A Multi-Level Optimizer For Burst Mode …nowick/asynctools/MLO-v0.9-tutorial.pdfMLO: A Multi-Level Optimizer For Burst Mode Asynchronous Controllers Walter Dearing Steven Nowick Computer

MLO

:A

Mul

ti-Le

vel O

ptim

izer

For

Bur

st M

ode

Asy

nchr

onou

s C

ontr

olle

rs

Wal

ter D

earin

g

Ste

ven

Now

ick

Com

pute

r Sci

ence

Dep

artm

ent

Col

umbi

a U

nive

rsity

This

wor

k w

as s

uppo

rted

by N

SF

ITR

Aw

ard

No.

NS

F-C

CR

-008

6036

and

by a

n In

itiat

ives

in S

cien

ce a

nd E

ngin

eerin

g (IS

E) g

rant

from

Col

umbi

a U

nive

rsity

(fro

m th

e O

ffice

of t

he E

xecu

tive

Vic

e P

resi

dent

for R

esea

rch)

2

MLO

: Acc

ess

Info

rmat

ion

!A

cces

sibl

e on

the

web

from

:

!In

itial

Rel

ease

!O

ne v

ersi

on �

for L

inux

Dis

tribu

tions

!In

clud

es

!C

ompl

ete

Tuto

rial

!D

ocum

enta

tion

!Ex

ampl

es

!To

ol re

quire

s P

ytho

n in

terp

rete

r to

run:

!C

onsu

lt R

EA

DM

E fo

r MLO

inst

alla

tion

info

rmat

ion

http

://w

ww

1.cs

.col

umbi

a.ed

u/~n

owic

k/as

ynct

ools

http

://w

ww

.pyt

hon.

org/

dow

nloa

d/

3

Out

line

!In

trodu

ctio

n to

MLO

(Mul

ti-Le

vel O

ptim

izer

)

!O

verv

iew

of f

eatu

res

!E

xam

ple

of fe

atur

es

!E

xam

ple

of g

ate

netw

orks

pro

duce

d

!B

uilt-

in v

erifi

er

!B

rief t

utor

ial

4

Intro

duct

ion

to M

LO

!In

trod

uctio

n to

MLO

(Mul

ti-Le

vel O

ptim

izer

)

!O

verv

iew

of f

eatu

res

!E

xam

ple

of fe

atur

es

!E

xam

ple

of g

ate

netw

orks

pro

duce

d

!B

uilt-

in v

erifi

er

!B

rief t

utor

ial

5

Intro

duct

ion

to M

LO!

MLO

is a

n in

tegr

ated

pos

t-pro

cess

ing

(i.e.

bac

kend

) too

l for

Min

imal

ist.

!Ta

rget

ed to

mul

ti-le

vel l

ogic

.

!In

con

trast

, Min

imal

ist c

urre

ntly

is ta

rget

ed to

two-

leve

l log

ic.

!D

esig

ned

to w

ork

on c

ombi

natio

nalh

azar

d-fr

ee lo

gic

for B

urst

Mod

e

cont

rolle

rs.

!U

ses

�haz

ard-

non-

incr

easi

ng�t

rans

form

s.

!O

utpu

t of M

LO is

Ver

ilog.

!M

LO is

a s

tand

alon

e to

ol ru

nnin

g fro

m th

e Li

nux

shel

l out

side

of

Min

imal

ist.

6

Intro

duct

ion

to M

LO

Mul

ti-le

vel t

ool f

low

BM

SPE

C (*

.bm

s)

MIN

IMAL

IST

PLA

file

(*.s

ol)

Tran

sitio

n fil

e(*

.pla

)

Mul

ti-Le

vel O

ptim

izer

Veril

ogfil

e(*

-ML.

v)

Two-

Leve

l

Mul

ti-Le

vel

New

7

Ove

rvie

ws

of fe

atur

es

!In

trodu

ctio

n to

MLO

(Mul

ti-Le

vel O

ptim

izer

)

!O

verv

iew

of f

eatu

res

!E

xam

ple

of fe

atur

es

!E

xam

ple

of g

ate

netw

orks

pro

duce

d

!B

uilt-

in v

erifi

er

!B

rief t

utor

ial

8

Feat

ure

Set

�O

verv

iew

MLO

has

man

y fe

atur

es w

hich

are

orth

ogon

al a

nd c

an

be u

sed

toge

ther

.

