ocbbs.cpp

Go to the documentation of this file.

#include <list>
#include <InterViews/resource.h>
 
#include "classreg.h"
#include "code.h"
#include "ivocvect.h"
#include "hoclist.h"
#include "bbs.h"
#include "bbsimpl.h"
#include "section.h"
#include "membfunc.h"
#include "multicore.h"
#include "nrnpy.h"
#include "utils/profile/profiler_interface.h"
#include "node_order_optim/node_order_optim.h"
#include <nrnmpi.h>
#include <cerrno>
 
#undef MD
#define MD 2147483647.
 
extern int vector_arg_px(int, double**);
Symbol* hoc_which_template(Symbol*);
extern double t;
extern void nrnmpi_source_var(), nrnmpi_target_var(), nrnmpi_setup_transfer();
extern int nrnmpi_spike_compress(int nspike, bool gid_compress, int xchng_meth);
extern int nrnmpi_splitcell_connect(int that_host);
extern int nrnmpi_multisplit(Section*, double x, int sid, int backbonestyle);
extern int nrn_set_timeout(int timeout);
extern void nrnmpi_gid_clear(int);
double nrnmpi_rtcomp_time_;
extern double nrn_multisend_receive_time(int);
extern void nrn_prcellstate(int gid, const char* suffix);
double nrnmpi_step_wait_;
#if NRNMPI
double nrnmpi_transfer_wait_;
double nrnmpi_splitcell_wait_;
#endif
#if NRNMPI
#include "nrnmpidec.h"
#else
static void nrnmpi_int_broadcast(int*, int, int) {}
static void nrnmpi_char_broadcast(char*, int, int) {}
static void nrnmpi_dbl_broadcast(double*, int, int) {}
#endif
extern double* nrn_mech_wtime_;
 
extern bool nrn_trajectory_request_per_time_step_;
 
extern size_t nrncore_write();
extern size_t nrnbbcore_register_mapping();
extern int nrncore_run(const char*);
extern int nrncore_is_enabled();
extern int nrncore_is_file_mode();
extern int nrncore_psolve(double tstop, int file_mode);
 
class OcBBS final: public BBS, public Resource {
  public:
    OcBBS(int nhost_request);
 
  public:
    double retval_ = 0.;
    int userid_ = 0;
    int next_local_ = 0;
};
 
OcBBS::OcBBS(int n)
    : BBS(n) {}
 
static bool posting_ = false;
static void pack_help(int, OcBBS*);
static void unpack_help(int, OcBBS*);
static int submit_help(OcBBS*);
static char* key_help();
 
void bbs_done() {
#if USEBBS
    Symbol* sym = hoc_lookup("ParallelContext");
    sym = hoc_which_template(sym);
    hoc_Item *q, *ql;
    ql = sym->u.ctemplate->olist;
    q = ql->next;
    if (q != ql) {
        Object* ob = OBJ(q);
        OcBBS* bbs = (OcBBS*) ob->u.this_pointer;
        if (bbs->is_master()) {
            bbs->done();
        }
    }
#endif
}
 
static int submit_help(OcBBS* bbs) {
    int id, i, firstarg, style;
    posting_ = true;
    bbs->pkbegin();
    i = 1;
    if (hoc_is_double_arg(i)) {
        bbs->pkint((id = (int) chkarg(i++, 0, MD)));
    } else {
        bbs->pkint((id = --bbs->next_local_));
    }
    bbs->pkint(0);  // space for working_id
    if (ifarg(i + 1)) {
#if 1
        int argtypes = 0;
        int ii = 1;
        if (hoc_is_str_arg(i)) {
            style = 1;
            bbs->pkint(style);  // "fname", arg1, ... style
            bbs->pkstr(gargstr(i++));
        } else {
            Object* ob = *hoc_objgetarg(i++);
            std::vector<char> pname{};
            if (neuron::python::methods.po2pickle) {
                pname = neuron::python::methods.po2pickle(ob);
            }
            if (!pname.empty()) {
                style = 3;
                bbs->pkint(style);  // pyfun, arg1, ... style
                bbs->pkpickle(pname);
            } else {
                style = 2;
                bbs->pkint(style);  // [object],"fname", arg1, ... style
                bbs->pkstr(ob->ctemplate->sym->name);
                bbs->pkint(ob->index);
                // printf("ob=%s\n", hoc_object_name(ob));
                bbs->pkstr(gargstr(i++));
            }
        }
        firstarg = i;
        for (; ifarg(i); ++i) {  // first is least significant
            if (hoc_is_double_arg(i)) {
                argtypes += 1 * ii;
            } else if (hoc_is_str_arg(i)) {
                argtypes += 2 * ii;
            } else if (is_vector_arg(i)) {  // hoc Vector
                argtypes += 3 * ii;
            } else {  // must be a PythonObject
                argtypes += 4 * ii;
            }
            ii *= 5;
        }
        // printf("submit style %d %s argtypes=%o\n", style, gargstr(firstarg-1), argtypes);
        bbs->pkint(argtypes);
        pack_help(firstarg, bbs);
#endif
    } else {
        if (hoc_is_str_arg(i)) {
            bbs->pkint(0);  // hoc statement style
            bbs->pkstr(gargstr(i));
        } else if (neuron::python::methods.po2pickle) {
            auto pname = neuron::python::methods.po2pickle(*hoc_objgetarg(i));
            bbs->pkint(3);  // pyfun with no arg style
            bbs->pkpickle(pname);
            bbs->pkint(0);  // argtypes
        }
    }
    posting_ = false;
    return id;
}
 
static double submit(void* v) {
    int id;
    OcBBS* bbs = (OcBBS*) v;
    id = submit_help(bbs);
    bbs->submit(id);
    return double(id);
}
 
