<s>
Dataflow	B-General_Concept
architecture	I-General_Concept
is	O
a	O
dataflow-based	O
computer	B-General_Concept
architecture	I-General_Concept
that	O
directly	O
contrasts	O
the	O
traditional	O
von	B-Architecture
Neumann	I-Architecture
architecture	I-Architecture
or	O
control	O
flow	O
architecture	O
.	O
</s>
<s>
Dataflow	B-General_Concept
architectures	I-General_Concept
have	O
no	O
program	B-General_Concept
counter	I-General_Concept
,	O
in	O
concept	O
:	O
the	O
executability	O
and	O
execution	O
of	O
instructions	O
is	O
solely	O
determined	O
based	O
on	O
the	O
availability	O
of	O
input	O
arguments	O
to	O
the	O
instructions	O
,	O
so	O
that	O
the	O
order	O
of	O
instruction	O
execution	O
may	O
be	O
hard	O
to	O
predict	O
.	O
</s>
<s>
Although	O
no	O
commercially	O
successful	O
general-purpose	O
computer	O
hardware	O
has	O
used	O
a	O
dataflow	B-General_Concept
architecture	I-General_Concept
,	O
it	O
has	O
been	O
successfully	O
implemented	O
in	O
specialized	O
hardware	O
such	O
as	O
in	O
digital	B-General_Concept
signal	I-General_Concept
processing	I-General_Concept
,	O
network	B-Protocol
routing	I-Protocol
,	O
graphics	O
processing	O
,	O
telemetry	O
,	O
and	O
more	O
recently	O
in	O
data	O
warehousing	O
,	O
and	O
artificial	B-Application
intelligence	I-Application
(	O
as	O
:	O
polymorphic	O
dataflow	B-Application
Convolution	O
Engine	O
,	O
structure-driven	O
,	O
dataflow	B-Application
scheduling	O
)	O
.	O
</s>
<s>
It	O
is	O
also	O
very	O
relevant	O
in	O
many	O
software	O
architectures	O
today	O
including	O
database	O
engine	O
designs	O
and	O
parallel	B-Operating_System
computing	I-Operating_System
frameworks	O
.	O
</s>
<s>
Synchronous	O
dataflow	B-General_Concept
architectures	I-General_Concept
tune	O
to	O
match	O
the	O
workload	O
presented	O
by	O
real-time	O
data	O
path	O
applications	O
such	O
as	O
wire	O
speed	O
packet	O
forwarding	O
.	O
</s>
<s>
Dataflow	B-General_Concept
architectures	I-General_Concept
that	O
are	O
deterministic	O
in	O
nature	O
enable	O
programmers	O
to	O
manage	O
complex	O
tasks	O
such	O
as	O
processor	O
load	O
balancing	O
,	O
synchronization	O
and	O
accesses	O
to	O
common	O
resources	O
.	O
</s>
<s>
Meanwhile	O
,	O
there	O
is	O
a	O
clash	O
of	O
terminology	O
,	O
since	O
the	O
term	O
dataflow	B-Application
is	O
used	O
for	O
a	O
subarea	O
of	O
parallel	B-Operating_System
programming	I-Operating_System
:	O
for	O
dataflow	B-Application
programming	I-Application
.	O
</s>
<s>
Hardware	O
architectures	O
for	O
dataflow	B-Application
was	O
a	O
major	O
topic	O
in	O
computer	B-General_Concept
architecture	I-General_Concept
research	O
in	O
the	O
1970s	O
and	O
early	O
1980s	O
.	O
</s>
<s>
Jack	O
Dennis	O
of	O
MIT	O
pioneered	O
the	O
field	O
of	O
static	O
dataflow	B-General_Concept
architectures	I-General_Concept
while	O
the	O
Manchester	O
Dataflow	B-Application
Machine	O
and	O
MIT	O
Tagged	O
Token	O
architecture	O
were	O
major	O
projects	O
in	O
dynamic	O
dataflow	B-Application
.	O
</s>
<s>
Building	O
content-addressable	B-Data_Structure
memory	I-Data_Structure
(	O
CAM	B-Data_Structure
)	O
large	O
enough	O
to	O
hold	O
all	O
of	O
the	O
dependencies	O
of	O
a	O
real	O
program	O
.	O
</s>
<s>
Nonetheless	O
,	O
out-of-order	B-General_Concept
execution	I-General_Concept
(	O
OOE	B-General_Concept
)	O
has	O
become	O
the	O
dominant	O
computing	O
paradigm	O
since	O
the	O
1990s	O
.	O
</s>
<s>
It	O
is	O
a	O
form	O
of	O
restricted	O
dataflow	B-Application
.	O
</s>
<s>
The	O
execution	O
window	O
follows	O
the	O
sequential	O
order	O
of	O
the	O
von	B-Architecture
Neumann	I-Architecture
architecture	I-Architecture
,	O
however	O
within	O
the	O
window	O
,	O
instructions	O
are	O
allowed	O
to	O
