超並列高速実行処理系 Hastega (ヘイスガ) ～ Lonestar ElixirConf 凱旋帰国

Erlang & Elixir Fest 2019
山崎 進

Qiitaで連載され，論文も複数本数発表され，さらには Lonestar ElixirConf でセンセーショナルな世界デビューを果たした超並列高速実行処理系 Hastega (ヘイスガ) の凱旋帰国講演です！

しかもおそらく本イベントのある6月までには，ファーストリリースバージョンを公開できるか，秒読みに入るのではないかと予測しています。

Hastega は，可読性に優れる Enum.map とパイプライン演算子を組み合わせた有限長のリスト処理を，SIMD命令やマルチコア，GPUを駆動するネイティブコードに変換して高速実行する処理系です。プロトタイプ実装では，純粋なElixirコードの4〜8倍にも及ぶ実行速度向上を達成しています。

本発表では，瞬時にわかる Hastega プログラミングのやり方を紹介し，動作原理や内部構造を紹介し，将来展望を示します。厨二病の血が騒ぐ発表に仕上げていきたいと思います。

Transcript

1. ௒ฒྻߴ଎࣮ߦॲཧܥ
   )BTUFHB
   ϔΠεΨ
   ʙ
   -POFTUBS
   &MJYJS$POG
   ֌ટؼࠃ ࢁ࡚
   ਐ 1 ˜
   
   4VTVNV
   :BNB[BLJ
2. ࣗݾ঺հ w 5XJUUFS
   
   'BDFCPPL
   
   (JU)VC
   
   (.BJM
   
 ͢΂ͯ
   [BDLZ
   
   w ๺۝भࢢཱେֶ
   ࠃࡍ؀ڥ޻ֶ෦
   ।ڭत
   w &MJYJS
   ਪ͠
   
   GVLVPLBFY
   
   )BTUFHB
   w φοδࣾձ࣮૷ݚڀηϯλʔ
   ηϯλʔ௕
   w ֶੜɾब৬ࢧԉ୲౰
   w 1FSTPOBM
   7JTJPO
   $P$SFBUPS
   w ,,4)J'5
   w ٕज़૬ஊɼڞಉݚڀґཔɼਐ࿏૬ஊɼ
 ద৬਍அͳͲɼਵ࣌ड෇·͢
    ˜
   
   4VTVNV
   :BNB[BLJ
3. )BTUFHB
   ϔΠεΨ 
 ͬͯԿʁ 3 ˜
   
   4VTVNV
   :BNB[BLJ
4. )BTUFHB
   ϔΠεΨ w ΢ΣϒΞϓϦέʔγϣϯͰ༻͍Δ
 ը૾ॲཧɾػցֶश౳Λ
   w &MJYJSͰ؆ܿʹهड़ͯ͠
   w αʔόʔɼ΢ΣϒΫϥΠΞϯτɼ
 Τοδ্ͷ$16(16ʹෛՙ෼ࢄ
 ͤ͞ͳ͕Βߴ଎ʹฒྻ࣮ߦͤ͞Δ
 ͜ͱΛ໨ࢦ͢ॲཧܥ
   w ໊শ͸''༝དྷ
 ஥ؒΛߴ଎Խͤ͞Δຐ๏
   w ৄࡉ͸ޙड़
    ˜
   
   4VTVNV
   :BNB[BLJ
5. -POFTUBS
   &MJYJS$POG
   ͬͯԿʁ 5 ˜
   
   4VTVNV
   :BNB[BLJ
6. -POFTUBS
   &MJYJS$POG w ౙʹςΩαεभΦʔεςΟϯͰ։࠵ ͞ΕΔ
   &MJYJS
   ͷΧϯϑΝϨϯε
   w -POFTUBS
   ͸ςΩαεभͷभضͷ
 γϯϘϧ
   w ೥͸݄೔ʙ݄೔ʹ։࠵
 IUUQTXXXMPOFTUBSFMJYJSDPOGDPN
   w &MJYJS
   ઐ໳ͷΧϯϑΝϨϯεͰ͋Δ &MJYJS$POG
   64ɼ&MJYJS$POG
   &6
   ʹ ͙࣍ୈҐͷن໛
    ˜
   
