Lift: a Functional Data-Parallel IR for High-Performance GPU Code Generation
- Submitting institution
-
University of Glasgow
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 11-04915
- Type
- E - Conference contribution
- DOI
-
10.1109/CGO.2017.7863730
- Title of conference / published proceedings
- 2017 IEEE/ACM International Symposium on Code Generation and Optimization (CGO)
- First page
- 74
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- February
- Year of publication
- 2017
- URL
-
http://eprints.gla.ac.uk/146596/
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
2
- Research group(s)
-
-
- Citation count
- 33
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- ORIGINALITY: Introduces a novel compiler Intermediate Representation (IR) that combines the flexibility to express GPU programs with complex optimizations, with performance equivalent to manually optimized code. SIGNIFICANCE: Has been published at the premier venue for compiler research, and was the highest cited paper of the conference that year. It has since been used for the design of Microsoft’s Brainwave AI accelerator (Senior Researcher) and Google’s MLIR framework (Research Scientist). RIGOUR: Extensive experiments have demonstrated that the proposed IR is capable of expressing highly optimised programs, quantified the effects of such optimisations, and evaluated the performance of diverse applications over it.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -