dpa2 model may need a fake communication op to deal with nloc == 0.

[github-actions]

this should be fixed after wrapping comm op as a pure c++ implementation.

Line: 135

deepmd-kit//source/api_cc/src/DeepPotPT.cc

Lines 132 to 142 in d0fe13c

	nghost, ntypes, 1, daparam, nall, aparam_nall);
	int nloc = nall_real - nghost_real;
	int nframes = 1;
	// TODO: dpa2 model may need a fake communication op to deal with nloc == 0.
	// this should be fixed after wrapping comm op as a pure c++ implementation.
	if (nloc == 0) {
	// no backward map needed
	ener.resize(nframes);
	// dforce of size nall * 3
	force.resize(static_cast<size_t>(nframes) * fwd_map.size() * 3);
	fill(force.begin(), force.end(), (VALUETYPE)0.0);

[njzjz]

It seems to me that we should support the situation for nloc == 0 everywhere, instead of writing a special case.

[wanghan-iapcm]

I agree.

[github-actions]

Closed in commit 7907142