1.G

ate

fan-

in li

mita

tion

2.N

egat

ive-

logi

c

3.C

ritic

al E

vent

Opt

imiz

er (C

EO)

!Tw

o m

odes

a)A

utom

ated

b)U

ser-

spec

ified

9

Feat

ure

Set

�G

oal o

f CE

O

Goa

l of C

EO

is to

redu

ce th

e cr

itica

l pat

hof

crit

ical

eve

nts.

!C

ritic

al e

vent

sar

e de

fined

as

a tra

nsiti

on o

f a p

rimar

y

outp

ut in

resp

onse

to a

n in

put t

rans

ition

!D

one

by m

ovin

g cr

itica

l inp

uts

clos

er to

the

prim

ary

outp

ut.

!In

tent

ion

is fo

r crit

ical

inpu

ts to

trav

el th

roug

h th

e ci

rcui

t

as fa

st a

pos

sibl

e.

10

Feat

ure

Set

�C

EO M

ode

1 -

Aut

omat

ed M

ode

CE

O d

efau

lts to

aut

omat

ed m

ode

if no

use

r inp

ut s

peci

fied.

!To

ol a

utom

atic

ally

opt

imiz

es li

kely

crit

ical

pat

hs.

!Fo

cuse

s op

timiz

atio

ns o

n pr

imar

y in

put-t

o-ou

tput

path

sin

volv

ed in

dyn

amic

tran

sitio

ns.

!(i.

e. 0"

1 / 1

"0)

11

Feat

ure

Set

�C

EO M

ode

2 -

Use

r-Sp

ecifi

ed C

ritic

al E

vent

s

!U

ser m

ay k

now

whi

ch p

rimar

y in

put(s

) are

impo

rtant

to c

ritic

al p

rimar

y ou

tput

.

!D

one

by d

efin

ing

a cr

itica

l tra

nsiti

on&

inpu

t/out

put

pair

at th

e B

urst

-Mod

e sp

ec le

vel.

!Fo

r par

ticul

ar o

utpu

ts s

peci

fied

durin

g tra

nsiti

on, u

ser

over

rides

auto

mat

ed m

ode.

!C

onsi

der t

he s

ituat

ion

on th

e ne

xt s

lide

whe

re u

ser s

peci

fies

inpu

t & o

utpu

ts o

ff sp

ecifi

c ar

cs.

Use

r can

spe

cify

det

ails

for C

EO

to u

se.

12

Feat

ure

Set

�C

EO M

ode

2 -

Use

r-Sp

ecifi

ed C

ritic

al E

vent

s

Cas

e 1:

Non

col

oriz

ed

arc.

Use

r-Sp

ecifi

ed

noth

ing

is c

ritic

al. D

efau

lts

to a

utom

ated

mod

e fo

r ev

ery

outp

ut.

Cas

e 2:

Som

e ou

tput

s co

loriz

ed, s

ome

outp

uts

not.

Both

use

r-spe

cifie

d da

ta a

nd a

utom

ated

ap

proa

ches

are

use

d to

de

term

ine

criti

calit

y.

ITEv

entR

eqw

ill u

se u

ser-

spec

ified

dat

a to

de

term

ine

criti

calit

y.

Ctr

incR

eqw

ill d

efau

lt to

au

tom

ated

mod

e to

de

term

ine

criti

calit

y.

Cas

e 3:

Ever

y ou

tput

is

colo

rized

. Aut

omat

ed

appr

oach

is n

ever

used

. In

tITR

eq-i

s cr

itica

l with

re

spec

t to

Ctr

IncR

eq-,

whi

le IT

Even

t2Ti

cks-

is

NO

T cr

itica

l to

Ctr

IncR

eq-

.

Use

r-Sp

ecifi

ed C

ritic

al A

rcs

Hig

hlig

hted

inR

ed

IntIT

Req

-

ITEv

ent2

Tick

s-/

Ctr

IncR

eq-

50 1

23

4

IntIT

Req

+ /

ITEv

entR

eq+

IntIT

Req

+ /

ITEv

entR

eq+

ITEv

ent2

Tick

s-

Ctr

IncA

ck+

/ Ctr

IncR

eq-

IntIT

Req

-/

ITEv

entR

eq-

IntIT

Req

+ /

ITEv

entR

eq+

CtIn

cAck

-/

ITEv

entR

eq+

ITEv

ent2

Tick

s-/

Ctr

IncR

eq+

ITEv

entR

eq-

13

Exa

mpl

e of

feat

ures

!In

trodu

ctio

n to

MLO

(Mul

ti-Le

vel O

ptim

izer

)

!O

verv

iew

of f

eatu

res

!Ex

ampl

e of

feat

ures

!E

xam

ple

of g

ate

netw

orks

pro

duce

d

!B

uilt-

in v

erifi

er

!B

rief t

utor

ial

14

Feat

ure

Set

-In

itial

Tw

o-Le

vel

Impl

emen

tatio

n (b

efor

e ap

plyi

ng M

LO)

The

next

four

slid

es

pres

ent d

iffer

ent M

LO

outp

ut e

xam

ples

. Fo

r eac

h

exam

ple,

the

star

ting

circ

uit (

inpu

t to

MLO

) is

this

circ

uit

Two-

leve

l Str

uctu

re fr

om M

inim

alis

t Out

put

15

Feat

ure

Set

Exa

mpl

e 1

-G

ate

Fan-

in L

imita

tion

Res

ult o

f MLO

: M

ulti-

Leve

l circ

uit w

ith A

ND

gat

e fa

n-in

lim

it of

2

16

Feat

ure

Set

Exa

mpl

e 2

-N

egat

ive

Logi

c

Res

ult o

f MLO

: M

ulti-

Leve

l Circ

uit u

sing

MLO

Neg

ativ

e Lo

gic

This

mod

e ca

refu

lly o

ptim

izes

on

ly h

azar

d no

n-in

crea

sing

sa

fe tr

ansf

orm

atio

ns

(DeM

orga

n�s

Law

).

Opt

imiz

atio

ns a

re a

lso

incl

uded

to c

aref

ully

elim

inat

e ex

tra

inve

rter

s.

17

Feat

ure

Set

Exa

mpl

e 3

-C

EO Res

ult o

f MLO

: M

ulti-

Leve

l Circ

uit a

fter M

LOC

EOis

use

d

Gat

e D

ecom

pose

d.In

put

intit

req

is m

ore

criti

cal t

o

outp

ut it

even

treq

than

ctrin

cack

�and

y0�

criti

cal p

rimar

y in

put-

to-o

utpu

t pat

h

18

Feat

ure

Set

Exa

mpl

e 4

-C

ombi

ned

Res

ult o

f MLO

: Mul

ti-Le

vel C

ircui

t with

neg

ativ

e lo

gic,

AN

D g

ate

fan-

in li

mit

of 2

, an

d C

EO.

Gat

e fa

n-in

lim

it of

2

Neg

ativ

e Lo

gic

CEO

Opt

imiz

es C

ritic

al P

ath

19

Exa

mpl

e of

gat

e ne

twor

ks

prod

uced

!In

trodu

ctio

n to

MLO

(Mul

ti-Le

vel O

ptim

izer

)

!O

verv

iew

of f

eatu

res

!E

xam

ple

of fe

atur

es

!Ex

ampl

e of

gat

e ne

twor

ks p

rodu

ced

!B

uilt-

in v

erifi

er

!B

rief t

utor

ial

20

Gat

e D

ecom

posi

tion

!G

ates

can

be

deco

mpo

sed

in tw

o w

ays

!C

EO

�re

sulti

ng n

etw

ork

will

be

casc

aded

!Fa

n-in

lim

itatio

ns �

resu

lting

net

wor

k w

ill b

e ba

lanc

ed

!N

etw

ork

can

be a

mix

ture

of t

wo

met

hods

.

!C

onsi

der a

6 in

put A

ND

gat

e�.

