Make Carpet's ASCII output independent of the number of processors

Issue #1706 closed
Erik Schnetter created an issue

See https://bitbucket.org/cactuscode/cactus/pull-request/3/cactus-ignore-empty-lines-in-test-case/diff. In addition to this change, one also needs to disable output of the ghost zones in CarpetIOASCII.

Keyword:

Comments (5)

  1. Ian Hinder
    • changed status to open
    • removed comment

    I have tested that this allows 1, 2, 4, 6, 10 and 11 process tests to all pass with WeylScal4/teukolskyID (after updating to use compact output format and to not output ghost zones). Please merge.

    Is it now guaranteed that the output will be independent of the process decomposition, or does this just work because of the specific process decomposition that Carpet has chosen, e.g. that higher-numbered components will always have higher 1D coordinate values along a line? Or does CarpetIOASCII deliberately output components in increasing order of the coordinate?

    I assume that this will not work for 2D or 3D output.

  2. Erik Schnetter reporter
    • removed comment

    No, it is not guaranteed that the output is independent of the process decomposition. However, it seems that Carpet happens to allocate processes so that components are ordered by their index ranges. If this is always the case, then things are always fine, even for 2d and 3d output.

  3. Log in to comment