static double context(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    submit_help(bbs);
    //  printf("%d context %s %s\n", bbs->myid(), hoc_object_name(*hoc_objgetarg(1)), gargstr(2));
    bbs->context();
    return 1.;
}
 
static double working(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    int id;
    bool b = bbs->working(id, bbs->retval_, bbs->userid_);
    hoc_return_type_code = HocReturnType::integer;
    if (b) {
        return double(id);
    } else {
        return 0.;
    }
}
 
static double retval(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->retval_;
}
 
static double userid(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return (double) bbs->userid_;
}
 
static double nhost(void* v) {
    hoc_return_type_code = HocReturnType::integer;
    return nrnmpi_numprocs;
}
 
static double nrn_rank(void* v) {
    hoc_return_type_code = HocReturnType::integer;
    return nrnmpi_myid;
}
 
static double nhost_world(void* v) {
    hoc_return_type_code = HocReturnType::integer;
    return nrnmpi_numprocs_world;
}
 
static double rank_world(void* v) {
    hoc_return_type_code = HocReturnType::integer;
    return nrnmpi_myid_world;
}
 
static double nhost_bbs(void* v) {
    hoc_return_type_code = HocReturnType::integer;
    return nrnmpi_numprocs_bbs;
}
 
static double rank_bbs(void* v) {
    hoc_return_type_code = HocReturnType::integer;
    return nrnmpi_myid_bbs;
}
 
static double subworlds(void* v) {
    int n = int(chkarg(1, 1, nrnmpi_numprocs_world));
#if NRNMPI
    nrnmpi_subworld_size(n);
#endif
    return 0.;
}
 
static double worker(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    bbs->worker();
    return 0.;
}
 
static double master_works(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    bbs->master_works(int(chkarg(1, 0, 1)));
    return 0.;
}
 
static double done(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    bbs->done();
    return 0.;
}
 
static void pack_help(int i, OcBBS* bbs) {
    if (!posting_) {
        bbs->pkbegin();
        posting_ = true;
    }
    for (; ifarg(i); ++i) {
        if (hoc_is_double_arg(i)) {
            bbs->pkdouble(*getarg(i));
        } else if (hoc_is_str_arg(i)) {
            bbs->pkstr(gargstr(i));
        } else if (is_vector_arg(i)) {
            int n;
            double* px;
            n = vector_arg_px(i, &px);
            bbs->pkint(n);
            bbs->pkvec(n, px);
        } else {  // must be a PythonObject
            auto s = neuron::python::methods.po2pickle(*hoc_objgetarg(i));
            bbs->pkpickle(s);
        }
    }
}
 
static double pack(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    pack_help(1, bbs);
    return 0.;
}
 
static double post(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    pack_help(2, bbs);
    posting_ = false;
    if (hoc_is_str_arg(1)) {
        bbs->post(gargstr(1));
    } else {
        char key[50];
        Sprintf(key, "%g", *getarg(1));
        bbs->post(key);
    }
    return 1.;
}
 
static void unpack_help(int i, OcBBS* bbs) {
    for (; ifarg(i); ++i) {
        if (hoc_is_pdouble_arg(i)) {
            *hoc_pgetarg(i) = bbs->upkdouble();
        } else if (hoc_is_str_arg(i)) {
            char* s = bbs->upkstr();
            char** ps = hoc_pgargstr(i);
            hoc_assign_str(ps, s);
            delete[] s;
        } else if (is_vector_arg(i)) {
            Vect* vec = vector_arg(i);
            int n = bbs->upkint();
            vec->resize(n);
            bbs->upkvec(n, vec->data());
        } else {
            hoc_execerror("pc.unpack can only unpack str, scalar, or Vector.",
                          "use pc.upkpyobj to unpack a Python Object");
        }
    }
}
 
static double unpack(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    unpack_help(1, bbs);
    return 1.;
}
 
static double upkscalar(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->upkdouble();
}
 
static const char** upkstr(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    char* s = bbs->upkstr();
    char** ps = hoc_pgargstr(1);
    hoc_assign_str(ps, s);
    delete[] s;
    return (const char**) ps;
}
 
static Object** upkvec(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    Vect* vec;
    int n = bbs->upkint();
    if (ifarg(1)) {
        vec = vector_arg(1);
        vec->resize(n);
    } else {
        vec = new Vect(n);
    }
    bbs->upkvec(n, vec->data());
    return vec->temp_objvar();
}
 
static Object** upkpyobj(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    std::vector<char> s = bbs->upkpickle();
    assert(neuron::python::methods.pickle2po);
    Object* po = neuron::python::methods.pickle2po(s);
    --po->refcount;
    return hoc_temp_objptr(po);
}
 
static Object** pyret(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->pyret();
}
Object** BBS::pyret() {
    assert(neuron::python::methods.pickle2po);
    Object* po = neuron::python::methods.pickle2po(impl_->pickle_ret_);
    impl_->pickle_ret_.clear();
    --po->refcount;
    return hoc_temp_objptr(po);
}
 
static Object** py_alltoall_type(int type) {
    assert(neuron::python::methods.mpi_alltoall_type);
    // for py_gather, py_broadcast, and py_scatter,
    // the second arg refers to the root rank of the operation (default 0)
    int size = 0;
    if (ifarg(2)) {
        size = int(chkarg(2, -1, 2.14748e9));
    }
    Object* po = neuron::python::methods.mpi_alltoall_type(size, type);
    return hoc_temp_objptr(po);
}
 