21

Gat

e D

ecom

posi

tion

-C

asca

ded

Gat

e is

dec

ompo

sed

base

d on

prio

rity

diffe

renc

e of

inpu

ts

Onl

y us

ed fo

r CEO

Low

est I

nput

Prio

rity

H

ighe

st In

put P

riorit

y

(s

tate

var

iabl

es o

nly)

22

Gat

e D

ecom

posi

tion

�B

alan

ced

Gat

e is

dec

ompo

sed

stric

tly b

ased

on

fan-

in li

mita

tion

Onl

y us

ed fo

rgat

efa

n-in

lim

itatio

n

All

Inpu

ts h

ave

the

sam

e pr

iorit

y

23

Gat

e D

ecom

posi

tion

�M

ixG

ate

is d

ecom

pose

d ba

sed

on b

oth

diffe

renc

e in

prio

rity

of in

puts

and

fan-

in li

mits

Com

bina

tion

ofga

tefa

n-in

lim

itatio

n an

dC

EO

Low

Prio

rity

M

ediu

m P

riorit

y

Hig

h Pr

iorit

y

24

Brie

f tut

oria

l

!In

trodu

ctio

n to

MLO

(Mul

ti-Le

vel O

ptim

izer

)

!O

verv

iew

of f

eatu

res

!E

xam

ple

of fe

atur

es

!E

xam

ple

of g

ate

netw

orks

pro

duce

d

!B

uilt-

in v

erifi

er

!B

rief t

utor

ial

25

Bui

lt-in

Ver

ifier

!A

verif

ier i

s bu

ilt in

to M

LO.

!Ve

rifie

s M

LO o

utpu

t is

func

tiona

lly c

orre

ctan

d ha

zard

free

!C

ompa

res

two-

leve

l stru

ctur

e (M

inim

alis

t out

put)

to m

ulti-

leve

l

stru

ctur

e (M

LO o

utpu

t).

!Ve

rifie

s sp

ecifi

c pr

oper

ties

hold

for e

ach

gate

net

wor

k in

mul

ti-le

vel

stru

ctur

e.

!M

ust b

e ex

plic

itly

set u

sing

com

man

d lin

e fla

g (n

ot ru

n by

def

ault)

26

Brie

f tut

oria

l

!In

trodu

ctio

n to

MLO

(Mul

ti-Le

vel O

ptim

izer

)

!O

verv

iew

of f

eatu

res

!E

xam

ple

of fe

atur

es

!E

xam

ple

of g

ate

netw

orks

pro

duce

d

!B

uilt-

in v

erifi

er

!B

rief t

utor

ial

27

Tuto

rial

!Th

e sp

ec fi

le h

p-ir.

bms

will

be u

sed

durin

g th

is tu

toria

l.

This

can

be

foun

d in

the

exam

ples

fold

er.

!E

very

com

man

d us

ed in

the

tuto

rial i

s al

so d

escr

ibed

in

the

tool

hel

p m

enu.

!To

dis

play

gen

eral

hel

p m

enu

>MLO.py--help

!To

dis

play

use

r-sp

ecifi

ed c

ritic

al e

vent

hel

p m

enu

>MLO.py--more_help

28

Tuto

rial -

Setu

pC

reat

e th

e fil

es n

eede

d fo

r MLO

to p

roce

ss

Step

1 �

Cre

ate

wor

king

dire

ctor

y> mkdir

MLO_tutorial

Step

2 �

Cop

y sp

ec fi

le> cp examples/hp-ir.bms

MLO_tutorial/

Step

3 �

Ente

r Dire

ctor

y> cd

MLO_tutorial

Step

4 �

Cre

ate

.pla

& .s

ol fi

les

>minimalist-speed hp-ir.bms

single-output fedback

Step

5 �

Verif

y fil

es c

reat

ed> Verify hp_IR-Fs.sol

and hp_IR-

Fs.pla

has been created

Two-

Leve

lcirc

uit f

rom

Min

imal

ist

Ass

umes

Min

imal

ist i

s lo

aded

and

path

s ar

e se

t cor

rect

ly!

