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
ぼくのかんがえたさいきょうCPU 2018 ベクトル演算
Search
houmei
April 24, 2018
Technology
0
170
ぼくのかんがえたさいきょうCPU 2018 ベクトル演算
houmei
April 24, 2018
Tweet
Share
More Decks by houmei
See All by houmei
ぼくのかんがえたさいきょうCPU 2018 スカラ演算
houmei
0
150
ぼくのかんがえたさいきょうCPU 2018 DATA
houmei
0
150
2018-BKSC-ALU
houmei
0
290
2017 CPU Architeciture
houmei
0
310
Other Decks in Technology
See All in Technology
Claude Code Subagents 再入門 ~cc-sddの実装で学んだこと~
gotalab555
6
8.6k
JAZUG 15周年記念 × JAT「AI Agent開発者必見:"今"のOracle技術で拡張するAzure × OCIの共存アーキテクチャ」
shisyu_gaku
1
160
Performance Insights 廃止から Database Insights 利用へ/transition-from-performance-insights-to-database-insights
emiki
0
280
フレームワークを意識させないワークショップづくり
keigosuda
0
190
20251007: What happens when multi-agent systems become larger? (CyberAgent, Inc)
ornew
1
290
All About Sansan – for New Global Engineers
sansan33
PRO
1
1.2k
20251010_HCCJP_AdaptiveCloudUpdates
sdosamut
0
130
ユーザーの声とAI検証で進める、プロダクトディスカバリー
sansantech
PRO
1
140
React19.2のuseEffectEventを追う
maguroalternative
0
250
WEBサービスを成り立たせるAWSサービス
takano0131
1
160
Sansan Engineering Unit 紹介資料
sansan33
PRO
1
3k
[Keynote] What do you need to know about DevEx in 2025
salaboy
0
170
Featured
See All Featured
How to Ace a Technical Interview
jacobian
280
24k
Learning to Love Humans: Emotional Interface Design
aarron
274
41k
Six Lessons from altMBA
skipperchong
29
4k
Why You Should Never Use an ORM
jnunemaker
PRO
59
9.6k
The Language of Interfaces
destraynor
162
25k
Product Roadmaps are Hard
iamctodd
PRO
54
11k
Building Better People: How to give real-time feedback that sticks.
wjessup
369
20k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
31
2.7k
Building an army of robots
kneath
306
46k
jQuery: Nuts, Bolts and Bling
dougneiner
65
7.9k
It's Worth the Effort
3n
187
28k
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
162
15k
Transcript
΅͘ͷ͔Μ͕͑ͨ ͍͖͞ΐ͏CPU 2018.4.24 @houmei ϕΫτϧԋࢉ 184݄24Ր༵
༰ • ΅͘ͷ͔Μ͕͍͖͑ͨ͞ΐ͏CPUͷղઆ • Ϩʔϯؒͷԋࢉʹ͍ͭͯʢϕΫτϧʣ twitter : @houmei blog :
ԼੈքౝͷܭࢉػΑ· 184݄24Ր༵
ϕΫτϧͱεΧϥ ɾσʔλશମΛ2ⁿόΠτ୯ҐͰׂͨ͠ ۠ըΛϨʔϯͱݺͿ ɾϨʔϯͯ͢ಉ͡ܕɺಉ͡αΠζ ɾϨʔϯ͕ͻͱ͚ͭͩͷͷΛεΧϥɺ ෳ͋ΔͷΛϕΫτϧͱݺͿ FP32 FP32 FP32 FP32
lane 0 lane 1 lane 2 lane 3 0 31 63 95 127 184݄24Ր༵
ݪଇ (1)ԋࢉ݁ՌRdͷσʔλαΠζɺϨʔϯɺܕ Ұ൪ͷιʔε(Ra)ʹ߹ΘͤΔ (2)ೋ൪ͷιʔε(Rb)ͷϨʔϯ͕RaͷϨʔϯ ΑΓগͳ͍߹܁Γฦ͠ద༻ (3)ೋ൪ͷιʔε(Rb)ͷϨʔϯ͕RaͷϨʔϯ ΑΓଟ͍߹ԼҐͷϨʔϯΛద༻ (4)Ϩʔϯؒͷԋࢉʹ͍ͭͯεΧϥԋࢉͷϧʔ ϧΛద༻ 184݄24Ր༵
ϕΫτϧ×ϕΫτϧʢ̍ʣ ɾRaͷϨʔϯ͕RbͷϨʔϯͱಉ͡߹ →ରԠ͢ΔϨʔϯͲ͏͠Ͱԋࢉ Lane3 D d D+d Lane2 C c
C+c Lane1 B ʴ b = B+b Lane0 A a A+a Ra Rb Rd 184݄24Ր༵
ϕΫτϧ×ϕΫτϧʢ̎ʣ ɾRaͷϨʔϯ͕RbΑΓগͳ͍߹ →ରԠ͢ΔRbͷԼҐϨʔϯͲ͏͠Ͱԋࢉ Lane3 d Lane2 c Lane1 B ʴ
b = B+b Lane0 A a A+a Ra Rb Rd 184݄24Ր༵
ϕΫτϧ×ϕΫτϧʢ̏ʣ ɾRaͷϨʔϯ͕RbΑΓଟ͍߹ →RbͷϨʔϯΛ܁Γฦ͠ద༻͠ԋࢉ Lane3 D D+b Lane2 C C+a Lane1
B ʴ b = B+b Lane0 A a A+a Ra Rb Rd 184݄24Ր༵
ϕΫτϧ×εΧϥ ɾRaͷϨʔϯ͕RbΑΓଟ͍߹ͱಉ͡ →RbΛ܁Γฦ͠ద༻͠ԋࢉ Lane3 D D+a Lane2 C C+a Lane1
B B+a Lane0 A ʴ a = A+a Ra Rb Rd 184݄24Ր༵
εΧϥ×ϕΫτϧ ɾRaͷϨʔϯ͕RbΑΓগͳ͍߹ͱಉ͡ →Rbͷ࠷ԼҐϨʔϯͰԋࢉ Lane3 d Lane2 c Lane1 b Lane0
A ʴ a ʹ A+a Ra Rb Rd 184݄24Ր༵
ଈͷѻ͍ • ԋࢉͷୈೋΦϖϥϯυଈͷࢦఆ͕Մೳ • ଈͷεΧϥσʔλͱͯ͠ѻΘΕΔ • ଈͷαΠζRaͷϨʔϯͷαΠζʹ߹Θͤූ ߸֦ு͞ΕΔ(.uम০ࢠͰθϩ֦ு) 184݄24Ր༵
ɹ ͭͮ͘ ΅͘ͷ͔Μ͕͍͖͑ͨ͞ΐ͏CPU 184݄24Ր༵