   4VTVNV
   :BNB[BLJ
7. -POFTUBS
   &MJYJS$POG
   ʹԿ͠ʹߦͬͨͷʁ 7 ˜
   
   4VTVNV
   :BNB[BLJ
8. )BTUFHB
   ւ֎σϏϡʔ w -POFTUBS
   &MJYJS$POG
   
   Ͱ )BTUFHB
   Λൃද͠·ͨ͠ʂ
    ˜
   
   4VTVNV
   :BNB[BLJ IUUQTZPVUVCFMZQRM(M,4P IUUQTXXXMPOFTUBSFMJYJSDPOGDPNTQFBLFSTTVTVNVZBNB[BLJ
9. ͲΜͳൃදͨ͠ͷʁ 9 ˜
   
   4VTVNV
   :BNB[BLJ
10. Hastega: Challenge for GPGPU on Elixir Susumu Yamazaki (ZACKY) Associate Professor at Univ. of Kitakyushu Adviser at fukuoka.ex
11. We’re three Samurais from fukuoka.ex, like “Les Trois Mousquetaires” (The Three Musketeers)
12. We’ll show you
 our spirits of
 Samurai and Zen…
13. Long, long distance… The world with
 destructive updates is…
14. Dystopia! © 2014, Teresa Prater
15. © 2011, Pavel Medzyun Their works become
 dystopia!
16. Our solution:
 Hastega!!!
17. Hastega • It’s a magic!! • the highest evolved magic to accelerate our party in Final Fantasy!
 (Stronger than Haste) • It will be the highest evolved technology to accelerate our machines in the Elixir ecosystem! It’s inspired by Flow
18. Let’s cast a spell of Hastega on Samurai
19. 4-8x faster than Elixir using Flow! and also faster than P-map
 (Parallel map) ↑ Slower ↓ Faster Flow
20. 3x+ faster than Python with GPU (CuPy) !!! ↑ Slower ↓ Faster
21. Only a little slower than native code (Rust)!! Dead heat!! ↑ Slower ↓ Faster
22. Overwhelming Effectiveness like a berserk Samurai
23. By the way,
 which is better, 
 recursive call and Enum.map?
24. Enum.map is simpler! It’s like Zen! ˜
    4UFQIBOF
    %"MV
25. Why is Enum.map Zen? ˜
    4UFQIBOF
    %"MV • Zen is the essential beauty • The essential of programming is data transformation • Enum.map describes only data transformation list = 1..1_000_000 |> Enum.to_list list 
 |> Enum.map(&foo(&1))
 |> Enum.map(&bar(&1))
26. 1..1_000_000 
 |> Enum.map(&foo(&1))
 |> Enum.map(&bar(&1)) • We propose to call it the Elixir Zen style to write in Enum.map • It is a good programming custom • Because it’s more readable and maintainable The Elixir Zen style
27. Then, what will happen? ˜
    4UFQIBOF
    %"MV
28. In Elixir on Erlang VM, the Elixir Zen style is 20 percents slower than recursive call list = 1..1_000_000 |> Enum.to_list list 
 |> Enum.map(&foo(&1))
 |> Enum.map(&bar(&1)) list
 |> func()
 
 def func( [] ), do: [] def func( [ head | tail ] ) do
 [ head |> foo |> bar 
 | func(tail) ]
 end 6msec ↑ Slower ↓ Faster Performance Evaluation
29. © 2011, Pavel Medzyun It’s Zen, but not Samurai…! It will cause dystopia!
30. • make the Elixir Zen styled code faster • by casting the spell of it on Samurai to be berserk • that is, to be transformed into the fastest native code, 
 using all computing resources, • not only multi-core CPUs (with SIMD instructions)
 but also GPUs Hastaga will… We feel it Wabi-Sabi
31. Hastega makes the most beautiful code to be transformed the fastest!
32. Do you wanna get power of Hastega, now?
33. I’m sorry, but the 1st practical use version of Hastega will be released before Summer, 2019 m(_ _)m
34. ͳΜͰ
    )BTUFHB
    Λ
 ։ൃͨ͠ͷʁ 34 ˜
    
    4VTVNV
    :BNB[BLJ
35. ৘ใྔരൃ ೥ʹσʔλͷ૯ྔ͕θλόΠτ
    ೥ʹθλόΠτʹୡ͢ΔݟࠐΈ 35 ˜
    
    4VTVNV
    :BNB[BLJ
36. ৘ใྔരൃ ೥ʹຖ݄ΤΫαόΠτ૿Ճ
    ೥ʹຖ݄ΤΫαόΠτ૿ՃݟࠐΈ
    ৘ใͷ૿Ճྔ͕Ճ଎ 36 ˜
    
