<s>
In	O
computing	O
,	O
single	B-Operating_System
program	I-Operating_System
,	I-Operating_System
multiple	I-Operating_System
data	I-Operating_System
(	O
SPMD	B-Operating_System
)	O
is	O
a	O
technique	O
employed	O
to	O
achieve	O
parallelism	B-Operating_System
;	O
it	O
is	O
a	O
subcategory	O
of	O
MIMD	B-Operating_System
.	O
</s>
<s>
SPMD	B-Operating_System
is	O
the	O
most	O
common	O
style	O
of	O
parallel	B-Operating_System
programming	I-Operating_System
.	O
</s>
<s>
It	O
is	O
also	O
a	O
prerequisite	O
for	O
research	O
concepts	O
such	O
as	O
active	B-Operating_System
messages	I-Operating_System
and	O
distributed	B-Operating_System
shared	I-Operating_System
memory	I-Operating_System
.	O
</s>
<s>
In	O
SPMD	B-Operating_System
,	O
multiple	O
autonomous	O
processors	O
simultaneously	O
execute	O
the	O
same	O
program	O
at	O
independent	O
points	O
,	O
rather	O
than	O
in	O
the	O
lockstep	B-General_Concept
that	O
SIMD	B-Device
or	O
SIMT	B-General_Concept
imposes	O
on	O
different	O
data	O
.	O
</s>
<s>
With	O
SPMD	B-Operating_System
,	O
tasks	O
can	O
be	O
executed	O
on	O
general	O
purpose	O
CPUs	B-Device
;	O
SIMD	B-Device
requires	O
vector	B-Operating_System
processors	I-Operating_System
to	O
manipulate	O
data	O
streams	O
.	O
</s>
<s>
SPMD	B-Operating_System
usually	O
refers	O
to	O
message	B-Architecture
passing	I-Architecture
programming	I-Architecture
on	O
distributed	B-Operating_System
memory	I-Operating_System
computer	O
architectures	O
.	O
</s>
<s>
A	O
distributed	B-Operating_System
memory	I-Operating_System
computer	O
consists	O
of	O
a	O
collection	O
of	O
independent	O
computers	O
,	O
called	O
nodes	O
.	O
</s>
<s>
Barrier	B-Operating_System
synchronization	O
may	O
also	O
be	O
implemented	O
by	O
messages	O
.	O
</s>
<s>
The	O
messages	O
can	O
be	O
sent	O
by	O
a	O
number	O
of	O
communication	O
mechanisms	O
,	O
such	O
as	O
TCP/IP	B-Protocol
over	O
Ethernet	O
,	O
or	O
specialized	O
high-speed	O
interconnects	O
such	O
as	O
Myrinet	B-General_Concept
and	O
Supercomputer	O
Interconnect	O
.	O
</s>
<s>
Nowadays	O
,	O
the	O
programmer	O
is	O
isolated	O
from	O
the	O
details	O
of	O
the	O
message	B-Architecture
passing	I-Architecture
by	O
standard	O
interfaces	O
,	O
such	O
as	O
PVM	B-Operating_System
and	O
MPI	B-Application
.	O
</s>
<s>
Distributed	B-Operating_System
memory	I-Operating_System
is	O
the	O
programming	O
style	O
used	O
on	O
parallel	O
supercomputers	O
from	O
homegrown	O
Beowulf	B-Operating_System
clusters	I-Operating_System
to	O
the	O
largest	O
clusters	O
on	O
the	O
Teragrid	B-Operating_System
.	O
</s>
<s>
On	O
a	O
shared	B-Operating_System
memory	I-Operating_System
machine	O
(	O
a	O
computer	O
with	O
several	O
CPUs	B-Device
that	O
access	O
the	O
same	O
memory	O
space	O
)	O
,	O
messages	O
can	O
be	O
sent	O
by	O
depositing	O
their	O
contents	O
in	O
a	O
shared	B-Operating_System
memory	I-Operating_System
area	O
.	O
</s>
<s>
This	O
is	O
often	O
the	O
most	O
efficient	O
way	O
to	O
program	O
shared	B-Operating_System
memory	I-Operating_System
computers	O
with	O
large	O
number	O
of	O
processors	O
,	O
especially	O
on	O
