HPVM: Performance, Programmability and Retargetability for Heterogeneous Parallel Systems
Scalable Parallel Computing Lab, SPCL @ ETH Zurich via YouTube
Overview
Explore a comprehensive lecture on the Heterogeneous Parallel Virtual Machine (HPVM) project, delivered by Vikram Adve at the SPCL_Bcast #40 event. Delve into the challenges of programming heterogeneous parallel systems and discover how HPVM aims to make these systems more accessible to both expert and non-expert developers. Learn about the project's focus on hardware-agnostic programming and its approach to achieving good performance on diverse hardware targets through compiler optimizations, autotuning, and design space exploration. Examine the innovative method of automatically generating retargetable and high-performance code generators for vector and matrix architectures using AutoLLVM IR. Gain insights into the project's ability to outperform mature production compilers across various benchmarks, demonstrating its potential to revolutionize programming for heterogeneous parallel systems in mobile devices and edge computing products.
Syllabus
[Bcast] HPVM: Performance, Programmability and Retargetability for Heterogeneous Parallel Systems
Taught by
Scalable Parallel Computing Lab, SPCL @ ETH Zurich