    4VTVNV
    :BNB[BLJ
37. ܭࢉೳྗ͕௥͍͔ͭͳ͍ )FOOFTZ
    
    1BUUFSTPO
 ೥ʹ$16ΫϩοΫप೾਺͸಄ଧͪʹ 37 ˜
    
    4VTVNV
    :BNB[BLJ
38. ͳͥΫϩοΫप೾਺Λ৳͹ͤͳ͍͔ʁ w ΫϩοΫप೾਺ͷ૿େ͸ফඅిྗͱൃ೤ྔͷ૿େΛট͘
    w ిݯڙڅྔͱ೤఻ಋ཰͓Αͼྫྷ٫ೳྗ͕௥͍͔ͭͳ͍
    w ʙ()[ఔ౓ͰৗԹͰ҆ఆಈ࡞͢ΔݶքΛܴ͍͑ͯΔ
    w ڝٕ༻ΦʔόʔΫϩοΫͰ͸ӷମ஠ૉ౳ʹਁ͚Δ౳ʹ ΑΓ()[Ҏ্Ͱಈ࡞͍ͤͯ͞Δ
    w ௨ৗͷ1$΍αʔόʔͰ͸ৗԹՔಇͰ͖Δ͜ͱ͕ཁ݅ 38 ˜
    
    4VTVNV
    :BNB[BLJ
39. ͔ΘΓʹίΞ਺͕૿Ճ w ೥Ҏ߱ͷϓϩηοαͷਐԽ
    w ºΫϩοΫप೾਺
    w ˓ίΞ਺
     ˜
    
    4VTVNV
    :BNB[BLJ *OUFM
    $PSF
    
 &YUSFNF
    9  ΫϩοΫ ()[ ίΞ਺  *OUFM
    $PSF
    J
 9&  ΫϩοΫ ()[ ίΞ਺ 
40. )BTUFHBͬͯ
 ͲΜͳͷʁ 40 ˜
    
    4VTVNV
    :BNB[BLJ
41. )BTUFHB
    ͷݪཧ
     ˜
    
    4VTVNV
    :BNB[BLJ @@
    c
    &OVNNBQ .GPP  
 c
    &OVNNBQ .CBS  
 
 EFGNPEVMF
    .
    EP
    
    
    EFG
    GPP
    O
    
    EP
    O
    
    
    
    
    EFG
    CBS
    O
    
    EP
    O
    
    
    FOE &MJYJS
    ;FO
    4UZMF
 &OVNNBQ
    ελΠϧ @@LFSOFM
    WPJE
    DBMD
    
    
    @@HMPCBM
    MPOH
    JOQVU
    
    
    @@HMPCBM
    MPOH
    PVUQVU
    \
    
    
    TJ[F@U
    J
    
    HFU@HMPCBM@JE  
    
    
    MPOH
    UFNQ
    
    JOQVU<J>
    
    
    UFNQ
    
    GPP UFNQ 
    
    
    UFNQ
    
    CBS UFNQ 
    
    
    PVUQVU<J>
    
    UFNQ
    ^ 0QFO$-
 Χʔωϧ෦ w ࠨͷίʔυ͸  ฒྻͷՄೳੑ͕͋Δ
    w ࠨͷίʔυ͔Βӈͷίʔυ΁ػցతʹม׵Ͱ͖Δ
    w ࠷৽(16ͷ Ҏ্ͷ4*.%ίΞͰฒྻ࣮ߦՄೳ
42. )BTUFHB
    ͷίʔυྫ
     ˜
    
    4VTVNV
    :BNB[BLJ EFGNPEVMF
    )
    EP
    
    
    SFRVJSF
    )BTUFHB
    
    
    JNQPSU
    )BTUFHB
    
    
    EFGIBTUFHB
    EP
    
    
    
    
    EFG
    GVOD MJTU
    EP
    
    
    
    
    
    
    
    
    MJTU
    
    
    
    
    
    
    
    
    c
    &OVNNBQ 
    
    
     

    
    
    
    
    
    
    
    c
    &OVNNBQ 
    
    
    
    
    
    
    
    FOE
    
    
    
    
    IBTUFHBTUVC
    
    
    FOE
    FOE w SFRVJSF
    )BTUFHB
    Ͱ
    )BTUFHB
    ΛಡࠐΈ
    w JNQPSU
    )BTUFHB
    ͰϞδϡʔϧ໊Λ෇ه
 ͤͣʹ
    EFGIBTUFHB
    ΍
    IBTUFHBTUVC
    ͕
 ར༻Ͱ͖Δ
    w EFGIBTUFHB
    EP
    ʜ
    FOE
    Ͱ
    ʜ
    ͷίʔυΛ
 ॲཧܥʹಡࠐΉ
    w IBTUFHBTUVC
    Ͱίʔυੜ੒Λߦ͏
43. -POFTUBS
    &MJYJS$POG
    ͷΈ΍͛࿩͸ʁ 43 ˜
    
