Cvode / SUNDIALS 6.0 Support

[GaryM]

New version of SUNDIALS/CVODE out: ​https://github.com/LLNL/sundials/releases

Might be quite a few changes as they've even released a script to help you convert!

Todo:

• Install on a test machine
• Get Chaste Building
• Put in the testing rotation

[GaryM]

See ​https://github.com/MichaelClerx/myokit/pull/823/files for the changes to Myokit to make this work

[kwabenaa]

• Created the CvodeContextManager Singleton to manage a shared SUNContext object required by Sundials 6.0+. A separate Singleton is created for each thread, using thread_local storage.
  • Passed Continuous Test Pack
  • Passed Nightly Test Pack
  • Passed Weekly Test Pack, except TestGenerateSteadyStateCrypt which already fails on develop branch (see #3089)
  • Passed Parallel Test Pack, except TestMonodomainProblemParallel / TestOutputDoesNotDependOnPrintTimestep

        Difference in printing time steps of 0.1 and 0.01 = 0.00325635 mV
        /.../TestMonodomainProblem.hpp:1474: Error: Expected (V_to_compare[0] == V_to_compare[1]) up to 3e-3 (0.0030), found (39.6539 != 39.6507)
        Failed
    

    Full test output: ​https://pastebin.com/Nnu3gwPj

• Modified relevant parts of Codegen boilerplate code.

• Reproducing failed test:
  • Install Sundials 6.0.0
  • Then,

    git clone --recursive --branch ticket_3084 https://github.com/kwabenantim/Chaste.git
    cd ./Chaste/build/
    cmake ../
    
    cd ./codegen_python3_venv
    source bin/activate
    pip uninstall -y chaste_codegen
    git clone --recursive --branch chaste_ticket_3084 https://github.com/kwabenantim/chaste-codegen.git
    cd ./chaste-codegen
    pip install -e .
    
    cd ../../
    make -j4 Parallel
    ctest -VV -R TestMonodomainProblemParallel$
    

Hi jmpf, do you have any thoughts on the parallel test, please?

[jmpf@…]

Replying to kwabenaa:

Hi jmpf, do you have any thoughts on the parallel test, please?

Sorry. I don't think that I was copied on this ticket. (I will be from now on though.)

[GaryM]

I think it's probably just worth changing the tolerance on the test here if it is just altering tolerance from 3e-3 to 3.5e-3. The results can change a tiny bit with sundials versions anyway, so not a huge surprise to me something like that would fail. What do you think Jmpfh?

[jmpf@…]

Replying to GaryM:

I think it's probably just worth changing the tolerance on the test here if it is just altering tolerance from 3e-3 to 3.5e-3. The results can change a tiny bit with sundials versions anyway, so not a huge surprise to me something like that would fail. What do you think Jmpfh?

Yes. Go a bit higher: 4e-3 or 5e-3. I would have committed to branch if there was one.

You see I started investigating this, but got distracted with other stuff. I basically found that the test has such a history of flakiness that it's not worth worrying whether the new cvode is "better" or "worse".

[kwabenaa]

Updating the ticket to reflect that the changes were merged into develop:

​https://github.com/Chaste/Chaste/pull/17

​https://github.com/Chaste/Chaste/pull/18

The tolerance of the parallel test was adjusted from 3e-3 to 4e-3.

The version of chaste_codgen was bumped up to 0.9.10 to include the ​changes made there to support Sundials 6.0.