Welcome to Harisankar Sadasivan's Homepage

About

Dr. Harisankar (Hari) Sadasivan is a Member of Technical Staff in the AMD AI Group, a faculty member (part-time) at the University of Washington, Seattle, and a Senior Member of IEEE. Hari’s research focuses on 3 Critical and Emergent Technologies (CET) that are identified as significant for US’s national security- ‘advanced computing’ for ‘artificial intelligence’ and ‘biotechnologies’. At AMD, Hari leads the AMD Center of Excellence in AI at UW Seattle and co-founded the AMD lifesciences group that collaborates with research labs worldwide. Hari received his PhD and Masters in CSE from the University of Michigan Ann Arbor where he was advised by Prof. Satish Narayanasamy. Hari’s PhD research focussed on HW-SW co-design for accelerated and portable long-read DNA sequencing. Hari’s PhD research has won the prestigious 2022 MICRO Top Picks with Honorable Mention (with multiple artifact badges). Hari is a Technical reviewer with several IEEE/ACM venues including the longest, oldest and the most prestigious formats. Hari has previously worked with NVIDIA and Samsung R&D.

Research

I envision a world where AI is advanced and performant enough to diagnose and find cures for all human diseases via techniques such as Precision Medicine. I realize that’s a big jump. So, I have broken down my interests for now.

• High Performance Artificial Intelligence (AI): Tall & Skinny GEMMs, Stream-K, Performance issues on multi-chiplet GPUs, LLM inference optimizations
• High Performance -omics: Long-read DNA alignment, Raw signal alignment, AI-based Basecalling, Metagenomics

Please shoot me an email if you are a potential collaborator/PI.

Teaching

AI’s demand for FLOPs has far outpaced what Moore’s Law could offer. This gap can only be bridged with aggressive software innovation. In order to do this, it is necessary to understand the AI software stack and how it maps to the GPU hardware. I designed and teach a course on AMD GPU Programming at the University of Washington Seattle. I am thankful to all content contributors (acknowledged on my slides). Students impressed me by porting several CUDA kernels to AMD’s portable ROCm HIP framework. We ported Dynamic Time Warping, Tall and Skinny GEMMs, ML-based Minimap2 seedingand explored Stream-K optimizations.

I am currently in the works of developing a new course at the University of Washington Seattle titled, “Matrix to Machines: GPU HW Design on FPGA for AI”. I’m on the lookout for a course development TA in fall. Email me if you are up for it.

Advising

I enjoy guiding motivated students:
-M. Emin Ozturk, University of Utah
-Juechu Dong and Xueshen Liu, U Michigan Ann Arbor

News

• [August. 2024] I was invited to present a lightning talk at the [IEEE New Era World Leaders AI Summit] in Seattle.
• [August. 2024] Our work on improving the work-partitioning for AI on GPUs is out. Here’s a [pre-print]
• [July. 2024] Our paper on attempting to set the direction for genomics acceleration for the next decades is out. Here’s a pre-print.
• [May. 2024] Our paper on GPU accelerated Minimap2 chaining is presented at Biosys, ASPLOS 2024. Here’s an AMD blog post detailing our collaboration.
• [April. 2024] Joined as a part-time course instructor in ECE, UW Seattle to teach GPU Programming. Here’s what my students like about the most popular course at UW-ECE in ‘SP24.
• [Jan. 2024] Our paper on GPU accelerated Dynamic Time Warping is published in the Journal of Biotechnology and Biomedicine.
• [Dec. 2023] Joined as a part-time course instructor at Paul G Allen School of CSE, UW Seattle to teach GPU Programming.

Recent Publications

Welcome to Harisankar Sadasivan's Home