Local Adjoints for Simultaneous Preaccumulations with Shared Inputs

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Blühdorn, Johannes, Gauger, Nicolas R.
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908912930783232
author Blühdorn, Johannes
Gauger, Nicolas R.
author_facet Blühdorn, Johannes
Gauger, Nicolas R.
contents In shared-memory parallel automatic differentiation, inputs that are shared among simultaneous thread-local preaccumulations lead to data races if Jacobians are accumulated with a single, shared vector of adjoint variables. In this work, we discuss the benefits and tradeoffs of re-enabling such preaccumulations by a transition to suitable local adjoints. We propose different vector- and map-based approaches for storing local adjoint variables and analyze them with respect to memory consumption, memory allocation, and adjoint variable access times in the context of simultaneous preaccumulations in multiple threads. We implement the approaches in CoDiPack and benchmark them in parallel discrete adjoint computations in the multiphysics simulation suite SU2.
format Preprint
id arxiv_https___arxiv_org_abs_2405_07819
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Local Adjoints for Simultaneous Preaccumulations with Shared Inputs
Blühdorn, Johannes
Gauger, Nicolas R.
Mathematical Software
D.1.3; G.1.4; G.4; J.2
In shared-memory parallel automatic differentiation, inputs that are shared among simultaneous thread-local preaccumulations lead to data races if Jacobians are accumulated with a single, shared vector of adjoint variables. In this work, we discuss the benefits and tradeoffs of re-enabling such preaccumulations by a transition to suitable local adjoints. We propose different vector- and map-based approaches for storing local adjoint variables and analyze them with respect to memory consumption, memory allocation, and adjoint variable access times in the context of simultaneous preaccumulations in multiple threads. We implement the approaches in CoDiPack and benchmark them in parallel discrete adjoint computations in the multiphysics simulation suite SU2.
title Local Adjoints for Simultaneous Preaccumulations with Shared Inputs
topic Mathematical Software
D.1.3; G.1.4; G.4; J.2
url https://arxiv.org/abs/2405.07819