An evaluation of code generation of dataflow languages on manycore architectures

Abstract

Today computer architectures are shifting from single core to manycores due to several reasons such as performance demands, power and heat limitations. However, shifting to manycores results in additional complexities, especially with regard to efficient development of applications. Hence there is a need to raise the abstraction level of development techniques for the manycores while exposing the inherent parallelism in the applications. One promising class of programming languages is dataflow languages and in this paper we evaluate and optimize the code generation for one such language, CAL. We have also developed a communication library to support the intercore communication. The code generation can target multiple architectures, but the results presented in this paper is focused on Adapteva's many core architecture Epiphany. We use the twodimensional inverse discrete cosine transform (2D-IDCT) as our benchmark and compare our code generation from CAL with a hand-written implementation developed in C. Several optimizations in the code generation as well as in the communication library are described, and we have observed that the most critical optimization is reducing the number of external memory accesses. Combining all optimizations we have been able to reduce the difference in execution time between auto-generated and handwritten implementations from a factor of 4.3x down to a factor of only 1.3x.

Dataflow Implementation of QR Decomposition on a Manycore

“…However, as the matrix size increases, the effect of communication decreases. In our previous works [17,14] we have experienced that the external memory access can be a bottleneck in the Epiphany architecture due to the slow link between the memory and the processing cores. Therefore we tested the implementations with and without using the external memory for storing the input matrix.…”

Threaded MPI programming model for the Epiphany RISC array processor

“…Additional APIs and frameworks that have been investigated include OpenCL within Erlang (based on OpenCL support in the COPRTHR SDK) [20], an APL to C compiler [21], and a CAL compilation framework [22]. Automatically generated code from the dataflow language CAL is compared with a C implementation of the 2D inverse discrete cosine transform targeting the Epiphany architecture in [23]. The development of OpenMP support has also been reported [2].…”

Kickstarting high-performance energy-efficient manycore architectures with Epiphany

“…We have also evaluated our approach of compiling CAL to parallel C code with support for message passing being implemented as a stand-alone library by using 1D-DCT algorithm [17]. Our preliminary results reveal that the handwritten implementation has only 1.3x better throughput performance with respect to the auto-generated implementation, which is quite competitive.…”