static Object** py_alltoall(void*) {
    return py_alltoall_type(1);
}
 
static Object** py_allgather(void*) {
    return py_alltoall_type(2);
}
 
static Object** py_gather(void*) {
    return py_alltoall_type(3);
}
 
static Object** py_broadcast(void*) {
    return py_alltoall_type(4);
}
 
static Object** py_scatter(void*) {
    return py_alltoall_type(5);
}
 
static char* key_help() {
    static char key[50];
    if (hoc_is_str_arg(1)) {
        return gargstr(1);
    } else {
        Sprintf(key, "%g", *getarg(1));
        return key;
    }
}
 
static double take(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    bbs->take(key_help());
    unpack_help(2, bbs);
    return 1.;
}
 
static double look(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    hoc_return_type_code = HocReturnType::boolean;
    if (bbs->look(key_help())) {
        unpack_help(2, bbs);
        return 1.;
    }
    return 0.;
}
 
static double look_take(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    hoc_return_type_code = HocReturnType::boolean;
    if (bbs->look_take(key_help())) {
        unpack_help(2, bbs);
        return 1.;
    }
    return 0.;
}
 
static double pctime(void* v) {
    return ((OcBBS*) v)->time();
}
 
static double vtransfer_time(void* v) {
    int mode = ifarg(1) ? int(chkarg(1, 0., 2.)) : 0;
    if (mode == 2) {
        return nrnmpi_rtcomp_time_;
#if NRNMPI
    } else if (mode == 1) {
        return nrnmpi_splitcell_wait_;
    } else {
        return nrnmpi_transfer_wait_;
    }
#else
    }
    return 0;
#endif
}
 
static double mech_time(void* v) {
    if (ifarg(1)) {
        if (nrn_mech_wtime_) {
            int i = (int) chkarg(1, 0, n_memb_func - 1);
            return nrn_mech_wtime_[i];
        }
    } else {
        if (!nrn_mech_wtime_) {
            nrn_mech_wtime_ = new double[n_memb_func];
        }
        for (int i = 0; i < n_memb_func; ++i) {
            nrn_mech_wtime_[i] = 0.0;
        }
    }
    return 0;
}
 
static double prcellstate(void* v) {
    nrn_prcellstate(int(*hoc_getarg(1)), hoc_gargstr(2));
    return 0;
}
 
static double wait_time(void* v) {
    double w = ((OcBBS*) v)->wait_time();
    return w;
}
 
static double step_time(void* v) {
    double w = ((OcBBS*) v)->integ_time();
#if NRNMPI
    w -= nrnmpi_transfer_wait_ + nrnmpi_splitcell_wait_;
#endif
    return w;
}
 
static double step_wait(void* v) {
    if (ifarg(1)) {
        nrnmpi_step_wait_ = chkarg(1, -1.0, 0.0);
    }
    double w = nrnmpi_step_wait_;
#if NRNMPI
    // sadly, no calculation of transfer and multisplit barrier times.
#endif
    if (w < 0.) {
        w = 0.0;
    }
    return w;
}
 
static double send_time(void* v) {
    int arg = ifarg(1) ? int(chkarg(1, 0, 20)) : 0;
    if (arg) {
        return nrn_multisend_receive_time(arg);
    }
    return ((OcBBS*) v)->send_time();
}
 
static double event_time(void* v) {
    return 0.;
}
 
static double integ_time(void* v) {
    return 0.;
}
 
static double set_gid2node(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    bbs->set_gid2node(int(chkarg(1, 0, MD)), int(chkarg(2, 0, MD)));
    return 0.;
}
 
static double gid_exists(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    hoc_return_type_code = HocReturnType::integer;
    return int(bbs->gid_exists(int(chkarg(1, 0, MD))));
}
 
static double cell(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    bbs->cell();
    return 0.;
}
 
static double threshold(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->threshold();
}
 
static double spcompress(void* v) {
    int nspike = -1;
    bool gid_compress = true;
    int xchng_meth = 0;
    if (ifarg(1)) {
        nspike = (int) chkarg(1, -1, MD);
    }
    if (ifarg(2)) {
        gid_compress = (chkarg(2, 0, 1) ? true : false);
    }
    if (ifarg(3)) {
        xchng_meth = (int) chkarg(3, 0, 15);
    }
    return (double) nrnmpi_spike_compress(nspike, gid_compress, xchng_meth);
}
 
static double splitcell_connect(void* v) {
    int that_host = (int) chkarg(1, 0, nrnmpi_numprocs - 1);
    // also needs a currently accessed section that is the root of this_tree
    nrnmpi_splitcell_connect(that_host);
    return 0.;
}
 
static double multisplit(void* v) {
    double x = -1.;
    Section* sec = NULL;
    int sid = -1;
    int backbone_style = 2;
    if (ifarg(1)) {
        nrn_seg_or_x_arg(1, &sec, &x);
        sid = (int) chkarg(2, 0, (double) (0x7fffffff));
    }
    if (ifarg(3)) {
        backbone_style = (int) chkarg(3, 0, 2);
    }
    // also needs a currently accessed section
    nrnmpi_multisplit(sec, x, sid, backbone_style);
    return 0.;
}
 
static double set_timeout(void* v) {
    int arg = 0;
    if (ifarg(1)) {
        arg = int(chkarg(1, 0, 10000));
    }
    arg = nrn_set_timeout(arg);
    return double(arg);
}
 
static double set_mpiabort_on_error(void*) {
    double ret = double(nrn_mpiabort_on_error_);
    if (ifarg(1)) {
        nrn_mpiabort_on_error_ = int(chkarg(1, 0, 1));
    }
    return ret;
}
 
