Define: _OPENMP -> AMREX_USE_OMP #1520
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
ax3l
requested review from
dpgrote,
lucafedeli88,
RemiLehe,
WeiqunZhang and
MaxThevenet
ax3l
commented
Nov 18, 2020
| @@ -22,7 +22,7 @@ PartPerGridFunctor::operator()(amrex::MultiFab& mf_dst, const int dcomp, const i | |||
| // Temporary MultiFab containing number of particles per grid. | |||
| // (stored as constant for all cells in each grid) | |||
| amrex::MultiFab ppg_mf(warpx.boxArray(m_lev), warpx.DistributionMap(m_lev), 1, ng); | |||
| #ifdef _OPENMP | |||
| #ifdef AMREX_USE_OMP | |||
| #pragma omp parallel | |||
There was a problem hiding this comment.
Missing if (amrex::Gpu::notInLaunchRegion())?
There was a problem hiding this comment.
Yes, we should add it if we want to support both in the same build.
There was a problem hiding this comment.
Actually, this is currently running on the host.
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
commented
Nov 18, 2020
ax3l
changed the title
ax3l
commented
Nov 18, 2020
Replace the define check of `_OPENMP` with the explicit backend control of `AMREX_USE_OMP` for parallel constructs. Doing so avoids that we accidentially turn on OpenMP, e.g. if a dependency pulls it in for linear algebra, I/O, etc. This can led to confusion if the user explicitly requested a serial build. Also, we might want to use OpenMP functionality here and there for auxiliary functions w/o having to use the AMReX OpenMP backend, i.e. because we compile for GPUs.
ax3l
force-pushed
the
topic-openmpDefine
branch
from
4599ebb to
0202f23
Compare
ax3l
changed the title
ax3l
commented
Jan 20, 2021
| @@ -51,7 +51,7 @@ PairWiseCoulombCollision::doCollisions (amrex::Real cur_time, MultiParticleConta | |||
| for (int lev = 0; lev <= species1.finestLevel(); ++lev){ | |||
|
|
|||
| // Loop over all grids/tiles at this level | |||
| #ifdef _OPENMP | |||
| #ifdef AMREX_USE_OMP | |||
| info.SetDynamic(true); | |||
There was a problem hiding this comment.
This looks odd here, I think we set this generally now as an input option.
There was a problem hiding this comment.
I agree with you that it should not be set like this...
ax3l
commented
Jan 20, 2021
| @@ -322,7 +322,7 @@ protected: | |||
| info.EnableTiling(pc_src.tile_size); | |||
| } | |||
|
|
|||
| #ifdef _OPENMP | |||
| #ifdef AMREX_USE_OMP | |||
| info.SetDynamic(true); | |||
There was a problem hiding this comment.
Do we still SetDynamic manually in code or generally as an input option?
There was a problem hiding this comment.
I mean warpx.do_dynamic_scheduling = 1, is that it?
WarpX/Source/Particles/WarpXParticleContainer.cpp
Lines 29 to 39 in 1f3191f
There was a problem hiding this comment.
Discussed on slack: those are set for particle loops generally since we know that those have computational inbalances.
We don't set this generally atm for MFIter loops, but should independently of this PR review if the three locations where we set this for MFIter loops are still up-to-date or if we want to remove those (or add to more locations). Dynamic scheduling comes with (some) overheads and we use atomics to get the next available work item.
ax3l
commented
Jan 20, 2021
| @@ -586,7 +586,7 @@ PhysicalParticleContainer::AddPlasma (int lev, RealBox part_realbox) | |||
| if (do_tiling && Gpu::notInLaunchRegion()) { | |||
| info.EnableTiling(tile_size); | |||
| } | |||
| #ifdef _OPENMP | |||
| #ifdef AMREX_USE_OMP | |||
| info.SetDynamic(true); | |||
There was a problem hiding this comment.
Do we still SetDynamic manually in code or generally as an input option?