    4VTVNV
    :BNB[BLJ
44. /FSWFT w &MJYJS
    Λ
    *P5
    
    ૊ࠐΈγεςϜʹ ಋೖ͢ΔιϦϡʔγϣϯ
    w -POFTUBS
    &MJYJS$POG
    
    Ͱ͸
    &MJYJS
    
    1IPFOJY
    ͱର౳ͳଘࡏͱ͠ ͯ঺հ͞Ε͍ͯͨ
    w ,FZOPUF
    ຊ
 1SFTFOUBUJPO
    ຊ
 5SBJOJOH
    ຊ
 -5
    ຊ
    w ৄ͘͠͸ߴ੉ઌੜͷൃදΛָ͠Έ ʹ͍ͯͩ͘͠͞ʂ
     ˜
    
    4VTVNV
    :BNB[BLJ IUUQTZPVUVCF*U*WOEU['.
45. 1IPFOJY
    -JWF7JFX w 1IPFOJY
    ͷ࡞ऀ
    $ISJT
    ͷൃද
 -JWF7JFX
    
    w ΋͏ϦϦʔε͞ΕͨͷͰɼ
 ࢼͨ͠ਓ΋ଟ͍ͷͰ͸ʁ
    w /FSWFT
    ͱ
    -JWF7JFX
    Λ઀ଓ ͨ͠σϞ΋͋Γ
     ˜
    
    4VTVNV
    :BNB[BLJ
46. -VODIJTPEF w ೔໨ͷϥϯνλΠϜʹ
 ߨԋऀઐ༻ͷ෦԰Ͱٞ࿦ͨ͠
 ༷ࢠΛऩ࿥
    w &MJYJS
    
    1IPFOJY
    
    /FSWFT
    ͷ
 ະདྷʹ͍ͭͯޠΓ߹͏
    w )BTUFHB
    ʹ͍ͭͯ΋Ұੜݒ໋ ͠Ό΂Γ·ͨ͠
    w ࠓɼฉ͍ͯ΋ɼ൒෼͘Β͍͠ ͔ฉ͖औΕͳ͍ɻɻɻ
     ˜
    
    4VTVNV
    :BNB[BLJ IUUQTQPEDBTUTNBSUMPHJDJPTQFDJBMMPOFTUBSFMJYJS
47. &MJYJSOFYU
    5IF
    3JWFS w ٕज़ίϛϡχςΟͷਐาͷྺ࢙Λ
 େՏͷྲྀΕʹͨͱ͑Δ
    w ্ྲྀ͔ΒԼྲྀʹྲྀΕΔʹͭΕͯ
 ίϛϡχςΟͷ͋Γํ͕ม࣭͢Δ
    w ίϛϡχςΟʹͲ͏ؔΘΔ΂͖͔
    w ίϛϡχςΟ΁ฦ͢ɼߩݙ͢Δ
    w ʮ7PUF
    ౤ථͤΑʯ
    w ʮ௨աّྱʯతηογϣϯ
    w ࠷΋ݟΔ΂͖ϓϨθϯςʔγϣϯ
     ˜
    
    4VTVNV
    :BNB[BLJ IUUQTZPVUVCF,8S81H*
48. &SMBOH
    &DPTZTUFN
    'PVOEBUJPO w &MJYJS
    ͷ࡞ऀ
    +PTÉ
    7BMJN
    ͷ
 ,FZOPUF
    ຊ໨
    ຊ໨΋ݟͯͶ
    w &MJYJSOFYU
    ͔ΒͷྲྀΕΛड͚
 &MJYJS
    ͷ฼ମͰ͋Δ
    &SMBOH
    ͱ
 Ͳ͏͍͏ؔ܎ੑΛ΋͔ͭʁ
 ʹ͍ͭͯ
    +PTÉ
    ͔ΒޠΒΕͨ
    w &SMBOH
    &DPTZTUFN
    'PVOEBUJPO
    ʹՃೖͯ͠ʮ7PUF
    ౤ථʯͯ͠
 ΄͍͠ͱݺͼ͔͚ͨ
    w &SMBOH
    ΁ͷϦεϖΫτΛ෷͍ɼ
 &MJYJS
    ͔Β୅දΛग़ͦ͏ͱ͍͏ಈ͖
     ˜
    
    4VTVNV
    :BNB[BLJ IUUQTZPVUVCFR$"2V51P.
49. +PTÉͱͷަྲྀ w )BTUFHB
    ߨԋޙɼ+PTÉ͕೤৺ ʹ࣭໰͖ͯͨ͠
    w +PTÉͷߨԋޙɼࢲ΋ٞ࿦͢Δ
    w )BTUFHB
    ͕
    +PTÉ
    ͷߏ૝͢Δ ϩʔυϚοϓ্ͷཁૉٕज़
 Ͱ͋Δ͜ͱΛ֬ೝ
    w ؼࠃޙϝʔϧͰจ௨͢Δ஥ʹ
     ˜
    
