Program

Workshop program – August 25th 2026, Pisa, Italy

25 August 2026

The HeteroPar 2026 technical program has been organized to guide participants through the major challenges and emerging directions in heterogeneous computing. The morning session begins with advances in programming models and software portability, highlighting AI-assisted code translation, portable SYCL applications, and standardized C++ parallel programming. The second session shifts the focus toward understanding and optimizing heterogeneous hardware through GPU microarchitectural characterization and energy-aware CPU--GPU power management. In the afternoon, the program broadens to system-level challenges, covering resource scheduling, distributed FPGA infrastructures, and portable frameworks for emerging computing paradigms. The workshop concludes with application-driven research demonstrating the use of novel heterogeneous accelerators for artificial intelligence and scientific computing. This progression reflects the heterogeneous computing stack, moving from software development and programming abstractions, through performance analysis and resource management, to cutting-edge applications that showcase the capabilities of next-generation heterogeneous platforms.

Session 5A: Workshop HeteroPar 1 -- Room B

Opening notes and Keynote presentation

09:00 - 09:05

Opening note

Prof. Andrea Bartolini, Program Chair

Short introduction of the workshop
09:05 - 10:30

Keynote: Accelerated computing for the masses: programming a (NVIDIA) GPU today is nothing like 10 years ago!

Filippo Spiga, NVIDIA

ABSTRACT. There was a time when GPUs were considered exotic and programming them was considered difficult. Fast forward a decade, advance in programming models helped democratize Accelerated Computing. The concept of “accelerator” itself has evolved, it is not longer just a PCIe card attached to a server. It is now often tightly coupled with a host CPU and the architecture spawns from one node to full rack-scale solutions.

In this talk, I present programming model approaches to improve the developer productivity by helping them exploiting the performance potential of modern tightly-coupled NVIDIA CPU+GPU systems. I will also highlight trends in modern and novel programming models approaches (e.g cuTILE) and languages within the NVIDIA CUDA ecosystem to support more efficient and agile use of latest hardware capabilities. To conclude, a glimpse of the role of AI and agentic tools in support of code development and optimization.

Morning Coffee Break

Session 6A: Workshop HeteroPar 2 - Programming Models and Performance Portability

This session explores modern approaches to programming heterogeneous systems, from AI-assisted code migration and performance-portable scientific applications to standardized C++ parallel programming models, highlighting techniques that simplify software portability across diverse architectures.

11:00 - 11:30

Execution-Guided LLM Translation of Parallel Programs: A CUDA-to-SYCL Study

Maximilian Hagn, Ruben Laso and Sascha Hunold

CUDA
SYCL
Code Transformation
LLM
11:30 - 12:00

PortLBM: A Portable Lattice Boltzmann Tool Leveraging SYCL on AMD, NVIDIA, and Intel GPUs

Alexander Strack, Marcel Graf, Alexander Van Craen and Dirk Pflüger

Lattice Boltzmann
SYCL
Portability
Heterogeneous Systems
GPU Programming
Real-time Visualization
12:00 - 12:30

pSTL-Bench on GPUs: Evaluating ISO C++ Parallel Algorithms across SYCL and CUDA

Saleh Jamali Golzar, Biagio Cosenza, Siegfried Benkner, Sascha Hunold and Ruben Laso

Performance portability
GPU
Benchmarking
C++
Heterogeneous computing
12:30 - 13:00

Microbenchmarking Modern AMD GPUs

Juan Ferrand, Ernesto Dufrechou and Pablo Ezzatti

Microbenchmarking
Memory
Divergence
AMD GPUs

Lunch Break

Session 7A: Workshop HeteroPar 3 – Resource Management and Emerging Heterogeneous Platforms

This session addresses system-level challenges in heterogeneous computing, including online GPU resource allocation, distributed FPGA development, and portable frameworks for emerging computing paradigms, illustrating how heterogeneous resources can be efficiently managed and exploited.

14:00 - 14:30

Energy-Performance Analysis of CPU-GPU Workloads Under Power Capping

Oksana Diakun, Paweł Czarnul and Jerzy Proficz

Power Capping
Heterogeneous Systems
Energy-Aware Computing
Green Computing
14:30 - 15:00

RCC-Preserving Online Resource Allocation for Heterogeneous GPU Sharing with Finite Availability Windows

Gennai Yuki, Yuichi Ohsita and Hideyuki Shimonishi

Heterogeneous GPU Scheduling
Online Resource Allocation
Finite Availability Windows
RCC-Preserving
15:00 - 15:30

Accelerating HLS-based FPGA Development in Distributed Heterogeneous Systems

Philipp Gündisch, Kenan Gündogan, Philipp Holzinger and Dietmar Fey

Distributed computing
Reconfigurable devices
Decentralized applications
Edge computing

Afternoon Coffee Break

Session 8A: Worksjop HeteroPar 4 – AI and Scientific Applications on Next-Generation Accelerators

The final session showcases innovative applications of heterogeneous computing, presenting novel accelerator architectures and GPU Tensor Core techniques that enable efficient execution of advanced AI workloads and scientific data analysis.

16:00 - 16:30

SYCL-HD: Parallel Framework for Hyperdimensional Computing

José Caires, Aleksandar Ilic and Leonel Sousa

Hyperdimensional Computing
SYCL
Graphics Processing Unit
16:30 - 17:00

Parameter Savings Are Not Memory Savings: Butterfly Attention on a Tile Processor

Philipp Hematty, S.-Kazem Shekofteh and Holger Fröning

Butterfly factorization
Attention
IPU
Graph compilation
Structured sparsity
Long context
17:00 - 17:30

Latent Motifs Discovery via Matrix-Accelerated IMF Modeling using GPU Tensor Cores

Hamid Moghadaspour, Ricardo Nobre, Nuno Neves and Gabriel Falcao

High-Performance GPU Computing
Empirical Mode Decomposition
GRACE/GRACE-FO Range-Rate Residuals