• No results found

Transformations for polyhedral process networks Meijer, S.

N/A
N/A
Protected

Academic year: 2021

Share "Transformations for polyhedral process networks Meijer, S."

Copied!
2
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Transformations for polyhedral process networks

Meijer, S.

Citation

Meijer, S. (2010, December 8). Transformations for polyhedral process networks. Retrieved from https://hdl.handle.net/1887/16221

Version: Corrected Publisher’s Version

License: Licence agreement concerning inclusion of doctoral thesis in the Institutional Repository of the University of Leiden

Downloaded from: https://hdl.handle.net/1887/16221

Note: To cite this publication please use the final published version (if applicable).

(2)

Stellingen

behorende bij het proefschrift

'Transformations for Polyhedral Process Networks' Sjoerd Meijer

1.

The derivation of polyhedral process networks is based on finding exact data dependence relations of program statements in static affine nested loop programs. Thus, in the derivation process, only data dependencies are made explicit and not the inherit parallelism of the application. Determining the degree of parallelism in the derived process networks is a totally different analysis.

2.

Application transformations cannot be applied and explored efficiently and effectively without

considering the application-to-architecture mapping and the communication/synchronization primitives that are provided by the target platform.

3.

The problem of ordering the different transformations, and on which process a particular transformation should be applied, has a very simple but yet powerful solution, i.e., first splitting all processes and subsequently merging different process instances results in perfectly balanced compound processes.

4.

For applications with fine-grained computations and corresponding implementation platforms, the gain of parallelization can easily be canceled by the costs for synchronization/communication.

5.

The commercialization of research work (i.e., valorization) should not be the responsibility of the researcher.

6.

The research field of embedded systems and software is a jungle of different hardware platforms and software tools. It should be unacceptable that experimental results are not reproducible as a result of the huge diversity in platforms/tools and commercial licenses.

7.

A lack of shame increases the chance of selfishness behavior.

8.

Near the end one starts to think about the beginning.

Referenties

GERELATEERDE DOCUMENTEN

License: Licence agreement concerning inclusion of doctoral thesis in the Institutional Repository of the University of Leiden. Downloaded

5 Appling Transformations in Combination 85 5.1 Impact of the Transformation on Performance

The third is a mapping specification describing how the processes of the PPN are as- signed to the processing elements of the hardware platform. The ESPAM tool takes these

It can be seen that process P0 is a source process because it does not read data from other processes, and that process P2 is a sink process because it does not write data to

Note that in this example, the first iterations of the second partition for the diagonal plane-cut and unfolding on the outermost loop i are the same, i.e., iteration (1, 0), but

Then we increase the workload of the producer processes and intentionally create a compound process that is the most compute intensive process. We check if this is captured by

Before introducing our solution in a more formal way, we show how our approach intuitively works for the examples discussed in Section 5.1. We have already shown 3 different

The first two classes of FIFO channels are easy to implement efficiently, as FIFOs from these classes are realized using just local (for producer and consumer processes) memories