<s>
Binary	B-Operating_System
Modular	I-Operating_System
Dataflow	I-Operating_System
Machine	I-Operating_System
(	O
BMDFM	B-Operating_System
)	O
is	O
a	O
software	O
package	O
that	O
enables	O
running	O
an	O
application	O
in	O
parallel	O
on	O
shared	O
memory	O
symmetric	B-Operating_System
multiprocessing	I-Operating_System
(	O
SMP	O
)	O
computers	O
using	O
the	O
multiple	O
processors	O
to	O
speed	O
up	O
the	O
execution	O
of	O
single	O
applications	O
.	O
</s>
<s>
BMDFM	B-Operating_System
automatically	O
identifies	O
and	O
exploits	O
parallelism	B-Operating_System
due	O
to	O
the	O
static	O
and	O
mainly	O
dynamic	O
scheduling	O
of	O
the	O
dataflow	B-Application
instruction	O
sequences	O
derived	O
from	O
the	O
formerly	O
sequential	O
program	O
.	O
</s>
<s>
The	O
BMDFM	B-Operating_System
dynamic	O
scheduling	O
subsystem	O
performs	O
a	O
symmetric	B-Operating_System
multiprocessing	I-Operating_System
(	O
SMP	O
)	O
emulation	B-Application
of	O
a	O
tagged-token	O
dataflow	B-Application
machine	O
to	O
provide	O
the	O
transparent	O
dataflow	B-Application
semantics	O
for	O
the	O
applications	O
.	O
</s>
<s>
Recent	O
commodity	O
SMP	O
machines	O
for	O
technical	O
computing	O
can	O
have	O
many	O
tightly	O
coupled	O
cores	O
(	O
good	O
examples	O
are	O
SMP	O
machines	O
based	O
on	O
multi-core	B-Architecture
processors	I-Architecture
from	O
Intel	O
(	O
Core	B-Device
or	O
Xeon	B-Device
)	O
or	O
IBM	O
(	O
Power	B-Device
)	O
)	O
.	O
</s>
<s>
Multi-core	B-Architecture
processors	I-Architecture
are	O
intended	O
to	O
exploit	O
a	O
thread-level	O
parallelism	B-Operating_System
,	O
identified	O
by	O
software	O
.	O
</s>
<s>
Hence	O
,	O
the	O
most	O
challenging	O
task	O
is	O
to	O
find	O
an	O
efficient	O
way	O
to	O
harness	O
power	B-Device
of	O
multi-core	B-Architecture
processors	I-Architecture
for	O
processing	O
an	O
application	O
program	O
in	O
parallel	O
.	O
</s>
<s>
Existent	O
OpenMP	O
paradigm	O
of	O
the	O
static	O
parallelization	B-Operating_System
with	O
a	O
fork-join	O
runtime	O
library	O
works	O
pretty	O
well	O
for	O
loop-intensive	O
regular	O
array-based	O
computations	O
only	O
,	O
however	O
,	O
compile-time	O
parallelization	B-Operating_System
methods	O
are	O
weak	O
in	O
general	O
and	O
almost	O
inapplicable	O
for	O
irregular	O
applications	O
:	O
</s>
<s>
The	O
BMDFM	B-Operating_System
technology	O
mainly	O
uses	O
dynamic	O
scheduling	O
to	O
exploit	O
parallelism	B-Operating_System
of	O
an	O
application	O
program	O
,	O
thus	O
,	O
BMDFM	B-Operating_System
avoids	O
mentioned	O
disadvantages	O
of	O
the	O
compile-time	O
methods	O
.	O
</s>
<s>
BMDFM	B-Operating_System
is	O
a	O
parallel	B-Operating_System
programming	I-Operating_System
environment	O
for	O
multi-core	B-Architecture
SMP	O
that	O
provides	O
:	O
</s>
<s>
Transparent	O
(	O
implicit	O
)	O
exploitation	O
of	O
parallelism	B-Operating_System
in	O
a	O
natural	O
and	O
load	O
balanced	O
manner	O
using	O
all	O
available	O
multi-core	B-Architecture
processors	I-Architecture
in	O
the	O
system	O
automatically	O
.	O
</s>
<s>
BMDFM	B-Operating_System
