35d888c · streamly · git.iko.soy

git.iko.soy/streamly

Categorize o-1/o-n benchmarks correctly

commit35d888c2f9484aea2ae3da3ee87e6f032862d062 authorHarendra Kumar <harendra@composewell.com> dateThu 9 Mar 2023, 16:21 (4 years ago) parent86cc020

10 files changed+65−42

compare with master

unifiedsplit

.hlint.yaml+1−3

@@ -24,14 +24,12 @@

2424- ignore: {name: "Use uncurry"}

2525- ignore: {name: "Redundant $!"}

2626- ignore: {name: "Use fmap"}

27+- ignore: {name: "Use ++"}

2728

2829# Warnings ignored in specific places

29−- ignore: {name: "Use ++", within: Stream.Transform}

3030- ignore: {name: "Use mapM", within: Stream.Transform}

3131- ignore: {name: "Use traverse", within: Stream.Transform}

3232- ignore: {name: "Redundant <*", within: Stream.Expand}

33−- ignore: {name: "Use ++", within: Stream.Reduce}

34−- ignore: {name: "Use ++", within: Stream.Split}

3533- ignore: {name: "Redundant bracket", within: Stream.Split}

3634- ignore: {name: "Use isDigit", within: Streamly.Internal.Unicode.Parser}

3735

benchmark/Streamly/Benchmark/Data/Fold.hs+1−1

@@ -306,7 +306,6 @@ o_1_space_serial_elimination value =

306306 , benchIOSink value "drainBy" (Stream.fold (FL.drainBy return))

307307 , benchIOSink value "drainN" (Stream.fold (FL.drainN value))

308308 , benchIOSink value "last" (Stream.fold FL.last)

309− , benchIOSink value "nub" (Stream.fold FL.nub)

310309 , benchIOSink value "length" (Stream.fold FL.length)

311310 , benchIOSink value "top" (Stream.fold $ FL.top 10)

312311 , benchIOSink value "bottom" (Stream.fold $ FL.bottom 10)

@@ -449,6 +448,7 @@ o_n_heap_serial value =

449448 , benchIOSink value "toStreamRev"

450449 (Stream.fold FL.toStreamRev

451450 :: Stream IO a -> IO (Stream Identity a))

451+ , benchIOSink value "nub" (Stream.fold FL.nub)

452452 ]

453453 , bgroup "key-value"

454454 [

benchmark/Streamly/Benchmark/Data/Fold/Window.hs+24−11

@@ -12,6 +12,7 @@ import qualified Streamly.Internal.Data.Ring.Unboxed as Ring

1212import qualified Streamly.Internal.Data.Fold.Window as Window

1313import qualified Streamly.Internal.Data.Stream.StreamD as Stream

1414

15+import Streamly.Benchmark.Common

1516import Gauge

1617

1718{-# INLINE source #-}

@@ -65,15 +66,9 @@ benchScanWith src len name f =

6566benchWithPostscan :: Int -> String -> Fold IO Double a -> Benchmark

6667benchWithPostscan = benchScanWith source

6768

68−{-# INLINE numElements #-}

69−numElements :: Int

70−numElements = 100000

71−

72−main :: IO ()

73−main =

74− defaultMain

75− [ bgroup

76− "fold"

69+o_1_space_folds :: Int -> [Benchmark]

70+o_1_space_folds numElements =

71+ [ bgroup "fold"

7772 [ benchWithFold numElements "minimum (window size 100)"

7873 (Window.minimum 100)

7974 , benchWithFold numElements "minimum (window size 1000)"

@@ -126,8 +121,11 @@ main =

126121 (Window.cumulative (Window.powerSum 2))

127122

128123 ]

129− , bgroup

130− "scan"

124+ ]

125+

126+o_1_space_scans :: Int -> [Benchmark]

127+o_1_space_scans numElements =

128+ [ bgroup "scan"

131129 [ benchWithPostscan numElements "minimum (window size 10)"

132130 (Window.minimum 10)

133131 -- Below window size 30 the linear search based impl performs better

@@ -176,3 +174,18 @@ main =

176174 (Ring.slidingWindow 1000 (Window.powerSum 2))

177175 ]

178176 ]

177+

178+moduleName :: String

179+moduleName = "Data.Fold.Window"

180+

181+main :: IO ()

182+main = runWithCLIOpts defaultStreamSize allBenchmarks

183+

184+ where

185+

186+ allBenchmarks value =

187+ [ bgroup (o_1_space_prefix moduleName) $ concat

188+ [ o_1_space_folds value

189+ , o_1_space_scans value

190+ ]