There was a problem hiding this comment.
offline discussed: see above
ax3l
commented
Jan 20, 2021
ax3l
commented
Jan 20, 2021
ax3l
commented
Jan 20, 2021
ax3l
commented
Jan 20, 2021
ax3l
commented
Jan 20, 2021
ax3l
commented
Jan 20, 2021
ax3l
commented
Jan 20, 2021
ax3l
commented
Jan 20, 2021
lucafedeli88
approved these changes
Jan 20, 2021
RemiLehe
approved these changes
Jan 21, 2021
mrowan137
pushed a commit
to mrowan137/WarpX
that referenced
this pull request
Feb 2, 2021
…rpX#1624) * Fix bug in momentum-conserving interpolation function * Reset checksum benchmark for CI test momentum-conserving-gather Unused Params: Check after Step 0 (ECP-WarpX#1596) An early check for unused parameters after the first step. Add Superparticle version of getParticlePosition. (ECP-WarpX#1640) * Add Superparticle version of GetParticlePosition. * move unpack_particle to a free function and rename * no longer need SuperPType * Update Source/Particles/Pusher/GetAndSetPosition.H * remove templating from get_particle_position * remove template * Add missing include Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Distribution mapping and cost plotting (ECP-WarpX#1444) * Distribution mapping and cost plotting Cleanup Cleanup Cleanup EOL Unused import * Cost initializes to 0.0 * plot slices of 3D * WIP * WIP docs * docs * docs * docs * docs * docs * docs * docs * docs * EOL SpeciesProperties: C, N, O, Cu (ECP-WarpX#1638) * SpeciesProperties: C, N, O, Cu Add more ionic species as pre-defined particle species. * Cu: Add to Ionization Energies Fix 2D: ParticleHistogram (ECP-WarpX#1635) Fix out-of-bounds access in particle histogram in non-3D. Fix FilterFunctor in non-3D (ECP-WarpX#1633) Fix an out-of-bounds access to positions in 2D & RZ for filter functors. openPMD: ionizationLevel (ECP-WarpX#1622) * openPMD: ionizationLevel Write out the ionizationLevel with openPMD. * openPMD Record Repetitions: Use Lambdas openPMD: Particle Filter (Container) (ECP-WarpX#1632) * openPMD: Particle Filter (Container) A fresh implementation of particle filters for openPMD diagnostics using the generalized `PhysicalParticleContainer` approach that we also use in plotfiles 🎉 * Ion Example: Filter & Coarsen Test diagnostics filter & coarsen functions in CI. Test reduced diagnostics (histograms). Define: _OPENMP -> AMREX_USE_OMP (ECP-WarpX#1520) * Define: _OPENMP -> AMREX_USE_OMP Replace the define check of `_OPENMP` with the explicit backend control of `AMREX_USE_OMP` for parallel constructs. Doing so avoids that we accidentially turn on OpenMP, e.g. if a dependency pulls it in for linear algebra, I/O, etc. This can led to confusion if the user explicitly requested a serial build. Also, we might want to use OpenMP functionality here and there for auxiliary functions w/o having to use the AMReX OpenMP backend, i.e. because we compile for GPUs. * Add missing amrex::Gpu::notInLaunchRegion Start: PerformanceHints After Init (ECP-WarpX#1630) Start a helper routine that gives performance hints after initialization of the simulation. Fix number of guard cells in PML with 2D PSATD (ECP-WarpX#1648) Remove unused lines about particle filters (ECP-WarpX#1646) Add parameter for default galilean velocity (ECP-WarpX#1097) * Add parameter for default galilean velocity * Apply suggestions from code review Co-authored-by: Edoardo Zoni <59625522+EZoni@users.noreply.github.com> * Fix bug in constructor of PhysicalParticleContainer * Use new input parameter in CI test Co-authored-by: Edoardo Zoni <59625522+EZoni@users.noreply.github.com> Co-authored-by: Edoardo Zoni <ezoni@lbl.gov> Cleaned up field diagnostic in picmi interface (ECP-WarpX#1649) Use pinned memory for tmp particles in diags. (ECP-WarpX#1644) * Use pinned memory for tmp particles in diags. * openPMD: pinned memory (tmp particles) Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Replace an always false test with an abort (ECP-WarpX#1655) * replace always false test with abort * fix bug [mini-PR] Broken link to scripts (ECP-WarpX#1651) CI: Ubuntu-latest -> 20.04 (ECP-WarpX#1654) Start Embedded Boundary development (ECP-WarpX#1641) * Start Embedded Boundary development Modify the build system for embedded boundary (EB) support. Currently EB is a compile time option that must be explicitly enabled. For GNU Make, one can enable it with `USE_EB=TRUE`, whereas for CMake, `-DAMReX_EB=ON`. Later we could decide to enable EB by default with all