/** (re-)allocation of the memory needed for the phys. values and if needed memory for the nodes located in the cpu memory
*/
static void lb_realloc_fluid_gpu() {
LB_TRACE (printf("#nodes \t %u \n", lbpar_gpu.number_of_nodes));
/**-----------------------------------------------------*/
/** reallocating of the needed memory for several structs */
/**-----------------------------------------------------*/
/**Struct holding calc phys values rho, j, phi of every node*/
size_t size_of_values = lbpar_gpu.number_of_nodes * sizeof(LB_values_gpu);
host_values = realloc(host_values, size_of_values);
LB_TRACE (fprintf(stderr,"lb_realloc_fluid_gpu \n"));
}
/** (re-) allocation of the memory needed for the particles (cpu part)
*/
void lb_realloc_particles_gpu(){
lbpar_gpu.number_of_particles = n_part;
LB_TRACE (printf("#particles realloc\t %u \n", lbpar_gpu.number_of_particles));
//fprintf(stderr, "%u \t \n", lbpar_gpu.number_of_particles);
/**-----------------------------------------------------*/
/** allocating of the needed memory for several structs */
/**-----------------------------------------------------*/
lbpar_gpu.your_seed = (unsigned int)i_random(max_ran);
LB_TRACE (fprintf(stderr,"test your_seed %u \n", lbpar_gpu.your_seed));
lb_realloc_particles_GPU_leftovers(&lbpar_gpu);
}
/** (Re-)initializes the fluid. */
void lb_reinit_parameters_gpu() {
int ii;
lbpar_gpu.time_step = (float)time_step;
for(ii=0;ii<LB_COMPONENTS;++ii){
lbpar_gpu.mu[ii] = 0.0;
if (lbpar_gpu.viscosity[ii] > 0.0) {
/* Eq. (80) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007). */
lbpar_gpu.gamma_shear[ii] = 1. - 2./(6.*lbpar_gpu.viscosity[ii]*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
if (lbpar_gpu.bulk_viscosity[ii] > 0.0) {
/* Eq. (81) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007). */
lbpar_gpu.gamma_bulk[ii] = 1. - 2./(9.*lbpar_gpu.bulk_viscosity[ii]*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
#ifdef SHANCHEN
if (lbpar_gpu.mobility[0] > 0.0) {
lbpar_gpu.gamma_mobility[0] = 1. - 2./(6.*lbpar_gpu.mobility[0]*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
#endif
if (temperature > 0.0) { /* fluctuating hydrodynamics ? */
lbpar_gpu.fluct = 1;
LB_TRACE (fprintf(stderr, "fluct on \n"));
/* Eq. (51) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007).*/
/* Note that the modes are not normalized as in the paper here! */
lbpar_gpu.mu[ii] = (float)temperature*lbpar_gpu.tau*lbpar_gpu.tau/c_sound_sq/(lbpar_gpu.agrid*lbpar_gpu.agrid);
/* lb_coupl_pref is stored in MD units (force)
* Eq. (16) Ahlrichs and Duenweg, JCP 111(17):8225 (1999).
* The factor 12 comes from the fact that we use random numbers
* from -0.5 to 0.5 (equally distributed) which have variance 1/12.
* time_step comes from the discretization.
