Translating affine nested-loop programs with dynamic loop bounds into Polyhedral Process Networks

Abstract

The Process Network (PN) is a suitable parallel model of computation (MoC) used to specify embedded streaming applications in a parallel form facilitating the efficient mapping onto embedded parallel execution platforms. Unfortunately, specifying an application using a parallel MoC is very difficult and highly error-prone task. To overcome the associated difficulties, an automated procedure exists for derivation of a specific polyhedral process networks (PPN) from static affine nested loop programs (SANLPs). This procedure is implemented in the pn complier. However, there are many applications, e.g., multimedia applications (MPEG coders/decoders, smart cameras, etc.) that have adaptive and dynamic behavior which can not be expressed as SANLPs. Therefore, in order to handle more dynamic multimedia applications, in this paper we address the important question whether we can relax some of the restrictions of the SANLPs while keeping the ability to perform compile-time analysis and to derive PPNs. Achieving this would significantly extend the range of applications that can be parallelized in an automated way. The main contribution of this paper is a first approach for automated translation of affine nested loops programs with dynamic loop bounds into input-output equivalent polyhedral process networks.

Automatic derivation of polyhedral process networks from while-loop affine programs

“…RELATED WORK The work presented in this paper is a significant extension to previous works [11], [12], [15] on systematic and automated derivation of process networks from affine nested loops programs. Turjan et al [15] proposed an automated derivation of process networks from static affine nested loop programs (SANLPs).…”

“…A for-loop with dynamic loop bounds can be replaced by dynamic if-condition with some modifications as it has been shown in [12]. However, a while-loop cannot be replaced by a for-loop with dynamic bounds.…”

“…In this class of affine nested loops programs, the conditions in if-statements may be dependent on some data that is unknown at compile-time and may change at run-time. Nadezhkin et al [12] further extended the class of WDP programs considered in [11] to programs with dynamic for-loop bounds (Dynloop) from which PPNs can be derived automatically. In contrast to the above mentioned techniques, our approach presented in this paper deals with affine nested loop programs with while-loops that currently cannot be handled by [11], [12], [15] to derive PPNs.…”

“…The number of iterations to converge is unknown at compiletime. Since the SVD algorithm cannot be specified as a static program or a program with dynamic if-conditions considered in [11] or for-loops with dynamic bounds considered in [12], the pn compiler [1] as well as techniques from [11], [12] are unable to handle the program in Figure 1(b). Therefore, in this paper, we propose a solution approach to this problem by introducing a novel procedure for automated translation of affine nested loops programs with while-loops (WLAP) into input-output equivalent PPNs.…”

“…We propose the following three relaxations to SANLP programs: 1) allow dynamic if-conditions; 2) allow for-loops with dynamic bounds; 3) allow while-loops. In [11], [12], the first two relaxations have been considered, i.e., how to translate affine nested-loop programs with dynamic ifconditions and for-loops with dynamic bounds into input-output equivalent PPNs in an automated way. In this paper, we consider the third relaxation, while-loops, which is the most difficult of all these relaxations.…”