image

RNDr. Jiří Filipovič, Ph.D.

I am the leader of the High Performance Computing research group at the CERIT-SC Centre, Institute of Computer Science, Masaryk University. In my research, I focus on scientific and high performance computing. I develop methods for automatic improvement of code performance for conventional CPUs and accelerators (GPUs, Intel MIC) such is source-to-source compilation and code optimization parameters autotuning. I am also involved in optimization and acceleration of several program tools. Last but not least, I develop simulation methods and tools used in computational chemistry and biology.

Additionally to my research activities, I teach class focused on accelerators programming in CUDA and OpenCL at Faculty of Informatics, Masaryk university and also OpenCL part of HPC class at Faculty of Computer Science, University of Vienna.

Work Experience

  • Present 2016

    Senior Researcher

    Masaryk University, Institute of Computer Science

  • 2017 2016

    Postdoc Researcher

    University of Vienna, Faculty of Computer Science

  • 2015 2013

    Postdoc Researcher

    Masaryk University, Faculty of Informatics

  • 2013 2006

    Researcher

    CESNET

  • 2011 2006

    Specialist

    Masaryk University, National Centre for Biomolecular Research

Education & Training

  • Ph.D. 2013

    Ph.D. in Informatics

    Masaryk University, Faculty of Informatics

  • M.Sc. 2008

    Mgr. (M.Sc. equivalent) in Applied Informatics

    Masaryk University, Faculty of Informatics

  • B.Sc. 2006

    Bc. (B.Sc. equivalent) in Applied Informatics

    Masaryk University, Faculty of Informatics

News

  • October 2021
    Using hardware performance counters to speed up autotuning convergence on GPUs is published in Journal of Parallel and Distributed Computing
    Our paper Using hardware performance counters to speed up autotuning convergence on GPUs has been published in Journal of Parallel and Distributed Computing. The paper introduces novel search method for auto-tuning spaces. The method uses performance counters to understand limitations of cuttently executed implementation and navigate tuning process towards softening the observed limitations.
  • October 2021
    Advances in Xmipp for Cryo–Electron Microscopy: From Xmipp to Scipion is published in Molecules
    Our paper Advances in Xmipp for Cryo–Electron Microscopy: From Xmipp to Scipion has been published in Molecules journal. The paper introduces changes in Xmipp software since 2013, discuss new features, used technologies and outline future research.
  • October 2021
    Approximating deformation fields for the analysis of continuous heterogeneity of biological macromolecules by 3D Zernike polynomials is published in IUCrJ
    Our paper Approximating deformation fields for the analysis of continuous heterogeneity of biological macromolecules by 3D Zernike polynomials has been published in IUCrJ journal. The paper introduces a novel algorithm for analysis of conformational changes of specimen molecules studied by cryo-electron microscopy.
  • May 2021
    Screening of world approved drugs against highly dynamical spike glycoprotein of SARS-CoV-2 using CaverDock and machine learning is published in Computational and Structural Biotechnology Journal
    Our paper Screening of world approved drugs against highly dynamical spike glycoprotein of SARS-CoV-2 using CaverDock and machine learning has been published in Computational and Structural Biotechnology Journal. The paper focus on using our CaverDock software on screening drugs in order to disable opening of SARS-CoV-2 spike glycoprotein.
  • May 2021
    Image Processing in Cryo-Electron Microscopy of Single Particles: The Power of Combining Methods is published in Structural Proteomics
    Our chapter Image Processing in Cryo-Electron Microscopy of Single Particles: The Power of Combining Methods has been published in Structural Proteomics. The chapter focus on combining mathematically different methods for improving resolution of data obtained by cryo-electron microscopy.
  • March 2021
    DeepAlign, a 3D Alignment Method based on Regionalized Deep Learning for Cryo-EM is published in Journal of Structural Biology
    Our paper DeepAlign, a 3D Alignment Method based on Regionalized Deep Learning for Cryo-EM has been published in Journal of Structural Biology. The paper focus on a 3D aligmnemt step in cryo-electron microscopy. It takes advantage of deep neural networks, which classifies particles' regions in 3D structure, improving the overall speed of the alignment.
  • February 2021
    A preprint of paper Using hardware performance counters to speed up autotuning convergence on GPUs is available
    A new preprint of the paper describing a novel method of searching autotuning space has been released. The paper reveals hardware performance counters to accelerate searching process of a tuning space consisting of any source-code tuning parameters. We have also released preprint of related data paper, which can be use to reproduce our result or develop alternative searchers.