*/
lbpar_gpu.lb_coupl_pref[ii] = sqrt(12.f*2.f*lbpar_gpu.friction[ii]*(float)temperature/lbpar_gpu.time_step);
lbpar_gpu.lb_coupl_pref2[ii] = sqrt(2.f*lbpar_gpu.friction[ii]*(float)temperature/lbpar_gpu.time_step);
} else {
/* no fluctuations at zero temperature */
lbpar_gpu.fluct = 0;
lbpar_gpu.lb_coupl_pref[ii] = 0.0;
lbpar_gpu.lb_coupl_pref2[ii] = 0.0;
}
LB_TRACE (fprintf(stderr,"lb_reinit_prarameters_gpu \n"));
}
reinit_parameters_GPU(&lbpar_gpu);
}
/** Performs a full initialization of
* the Lattice Boltzmann system. All derived parameters
* and the fluid are reset to their default values. */
void lb_init_gpu() {
LB_TRACE(printf("Begin initialzing fluid on GPU\n"));
/** set parameters for transfer to gpu */
lb_reinit_parameters_gpu();
lb_realloc_particles_gpu();
lb_init_GPU(&lbpar_gpu);
gpu_init_particle_comm();
cuda_bcast_global_part_params();
LB_TRACE(printf("Initialzing fluid on GPU successful\n"));
}
/** allocation of the needed memory for phys. values and particle data residing in the cpu memory
*/
void lb_pre_init_gpu() {
lbpar_gpu.number_of_particles = 0;
LB_TRACE (fprintf(stderr,"#nodes \t %u \n", lbpar_gpu.number_of_nodes));
/*-----------------------------------------------------*/
/* allocating of the needed memory for several structs */
/*-----------------------------------------------------*/
/* Struct holding calc phys values rho, j, phi of every node */
size_t size_of_values = lbpar_gpu.number_of_nodes * sizeof(LB_values_gpu);
host_values = (LB_values_gpu*)malloc(size_of_values);
LB_TRACE (fprintf(stderr,"lb_pre_init_gpu \n"));
}
/** (re-) allocation of the memory need for the particles (cpu part)*/
void lb_realloc_particles_gpu(){
lbpar_gpu.number_of_particles = n_total_particles;
LB_TRACE (printf("#particles realloc\t %u \n", lbpar_gpu.number_of_particles));
/**-----------------------------------------------------*/
/** allocating of the needed memory for several structs */
/**-----------------------------------------------------*/
/**Allocate struct for particle forces */
size_t size_of_forces = lbpar_gpu.number_of_particles * sizeof(LB_particle_force_gpu);
host_forces = realloc(host_forces, size_of_forces);
lbpar_gpu.your_seed = (unsigned int)i_random(max_ran);
LB_TRACE (fprintf(stderr,"test your_seed %u \n", lbpar_gpu.your_seed));
lb_realloc_particle_GPU(&lbpar_gpu, &host_data);
}
/** (Re-)initializes the fluid according to the given value of rho. */
void lb_reinit_fluid_gpu() {
lbpar_gpu.your_seed = (unsigned int)i_random(max_ran);
lb_init_GPU(&lbpar_gpu);
LB_TRACE (fprintf(stderr,"lb_reinit_fluid_gpu \n"));
}
/** (Re-)initializes the fluid. */
void lb_reinit_parameters_gpu() {
lbpar_gpu.mu = 0.0;
lbpar_gpu.time_step = (float)time_step;
if (lbpar_gpu.viscosity > 0.0) {
/* Eq. (80) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007). */
lbpar_gpu.gamma_shear = 1. - 2./(6.*lbpar_gpu.viscosity*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
if (lbpar_gpu.bulk_viscosity > 0.0) {
/* Eq. (81) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007). */
lbpar_gpu.gamma_bulk = 1. - 2./(9.*lbpar_gpu.bulk_viscosity*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
if (temperature > 0.0) { /* fluctuating hydrodynamics ? */
lbpar_gpu.fluct = 1;
LB_TRACE (fprintf(stderr, "fluct on \n"));
/* Eq. (51) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007).*/
/* Note that the modes are not normalized as in the paper here! */
lbpar_gpu.mu = (float)temperature/c_sound_sq*lbpar_gpu.tau*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid);
//lbpar_gpu->mu *= agrid*agrid*agrid; // Marcello's conjecture
/* lb_coupl_pref is stored in MD units (force)
* Eq. (16) Ahlrichs and Duenweg, JCP 111(17):8225 (1999).
* The factor 12 comes from the fact that we use random numbers
* from -0.5 to 0.5 (equally distributed) which have variance 1/12.
* time_step comes from the discretization.
