MeshToParticle attempts out of bounds access?

[iljah]

Using code in #4801 with addition std::cout<<"data2 "<<data2<<" ijk "<<i<<","<<j<<","<<k<<" c "<<comp<<std::endl; I get data2 (((0,0,0),(1,1,0)),3) ijk 1,1,1 c 0 so does MeshToParticle try to access non-existing z index 1 even though (periodic) grid is only 2x2x1? Am I doing something wrong?

[aonexyz]

From the output you posted, your observation is correct: the tuple
data2 (((0,0,0),(1,1,0)),3) with ijk = 1,1,1 indicates that
MeshToParticle is attempting to access index k = 1 in the Z-direction.

Since your grid is 2 × 2 × 1, the only valid k index is 0.
So when MeshToParticle queries k = 1, it is effectively outside the
range of the z dimension.

This does not necessarily mean your code is wrong it often happens
when the interpolation stencil assumes a 3D neighborhood even for grids
with Nz = 1. In fully periodic setups, the operator typically wraps
indices automatically, but when Nz = 1 the wrap still produces k = 1
before being reduced modulo 1 → 0.

In other words:

• the interpolation code asks for (i,j,k = 1)
• but since Nz = 1, k = 1 is “out of bounds” in the logical array
• the periodic wrap is expected to map k = 1 → 0
• if wrap isn’t applied correctly, you see log output with k = 1

So you are not doing anything wrong this is an edge case when
Nz = 1 (a “2.5D” mesh). The interpolation code still builds a 3D
stencil and temporarily generates an index that looks invalid before
wrapping.

If your run does not crash and the data access eventually resolves to
k = 0 internally, then this is safe. If it crashes or reads invalid
memory, then it likely needs an additional bounds/wrap check in the
z-direction for Nz = 1 grids.

If you can share:

• the grid dimensions passed into MeshToParticle
• your boundary/periodicity flags
• the exact call site for MeshToParticle

we can confirm whether the z-wrap logic is being applied correctly.

[iljah]

Thanks, all code is in #4801

[iljah]

Specifically grid dimensions I guess are: amrex::Box domain{IntVect{0, 0, 0}, IntVect{1, 1, 0}};, periodicity: amrex::Array<int, 3>{1, 1, 1} and call site:

	amrex::MeshToParticle(particle_container, V, 0, [=](auto& particle1, const auto& data1){
		amrex::ParticleInterpolator::Nearest interp(particle1, grid_start, inv_dr);
		interp.MeshToParticle(particle1, data1, 0, Vx_i, 3,
			[=](const auto& data2, int i, int j, int k, int comp){
				return data2(i, j, k, comp);
			},
			[=](auto& particle2, int comp, auto value){
				particle2.rdata(comp) += value;
			});
	});

[iljah]

Looks like things work if particles are in first half (closer to negative side) of cell/grid. In that case valid i,j,k are given to lambda. But if I create particles with coordinates > 0 then too large values are used for indices in those dimensions even if grid consists of more than one cell in that dimension. For example: #4801 (comment)

[iljah]

Neareast neighbor interpolator assuming vertex data explains why interpolation doesn't work if particle is located in last half of last cell (#4810), but that still leaves the question of how I should prepare my grid for interpolation so that interpolator doesn't abort with invalid index even though I should be using periodic grid.

[iljah]

Out of bounds access happens in any dimension, I just happened to first run into this in Z dim.