futhark-0.25.7: src/Futhark/CodeGen/ImpGen/GPU/SegRed.hs
{-# LANGUAGE TypeFamilies #-}
-- | We generate code for non-segmented/single-segment SegRed using
-- the basic approach outlined in the paper "Design and GPGPU
-- Performance of Futhark’s Redomap Construct" (ARRAY '16). The main
-- deviations are:
--
-- * While we still use two-phase reduction, we use only a single
-- kernel, with the final workgroup to write a result (tracked via
-- an atomic counter) performing the final reduction as well.
--
-- * Instead of depending on storage layout transformations to handle
-- non-commutative reductions efficiently, we slide a
-- @groupsize@-sized window over the input, and perform a parallel
-- reduction for each window. This sacrifices the notion of
-- efficient sequentialisation, but is sometimes faster and
-- definitely simpler and more predictable (and uses less auxiliary
-- storage).
--
-- For segmented reductions we use the approach from "Strategies for
-- Regular Segmented Reductions on GPU" (FHPC '17). This involves
-- having two different strategies, and dynamically deciding which one
-- to use based on the number of segments and segment size. We use the
-- (static) @group_size@ to decide which of the following two
-- strategies to choose:
--
-- * Large: uses one or more groups to process a single segment. If
-- multiple groups are used per segment, the intermediate reduction
-- results must be recursively reduced, until there is only a single
-- value per segment.
--
-- Each thread /can/ read multiple elements, which will greatly
-- increase performance; however, if the reduction is
-- non-commutative we will have to use a less efficient traversal
-- (with interim group-wide reductions) to enable coalesced memory
-- accesses, just as in the non-segmented case.
--
-- * Small: is used to let each group process *multiple* segments
-- within a group. We will only use this approach when we can
-- process at least two segments within a single group. In those
-- cases, we would allocate a /whole/ group per segment with the
-- large strategy, but at most 50% of the threads in the group would
-- have any element to read, which becomes highly inefficient.
module Futhark.CodeGen.ImpGen.GPU.SegRed
( compileSegRed,
compileSegRed',
DoSegBody,
)
where
import Control.Monad
import Data.List (genericLength, zip7)
import Data.Maybe
import Futhark.CodeGen.ImpCode.GPU qualified as Imp
import Futhark.CodeGen.ImpGen
import Futhark.CodeGen.ImpGen.GPU.Base
import Futhark.Error
import Futhark.IR.GPUMem
import Futhark.IR.Mem.LMAD qualified as LMAD
import Futhark.Transform.Rename
import Futhark.Util (chunks)
import Futhark.Util.IntegralExp (divUp, quot, rem)
import Prelude hiding (quot, rem)
-- | The maximum number of operators we support in a single SegRed.
-- This limit arises out of the static allocation of counters.
maxNumOps :: Int32
maxNumOps = 10
-- | Code generation for the body of the SegRed, taking a continuation
-- for saving the results of the body. The results should be
-- represented as a pairing of a t'SubExp' along with a list of
-- indexes into that t'SubExp' for reading the result.
type DoSegBody = ([(SubExp, [Imp.TExp Int64])] -> InKernelGen ()) -> InKernelGen ()
-- | Compile 'SegRed' instance to host-level code with calls to
-- various kernels.
compileSegRed ::
Pat LetDecMem ->
SegLevel ->
SegSpace ->
[SegBinOp GPUMem] ->
KernelBody GPUMem ->
CallKernelGen ()
compileSegRed pat lvl space reds body = do
emit $ Imp.DebugPrint "\n# SegRed" Nothing
KernelAttrs _ _ num_groups group_size <- lvlKernelAttrs lvl
let grid = KernelGrid num_groups group_size
compileSegRed' pat grid space reds $ \red_cont ->
compileStms mempty (kernelBodyStms body) $ do
let (red_res, map_res) = splitAt (segBinOpResults reds) $ kernelBodyResult body
sComment "save map-out results" $ do
let map_arrs = drop (segBinOpResults reds) $ patElems pat
zipWithM_ (compileThreadResult space) map_arrs map_res
red_cont $ map ((,[]) . kernelResultSubExp) red_res
emit $ Imp.DebugPrint "" Nothing
-- | Like 'compileSegRed', but where the body is a monadic action.