29

Tuto

rial �

Gat

e Fa

n-In

Lim

itatio

nIm

plem

ent c

ircui

t with

gat

e fa

n-in

lim

it of

2 fo

r AN

D g

ates

. Th

is c

an b

edo

ne u

sing

two

diffe

rent

met

hods

(res

ults

will

be th

e sa

me)

Step

1 �

Run

MLO

> MLO.py

–A 2 –d hp_IR-Fs.sol

Step

2 �

View

Res

ults

> less hp_IR-Fs-ML.v

Step

3 �

Verif

y R

esul

ts> Compare output with circuit on

next page

Step

1 �

Run

MLO

> MLO.py

–M 2 –d hp_IR-Fs.sol

Step

2 �

View

Res

ults


Step

3 �

Verif

y R

esul

ts> Compare output with circuit on

next page

Met

hod

1: I

ndep

ende

ntly

M

etho

d 2:

Glo

bally

S

peci

fy fa

n-in

lim

it fo

r AN

D g

ates

onl

y

S

peci

fy fa

n-in

lim

it fo

r eve

ry g

ate

type

30

Tuto

rial �

Gat

e Fa

n-In

Lim

itatio

nmodule hp_IR_Fs(intitreq, itevent2ticks, ctrincack, iteventreq, ctrincreq);

input intitreq, itevent2ticks, ctrincack;

output iteventreq, ctrincreq;

wire neg_ctrincack, neg_ctrincreq_i, neg_itevent2ticks, ctrincreq_i, ctrincreq, _i1;

wire _i2, _i3;

// Wires needed for multi-level

wire _i4, _i5, _i6;

// Feedback variables

assign ctrincreq_i= ctrincreq;

// Negative input literals

not (neg_ctrincack, ctrincack);

not (neg_ctrincreq_i, ctrincreq_i);

not (neg_itevent2ticks, itevent2ticks);

// First plane of logic

// --Network implementing gate _i3 --

and (_i3, _i4, neg_ctrincack);

and (_i4, intitreq, ctrincreq_i);


assign _i2 = itevent2ticks;


and (_i1, _i5, _i6);

and (_i5, neg_ctrincack, neg_ctrinreq_i);

and (_i6, intitreq, neg_itevent2ticks);

// Second plane of logic

// --Network implementing gate CtrIncReq--

or (ctrincreq, _i3, _i2);

// --Network implementing gate ITEventReq --

assign iteventreq= _i1;

endmodule

MLO

Out

put:

2 A

ND

gat

es h

ave

been

dec

ompo

sed

base

d on

fan-

in li

mit

of 2

.

31

Tuto

rial �

Neg

ativ

e Lo

gic

Impl

emen

t circ

uit u

sing

neg

ativ

e lo

gic.

Step

1 �

Run

MLO

> MLO.py

–n –d hp_IR-Fs.sol

Step

2 �

View

Res

ults


Step

3 �

Verif

y R

esul

ts> Compare output with

circuit on next page

MLO

Res

ult:

Mul

ti-Le

vel C

ircui

t usi

ng n

egat

ive

logi

c on

ly

32

Tuto

rial �

Neg

ativ

e Lo

gic

module hp_IR_Fs(intitreq, itevent2ticks, ctrincack, iteventreq, ctrincreq);



wire neg__i2, neg_ctrincack, neg_intitreq, ctrincreq_i, ctrincreq, _i1;

wire _i2, _i3;


wire neg__i3;




not (neg__i2, _i2);


not (neg_intitreq, intitreq);


// --Network implementing gate neg__i3 --

nand(neg__i3, neg_ctrincack, intitreq, ctrincreq_i);




nor (_i1, ctrincack, ctrincreq_i, neg_intitreq, itevent2ticks);



nand(ctrincreq, neg__i3, neg__i2);



endmodule

MLO

Out

put:

Onl

y N

OT/

NO

R/N

AN

D g

ates

use

d.

33

Tuto

rial �

CE

O C

ritic

al E

vent

O

ptim

izer

(Aut

omat

ed M

ode)

Run

MLO

usi

ng d

efau

lts (C

EO o

n, n

o ga

te fa

n-in

lim

its, a

nd n

one

gativ

e lo

gic)

.

