/
2 nd  Data Prefetching  Championship 2 nd  Data Prefetching  Championship

2 nd Data Prefetching Championship - PowerPoint Presentation

danika-pritchard
danika-pritchard . @danika-pritchard
Follow
432 views
Uploaded On 2018-03-21

2 nd Data Prefetching Championship - PPT Presentation

Results and Awards Seth Pugsley Thanks Big thanks to Hyesoon Kim and the Program Committee Babak Falsafi Mike Ferdman Aamer Jaleel Daniel Jiménez Calvin Lin Moin Qureshi ID: 659114

prefetching scores prefetcher place scores prefetching place prefetcher time bandwidth ipc sendag awards3rd works ibrahim burak karsli mustafa cavus

Share:

Link:

Embed:

Download Presentation from below link

Download Presentation The PPT/PDF document "2 nd Data Prefetching Championship" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

Slide1

2nd Data Prefetching ChampionshipResults and Awards

Seth

PugsleySlide2

Thanks!Big thanks to Hyesoon Kim and the Program Committee

Babak

Falsafi

Mike

Ferdman

Aamer

Jaleel

Daniel Jiménez Calvin Lin

Moin

Qureshi

Eric Rotenberg Thomas

Wenisch

Thanks to Alaa Alameldeen and Chris Wilkerson at Intel Labs

Thanks also to submission chair

Hyojong

KimSlide3

DPC2Sim ParametersSingle Core - 3.2 GHz, 6-wide, 256 ROB3 level cache hierarchy

16 KB L1D, 128 KB L2, 1 MB L3

1 channel 64-bit 1600 MT/s DDR3 channel

Prefetching all done at the L2 level

L2 read event is the entry point into the

prefetcher

Prefetches

inserted into the L2 read queue

void

l2_prefetcher_operate(

cpu_num

,

addr

,

PC,

cache_hit

);

MSHR, read queue occupancy, cycle timeSlide4

Championship Scoring4 configurationsDefault (no knobs)

Small LLC

Low bandwidth

Scrambled loads

Score for each configuration

Geomean

((

Prefetcher

IPC) / (No

prefetcher

IPC))

Final score is sum of 4 scores

(20 SPEC CPU 2006 workloads) x (3 traces/workload) x (1 B instructions/trace) x (4 configurations) = 240 B simulated instructionsSlide5

Total ScoresSlide6

Total ScoresSlide7

Default Configuration ScoresSlide8

Small LLC ScoresSlide9

Low Bandwidth ScoresSlide10

Scrambled Loads ScoresSlide11

Effect of Scrambling Loads:IPC(scrambled)/IPC(default)Slide12

Idealized Total Score:Max(All Prefetchers)Slide13

Accepted Workshop PrefetchersScoresSlide14

AwardsSlide15

Awards3rd place -

Prefetching On-time and When It

Works

Ibrahim Burak Karsli, Mustafa Cavus,

and Resit SendagSlide16

Awards3rd place - Prefetching On-time and When It Works

Ibrahim Burak Karsli, Mustafa Cavus,

and Resit Sendag

2nd place -

Towards Bandwidth-Efficient Prefetching with Slim

AMPM

Vinson

Young and Ajit

KrisshnaSlide17

Awards3rd place - Prefetching On-time and When It Works

Ibrahim Burak Karsli, Mustafa Cavus,

and Resit Sendag

2nd place -

Towards Bandwidth-Efficient Prefetching with Slim

AMPM

Vinson

Young and Ajit Krisshna

1st place

- A Best-Offset

Prefetcher

Pierre MichaudSlide18

Thanks for participating!