static double gid_clear(void* v) {
    int arg = 0;
    if (ifarg(1)) {
        arg = int(chkarg(1, 0, 4));
    }
    nrnmpi_gid_clear(arg);
    return 0.;
}
 
static double outputcell(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    int gid = int(chkarg(1, 0., MD));
    bbs->outputcell(gid);
    return 0.;
}
 
static double spike_record(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    IvocVect* spikevec = vector_arg(2);
    IvocVect* gidvec = vector_arg(3);
    if (hoc_is_object_arg(1) && is_vector_arg(1)) {
        IvocVect* gids = vector_arg(1);
        bbs->spike_record(gids, spikevec, gidvec);
    } else {
        int gid = int(chkarg(1, -1., MD));
        bbs->spike_record(gid, spikevec, gidvec);
    }
    return 0.;
}
 
static double psolve(void* v) {
    nrn::Instrumentor::phase_begin("psolve");
    OcBBS* bbs = (OcBBS*) v;
    double tstop = chkarg(1, t, 1e9);
    int enabled = nrncore_is_enabled();
    int file_mode = nrncore_is_file_mode();
    if (enabled == 1) {
        nrncore_psolve(tstop, file_mode);
    } else if (enabled == 0) {
        // Classic case
        bbs->netpar_solve(tstop);
    }
    nrn::Instrumentor::phase_end("psolve");
    return double(enabled);
}
 
static double set_maxstep(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->netpar_mindelay(chkarg(1, 1e-6, 1e9));
}
 
static double spike_stat(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    int nsend, nsendmax, nrecv, nrecv_useful;
    hoc_return_type_code = HocReturnType::integer;
    nsend = nsendmax = nrecv = nrecv_useful = 0;
    bbs->netpar_spanning_statistics(&nsend, &nsendmax, &nrecv, &nrecv_useful);
    if (ifarg(1)) {
        *hoc_pgetarg(1) = nsend;
    }
    if (ifarg(2)) {
        *hoc_pgetarg(2) = nrecv;
    }
    if (ifarg(3)) {
        *hoc_pgetarg(3) = nrecv_useful;
    }
    return double(nsendmax);
}
 
static double maxhist(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    IvocVect* vec = ifarg(1) ? vector_arg(1) : nullptr;
    if (vec) {
        hoc_obj_ref(vec->obj_);
    }
    vec = bbs->netpar_max_histogram(vec);
    if (vec) {
        hoc_obj_unref(vec->obj_);
    }
    return 0.;
}
 
static double source_var(void*) {  // &source_variable, source_global_index
    // At BEFORE BREAKPOINT, the value of variable is sent to the
    // target machine(s).  This can only be executed on the
    // source machine (where source_variable exists).
    nrnmpi_source_var();
    return 0.;
}
 
static double target_var(void*) {  // &target_variable, source_global_index
    // At BEFORE BREAKPOINT, the value of the target_variable is set
    // to the value of the source variable associated
    // with the source_global_index.  This can only be executed on the
    // target machine (where target_variable exists).
    nrnmpi_target_var();
    return 0.;
}
 
static double setup_transfer(void*) {  // after all source/target and before init and run
    nrnmpi_setup_transfer();
    return 0.;
}
 
static double barrier(void*) {
    // return wait time
    double t = 0.;
#if NRNMPI
    if (nrnmpi_numprocs > 1) {
        t = nrnmpi_wtime();
        nrnmpi_barrier();
        t = nrnmpi_wtime() - t;
    }
    errno = 0;
#endif
    return t;
}
 
static double allreduce(void*) {
    // type 1,2,3 sum, max, min
    if (hoc_is_object_arg(1)) {
        Vect* vec = vector_arg(1);
        int n = vec->size();
        if (n == 0) {
            return 0.0;
        }
#if NRNMPI
        if (nrnmpi_numprocs > 1) {
            int type = (int) chkarg(2, 1, 3);
            double* px = vector_vec(vec);
            double* dest = new double[n];
            nrnmpi_dbl_allreduce_vec(px, dest, n, type);
            for (int i = 0; i < n; ++i) {
                px[i] = dest[i];
            }
            delete[] dest;
        }
        errno = 0;
#endif
        return 0.;
    } else {
        double val = *getarg(1);
#if NRNMPI
        if (nrnmpi_numprocs > 1) {
            int type = (int) chkarg(2, 1, 3);
            val = nrnmpi_dbl_allreduce(val, type);
        }
        errno = 0;
#endif
        return val;
    }
}
 
static double allgather(void*) {
    double val = *getarg(1);
    Vect* vec = vector_arg(2);
    vector_resize(vec, nrnmpi_numprocs);
    double* px = vector_vec(vec);
 
#if NRNMPI
    if (nrnmpi_numprocs > 1) {
        nrnmpi_dbl_allgather(&val, px, 1);
        errno = 0;
    } else {
        px[0] = val;
    }
#else
    px[0] = val;
#endif
    return 0.;
}
 