compileSegRed' ::
Pat LetDecMem ->
KernelGrid ->
SegSpace ->
[SegBinOp GPUMem] ->
DoSegBody ->
CallKernelGen ()
compileSegRed' pat grid space reds body
| genericLength reds > maxNumOps =
compilerLimitationS $
"compileSegRed': at most " ++ show maxNumOps ++ " reduction operators are supported."
| [(_, Constant (IntValue (Int64Value 1))), _] <- unSegSpace space =
nonsegmentedReduction pat num_groups group_size space reds body
| otherwise = do
let group_size' = pe64 $ unCount group_size
segment_size = pe64 $ last $ segSpaceDims space
use_small_segments = segment_size * 2 .<. group_size'
sIf
use_small_segments
(smallSegmentsReduction pat num_groups group_size space reds body)
(largeSegmentsReduction pat num_groups group_size space reds body)
where
num_groups = gridNumGroups grid
group_size = gridGroupSize grid
-- | Prepare intermediate arrays for the reduction. Prim-typed
-- arguments go in local memory (so we need to do the allocation of
-- those arrays inside the kernel), while array-typed arguments go in
-- global memory. Allocations for the former have already been
-- performed. This policy is baked into how the allocations are done
-- in ExplicitAllocations.
intermediateArrays ::
Count GroupSize SubExp ->
SubExp ->
SegBinOp GPUMem ->
InKernelGen [VName]
intermediateArrays (Count group_size) num_threads (SegBinOp _ red_op nes _) = do
let red_op_params = lambdaParams red_op
(red_acc_params, _) = splitAt (length nes) red_op_params
forM red_acc_params $ \p ->
case paramDec p of
MemArray pt shape _ (ArrayIn mem _) -> do
let shape' = Shape [num_threads] <> shape
sArray "red_arr" pt shape' mem $
LMAD.iota 0 (map pe64 $ shapeDims shape')
_ -> do
let pt = elemType $ paramType p
shape = Shape [group_size]
sAllocArray "red_arr" pt shape $ Space "local"
-- | Arrays for storing group results.
--
-- The group-result arrays have an extra dimension because they are
-- also used for keeping vectorised accumulators for first-stage
-- reduction, if necessary. If necessary, this dimension has size
-- group_size, and otherwise 1. When actually storing group results,
-- the first index is set to 0.
groupResultArrays ::
Count NumGroups SubExp ->
Count GroupSize SubExp ->
[SegBinOp GPUMem] ->
CallKernelGen [[VName]]
groupResultArrays (Count virt_num_groups) (Count group_size) reds =
forM reds $ \(SegBinOp _ lam _ shape) ->
forM (lambdaReturnType lam) $ \t -> do
let pt = elemType t
extra_dim
| primType t, shapeRank shape == 0 = intConst Int64 1
| otherwise = group_size
full_shape = Shape [extra_dim, virt_num_groups] <> shape <> arrayShape t
-- Move the groupsize dimension last to ensure coalesced
-- memory access.
perm = [1 .. shapeRank full_shape - 1] ++ [0]
sAllocArrayPerm "segred_tmp" pt full_shape (Space "device") perm
nonsegmentedReduction ::
Pat LetDecMem ->
Count NumGroups SubExp ->
Count GroupSize SubExp ->
SegSpace ->
[SegBinOp GPUMem] ->
DoSegBody ->
CallKernelGen ()
nonsegmentedReduction segred_pat num_groups group_size space reds body = do
let (gtids, dims) = unzip $ unSegSpace space
dims' = map pe64 dims
num_groups' = fmap pe64 num_groups
group_size' = fmap pe64 group_size
global_tid = Imp.le64 $ segFlat space
w = last dims'
counter <- genZeroes "counters" $ fromIntegral maxNumOps
reds_group_res_arrs <- groupResultArrays num_groups group_size reds
num_threads <-
dPrimV "num_threads" $
unCount num_groups' * unCount group_size'
sKernelThread "segred_nonseg" (segFlat space) (defKernelAttrs num_groups group_size) $ do
constants <- kernelConstants <$> askEnv
sync_arr <- sAllocArray "sync_arr" Bool (Shape [intConst Int32 1]) $ Space "local"
reds_arrs <- mapM (intermediateArrays group_size (tvSize num_threads)) reds
-- Since this is the nonsegmented case, all outer segment IDs must
-- necessarily be 0.