NUMA	B-Operating_System
machines	O
,	O
where	O
memory	O
is	O
local	O
to	O
processors	O
and	O
accessing	O
memory	O
of	O
another	O
processor	O
takes	O
longer	O
.	O
</s>
<s>
SPMD	B-Operating_System
on	O
a	O
shared	B-Operating_System
memory	I-Operating_System
machine	O
is	O
usually	O
implemented	O
by	O
standard	O
(	O
heavyweight	O
)	O
processes	B-Operating_System
.	O
</s>
<s>
Unlike	O
SPMD	B-Operating_System
,	O
shared	B-Operating_System
memory	I-Operating_System
multiprocessing	B-Operating_System
(	O
both	O
symmetric	B-Operating_System
multiprocessing	I-Operating_System
,	O
SMP	O
,	O
and	O
non-uniform	B-Operating_System
memory	I-Operating_System
access	I-Operating_System
,	O
NUMA	B-Operating_System
)	O
presents	O
the	O
programmer	O
with	O
a	O
common	O
memory	O
space	O
and	O
the	O
possibility	O
to	O
parallelize	O
execution	O
by	O
having	O
the	O
program	O
take	O
different	O
paths	O
on	O
different	O
processors	O
.	O
</s>
<s>
The	O
current	O
standard	O
interface	O
for	O
shared	B-Operating_System
memory	I-Operating_System
multiprocessing	B-Operating_System
is	O
OpenMP	B-Application
.	O
</s>
<s>
It	O
is	O
usually	O
implemented	O
by	O
lightweight	O
processes	B-Operating_System
,	O
called	O
threads	B-Operating_System
.	O
</s>
<s>
A	O
distributed	B-Operating_System
memory	I-Operating_System
program	O
using	O
MPI	B-Application
may	O
run	O
on	O
a	O
collection	O
of	O
nodes	O
.	O
</s>
<s>
Each	O
node	O
may	O
be	O
a	O
shared	B-Operating_System
memory	I-Operating_System
computer	O
and	O
execute	O
in	O
parallel	O
on	O
multiple	O
CPUs	B-Device
using	O
OpenMP	B-Application
.	O
</s>
<s>
Within	O
each	O
CPU	B-Device
,	O
SIMD	B-Device
vector	O
instructions	O
(	O
usually	O
generated	O
automatically	O
by	O
the	O
compiler	O
)	O
and	O
superscalar	B-General_Concept
instruction	O
execution	O
(	O
usually	O
handled	O
transparently	O
by	O
the	O
CPU	B-Device
itself	O
)	O
,	O
such	O
as	O
pipelining	B-General_Concept
and	O
the	O
use	O
of	O
multiple	O
parallel	O
functional	O
units	O
,	O
are	O
used	O
for	O
maximum	O
single	O
CPU	B-Device
speed	O
.	O
</s>
<s>
SPMD	B-Operating_System
was	O
proposed	O
first	O
in	O
1983	O
by	O
Michel	O
Auguin	O
(	O
University	O
of	O
Nice	O
Sophia-Antipolis	O
)	O
and	O
François	O
Larbey	O
(	O
Thomson/Sintra	O
)	O
in	O
the	O
OPSILA	O
parallel	B-Operating_System
computer	I-Operating_System
and	O
next	O
in	O
1984	O
by	O
Frederica	O
Darema	O
at	O
IBM	O
for	O
highly	O
parallel	B-Operating_System
machines	I-Operating_System
like	O
the	O
RP3	O
(	O
the	O
IBM	O
Research	O
Parallel	B-Operating_System
Processor	I-Operating_System
Prototype	O
)	O
,	O
in	O
an	O
unpublished	O
IBM	O
memo	O
.	O
</s>
<s>
By	O
the	O
late	O
1980s	O
,	O
there	O
were	O
many	O
distributed	O
computers	O
with	O
proprietary	O
message	B-Architecture
passing	I-Architecture
libraries	O
.	O
</s>
<s>
The	O
first	O
SPMD	B-Operating_System
standard	O
was	O
PVM	B-Operating_System
.	O
</s>
<s>
The	O
current	O
de	O
facto	O
standard	O
is	O
MPI	B-Application
.	O
</s>
<s>
The	O
Cray	O
parallel	O
directives	O
were	O
a	O
direct	O
predecessor	O
of	O
OpenMP	B-Application
.	O
</s>
