Polyhedral code optimization in the multicore era
DEIB - Building 20, NECST Lab
June 28th, 2016
9.00 am - 11.30 am
Contact:
Marco Domenico Santambrogio
Research Line:
System architectures
June 28th, 2016
9.00 am - 11.30 am
Contact:
Marco Domenico Santambrogio
Research Line:
System architectures
Abstract
On June 28th, 2016 at 9.00 am a workshop on "Polyhedral code optimization in the multicore era" will be held at NECST Lab.
The evolution of multi-cores and GPGPUs highly increased the programming complexity in both mass-market architectures and high performance computing (HPC) architectures. Therefore, there is a pressing need to make parallel programming techniques and related code optimizations accessible to a larger public. In particular, supercomputing today requires a more global approach, from the design of numerical methods to extensive hardware considerations, in interaction with languages and compilers, to take into account both the complexity of architectures and the needs of their non-expert users.
Research communities in computer science (architecture, compilation) and applied mathematics (numerical simulation) are not always aware of this need. Instead, it is important to favor their interaction and the spread of ideas between the two.
Automatic code optimizations and tools also require a better evaluation of their applicability.
This seminar will cover the main techniques and tools and their applicability to perform program optimizations via polyhedral techniques together with suitable schemes for numerical simulation in the context of HPC.
The workshop will include three talks:
1. Introduction to the polyhedral model: applicability and optimizations, by Marco Rabozzi;
2. Overview of automatic optimization via polyhedral tools, by Giuseppe Natale;
3. Cost and performance models and architectures: current and future trends, by Alberto Scolari.
The evolution of multi-cores and GPGPUs highly increased the programming complexity in both mass-market architectures and high performance computing (HPC) architectures. Therefore, there is a pressing need to make parallel programming techniques and related code optimizations accessible to a larger public. In particular, supercomputing today requires a more global approach, from the design of numerical methods to extensive hardware considerations, in interaction with languages and compilers, to take into account both the complexity of architectures and the needs of their non-expert users.
Research communities in computer science (architecture, compilation) and applied mathematics (numerical simulation) are not always aware of this need. Instead, it is important to favor their interaction and the spread of ideas between the two.
Automatic code optimizations and tools also require a better evaluation of their applicability.
This seminar will cover the main techniques and tools and their applicability to perform program optimizations via polyhedral techniques together with suitable schemes for numerical simulation in the context of HPC.
The workshop will include three talks:
1. Introduction to the polyhedral model: applicability and optimizations, by Marco Rabozzi;
2. Overview of automatic optimization via polyhedral tools, by Giuseppe Natale;
3. Cost and performance models and architectures: current and future trends, by Alberto Scolari.