Add conservative projection functionality to C-Fortran API by Fuad-HH · Pull Request #349 · SCOREC/pcms

Fuad-HH · 2026-08-02T22:29:19Z

Produces visually correct results after merging Fix std::array Bug for GPU kernels - For Conservative Transfer #348. Plotted den_neut and temp_neut from xgc.neutral.*.bp files.
The petsc MPI comm changes are temp fix for some error
0-D plots are used for verification
Rebased to the develop branch

- runs but results are incorrect - followed the python example - the petsc mpi comm changes are temp fix for some error

replace pcms::Field with pcms::Function

jacobmerson · 2026-08-12T21:08:43Z

I will launch self-hosted build in a few minutes. Currently updating the dependencies [redev].

Copilot

Pull request overview

This PR extends the C/Fortran API surface to support a cached, mesh-intersection-based conservative projection workflow, and adjusts PETSc object creation/abort communicators to use PETSC_COMM_SELF for sequential PETSc objects (improving correctness for MPI runs and GPU/Kokkos paths).

Changes:

Switch sequential PETSc Vec/Mat/KSP creation and CHKERRABORT calls from PETSC_COMM_WORLD to PETSC_COMM_SELF in conservative transfer components.
Add a new conservative projection handle + functions to the C API, and expose them through the SWIG-generated Fortran module/wrappers.
Implement conservative projection context creation, apply, and destroy logic in the C API using Omega_h + PCMS MeshFields + PETSc.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/pcms/transfer/omega_h_mc_rhs_integrator.cpp	Use `PETSC_COMM_SELF` for sequential PETSc RHS vector creation/assembly aborts.
src/pcms/transfer/omega_h_mass_integrator.cpp	Use `PETSC_COMM_SELF` for sequential PETSc mass matrix creation/assembly aborts.
src/pcms/transfer/omega_h_intersection_rhs_integrator.cpp	Use `PETSC_COMM_SELF` for sequential PETSc RHS vector creation/assembly aborts.
src/pcms/transfer/conservative_projection_solver.cpp	Use `PETSC_COMM_SELF` for sequential KSP creation/solve error handling.
src/pcms/fortranapi/pcms_interpolator.f90	Expose new conservative projection handle and entry points to Fortran (SWIG output).
src/pcms/fortranapi/interpolator.i	Add SWIG interface declarations for the conservative projection C API.
src/pcms/fortranapi/interpolator_wrap.c	Add SWIG-generated C wrapper glue for the new conservative projection API.
src/pcms/capi/interpolator.h	Declare/document the new conservative projection C API and handle type.
src/pcms/capi/interpolator.cpp	Implement the new conservative projection C API (context + create/apply/destroy).

Suppressed comments (1)

src/pcms/capi/interpolator.cpp:223

pcms_conservative_projection_get_target_size() unconditionally casts and dereferences projection.pointer. If the handle is null/invalid this will crash; return 0 and emit an error instead.

int pcms_conservative_projection_get_target_size(
  PcmsConservativeProjectionHandle projection)
{
  auto* ctx =
    reinterpret_cast<ConservativeProjectionContext*>(projection.pointer);
  return ctx->target_space->GetLayout()->GetNumOwnedDofHolder();
}

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

+int pcms_conservative_projection_get_source_size(
+  PcmsConservativeProjectionHandle projection)
+{
+  auto* ctx =
+    reinterpret_cast<ConservativeProjectionContext*>(projection.pointer);
+  return ctx->source_space->GetLayout()->GetNumOwnedDofHolder();
+}


+PcmsConservativeProjectionHandle pcms_create_conservative_projection(
+  const char* source_mesh_name, int source_order, const char* target_mesh_name,
+  int target_order)
+{
+  auto* ctx = new ConservativeProjectionContext(
+    source_mesh_name, target_mesh_name, source_order, target_order);
+  return {reinterpret_cast<void*>(ctx)};
+}


+ * @brief Apply conservative projection
+ * @param projection Handle to the projection
+ * @param source_data Flat array of source field values (size = source_size)
+ * @param source_size Number of source DOF holders
+ * @param target_data Flat array to receive target field values (size =
+ * target_size)
+ * @param target_size Number of target DOF holders


+void pcms_conservative_projection_apply(
+  PcmsConservativeProjectionHandle projection, void* source_data,
+  int source_size, void* target_data, int target_size)
+{
+  auto* ctx =
+    reinterpret_cast<ConservativeProjectionContext*>(projection.pointer);
+
+  // Copy source data into internal field
+  auto source_view = pcms::Rank2View<const pcms::Real, pcms::HostMemorySpace>(
+    reinterpret_cast<pcms::Real*>(source_data), source_size, 1);
+  ctx->source_field.SetDOFHolderDataHost(source_view);
+
+  // Apply conservative projection
+  ctx->projection->Apply(ctx->source_field, ctx->target_field);
+
+  // Copy result back to user buffer
+  auto target_view = ctx->target_field.GetDOFHolderDataHost();
+  auto flat = pcms::FlattenToRank1View(target_view);
+  std::memcpy(target_data, flat.data_handle(),
+              static_cast<std::size_t>(target_size) * sizeof(double));
+}


jacobmerson

I commented in a number of places, but the biggest thing is that with the new API, you should have everything except the creation function be called on a generic PcmsTransferOperatorHandle which holds the base pointer to the TransferOperator. That way the same functions can be used for every field transfer method in the exact same way. I.e., the only thing you need to do to swap out the transfer method you are using is swap out the constructor.

