Skip to Main content Skip to Navigation

Analysis and Compilation of Parallel Programming Languages

Abstract : Traditional compilation faces numerous challenges with program optimizations for parallel architectures. A particular challenge is the design of proper intermediate languages and representations to enable the application of relevant optimization techniques.Various parallel intermediate representations and languages have been proposed.To overcome this issue, different alternatives are more and more exploitedsuch as empirical autotuning or interactive compilation. Such alernatives require fondamentally different typesof intermediates languages such as transformation meta-languages. In this thesis, we study transformation meta-languages for numerical applications: wa particularly address four questions:(i) How do we introduce domain-specific expressiveness?(ii) How do we rethink their design to enhance their flexibility in composing optimizations paths and generating multiple program variants?(iii) How far can we introduce NUMA (Non-Uniform Memory Access) awareness?(iv) As a new class of meta-languages, how do we formalize their semantics? We answer these questions through the design and semantics of TeML, a tensor optimizations meta-language.
Document type :
Complete list of metadata

Cited literature [135 references]  Display  Hide  Download
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, March 27, 2019 - 1:06:23 PM
Last modification on : Wednesday, November 17, 2021 - 12:31:54 PM
Long-term archiving on: : Friday, June 28, 2019 - 2:08:16 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02081178, version 1


Adilla Susungi. Analysis and Compilation of Parallel Programming Languages. Computation and Language [cs.CL]. Université Paris sciences et lettres, 2018. English. ⟨NNT : 2018PSLEM046⟩. ⟨tel-02081178⟩



Record views


Files downloads