*/
lbpar_gpu.lb_coupl_pref = sqrt(12.f*2.f*lbpar_gpu.friction*(float)temperature/lbpar_gpu.time_step);
lbpar_gpu.lb_coupl_pref2 = sqrt(2.f*lbpar_gpu.friction*(float)temperature/lbpar_gpu.time_step);
} else {
/* no fluctuations at zero temperature */
lbpar_gpu.fluct = 0;
lbpar_gpu.lb_coupl_pref = 0.0;
lbpar_gpu.lb_coupl_pref2 = 0.0;
}
LB_TRACE (fprintf(stderr,"lb_reinit_prarameters_gpu \n"));
}
/** Performs a full initialization of
* the Lattice Boltzmann system. All derived parameters
* and the fluid are reset to their default values. */
void lb_init_gpu() {
/** set parameters for transfer to gpu */
lb_reinit_parameters_gpu();
lb_realloc_particles_gpu();
lb_realloc_fluid_gpu();
lb_init_GPU(&lbpar_gpu);
LB_TRACE (fprintf(stderr,"lb_init_gpu \n"));
}
/** lattice boltzmann update gpu called from integrate.c
*/
void lattice_boltzmann_update_gpu() {
int factor = (int)round(lbpar_gpu.tau/time_step);
fluidstep += 1;
if (fluidstep>=factor) {
fluidstep=0;
lb_integrate_GPU();
LB_TRACE (fprintf(stderr,"lb_integrate_GPU \n"));
}
}
/** (Re-)initializes the fluid according to the given value of rho. */
void lb_reinit_fluid_gpu() {
//lbpar_gpu.your_seed = (unsigned int)i_random(max_ran);
lb_reinit_parameters_gpu();
//#ifdef SHANCHEN
// lb_calc_particle_lattice_ia_gpu();
// copy_forces_from_GPU();
//#endif
if(lbpar_gpu.number_of_nodes != 0){
lb_reinit_GPU(&lbpar_gpu);
lbpar_gpu.reinit = 1;
}
LB_TRACE (fprintf(stderr,"lb_reinit_fluid_gpu \n"));
}
/**copy forces from gpu to cpu and call mpi routines to add forces to particles
*/
void lb_send_forces_gpu(){
if (transfer_momentum_gpu) {
if(this_node == 0){
if (lbpar_gpu.number_of_particles) lb_copy_forces_GPU(host_forces);
LB_TRACE (fprintf(stderr,"lb_send_forces_gpu \n"));
#if 0
for (i=0;i<n_total_particles;i++) {
fprintf(stderr, "%i particle forces , %f %f %f \n", i, host_forces[i].f[0], host_forces[i].f[1], host_forces[i].f[2]);
}
#endif
}
mpi_send_forces_lb(host_forces);
}
}
/** Parser for the \ref lbfluid command gpu.
*/
int tclcommand_lbfluid_gpu(Tcl_Interp *interp, int argc, char **argv) {
#ifdef LB_GPU
int err = TCL_OK;
int change = 0;
while (argc > 0) {
if (ARG0_IS_S("grid") || ARG0_IS_S("agrid"))
err = lbfluid_parse_agrid(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("tau"))
err = lbfluid_parse_tau(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("density") || ARG0_IS_S("dens"))
err = lbfluid_parse_density(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("viscosity") || ARG0_IS_S("visc"))
err = lbfluid_parse_viscosity(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("bulk_viscosity") || ARG0_IS_S("b_visc"))
err = lbfluid_parse_bulk_visc(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("friction") || ARG0_IS_S("coupling"))
err = lbfluid_parse_friction(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("ext_force"))
err = lbfluid_parse_ext_force(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("gamma_odd"))
err = lbfluid_parse_gamma_odd(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("gamma_even"))
err = lbfluid_parse_gamma_even(interp, argc-1, argv+1, &change);
else {
Tcl_AppendResult(interp, "unknown feature \"", argv[0],"\" of lbfluid", (char *)NULL);
err = TCL_ERROR ;
}
if (err == TCL_ERROR) return TCL_ERROR;
argc -= (change + 1);
argv += (change + 1);
