[WIP] Tag / debug tape recordings #2343

oleburghardt · 2024-08-15T22:42:53Z

Proposed Changes

Add feature to record a tape for debug purposes (as a tool for AD / discrete adjoint / .. developers), using CoDiPack's codi::RealReverseTag.

Related Work

Supposed to help finding remaining errors in #2317.

PR Checklist

Put an X by all that apply. You can fill this out after submitting the PR. If you have any questions, don't hesitate to ask! We want to help. These are a guide for you to know what the reviewers will be looking for in your contribution.

I am submitting my contribution to the develop branch.
My contribution generates no new compiler warnings (try with --warnlevel=3 when using meson).
My contribution is commented and consistent with SU2 style (https://su2code.github.io/docs_v7/Style-Guide/).
I used the pre-commit hook to prevent dirty commits and used pre-commit run --all to format old commits.
I have added a test case that demonstrates my contribution, if necessary.
I have updated appropriate documentation (Tutorials, Docs Page, config_template.cpp), if necessary.

…l write restart files during runtime (there is a bug, potentially when clearing the tape).

oleburghardt · 2024-08-16T07:26:39Z

SU2_CFD/src/drivers/CDiscAdjMultizoneDriver.cpp

@@ -744,7 +744,7 @@
    if (kind_recording == RECORDING::SOLUTION_VARIABLES) {
      cout << " Objective function                   : " << ObjFunc;
      if (driver_config->GetWrt_AD_Statistics()){
-        cout << " (" << ObjFunc_Index << ")\n";
+        // cout << " (" << ObjFunc_Index << ")\n";


This will be resolved inside CoDiPack :)

pcarruscag

Looks good, can you document how it would be used?

…zation.hpp (to cover RealReverseTag).

oleburghardt · 2024-08-25T22:44:06Z

Looks good, can you document how it would be used?

Sure. Added an option so that you can compile the AD build using the tag type (-Dcodi-tape=Tag) for now. You will have to check out the current develop branch of CoDiPack.
I'm thinking about writing a "SetRecording_Debug" next. Together with config file options and so on... documentation will follow.

oleburghardt · 2024-08-25T22:47:41Z

Common/include/parallelization/vectorization.hpp

  static constexpr bool StoreAsRef = true;

 private:
-  alignas(Size * sizeof(Scalar)) Scalar x_[N];
+   alignas(Size * 32) Scalar x_[N];


Would you be okay with replacing 32 by a function to find the next power of 2? (Requirement of alignas.) Or could this be done any better?

For what type is the result not a power of 2?
If we use a fixed number instead of sizeof, arrays of this type may take more space than necessary.

The reverse tag type is 24 I think. If sizeof(Scalar) is already a power of 2 that function should of course not change it :)

I see, so the tag doesn't have a float/double part? Then it should be ok to round Size * sizeof(Scalar) to the next power of 2.
Or maybe using Size * alignof(Scalar), that may be more generic and achieve the desired result.

Well it does contain at least a double for the primal value (show must go on). I'll check or ask what the remaining bytes are for besides two ints for index and tag. But sizeof(RealReverseTag) = 24.

…code/SU2 into feature_mz_adjoint_for_turbo

…eature_mz_adjoint_for_turbo

…his is a temporary change to avoid false positives when a tag tape is used. They might be unnecessary anyway though.

…te adjoint solver.

…g mode finding :-))

oleburghardt · 2024-09-24T15:59:43Z

These are WIP commits just in order to make the debug mode already available to e.g. @joshkellyjak :-)