  • October 2020
    A preprint of paper Screening of World Approved Drugs against Highly Dynamical Spike Glycoprotein SARS-CoV-2 using CaverDock and Machine Learning is available
    A new preprint of the paper focusing to usage CaverDock 1.1 to screen drugs havin potential to stabilize COVID-19 spike glycoprotein has been released.
  • October 2020
    Kernel Tuning Toolkit 1.3 has been released
    We have released Kernel Tuning Toolkit 1.3. The latest version adds interoperability with external tools via user-provided structures, such as memory objects, queues etc. It also add API for better configuration of searchers.
  • August 2020
    Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps is published in Concurrency and Computation: Practice and Experience
    Our paper Exploiting historical data: Pruning autotuning spaces and estimating the number of tuning steps has been published in Electronics. The paper shows how to use historical data from code optimization parameters autotuning for estimating tuning cost (number of autotuning steps) as well as pruning the autotuning space. The evaluation of proposed method is performed with GPUs and CPUs.
  • June 2020
    FlexAlign: An Accurate and Fast Algorithm for Movie Alignment in Cryo-Electron Microscopy is published in Electronics
    Our paper FlexAlign: An Accurate and Fast Algorithm for Movie Alignment in Cryo-Electron Microscopy has been published in Electronics. The manuscript introduces and evaluates a new algorithm used for movie alignment in cryo-electron microscopy. The algorithm performs a global and elastic local registration of the movie frames using Cross-Correlation and B-spline interpolation for high precision.
  • April 2020
    CaverDock 1.1 has been released
    We have released CaverDock 1.1. The major contribution of the new version is implementation of a new heuristic, which improves parallel scaling, improves upper-bound trajectory computation and allows to generate multiple ligand's trajectories. CaverDock 1.1 also contains multiple improvement, e.g., better control of ligand position at the active site or automatic adding flexible residues in receptor.
  • February 2020
    Kernel Tuning Toolkit 1.2 has been released
    We have released Kernel Tuning Toolkit 1.2. The latest version improves profiling support (for both legacy and current profiling APIs for AMD and NVIDIA). It also improves support of Vulcan API.
  • February 2020
    A Benchmark Set of Highly-efficient CUDA and OpenCL Kernels and its Dynamic Autotuning with Kernel Tuning Toolkit is published in Future Generation Computer Systems
    Our paper A Benchmark Set of Highly-efficient CUDA and OpenCL Kernels and its Dynamic Autotuning with Kernel Tuning Toolkit has been published in Future Generation Computer Systems. The manuscript introduces and evaluates a comprehensive set of benchmarks for autotuning of code optimization parameters. Moreover, it introduces dynamic autotuning feature implemented in Kernel Tuning Toolkit and experimentally shows that many of our benchmarks can be autotuned during runtime with acceptable overhead.
  • October 2019
    A preprint of the paper A Benchmark Set of Highly-efficient CUDA and OpenCL Kernels and its Dynamic Autotuning with Kernel Tuning Toolkit is available
    We have prepared a preprint A Benchmark Set of Highly-efficient CUDA and OpenCL Kernels and its Dynamic Autotuning with Kernel Tuning Toolkit. The manuscript introduces and evaluates a comprehensive set of benchmarks for autotuning of code optimization parameters. Moreover, it introduces dynamic autotuning feature implemented in Kernel Tuning Toolkit and experimentally shows that many of our benchmarks can be autotuned during runtime with acceptable overhead.
  • October 2019
    Fast Screening of Inhibitor Binding/Unbinding Using Novel Software Tool CaverDock is published in Frontiers in Chemistry
    Our paper Fast Screening of Inhibitor Binding/Unbinding Using Novel Software Tool CaverDock is published in Frontiers in Chemistry. The paper shows how our tool CaverDock can be used for screening databases of drugs against pharmacologically relevant targets.
  • July 2019
    Exploiting historical data: pruning autotuning spaces and estimating the number of tuning steps will be presented at HeteroPar 2019
    Our paper "Exploiting historical data: pruning autotuning spaces and estimating the number of tuning steps" has been accepted and will be presented on HeteroPar 2019 workshop (co-located with Euro-Par 2019 conference).
  • May 2019
    Caver Web 1.0: identification of tunnels and channels in proteins and analysis of ligand transport is published in Nucleid Acids Research
    Our paper Caver Web 1.0: identification of tunnels and channels in proteins and analysis of ligand transport has been published in Nucleid Acids Research. The paper introduces Caver Web 1.0, a service for comprehensive analysis of protein tunnels and channels, and study of the ligands’ transport through these transport pathways. It provides easy-to-use web-based interface allowing calculations with Caver 3.02 and CaverDock 1.0.
