Merge pull request #724 from cjknight/comm-nprocs-opt (43b394c6) · Commits · 郑智淋 / lammps

doc/src/kspace_modify.txt

+9 −1

Original line number	Diff line number	Diff line
		@@ -13,7 +13,7 @@ kspace_modify command :h3
		kspace_modify keyword value ... :pre

		one or more keyword/value pairs may be listed :ulb,l
		keyword = {mesh} or {order} or {order/disp} or {mix/disp} or {overlap} or {minorder} or {force} or {gewald} or {gewald/disp} or {slab} or (nozforce} or {compute} or {cutoff/adjust} or {fftbench} or {collective} or {diff} or {kmax/ewald} or {force/disp/real} or {force/disp/kspace} or {splittol} or {disp/auto}:l
		keyword = {mesh} or {order} or {order/disp} or {mix/disp} or {overlap} or {minorder} or {force} or {gewald} or {gewald/disp} or {slab} or (nozforce} or {compute} or {cutoff/adjust} or {fftbench} or {collective} or {remap/dup} or {diff} or {kmax/ewald} or {force/disp/real} or {force/disp/kspace} or {splittol} or {disp/auto}:l
		{mesh} value = x y z
		x,y,z = grid size in each dimension for long-range Coulombics
		{mesh/disp} value = x y z
		@@ -40,6 +40,7 @@ keyword = {mesh} or {order} or {order/disp} or {mix/disp} or {overlap} or {minor
		{pressure/scalar} value = {yes} or {no}
		{fftbench} value = {yes} or {no}
		{collective} value = {yes} or {no}
		{remap/dup} value = {yes} or {no}
		{diff} value = {ad} or {ik} = 2 or 4 FFTs for PPPM in smoothed or non-smoothed mode
		{kmax/ewald} value = kx ky kz
		kx,ky,kz = number of Ewald sum kspace vectors in each dimension
		@@ -232,6 +233,13 @@ This is faster on IBM BlueGene machines, and may also be faster on
		other machines if they have an efficient implementation of MPI
		collective operations and adequate hardware.

		The {remap/dup} keyword applies only to PPPM. It is set to {no} by
		default. If this option is set to {yes}, LAMMPS will duplicate the
		MPI communicator in the final remap plan for 3d-FFT operations. In the
		final plan, this is much faster on a large number of MPI ranks than
		iteratively generating the list of all participating MPI ranks, which
		in this case is all MPI ranks in communicator.

		The {diff} keyword specifies the differentiation scheme used by the
		PPPM method to compute forces on particles given electrostatic
		potentials on the PPPM mesh. The {ik} approach is the default for

src/KSPACE/pppm.cpp

+1 −1

Original line number	Diff line number	Diff line
		@@ -839,7 +839,7 @@ void PPPM::allocate()
		remap = new Remap(lmp,world,
		nxlo_in,nxhi_in,nylo_in,nyhi_in,nzlo_in,nzhi_in,
		nxlo_fft,nxhi_fft,nylo_fft,nyhi_fft,nzlo_fft,nzhi_fft,
		1,0,0,FFT_PRECISION,collective_flag,1);
		1,0,0,FFT_PRECISION,collective_flag,remapdup_flag);

		// create ghost grid object for rho and electric field communication

src/kspace.cpp

+10 −0

Original line number	Diff line number	Diff line
		@@ -57,6 +57,10 @@ KSpace::KSpace(LAMMPS lmp, int narg, char *arg) : Pointers(lmp)
		collective_flag = 0;
		#endif

		// default is to iteratively generate list of participating MPI ranks

		remapdup_flag = 0;

		kewaldflag = 0;

		order_6 = 5;
		@@ -523,6 +527,12 @@ void KSpace::modify_params(int narg, char **arg)
		else if (strcmp(arg[iarg+1],"no") == 0) collective_flag = 0;
		else error->all(FLERR,"Illegal kspace_modify command");
		iarg += 2;
		} else if (strcmp(arg[iarg],"remap/dup") == 0) {
		if (iarg+2 > narg) error->all(FLERR,"Illegal kspace_modify command");
		if (strcmp(arg[iarg+1],"yes") == 0) remapdup_flag = 1;
		else if (strcmp(arg[iarg+1],"no") == 0) remapdup_flag = 0;
		else error->all(FLERR,"Illegal kspace_modify command");
		iarg += 2;
		} else if (strcmp(arg[iarg],"diff") == 0) {
		if (iarg+2 > narg) error->all(FLERR,"Illegal kspace_modify command");
		if (strcmp(arg[iarg+1],"ad") == 0) differentiation_flag = 1;

src/kspace.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -89,6 +89,7 @@ class KSpace : protected Pointers {
		int compute_flag; // 0 if skip compute()
		int fftbench; // 0 if skip FFT timing
		int collective_flag; // 1 if use MPI collectives for FFT/remap
		int remapdup_flag; // 1 if hint to duplicate MPI comm
		int stagger_flag; // 1 if using staggered PPPM grids

		double splittol; // tolerance for when to truncate splitting

Admin message