Welcome Dinner

Welcome

• Alberto Garfagnini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino School Introduction

School Introduction

• Timothy Mattson (Human Learning Group)

Room: Rivellino School Introduction

Environment setup

• Daniele Lattanzio (Istituto Nazionale di Fisica Nucleare)
• Daniele Cesini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Parallel architectures

• Timothy Mattson (Human Learning Group)

Room: Rivellino

Floating-point and vectorization

• Timothy Mattson (Human Learning Group)

Room: Rivellino

Floating-point and vectorization

• Timothy Mattson (Human Learning Group)

Room: Rivellino

Consolidation Room: Rivellino

Students lightning presentations Room: Rivellino

Efficient C++ programming

• Francesco Giacomini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Efficient C++ programming

• Francesco Giacomini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Efficient C++ programming

• Francesco Giacomini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Consolidation Room: Rivellino

Students lightning presentations Room: Rivellino

Efficient C++ programming

• Francesco Giacomini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Efficient C++ programming

• Francesco Giacomini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Efficient C++ programming

• Francesco Giacomini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Consolidation Room: Rivellino

Seminar: "Use and Abuse of Random Numbers"

• Timothy Mattson (Human Learning Group)

Room: Rivellino There are a wide range of problems where we find useful answers by “sampling” a solution space. These answers work if the samples are completely uncorrelated, that is, if they are randomly selected. Generating numbers that appear to be random is easy to do. The problem is it’s easy to generate sets of such numbers that seem to be randomly distributed but actually aren't. In this talk, we’ll discuss how to work with numbers that appear to be random and to do so in ways that aren’t stupid.

Introduction to Parallel Programming with OpenMP

• Timothy Mattson (Human Learning Group)

Room: Rivellino

Introduction to Parallel Programming with OpenMP

• Timothy Mattson (Human Learning Group)

Room: Rivellino

Working with threads - The C++ memory model

• Timothy Mattson (Human Learning Group)

Room: Rivellino

Consolidation Room: Rivellino

Efficient Memory Usage

• Wahid Redjeb

Room: Rivellino

Efficient Memory Usage

• Wahid Redjeb

Room: Rivellino

Debugging

• Francesco Giacomini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Consolidation Room: Rivellino

Efficient C++ programming

• Simone Balducci (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Python Bindings

• Simone Balducci (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Consolidation / Study time Room: Rivellino

Poster session party (upstairs in the Rocca Castle) Room: Sala Calligari

OpenMP and friends: task parallelism

• Timothy Mattson (Human Learning Group)

Room: Rivellino

OpenMP beyond Symmetric Multiprocessors

• Timothy Mattson (Human Learning Group)

Room: Rivellino

GPU Programming

• Simone Balducci (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Consolidation Room: Rivellino

Seminar: Radiation-Induced Faults in Modern HPC Systems

• Filippo Reghenzani (PoliMI)

Room: Rivellino Radiation-Induced Faults in Modern HPC Systems Reliability is an increasingly important concern in high-performance and scientific computing, where large-scale systems integrate complex processors, memories, and accelerators and are expected to execute long-running, computation-intensive applications. Among the possible sources of failures, transient faults induced by energetic particles, such as atmospheric neutrons, can affect hardware components and propagate to software, potentially leading to incorrect results or application failures. In this talk, we will introduce the main mechanisms behind radiation-induced transient faults and discuss their impact on modern HPC systems, with particular attention to emerging challenges in GPU-accelerated platforms. Experimental methodologies based on particle accelerators will be presented as a powerful tool for characterizing hardware sensitivity, identifying vulnerable components, and studying fault propagation up to scientific applications. Finally, the talk will discuss software-based techniques for detecting, tolerating, and mitigating transient faults.

GPU Programming

• Simone Balducci (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

GPU Programming

• Simone Balducci (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Cluster computing with MPI

• Daniele Cesini (Istituto Nazionale di Fisica Nucleare)

Room: Rivellino

Cluster computing with MPI Room: Rivellino

Seminar: High-Performance GPU Computing in R for Bioinformatics

• Gabriele Sales (Università di Padova)

Room: Rivellino The explosion of single-cell measurement techniques has created a computational bottleneck in bioinformatics, where the need for rapid exploration often clashes with the performance limitations of high-level languages. While R is the standard for statistical analysis due to its expressivity and ecosystem, extending R capabilities into the realm of GPU-accelerated computing is still challenging. In this talk, I will examine recent trends in GPU acceleration for R, focusing on the transition from traditional CPU-based implementations to CUDA-accelerated kernels, presenting a case study where an API-equivalent rewrite of a core library resulted in speedups of several orders of magnitude.

Consolidation Room: Rivellino

Consolidation Room: Rivellino

Students feedback Room: Rivellino

Exam Room: Rivellino

Discussion Room: Rivellino