<s>
oneAPI	B-Operating_System
Threading	B-Application
Building	I-Application
Blocks	I-Application
(	O
oneTBB	O
;	O
formerly	O
Threading	B-Application
Building	I-Application
Blocks	I-Application
or	O
TBB	O
)	O
,	O
is	O
a	O
C++	B-Application
template	I-Application
library	B-Library
developed	O
by	O
Intel	O
for	O
parallel	B-Operating_System
programming	I-Operating_System
on	O
multi-core	B-Architecture
processors	I-Architecture
.	O
</s>
<s>
Using	O
TBB	O
,	O
a	O
computation	O
is	O
broken	O
down	O
into	O
tasks	B-Operating_System
that	O
can	O
run	O
in	O
parallel	O
.	O
</s>
<s>
The	O
library	B-Library
manages	O
and	O
schedules	O
threads	B-Operating_System
to	O
execute	O
these	O
tasks	B-Operating_System
.	O
</s>
<s>
A	O
oneTBB	O
program	O
creates	O
,	O
synchronizes	O
,	O
and	O
destroys	O
graphs	O
of	O
dependent	O
tasks	B-Operating_System
according	O
to	O
algorithms	O
,	O
i.e.	O
</s>
<s>
high-level	O
parallel	B-Operating_System
programming	I-Operating_System
paradigms	O
(	O
a.k.a.	O
</s>
<s>
Algorithmic	B-Language
Skeletons	I-Language
)	O
.	O
</s>
<s>
Tasks	B-Operating_System
are	O
then	O
executed	O
respecting	O
graph	O
dependencies	O
.	O
</s>
<s>
This	O
approach	O
groups	O
TBB	O
in	O
a	O
family	O
of	O
techniques	O
for	O
parallel	B-Operating_System
programming	I-Operating_System
aiming	O
to	O
decouple	O
the	O
programming	O
from	O
the	O
particulars	O
of	O
the	O
underlying	O
machine	O
.	O
</s>
<s>
oneTBB	O
implements	O
work	B-Operating_System
stealing	I-Operating_System
to	O
balance	O
a	O
parallel	O
workload	O
across	O
available	O
processing	O
cores	O
in	O
order	O
to	O
increase	O
core	O
utilization	O
and	O
therefore	O
scaling	O
.	O
</s>
<s>
This	O
dynamic	O
capability	O
decouples	O
the	O
programmer	O
from	O
the	O
machine	O
,	O
allowing	O
applications	O
written	O
using	O
the	O
library	B-Library
to	O
scale	O
to	O
utilize	O
the	O
available	O
processing	O
cores	O
with	O
no	O
changes	O
to	O
the	O
source	O
code	O
or	O
the	O
executable	O
program	O
file	O
.	O
</s>
<s>
In	O
a	O
2008	O
assessment	O
of	O
the	O
work	B-Operating_System
stealing	I-Operating_System
implementation	O
in	O
TBB	O
,	O
researchers	O
from	O
Princeton	O
University	O
found	O
that	O
it	O
was	O
suboptimal	O
for	O
large	O
numbers	O
of	O
processors	O
cores	O
,	O
causing	O
up	O
to	O
47%	O
of	O
computing	O
time	O
spent	O
in	O
scheduling	O
overhead	O
when	O
running	O
certain	O
benchmarks	O
on	O
a	O
32-core	O
system	O
.	O
</s>
<s>
oneTBB	O
,	O
like	O
the	O
STL	B-Application
(	O
and	O
the	O
part	O
of	O
the	O
C++	B-Language
standard	O
library	B-Library
based	O
on	O
it	O
)	O
,	O
uses	O
templates	B-Application
extensively	O
.	O
</s>
<s>
This	O
has	O
the	O
advantage	O
of	O
low-overhead	O
polymorphism	B-Application
,	O
since	O
templates	B-Application
are	O
a	O
compile-time	O
construct	O
which	O
modern	O
C++	B-Language
compilers	B-Language
can	O
largely	O
optimize	O
away	O
.	O
</s>
<s>
oneTBB	O
is	O
available	O
commercially	O
as	O
a	O
binary	O
distribution	O
with	O
support	O
,	O
and	O
as	O
open-source	B-Application
software	I-Application
in	O
both	O
source	O
and	O
binary	O
forms	O
.	O
</s>
<s>
oneTBB	O
does	O
not	O
provide	O
guarantees	O
of	O
determinism	B-Application
or	O
freedom	O
from	O
data	O
races	O
.	O
</s>
<s>
oneTBB	O
is	O
a	O
collection	O
of	O
components	O
for	O
parallel	B-Operating_System
programming	I-Operating_System
:	O
</s>