regular geometry. Add a simple geometry initialization function, WarpX::InitEB. By default, the geometry is all regular. A few basic types such as box, cylinder, plane, sphere, etc. are supported via ParmParse runtime parameters. See `amrex/Src/EB/AMReX_EB2.cpp` for more details. Later, we could build more complex geometry using constructive solid geometry (CSG). ( https://en.wikipedia.org/wiki/Constructive_solid_geometry ) See `amrex/Tutorials/EB/GeometryGeneration` for an example of CSG. There is also a STL ( https://en.wikipedia.org/wiki/STL_(file_format) ) approach under development in AMReX. Add a new member, m_factory, to WarpX class. This object can be used to obtain geometry information such as whether a cell is cut, edge centroids, etc. Currently we are not using these factories to build MultiFabs for field data to embed the geometry information into the data containers. We could do that later if it is needed or it makes things more convenient. Nevertheless, this should be sufficient to start the EB development of the field solver. It's not clear to me yet how many ghost cells are needed for the geometry information. It's currently one, and can be adjusted. In the future, when particle and embedded boundary interaction is considered, we can use `amrex::FillSignedDistance` function to obtain signed distance function on the nodes. With that information, one should be able to determine where and when a particle collides with the embedded boundary. * resize factory vector * CMake: Require AMReX_EB for embedded boundaries * Add AMReX_Config.H et al. Explicit includes are most robust to make sure defines are set with future refactorings. * rename Factory fieldFactory Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Add filter for particle histogram (ECP-WarpX#1643) * Add filter for particle histogram * Update benchmarks * Update benchmark again * Use serialize_ics in test + proper usage of RandomEngine * Laser-Ion Acc. (2D3V): Test Hist Filter Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Take time step into account to compute guard cells for J and rho (ECP-WarpX#1607) * Use IntVect for ng_J and ng_rho * Compute guard cells for J and rho based on dt * Reset some CI benchmarks * Fix rebase commit * Add back +1 cell for rho: fix remaining out-of-bound accesses * Simplify ASSERTS using new interface of amrex::numParticlesOutOfRange Modern setup.py: CMake-Driven, Multi-Dimensional (ECP-WarpX#1647) This adds a new, project-centric setup.py file. With this file, all dimensions (2D, 3D, RZ) of WarpX can be built and packaged at once, using the CMake build logic. Build & install: ```bash pip wheel -v . pip install *whl ``` Updates for electrostatic solver (ECP-WarpX#1604) * Fix electrostatic solver with momentum conservation * Fix electrostatic, adding call to FillBoundaryAux * For electrostatic, removed unneeded extra calls to UpdateAux and FillBoundary * For electrostatic, calculate fields at the end of the time step * Updated ElectrostaticSphere analysis script to use fields from end of time step Reset broken benchmark for CI test initial_distribution (ECP-WarpX#1661) make sure we redefine the tmp particle tiles when we load balance. (ECP-WarpX#1658) [mini-PR] Add particle tiling documentation (ECP-WarpX#1665) * tiling wip * do_tiling documentation * Update Docs/source/running_cpp/parameters.rst Co-authored-by: Luca Fedeli <luca.fedeli@for.unipi.it> Co-authored-by: Luca Fedeli <luca.fedeli@for.unipi.it> fix inconsistent formatting (ECP-WarpX#1666) parameter.rst fixes (ECP-WarpX#1663) requirements.txt: fix overspecification (ECP-WarpX#1668) I accidentally added the patch-level for version-compatible matching This removes the patch-level for `~=` matching for the packages that have a >=1 major version already. scipy 1.5+ (ECP-WarpX#1670) Sufficient since it was last release in december and works well. The 3.5 release series still builds wheels for Python 3.6 for Ubuntu oldstable (18.04). Remove redundant "do_qed" option in inputfile (ECP-WarpX#1667) * removed redundant do_qed option in inputfile * fixed bug remove semicolons (ECP-WarpX#1662) fixTypo for pml has particles parameter for docs (ECP-WarpX#1671) load balance efficiency reduced diagnostic eol minor eol
mrowan137
pushed a commit
to mrowan137/WarpX
that referenced
this pull request
Feb 3, 2021
…rpX#1624) * Fix bug in momentum-conserving interpolation function * Reset checksum benchmark for CI test momentum-conserving-gather Unused Params: Check after Step 0 (ECP-WarpX#1596) An early check for unused parameters after the first step. Add Superparticle version of getParticlePosition. (ECP-WarpX#1640) * Add Superparticle version of GetParticlePosition. * move unpack_particle to a free function and rename * no longer need SuperPType * Update Source/Particles/Pusher/GetAndSetPosition.H * remove templating from get_particle_position * remove template * Add missing include Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Distribution mapping and cost plotting (ECP-WarpX#1444) * Distribution mapping and cost plotting Cleanup Cleanup Cleanup EOL Unused import * Cost initializes to 0.0 * plot slices of 3D * WIP * WIP docs * docs * docs * docs * docs * docs * docs * docs * docs * EOL SpeciesProperties: C, N, O, Cu (ECP-WarpX#1638) * SpeciesProperties: C, N, O, Cu Add more ionic species as pre-defined particle species. * Cu: Add to Ionization Energies Fix 2D: ParticleHistogram (ECP-WarpX#1635) Fix out-of-bounds access in particle histogram in non-3D. Fix FilterFunctor in non-3D (ECP-WarpX#1633) Fix an out-of-bounds access to positions in 2D & RZ for filter functors. openPMD: ionizationLevel (ECP-WarpX#1622) * openPMD: ionizationLevel Write out the ionizationLevel with openPMD. * openPMD Record Repetitions: Use Lambdas openPMD: Particle Filter (Container) (ECP-WarpX#1632) * openPMD: Particle Filter (Container) A fresh implementation of particle filters for openPMD diagnostics using the generalized `PhysicalParticleContainer` approach that we also use in plotfiles 🎉 * Ion Example: Filter & Coarsen Test diagnostics filter & coarsen functions in CI. Test reduced diagnostics (histograms). Define: _OPENMP -> AMREX_USE_OMP (ECP-WarpX#1520) * Define: _OPENMP -> AMREX_USE_OMP Replace the define check of `_OPENMP` with the explicit backend control of `AMREX_USE_OMP` for parallel constructs. Doing so avoids that we accidentially turn on OpenMP, e.g. if a dependency pulls it in for linear algebra, I/O, etc. This can led to confusion if the user explicitly requested a serial build. Also, we might want to use OpenMP functionality here and there for auxiliary functions w/o having to use the AMReX OpenMP backend, i.e. because we compile for GPUs. * Add missing amrex::Gpu::notInLaunchRegion Start: PerformanceHints After Init (ECP-WarpX#1630) Start a helper routine that gives performance hints after initialization of the simulation. Fix number of guard cells in PML with 2D PSATD (ECP-WarpX#1648) Remove unused lines about particle filters (ECP-WarpX#1646) Add parameter for default galilean velocity (ECP-WarpX#1097) * Add parameter for default galilean velocity * Apply suggestions from code review Co-authored-by: Edoardo Zoni <59625522+EZoni@users.noreply.github.com> * Fix bug in constructor of PhysicalParticleContainer * Use new input parameter in CI test Co-authored-by: Edoardo Zoni <59625522+EZoni@users.noreply.github.com> Co-authored-by: Edoardo Zoni <ezoni@lbl.gov> Cleaned up field diagnostic in picmi interface (ECP-WarpX#1649) Use pinned memory for tmp particles in diags. (ECP-WarpX#1644) * Use pinned memory for tmp particles in diags. * openPMD: pinned memory (tmp particles) Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Replace an always false test with an abort (ECP-WarpX#1655) * replace always false test with abort * fix bug [mini-PR] Broken link to scripts (ECP-WarpX#1651) CI: Ubuntu-latest -> 20.04 (ECP-WarpX#1654) Start Embedded Boundary development (ECP-WarpX#1641) * Start Embedded Boundary development Modify the build system for embedded boundary (EB) support. Currently EB is a compile time option that must be explicitly enabled. For GNU Make, one can enable it with `USE_EB=TRUE`, whereas for CMake, `-DAMReX_EB=ON`. Later we could decide to enable EB by default with all regular geometry. Add a simple geometry initialization function, WarpX::InitEB. By default, the geometry is all regular. A few basic types such as box, cylinder, plane, sphere, etc. are supported via ParmParse runtime parameters. See `amrex/Src/EB/AMReX_EB2.cpp` for more details. Later, we could build more complex geometry using constructive solid geometry (CSG). ( https://en.wikipedia.org/wiki/Constructive_solid_geometry ) See `amrex/Tutorials/EB/GeometryGeneration` for an example of CSG. There is also a STL ( https://en.wikipedia.org/wiki/STL_(file_format) ) approach