Merge remote-tracking branch 'origin/hongzh/adjoint-parallel'

Merge remote-tracking branch 'origin/barry/update-petscfileretrieve'

Merge remote-tracking branch 'origin/barry/fix-use-objectid-not-object-for-comparison-in-ts'

Merge remote-tracking branch 'origin/barry/fix/even-huger-flaw-in-ts'

Merge remote-tracking branch 'origin/barry/fix-huge-flaw-in-ts'

Merge remote-tracking branch 'origin/barry/feature-ts-monitor-solution'

Make ASCII PC/KSP/SNES/TSView() code and output in a standard style

The code that prints the ASCII view for solvers was occasionally inconsistent:
1) TSView printed the type specific information aft

Make ASCII PC/KSP/SNES/TSView() code and output in a standard style

The code that prints the ASCII view for solvers was occasionally inconsistent:
1) TSView printed the type specific information after the general info while
all other solvers printed it before (right after the name of the type is printed)
2) KSPView consistently printed the name of the subtype on EACH line of the subtypes
output as did a few of the PC and SNES viewers. Since they are all printed indented
directly below the subtype name there is no reason to print this information on each line
3) TSView printed output about number of linear solvers and SNES even for explicit
methods, which is goofy.
4) a few other minor formatting and consistency issues where fix.

Commit-type: style-fix, feature
Time: 16 hours

show more ...

fix problems caused by interface change and TSSetExactFinalTime

stability_9bus examples

TSSetFromOptions: Give subtypes a chance to set the default adapt type

TSMonitorDrawSolutionPhase: Add missing draw pause

Force users to provide Jacp matrix

Users need to make sure that the JacobianP matrix can be multiplied by the mu vectors from right.

Change TSSetCostIntegrand interface

Originally PETSc creates a sequential vec to hold the integral values for users. This is ok for common senarios where there is just one integral to calculate. Whe

Change TSSetCostIntegrand interface

Originally PETSc creates a sequential vec to hold the integral values for users. This is ok for common senarios where there is just one integral to calculate. When there are multiple integrals (e.g. coming from multiple cost functions), one might want to do it in parallel. The new interface allows users to provide their own vec to support parallelization. If the users do not provide one, a sequential vec will be created.

show more ...

Fix TSAdjointMonitorGradient -> TSAdjointMonitorSensi

Commit-type: bug-fix

Change TSAdjointMonitorGradient() to TSAdjointMonitorSensi() to match other usage in TS

Commit-type: style-fix
Funded-by: IMEX
Project: adjoints
Reported-by: Hong Zhang <hongzang@anl.gov>

Do not use pointer values to compare two objects

One of the objects may have been destroyed and then recreated at the same memory
location. PetscObjectIds are unique so compare with them instead.

C

Do not use pointer values to compare two objects

One of the objects may have been destroyed and then recreated at the same memory
location. PetscObjectIds are unique so compare with them instead.

Commit-type: bug-fix
Funded-by: ECP
Time: .2 hours

show more ...

Add simple TSAdjointMonitorGradient() and improve manual pages for possible RK schemes

Commit-type: feature,documentation
Funded-by: IMEX
Project: ajoints
Time: .5 hours

TSAdapt: Set TSADAPTNONE as default adapt type

* TSCreate() set TSADAPTNONE as default adapt type
* TSSetType() set TSADAPTNONE before running the constructor
* Add missing error checking in TSXXXSe

TSAdapt: Set TSADAPTNONE as default adapt type

* TSCreate() set TSADAPTNONE as default adapt type
* TSSetType() set TSADAPTNONE before running the constructor
* Add missing error checking in TSXXXSetType()

show more ...

TSAdapt: Refactor handling of default adapt type

* Let TS subtypes state its preference for adaptivity
* Methods without embedded error estimators prefer TSADAPTNONE
* If not set, use TS_EXACTFINALT

TSAdapt: Refactor handling of default adapt type

* Let TS subtypes state its preference for adaptivity
* Methods without embedded error estimators prefer TSADAPTNONE
* If not set, use TS_EXACTFINALTIME_MATCHSTEP when using adaptivity

* Remove command line option -ts_theta_adapt
* Use `-ts_adapt_type basic` to turn on adaptivity in TSTHETA

* Remove command line option -ts_alpha_adapt
* Remove routine TSAlphaUseAdapt() and TSAlpha2UseAdapt()
* Use `-ts_adapt_type basic` to turn on adaptivity in TSALPHA/TSALPHA2

show more ...