forM_ gtids $ \v -> dPrimV_ v (0 :: Imp.TExp Int64)
let num_elements = Imp.elements w
elems_per_thread =
num_elements
`divUp` Imp.elements (sExt64 (kernelNumThreads constants))
slugs <-
mapM (segBinOpSlug (kernelLocalThreadId constants) (kernelGroupId constants)) $
zip3 reds reds_arrs reds_group_res_arrs
reds_op_renamed <-
reductionStageOne
constants
(zip gtids dims')
num_elements
global_tid
elems_per_thread
(tvExp num_threads)
slugs
body
let segred_pes =
chunks (map (length . segBinOpNeutral) reds) $
patElems segred_pat
forM_ (zip7 reds reds_arrs reds_group_res_arrs segred_pes slugs reds_op_renamed [0 ..]) $
\(SegBinOp _ red_op nes _, red_arrs, group_res_arrs, pes, slug, red_op_renamed, i) -> do
let (red_x_params, red_y_params) = splitAt (length nes) $ lambdaParams red_op
reductionStageTwo
constants
pes
(kernelGroupId constants)
0
[0]
0
(sExt64 $ kernelNumGroups constants)
slug
red_x_params
red_y_params
red_op_renamed
nes
1
counter
(fromInteger i)
sync_arr
group_res_arrs
red_arrs
smallSegmentsReduction ::
Pat LetDecMem ->
Count NumGroups SubExp ->
Count GroupSize SubExp ->
SegSpace ->
[SegBinOp GPUMem] ->
DoSegBody ->
CallKernelGen ()
smallSegmentsReduction (Pat segred_pes) num_groups group_size space reds body = do
let (gtids, dims) = unzip $ unSegSpace space
dims' = map pe64 dims
segment_size = last dims'
-- Careful to avoid division by zero now.
segment_size_nonzero <-
dPrimVE "segment_size_nonzero" $ sMax64 1 segment_size
let num_groups' = fmap pe64 num_groups
group_size' = fmap pe64 group_size
num_threads <- dPrimV "num_threads" $ unCount num_groups' * unCount group_size'
let num_segments = product $ init dims'
segments_per_group = unCount group_size' `quot` segment_size_nonzero
required_groups = sExt32 $ num_segments `divUp` segments_per_group
emit $ Imp.DebugPrint "# SegRed-small" Nothing
emit $ Imp.DebugPrint "num_segments" $ Just $ untyped num_segments
emit $ Imp.DebugPrint "segment_size" $ Just $ untyped segment_size
emit $ Imp.DebugPrint "segments_per_group" $ Just $ untyped segments_per_group
emit $ Imp.DebugPrint "required_groups" $ Just $ untyped required_groups
sKernelThread "segred_small" (segFlat space) (defKernelAttrs num_groups group_size) $ do
constants <- kernelConstants <$> askEnv
reds_arrs <- mapM (intermediateArrays group_size (Var $ tvVar num_threads)) reds
-- We probably do not have enough actual workgroups to cover the
-- entire iteration space. Some groups thus have to perform double
-- duty; we put an outer loop to accomplish this.
virtualiseGroups SegVirt required_groups $ \group_id' -> do
-- Compute the 'n' input indices. The outer 'n-1' correspond to
-- the segment ID, and are computed from the group id. The inner
-- is computed from the local thread id, and may be out-of-bounds.
let ltid = sExt64 $ kernelLocalThreadId constants
segment_index =
(ltid `quot` segment_size_nonzero)
+ (sExt64 group_id' * sExt64 segments_per_group)
index_within_segment = ltid `rem` segment_size
dIndexSpace (zip (init gtids) (init dims')) segment_index
dPrimV_ (last gtids) index_within_segment
let out_of_bounds =
forM_ (zip reds reds_arrs) $ \(SegBinOp _ _ nes _, red_arrs) ->
forM_ (zip red_arrs nes) $ \(arr, ne) ->
copyDWIMFix arr [ltid] ne []
in_bounds =
body $ \red_res ->
sComment "save results to be reduced" $ do
let red_dests = map (,[ltid]) (concat reds_arrs)
forM_ (zip red_dests red_res) $ \((d, d_is), (res, res_is)) ->
copyDWIMFix d d_is res res_is
sComment "apply map function if in bounds" $
sIf
( segment_size
.>. 0
.&&. isActive (init $ zip gtids dims)
.&&. ltid
.<. segment_size
* segments_per_group
)
in_bounds
out_of_bounds
sOp $ Imp.ErrorSync Imp.FenceLocal -- Also implicitly barrier.
