12.01.2014 Views

Curriculum vitae (pdf, en, 50 KB, 10/23/13)

Curriculum vitae (pdf, en, 50 KB, 10/23/13)

Curriculum vitae (pdf, en, 50 KB, 10/23/13)

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

creating local schedulers, associated with differ<strong>en</strong>t execution cores and managing a subset of the C++ functions.<br />

A c<strong>en</strong>tralized scheduler is in charge of synchronizing local schedulers [IC.11].<br />

GP-GPUs offer a massive level of parallelism that can be exploited to speed up computation. In detail,<br />

discrete GP-GPUs can be used to run code that simulates the starting system, while integrated GP-GPUs<br />

(or APUs) can be exploited as hardware accelerators for running software code. Sara Vinco proposed two<br />

flows, targeting each of such sc<strong>en</strong>arios. SAGA proposed a flow to simulate RTL HDL code on GP-GPUs.<br />

RTL processes are partitioned in dataflows according to inter-process dep<strong>en</strong>d<strong>en</strong>cies, and such dataflows are<br />

th<strong>en</strong> associated with differ<strong>en</strong>t cores. Multiple copies of the same dataflow can be run in parallel to increase<br />

throughput and to run more instancies at one time [IC.12], [IC.17], [IC.14]. G<strong>en</strong>erating code for APUs allows<br />

to modify the internal structure of the code, in order to get more performing code. To this ext<strong>en</strong>t, [IC.19]<br />

proposed an approach to exploit both inter-process dep<strong>en</strong>d<strong>en</strong>cies and pipeline structure of the starting designs<br />

to achieve massively parallel execution in software on APUs.<br />

The code g<strong>en</strong>eration flows have be<strong>en</strong> accompanied by the implem<strong>en</strong>tation of an effici<strong>en</strong>t data type library,<br />

reproducing hardware data types (e.g., multi-value logic) in a very effici<strong>en</strong>t and performing way. This allows<br />

to preserve compatibility with respect to the starting code, without slowing down the execution or adding the<br />

overhead of hardware libraries (e.g., SystemC) [JR.2], [IC.7].<br />

7

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!