Scribd
Upload
73 views14 pages

Coolpression: A Hybrid Significance Compression Technique For Reducing Energy in Caches

1) CoolPression is a hybrid cache compression technique that uses both bit-level and byte-level compression schemes to reduce energy usage in caches. It can save over 35% more energy than applying each scheme independently. 2) The technique counts leading zeroes and ones in byte values to compress data. It selectively reads and writes only significant bits to reduce switching activity. 3) Evaluation shows the technique reduces energy usage by 30-45% for a 16KB data cache and 5-10% for a 16KB instruction cache with little performance overhead.

Uploaded by

larryshi
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
73 views14 pages

Coolpression: A Hybrid Significance Compression Technique For Reducing Energy in Caches

1) CoolPression is a hybrid cache compression technique that uses both bit-level and byte-level compression schemes to reduce energy usage in caches. It can save over 35% more energy than applying each scheme independently. 2) The technique counts leading zeroes and ones in byte values to compress data. It selectively reads and writes only significant bits to reduce switching activity. 3) Evaluation shows the technique reduces energy usage by 30-45% for a 16KB data cache and 5-10% for a 16KB instruction cache with little performance overhead.

Uploaded by

larryshi
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

CoolPression: A Hybrid Significance Compression Technique for Reducing Energy in Caches

Mrinmoy Ghosh Dr Hsien Hsin Lee Weidong Shi


School of Electrical and Computer Engineering

Georgia Institute of Technology


Mrinmoy Ghosh 1 September 15, 2004

Hot Caches
Mem. Controller 19% Bus Interface Unit 12% Data Cache 14% Instructio n Cache 7% Data Path 32% Integer Units 16%

Alpha 21264

D Cache 19%

I Cache 25%

Other 4% Clocks 4% SysCtl 3% BIU PATag 8% CP15 RAM 2% 1%

D MMU 5% I MMU 4%

ARM 920T

Intel Pentium 4 (Willamette)

ARM 9 25%

Mrinmoy Ghosh

Motivation
1000000000 100000000 10000000 1000000 100000 10000 1000 100 10 1 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 Max Min Avg

Mrinmoy Ghosh

Salient Features
Reuse most significant byte Counting granularity in bits rather than bytes Compression scheme is a hybrid of two schemes, where the better scheme is chosen dynamically Each scheme gives power savings of around 30-45% if applied independently, and the hybrid scheme saves around 3-5% more than either of the schemes.
Mrinmoy Ghosh
4

CoolPression Cache
CE Bit

ZIBs
CE Bit 66bits bits 36 bits

count bits count bits

SRAM Cell Array


32- count Sense Amps 37 bits

SRAM Cell Array Sense Amps


37 Data from Cache

CoolCount Circuit
Data Out 32

33

Step 2a: Read only 32 count bits and append with leading zeroes or ones 33 33 CE Bit 36 bits

Circuit CoolCount Circuit CoolCount Data Out 32

Bitline Enable Lines Step 1: Read In First 7 bits and the ZIBs

SRAM Cell Array Sense Amps


37 Data from Cache

CoolCount Circuit
Data Out Mrinmoy Ghosh 32

33

Step 2b: Read Bytes that are not zeroes

Counting Leading Zeroes And Ones


7 6 5 4 3 2 1 0

Priority Encoder

2
Mrinmoy Ghosh

0
6

No of Leading Zeroes or Ones -1

Counting Leading Zeroes And Ones 0


7 6 5 4 3 2 1

Priority Encoder

No of Leading Zeroes or Ones -1


Mrinmoy Ghosh
7

Bitline Precharge Enabling Circuit

Mrinmoy Ghosh

Read Data From Cache


Read in Count Enable (CE) Bit and First 6 bits of data

CE ==1

Yes

Enable Least Significant ~count bit lines

No Read Data for bytes where ZIB is not enabled and make the other bytes zero Mrinmoy Ghosh Read Data From Least Significant ~count bit lines and append with count leading zeroes or ones

Write Data To Cache


Count Number of Leading Zeroes or Ones Check for Bytes which are zero

Count >8

Yes

Set CE bit to one and Enable Most Significant 6 bits lines and Least Significant ~count bit lines

No Set CE bit to 0 and Write Data to Cache setting ZIBs where necessary

Write Encoded Data to Cache

Mrinmoy Ghosh

10

Effect on Performance
Normal Cache
2.5 2 1.5 IP C 1 0.5 0 Crafty Gcc Gzip Mcf Parser Twolf Vortex VPR Avg

Coolcount Cache

Mrinmoy Ghosh

11

Results
16K Data Cache
1 .2 1 0.8 0.6

Dcache Base Dcache DZC

Dcache CoolCount Dcache CoolPression

0.4 0.2 0 Bzip 2 Craf t y GCC GZIP M CF Parser V ort ex Vpr A vg

16K Instruction Cache


Icache Base
1 .05 1 0.95 0.9 0.85 0.8 0.75 Bzip2 Craft y GCC GZIP M CF Parser Vortex Vpr Avg

Icache CoolCount

Icache DZC

Icache CoolPression

Mrinmoy Ghosh

12

Conclusions
System Transparent Hybrid Zero Compression Scheme Bit level and Byte level compressibility used to save power Energy Savings of over 35% over baseline cache Potential Use at other places where data transfer takes place like L2 Cache to Memory
Mrinmoy Ghosh
13

Thank You

Mrinmoy Ghosh

14

You might also like