<s>
In	O
computer	B-General_Concept
science	I-General_Concept
,	O
partitioned	B-Application
global	I-Application
address	I-Application
space	I-Application
(	O
PGAS	O
)	O
is	O
a	O
parallel	B-Application
programming	I-Application
model	I-Application
paradigm	O
.	O
</s>
<s>
PGAS	O
is	O
typified	O
by	O
communication	O
operations	O
involving	O
a	O
global	O
memory	O
address	B-General_Concept
space	I-General_Concept
abstraction	O
that	O
is	O
logically	O
partitioned	O
,	O
where	O
a	O
portion	O
is	O
local	O
to	O
each	O
process	O
,	O
thread	O
,	O
or	O
processing	O
element	O
.	O
</s>
<s>
The	O
novelty	O
of	O
PGAS	O
is	O
that	O
the	O
portions	O
of	O
the	O
shared	B-Operating_System
memory	I-Operating_System
space	O
may	O
have	O
an	O
affinity	O
for	O
a	O
particular	O
process	O
,	O
thereby	O
exploiting	O
locality	B-General_Concept
of	I-General_Concept
reference	I-General_Concept
in	O
order	O
to	O
improve	O
performance	O
.	O
</s>
<s>
A	O
PGAS	O
memory	O
model	O
is	O
featured	O
in	O
various	O
parallel	O
programming	O
languages	O
and	O
libraries	O
,	O
including	O
:	O
Coarray	B-Language
Fortran	I-Language
,	O
Unified	B-Operating_System
Parallel	I-Operating_System
C	I-Operating_System
,	O
,	O
Fortress	B-Language
,	O
Chapel	B-Language
,	O
X10	B-Language
,	O
,	O
,	O
Global	B-Library
Arrays	I-Library
,	O
and	O
SHMEM	B-Operating_System
.	O
</s>
<s>
The	O
PGAS	O
paradigm	O
is	O
now	O
an	O
integrated	O
part	O
of	O
the	O
Fortran	B-Application
language	I-Application
,	O
as	O
of	O
Fortran	B-Application
2008	O
which	O
standardized	O
coarrays	O
.	O
</s>
<s>
Many	O
PGAS	O
systems	O
combine	O
the	O
advantages	O
of	O
a	O
SPMD	B-Operating_System
programming	O
style	O
for	O
distributed	O
memory	O
systems	O
(	O
as	O
employed	O
by	O
MPI	B-Application
)	O
with	O
the	O
data	O
referencing	O
semantics	O
of	O
shared	B-Operating_System
memory	I-Operating_System
systems	O
.	O
</s>
<s>
In	O
contrast	O
to	O
message	B-Architecture
passing	I-Architecture
,	O
PGAS	O
programming	O
models	O
frequently	O
offer	O
one-sided	O
communication	O
operations	O
such	O
as	O
Remote	O
Memory	O
Access	O
(	O
RMA	O
)	O
,	O
whereby	O
one	O
processing	O
element	O
may	O
directly	O
access	O
memory	O
with	O
affinity	O
to	O
a	O
different	O
(	O
potentially	O
remote	O
)	O
process	O
,	O
without	O
explicit	O
semantic	O
involvement	O
by	O
the	O
passive	O
target	O
process	O
.	O
</s>
<s>
PGAS	O
offers	O
more	O
efficiency	O
and	O
scalability	O
than	O
traditional	O
shared-memory	B-Operating_System
approaches	O
with	O
a	O
flat	O
address	B-General_Concept
space	I-General_Concept
,	O
because	O
hardware-specific	O
data	B-General_Concept
locality	I-General_Concept
can	O
be	O
explicitly	O
exposed	O
in	O
the	O
semantic	O
partitioning	O
of	O
the	O
address	B-General_Concept
space	I-General_Concept
.	O
</s>
<s>
A	O
variant	O
of	O
the	O
PGAS	O
paradigm	O
,	O
asynchronous	O
partitioned	B-Application
global	I-Application
address	I-Application
space	I-Application
(	O
APGAS	O
)	O
augments	O
the	O
programming	O
model	O
with	O
facilities	O
for	O
both	O
local	O
and	O
remote	O
asynchronous	O
task	O
creation	O
.	O
</s>
<s>
Two	O
programming	O
languages	O
that	O
use	O
this	O
model	O
are	O
Chapel	B-Language
and	O
X10	B-Language
.	O
</s>
<s>
The	O
Adapteva	B-Application
Epiphany	I-Application
architecture	O
is	O
a	O
manycore	B-General_Concept
network	B-Architecture
on	I-Architecture
a	I-Architecture
chip	I-Architecture
processor	O
with	O
scratchpad	B-General_Concept
memory	I-General_Concept
addressable	O
between	O
cores	O
.	O
</s>