// This function takes 3 arguments:
//   - vsrc (In)
//   - vscnt (In)
//   - vdest (Out)
static double alltoall(void*) {
    int np = nrnmpi_numprocs;
    const Vect* vsrc = vector_arg(1);
    const Vect* vscnt = vector_arg(2);
    Vect* vdest = vector_arg(3);
    std::size_t ns = vsrc->size();
    if (vscnt->size() != np) {
        hoc_execerror("size of source counts vector is not nhost", nullptr);
    }
    const std::vector<int> scnt(vscnt->begin(), vscnt->end());  // cast from double to int
    std::vector<int> sdispl(np + 1);
    for (int i = 0; i < np; ++i) {
        sdispl[i + 1] = sdispl[i] + scnt[i];
    }
    if (ns != sdispl[np]) {
        hoc_execerror("sum of source counts is not the size of the src vector", nullptr);
    }
    if (nrnmpi_numprocs > 1) {
#if NRNMPI
        std::vector<int> rcnt(np);
        std::vector<int> c(np, 1);
        std::vector<int> rdispl(np + 1);
        std::iota(rdispl.begin(), rdispl.end(), 0);
 
        nrnmpi_int_alltoallv(
            scnt.data(), c.data(), rdispl.data(), rcnt.data(), c.data(), rdispl.data());
        for (int i = 0; i < np; ++i) {
            rdispl[i + 1] = rdispl[i] + rcnt[i];
        }
        vdest->resize(rdispl[np]);
        nrnmpi_dbl_alltoallv(
            vsrc->data(), scnt.data(), sdispl.data(), vdest->data(), rcnt.data(), rdispl.data());
#endif
    } else {
        vdest->resize(ns);
        std::copy(vsrc->begin(), vsrc->end(), vdest->begin());
    }
    return 0.;
}
 
static double broadcast(void*) {
    int srcid = int(chkarg(2, 0, nrnmpi_numprocs - 1));
    int cnt = 0;
#if NRNMPI
    if (nrnmpi_numprocs > 1) {
        if (hoc_is_str_arg(1)) {
            char* s;
            if (srcid == nrnmpi_myid) {
                s = gargstr(1);
                cnt = strlen(s) + 1;
            }
            nrnmpi_int_broadcast(&cnt, 1, srcid);
            if (srcid != nrnmpi_myid) {
                s = new char[cnt];
            }
            nrnmpi_char_broadcast(s, cnt, srcid);
            if (srcid != nrnmpi_myid) {
                hoc_assign_str(hoc_pgargstr(1), s);
                delete[] s;
            }
        } else {
            Vect* vec = vector_arg(1);
            if (srcid == nrnmpi_myid) {
                cnt = vec->size();
            }
            nrnmpi_int_broadcast(&cnt, 1, srcid);
            if (srcid != nrnmpi_myid) {
                vec->resize(cnt);
            }
            nrnmpi_dbl_broadcast(vector_vec(vec), cnt, srcid);
        }
    } else {
#else
    {
#endif
        if (hoc_is_str_arg(1)) {
            cnt = strlen(gargstr(1));
        } else {
            cnt = vector_arg(1)->size();
        }
    }
    return double(cnt);
}
 
static double nthrd(void*) {
    bool ip{true};
    hoc_return_type_code = HocReturnType::integer;
    if (ifarg(1)) {
        if (ifarg(2)) {
            ip = bool(chkarg(2, 0, 1));
        }
        nrn_threads_create(int(chkarg(1, 1, 1e5)), ip);
    }
    return double(nrn_nthread);
}
 
static double number_of_worker_threads(void*) {
    hoc_return_type_code = HocReturnType::integer;
    return nof_worker_threads();
}
 
static double partition(void*) {
    Object* ob = 0;
    int it;
    if (ifarg(2)) {
        ob = *hoc_objgetarg(2);
        if (ob) {
            check_obj_type(ob, "SectionList");
        }
    }
    if (ifarg(1)) {
        it = (int) chkarg(1, 0, nrn_nthread - 1);
        nrn_thread_partition(it, ob);
    } else {
        for (it = 0; it < nrn_nthread; ++it) {
            nrn_thread_partition(it, ob);
        }
    }
    return 0.0;
}
 
static Object** get_partition(void*) {
    return nrn_get_thread_partition(int(chkarg(1, 0, nrn_nthread - 1)));
    ;
}
 
static double thread_stat(void*) {
    // nrn_thread_stat was called here but didn't do anything
    return 0.0;
}
 
static double thread_busywait(void*) {
    int old = nrn_allow_busywait(int(chkarg(1, 0, 1)));
    return double(old);
}
 
static double thread_how_many_proc(void*) {
    hoc_return_type_code = HocReturnType::integer;
    int i = nrn_how_many_processors();
    return double(i);
}
 
static double optimize_node_order(void*) {
    hoc_return_type_code = HocReturnType::integer;
    if (ifarg(1)) {
        neuron::nrn_optimize_node_order(int(chkarg(1, 0, 2)));
    }
    return double(neuron::interleave_permute_type);
}
 
static double sec_in_thread(void*) {
    hoc_return_type_code = HocReturnType::boolean;
    Section* sec = chk_access();
    return double(sec->pnode[0]->_nt->id);
}
 
static double thread_ctime(void*) {
    int i;
#if 1
    if (ifarg(1)) {
        i = int(chkarg(1, 0, nrn_nthread));
        return nrn_threads[i]._ctime;
    } else {
        for (i = 0; i < nrn_nthread; ++i) {
            nrn_threads[i]._ctime = 0.0;
        }
    }
#endif
    return 0.0;
}
 
static double nrn_thread_t(void*) {
    int i;
    i = int(chkarg(1, 0, nrn_nthread));
    return nrn_threads[i]._t;
}
 
static double thread_dt(void*) {
    int i;
    i = int(chkarg(1, 0, nrn_nthread));
    return nrn_threads[i]._dt;
}
 
static double nrncorewrite_argvec(void*) {
    if (ifarg(2) && !(hoc_is_object_arg(2) && is_vector_arg(2))) {
        hoc_execerror("nrnbbcore_write: optional second arg is not a Vector", nullptr);
    }
    return double(nrncore_write());
}
 
static double print_memory_stats(void*) {
    neuron::container::MemoryUsage local_memory_usage = neuron::container::local_memory_usage();
 