let crossesSegment from to =
(sExt64 to - sExt64 from) .>. (sExt64 to `rem` segment_size)
sWhen (segment_size .>. 0) $
sComment "perform segmented scan to imitate reduction" $
forM_ (zip reds reds_arrs) $ \(SegBinOp _ red_op _ _, red_arrs) ->
groupScan
(Just crossesSegment)
(sExt64 $ tvExp num_threads)
(segment_size * segments_per_group)
red_op
red_arrs
sOp $ Imp.Barrier Imp.FenceLocal
sComment "save final values of segments"
$ sWhen
( sExt64 group_id'
* segments_per_group
+ sExt64 ltid
.<. num_segments
.&&. ltid
.<. segments_per_group
)
$ forM_ (zip segred_pes (concat reds_arrs))
$ \(pe, arr) -> do
-- Figure out which segment result this thread should write...
let flat_segment_index =
sExt64 group_id' * segments_per_group + sExt64 ltid
gtids' =
unflattenIndex (init dims') flat_segment_index
copyDWIMFix
(patElemName pe)
gtids'
(Var arr)
[(ltid + 1) * segment_size_nonzero - 1]
-- Finally another barrier, because we will be writing to the
-- local memory array first thing in the next iteration.
sOp $ Imp.Barrier Imp.FenceLocal
largeSegmentsReduction ::
Pat LetDecMem ->
Count NumGroups SubExp ->
Count GroupSize SubExp ->
SegSpace ->
[SegBinOp GPUMem] ->
DoSegBody ->
CallKernelGen ()
largeSegmentsReduction segred_pat num_groups group_size space reds body = do
let (gtids, dims) = unzip $ unSegSpace space
dims' = map pe64 dims
num_segments = product $ init dims'
segment_size = last dims'
num_groups' = fmap pe64 num_groups
group_size' = fmap pe64 group_size
(groups_per_segment, elems_per_thread) <-
groupsPerSegmentAndElementsPerThread
segment_size
num_segments
num_groups'
group_size'
virt_num_groups <-
dPrimV "virt_num_groups" $
groups_per_segment * num_segments
num_threads <-
dPrimV "num_threads" $
unCount num_groups' * unCount group_size'
threads_per_segment <-
dPrimV "threads_per_segment" $
groups_per_segment * unCount group_size'
emit $ Imp.DebugPrint "# SegRed-large" Nothing
emit $ Imp.DebugPrint "num_segments" $ Just $ untyped num_segments
emit $ Imp.DebugPrint "segment_size" $ Just $ untyped segment_size
emit $ Imp.DebugPrint "virt_num_groups" $ Just $ untyped $ tvExp virt_num_groups
emit $ Imp.DebugPrint "num_groups" $ Just $ untyped $ Imp.unCount num_groups'
emit $ Imp.DebugPrint "group_size" $ Just $ untyped $ Imp.unCount group_size'
emit $ Imp.DebugPrint "elems_per_thread" $ Just $ untyped $ Imp.unCount elems_per_thread
emit $ Imp.DebugPrint "groups_per_segment" $ Just $ untyped groups_per_segment
reds_group_res_arrs <- groupResultArrays (Count (tvSize virt_num_groups)) group_size reds
-- In principle we should have a counter for every segment. Since
-- the number of segments is a dynamic quantity, we would have to
-- allocate and zero out an array here, which is expensive.
-- However, we exploit the fact that the number of segments being
-- reduced at any point in time is limited by the number of
-- workgroups. If we bound the number of workgroups, we can get away
-- with using that many counters. FIXME: Is this limit checked
-- anywhere? There are other places in the compiler that will fail
-- if the group count exceeds the maximum group size, which is at
-- most 1024 anyway.
let num_counters = fromIntegral maxNumOps * 1024
counter <- genZeroes "counters" num_counters
sKernelThread "segred_large" (segFlat space) (defKernelAttrs num_groups group_size) $ do
constants <- kernelConstants <$> askEnv
reds_arrs <- mapM (intermediateArrays group_size (tvSize num_threads)) reds
sync_arr <- sAllocArray "sync_arr" Bool (Shape [intConst Int32 1]) $ Space "local"
-- We probably do not have enough actual workgroups to cover the
-- entire iteration space. Some groups thus have to perform double
-- duty; we put an outer loop to accomplish this.
