Skip to content

Latest commit

 

History

History
38 lines (29 loc) · 1.61 KB

2023-01-09.md

File metadata and controls

38 lines (29 loc) · 1.61 KB

Binary Sparse Format Working Group Meeting #7 - January 9, 2023

Attendees

  • Ben Brock
  • Tim Davis
  • Jim Kitchen
  • Tim Mattson
  • Scott McMillan
  • Erik Welch
  • Isaac Virshup
  • Will Kimmerer

Agenda

  • Discuss Tim Davis' matrix C struct

  • Discuss Ben's work on HDF5 bindings, as well as issues with HDF5.

    • It is not straightforward to determine the type of an HDF5 dataset.

Summary of Discussion

Tim discussed the C matrix he defined for binsparse. This consists of the following:

  • Matrix dimensions and number of values
  • Types in the matrix
  • Number of axes
  • Values pointer
  • A pointer to multiple axes
    • Each axis contains a "pointer" or "index"

Ben discussed his implementation of C bindings.

  • One major issue is that HDF5 does not have an easy mechanism in C/C++ to determine the type of a dataset.
    • You are limited to individual queries like type class (e.g. integer, float), endianness, and size.
    • Ben implemented a helper function that does some of this work on top of HDF5.
    • To help with this issue, we should list the types in JSON.
    • Users could specify what types to store the matrix to disk with, even if those don't necessarily match the types stored in memory. HDF5 should handle this automatically.

Outcomes