The pochoir stencil compiler

Abstract

A stencil computation repeatedly updates each point of a ddimensional grid as a function of itself and its near neighbors. Parallel cache-efficient stencil algorithms based on "trapezoidal decompositions" are known, but most programmers find them difficult to write. The Pochoir stencil compiler allows a programmer to write a simple specification of a stencil in a domain-specific stencil language embedded in C++ which the Pochoir compiler then translates into high-performing Cilk code that employs an efficient parallel cache-oblivious algorithm. Pochoir supports general d-dimensional stencils and handles both periodic and aperiodic boundary conditions in one unified algorithm. The Pochoir system provides a C++ template library that allows the user's stencil specification to be executed directly in C++ without the Pochoir compiler (albeit more slowly), which simplifies user debugging and greatly simplified the implementation of the Pochoir compiler itself. A host of stencil benchmarks run on a modern multicore machine demonstrates that Pochoir outperforms standard parallelloop implementations, typically running 2-10 times faster. The algorithm behind Pochoir improves on prior cache-efficient algorithms on multidimensional grids by making "hyperspace" cuts, which yield asymptotically more parallelism for the same cache efficiency.

Performance Analysis of a High-Level Abstractions-Based Hydrocode on Future Computing Systems

“…These include Firedrake [1], FENiCS [25] and Liszt [14], OP2 [6] for unstructured mesh applications and Paraiso [24], Ypnos [26], Pochoir [30] and SBLOCK [12] for explicit stencil based applications (structured mesh applications).…”

“…Pochoir [30] is a compiler and runtime system for implementing stencil computations on multi-core processors. The main aim of the project is to generate cache efficient multi-threaded CPU code for structured mesh (i.e.…”

ExaStencils: Advanced Stencil-Code Engineering

“…Alone for the domain of stencil computations, there are, e.g., Liszt [13] (or the newer DeLite), Pochoir [39], and PATUS [11]. Each one of these is pursuing specific goals: Liszt adds abstractions to Java to make stencils programming easier, also for unstructured problems; Pochoir employs a divide-andconquer skeleton on top of the parallel C extension Cilk to make stencil computations cache-oblivious; PATUS achieves performance by auto-tuning.…”

Auto-tuning the Matrix Powers Kernel with SEJITS

“…The technology has since been broadly disseminated in the ATLAS package [12]. Auto-tuning libraries include OSKI (sparse linear algebra) [13], SPIRAL (Fast Fourier Transforms) [14], and stencils [15,16], in each case showing large performance improvements over non-autotuned implementations. With the exception of SPIRAL and Pochoir, all of these code generators use ad-hoc Perl or C with simple string replacement, unlike the template and tree manipulation systems provided by SEJITS.…”