Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Compare Benchmarks and Compiler Optimization I...
Search
po3rin
May 27, 2019
Programming
4
850
Compare Benchmarks and Compiler Optimization In Go
Go(Un) Conference #6
po3rin
May 27, 2019
Tweet
Share
More Decks by po3rin
See All by po3rin
自作LLM Native GORM Pluginで実現する AI Agentバックテスト基盤構築
po3rin
2
240
麻雀点数計算問題生成タスクから学ぶ Single Agentの限界と Agentic Workflowの底力
po3rin
5
2.1k
LLMOpsのパフォーマンスを支える技術と現場で実践した改善
po3rin
8
1.1k
LLMは麻雀を知らなすぎるから俺が教育してやる
po3rin
4
2.4k
M3におけるCI/CDパイプラインを新鮮に保つ仕組み
po3rin
0
380
Go Templateを使った フロントエンド開発を採用した 経緯と所感
po3rin
1
360
検索基盤移行時の思考
po3rin
0
4.3k
Go1.19で採用された Pattern-defeating Quicksort の紹介
po3rin
7
6.2k
Elasticsearchを負荷から守るesguard
po3rin
0
310
Other Decks in Programming
See All in Programming
Чего вы не знали о строках в Python – Василий Рябов, PythoNN
sobolevn
0
150
iOS 17で追加されたSubscriptionStoreView を利用して5分でサブスク実装チャレンジ
natmark
0
500
AIで開発生産性を上げる個人とチームの取り組み
taniigo
0
130
開発生産性を上げるための生成AI活用術
starfish719
1
150
Reduxモダナイズ 〜コードのモダン化を通して、将来のライブラリ移行に備える〜
pvcresin
2
670
CSC509 Lecture 01
javiergs
PRO
1
430
monorepo の Go テストをはやくした〜い!~最小の依存解決への道のり~ / faster-testing-of-monorepos
convto
2
220
(Extension DC 2025) Actor境界を越える技術
teamhimeh
1
210
育てるアーキテクチャ:戦い抜くPythonマイクロサービスの設計と進化戦略
fujidomoe
1
150
Conquering Massive Traffic Spikes in Ruby Applications with Pitchfork
riseshia
0
140
私はどうやって技術力を上げたのか
yusukebe
43
17k
非同期jobをtransaction内で 呼ぶなよ!絶対に呼ぶなよ!
alstrocrack
0
480
Featured
See All Featured
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
48
9.7k
Product Roadmaps are Hard
iamctodd
PRO
54
11k
Responsive Adventures: Dirty Tricks From The Dark Corners of Front-End
smashingmag
252
21k
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
34
6.1k
