Paper

[hyschive]

Tasks

• [ ] Release the code (#8)
• [ ] Work with Matt
• [ ] Where to submit (e.g., ApJS)?

[cindytsai]

Example Papers

ApJS

• yt
• Trident
• Galaxy Cluster Merger Catalog

  JOSS

• ytree
• Enzo

[cindytsai]

Outline of the Paper

• TODO
  • [ ] Check do I miss any citations.
  • [ ] Add Enzo in Section 4.
• Need Discussions:

  • How to show memory consumption in libyt?

    ABSTRACT

1. INTRODUCTION

• Why develop libyt? Motivation.
  • Some other situ-analyze tool sets. (ex: Catalyst)
• Advantage and usage of libyt?
• Introducing libyt:
  • libyt originated in Enzo.
  • Mention libyt has already and can be implemented in simulation codes (ex: Gamer, Enzo).
  • Basic idea of libyt.
  • libyt is built with C++ and Python. It is aimed for all C++ simulation codes that are supported by yt.
  • Example.
  • A time series of very big data sets.
  • Original Python script and Inline script are basically the same. (Section 3)
• Outline of this paper.

2. CODE METHOD

2.1 Big Picture of libyt

2.2 Embedding Python in C++ Application

• Brief description of Python and NumPy API.
• Creation of libyt Python module and its C extended Python methods.
• Loading user's inline Python script.
• How it loads data from C++ to Python.

2.3 Connecting libyt to yt

• View libyt as another yt frontend.
• How libyt inherit other frontend's field definitions and yt's field definitions.

2.4 Supported yt Functionalities

• Supported yt functionalities.
  • Table.

2.5 Parallelism

• Structure of libyt and Python under MPI.
• How libyt is paralleled is heavily rely on yt.
  • Some distribute jobs according to where the grids data are stored. (ex: Projection)
  • Others. (ex: Slice)
• How to deal with getting non-local data and restrictions.
  • RMA MPI
  • C++ class

2.6 Collecting Grid Information

• Collecting hierarchy from each rank.

2.7 Loading Intrinsic Fields

• Definition of in-memory field data.
• Connect in-memory field data to Python.
• Supporting ghost zone.

2.8 Computing Derived Fields

• Definition of derived fields.
• Why do we need this?
  • Some calculations cannot be done in yt. (ex: EoS, #19 )
  • Save memory space. (ex: MHD face-centered to cell-centered.)
• Face-centered data.
• How required data is generated and passed back to Python.

2.9 Supporting Particle Data

• Particle data type may be discrete and is not a contiguous array in memory space.
• How to collect discrete data and pass back to Python.

3. DEMONSTRATION: HOW TO IMPLEMENT LIBYT TO SIMULATION CODE

• A C++ library for yt.
• Pseudo-code or procedure graph of simulation code combined with libyt.
• A list of libyt APIs.
• [x] Display code block.

3.1 Initialization

3.2 Setting yt Parameters

3.3 Setting Code-Specific Parameters

3.4 Setting Fields Information

3.5 Adding Derived Field

3.6 Setting Particles Information

3.7 Setting Local Grids Information

3.8 Committing Your Settings

3.9 Performing Inline-Analysis

• How to use yt_inline and yt_inline_argument
• How to convert your post-processing script to inline script.
• Distinguishing libyt fields and yt defined fields

3.10 Freeing Resource

3.11 Finalization

4. SCIENTIFIC APPLICATIONS

• Gamer

  • 19

  • CCSN simulations
  • Vortex Run.
• Enzo
  • Time Consumption - Number of Nodes graph.
  • Inline vs No-inline
  • Inline vs Post-process (plan B)
  • Memory Consumption - Number of Nodes Graph

    5. DISCUSSION AND FUTURE WORK

• Open Source Software
  • Build test on GitHub Action with Python 3.7, 3.8, 3.9, 3.10.
• Performance
  • Time-Number of Processors Graph.
• Limitations of libyt
  • Description and what causes the limitations.
  • Some yt functionalities still needs additional data when saving figure. Ex: Annotations.
  • Some yt functionalities has asymmetric data IO like volume rendering.
• Support Fortran (ex: Flash).
• Supporting Dask
• Support ParaView for inline-rendering. libyt has the potential to give another way for simulation codes to support inline-analysis in ParaView.

Reference

• bibliography