Leveraging MLIR for Loop Vectorization and GPU Porting of FFT Libraries

被引：0

作者：

He, Yifei ^{[1
]}

Podobas, Artur ^{[1
]}

Markidis, Stefano ^{[1
]}

机构：

[1] KTH Royal Inst Technol, Stockholm, Sweden

来源：

EURO-PAR 2023: PARALLEL PROCESSING WORKSHOPS, PT I, EURO-PAR 2023 | 2024年 / 14351卷

基金：

瑞典研究理事会;

关键词：

FFTc; Automatic Loop Vectorization; GPU Porting; LLVM; MLIR;

D O I：

10.1007/978-3-031-50684-0_16

中图分类号：

TP31 [计算机软件];

学科分类号：

081202 ; 0835 ;

摘要：

FFTc is a Domain-Specific Language (DSL) for designing and generating Fast Fourier Transforms (FFT) libraries. The FFTc uniqueness is that it leverages and extend Multi-Level Intermediate Representation (MLIR) dialects to optimize FFT code generation. In this work, we present FFTc extensions and improvements such as the possibility of using different data layout for complex-value arrays, and sparsification to enable efficient vectorization, and a seamless porting of FFT libraries to GPU systems. We show that, on CPUs, thanks to vectorization, the performance of the FFTc-generated FFT is comparable to performance of FFTW, a state-of-the-art FFT libraries. We also present the initial performance results for FFTc on Nvidia GPUs.

引用

页码：207 / 218

页数：12