191+ ]

benchmark/Streamly/Benchmark/Data/Parser.hs+8−5

@@ -723,14 +723,11 @@ o_1_space_serial value =

723723 , benchIOSink value "dropWhile" $ dropWhile value

724724 , benchIOSink value "takeStartBy" $ takeStartBy value

725725 , benchIOSink value "takeEndBy_" $ takeEndBy_ value

726− , benchIOSrc sourceEscapedFrames value "takeFramedByEsc_"

727− $ takeFramedByEsc_ value

728726 , benchIOSink value "groupBy" $ groupBy

729727 , benchIOSink value "groupByRolling" $ groupByRolling

730728 , benchIOSink value "wordBy" $ wordBy value

731729 , benchIOSink value "sepBy (words)" sepByWords

732730 , benchIOSink value "sepByAll (words)" sepByAllWords

733− , benchIOSink value "sepBy1" sepBy1

734731 , benchIOSink value "sepBy1 (words)" sepByWords1

735732 , benchIOSink value "deintercalate" $ deintercalate value

736733 , benchIOSink value "deintercalate1" $ deintercalate1 value

@@ -748,7 +745,6 @@ o_1_space_serial value =

748745 , benchIOSink value "monad2" $ monad value

749746 , benchIOSink value "monad4" $ monad4 value

750747 , benchIOSink value "monad8" $ monad8 value

751− , benchIOSink value "monad16" $ monad16 value

752748 -- Alternative

753749 , benchIOSink value "alt2parseMany" $ altSmall value

754750 , benchIOSink value "alt2" $ alt2 value

@@ -783,7 +779,6 @@ o_1_space_serial value =

783779 , benchIOSink value "shortest" $ shortestAllAny value

784780 , benchIOSink value "longest" $ longestAllAny value

785781 -}

786− , benchIOSink value "listEqBy" (listEqBy value)

787782 , benchIOSink value "streamEqBy" (streamEqBy value)

788783 ]

789784

@@ -821,13 +816,21 @@ o_n_heap_serial value =

821816 -- lookahead benchmark holds the entire input till end

822817 benchIOSink value "lookAhead" $ lookAhead value

823818

819+ -- o-n-heap because of backtracking

820+ , benchIOSrc sourceEscapedFrames value "takeFramedByEsc_"

821+ $ takeFramedByEsc_ value

822+

824823 -- non-linear time complexity (parserD)

825824 , benchIOSink value "split_" $ split_ value

825+ -- XXX Takes lot of space when run on a long stream, why?

826+ , benchIOSink value "monad16" $ monad16 value

826827

827828 -- These show non-linear time complexity.

828829 -- They accumulate the results in a list.

830+ , benchIOSink value "sepBy1" sepBy1

829831 , benchIOSink value "manyAlt" manyAlt

830832 , benchIOSink value "someAlt" someAlt

833+ , benchIOSink value "listEqBy" (listEqBy value)

831834 ]

832835

833836-- accumulate results in a list in IO

benchmark/Streamly/Benchmark/Data/Stream/Concurrent.hs+1−1

@@ -17,4 +17,4 @@ moduleName = "Data.Stream.Concurrent"

1717-------------------------------------------------------------------------------

1818

1919main :: IO ()

20−main = runWithCLIOpts defaultStreamSize (allBenchmarks moduleName id)

20+main = runWithCLIOpts defaultStreamSize (allBenchmarks moduleName False id)

benchmark/Streamly/Benchmark/Data/Stream/ConcurrentCommon.hs+6−5

@@ -193,15 +193,16 @@ o_1_space_outerProduct value f =

193193-- Benchmark sets

194194-------------------------------------------------------------------------------

195195

196−allBenchmarks :: String -> (Config -> Config) -> Int -> [Benchmark]

197−allBenchmarks moduleName modifier value =

196+allBenchmarks :: String -> Bool -> (Config -> Config) -> Int -> [Benchmark]

197+allBenchmarks moduleName wide modifier value =

198198 [ bgroup (o_1_space_prefix moduleName) $ concat

199199 [ o_1_space_mapping value modifier

200200 , o_1_space_concatFoldable value modifier

201201 , o_1_space_concatMap value modifier

202− , o_1_space_outerProduct value modifier

203202 , o_1_space_joining value modifier

203+ ] ++ if wide then [] else o_1_space_outerProduct value modifier

204+ , bgroup (o_n_heap_prefix moduleName) $ concat

205+ [ if wide then o_1_space_outerProduct value modifier else []

206+ , o_n_heap_buffering value modifier

204207 ]

205− , bgroup (o_n_heap_prefix moduleName)

206− (o_n_heap_buffering value modifier)