Add physically reasonably dt_min and max for tchem examples
Plot dt as a function of time using log scale
add simple script for running tchem cases

Commit-type: style-fix, feature, example
Funded-by

Add physically reasonably dt_min and max for tchem examples
Plot dt as a function of time using log scale
add simple script for running tchem cases

Commit-type: style-fix, feature, example
Funded-by: IMEX
Project: chemistry

show more ...

Fix bug where user provided only RHSJacobian and uses -snes_mf_operator

trouble because IJacobian now thinks there are separate IJacobian and
RHS matrices. Make TS aware that Jacobian is managed by

Fix bug where user provided only RHSJacobian and uses -snes_mf_operator

trouble because IJacobian now thinks there are separate IJacobian and
RHS matrices. Make TS aware that Jacobian is managed by SNES so that TS should
not adjust the matrix for the timestep

Commit-type: bug-fix
Funded-by: IMEX
Time: 2 hour
Reported-by: Ed Bueler <elbueler@alaska.edu>

show more ...

Fix flaw with TSSetRHSJacobian() and no TSSetIJacobian() where DM provides the Jacobian matrix

For example src/ts/examples/tutorials/advection-diffusion-reaction/ex5.c
TS would create the Jacobian f

Fix flaw with TSSetRHSJacobian() and no TSSetIJacobian() where DM provides the Jacobian matrix

For example src/ts/examples/tutorials/advection-diffusion-reaction/ex5.c
TS would create the Jacobian from the DM in TSGetRHSMats_Private() and then make
a copy of it resulting in unneeded matrix creation and unneeded MatAXPY computations at
each Jacobian computation. MatAXPY() would take up to 80% of time.

The management of Jacobians for the various cases of implicit, explicit, IMEX, with and without RHSJacobian
or IJacobian is excessively convoluted and should be refactored to make the code much simplier and
clearer.

Commit-type: bug-fix
Funded-by: IMEX
Project: IMEX
Time: .7 hours
Reported-by: Hong Zhang <hongzhang@anl.gov>

show more ...

Merge remote-tracking branch 'origin/barry/improve-ts-trajectory-viz'

Fortran: Add error checking for incorrect PETSC_NULL_XXX argument when expecting function

Commit-type: enhancement
Funded-by: ECP
Project: ECP/PETSc
Time: .8 hours
Reported-by: Praveen C <cpraveen@

Fortran: Add error checking for incorrect PETSC_NULL_XXX argument when expecting function

Commit-type: enhancement
Funded-by: ECP
Project: ECP/PETSc
Time: .8 hours
Reported-by: Praveen C <cpraveen@gmail.com>

show more ...

Fix flaw with TSSetRHSJacobian() and no TSSetIJacobian() where DM provides the Jacobian matrix

For example src/ts/examples/tutorials/advection-diffusion-reaction/ex5.c
TS would create the Jacobian f

Fix flaw with TSSetRHSJacobian() and no TSSetIJacobian() where DM provides the Jacobian matrix

For example src/ts/examples/tutorials/advection-diffusion-reaction/ex5.c
TS would create the Jacobian from the DM in TSGetRHSMats_Private() and then make
a copy of it resulting in unneeded matrix creation and unneeded MatAXPY computations at
each Jacobian computation. MatAXPY() would take up to 80% of time.

The management of Jacobians for the various cases of implicit, explicit, IMEX, with and without RHSJacobian
or IJacobian is excessively convoluted and should be refactored to make the code much simplier and
clearer.

Commit-type: bug-fix
Funded-by: IMEX
Project: IMEX
Time: .7 hours
Reported-by: Hong Zhang <hongzhang@anl.gov>

show more ...

A variety of improvements/fixes for saving and accessing trajectories.

Fixed bugs in PetscReadBinaryTrajectory.m
Added PetscBinaryIOTrajectory.py to read in trajectories in Python
Fixed bug in all t

A variety of improvements/fixes for saving and accessing trajectories.

Fixed bugs in PetscReadBinaryTrajectory.m
Added PetscBinaryIOTrajectory.py to read in trajectories in Python
Fixed bug in all the trajectory file saving routines that used PETSC_COMM_WORLD instead of the MPI_Comm associated with the TS object.

Funded-by: IMEX
Project: adjoints
Time: 2 hours
Reported-by: Shashikant M. Aithal <aithal@cels.anl.gov>

show more ...