Allowed pppm/gpu to fall back to pppm for triclinic boxes (bd87b112) · Commits · 郑智淋 / lammps

src/GPU/pppm_gpu.cpp

+28 −15

Original line number	Original line	Diff line number	Diff line
	@@ -91,7 +91,6 @@ void PPPM_GPU_API(forces)(double **f);

	PPPMGPU::PPPMGPU(LAMMPS *lmp) : PPPM(lmp)		PPPMGPU::PPPMGPU(LAMMPS *lmp) : PPPM(lmp)
	{		{
	triclinic_support = 0;
	density_brick_gpu = vd_brick = NULL;		density_brick_gpu = vd_brick = NULL;
	kspace_split = false;		kspace_split = false;
	im_real_space = false;		im_real_space = false;
	@@ -210,6 +209,7 @@ void PPPMGPU::compute(int eflag, int vflag)
	cg_peratom->setup();		cg_peratom->setup();
	}		}

			if (triclinic == 0) {
	bool success = true;		bool success = true;
	int flag=PPPM_GPU_API(spread)(nago, atom->nlocal, atom->nlocal +		int flag=PPPM_GPU_API(spread)(nago, atom->nlocal, atom->nlocal +
	atom->nghost, atom->x, atom->type, success,		atom->nghost, atom->x, atom->type, success,
	@@ -219,6 +219,7 @@ void PPPMGPU::compute(int eflag, int vflag)
	error->one(FLERR,"Insufficient memory on accelerator");		error->one(FLERR,"Insufficient memory on accelerator");
	if (flag != 0)		if (flag != 0)
	error->one(FLERR,"Out of range atoms - cannot compute PPPM");		error->one(FLERR,"Out of range atoms - cannot compute PPPM");
			}

	// convert atoms from box to lamda coords		// convert atoms from box to lamda coords

	@@ -229,9 +230,10 @@ void PPPMGPU::compute(int eflag, int vflag)
	}		}

	// If need per-atom energies/virials, also do particle map on host		// If need per-atom energies/virials, also do particle map on host
	// concurrently with GPU calculations		// concurrently with GPU calculations,
			// or if the box is triclinic, particle map is done on host

	if (evflag_atom) {		if (evflag_atom \|\| triclinic) {

	// extend size of per-atom arrays if necessary		// extend size of per-atom arrays if necessary

	@@ -244,14 +246,24 @@ void PPPMGPU::compute(int eflag, int vflag)
	particle_map();		particle_map();
	}		}

			// if the box is triclinic,
			// map my particle charge onto my local 3d density grid on the host

			if (triclinic) make_rho();

	double t3 = MPI_Wtime();		double t3 = MPI_Wtime();

	// all procs communicate density values from their ghost cells		// all procs communicate density values from their ghost cells
	// to fully sum contribution in their 3d bricks		// to fully sum contribution in their 3d bricks
	// remap from 3d decomposition to FFT decomposition		// remap from 3d decomposition to FFT decomposition

			if (triclinic == 0) {
	cg->reverse_comm(this,REVERSE_RHO_GPU);		cg->reverse_comm(this,REVERSE_RHO_GPU);
	brick2fft_gpu();		brick2fft_gpu();
			} else {
			cg->reverse_comm(this,REVERSE_RHO);
			PPPM::brick2fft();
			}

	// compute potential gradient on my FFT grid and		// compute potential gradient on my FFT grid and
	// portion of e_long on this proc's FFT grid		// portion of e_long on this proc's FFT grid
	@@ -279,7 +291,8 @@ void PPPMGPU::compute(int eflag, int vflag)
	// calculate the force on my particles		// calculate the force on my particles

	FFT_SCALAR qscale = force->qqrd2e * scale;		FFT_SCALAR qscale = force->qqrd2e * scale;
	PPPM_GPU_API(interp)(qscale);		if (triclinic == 0) PPPM_GPU_API(interp)(qscale);
			else fieldforce();

	// per-atom energy/virial		// per-atom energy/virial
	// energy includes self-energy correction		// energy includes self-energy correction

Admin message