Step

1 �

Run

MLO

> MLO.py

hp_IR-Fs.sol

Step

2 �

View

Res

ults


Step

3 �

Verif

y R

esul

ts> Compare output with

circuit on next page

Mul

ti-Le

vel c

ircui

t afte

r CEO

is u

sed

No

switc

h se

lect

ed: d

efau

lt is

CE

O in

aut

o m

ode

criti

cal p

rimar

y

inpu

t-to-

outp

ut p

ath

34

Tuto

rial �

CE

Omodule hp_IR_Fs(intitreq, itevent2ticks, ctrincack, iteventreq, ctrincreq);



wire neg_ctrincack, neg_itevent2ticks, neg_ctrincreq_i, ctrincreq_i, ctrincreq, _i1;

wire _i2, _i3;


wire _i4;









and (_i3, neg_ctrincack, intitreq, ctrincreq_i);




and (_i1, intitreq, _i4);

and (_i4, neg_itevent2ticks, neg_ctrincack, neg_ctrincreq_i);






endmodule

MLO

Out

put:

Gat

e D

ecom

pose

d. I

nput

intit

req

is

mor

e cr

itica

l to

outp

ut it

even

treq

than

ne

g_ct

rinca

ckan

d ne

g_ct

rincr

eq_i

35

Tuto

rial �

Feat

ures

Mix

Feat

ures

can

be

mix

ed a

nd m

atch

ed.

This

exa

mpl

e do

es n

ot u

tiliz

e C

EO, b

ut it

doe

s ta

rget

neg

ativ

e lo

gic

circ

uit a

nd li

mits

fan-

in to

2.

Step

1 �

Run

MLO

> MLO.py

–d –n –R 2 –N 2 hp_IR-Fs.sol

Step

2 �

View

Res

ults


Step

3 �

Verif

y R

esul

ts> Compare output with circuit on next

page

36

Tuto

rial �

Feat

ures

Mix

module hp_IR_Fs(intitreq, itevent2ticks, ctrincack, iteventreq, ctrincreq);



wire neg__i2, neg_ctrincack, neg_intitreq, neg_ctrincreq_i, neg_itevent2ticks, ctrincreq_i;

wire ctrincreq, _i1, _i2, _i3;


wire _i4, _i5, _i6, neg__i3;




not (neg__i2, _i2);


not (neg_intitreq, intitreq);




// --Network implementing gate neg__i3 --

nand(neg__i3, _i4, neg_ctrincack);

nor (_i4, neg_intitreq, neg_ctrincreq_i);




nor (_i1, _i5, _i6);

nand(_i5, neg_ctrincack, neg_ctrincreq_i);

nand(_i6, intitreq, neg_itevent2ticks);



nand(ctrincreq, neg__i3, neg__i2);



endmodule

MLO

Out

put:

Onl

y N

OT/

NO

R/N

AN

D g

ates

us

ed a

nd fa

n-in

lim

ited

to 2

37

Tuto

rial �

Use

r-S

peci

fied

Crit

ical

E

vent

s -O

verv

iew

!M

LO o

rigin

ally

nee

ded

only

two

files

to ru

n (*

.pla

and

*.so

l)

!To

util

ize

Use

r-S

peci

fied

Crit

ical

Eve

nts,

MLO

now

nee

ds tw

o

addi

tiona

l file

s:

1.st

ate_

info

.txt

2.*.

bms

spec

-file

38

Tuto

rial �

Use

r-S

peci

fied

Crit

ical

E

vent

s �

stat

e_in

fo.tx

tS

tate

min

imiz

atio

n da

ta is

nee

ded

to u

tiliz

e U

ser-

Spe

cifie

d C

ritic

al E

vent

s

!Th

is in

form

atio

n is

onl

y fo

und

on s

tand

ard

outp

ut d

urin

g ru

n of

Min

imal

ist.

!Th

eref

ore

new

pro

gram

cap

utur

e_m

in_d

ata

is n

eede

d to

redi

rect

Min

imal

ist

outp

ut to

sta

te_i

nfo.

txt

!R

uns

from

linu

xsh

ell

!W

rapp

er a

roun

d M

inim

alis

t cal

l

!Im

porta

nt to

not

e th

at c

aptu

re_m

in_d

ata

only

wor

ks w

ith s

ingl

e-ru

n

Min

imal

ist s

crip

ts.

