Conference on Adaptive Hardware and Systems (AHS'14) - FlexTiles FPGA Emulation

www.flextiles.eu

FlexTiles

18.0

7.20

14/A

HS

Benedikt JanßenRuhr-University Bochum (RUB)

FlexTiles Workshop at AHS 2014

FPGA-Based Emulation of the FlexTiles Platform

2 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Global View

GPP Node

AI

DSP FPGA Fabric

NI

GPP Node

NI

Network-on-Chip (NoC)

NI NI NI

AI AI

NI

Reconfig. Control

DDR Node

NI

Tile TileGPP Node

NI

I/O

NI

HW Acc. HW Acc.

Tile

3 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Development Platform

Software Emulator

Based on Open Virtual Platform

Enable Software prototyping

Hardware Emulator

Sundance SMT166 Based on 2 Virtex-6 FPGAs

24-pin ATX connector

Based on the TU/e platform

Enable Hw/Sw prototyping

4 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Development Platform

System Components

GPP Node MicroBlaze soft-core processors

Network on Chip Nodes connected via Network Interfaces

Network Interface Device Transport Layer (DTL) protocol

Accelerator Interface Newly developed within this project

Accelerators Micro-programmed

Data-flow

GPP Node

AI

Accelerator

NI

NoC

NI

5 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Network on Chip

Network on Chip

AElite NoC

Network Interface

Main Task Data into packets NoC

NoC data out of packets

Device Transport Layer Master / Slave

Data flow Load / Store

Stream

GPP Node

AI

Accelerator

NI

NoC

NI

6 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Accelerator Interface

Main Tasks

Control accelerators

Provide a unique interface

Components

Accelerator control

Protocol parser

GPP Node

AI

Accelerator

NI

NoC

NI

AC

7 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

AI : DTL-To-AI

Device Transaction Layer (DTL)

Command (dtl_cmd)

Write (dtl_wr)

Read (dtl_rd)

DTL-To-AI Structure

GPP Node

AI

Accelerator

NI

NoC

NI

DTL2AIAC

8 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

AI : Accelerator Control

Accelerator Control Structure GPP Node

AI

Accelerator

NI

NoC

NI

DTL2AIAC

in

out

9 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Accelerators

Accelerator Types

Micro-programmed: FlexTiles DSP

Data-flow: Accelerators on eFPGA

Lite Accelerator

Example/Test Application

Implements 2 register banks

3 FIFOs

Adding two sequential values

GPP Node

AI

Accelerator

NI

NoC

NI

10 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Software Functions

Accelerator Software Functions

send_rqst (int acc_number, int size, int ch_id, int addr, char type)

send_data (int acc_number, int size, int ch_id, int addr, int* array)

read_data (int acc_number, int size, int addr, int* array)

write_register (int acc_number, int ch_id, int addr, int value)

read_register (int acc_number, char regis)

GPP Node

AI

Accelerator

NI

NoC

NI

11 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Development Platform Configuration

Xilinx Platform Studio Project

TUe Platform configured via XML file

Xilinx Microprocessor Project file (XMP)

Platform Configuration

User Constraint File (UCF)

Microprocessor Hardware Specification (MHS)

Microprocessor Software Specification (MSS)

12 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Adding Accelerators

Accelerator Integration

Peripheral Core (PCORE)

Add Accelerator files to pcore directory

Add Accelerator entry to MHS

Configuration Files

Black Box Definition file (BBD)

Peripheral Analyze Order file (PAO)

Microprocessor Peripheral Definition (MPD)

13 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Implementation

XPS Command Line Mode

xps -nw -scr system.tcl system.xmp

Simple Design:

GPP

AI+ Acc

DebugMonitor

NoC

14 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Implementation Results

Results

Standalone platform Occupied Slices: 5%

IOB: 1%

RAMB36E1: 13%

RAMB18E1: 0%

BUFG: 9%

Platform with lite accelerator Occupied Slices: 7%

IOB: 1%

RAMB36E1: 13%

RAMB18E1: 1%

BUFG: 12%

GPP Node

NI

NoC

Monitor Node

GPP Node

AI

Lite Accelerator

NI

NoC

NI

Monitor Node

15 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Implementation Results

Results

Platform with DSP Occupied Slices: 62%

IOB: 1%

RAMB36E1: 15%

RAMB18E1: 1%

BUFG: 9%

GPP Node

AI

DSP

NI

NoC

NI

Monitor Node

16 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Challenges

Platform with DSP

Unroutable Signals (Rats Nests)

Colors

Monitor: yellow

GPP: cyan

NoC: purple

AI: green

DSP: red

17 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Summary

Emulator Goal

Run demo application like optical flow

NoC

GPP Node

NI

NI

GPP Node

GPP Node

NI

NI

GPP Node

NI

NI

NI

NI

AI

Acc

AI

Acc

Acc

AI

Acc

AI

NoC

AURORA

18 /

18.0

7.20

15/A

HS

14

The

info

rmat

ion

cont

aine

d in

thi

s do

cum

ent

and

any

atta

chm

ents

are

the

pro

pert

y of

Fle

xTile

s co

nsor

tium

. Y

ou a

re h

ereb

y no

tifie

d th

at a

ny r

evie

w,

diss

emin

atio

n, d

istr

ibut

ion,

co

pyin

g or

oth

erw

ise

use

of t

his

docu

men

t m

ust

be d

one

in a

ccor

danc

e w

ith t

he C

A o

f th

e pr

ojec

t (T

RT

/DJ/

6244

1278

5.20

11).

Tem

plat

e ve

rsio

n 1

.0

18

Many Thanks for Your Attention!

www.flextiles.eu

Benedikt.Janssen@rub.de

The last Slide