The school will focus on the theme of Scientific Software for Heterogeneous Architectures. The complete programme will offer 22 hours of lectures and hands-on exercises, and a student presentations session.
-
Introduction lecture
Preparing for the HL-LHC computational challenge
- HEP data processing and analysis workflows
- Upgrades of the LHC accelerator and experiments
- Evolution of hardware and computing infrastructure
- Impact on HEP data processing software
-
Track 1: CPU Architecture and High Performance
4 hours of lectures and 2 hours of hands-on exercises
CPU Hardware Architecture and Evolution- Hardware evolution of the CPU
- Memory hierarchy, caching, NUMA
- Microarchitecture of modern CPUs
Performance Analysis on Modern CPUs
- Performance analysis tools for Linux
- CPU features for performance analysis
- Top-down microarchitecture analysis
Low-level Performance Optimization Guidelines
- Main sources of performance bottlenecks
- Floating point arithmetics performance
- Advanced low-level performance tuning
Data-Oriented Design
- Principles of data-oriented design
- Memory access and data-type profiling
- Data structure performance optimization
-
Track 2: Parallel and Optimised Scientific Software
5 hours of lectures and 2 hours of hands-on exercises
Writing parallel software- Amdahl's and Gustafson's laws
- Asynchronous execution
- Finding concurrency, task vs. data parallelism
- Using threading in C++ and Python, comparison with multi-process
- Resource protection and thread safety
- Locks, thread local storage, atomic operations
Writing efficient software
- virtues of functional programming
- practical usage in C++ and why it's efficient
- how to help the compiler to produce faster code
- doing more at compile time
- Templating versus inheritance, pros and cons of virtual inheritance
Optimizing existing large codebase
- Measuring performance, tools and key indicators
- Improving memory handling
- The nightmare of thread safety
- Code modernization and low level optimizations
- Data structures for efficient computation in modern C++
Practical vectorization
- Measuring vectorization level
- What to expect from vectorization
- Preparing code for vectorization
- Vectorizing techniques in C++: intrinsics, libraries, autovectorization
-
Track 3: Programming for Heterogeneous Architectures
4 hours of lectures and 4 hours of hands-on exercises
Scientific computing on heterogeneous architectures- Introduction to heterogeneous architectures and the performance challenge
- From general to specialized: Hardware accelerators and applications
- Type of workloads ideal for different accelerators
- Trade-offs between multi-core and many-core architectures
- Implications of heterogeneous hardware on the design and architecture of scientific software
- Embarrassingly parallel scientific applications in HPC and CERN
Programming for GPUs
- From SIMD to SPMD, a programming model transition
- Thread and memory organization
- Basic building blocks of a GPU program
- Control flow, synchronization, atomics
Performant programming for GPUs
- Data locality, coalesced memory accesses, tiled data processing
- GPU streams, pipelined memory transfers
- Under the hood: branchless, warps, masked execution
- Debugging and profiling a GPU application
Design patterns and best practices
- Good practices: single precision, floating point rounding, avoid register spilling, prefer single source
- Other standards: SYCL, HIP, OpenCL
- Middleware libraries and cross-architecture compatibility
- Reusable parallel design patterns with real-life applications
-
Additional lectures
Student lightning talks session