In o

ther

wor

ds, i

t doe

sn�t

wor

k fo

r scr

ipt s

uite

s

39

Tuto

rial �

Use

r-S

peci

fied

Crit

ical

E

vent

s �

stat

e_in

fo.tx

tS

ynta

x R

ules

:

1.ca

putu

re_m

in_d

ata

sim

ply

prec

edes

Min

imal

ist c

all.

�Fo

r exa

mpl

e, if

use

r wan

ts to

cap

ture

dat

a of

min

imal

ist-a

rea

scrip

t usi

ng m

ulti-

outp

ut ru

n ty

pe, t

he s

ynta

x w

ould

be

> capture_min_dataminimalist-area hp-ir.bmsmulti-output

2.Sy

ntax

for M

inim

alis

t com

man

d lin

e op

tions

are

sam

e as

if

capt

ure_

min

_dat

aw

as n

otbe

ing

used

.

40

Tuto

rial �

Use

r-S

peci

fied

Crit

ical

E

vent

s �

*.bm

ssp

ec-fi

le

!C

ritic

al e

vent

s w

ill b

e sp

ecifi

ed b

y pr

ovid

ing

criti

cal i

nput

/out

put p

airs

for

spec

ified

tran

sitio

ns in

the

burs

t-mod

e sp

ecifi

catio

n.

!If

all i

nput

s ar

e cr

itica

l for

out

put d

urin

g tra

nsiti

on, w

ildca

rd(�

*�) c

an b

e

used

.

!N

o bl

ank

lines

can

sur

roun

d an

y in

serte

d us

er-s

peci

fied

criti

cal

com

men

ts(�

; CR

ITIC

AL�

stat

emen

ts) i

n th

e *.

bms

file.

!C

omm

ent m

ust i

mm

edia

tely

pre

cede

the

trans

ition

that

it is

des

crib

ing.

Burs

t Mod

e S

pec

file

has

to b

e ed

ited

to s

peci

fy c

ritic

al e

vent

s.

41

Tuto

rial �

Use

r-Spe

cifie

d C

ritic

al

Eve

nts

-*.b

ms

spec

-file

Syn

tax

Rul

es (r

edco

mm

ent):

1.To

spe

cify

that

the

inpu

t CtrI

ncA

ck+

is c

ritic

al to

CtrI

ncR

eq-d

urin

g tra

nsiti

on fr

om

curre

nt s

tate

3 to

nex

t sta

te 4

(not

e th

at th

is is

als

o sa

ying

that

ITev

ent2

Tick

s is

NO

Tcr

itica

l): ;CRITICAL: CtrIncAck+/CtrIncReq-

3 4 ITevent2Ticks-CtrIncAck+ | CtrIncReq-

2.To

spe

cify

all

inpu

ts a

re c

ritic

al to

CtrI

ncR

eq-f

rom

cur

rent

sta

te 3

to n

ext s

tate

5:

;CRITICAL: */CtrIncReq-


3.A

noth

er w

ay to

spe

cify

all

inpu

ts a

re c

ritic

al (s

ame

as R

ule

#2)t

o C

trInc

Req

-dur

ing

a tra

nsiti

on fr

om c

urre

nt s

tate

3 to

nex

t sta

te 5

:;CRITICAL: ITevent2Ticks-/CtrIncReq-, CtrIncAck+/CtrIncReq-


4.If

no in

put/o

utpu

ts a

re m

arke

d cr

itica

l, au

tom

ated

CE

O is

use

d. A

utom

ated

CE

O is

us

ed to

rate

crit

ical

ity o

f ITe

vent

2Tic

ks a

nd C

trInc

Act

with

resp

ect t

o C

trInc

Req

:3 5 ITevent2Ticks-CtrIncAck+ | CtrIncReq-

42

Tuto

rial �

Use

r-Spe

cifie

d C

ritic

al

Eve

nts

Ther

efor

e, th

ere

are

thre

e st

eps

to u

se U

ser-

Spec

ified

Crit

ical

Eve

nts

1.C

aptu

re o

utpu

t of M

inim

alis

t.

!R

un M

inim

alis

t with

cap

ture

_min

_dat

ato

cre

ate

stat

e_in

fo.tx

t

2.Sp

ecify

crit

ical

eve

nts

in s

pec

file

(*.b

ms)

.

