Automatic Heterogeneous Runtime Using Signal Processing Domain-Specific and Parallel Patterns

Parallel and signal processing patterns for large-scale radio data applications have been captured with a new domain-specific language (DSL), OptiSDR. The intermediate representations (IR) of the code are optimized at the frontend with the Delite compiler targeted for heterogeneous computing archite...

Full description

Saved in:
Bibliographic Details
Published in:International journal of parallel programming Vol. 53; no. 2; p. 8
Main Authors: Zaidi, Yaseen, Winberg, Simon
Format: Journal Article
Language:English
Published: New York Springer US 01.04.2025
Springer Nature B.V
Subjects:
Computation
Computer architecture
Computer Science
Data compression
Domain specific languages
Language
Linear algebra
Optimization
Processor Architectures
Remote sensing
Run time (computers)
Semantics
Signal processing
Software
Software Engineering/Programming and Operating Systems
Theory of Computation
Compiler
Domain-specific language (DSL)
Heterogeneous computing architecture (HCA)
ISSN:0885-7458, 1573-7640
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Parallel and signal processing patterns for large-scale radio data applications have been captured with a new domain-specific language (DSL), OptiSDR. The intermediate representations (IR) of the code are optimized at the frontend with the Delite compiler targeted for heterogeneous computing architecture (HCA). The design flow begins at the abstract programming model and, at a later stage, addresses tuning of the domain applications to the target HCA comprising GPUs and MCPU, as well as scheduling and optimal memory access. For a variety of signal processing use cases, the framework has shown up to two orders of magnitude higher computational speed than MPI and C++ and comparable performance to architecture-specific CUDA. Additional gains are ease of use, scalable computing, and efficient runtime.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0885-7458
1573-7640
DOI:10.1007/s10766-025-00790-7