combines	O
the	O
advantages	O
of	O
known	O
architectural	O
principles	O
into	O
a	O
single	O
hybrid	O
architecture	O
that	O
is	O
able	O
to	O
exploit	O
implicit	B-Operating_System
parallelism	I-Operating_System
of	O
the	O
applications	O
having	O
negligible	O
dynamic	O
scheduling	O
overhead	O
and	O
no	O
bottlenecks	O
.	O
</s>
<s>
Mainly	O
,	O
the	O
basic	O
dataflow	B-Application
principle	O
is	O
used	O
.	O
</s>
<s>
The	O
dataflow	B-Application
principle	O
says	O
:	O
"	O
An	O
instruction	O
or	O
a	O
function	O
can	O
be	O
executed	O
as	O
soon	O
as	O
all	O
its	O
arguments	O
are	O
ready	O
.	O
</s>
<s>
A	O
dataflow	B-Application
machine	O
manages	O
the	O
tags	O
for	O
every	O
piece	O
of	O
data	O
at	O
runtime	O
.	O
</s>
<s>
The	O
main	O
feature	O
of	O
BMDFM	B-Operating_System
is	O
to	O
provide	O
a	O
conventional	O
programming	O
paradigm	O
at	O
the	O
top	O
level	O
,	O
so-called	O
transparent	O
dataflow	B-Application
semantics	O
.	O
</s>
<s>
A	O
user	O
understands	O
BMDFM	B-Operating_System
as	O
a	O
virtual	B-Architecture
machine	I-Architecture
(	O
VM	O
)	O
,	O
which	O
runs	O
all	O
statements	O
of	O
an	O
application	O
program	O
in	O
parallel	O
,	O
having	O
all	O
parallelizing	O
and	O
synchronizing	O
mechanisms	O
fully	O
transparent	O
.	O
</s>
<s>
The	O
two	O
first	O
statements	O
are	O
independent	O
,	O
so	O
a	O
dataflow	B-Application
engine	O
of	O
BMDFM	B-Operating_System
can	O
run	O
them	O
on	O
different	O
processors	O
or	O
processor	O
's	O
cores	O
.	O
</s>
<s>
The	O
dataflow	B-Application
engine	O
recognizes	O
dependencies	O
automatically	O
because	O
of	O
its	O
ability	O
to	O
build	O
a	O
dataflow	B-Application
graph	O
dynamically	O
at	O
runtime	O
.	O
</s>
<s>
Additionally	O
,	O
the	O
dataflow	B-Application
engine	O
correctly	O
orders	O
the	O
output	O
stream	O
to	O
output	O
the	O
results	O
sequentially	O
.	O
</s>
<s>
The	O
dataflow	B-Application
engine	O
of	O
BMDFM	B-Operating_System
will	O
keep	O
variables	O
"	O
a	O
"	O
and	O
"	O
b	O
"	O
under	O
unique	O
contexts	O
for	O
each	O
iteration	O
.	O
</s>
<s>
Therefore	O
,	O
the	O
dataflow	B-Application
engine	O
can	O
exploit	O
both	O
local	O
parallelism	B-Operating_System
within	O
the	O
iteration	O
and	O
global	O
parallelism	B-Operating_System
as	O
well	O
running	O
multiple	O
iterations	O
simultaneously	O
.	O
</s>
<s>
BMDFM	B-Operating_System
is	O
a	O
convenient	O
parallel	B-Operating_System
programming	I-Operating_System
environment	O
and	O
an	O
efficient	O
runtime	O
engine	O
for	O
multi-core	B-Architecture
SMP	O
due	O
to	O
the	O
MIMD	O
unification	O
of	O
several	O
architectural	O
paradigms	O
(	O
von-Neumann	O
,	O
SMP	O
and	O
dataflow	B-Application
)	O
:	O
</s>
<s>
At	O
first	O
,	O
it	O
is	O
a	O
hybrid	O
dataflow	B-Application
emulator	B-Application
running	O
multithreadedly	O
on	O
commodity	O
SMP	O
.	O
</s>
<s>
The	O
SMP	O
ensures	O
MIMD	O
while	O
dataflow	B-Application
exploits	O
implicit	B-Operating_System
parallelism	I-Operating_System
.	O
</s>
<s>
At	O
second	O
,	O
it	O
is	O
a	O
hybrid	O
multithreaded	O
dataflow	B-Application
