries j (tvExp chunk_offset) dims' vec_is pes scan
                  )
                  ( forM_ (zip (yParams scan) (segBinOpNeutral scan)) $ \(p, ne) ->
                      copyDWIMFix (paramName p) [] ne []
                  )

              sComment "combine with carry and write to local memory" $
                compileStms mempty (bodyStms $ lambdaBody scan_op) $
                  forM_ (zip3 rets local_arrs $ map resSubExp $ bodyResult $ lambdaBody scan_op) $
                    \(t, arr, se) ->
                      copyDWIMFix arr [localArrayIndex constants t] se []

              let crossesSegment' = do
                    f <- crossesSegment
                    Just $ \from to ->
                      let from' = sExt64 from + tvExp chunk_offset
                          to' = sExt64 to + tvExp chunk_offset
                       in f from' to'

              sOp $ Imp.ErrorSync fence

              -- We need to avoid parameter name clashes.
              scan_op_renamed <- renameLambda scan_op
              groupScan
                crossesSegment'
                (sExt64 $ tvExp num_threads)
                (sExt64 $ kernelGroupSize constants)
                scan_op_renamed
                local_arrs

              sComment "threads in bounds write partial scan result" $
                sWhen in_bounds $
                  forM_ (zip3 rets pes local_arrs) $ \(t, pe, arr) ->
                    copyDWIMFix
                      (patElemName pe)
                      (map Imp.vi64 gtids ++ vec_is)
                      (Var arr)
                      [localArrayIndex constants t]

              barrier

              let load_carry =
                    forM_ (zip local_arrs scan_x_params) $ \(arr, p) ->
                      copyDWIMFix
                        (paramName p)
                        []
                        (Var arr)
                        [ if primType $ paramType p
                            then sExt64 (kernelGroupSize constants) - 1
                            else
                              (sExt64 (kernelGroupId constants) + 1)
                                * sExt64 (kernelGroupSize constants) - 1
                        ]
                  load_neutral =
                    forM_ (zip nes scan_x_params) $ \(ne, p) ->
                      copyDWIMFix (paramName p) [] ne []

              sComment "first thread reads last element as carry-in for next iteration" $ do
                crosses_segment <- dPrimVE "crosses_segment" $
                  case crossesSegment of
                    Nothing -> false
                    Just f ->
                      f
                        ( tvExp chunk_offset
                            + sExt64 (kernelGroupSize constants) -1
                        )
                        ( tvExp chunk_offset
                            + sExt64 (kernelGroupSize constants)
                        )
                should_load_carry <-
                  dPrimVE "should_load_carry" $
                    kernelLocalThreadId constants .==. 0 .&&. bNot crosses_segment
                sWhen should_load_carry load_carry
                when array_scan barrier
                sUnless should_load_carry load_neutral

              barrier

  return (num_threads, elems_per_group, crossesSegment)

scanStage2 ::
  Pat GPUMem ->
  TV Int32 ->
  Imp.TExp Int64 ->
  Count NumGroups SubExp ->
  CrossesSegment ->
  SegSpace ->
  [SegBinOp GPUMem] ->
  CallKernelGen ()
scanStage2 (Pat all_pes) stage1_num_threads elems_per_group num_groups crossesSegment space scans = do
  let (gtids, dims) = unzip $ unSegSpace space
      dims' = map toInt64Exp dims

  -- Our group size is the number of groups for the stage 1 kernel.
  let group_size = Count $ unCount num_groups
      group_size' = fmap toInt64Exp group_size

  let crossesSegment' = do
        f <- crossesSegment
        Just $ \from to ->
          f
            ((sExt64 from + 1) * elems_per_group - 1)
            ((sExt64 to + 1) * elems_per_group - 1)

  sKernelThread "scan_stage2" 1 group_size' (segFlat space) $ do
    constants <- kernelConstants <$> askEnv
    per_scan_local_arrs <- makeLocalArrays group_size (tvSize stage1_num_threads) scans
    let per_scan_rets = map (lambdaReturnType . segBinOpLambda) scans
        per_scan_pes = segBinOpChunks scans all_pes

    flat_idx <-
      dPrimV "flat_idx" $
        (sExt64 (kernelLocalThreadId constants) + 1) * elems_per_group - 1
    -- Construct segment indices.
    zipWithM_ dPrimV_ gtids $ unflattenIndex dims' $ tvExp flat_idx

    forM_ (zip4 scans per_scan_local_arrs per_scan_rets per_scan_pes) $
      \(SegBinOp _ scan_op nes vec_shape, local_arrs, rets, pes) ->
        sLoopNest vec_shape $ \vec_is -> do
          let glob_is = map Imp.vi64 gtids ++ vec_is

              in_bounds =
                foldl1 (.&&.) $ zipWith (.<.) (map Imp.vi64 gtids) dims'

              when_in_bounds = forM_ (zip3 rets local_arrs pes) $ \(t, arr, pe) ->
                copyDWIMFix
                  arr
                  [localArrayIndex constants t]
                  (Var $ patElemName pe)
                  glob_is

              when_out_of_bounds = forM_ (zip3 rets local_arrs nes) $ \(t, arr, ne) ->
                copyDWIMFix arr [localArrayIndex constants t] ne []
              (_, _, barrier) =
                barrierFor scan_op

          sComment "threads in bound read carries; others get neutral element" $
            sIf in_bounds when_in_bounds when_out_of_bounds

