Improve GPU performance of fluid-* interact

[efaulhaber]

This PR rewrites the interact! function for fluid-* interaction and is part of #1131.

• The foreach_point_neighbor loop is now unrolled into an @threaded for particle and foreach_neighbor.
• With this change, I can now load data for particle a once instead of loading it again for each neighbor.
• This unrolled loop now also allows me to accumulate the dv values over all neighbors and write to dv only once per particle.

[codecov]

Codecov Report

❌ Patch coverage is 94.73684% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.10%. Comparing base (a3f1139) to head (07c71e4).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/general/neighborhood_search.jl	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1116      +/-   ##
==========================================
- Coverage   89.17%   89.10%   -0.08%     
==========================================
  Files         128      128              
  Lines        9925     9925              
==========================================
- Hits         8851     8844       -7     
- Misses       1074     1081       +7     

Flag	Coverage Δ
total	89.10% <94.73%> (-0.08%)	⬇️
unit	67.62% <81.57%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

Refactors the fluid-* interact! implementation to improve GPU performance by unrolling the point-neighbor iteration into a per-particle threaded loop and reducing repeated loads/writes.

Changes:

• Rewrites WCSPH interact! to thread over particles, use foreach_neighbor, and accumulate per-particle RHS contributions before writing.
• Introduces foreach_neighbor wrapper with a GPU-unsafe fast path (bounds-check-free) for neighbor traversal.
• Adds a 3D WCSPH ContinuityDensity fast path that combines velocity+density loads using SIMD wide loads; updates correction and pressure helper APIs accordingly.

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/schemes/fluid/weakly_compressible_sph/rhs.jl	Main WCSPH interact! refactor; adds neighbor_pressure and velocity_and_density helpers + SIMD load fast path.
src/schemes/fluid/implicit_incompressible_sph/rhs.jl	Switches to neighbor_pressure helper to avoid redundant pressure loads / enable mirroring behavior.
src/general/neighborhood_search.jl	Adds foreach_neighbor wrapper with a GPU specialization calling PointNeighbors.foreach_neighbor_unsafe.
src/general/corrections.jl	Adjusts free_surface_correction API to accept (rho_a, rho_b) and compute rho_mean internally.
src/TrixiParticles.jl	Adds SIMD import required by the new 3D fast path.
Project.toml	Adds SIMD dependency and bumps PointNeighbors compat to 0.6.6.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[efaulhaber]

/run-gpu-tests

[efaulhaber]

/run-gpu-tests

[LasNikas]

The following remarks can also be resolved by creating an issue for each one:

• There is still a PointNeighbors.foreach_neighbor call in mirroring.jl. Should that also be changed to foreach_neighbor in this PR?
• Could the update_shifting methods also be optimized in this way?

[efaulhaber]

There is still a PointNeighbors.foreach_neighbor call in mirroring.jl. Should that also be changed to foreach_neighbor in this PR?

No. This PointNeighbors.foreach_neighbor is the internal one that has the position as an argument. This is only supported by the GridNeighborhoodSearch and not yet documented.

Could the update_shifting methods also be optimized in this way?

Yes, absolutely. So far, we have TLSPH RHS (merged), TLSPH deformation gradient and stress tensor (merged), WCSPH RHS (this PR). Next are IISPH and EDAC RHS, open boundary RHS, shifting updates, etc.

[svchb]

/run-gpu-tests