}
mpi_bcast_parameter(FIELD_LATTICE_SWITCH) ;
/* thermo_switch is retained for backwards compatibility */
thermo_switch = (thermo_switch | THERMO_LB);
mpi_bcast_parameter(FIELD_THERMO_SWITCH);
LB_TRACE (fprintf(stderr,"tclcommand_lbfluid_gpu parser ok \n"));
return err;
#else /* !defined LB_GPU */
Tcl_AppendResult(interp, "LB_GPU is not compiled in!", NULL);
return TCL_ERROR;
#endif
}
void lattice_boltzmann_update_gpu() {
int factor = (int)round(lbpar_gpu.tau/time_step);
fluidstep += 1;
if (fluidstep>=factor) {
fluidstep=0;
lb_integrate_GPU();
#ifdef SHANCHEN
if(lbpar_gpu.remove_momentum) lb_remove_fluid_momentum_GPU();
#endif
LB_TRACE (fprintf(stderr,"lb_integrate_GPU \n"));
}
}
/** Calculate particle lattice interactions called from forces.c
*/
void lb_calc_particle_lattice_ia_gpu() {
if (transfer_momentum_gpu) {
mpi_get_particles_lb(host_data);
if(this_node == 0){
#if 0
for (i=0;i<n_total_particles;i++) {
fprintf(stderr, "%i particle posi: , %f %f %f\n", i, host_data[i].p[0], host_data[i].p[1], host_data[i].p[2]);
}
#endif
if(lbpar_gpu.number_of_particles) lb_particle_GPU(host_data);
LB_TRACE (fprintf(stderr,"lb_calc_particle_lattice_ia_gpu \n"));
}
}
}
static int lbfluid_parse_agrid(Tcl_Interp *interp, int argc, char *argv[], int *change) {
double agrid;
if (argc < 1) {
Tcl_AppendResult(interp, "agrid requires 1 argument", (char *)NULL);
return TCL_ERROR;
}
if (!ARG0_IS_D(agrid)) {
Tcl_AppendResult(interp, "wrong argument for agrid", (char *)NULL);
return TCL_ERROR;
}
if (agrid <= 0.0) {
Tcl_AppendResult(interp, "agrid must be positive", (char *)NULL);
return TCL_ERROR;
}
*change = 1;
lbpar_gpu.agrid = (float)agrid;
lbpar_gpu.dim_x = (unsigned int)floor(box_l[0]/agrid);
lbpar_gpu.dim_y = (unsigned int)floor(box_l[1]/agrid);
lbpar_gpu.dim_z = (unsigned int)floor(box_l[2]/agrid);
unsigned int tmp[3];
tmp[0] = lbpar_gpu.dim_x;
tmp[1] = lbpar_gpu.dim_y;
tmp[2] = lbpar_gpu.dim_z;
/* sanity checks */
int dir;
for (dir=0;dir<3;dir++) {
/* check if box_l is compatible with lattice spacing */
if (fabs(box_l[dir]-tmp[dir]*agrid) > ROUND_ERROR_PREC) {
char *errtxt = runtime_error(128);
ERROR_SPRINTF(errtxt, "{097 Lattice spacing agrid=%f is incompatible with box_l[%i]=%f} ",agrid,dir,box_l[dir]);
}
}
lbpar_gpu.number_of_nodes = lbpar_gpu.dim_x * lbpar_gpu.dim_y * lbpar_gpu.dim_z;
LB_TRACE (printf("#nodes \t %u \n", lbpar_gpu.number_of_nodes));
return TCL_OK;
}
/** printing the hole fluid field to file with order x+y*dim_x+z*dim_x*dim_y */
int tclcommand_lbprint_gpu(ClientData data, Tcl_Interp *interp, int argc, char **argv) {
int err = TCL_OK;
int change = 0;
int vtk = 0;
argc--; argv++;
if (argc < 1) {
Tcl_AppendResult(interp, "too few arguments to \"lbprint\"", (char *)NULL);
err = TCL_ERROR;
}
else while (argc > 0) {
if (ARG0_IS_S("u") || ARG0_IS_S("velocity") || ARG0_IS_S("v")){
argc--; argv++;
if (ARG0_IS_S("vtk")){
vtk = 1;
err = lbprint_parse_velocity(interp, argc-1, argv+1, &change, vtk);
}
else
err = lbprint_parse_velocity(interp, argc, argv, &change, vtk);
}
else if (ARG0_IS_S("rho") || ARG0_IS_S("density"))
err = lbprint_parse_density(interp, argc-1, argv+1, &change);
else if (ARG0_IS_S("stresstensor")){
//err = lbprint_parse_stresstensor(interp, argc-1, argv+1, &change);