          barrier

          groupScan
            crossesSegment'
            (sExt64 $ tvExp stage1_num_threads)
            (sExt64 $ kernelGroupSize constants)
            scan_op
            local_arrs

          sComment "threads in bounds write scanned carries" $
            sWhen in_bounds $
              forM_ (zip3 rets pes local_arrs) $ \(t, pe, arr) ->
                copyDWIMFix
                  (patElemName pe)
                  glob_is
                  (Var arr)
                  [localArrayIndex constants t]

scanStage3 ::
  Pat GPUMem ->
  Count NumGroups SubExp ->
  Count GroupSize SubExp ->
  Imp.TExp Int64 ->
  CrossesSegment ->
  SegSpace ->
  [SegBinOp GPUMem] ->
  CallKernelGen ()
scanStage3 (Pat all_pes) num_groups group_size elems_per_group crossesSegment space scans = do
  let num_groups' = fmap toInt64Exp num_groups
      group_size' = fmap toInt64Exp group_size
      (gtids, dims) = unzip $ unSegSpace space
      dims' = map toInt64Exp dims
  required_groups <-
    dPrimVE "required_groups" $
      sExt32 $ product dims' `divUp` sExt64 (unCount group_size')

  sKernelThread "scan_stage3" num_groups' group_size' (segFlat space) $
    virtualiseGroups SegVirt required_groups $ \virt_group_id -> do
      constants <- kernelConstants <$> askEnv

      -- Compute our logical index.
      flat_idx <-
        dPrimVE "flat_idx" $
          sExt64 virt_group_id * sExt64 (unCount group_size')
            + sExt64 (kernelLocalThreadId constants)
      zipWithM_ dPrimV_ gtids $ unflattenIndex dims' flat_idx

      -- Figure out which group this element was originally in.
      orig_group <- dPrimV "orig_group" $ flat_idx `quot` elems_per_group
      -- Then the index of the carry-in of the preceding group.
      carry_in_flat_idx <-
        dPrimV "carry_in_flat_idx" $
          tvExp orig_group * elems_per_group - 1
      -- Figure out the logical index of the carry-in.
      let carry_in_idx = unflattenIndex dims' $ tvExp carry_in_flat_idx

      -- Apply the carry if we are not in the scan results for the first
      -- group, and are not the last element in such a group (because
      -- then the carry was updated in stage 2), and we are not crossing
      -- a segment boundary.
      let in_bounds =
            foldl1 (.&&.) $ zipWith (.<.) (map Imp.vi64 gtids) dims'
          crosses_segment =
            fromMaybe false $
              crossesSegment
                <*> pure (tvExp carry_in_flat_idx)
                <*> pure flat_idx
          is_a_carry = flat_idx .==. (tvExp orig_group + 1) * elems_per_group - 1
          no_carry_in = tvExp orig_group .==. 0 .||. is_a_carry .||. crosses_segment

      let per_scan_pes = segBinOpChunks scans all_pes
      sWhen in_bounds $
        sUnless no_carry_in $
          forM_ (zip per_scan_pes scans) $
            \(pes, SegBinOp _ scan_op nes vec_shape) -> do
              dScope Nothing $ scopeOfLParams $ lambdaParams scan_op
              let (scan_x_params, scan_y_params) =
                    splitAt (length nes) $ lambdaParams scan_op

              sLoopNest vec_shape $ \vec_is -> do
                forM_ (zip scan_x_params pes) $ \(p, pe) ->
                  copyDWIMFix
                    (paramName p)
                    []
                    (Var $ patElemName pe)
                    (carry_in_idx ++ vec_is)

                forM_ (zip scan_y_params pes) $ \(p, pe) ->
                  copyDWIMFix
                    (paramName p)
                    []
                    (Var $ patElemName pe)
                    (map Imp.vi64 gtids ++ vec_is)

                compileBody' scan_x_params $ lambdaBody scan_op

                forM_ (zip scan_x_params pes) $ \(p, pe) ->
                  copyDWIMFix
                    (patElemName pe)
                    (map Imp.vi64 gtids ++ vec_is)
                    (Var $ paramName p)
                    []

-- | Compile 'SegScan' instance to host-level code with calls to
-- various kernels.
compileSegScan ::
  Pat GPUMem ->
  SegLevel ->
  SegSpace ->
  [SegBinOp GPUMem] ->
  KernelBody GPUMem ->
  CallKernelGen ()
compileSegScan pat lvl space scans kbody = do
  -- Since stage 2 involves a group size equal to the number of groups
  -- used for stage 1, we have to cap this number to the maximum group
  -- size.
  stage1_max_num_groups <- dPrim "stage1_max_num_groups" int64
  sOp $ Imp.GetSizeMax (tvVar stage1_max_num_groups) SizeGroup

  stage1_num_groups <-
    fmap (Imp.Count . tvSize) $
      dPrimV "stage1_num_groups" $
        sMin64 (tvExp stage1_max_num_groups) $
          toInt64Exp $ Imp.unCount $ segNumGroups lvl

  (stage1_num_threads, elems_per_group, crossesSegment) <-
    scanStage1 pat stage1_num_groups (segGroupSize lvl) space scans kbody

  emit $ Imp.DebugPrint "elems_per_group" $ Just $ untyped elems_per_group

  scanStage2 pat stage1_num_threads elems_per_group stage1_num_groups crossesSegment space scans
  scanStage3 pat (segNumGroups lvl) (segGroupSize lvl) elems_per_group crossesSegment space scans