#if NRNMPI
    neuron::container::MemoryStats memory_stats;
    nrnmpi_memory_stats(memory_stats, local_memory_usage);
    nrnmpi_print_memory_stats(memory_stats);
#else
    print_memory_usage(local_memory_usage);
#endif
 
    return 1.0;
}
 
static double nrncorewrite_argappend(void*) {
    if (ifarg(2) && !hoc_is_double_arg(2)) {
        hoc_execerror(
            "nrncore_write: optional second arg is not a number (True or False append flag)",
            nullptr);
    }
    return double(nrncore_write());
}
 
static double nrncorerun(void*) {
    if (ifarg(2)) {
        nrn_trajectory_request_per_time_step_ = chkarg(2, 0., 1.) != 0.0;
    }
    return double(nrncore_run(gargstr(1)));
}
 
static double nrnbbcore_register_mapping(void*) {
    return double(nrnbbcore_register_mapping());
}
 
static Object** gid2obj(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->gid2obj(int(chkarg(1, 0, MD)));
}
 
static Object** gid2cell(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->gid2cell(int(chkarg(1, 0, MD)));
}
 
static Object** gid_connect(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    return bbs->gid_connect(int(chkarg(1, 0, MD)));
}
 
static Member_func members[] = {{"submit", submit},
                                {"working", working},
                                {"retval", retval},
                                {"userid", userid},
                                {"pack", pack},
                                {"post", post},
                                {"unpack", unpack},
                                {"upkscalar", upkscalar},
                                {"take", take},
                                {"look", look},
                                {"look_take", look_take},
                                {"runworker", worker},
                                {"master_works_on_jobs", master_works},
                                {"done", done},
                                {"id", nrn_rank},
                                {"nhost", nhost},
                                {"id_world", rank_world},
                                {"nhost_world", nhost_world},
                                {"id_bbs", rank_bbs},
                                {"nhost_bbs", nhost_bbs},
                                {"subworlds", subworlds},
                                {"context", context},
 
                                {"time", pctime},
                                {"wait_time", wait_time},
                                {"step_time", step_time},
                                {"step_wait", step_wait},
                                {"send_time", send_time},
                                {"event_time", event_time},
                                {"integ_time", integ_time},
                                {"vtransfer_time", vtransfer_time},
                                {"mech_time", mech_time},
                                {"timeout", set_timeout},
                                {"mpiabort_on_error", set_mpiabort_on_error},
 
                                {"set_gid2node", set_gid2node},
                                {"gid_exists", gid_exists},
                                {"outputcell", outputcell},
                                {"cell", cell},
                                {"threshold", threshold},
                                {"spike_record", spike_record},
                                {"psolve", psolve},
                                {"set_maxstep", set_maxstep},
                                {"spike_statistics", spike_stat},
                                {"max_histogram", maxhist},
                                {"spike_compress", spcompress},
                                {"gid_clear", gid_clear},
                                {"prcellstate", prcellstate},
 
                                {"source_var", source_var},
                                {"target_var", target_var},
                                {"setup_transfer", setup_transfer},
                                {"splitcell_connect", splitcell_connect},
                                {"multisplit", multisplit},
 
                                {"barrier", barrier},
                                {"allreduce", allreduce},
                                {"allgather", allgather},
                                {"alltoall", alltoall},
                                {"broadcast", broadcast},
 
                                {"nthread", nthrd},
                                {"nworker", number_of_worker_threads},
                                {"partition", partition},
                                {"thread_stat", thread_stat},
                                {"thread_busywait", thread_busywait},
                                {"thread_how_many_proc", thread_how_many_proc},
                                {"optimize_node_order", optimize_node_order},
                                {"sec_in_thread", sec_in_thread},
                                {"thread_ctime", thread_ctime},
                                {"dt", thread_dt},
                                {"t", nrn_thread_t},
 
                                {"nrnbbcore_write", nrncorewrite_argvec},
                                {"nrncore_write", nrncorewrite_argappend},
                                {"nrnbbcore_register_mapping", nrnbbcore_register_mapping},
                                {"nrncore_run", nrncorerun},
                                {"print_memory_stats", print_memory_stats},
 
                                {0, 0}};
 
static Member_ret_str_func retstr_members[] = {{"upkstr", upkstr}, {0, 0}};
 
static Member_ret_obj_func retobj_members[] = {{"upkvec", upkvec},
                                               {"gid2obj", gid2obj},
                                               {"gid2cell", gid2cell},
                                               {"gid_connect", gid_connect},
                                               {"get_partition", get_partition},
                                               {"upkpyobj", upkpyobj},
                                               {"pyret", pyret},
                                               {"py_alltoall", py_alltoall},
                                               {"py_allgather", py_allgather},
                                               {"py_gather", py_gather},
                                               {"py_broadcast", py_broadcast},
                                               {"py_scatter", py_scatter},
                                               {0, 0}};
 
static void* cons(Object*) {
    // not clear at moment what is best way to handle nested context
    int i = -1;
    if (ifarg(1)) {
        i = int(chkarg(1, 0, 10000));
    }
    OcBBS* bbs = new OcBBS(i);
    bbs->ref();
    return bbs;
}
 
static void destruct(void* v) {
    OcBBS* bbs = (OcBBS*) v;
    bbs->unref();
}
 
void ParallelContext_reg() {
    class2oc("ParallelContext", cons, destruct, members, retobj_members, retstr_members);
}
 