!Ad

d ;CRITICAL...

stat

emen

ts

3.R

un M

LOw

ith p

rope

r com

man

d lin

e op

tions

.

!M

ust u

se �

U [*

.bm

ssp

ec-fi

le] o

ptio

n.

!O

nly

have

to s

peci

fy *

.bm

san

d *.s

ol fi

le, M

LO a

ssum

es

stat

e_in

fo.tx

t and

*.pl

aar

e in

sam

e di

rect

ory

as *.

sol.

43

Tuto

rial �

Use

r-Spe

cifie

d C

ritic

al

Even

ts: C

ompl

ete

Run

Spe

cify

Use

r-Spe

cifie

d C

ritic

al E

vent

sSt

ep 1

�C

aptu

re M

inim

alis

t Out

put

> capture_min_data

minimalist-speed hp-ir.bms

single-output

Step

2 �

Edit

spec

file

>emacs

(or vi) hp-ir.bms

Step

3 �

Add

crit

ical

eve

nts

to s

pec

file

(def

ault

auto

mod

e w

ill b

e us

ed fo

r oth

er e

vent

s)

> add line in red to file:

;CRITICAL: CtrIncAck+/CtrIncReq-

34

ITEvent2Ticks-

CtrIncAck+

| CtrIncReq-

Step

4 �

Run

MLO

> MLO.py

–U hp-ir.bms

hp_IR-s.sol

Step

5 �

View

Res

ults

> less hp_IR-s-ML.v

Step

6 �

Verif

y R

esul

ts> Compare output with circuit on next page

44

Tuto

rial �

Use

r Def

ined

Crit

ical

E

vent

smodule hp_IR_s(intitreq, itevent2ticks, ctrincack, iteventreq, ctrincreq);



wire neg_ctrincack, neg_itevent2ticks, neg_y0_i, y0_i, y0, _i1;

wire _i2, _i3, _i4, _i5;


wire _i6, _i7;


assign y0_i = y0;




not (neg_y0_i, y0_i);



and (_i5, neg_ctrincack, _i6);

and (_i6, intitreq, y0_i);




and (_i3, intitreq, _i7);

and (_i7, neg_itevent2ticks, neg_ctrincack, neg_y0_i);


and (_i2, neg_ctrincack, intitreq, y0_i);






// --Network implementing gate y0 --

or (y0, _i2, _i1);

// --Network implementing gate ITEventReq--


endmodule

Gat

e de

com

pose

d us

ing

fact

that

us

er-s

peci

fied

neg_

ctrin

cack

is

criti

cal t

o ct

rincr

eq

Not

ice

that

Aut

omat

ed C

EO

is

still

used

on

outp

uts

not

spec

ified

by

user

45

Tuto

rial �

Ver

ifier

Run

MLO

usi

ng w

ith C

EO a

nd u

se th

e ve

rifie

r to

ensu

re o

utpu

t is

func

tiona

lly c

orre

ct a

nd h

azar

d fre

e

Step

1 �

Run

MLO

> MLO.py

-V hp_IR-s.sol

Step

2 �

Verif

y M

LO o

utpu

t pas

ses

verif

ier (

verif

y su

cces

s m

essa

ge is

prin

ted)

> check standard output

-----------------------------------

Verification Results

Output has been verified -

Everything OK.

----------------------------------

If ou

tput

pas

ses

built

-in

verif

ier,

a su

cces

s m

essa

ge w

ill be

prin

ted

to s

tand

ard

outp

ut.

46

Rec

ap

!M

LO u

ses

Min

imal

ist t

wo-

leve

l log

ic o

utpu

t and

prod

uces

an

optim

ized

mul

ti-le

vel c

ircui

t

spec

ified

in V

erilo

g.

!To

ol h

as m

ultip

le fe

atur

es in

clud

ing

nega

tive

logi

c im

plem

enta

tions

, crit

ical

-eve

nt o

ptim

izer

,

and

gate

fan-

in li

mita

tions

.

A Multi-Level Optimizer For Burst Mode …nowick/asynctools/MLO-v0.9-tutorial.pdfMLO: A Multi-Level Optimizer For Burst Mode Asynchronous Controllers Walter Dearing Steven Nowick Computer

Documents