Efficient nested loop pipelining in high level synthesis using polyhedral bubble insertion

Abstract

Loop pipelining is a key transformation in highlevel synthesis tools as it helps maximizing both computational throughput and hardware utilization. Nevertheless, it somewhat looses its efficiency when dealing with small trip-count inner loops, as the pipeline latency overhead quickly limits its efficiency. Even if it is possible to overcome this limitation by pipelining the execution of a whole loop nest, the applicability of nested loop pipelining has so far been limited to a very narrow subset of loops, namely perfectly nested loops with constant bounds. In this work we propose to extend the applicability of nested-loop pipelining to imperfectly nested loops with affine dependencies by leveraging on the so-called polyhedral model. We show how such loop nest can be analyzed, and under certain conditions, how one can modify the source code in order to allow nested loop pipeline to be applied using a method called polyhedral bubble insertion. We also discuss the implementation of our method in a source-to-source compiler specifically targeted at High-Level Synthesis tools.

Runtime dependency analysis for loop pipelining in high-level synthesis

“…The performance comparison of these implementations is summarized in tables Table 3 and Table 4 for LEC-HLS and FPGA-HLS tools respectively. We compare the results obtained using LEC-HLS tool with the results of approach proposed by Morvan et al [MDQ11].…”

“…We refer to this implementation as pipeline-1D. For such kernels, our technique is applied on the coalesced loop [10], which allows pipelining of outer levels of the loopnest in addition to the inner-most loop.…”

“…The first set of kernels exposes partially-static memory dependencies. They are the same as those used by Morvan et al [10] in their work and correspond to loop nest that underwent a loop coalescing transformation. This set includes BBFIR, Matrix-mult, Floyd-Warshall and Jacobi-2D kernels.…”

“…Partially-static dependences consist of accesses that can be statically analyzed but for which the compiler identifies a subset of the loop iteration domain where alias happens (see Appendix C.1). Some of these references may be good candidates for our technique which could be more area efficient than fully static solutions [10].…”

“…Source-to-source (S2S) compilers are well known in the parallel computing community [7,8], but are rarely used in the context of HLS [9,10].…”