// A BBS message is something to execute.
// This helper execute depending of the style of the message.
// style == 0:
//     This is a string that is a hoc statement, execute it with `hoc_obj_run`.
//     Return `nullptr`, set `size` argument to 0.
//     Not listen to `exec`.
//     No arguments.
// style == 1:
//     This is a function name as a string followed by a list of arguments.
//     Return `nullptr` and set `size` to 0.
//     Dry run if `exec` is false.
// style == 2:
//     From a template, find a object with an id and inside this object find a member
//     with a name as a string. Execute this member with arguments.
//     Return `nullptr` and set `size` to 0.
//     Dry run if `exec` is false.
// style == 3:
//     A python pickle (https://docs.python.org/3/library/pickle.html) followed by arguments.
//     Return a string that is of size `size`.
//     Dry run if `exec` is false.
std::vector<char> BBSImpl::execute_helper(int id, bool exec) {
    int subworld = (nrnmpi_numprocs > 1 && nrnmpi_numprocs_bbs < nrnmpi_numprocs_world);
    int style = upkint();
    if (subworld) {
        assert(nrnmpi_myid == 0);
        int info[2];
        info[0] = id;
        info[1] = style;
        nrnmpi_int_broadcast(info, 2, 0);
    }
    std::vector<char> rs{};
    switch (style) {
    case 0: {
        char* statement = upkstr();
        if (subworld) {
            int size = strlen(statement) + 1;
            nrnmpi_int_broadcast(&size, 1, 0);
            nrnmpi_char_broadcast(statement, size, 0);
        }
        hoc_obj_run(statement, nullptr);
        delete[] statement;
    } break;
    default: {
        std::vector<char> python_pickle{};  // Only for style == 3
        Symbol* fname = nullptr;
        Object* ob = nullptr;
        std::list<char*> sarg;  // Store the strings pointer to delete[] them later
                                // Use a list because, we push pointers of the object into
                                // the hoc stack
        int narg = 0;           // total number of args
        if (style == 2) {       // object first
            char* template_name = upkstr();
            int object_index = upkint();  // object index
            Symbol* sym = hoc_lookup(template_name);
            if (sym) {
                sym = hoc_which_template(sym);
            }
            if (!sym) {
                hoc_execerror(template_name, "is not a template");
            }
            hoc_Item *q, *ql;
            ql = sym->u.ctemplate->olist;
            ITERATE(q, ql) {
                ob = OBJ(q);
                if (ob->index == object_index) {
                    break;
                }
                ob = nullptr;
            }
            if (!ob) {
                fprintf(stderr,
                        "%s[%d] is not an Object in this process\n",
                        template_name,
                        object_index);
                hoc_execerror("ParallelContext execution error", nullptr);
            }
            delete[] template_name;
            char* fname_str = upkstr();
            fname = hoc_table_lookup(fname_str, sym->u.ctemplate->symtable);
            if (!fname) {
                fprintf(stderr, "%s not a function in %s\n", fname_str, hoc_object_name(ob));
                hoc_execerror("ParallelContext execution error", nullptr);
            }
            delete[] fname_str;
            if (subworld) {
                hoc_execerror("with subworlds, this submit style not implemented", nullptr);
            }
        } else if (style == 3) {  // Python callable
            python_pickle = upkpickle();
            if (subworld) {
                int size = static_cast<int>(python_pickle.size());
                nrnmpi_int_broadcast(&size, 1, 0);
                nrnmpi_char_broadcast(python_pickle.data(), size, 0);
            }
        } else {
            char* fname_str = upkstr();
            if (subworld) {
                int size = strlen(fname_str) + 1;
                nrnmpi_int_broadcast(&size, 1, 0);
                nrnmpi_char_broadcast(fname_str, size, 0);
            }
            fname = hoc_lookup(fname_str);
            if (!fname) {
                fprintf(stderr, "%s not a function in %s\n", fname_str, hoc_object_name(ob));
                hoc_execerror("ParallelContext execution error", nullptr);
            }
            delete[] fname_str;
        }
 
        int argtypes = upkint();  // first is least signif
        if (subworld) {
            // printf("%d exec argtypes = %d\n", nrnmpi_myid_world, argtypes);
            nrnmpi_int_broadcast(&argtypes, 1, 0);
        }
        for (int i = 0, j = argtypes; (i = j % 5) != 0; j /= 5) {
            ++narg;
            if (i == 1) {
                double x = upkdouble();
                if (subworld) {
                    nrnmpi_dbl_broadcast(&x, 1, 0);
                }
                hoc_pushx(x);
            } else if (i == 2) {
                sarg.push_back(upkstr());
                if (subworld) {
                    int size = strlen(sarg.back()) + 1;
                    nrnmpi_int_broadcast(&size, 1, 0);
                    nrnmpi_char_broadcast(sarg.back(), size, 0);
                }
                hoc_pushstr(&(sarg.back()));
            } else if (i == 3) {
                int n = upkint();
                if (subworld) {
                    nrnmpi_int_broadcast(&n, 1, 0);
                }
                Vect* vec = new Vect(n);
                upkvec(n, vec->data());
                if (subworld) {
                    nrnmpi_dbl_broadcast(vec->data(), n, 0);
                }
                hoc_pushobj(vec->temp_objvar());
            } else {  // PythonObject
                auto s = upkpickle();
                int size = static_cast<int>(s.size());
                if (subworld) {
                    nrnmpi_int_broadcast(&size, 1, 0);
                    nrnmpi_char_broadcast(s.data(), size, 0);
                }
                assert(neuron::python::methods.pickle2po);
                Object* po = neuron::python::methods.pickle2po(s);
                --po->refcount;
                hoc_pushobj(hoc_temp_objptr(po));
            }
        }
        if (style == 3) {
            assert(neuron::python::methods.call_picklef);
            if (exec) {
                rs = neuron::python::methods.call_picklef(python_pickle, narg);
            }
            hoc_ac_ = 0.;
        } else {
            if (exec) {
                hoc_ac_ = hoc_call_objfunc(fname, narg, ob);
            } else {
                hoc_ac_ = 0.;
            }
        }
        for (auto& arg: sarg) {
            delete[] arg;
        }
    } break;
    }
    return rs;
}
 
void BBSImpl::subworld_worker_execute() {
    // execute the same thing that execute_worker is executing. This
    // is done for all the nrnmpi_myid_bbs == -1 workers associated with
    // the specific nrnmpi_myid == 0 with nrnmpi_myid_bbs >= 0.
    // All the nrnmpi/mpispike.cpp functions can be used since the
    // proper communicators for a subworld are used by those functions.
    // The broadcast functions are particularly useful and those are
    // how execute_worker passes messages into here.
 