jacobmerson · 2026-08-14T02:29:57Z

+
+} // namespace
+
+PcmsConservativeProjectionHandle pcms_create_conservative_projection(


Can this return a TransferOperatorHandle? With the new API, all of the transfer operators should be able to be used the same way after construction.

jacobmerson · 2026-08-14T02:30:43Z

+  return {reinterpret_cast<void*>(ctx)};
+}
+
+int pcms_conservative_projection_get_source_size(


pcms_transfer_get_source_size(TransferOperatorHandle)?

jacobmerson · 2026-08-14T02:31:14Z

+  return ctx->source_space->GetLayout()->GetNumOwnedDofHolder();
+}
+
+int pcms_conservative_projection_get_target_size(


pcms_transfer_get_target_size(TransferOperatorHandle)?

jacobmerson · 2026-08-14T02:31:42Z

+  return ctx->target_space->GetLayout()->GetNumOwnedDofHolder();
+}
+
+void pcms_conservative_projection_apply(


pcms_transfer_apply(TransferOperatorHandle)

jacobmerson · 2026-08-14T02:33:56Z

+
+  // Copy result back to user buffer
+  auto target_view = ctx->target_field.GetDOFHolderDataHost();
+  auto flat = pcms::FlattenToRank1View(target_view);


I suspect we can probably avoid this memcopy if the target field is constructed based on the target field pointer. I think the XGC field data is non-owning specifically for this reason, however I implemented that a while ago so it needs to be checked.

jacobmerson · 2026-08-14T02:34:22Z

+              static_cast<std::size_t>(target_size) * sizeof(double));
+}
+
+void pcms_destroy_conservative_projection(


pcms_destroy_transfer_operator(...

Fuad-HH requested a deployment to asics August 2, 2026 22:29 — with GitHub Actions Waiting

Add conservative projection functionality to interpolator

113d36d

- runs but results are incorrect - followed the python example - the petsc mpi comm changes are temp fix for some error

Fuad-HH force-pushed the conservative-api branch from 48e54d1 to 113d36d Compare August 11, 2026 18:30

Fuad-HH requested a deployment to asics August 11, 2026 18:30 — with GitHub Actions Waiting

refactor to changed in dev branch

46d5185

replace pcms::Field with pcms::Function

Fuad-HH requested a deployment to asics August 12, 2026 20:58 — with GitHub Actions Waiting

format interpolator.h and .cpp

ba69014

Fuad-HH temporarily deployed to asics August 12, 2026 21:00 — with GitHub Actions Inactive

Fuad-HH marked this pull request as ready for review August 12, 2026 21:09

Fuad-HH requested review from jacobmerson and a lite review from Copilot August 12, 2026 21:09

Copilot started reviewing on behalf of Fuad-HH August 12, 2026 21:10 View session

Copilot AI reviewed Aug 12, 2026

View reviewed changes

jacobmerson reviewed Aug 14, 2026

View reviewed changes

Navigation Menu

Uh oh!

Add conservative projection functionality to C-Fortran API - #349

Add conservative projection functionality to C-Fortran API#349
Fuad-HH wants to merge 3 commits into
SCOREC:developfrom
Fuad-HH:conservative-api

Fuad-HH commented Aug 2, 2026 •

edited

Loading

Uh oh!

jacobmerson commented Aug 12, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

jacobmerson left a comment

Uh oh!

jacobmerson Aug 14, 2026

Uh oh!

jacobmerson Aug 14, 2026

Uh oh!

jacobmerson Aug 14, 2026

Uh oh!

jacobmerson Aug 14, 2026

Uh oh!

jacobmerson Aug 14, 2026

Uh oh!

jacobmerson Aug 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants


		} // namespace

		PcmsConservativeProjectionHandle pcms_create_conservative_projection(

Uh oh!

Conversation

Fuad-HH commented Aug 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jacobmerson commented Aug 12, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

jacobmerson left a comment

Choose a reason for hiding this comment

Uh oh!

jacobmerson Aug 14, 2026

Choose a reason for hiding this comment

Uh oh!

jacobmerson Aug 14, 2026

Choose a reason for hiding this comment

Uh oh!

jacobmerson Aug 14, 2026

Choose a reason for hiding this comment

Uh oh!

jacobmerson Aug 14, 2026

Choose a reason for hiding this comment

Uh oh!

jacobmerson Aug 14, 2026

Choose a reason for hiding this comment

Uh oh!

jacobmerson Aug 14, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Fuad-HH commented Aug 2, 2026 •

edited

Loading