runtime	O
engine	O
controlled	O
by	O
a	O
von-Neumann	O
front-end	O
VM	O
.	O
</s>
<s>
The	O
dataflow	B-Application
runtime	O
engine	O
executes	O
tagged-token	O
contextual	O
parallel	O
instructions	O
(	O
opposite	O
to	O
the	O
restricted	O
fork-join	O
paradigm	O
)	O
while	O
the	O
von-Neumann	O
front-end	O
VM	O
initializes	O
contexts	O
and	O
feeds	O
the	O
dataflow	B-Application
runtime	O
engine	O
with	O
marshaled	O
clusters	O
of	O
instructions	O
.	O
</s>
<s>
The	O
von-Neumann	O
front-end	O
VM	O
tries	O
statically	O
to	O
split	O
an	O
application	O
into	O
parallel	O
marshaled	O
clusters	O
of	O
instructions	O
while	O
the	O
dataflow	B-Application
runtime	O
engine	O
complements	O
the	O
static	O
parallelizing	O
methods	O
dynamically	O
.	O
</s>
<s>
BMDFM	B-Operating_System
is	O
intended	O
for	O
use	O
in	O
a	O
role	O
of	O
the	O
parallel	O
runtime	O
engine	O
(	O
instead	O
of	O
conventional	O
fork-join	O
runtime	O
library	O
)	O
able	O
to	O
run	O
irregular	O
applications	O
automatically	O
in	O
parallel	O
.	O
</s>
<s>
Due	O
to	O
the	O
transparent	O
dataflow	B-Application
semantics	O
on	O
top	O
,	O
BMDFM	B-Operating_System
is	O
a	O
simple	O
parallelization	B-Operating_System
technique	O
for	O
application	O
programmers	O
and	O
,	O
at	O
the	O
same	O
time	O
,	O
is	O
a	O
much	O
better	O
parallel	B-Operating_System
programming	I-Operating_System
and	O
compiling	O
technology	O
for	O
multi-core	B-Architecture
SMP	O
computers	O
.	O
</s>
<s>
The	O
basic	O
concept	O
of	O
BMDFM	B-Operating_System
relies	O
on	O
underlying	O
commodity	O
SMP	O
hardware	O
,	O
which	O
is	O
available	O
on	O
the	O
market	O
.	O
</s>
<s>
Normally	O
,	O
SMP	O
vendors	O
provide	O
their	O
own	O
SMP	O
Operating	O
System	O
(	O
OS	O
)	O
with	O
an	O
SVR4/POSIX	O
UNIX	O
interface	O
(	O
Linux	B-Application
,	O
HP-UX	O
,	O
SunOS/Solaris	O
,	O
Tru64OSF1	O
,	O
IRIX	O
,	O
AIX	O
,	O
BSD	O
,	O
MacOS	B-Application
,	O
etc	O
.	O
)	O
.	O
</s>
<s>
On	O
top	O
of	O
an	O
SMP	O
OS	O
,	O
the	O
multithreaded	O
dataflow	B-Application
runtime	O
engine	O
performs	O
a	O
software	B-Application
emulation	I-Application
of	O
the	O
dataflow	B-Application
machine	O
.	O
</s>
<s>
Such	O
a	O
virtual	B-Architecture
machine	I-Architecture
has	O
interfaces	O
to	O
the	O
virtual	B-Architecture
machine	I-Architecture
language	O
and	O
to	O
C	O
providing	O
the	O
transparent	O
dataflow	B-Application
semantics	O
for	O
conventional	O
programming	O
.	O
</s>
<s>
BMDFM	B-Operating_System
is	O
built	O
as	O
a	O
hybrid	O
of	O
several	O
architectural	O
principles	O
:	O
</s>
<s>
Implicit	O
parallel	O
execution	O
is	O
ensured	O
by	O
dataflow	B-Application
emulation	B-Application
.	O
</s>
<s>
Von-Neumann	O
computational	O
principle	O
is	O
good	O
to	O
implement	O
the	O
Front-end	O
Control	O
Virtual	B-Architecture
Machine	I-Architecture
.	O
</s>
<s>
The	O
BMDFM	B-Operating_System
dynamic	O
scheduling	O
subsystem	O
is	O
an	O
efficient	O
SMP	O
emulator	B-Application
of	O
the	O
tagged-token	O
dataflow	B-Application
machine	O
.	O
</s>
<s>
The	O
Shared	O
Memory	O
Pool	O
is	O
divided	O
