Vectorization for digital signal processors via equality saturation
Векторизация для цифровых сигнальных процессоров с использованием насыщения равенствами
2021-04-11
SCID: 54.1/tgery345
Discuss with AI
auto-vectorizationdata layoutsdigital signal processorsequality saturationsmall linear algebra kernels
Figures from the paper
Abstract (AI)
Applications targeting digital signal processors (DSPs) benefit from fast implementations of small linear algebra kernels. While existing auto-vectorizing compilers are effective at extracting performance from large kernels, they struggle to invent the complex data movements necessary to optimize small kernels. To get the best performance, DSP engineers must hand-write and tune specialized small kernels for a wide spectrum of applications and architectures. We present Diospyros, a search-based compiler that automatically finds efficient vectorizations and data layouts for small linear algebra kernels. Diospyros combines symbolic evaluation and equality saturation to vectorize computations with irregular structure. We show that a collection of Diospyros-compiled kernels outperform implementations from existing DSP libraries by 3.1× on average, that Diospyros can generate kernels that are competitive with expert-tuned code, and that optimizing these small kernels offers end-to-end speedup for a DSP application.
Key Findings
1
Diospyros is a search-based compiler that automatically discovers efficient vectorizations and data layouts for small linear algebra kernels targeting DSPs.
2
Diospyros-compiled kernels outperform implementations from existing DSP libraries by an average of 3.1×.
3
Generated kernels are competitive with expert-tuned implementations, reducing the need for manually specialized DSP kernels.
4
Optimizing small kernels with Diospyros delivers end-to-end speedup for a DSP application.
5
The compiler combines symbolic evaluation with equality saturation to vectorize computations involving irregular data movement and structure.
Research Object
small linear algebra kernels targeting digital signal processors (DSPs)
Research Subject
automatic vectorization and data-layout optimization, including efficient irregular data movements and performance
Publication Details
Publication Date
2021-04-11
Journal
Publisher
ISSN
Access Type
Author Information
Download PDF
Subscribe to digest