<s>
There	O
are	O
three	O
main	O
components	O
to	O
MADNESS	B-Operating_System
.	O
</s>
<s>
that	O
aims	O
to	O
increases	O
programmer	O
productivity	O
and	O
code	O
performance/scalability	O
while	O
maintaining	O
backward	O
compatibility	O
with	O
current	O
programming	O
tools	O
such	O
as	O
the	O
message-passing	B-Application
interface	I-Application
and	O
Global	B-Library
Arrays	I-Library
.	O
</s>
<s>
The	O
numerical	O
capabilities	O
built	O
upon	O
the	O
parallel	O
tools	O
provide	O
a	O
high-level	O
environment	O
for	O
composing	O
and	O
solving	O
numerical	B-General_Concept
problems	I-General_Concept
in	O
many	O
(	O
1-6	O
+	O
)	O
dimensions	O
.	O
</s>
<s>
It	O
is	O
open-source	B-Application
,	O
has	O
an	O
object-oriented	B-Language
design	O
,	O
and	O
is	O
designed	O
to	O
be	O
a	O
parallel	B-Operating_System
processing	I-Operating_System
program	O
for	O
computers	O
with	O
up	O
to	O
millions	O
of	O
cores	O
running	O
already	O
on	O
the	O
Cray	B-Device
XT5	I-Device
at	O
Oak	O
Ridge	O
National	O
Laboratory	O
and	O
the	O
IBM	B-Operating_System
Blue	I-Operating_System
Gene	I-Operating_System
at	O
Argonne	O
National	O
Laboratory	O
.	O
</s>
<s>
The	O
small	O
matrix	O
multiplication	O
(	O
relative	O
to	O
large	O
,	O
BLAS-optimized	O
matrices	O
)	O
is	O
the	O
primary	O
computational	O
kernel	O
in	O
MADNESS	B-Operating_System
;	O
thus	O
,	O
an	O
efficient	O
implement	O
on	O
modern	O
CPUs	B-Device
is	O
an	O
ongoing	O
research	O
effort	O
.	O
</s>
<s>
Adapting	O
the	O
irregular	O
computation	O
in	O
MADNESS	B-Operating_System
to	O
heterogeneous	O
platforms	O
is	O
nontrivial	O
due	O
to	O
the	O
size	O
of	O
the	O
kernel	O
,	O
which	O
is	O
too	O
small	O
to	O
be	O
offloaded	O
via	O
compiler	O
directives	O
(	O
e.g.	O
</s>
<s>
MADNESS	B-Operating_System
was	O
recognized	O
by	O
the	O
R&D	O
100	O
Awards	O
in	O
2011	O
.	O
</s>
