Programming tensor cores from an image processing DSL

Tensor Cores (TCUs) are specialized units first introduced by NVIDIA in the Volta microarchitecture in order to accelerate matrix multiplications for deep learning and linear algebra workloads. While these units have proved to be capable of providing significant speedups for specific applications, their programmability remains difficult for the...

Schedule Synthesis for Halide Pipelines on GPUs

The Halide DSL and compiler have enabled high-performance code generation for image processing pipelines targeting heterogeneous architectures through the separation of algorithmic description and optimization schedule. However, automatic schedule generation is currently only possible for multi-core CPU architectures. As a result, expert...

The semantic snake charmer search engine: A tool to facilitate data science in high-tech industry domains

The booming popularity of data science is also affecting high-tech industries. However, since these usually have different core competencies - building cyber-physical systems rather than e.g. machine learning or data mining algorithms - delving into data science by domain experts such as system engineers or architects might be more cumbersome...

Schedule synthesis for halide pipelines through reuse analysis

Efficient code generation for image processing applications continues to pose a challenge in a domain where high performance is often necessary to meet real-time constraints. The inherently complex structure found in most image-processing pipelines, the plethora of transformations that can be applied to optimize the performance of an...

The Semantic Snake Charmer Search Engine: A Tool to Facilitate Data Science in High-tech Industry Domains

The booming popularity of data science is also affecting high-tech industries. However, since these usually have different core competencies — building cyber-physical systems rather than e.g. machine learning or data mining algorithms — delving into data science by domain experts such as system engineers or architects might be more cumbersome...

Loop transformations leveraging hardware prefetching

Memory-bound applications heavily depend on the bandwidth of the system in order to achieve high performance. Improving temporal and/or spatial locality through loop transformations is a common way of mitigating this dependency. However, choosing the right combination of optimizations is not a trivial task, due to the fact that most of them...

Performance prediction for families of data-intensive software applications

Performance is a critical system property of any system, in particular of data-intensive systems, such as image processing systems. We describe a performance engineering method for families of data-intensive systems that is both simple and accurate; the performance of new family members is predicted using models of existing family members. The...

Online scheduling of 2-re-entrant flexible manufacturing systems

Online scheduling of operations is essential to optimize productivity of flexible manufacturing systems (FMSs) where manufacturing requests arrive on the fly. An FMS processes products according to a particular flow through processing stations. This work focusses on online scheduling of re-entrant FMSs with flows using processing stations where...

A Fast Estimator of Performance with Respect to the Design Parameters of Self Re-Entrant Flowshops

Self re-entrant flowshops consist of machines which process jobs several times. They are found in applications like TFT-LCD assembly, LED manufacturing and industrial printing. The structure of a self re-entrant flowshop influences its performance. To get better performance while reducing costs a fast performance estimation method can be used to...

A blueprint for system-level performance modeling of software-intensive embedded systems

Exploration of design alternatives and estimation of their key performance metrics such as latency and energy consumption is essential for making the proper design decisions in the early phases of system development. Often, high-level models of the dynamic behavior of the system are used for the analysis of design alternatives. Our work presents...

Online heuristic for the Multi-Objective Generalized traveling salesman problem

Today’s manufacturing systems are typically complex cyber-physical systems where the physical and control aspects interact with the scheduling decisions. Optimizing such facilities requires ordering jobs and configuring the manufacturing system for each job. This optimization problem can be described as a Multi-Objective Generalized TSP where...

Performance engineering for industrial embedded data-processing systems

Performance is a key aspect of many embedded systems, embedded data processing systems in particular. System performance can typically only be measured in the later stages of system development. To avoid expensive re-work in the final stages of development, it is essential to have accurate performance estimations in the early stages. For this...

A blueprint for system-level performance modeling of software-intensive embedded systems

Exploration of design alternatives and estimation of their key performance metrics such as latency and energy consumption is essential for making the proper design decisions in the early phases of system development. Often, highlevel models of the dynamic behavior of the system are used for the analysis of design alternatives. Our work presents...

Aspects of Adaptive Systems Engineering: A Professional Printing Case

Adaptive systems engineering comprises two individual themes, adaptive systems and systems engineering, and their interaction. In the Octopus project, some challenges that arise from these themes have been addressed in the realm of professional printers. This chapter serves to place these challenges in a common context, which is done along the...

Adaptivity in Professional Printing Systems

There is a constant pressure on developers of embedded systems to simultaneously increase system functionality and to decrease development costs. Aviable way to obtain a better system performance with the same physical hardware is adaptivity: a system should be able to adapt itself to dynamically changing circumstances. The development of...