in	O
three	O
main	O
parts	O
:	O
input/output	B-General_Concept
ring	O
buffer	O
port	O
(	O
IORBP	O
)	O
,	O
data	O
buffer	O
(	O
DB	O
)	O
,	O
and	O
operation	O
queue	O
(	O
OQ	O
)	O
.	O
</s>
<s>
The	O
front-end	O
control	O
virtual	B-Architecture
machine	I-Architecture
schedules	O
an	O
input	O
application	O
program	O
statically	O
and	O
puts	O
clustered	O
instructions	O
and	O
data	O
of	O
the	O
input	O
program	O
into	O
the	O
IORBP	O
.	O
</s>
<s>
The	O
context	O
is	O
a	O
special	O
unique	O
identifier	O
representing	O
a	O
copy	O
of	O
data	O
within	O
different	O
iteration	O
bodies	O
accordingly	O
to	O
the	O
tagged-token	O
dataflow	B-Application
architecture	O
.	O
</s>
<s>
In	O
order	O
to	O
allow	O
several	O
processes	O
accessing	O
the	O
same	O
data	O
concurrently	O
,	O
the	O
BMDFM	B-Operating_System
dynamic	O
scheduler	O
locks	O
objects	O
in	O
the	O
shared	O
memory	O
pool	O
via	O
SVR4/POSIX	O
semaphore	O
operations	O
.	O
</s>
<s>
Every	O
machine	O
supporting	O
ANSI	O
C	O
and	O
POSIX	O
;	O
UNIX	B-Operating_System
System	I-Operating_System
V	I-Operating_System
(	O
SVR4	O
)	O
may	O
run	O
BMDFM	B-Operating_System
.	O
</s>
<s>
BMDFM	B-Operating_System
is	O
provided	O
as	O
full	O
multi-threaded	O
versions	O
for	O
:	O
</s>
<s>
x86	B-Operating_System
:	O
Linux/32	O
,	O
FreeBSD/32	O
,	O
OpenBSD/32	O
,	O
NetBSD/32	O
,	O
MacOS/32	O
,	O
SunOS/32	O
,	O
UnixWare/32	O
,	O
Minix/32	O
,	O
Android/32	O
,	O
Win-Cygwin/32	O
,	O
Win-UWIN/32	O
,	O
Win-SFU-SUA/32	O
;	O
</s>
<s>
x86-64	B-Device
:	O
Linux/64	O
,	O
FreeBSD/64	O
,	O
OpenBSD/64	O
,	O
NetBSD/64	O
,	O
MacOS/64	O
,	O
SunOS/64	O
,	O
Android/64	O
,	O
Win-Cygwin/64	O
;	O
</s>
<s>
VAX	B-Device
:	O
Ultrix/32	O
;	O
</s>
<s>
Alpha	B-Device
:	O
Tru64OSF1/64	O
,	O
Linux/64	O
,	O
FreeBSD/64	O
,	O
OpenBSD/64	O
;	O
</s>
<s>
IA-64	B-General_Concept
:	O
HP-UX/32	O
,	O
HP-UX/64	O
,	O
Linux/64	O
,	O
FreeBSD/64	O
;	O
</s>
<s>
XeonPhiMIC	B-General_Concept
:	O
Linux/64	O
;	O
</s>
<s>
MCST-Elbrus	B-General_Concept
:	O
Linux/32	O
,	O
Linux/64	O
;	O
</s>
<s>
PA-RISC	B-Device
:	O
HP-UX/32	O
,	O
HP-UX/64	O
,	O
Linux/32	O
;	O
</s>
<s>
SPARC	B-Architecture
:	O
SunOS/32	O
,	O
SunOS/64	O
,	O
Linux/32	O
,	O
Linux/64	O
,	O
FreeBSD/64	O
,	O
OpenBSD/64	O
;	O
</s>
<s>
MIPS	B-Device
:	O
IRIX/32	O
,	O
IRIX/64	O
,	O
Linux/32	O
,	O
Linux/64	O
;	O
</s>
<s>
MIPSel	B-Device
:	O
Linux/32	O
,	O
Linux/64	O
,	O
Android/32	O
,	O
Android/64	O
;	O
</s>
<s>
PowerPC	B-Architecture
:	O
AIX/32	O
,	O
AIX/64	O
,	O
MacOS/32	O
,	O
MacOS/64	O
,	O
Linux/32	O
,	O
Linux/64	O
,	O
FreeBSD/32	O
,	O
FreeBSD/64	O
;	O
</s>
<s>
PowerPCle	O
:	O
Linux/32	O
,	O
Linux/64	O
;	O
</s>
<s>
S/390	B-Device
:	O
Linux/32	O
,	O
Linux/64	O
;	O
</s>
<s>
M68000	B-Device
:	O
Linux/32	O
;	O
</s>
<s>
ARM	B-Architecture
:	O
Linux/32	O
,	O
Linux/64	O
,	O
FreeBSD/64	O
,	O
Android/32	O
,	O
Android/64	O
,	O
MacOS/64	O
;	O
</s>
<s>
ARMbe	B-Architecture
:	O
Linux/64	O
;	O
</s>
<s>
RISC-V	B-Device
:	O
Linux/32	O
,	O
Linux/64	O
;	O
</s>
<s>
and	O
a	O
limited	O
single-threaded	O
version	O
for	O
x86	B-Operating_System
:	O
Win/32	O
.	O
</s>