be	O
completed	O
in	O
data	O
dependency	O
order	O
.	O
</s>
<s>
This	O
is	O
accomplished	O
in	O
CPUs	B-Device
that	O
dynamically	O
tag	O
the	O
data	O
dependencies	O
of	O
the	O
code	O
in	O
the	O
execution	O
window	O
.	O
</s>
<s>
The	O
logical	O
complexity	O
of	O
dynamically	O
keeping	O
track	O
of	O
the	O
data	O
dependencies	O
,	O
restricts	O
OOE	B-General_Concept
CPUs	B-Device
to	O
a	O
small	O
number	O
of	O
execution	B-General_Concept
units	I-General_Concept
(	O
2-6	O
)	O
and	O
limits	O
the	O
execution	O
window	O
sizes	O
to	O
the	O
range	O
of	O
32	O
to	O
200	O
instructions	O
,	O
much	O
smaller	O
than	O
envisioned	O
for	O
full	O
dataflow	B-Application
machines	O
.	O
</s>
<s>
Designs	O
that	O
use	O
conventional	O
memory	O
addresses	O
as	O
data	O
dependency	O
tags	O
are	O
called	O
static	O
dataflow	B-Application
machines	O
.	O
</s>
<s>
Designs	O
that	O
use	O
content-addressable	B-Data_Structure
memory	I-Data_Structure
(	O
CAM	B-Data_Structure
)	O
are	O
called	O
dynamic	O
dataflow	B-Application
machines	O
.	O
</s>
<s>
They	O
use	O
tags	O
in	O
memory	O
to	O
facilitate	O
parallelism	B-Operating_System
.	O
</s>
<s>
Normally	O
,	O
in	O
the	O
control	O
flow	O
architecture	O
,	O
compilers	B-Language
analyze	O
program	O
source	O
code	O
for	O
data	O
dependencies	O
between	O
instructions	O
in	O
order	O
to	O
better	O
organize	O
the	O
instruction	O
sequences	O
in	O
the	O
binary	O
output	O
files	O
.	O
</s>
<s>
Binaries	O
compiled	B-Language
for	O
a	O
dataflow	B-Application
machine	O
contain	O
this	O
dependency	O
information	O
.	O
</s>
<s>
A	O
dataflow	B-Application
compiler	B-Language
records	O
these	O
dependencies	O
by	O
creating	O
unique	O
tags	O
for	O
each	O
dependency	O
instead	O
of	O
using	O
variable	O
names	O
.	O
</s>
<s>
Compiler	B-Language
detects	O
the	O
loops	O
,	O
break	O
statements	O
and	O
various	O
programming	O
control	O
syntax	O
for	O
data	B-Application
flow	I-Application
.	O
</s>
<s>
Programs	O
are	O
loaded	O
into	O
the	O
CAM	B-Data_Structure
of	O
a	O
dynamic	O
dataflow	B-Application
computer	O
.	O
</s>
<s>
When	O
all	O
of	O
the	O
tagged	O
operands	O
of	O
an	O
instruction	O
become	O
available	O
(	O
that	O
is	O
,	O
output	O
from	O
previous	O
instructions	O
and/or	O
user	O
input	O
)	O
,	O
the	O
instruction	O
is	O
marked	O
as	O
ready	O
for	O
execution	O
by	O
an	O
execution	B-General_Concept
unit	I-General_Concept
.	O
</s>
<s>
Once	O
an	O
instruction	O
is	O
completed	O
by	O
an	O
execution	B-General_Concept
unit	I-General_Concept
,	O
its	O
output	O
data	O
is	O
sent	O
(	O
with	O
its	O
tag	O
)	O
to	O
the	O
CAM	B-Data_Structure
.	O
</s>
<s>
In	O
this	O
way	O
,	O
subsequent	O
instructions	O
are	O
executed	O
in	O
proper	O
order	O
,	O
avoiding	O
race	B-Operating_System
conditions	I-Operating_System
.	O
</s>
<s>
An	O
instruction	O
,	O
along	O
with	O
its	O
required	O
data	O
operands	O
,	O
is	O
transmitted	O
to	O
an	O
execution	B-General_Concept
unit	I-General_Concept
as	O
a	O
packet	O
,	O
also	O
called	O
an	O
instruction	O
token	O
.	O
</s>
<s>
Similarly	O
,	O
output	O
data	O
is	O
transmitted	O
back	O
to	O
the	O
CAM	B-Data_Structure
as	O
a	O
data	O
token	O
.	O
</s>
<s>
Dataflow	B-Application
networks	O
deliver	O
the	O
instruction	O
tokens	O
to	O
the	O
execution	B-General_Concept
units	I-General_Concept
and	O
return	O
the	O
data	O
tokens	O
to	O
the	O
CAM	B-Data_Structure
.	O
</s>
<s>
In	O
contrast	O
to	O
the	O
conventional	O
von	B-Architecture
Neumann	I-Architecture
architecture	I-Architecture
,	O
data	O
tokens	O
are	O
not	O
permanently	O
stored	O
in	O
memory	O
,	O
rather	O
they	O
are	O
transient	O
messages	O
that	O
only	O
exist	O
when	O
in	O
transit	O
to	O
the	O
instruction	O
storage	O
.	O
</s>