Optimising Largest Contentful Paint
csswizardry
37
3.4k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
140
34k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
667
120k
Performance Is Good for Brains [We Love Speed 2024]
tammyeverts
12
1.1k
Art, The Web, and Tiny UX
lynnandtonic
303
21k
CSS Pre-Processors: Stylus, Less & Sass
bermonpainter
358
30k
Automating Front-end Workflow
addyosmani
1371
200k
Building an army of robots
kneath
306
46k
Transcript
Compare Benchmarks and Compiler Optimization In Go Go Un Conference
May 27, 2019 @ po3rin
@po3rin Software Enginner @ Shiroyagi Corporation QBHF Golang / Python
/ Rust / Docker / AWS / Elasticsearch
QBHF Introduction \
QBHF 044ʹίϛοτΛࢼΈΔࡍʹɺʮͳͥ ͜ͷ࣮͕ྑ͍͔ʯΛূ໌͢Δҝʹ (Pʹ͓͚ΔϕϯνϚʔΫʹ͍ͭͯௐ ͨͷͰͦΕΛൃද͠·͢ɻ Benchmark of Go Ὂ
QBHF ɾϕϯνϚʔΫΛʮূ໌ʯʹ͑ΔΑ͏ʹͳΔ ɾϕϯνϚʔΫٻ͔ΒίϯύΠϥ࠷దԽΛআ͘ Ὂ the porpose of this talk
QBHF Review how to take a benchmark \
QBHF Review how to take a benchmark Ὂ ֤ϕϯνϚʔΫؔC/ճ෮ ͞Ε·͢ɻσϑΥϧτͰC/
͔Β࢝·Γ·͕͢ɺϕϯνϚʔ Ϋػೳ͕ඵҎʹྃͨ͠߹ C/͕૿Ճͯ͠ϕϯνϚʔΫ͕࠶ ࣮ߦ͞Ε·͢ɻ
QBHF Review how to take a benchmark Ὂ ݁ՌΛݟΔͱͱ͍͏TV⒏Y͕͍͍ͭͯ·͢ɻ͜Ε͜ͷςετΛ࣮ߦ͢ΔͨΊ ʹ༻͞Εͨ(0."9130$4ͷͰ͢ɻ͜ͷσϑΥϧτͰىಈ࣌ʹ(Pϓϩ
ηεʹݟ͑Δ$16ͷʹͳΓ·͢ɻ
QBHF -cpu Ὂ (0."9130$4DQVϑϥάͰมߋͰ͖·͢ɻνʔϜؒͰϕϯνϚʔΫΛ͍ճ ࣌͢$16ͷ͕ϕϯνϚʔΫʹӨڹΛ༩͑ͳ͍Α͏ʹҙ͕ඞཁͰ͢ɻ
QBHF -benchtime Ὂ ෮ճΛ૿͢ҝʹ CFODIUJNFϑϥάΛ༻ ͯ͠ϕϯνϚʔΫ࣌ؒΛ ૿͢͜ͱ͕Ͱ͖·͢ɻ (P͔ΒCFODIUJNF ϑϥά෮ճΛࢦఆ Ͱ͖·͢ɻ
QBHF Benchmark cost avoidance Ὂ C3FTFU5JNFS ͰηοτΞοϓͰ ൃੜ͢ΔίετΛճආͰ͖·͢ɻ ϧʔϓͷ෮͝ͱʹίετ͕ߴ͍ ηοτΞοϓ͕͋Δ߹ɺ
C4UPQ5JNFS ͓Αͼ C4UBSU5JNFS Λ༻͠·͢ɻ
QBHF Check allocations Ὂ ΞϩέʔγϣϯͷͱαΠζɺϕϯνϚʔΫͱڧ͘૬͍ؔͯ͠·͢ɻΞϩέʔ γϣϯͷΛϕϯνϚʔΫͰ֬ೝ͢Δ࣌CFODINFNΛ͍·͢ɻ
QBHF Benchmark stability \
QBHF Benchmark stability Ὂ ඦສ·ͨेԯճ΄Ͳ෮࣮ߦ͞ ΕΔϕϯνϚʔΫ͕OTdNTͷൣғ ͷͳΔ߹ɺϕϯνϚʔΫε έʔϦϯάɺϝϞϦہॴੑͳͲ༷ʑͳ ཁҼʹΑΓෆ҆ఆʹͳ͍ͬͯ·͢ɻ
QBHF Benchmark stability Ὂ ͜ͷΑ͏ͳ߹DPVOUϑ ϥάΛ༻ͯ͠ɺϕϯν ϚʔΫΛෳճ࣮ߦ͢Δ͜ ͱͰϕϯνϚʔΫͷࢄ ؚΊͯ֬ೝ͢Δͷ͕ಘࡦͰ ͢ɻ