  • May 2019
    CaverDock: A Molecular Docking-Based Tool to Analyse Ligand Transport through Protein Tunnels and Channels is published in Bioinformatics
    Our paper CaverDock: A Molecular Docking-Based Tool to Analyse Ligand Transport through Protein Tunnels and Channels has been published in Bioinformatics. The paper extensively evaluates CaverDock and is therefore complementar to IEEE/TCBB paper. CaverDock implements a novel method for the analysis of transport processes in proteins.
  • April 2019
    Kernel Tuning Toolkit 1.1 has been released
    We have released Kernel Tuning Toolkit 1.1. The latest version newly supports Vulcan API, together with already supported OpenCL and CUDA. It is also capable of profiling CUDA kernels, so a developer has detailed feedback on tuning parameters effect.
  • March 2019
    CaverDock: A Novel Method for the Fast Analysis of Ligand Transport is published in IEEE/ACM TCBB
    Our paper CaverDock: A Novel Method for the Fast Analysis of Ligand Transport has been published in IEEE/ACM Transactions on Computational Biology and Bioinformatics. The paper introduces a novel method for the analysis of transport processes in proteins and its implementation called CaverDock. Our method uses a restrained molecular docking, allowing it to compute energy of transport processes much faster than molecular dynamic-based methods. CaverDock have broad applications in the fields of computational enzymology, drug design and protein engineering.
  • March 2019
    GPU acceleration of 3D Fourier reconstruction is published in Sage IJHPCA
    Our paper A GPU acceleration of 3-D Fourier reconstruction in cryo-EM has been published in The International Journal of High Performance Computing Applications. The paper introduces a novel GPU-friendly algorithm for 3D Fourier reconstruction (a computationaly-demanding part of reconstruction pipeline in cryo-electron microscopy). We have used our Kernel Tuning Toolkit to auto-tune the code for various GPU architectures. The code presented in the paper is integrated in a Xmipp 3.19.
  • October 2018
    Autotuning of cuFFT setup is to be presented at ANDARE 2018
    Our paper Performance analysis and autotuning setup of the cuFFT library is to be presented on ANDARE workshop (part of PACT conference). The paper evaluates the performance of cuFFT library and introduces a method to speed it up by introducing small changes in input data size. The cuFFTAdvisor tool, introduced in the paper, is available here.
  • September 2018
    Preprint discussing the implementation of CaverDock is available
    We have published a preprint of our paper "CaverDock: A Novel Method for the Fast Analysis of Ligand Transport", describing an implementation of the novel method for analysis of tunnels in proteins.
  • July 2018
    Kernel Tuning Toolkit 1.0 has been released
    We have released Kernel Tuning Toolkit 1.0. Version 1.0 is capable of performing dynamic tuning (i.e. just in time recompilation of tunable kernels) and global optimizations of sets of kernels, where tuning parameters affect multiple kernels and/or host code.
  • March 2018
    GPU-accelerated 3D Fourier Reconstruction has been integrated into Xmipp
    We have successfully implemented the autotuned and GPU-accelerated version of 3D Fourier Reconstruction. Our implementation is integrated into Xmipp. The paper will follow.).
  • October 2017
    CaverDock 1.0 has been released
    Our tool for evaluation of ligand's binding or unbinding trajectory with molecular docking has been released. It can be downloaded at https://loschmidt.chemi.muni.cz/caverdock/. The papers describing CaverDock functionality and evaluating it are in preparation.
  • August 2017
    Autotuning of OpenCL/CUDA with global kernel optimization will be presented at ANDARE 2017
    Our paper "Autotuning of OpenCL Kernels with Global Optimizations" has been accepted on ANDARE workshop (part of PACT conference).
  • July 2017
    CaverDock will be presented at ISMB/ECCB 2017
    We are currently finishing CaverDock tool for release. I will present poster focused on basic principles of novel transport method analysis implemented in CaverDock as well as preliminary evaluation of the tool at ISMB/ECCB conference.
  • May 2017
    CERIT-SC HPC Research Group has been established
    The CERIT-SC Centre is a Czech national centre operating computing and storage infrastructure for research and development. The centre cooperates on many interdisciplinary research projects for a couple of years. There are two newly established research groups in the centre, focusing on High Performance Computing and Big Data. Those research groups will focus on general, but still widely applicable research topics, helping infrastructure and their partners to solve challenging research questions requiring huge amount of computation or non-trivial data analysis.
  • January 2017
    This website has been created
    I have finally found some time to create my personal web-site!