    // printf("%d enter subworld_worker_execute\n", nrnmpi_myid_world);
    int info[2];
    // wait for something to do
    nrnmpi_int_broadcast(info, 2, 0);
    // info[0] = -1 means it was from a pc.context. Also -2 means
    // DONE.
    // printf("%d subworld_worker_execute info %d %d\n", nrnmpi_myid_world, info[0], info[1]);
    int id = info[0];
    if (id == -2) {  // DONE, so quit.
        done();
    }
    hoc_ac_ = double(id);
    int style = info[1];
    if (style == 0) {  // hoc statement form
        int size;
        nrnmpi_int_broadcast(&size, 1, 0);  // includes terminator
        char* s = new char[size];
        nrnmpi_char_broadcast(s, size, 0);
        hoc_obj_run(s, nullptr);
        delete[] s;
        // printf("%d leave subworld_worker_execute\n", nrnmpi_myid_world);
        return;
    }
    int i, j;
    int npickle;
    std::vector<char> s{};
    Symbol* fname = 0;
    Object* ob = nullptr;
    char* sarg[20];  // up to 20 arguments may be strings
    int ns = 0;      // number of args that are strings
    int narg = 0;    // total number of args
 
    if (style == 3) {  // python callable
        nrnmpi_int_broadcast(&npickle, 1, 0);
        s.resize(npickle);
        nrnmpi_char_broadcast(s.data(), npickle, 0);
    } else if (style == 1) {  // hoc function
        int size;
        nrnmpi_int_broadcast(&size, 1, 0);  // includes terminator
        // printf("%d subworld hoc function string size = %d\n", nrnmpi_myid_world, size);
        s.resize(size);
        nrnmpi_char_broadcast(s.data(), size, 0);
        fname = hoc_lookup(s.data());
        if (!fname) {
            return;
        }  // error raised by sender
    } else {
        return;  // no others implemented, error raised by sender
    }
 
    // now get the args
    int argtypes;
    nrnmpi_int_broadcast(&argtypes, 1, 0);
    // printf("%d subworld argtypes = %d\n", nrnmpi_myid_world, argtypes);
    for (j = argtypes; (i = j % 5) != 0; j /= 5) {
        ++narg;
        if (i == 1) {  // double
            double x;
            nrnmpi_dbl_broadcast(&x, 1, 0);
            // printf("%d subworld scalar = %g\n", nrnmpi_myid_world, x);
            hoc_pushx(x);
        } else if (i == 2) {  // string
            int size;
            nrnmpi_int_broadcast(&size, 1, 0);
            sarg[ns] = new char[size];
            nrnmpi_char_broadcast(sarg[ns], size, 0);
            hoc_pushstr(sarg + ns);
            ns++;
        } else if (i == 3) {  // Vector
            int n;
            nrnmpi_int_broadcast(&n, 1, 0);
            Vect* vec = new Vect(n);
            nrnmpi_dbl_broadcast(vec->data(), n, 0);
            hoc_pushobj(vec->temp_objvar());
        } else {  // PythonObject
            int n;
            nrnmpi_int_broadcast(&n, 1, 0);
            std::vector<char> s(n);
            nrnmpi_char_broadcast(s.data(), n, 0);
            Object* po = neuron::python::methods.pickle2po(s);
            --po->refcount;
            hoc_pushobj(hoc_temp_objptr(po));
        }
    }
 
    if (style == 3) {
        auto rs = neuron::python::methods.call_picklef(s, narg);
        assert(!rs.empty());
    } else {
        // printf("%d subworld hoc call %s narg=%d\n", nrnmpi_myid_world, fname->name, narg);
        hoc_call_objfunc(fname, narg, ob);
        // printf("%d subworld return from hoc call %s\n", nrnmpi_myid_world, fname->name);
    }
    for (i = 0; i < ns; ++i) {
        delete[] sarg[i];
    }
}
 
void BBSImpl::return_args(int id) {
    // the message has been set up by the subclass
    // perhaps it would be better to do this directly
    // and avoid the meaningless create and delete.
    // but then they all would have to know this format
    char* s;
    // printf("BBSImpl::return_args(%d):\n", id);
    upkint();  // userid
    /* int wid = */ upkint();
    int style = upkint();
    // printf("message userid=%d style=%d\n", i, style);
    switch (style) {
    case 0:
        s = upkstr();  // the statement
                       // printf("statement |%s|\n", s);
        delete[] s;
        break;
    case 2:            // obj first
        s = upkstr();  // template name
        upkint();      // instance index
                       // printf("object %s[%d]\n", s, i);
        delete[] s;
        // fall through
    case 1:
        s = upkstr();  // fname
        upkint();      // arg manifest
                       // printf("fname=|%s| manifest=%o\n", s, i);
        delete[] s;
        break;
    case 3:
        auto pickle = upkpickle();  // pickled callable
        upkint();                   // arg manifest
        break;
    }
    // now only args are left and ready to unpack.
}