virtualiseGroups SegVirt (sExt32 (tvExp virt_num_groups)) $ \group_id -> do
let segment_gtids = init gtids
w = last dims
local_tid = kernelLocalThreadId constants
flat_segment_id <-
dPrimVE "flat_segment_id" $
group_id `quot` sExt32 groups_per_segment
global_tid <-
dPrimVE "global_tid" $
(sExt64 group_id * sExt64 (unCount group_size') + sExt64 local_tid)
`rem` (sExt64 (unCount group_size') * groups_per_segment)
let first_group_for_segment = sExt64 flat_segment_id * groups_per_segment
dIndexSpace (zip segment_gtids (init dims')) $ sExt64 flat_segment_id
dPrim_ (last gtids) int64
let num_elements = Imp.elements $ pe64 w
slugs <-
mapM (segBinOpSlug local_tid group_id) $
zip3 reds reds_arrs reds_group_res_arrs
reds_op_renamed <-
reductionStageOne
constants
(zip gtids dims')
num_elements
global_tid
elems_per_thread
(tvExp threads_per_segment)
slugs
body
let segred_pes =
chunks (map (length . segBinOpNeutral) reds) $
patElems segred_pat
multiple_groups_per_segment =
forM_ (zip7 reds reds_arrs reds_group_res_arrs segred_pes slugs reds_op_renamed [0 ..]) $
\(SegBinOp _ red_op nes _, red_arrs, group_res_arrs, pes, slug, red_op_renamed, i) -> do
let (red_x_params, red_y_params) =
splitAt (length nes) $ lambdaParams red_op
reductionStageTwo
constants
pes
group_id
flat_segment_id
(map Imp.le64 segment_gtids)
(sExt64 first_group_for_segment)
groups_per_segment
slug
red_x_params
red_y_params
red_op_renamed
nes
(fromIntegral num_counters)
counter
(fromInteger i)
sync_arr
group_res_arrs
red_arrs
one_group_per_segment =
comment "first thread in group saves final result to memory" $
forM_ (zip slugs segred_pes) $ \(slug, pes) ->
sWhen (local_tid .==. 0) $
forM_ (zip pes (slugAccs slug)) $ \(v, (acc, acc_is)) ->
copyDWIMFix (patElemName v) (map Imp.le64 segment_gtids) (Var acc) acc_is
sIf (groups_per_segment .==. 1) one_group_per_segment multiple_groups_per_segment
-- Careful to avoid division by zero here. We have at least one group
-- per segment.
groupsPerSegmentAndElementsPerThread ::
Imp.TExp Int64 ->
Imp.TExp Int64 ->
Count NumGroups (Imp.TExp Int64) ->
Count GroupSize (Imp.TExp Int64) ->
CallKernelGen
( Imp.TExp Int64,
Imp.Count Imp.Elements (Imp.TExp Int64)
)
groupsPerSegmentAndElementsPerThread segment_size num_segments num_groups_hint group_size = do
groups_per_segment <-
dPrimVE "groups_per_segment" $
unCount num_groups_hint `divUp` sMax64 1 num_segments
elements_per_thread <-
dPrimVE "elements_per_thread" $
segment_size `divUp` (unCount group_size * groups_per_segment)
pure (groups_per_segment, Imp.elements elements_per_thread)
-- | A SegBinOp with auxiliary information.
data SegBinOpSlug = SegBinOpSlug
{ slugOp :: SegBinOp GPUMem,
-- | The arrays used for computing the intra-group reduction
-- (either local or global memory).
slugArrs :: [VName],
-- | Places to store accumulator in stage 1 reduction.
