Skip to content
Snippets Groups Projects

Corrected names in testSingle.c to reflect part struct change.

Merged James Willis requested to merge intrinsic-vectorisation into master

Merge request reports

Loading
Loading

Activity

Filter activity
  • Approvals
  • Assignees & reviewers
  • Comments (from bots)
  • Comments (from users)
  • Commits & branches
  • Edits
  • Labels
  • Lock status
  • Mentions
  • Merge request status
  • Tracking
  • James Willis mentioned in merge request !204 (merged)

    mentioned in merge request !204 (merged)

  • Still more to do:

    testSPHStep.c(196): error: struct "<unnamed>" has no field "c"
        message("c=%f", p->force.c);

    Make sure you run make check.

  • I corrected that issue, but when I ran make check I got tests failing. After you merged the Gadget2-update-part branch, were you getting all tests to pass for both the Gadget2 and Default SPH scheme?

  • James Willis Added 2 commits:

    Added 2 commits:

    • 2f9c21fe - Renamed variable and corrected memory issue with xparts.
    • cf666beb - Increased tolerance.
  • I've not checked the Default scheme tests before, so if they are failing pass the issue back to the author. Just after successful compilation from you. Note not in today, so about to go offline..

  • Okay, no worries. I got the Gadget2 scheme to pass all tests, but the Default fails testPair and testPairPerturbed, I will speak to Matthieu about it.

  • James, the branch test_125 has all the updated tests working and running with all 3 SPH schemes. You may want to merge that into your branch as it could solve some of the problems you are facing. Also, I am not a big fan of increasing the tolerance unless there is a good physically motivated reason for it.

  • I increased the tolerance for rho_dh due to these errors in the Gadget2 scheme:

    Tolerances read from file
    Checking the first 216 particles.
    Absolute difference larger than tolerance (2.000000e-05) for particle 17, column 9:
        File 1:           a = 4.805696e+01
        File 2:           b = 4.805698e+01
    Difference:       |a-b| = 2.000000e-05
    
    Absolute difference larger than tolerance (2.000000e-05) for particle 27, column 9:
        File 1:           a = 4.832451e+01
        File 2:           b = 4.832453e+01
    Difference:       |a-b| = 2.000000e-05
    
    Absolute difference larger than tolerance (2.000000e-05) for particle 81, column 9:
        File 1:           a = 4.913175e+01
        File 2:           b = 4.913177e+01
    Difference:       |a-b| = 2.000000e-05
    
    Absolute difference larger than tolerance (2.000000e-05) for particle 159, column 9:
        File 1:           a = 4.716535e+01
        File 2:           b = 4.716537e+01
    Difference:       |a-b| = 2.000000e-05
    
    Absolute difference larger than tolerance (2.000000e-05) for particle 179, column 9:
        File 1:           a = 4.802430e+01
        File 2:           b = 4.802432e+01
    Difference:       |a-b| = 2.000000e-05
    
    Absolute difference larger than tolerance (2.000000e-05) for particle 200, column 9:
        File 1:           a = 4.809036e+01
        File 2:           b = 4.809039e+01
    Difference:       |a-b| = 3.000000e-05
    
    FAIL: test27cellsPerturbed.sh

    I will try and merge my latest doself1 branch with master and test_125.

  • Ok. rho_dh is not the most critical quantity but do we know what changed ?

  • Sorry, I just looked again it's wcount. It will be because the Gadget2 scheme will call the vectorised versions of non-sym and sym density interactions now. Whereas before it was just using the serial code.

  • Ok. So if you compile with --disable-vec the old tolerances are still good ?

  • Yep, the old tolerances are fine.

  • @jwillis did you merge the test_125 branch into this one ?

  • No, I merged it into my doself-vectorisation branch.

  • Ok. Let me handle the merges to master then.

  • Reassigned to @matthieu

  • Matthieu Schaller Status changed to merged

    Status changed to merged

  • mentioned in commit 352f3185

Please register or sign in to reply
Loading