[BUG] Native-spin DPA4C graph path fails on zero-node graphs

[njzjz-bot]

Summary

The portable native-spin DPA4C graph path crashes when the graph has zero nodes. The spin-free path and the compressed native-spin CPU path handle this empty-node case.

This reproduces on origin/master at 8cfd46e37448.

Reproduction

import torch

from deepmd.dpmodel.utils.neighbor_graph import NeighborGraph
from deepmd.pt_expt.descriptor.dpa4c import DescrptDPA4C

d = DescrptDPA4C(
    rcut=3.0,
    ntypes=2,
    channels=8,
    lmax=2,
    n_radial=4,
    precision="float32",
    use_spin=[True, False],
).eval()

g = NeighborGraph(
    n_node=torch.zeros(1, dtype=torch.int64),
    n_local=torch.zeros(1, dtype=torch.int64),
    edge_index=torch.empty((2, 0), dtype=torch.int64),
    edge_vec=torch.empty((0, 3), dtype=torch.float32),
    edge_mask=torch.empty((0,), dtype=torch.bool),
)
d.call_graph(
    g,
    torch.empty((0,), dtype=torch.int64),
    spin=torch.empty((0, 3), dtype=torch.float32),
)

Actual result:

deepmd/dpmodel/descriptor/dpa4c_nn/spin.py:328
RuntimeError: cannot reshape tensor of 0 elements into shape [0, -1]

Cause and expected behavior

The spin/geometric cross Gram is reshaped to (0, -1), for which PyTorch cannot infer the second dimension. The width is already known and can be stated explicitly as self.quadrupole_width * self.degree_channels[2].

The call should return an empty descriptor with shape (0, descriptor.get_dim_out()). A regression test should cover call_graph() on the same empty native-spin graph already covered by the compressed CPU reference.

---

Coding agent: Codex
Codex version: codex-cli 0.149.0
Model: gpt-5.6-sol
Reasoning effort: xhigh