slugAccs :: [(VName, [Imp.TExp Int64])]
}
slugBody :: SegBinOpSlug -> Body GPUMem
slugBody = lambdaBody . segBinOpLambda . slugOp
slugParams :: SegBinOpSlug -> [LParam GPUMem]
slugParams = lambdaParams . segBinOpLambda . slugOp
slugNeutral :: SegBinOpSlug -> [SubExp]
slugNeutral = segBinOpNeutral . slugOp
slugShape :: SegBinOpSlug -> Shape
slugShape = segBinOpShape . slugOp
slugsComm :: [SegBinOpSlug] -> Commutativity
slugsComm = mconcat . map (segBinOpComm . slugOp)
accParams, nextParams :: SegBinOpSlug -> [LParam GPUMem]
accParams slug = take (length (slugNeutral slug)) $ slugParams slug
nextParams slug = drop (length (slugNeutral slug)) $ slugParams slug
segBinOpSlug :: Imp.TExp Int32 -> Imp.TExp Int32 -> (SegBinOp GPUMem, [VName], [VName]) -> InKernelGen SegBinOpSlug
segBinOpSlug local_tid group_id (op, group_res_arrs, param_arrs) =
SegBinOpSlug op group_res_arrs
<$> zipWithM mkAcc (lambdaParams (segBinOpLambda op)) param_arrs
where
mkAcc p param_arr
| Prim t <- paramType p,
shapeRank (segBinOpShape op) == 0 = do
acc <- dPrim (baseString (paramName p) <> "_acc") t
pure (tvVar acc, [])
| otherwise =
pure (param_arr, [sExt64 local_tid, sExt64 group_id])
computeThreadChunkSize ::
Commutativity ->
Imp.TExp Int64 ->
Imp.TExp Int64 ->
Imp.Count Imp.Elements (Imp.TExp Int64) ->
Imp.Count Imp.Elements (Imp.TExp Int64) ->
TV Int64 ->
ImpM rep r op ()
computeThreadChunkSize Commutative threads_per_segment thread_index elements_per_thread num_elements chunk_var =
chunk_var
<-- sMin64
(Imp.unCount elements_per_thread)
((Imp.unCount num_elements - thread_index) `divUp` threads_per_segment)
computeThreadChunkSize Noncommutative _ thread_index elements_per_thread num_elements chunk_var = do
starting_point <-
dPrimV "starting_point" $
thread_index * Imp.unCount elements_per_thread
remaining_elements <-
dPrimV "remaining_elements" $
Imp.unCount num_elements - tvExp starting_point
let no_remaining_elements = tvExp remaining_elements .<=. 0
beyond_bounds = Imp.unCount num_elements .<=. tvExp starting_point
sIf
(no_remaining_elements .||. beyond_bounds)
(chunk_var <-- 0)
( sIf
is_last_thread
(chunk_var <-- Imp.unCount last_thread_elements)
(chunk_var <-- Imp.unCount elements_per_thread)
)
where
last_thread_elements =
num_elements - Imp.elements thread_index * elements_per_thread
is_last_thread =
Imp.unCount num_elements
.<. (thread_index + 1)
* Imp.unCount elements_per_thread
reductionStageZero ::
KernelConstants ->
[(VName, Imp.TExp Int64)] ->
Imp.Count Imp.Elements (Imp.TExp Int64) ->
Imp.TExp Int64 ->
Imp.Count Imp.Elements (Imp.TExp Int64) ->
Imp.TExp Int64 ->
[SegBinOpSlug] ->
DoSegBody ->
InKernelGen ([Lambda GPUMem], InKernelGen ())
reductionStageZero constants ispace num_elements global_tid elems_per_thread threads_per_segment slugs body = do
let (gtids, _dims) = unzip ispace
gtid = mkTV (last gtids) int64
local_tid = sExt64 $ kernelLocalThreadId constants
-- Figure out how many elements this thread should process.
chunk_size <- dPrim "chunk_size" int64
computeThreadChunkSize
(slugsComm slugs)
threads_per_segment
(sExt64 global_tid)
elems_per_thread
num_elements
chunk_size
dScope Nothing $ scopeOfLParams $ concatMap slugParams slugs
sComment "neutral-initialise the accumulators" $
forM_ slugs $ \slug ->
forM_ (zip (slugAccs slug) (slugNeutral slug)) $ \((acc, acc_is), ne) ->
sLoopNest (slugShape slug) $ \vec_is ->
copyDWIMFix acc (acc_is ++ vec_is) ne []
slugs_op_renamed <- mapM (renameLambda . segBinOpLambda . slugOp) slugs
let doTheReduction =
forM_ (zip slugs_op_renamed slugs) $ \(slug_op_renamed, slug) ->
sLoopNest (slugShape slug) $ \vec_is -> do
comment "to reduce current chunk, first store our result in memory" $ do
forM_ (zip (slugParams slug) (slugAccs slug)) $ \(p, (acc, acc_is)) ->
copyDWIMFix (paramName p) [] (Var acc) (acc_is ++ vec_is)
forM_ (zip (slugArrs slug) (slugParams slug)) $ \(arr, p) ->
when (primType $ paramType p) $
copyDWIMFix arr [local_tid] (Var $ paramName p) []
sOp $ Imp.ErrorSync Imp.FenceLocal -- Also implicitly barrier.
