PPIR: Parallel Pattern Intermediate Representation

Schmitz, Adrian; Miller, Julian; Trumper, L. C.; Müller, Matthias S.

doi:10.1109/hipar54615.2021.00009

Cited by 1 publication

(1 citation statement)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Several pieces of prior art discuss intermediate representations for parallelism, such as Tapir [15] for representing Cilk [49] in LLVM; OpenMPIR [50] for representing OpenMP in LLVM, PPIR [51] for pattern trees, and the MLIR OpenMP Dialect; as well as SDf3 [52] for visually representing concurrency as a control-flow graph. These works primarily focus on the representation for their particular style of parallelism (e.g.…”

Section: B Parallel Portablity/ir and Openmp Optimizationsmentioning

confidence: 99%

High-Performance GPU-to-CPU Transpilation and Optimization via High-Level Parallel Constructs

Moses¹,

Ivanov²,

Domke³

et al. 2022

Preprint

View full text Add to dashboard Cite

While parallelism remains the main source of performance, architectural implementations and programming models change with each new hardware generation, often leading to costly application re-engineering. Most tools for performance portability require manual and costly application porting to yet another programming model.We propose an alternative approach that automatically translates programs written in one programming model (CUDA), into another (CPU threads) based on Polygeist/MLIR. Our approach includes a representation of parallel constructs that allows conventional compiler transformations to apply transparently and without modification and enables parallelism-specific optimizations. We evaluate our framework by transpiling and optimizing the CUDA Rodinia benchmark suite for a multicore CPU and achieve a 76% geomean speedup over handwritten OpenMP code. Further, we show how CUDA kernels from PyTorch can efficiently run and scale on the CPU-only Supercomputer Fugaku without user intervention. Our PyTorch compatibility layer making use of transpiled CUDA PyTorch kernels outperforms the PyTorch CPU native backend by 2.7×.

show abstract

Section: B Parallel Portablity/ir and Openmp Optimizationsmentioning

confidence: 99%