207208 ]

benchmark/Streamly/Benchmark/Data/Stream/ConcurrentEager.hs+1−1

@@ -22,4 +22,4 @@ main :: IO ()

2222main =

2323 runWithCLIOpts

2424 defaultStreamSize

25− (allBenchmarks moduleName (Async.eager True))

25+ (allBenchmarks moduleName True (Async.eager True))

benchmark/Streamly/Benchmark/Data/Stream/ConcurrentInterleaved.hs+1−1

@@ -22,4 +22,4 @@ main :: IO ()

2222main =

2323 runWithCLIOpts

2424 defaultStreamSize

25− (allBenchmarks moduleName (Async.interleaved True))

25+ (allBenchmarks moduleName True (Async.interleaved True))

benchmark/Streamly/Benchmark/Data/Stream/ConcurrentOrdered.hs+1−1

@@ -22,4 +22,4 @@ main :: IO ()

2222main =

2323 runWithCLIOpts

2424 defaultStreamSize

25− (allBenchmarks moduleName (Async.ordered True))

25+ (allBenchmarks moduleName False (Async.ordered True))

benchmark/bench-runner/Main.hs+21−13

@@ -47,6 +47,16 @@ rtsOpts exeName benchName0 = unwords [general, exeSpecific, benchSpecific]

4747 | "Prelude.WAsync/o-n-space.monad-outer-product." `isPrefixOf` benchName =

4848 "-K4M"

4949

50+ -- This module is dev only now, and can be removed at some point

51+ | "Data.Stream.StreamDK/o-1-space.grouping.classifySessionsOf"

52+ `isPrefixOf` benchName = "-K512K"

53+ | "Data.Stream.StreamDK/o-n-space.foldr.foldrM/"

54+ `isPrefixOf` benchName = "-K4M"

55+ | "Data.Stream.StreamDK/o-n-space.iterated."

56+ `isPrefixOf` benchName = "-K4M -M64M"

57+ | "Data.Stream.StreamDK/o-n-space.traversable."

58+ `isPrefixOf` benchName = "-K2M"

59+

5060 -----------------------------------------------------------------------

5161

5262 | "Data.Stream.StreamD/o-n-space.elimination.toList" == benchName =

@@ -63,19 +73,7 @@ rtsOpts exeName benchName0 = unwords [general, exeSpecific, benchSpecific]

6373 | "Data.Stream/o-n-space.iterated."

6474 `isPrefixOf` benchName = "-K4M"

6575

66− | "Data.Stream.StreamDK/o-1-space.grouping.classifySessionsOf"

67− `isPrefixOf` benchName = "-K512K"

68− | "Data.Stream.StreamDK/o-n-space.foldr.foldrM/"

69− `isPrefixOf` benchName = "-K4M"

70− | "Data.Stream.StreamDK/o-n-space.iterated."

71− `isPrefixOf` benchName = "-K4M -M64M"

72− | "Data.Stream.StreamDK/o-n-space.traversable."

73− `isPrefixOf` benchName = "-K2M"

74−

75− | "Data.Stream.ConcurrentInterleaved/o-1-space.monad-outer-product.toNullAp"

76− `isPrefixOf` benchName = "-M32M"

77−

78− | "Data.Stream.ConcurrentEager/o-1-space.monad-outer-product.toNullAp"

76+ | "Data.Stream.ConcurrentEager/o-n-space.monad-outer-product.toNullAp"

7977 `isPrefixOf` benchName = "-M768M"

8078 | "Data.Stream.ConcurrentEager/o-1-space."

8179 `isPrefixOf` benchName = "-M128M"

@@ -89,6 +87,16 @@ rtsOpts exeName benchName0 = unwords [general, exeSpecific, benchSpecific]

8987 | "Data.Array.Generic/o-1-space.transformationX4.map"

9088 `isPrefixOf` benchName = "-M32M"

9189

90+ -- XXX For --long option, need to check why so much heap is required.

91+ -- Note, if we remove the chunked stream module we need to keep the

92+ -- chunked stream benchmarks in the stream module.

93+ | "Data.Array.Stream/o-1-space"

94+ `isPrefixOf` benchName = "-K4M -M512M"

95+ -- XXX Takes up to 160MB heap for --long, we use chunked stream for

96+ -- this, so the reason may be related to chunked streams.

97+ | "Data.Parser.ParserK/o-1-space"

98+ `isPrefixOf` benchName = "-K4M -M256M"

99+

92100 -----------------------------------------------------------------------

93101

94102 | "Unicode.Char/o-1-space." `isPrefixOf` benchName = "-M32M"