Ing. Petr Hodač

Ing. Petr Hodač

Department of Theoretical Computer Science

Constant propagation is a vital optimization in modern compilers. This thesis discusses the design and implementation of a new function attribute and a framework extending the call graph to make constant propagation to OpenMP kernels possible. We show that constant propagation and inter-procedural optimizations can be performed when parallelizing with OpenMP on concrete proof-of-concept examples, though our real-world benchmark results are inconclusive. Further work is required to maximize the framework's potential.

Thesis on DSpace

Ing. Petr Hodač

doc. Mgr. Jan Hubička, Ph.D.

Department of Computer Systems

Modern computing systems introduce execution time variations during bench- marking due to OS orchestration, shared hardware resources, and microar- chitectural behaviors. While extensive literature exists on optimizing systems for performance, throughput, and latency, achieving determinism in general- purpose applications remains a niche area of research. This paper presents a methodology to quantify execution noise and identify its potential sources utilizing kernel-level monitoring capabilities. Ansible is utilized to construct a fully automated testing environment tailored for modern architectures run- ning Linux. This framework manages precise system configurationsincluding compiler versions, optimization flags, and resource isolation methodswhile streamlining the collection of system metrics, test execution, and subsequent data interpretation. Finally, this paper presents the gathered measurements, analyzes the implications of the data, and explores the limits of benchmarking the GCC compiler on the SPEC CPU 2017 suite.

Thesis on DSpace