groupReduce (sExt32 (kernelGroupSize constants)) slug_op_renamed (slugArrs slug)
sOp $ Imp.Barrier Imp.FenceLocal
sComment "first thread saves the result in accumulator" $
sWhen (local_tid .==. 0) $
forM_ (zip (slugAccs slug) (lambdaParams slug_op_renamed)) $ \((acc, acc_is), p) ->
copyDWIMFix acc (acc_is ++ vec_is) (Var $ paramName p) []
-- If this is a non-commutative reduction, each thread must run the
-- loop the same number of iterations, because we will be performing
-- a group-wide reduction in there.
let comm = slugsComm slugs
(bound, check_bounds) =
case comm of
Commutative -> (tvExp chunk_size, id)
Noncommutative ->
( Imp.unCount elems_per_thread,
sWhen (tvExp gtid .<. Imp.unCount num_elements)
)
sFor "i" bound $ \i -> do
gtid
<-- case comm of
Commutative ->
global_tid + threads_per_segment * i
Noncommutative ->
let index_in_segment = global_tid `quot` kernelGroupSize constants
in sExt64 local_tid
+ (index_in_segment * Imp.unCount elems_per_thread + i)
* kernelGroupSize constants
check_bounds $
sComment "apply map function" $
body $ \all_red_res -> do
let slugs_res = chunks (map (length . slugNeutral) slugs) all_red_res
forM_ (zip slugs slugs_res) $ \(slug, red_res) ->
sLoopNest (slugShape slug) $ \vec_is -> do
sComment "load accumulator" $
forM_ (zip (accParams slug) (slugAccs slug)) $ \(p, (acc, acc_is)) ->
copyDWIMFix (paramName p) [] (Var acc) (acc_is ++ vec_is)
sComment "load new values" $
forM_ (zip (nextParams slug) red_res) $ \(p, (res, res_is)) ->
copyDWIMFix (paramName p) [] res (res_is ++ vec_is)
sComment "apply reduction operator"
$ compileStms mempty (bodyStms $ slugBody slug)
$ sComment "store in accumulator"
$ forM_
( zip
(slugAccs slug)
(map resSubExp $ bodyResult $ slugBody slug)
)
$ \((acc, acc_is), se) ->
copyDWIMFix acc (acc_is ++ vec_is) se []
case comm of
Noncommutative -> do
doTheReduction
sComment "first thread keeps accumulator; others reset to neutral element" $ do
let reset_to_neutral =
forM_ slugs $ \slug ->
forM_ (zip (slugAccs slug) (slugNeutral slug)) $ \((acc, acc_is), ne) ->
sLoopNest (slugShape slug) $ \vec_is ->
copyDWIMFix acc (acc_is ++ vec_is) ne []
sUnless (local_tid .==. 0) reset_to_neutral
_ -> pure ()
sOp $ Imp.ErrorSync Imp.FenceLocal
pure (slugs_op_renamed, doTheReduction)
reductionStageOne ::
KernelConstants ->
[(VName, Imp.TExp Int64)] ->
Imp.Count Imp.Elements (Imp.TExp Int64) ->
Imp.TExp Int64 ->
Imp.Count Imp.Elements (Imp.TExp Int64) ->
Imp.TExp Int64 ->
[SegBinOpSlug] ->
DoSegBody ->
InKernelGen [Lambda GPUMem]
reductionStageOne constants ispace num_elements global_tid elems_per_thread threads_per_segment slugs body = do
(slugs_op_renamed, doTheReduction) <-
reductionStageZero constants ispace num_elements global_tid elems_per_thread threads_per_segment slugs body
case slugsComm slugs of
Noncommutative -> pure ()
Commutative -> doTheReduction
pure slugs_op_renamed
reductionStageTwo ::
KernelConstants ->
[PatElem LetDecMem] ->
Imp.TExp Int32 ->
Imp.TExp Int32 ->
[Imp.TExp Int64] ->
Imp.TExp Int64 ->
Imp.TExp Int64 ->
SegBinOpSlug ->
[LParam GPUMem] ->
[LParam GPUMem] ->
Lambda GPUMem ->
[SubExp] ->
Imp.TExp Int32 ->
VName ->
Imp.TExp Int32 ->
VName ->
[VName] ->
[VName] ->
InKernelGen ()
reductionStageTwo
constants
segred_pes
group_id
flat_segment_id
segment_gtids
first_group_for_segment
groups_per_segment
slug
red_x_params
red_y_params
red_op_renamed
nes
num_counters
counter
counter_i
sync_arr
group_res_arrs
red_arrs = do
let local_tid = kernelLocalThreadId constants
group_size = kernelGroupSize constants
old_counter <- dPrim "old_counter" int32
(counter_mem, _, counter_offset) <-
fullyIndexArray
counter
[ sExt64 $
counter_i * num_counters
+ flat_segment_id `rem` num_counters
]
comment "first thread in group saves group result to global memory" $
sWhen (local_tid .==. 0) $ do
forM_ (take (length nes) $ zip group_res_arrs (slugAccs slug)) $ \(v, (acc, acc_is)) ->
copyDWIMFix v [0, sExt64 group_id] (Var acc) acc_is
sOp $ Imp.MemFence Imp.FenceGlobal
-- Increment the counter, thus stating that our result is
-- available.