QBHF Benchmark stability Ὂ ҰํͰϕϯνϚʔΫͷ҆ఆΛݟΔͷʹศརͳπʔϧ͕͋Γ· ͢ɻ3VTT$PYʹΑΔCFODITUBUͱ͍͏πʔϧΛհ͠·͢ɻ
QBHF Benchmark stability Ὂ CFODITUBUҰ࿈ͷϕϯνϚʔΫ ςετΛ࣮ߦͯ͠ɺͦΕΒ͕ͲΕ ΄Ͳ҆ఆ͍ͯ͠Δ͔Λڭ͑ͯ͘Ε ·͢ɻ
QBHF Comparing benchmarks \
QBHF Comparing benchmarks Ὂ ϕϯνϚʔΫؒͷύϑΥʔϚϯεͷࠩΛஅ͢Δͷ໘Ͱ͕͢ɺ CFODITUBU͜ͷղܾ͠·͢ɻ
QBHF ૣ'JC Λվྑ͍ͨ͠ͷͰ͕͢ɺίʔυΛվྑͨ͠ޙͰɺ ͏ҰվྑલͷϕϯνϚʔΫΛऔΓ͍͕ͨ࣌ग़͖ͯͨΒ Ͳ͏͠·͠ΐ͏͔ɻ࣮HPUFTUʹલճͷϕϯνϚʔΫ݁ ՌΛੜͨ͠όΠφϦΛอଘ͓ͯ͘͜͠ͱ͕Ͱ͖ΔػೳΛ ఏڙ͢ΔDϑϥοά͕ଘࡏ͠·͢ɻվྑલͷόΠφϦ໊ UFTU͔ΒHPMEFOʹมߋ͢Δͷ͕௨ྫͷΑ͏Ͱ͢ɻ -c Ὂ
QBHF ϕϯνϚʔΫΛൺֱ͢Δҝ ʹ࠶ؼݺͼग़͠Λ̍ͭݮΒ ͠·͢ɻ Reduce recursive calls Ὂ
QBHF 'JC Ͱ'JC ͱൺͯͷվྑ͕֬ೝͰ͖·͢ɻιʔείʔυͷมߋ ޙʹͲͷ͘Β͍ͷվળ͕͋ͬͨͷ͔Λূ໌͢Δͷʹ༗༻Ͱ͢ɻࢄ͕େ͖͍ϕϯ νϚʔΫΛൺֱ͢Δͱ͖ҙɻ Comparing benchmarks
Ὂ
QBHF Oɺ༗ޮͩͱݟͳ͞ΕͨσʔλͷݸΛද͠·͢ɻσʔλͷغ٫͕ˋΛ͑ Δͱൺֱ͢Δαϯϓϧ͕গͳ͗͢ΔՄೳੑ͕͋Γ·͢ɻ Q͕Λ͑Δ͜ͱϕϯνϚʔΫ͕౷ܭతʹ༗ҙͰͳ͍͜ͱΛҙຯ͠·͢ɻ Qʹ͍ͭͯԼه͕ৄ͍͠Ͱ͢ɻ ౷ܭֶతݕఆͷ1ɺ౷ܭֶతʹ༗ҙɺ༗ҙࠩɺ༗ҙਫ४ͱԿ͔ʁ IUUQUPVLFJMJOLCBTJDTUBUJTUJDTQWBMVF@BOE@TJHOJpDBODF Comparing benchmarks Ὂ
QBHF Watch out for compiler optimisations \
QBHF ਐͰදͨ͠ͱཱ͖͍ͬͯΔ ϏοτͷΛฦ͠·͢ɻ͜ͷؔ ͷϕϯνϚʔΫΛͱΓ·͠ΐ ͏ɻ compiler optimisations Ὂ
QBHF ͜ͷ݁Ռͷඵ֓ͶΫϩοΫपͰ͢ɻΑͬͯ͜ͷ͔ͳΓ͓͔͍͠Ͱ͢ɻ $16ΫϩοΫ৴߸ʹ߹Θͤͯಈ࡞͠·͢ QPQDOUϦʔϑؔ ଞͷؔݺͼग़͠Λ͠ͳ͍ ʹͳ͍ͬͯ·͢ɻɻίϯύΠϥ͜ ͷؔΛΠϯϥΠϯల։Ͱ͖·͢ɻͦͯ͠QPQDOUɺͲͷάϩʔόϧมͷঢ়ଶʹ Өڹ͠·ͤΜɻ͕ͨͬͯ͠ɺݺͼग़ࣗ͠ମ͕ഉআ͞Ε͍ͯ·͢ɻ compiler
optimisations Ὂ
QBHF ϕϯνϚʔΫΛػೳͤ͞ΔͨΊʹΠϯ ϥΠϯԽΛແޮʹ͢Δ͜ͱଞͷϕϯ νϚʔΫʹӨڹ͕͋ΔͷͰ͓͢͢Ί ͠·ͤΜɻ্̎ͭίϯύΠϥ͕ϧʔ ϓຊମΛ࠷దԽͰ͖ͳ͍Α͏ʹ͢Δͨ Ίͷਪํ๏Ͱ͢ɻ compiler optimisations Ὂ
QBHF JOMJOJOHͳͲͷίϯύΠϥ࠷దԽͷঢ়گΛ֬ೝ͢ΔʹHDqBHTΛ͍·͢ɻ -gcflags Ὂ
QBHF ͞ΒʹڧྗͳJOMJOJOH͕(P͔ΒೖͬͯΔ IUUQTEPDTHPPHMFDPNQSFTFOUBUJPOE8DCMQKQGF,X":'0NK18.@RN/RM2L/B-K1PFEJUTMJEFJEQ Mid Stack inlining Ὂ
QBHF Conclusion \
QBHF (Pʹ͓͚ΔϕϯνϚʔΫ؆୯ʹऔΕΔ͕ɺ͍ํʹҙ͕ඞཁɻ ϕϯνϚʔΫͷڥͷఏࣔ౷Ұ ϕϯνϚʔΫͷ҆ఆੑͷ֬ೝ ͓͔͍͘͠Β͍͍࣌ϕϯνϚʔΫ࣌ͷίϯύΠϥ࠷దԽͷڍಈΛ֬ೝ Conclusion Ὂ
QBHF Additional Talk \
QBHF (P$POGFSFODF'VLVPLB Ͱొஃ͠·͢ʂ
Benchmark and Compiler Optimization In Go Go Un Conference May
27, 2019 @ po3rin