Tcl_AppendResult(interp, "\"lbprint stresstensor\" is not available by default due to memory saving, pls ensure availablity of pi[6] (see lbgpu.h) and lbprint_parse_stresstensor()", (char *)NULL);
err = TCL_ERROR;
}
else {
Tcl_AppendResult(interp, "unknown feature \"", argv[0],"\" of lbprint", (char *)NULL);
err = TCL_ERROR ;
}
argc -= (change + 1);
argv += (change + 1);
LB_TRACE (fprintf(stderr,"tclcommand_lbprint_gpu parser ok \n"));
}
return err;
}
/** (Re-)initializes the fluid. */
void lb_reinit_parameters_gpu() {
int ii;
lbpar_gpu.time_step = (float)time_step;
for(ii=0;ii<LB_COMPONENTS;++ii){
lbpar_gpu.mu[ii] = 0.0;
if (lbpar_gpu.viscosity[ii] > 0.0) {
/* Eq. (80) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007). */
lbpar_gpu.gamma_shear[ii] = 1. - 2./(6.*lbpar_gpu.viscosity[ii]*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
if (lbpar_gpu.bulk_viscosity[ii] > 0.0) {
/* Eq. (81) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007). */
lbpar_gpu.gamma_bulk[ii] = 1. - 2./(9.*lbpar_gpu.bulk_viscosity[ii]*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
#ifdef SHANCHEN
if (lbpar_gpu.mobility[0] > 0.0) {
lbpar_gpu.gamma_mobility[0] = 1. - 2./(6.*lbpar_gpu.mobility[0]*lbpar_gpu.tau/(lbpar_gpu.agrid*lbpar_gpu.agrid) + 1.);
}
#endif
if (temperature > 0.0) { /* fluctuating hydrodynamics ? */
lbpar_gpu.fluct = 1;
LB_TRACE (fprintf(stderr, "fluct on \n"));
/* Eq. (51) Duenweg, Schiller, Ladd, PRE 76(3):036704 (2007).*/
/* Note that the modes are not normalized as in the paper here! */
lbpar_gpu.mu[ii] = (float)temperature*lbpar_gpu.tau*lbpar_gpu.tau/c_sound_sq/(lbpar_gpu.agrid*lbpar_gpu.agrid);
/* lb_coupl_pref is stored in MD units (force)
* Eq. (16) Ahlrichs and Duenweg, JCP 111(17):8225 (1999).
* The factor 12 comes from the fact that we use random numbers
* from -0.5 to 0.5 (equally distributed) which have variance 1/12.
* time_step comes from the discretization.
*/
lbpar_gpu.lb_coupl_pref[ii] = sqrt(12.f*2.f*lbpar_gpu.friction[ii]*(float)temperature/lbpar_gpu.time_step);
lbpar_gpu.lb_coupl_pref2[ii] = sqrt(2.f*lbpar_gpu.friction[ii]*(float)temperature/lbpar_gpu.time_step);
} else {
/* no fluctuations at zero temperature */
lbpar_gpu.fluct = 0;
lbpar_gpu.lb_coupl_pref[ii] = 0.0;
lbpar_gpu.lb_coupl_pref2[ii] = 0.0;
}
LB_TRACE (fprintf(stderr,"lb_reinit_prarameters_gpu \n"));
}
#ifdef ELECTROKINETICS
if (ek_initialized) {
lbpar_gpu.dim_x = (unsigned int) round(box_l[0] / lbpar_gpu.agrid); //TODO code duplication with lb.c start
lbpar_gpu.dim_y = (unsigned int) round(box_l[1] / lbpar_gpu.agrid);
lbpar_gpu.dim_z = (unsigned int) round(box_l[2] / lbpar_gpu.agrid);
unsigned int tmp[3];
tmp[0] = lbpar_gpu.dim_x;
tmp[1] = lbpar_gpu.dim_y;
tmp[2] = lbpar_gpu.dim_z;
/* sanity checks */
int dir;
for (dir=0;dir<3;dir++) {
/* check if box_l is compatible with lattice spacing */
if (fabs(box_l[dir] - tmp[dir] * lbpar_gpu.agrid) > 1.0e-3) {
ostringstream msg;
msg <<"Lattice spacing lbpar_gpu.agrid= "<< lbpar_gpu.agrid << " is incompatible with box_l[" << dir << "]=" << box_l[dir];
runtimeError(msg);
}
}
lbpar_gpu.number_of_nodes = lbpar_gpu.dim_x * lbpar_gpu.dim_y * lbpar_gpu.dim_z;
lbpar_gpu.tau = (float) time_step; //TODO code duplication with lb.c end
}
#endif
LB_TRACE (fprintf(stderr,"lb_reinit_prarameters_gpu \n"));
reinit_parameters_GPU(&lbpar_gpu);
}