under development in AMReX. Add a new member, m_factory, to WarpX class. This object can be used to obtain geometry information such as whether a cell is cut, edge centroids, etc. Currently we are not using these factories to build MultiFabs for field data to embed the geometry information into the data containers. We could do that later if it is needed or it makes things more convenient. Nevertheless, this should be sufficient to start the EB development of the field solver. It's not clear to me yet how many ghost cells are needed for the geometry information. It's currently one, and can be adjusted. In the future, when particle and embedded boundary interaction is considered, we can use `amrex::FillSignedDistance` function to obtain signed distance function on the nodes. With that information, one should be able to determine where and when a particle collides with the embedded boundary. * resize factory vector * CMake: Require AMReX_EB for embedded boundaries * Add AMReX_Config.H et al. Explicit includes are most robust to make sure defines are set with future refactorings. * rename Factory fieldFactory Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Add filter for particle histogram (ECP-WarpX#1643) * Add filter for particle histogram * Update benchmarks * Update benchmark again * Use serialize_ics in test + proper usage of RandomEngine * Laser-Ion Acc. (2D3V): Test Hist Filter Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Take time step into account to compute guard cells for J and rho (ECP-WarpX#1607) * Use IntVect for ng_J and ng_rho * Compute guard cells for J and rho based on dt * Reset some CI benchmarks * Fix rebase commit * Add back +1 cell for rho: fix remaining out-of-bound accesses * Simplify ASSERTS using new interface of amrex::numParticlesOutOfRange Modern setup.py: CMake-Driven, Multi-Dimensional (ECP-WarpX#1647) This adds a new, project-centric setup.py file. With this file, all dimensions (2D, 3D, RZ) of WarpX can be built and packaged at once, using the CMake build logic. Build & install: ```bash pip wheel -v . pip install *whl ``` Updates for electrostatic solver (ECP-WarpX#1604) * Fix electrostatic solver with momentum conservation * Fix electrostatic, adding call to FillBoundaryAux * For electrostatic, removed unneeded extra calls to UpdateAux and FillBoundary * For electrostatic, calculate fields at the end of the time step * Updated ElectrostaticSphere analysis script to use fields from end of time step Reset broken benchmark for CI test initial_distribution (ECP-WarpX#1661) make sure we redefine the tmp particle tiles when we load balance. (ECP-WarpX#1658) [mini-PR] Add particle tiling documentation (ECP-WarpX#1665) * tiling wip * do_tiling documentation * Update Docs/source/running_cpp/parameters.rst Co-authored-by: Luca Fedeli <luca.fedeli@for.unipi.it> Co-authored-by: Luca Fedeli <luca.fedeli@for.unipi.it> fix inconsistent formatting (ECP-WarpX#1666) parameter.rst fixes (ECP-WarpX#1663) requirements.txt: fix overspecification (ECP-WarpX#1668) I accidentally added the patch-level for version-compatible matching This removes the patch-level for `~=` matching for the packages that have a >=1 major version already. scipy 1.5+ (ECP-WarpX#1670) Sufficient since it was last release in december and works well. The 3.5 release series still builds wheels for Python 3.6 for Ubuntu oldstable (18.04). Remove redundant "do_qed" option in inputfile (ECP-WarpX#1667) * removed redundant do_qed option in inputfile * fixed bug remove semicolons (ECP-WarpX#1662) fixTypo for pml has particles parameter for docs (ECP-WarpX#1671) Move getCosts Always synchronize nodal points of PML MultiFabs (ECP-WarpX#1669) * Always synchronize nodal points of PML MultiFabs * Reset benchmark for CI test pml_x_psatd * Use new separate functions NodalSyncPML minor
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Replace the define check of
_OPENMPwith the explicit backend control ofAMREX_USE_OMPfor compute constructs.Doing so avoids that we accidentially turn on OpenMP, e.g. if a dependency pulls it in for linear algebra, I/O, etc. This can led to confusion if the user explicitly requested a serial build. Also, we might want to use OpenMP functionality here and there for auxiliary functions w/o having to use the AMReX OpenMP backend, i.e. because we compile for GPUs.
Related to: AMReX-Codes/amrex#1558 (comment) AMReX-Codes/amrex#1560