    4VTVNV
    :BNB[BLJ
50. +PTÉ͕໌Β͔ʹͨ͠
 ڻዼͷࣄ࣮ 50 ˜
    
    4VTVNV
    :BNB[BLJ
51. ܦҢ w +PTÉ͸
    )BTUFHB
    ʹͱͯ΋ڵຯΛ࣋ͬͯ͘Ε
    w ͳΜͱɼࢲͷॻ͍ͨ೔ຊޠͷ࿦จΛࣗ෼Ͱ຋༁ͯ͠ಡΜͰ͘Εͨ
    w ͦͷͱ͖ʹ͜Μͳϝʔϧ͕ૹΒΕ͖ͯͨ 51 ˜
    
    4VTVNV
    :BNB[BLJ
52. ˜
    
    4VTVNV
    :BNB[BLJ
53. ˜
    
    4VTVNV
    :BNB[BLJ
54. ˜
    
    4VTVNV
    :BNB[BLJ
55. ˜
    
    4VTVNV
    :BNB[BLJ
56. ʜͰɼ)BTUFHB
    ͷ
 ϦϦʔε͸ɼ·ͩͳͷʁ 56 ˜
    
    4VTVNV
    :BNB[BLJ
57. ·ͩͰ͢ʂ
 +PTÉ
    ͷॿݴͱֶੜͷ
    KPJO
    Ͱ
 ΑΓύϫʔΞοϓͯ͠ઈࢍ։ൃதʂ 57 ˜
    
    4VTVNV
    :BNB[BLJ
58. Our adventure of Samurai, Zen and Wabi-Sabi continues…
59. I’ll be back! to share new research results
60. THE END
61. 0OF
    NPSF
    UIJOH 61 ˜
    
    4VTVNV
    :BNB[BLJ
62. &MJYJS$POG
    64
    
    ݄೔ʙ೔ w )BTUFHB
    ͷൃදΛ͠·͢ʂ
    w ৽ͨʹݚڀ։ൃͨ͠৽ٕज़Λ ൃද͠·͢ʂ
    w ϓϩάϥϜ͸੝Γͩ͘͞Μ
    w ΈΜͳͰߦ͜͏Αʂ
    w ੈքͷைྲྀΛײ͡Α͏ʂ
     ˜
    
    4VTVNV
    :BNB[BLJ IUUQTFMJYJSDPOGDPNTQFBLFST
63. ݚڀʹ͝ڠྗ͍ͩ͘͞ʂ w ͞ΒͳΔίʔυ࠷దԽͷݚڀʹඞཁͳ࣮ߦ࣌৘ใΛऩू͢ΔͨΊͷϓϩϑΝΠϥ Λ೥౓தʹ։ൃ͢Δ༧ఆͰ͢
    w ऩू͢Δ৘ใͷྫ
 $16ͷछྨ
    
    ϝϞϦɾΩϟογϡϝϞϦ༰ྔ
    
    ϦεταΠζɾཁૉͷܕ
    
 ϓϩηε਺
    
    ΠςϨʔγϣϯதίʔυྔ
    
    &MJYJS
    ;FO
    ελΠϧίʔυͷׂ߹ͳͲ
    w Ͱ͖Ε͹࣮ӡ༻ʹ͍ۙ৘ใΛऩू͍ͨ͠ͱߟ͍͑ͯ·͢
    w ࢲୡͷνʔϜ͚ͩͰσʔλΛಠ઎͢ΔΘ͚Ͱ͸ͳ͘ɼ౷ܭॲཧ্ͨ͠Ͱɼ
 اۀ໊ɾݸਓ໊౳Λ໌͔ͣ͞ʹɼݚڀऀ޲͚ʹσʔλఏڙ͍ͨ͠ͱߟ͍͑ͯ·͢
    w ͝ڠྗ͍͚ͨͩΔاۀɾݸਓͷํ͸ɼ͝࿈བྷ͍ͩ͘͞ʂ
 ࢁ࡚
    ਐ
    [BDLZ!LJUBLZVVBDKQ
    63 ˜
    
    4VTVNV
    :BNB[BLJ
64. ݄೔ͷ༧ఆΛۭ͚ͯͶʂ
    ࠃ಺ͰͷΠϕϯτاըத 64 ˜
    
    4VTVNV
    :BNB[BLJ