sOp
$ Imp.Atomic DefaultSpace
$ Imp.AtomicAdd
Int32
(tvVar old_counter)
counter_mem
counter_offset
$ untyped (1 :: Imp.TExp Int32)
-- Now check if we were the last group to write our result. If
-- so, it is our responsibility to produce the final result.
sWrite sync_arr [0] $ untyped $ tvExp old_counter .==. groups_per_segment - 1
sOp $ Imp.Barrier Imp.FenceGlobal
is_last_group <- dPrim "is_last_group" Bool
copyDWIMFix (tvVar is_last_group) [] (Var sync_arr) [0]
sWhen (tvExp is_last_group) $ do
-- The final group has written its result (and it was
-- us!), so read in all the group results and perform the
-- final stage of the reduction. But first, we reset the
-- counter so it is ready for next time. This is done
-- with an atomic to avoid warnings about write/write
-- races in oclgrind.
sWhen (local_tid .==. 0) $
sOp $
Imp.Atomic DefaultSpace $
Imp.AtomicAdd Int32 (tvVar old_counter) counter_mem counter_offset $
untyped $
negate groups_per_segment
sLoopNest (slugShape slug) $ \vec_is -> do
unless (null $ slugShape slug) $
sOp (Imp.Barrier Imp.FenceLocal)
-- There is no guarantee that the number of workgroups for the
-- segment is less than the workgroup size, so each thread may
-- have to read multiple elements. We do this in a sequential
-- way that may induce non-coalesced accesses, but the total
-- number of accesses should be tiny here.
comment "read in the per-group-results" $ do
read_per_thread <-
dPrimVE "read_per_thread" $
groups_per_segment `divUp` sExt64 group_size
forM_ (zip red_x_params nes) $ \(p, ne) ->
copyDWIMFix (paramName p) [] ne []
sFor "i" read_per_thread $ \i -> do
group_res_id <-
dPrimVE "group_res_id" $
sExt64 local_tid * read_per_thread + i
index_of_group_res <-
dPrimVE "index_of_group_res" $
first_group_for_segment + group_res_id
sWhen (group_res_id .<. groups_per_segment) $ do
forM_ (zip red_y_params group_res_arrs) $
\(p, group_res_arr) ->
copyDWIMFix
(paramName p)
[]
(Var group_res_arr)
([0, index_of_group_res] ++ vec_is)
compileStms mempty (bodyStms $ slugBody slug) $
forM_ (zip red_x_params $ map resSubExp $ bodyResult $ slugBody slug) $ \(p, se) ->
copyDWIMFix (paramName p) [] se []
forM_ (zip red_x_params red_arrs) $ \(p, arr) ->
when (primType $ paramType p) $
copyDWIMFix arr [sExt64 local_tid] (Var $ paramName p) []
sOp $ Imp.ErrorSync Imp.FenceLocal
sComment "reduce the per-group results" $ do
groupReduce (sExt32 group_size) red_op_renamed red_arrs
sComment "and back to memory with the final result" $
sWhen (local_tid .==. 0) $
forM_ (zip segred_pes $ lambdaParams red_op_renamed) $ \(pe, p) ->
copyDWIMFix
(patElemName pe)
(segment_gtids ++ vec_is)
(Var $ paramName p)
[]