Tesi etd-02072026-162809 |
Link copiato negli appunti
Tipo di tesi
Tesi di laurea magistrale
Autore
STANGANINI, SIMONE
Indirizzo email
s.stanganini@studenti.unipi.it,stanganinisimone@gmail.com
URN
etd-02072026-162809
Titolo
AMM Scheduler: Design and Implementation of a Runtime Scheduling System for Asymmetric Heterogeneous Computing
Dipartimento
INFORMATICA
Corso di studi
INFORMATICA
Relatori
relatore Danelutto, Marco
Parole chiave
- CUDA
- GEMM
- MatMul
- OpenVINO
- Scheduler
- SYCL
Data inizio appello
27/02/2026
Consultabilità
Tesi non consultabile
Riassunto (Inglese)
Riassunto (Italiano)
This thesis proposes the design and implementation of a scheduler capable of assigning computational tasks (matrix multiplication streams) to various accelerators found on modern CPUs to maximize both utilization and efficiency. The project involved the use and integration of several backends, including CUDA, SYCL, OpenVINO, and OpenBLAS. Finally, the thesis presents experimental results that validate the design choices, obtained on an Intel multi-core architecture equipped with an NVIDIA GPU, as well as Intel GPU and NPU components.
File
| Nome file | Dimensione |
|---|---|
Tesi non consultabile. |
|