Dev

benchmarks/engines/busyring/arbor/ring.cpp

@@ -32,7 +32,7 @@ using arb::cell_size_type;
 using arb::cell_member_type;
 using arb::cell_kind;
 using arb::time_type;
-using arb::cell_probe_address;
+using arb::cable_probe_membrane_voltage;
 
 // Writes voltage trace as a json file.
 void write_trace_json(std::string fname, const arb::trace_data<double>& trace);
@@ -127,12 +127,9 @@ class ring_recipe: public arb::recipe {
     }
 
     arb::probe_info get_probe(cell_member_type id) const override {
-        // Get the appropriate kind for measuring voltage.
-        cell_probe_address::probe_kind kind = cell_probe_address::membrane_voltage;
         // Measure at the soma.
-        arb::segment_location loc(0, 0.0);
-
-        return arb::probe_info{id, kind, cell_probe_address{loc, kind}};
+        arb::mlocation loc{0, 0.0};
+        return arb::probe_info{id, 0, cable_probe_membrane_voltage{loc}};
     }
 
 private:
@@ -146,8 +143,7 @@ class ring_recipe: public arb::recipe {
 struct cell_stats {
     using size_type = unsigned;
     size_type ncells = 0;
-    size_type nsegs = 0;
-    size_type ncomp = 0;
+    size_type nbranch = 0;
 
     cell_stats(arb::recipe& r) {
 #ifdef ARB_MPI_ENABLED
@@ -158,30 +154,26 @@ struct cell_stats {
         size_type cells_per_rank = ncells/nranks;
         size_type b = rank*cells_per_rank;
         size_type e = (rank==nranks-1)? ncells: (rank+1)*cells_per_rank;
-        size_type nsegs_tmp = 0;
-        size_type ncomp_tmp = 0;
+        size_type nbranch_tmp = 0;
         for (size_type i=b; i<e; ++i) {
             auto c = arb::util::any_cast<arb::cable_cell>(r.get_cell_description(i));
-            nsegs_tmp += c.num_segments();
-            ncomp_tmp += c.num_compartments();
+            nbranch_tmp += c.morphology().num_branches();
         }
-        MPI_Allreduce(&nsegs_tmp, &nsegs, 1, MPI_UNSIGNED, MPI_SUM, MPI_COMM_WORLD);
-        MPI_Allreduce(&ncomp_tmp, &ncomp, 1, MPI_UNSIGNED, MPI_SUM, MPI_COMM_WORLD);
+        MPI_Allreduce(&nbranch_tmp, &nbranch, 1, MPI_UNSIGNED, MPI_SUM, MPI_COMM_WORLD);
 #else
         ncells = r.num_cells();
         for (size_type i=0; i<ncells; ++i) {
             auto c = arb::util::any_cast<arb::cable_cell>(r.get_cell_description(i));
-            nsegs += c.num_segments();
-            ncomp += c.num_compartments();
+            nbranch += c.morphology().num_branches();
         }
 #endif
     }
 
     friend std::ostream& operator<<(std::ostream& o, const cell_stats& s) {
         return o << "cell stats: "
                  << s.ncells << " cells; "
-                 << s.nsegs << " segments; "
-                 << s.ncomp << " compartments.";
+                 << s.nbranch << " branches; "
+                 << 0 << " compartments; ";
     }
 };
 
@@ -336,13 +328,11 @@ double interp(const std::array<T,2>& r, unsigned i, unsigned n) {
 }
 
 arb::cable_cell branch_cell(arb::cell_gid_type gid, const cell_parameters& params) {
-    arb::cable_cell cell;
-
-    cell.default_parameters.axial_resistivity = 100;
+    arb::sample_tree tree;
 
     // Add soma.
-    auto soma = cell.add_soma(12.6157/2.0); // For area of 500 μm².
-    soma->add_mechanism("hh");
+    double soma_radius = 12.6157/2.0;
+    tree.append(arb::mnpos, {{0,0,0,soma_radius}, 1}); // For area of 500 μm².
 
     std::vector<std::vector<unsigned>> levels;
     levels.push_back({0});
@@ -353,7 +343,7 @@ arb::cable_cell branch_cell(arb::cell_gid_type gid, const cell_parameters& param
 
     double dend_radius = 0.5; // Diameter of 1 μm for each cable.
 
-    unsigned nsec = 1;
+    double dist_from_soma = soma_radius;
     for (unsigned i=0; i<params.max_depth; ++i) {
         // Branch prob at this level.
         double bp = interp(params.branch_probs, i, params.max_depth);
@@ -366,37 +356,51 @@ arb::cable_cell branch_cell(arb::cell_gid_type gid, const cell_parameters& param
         for (unsigned sec: levels[i]) {
             for (unsigned j=0; j<2; ++j) {
                 if (dis(gen)<bp) {
-                    sec_ids.push_back(nsec++);
-                    auto dend = cell.add_cable(sec, arb::section_kind::dendrite, dend_radius, dend_radius, l);
-                    dend->set_compartments(nc);
-                    dend->add_mechanism("pas");
+                    auto z = dist_from_soma;
+                    auto p = tree.append(sec, {{0,0,z,dend_radius}, 3});
+                    if (nc>1) {
+                        auto dz = l/nc;
+                        for (unsigned k=1; k<nc; ++k) {
+                            p = tree.append(p, {{0,0,z+k*dz, dend_radius}, 3});
+                        }
+                    }
+                    sec_ids.push_back(tree.append(p, {{0,0,z+l,dend_radius}, 3}));
                 }
             }
         }
         if (sec_ids.empty()) {
             break;
         }
         levels.push_back(sec_ids);
+
+        dist_from_soma += l;
     }
 
+    arb::label_dict d;
+
+    using arb::reg::tagged;
+    d.set("soma",      tagged(1));
+    d.set("dendrites", join(tagged(3), tagged(4)));
+
+    arb::cable_cell cell(arb::morphology(tree, true), d);
+
+    cell.paint("soma", "hh");
+    cell.paint("dendrites", "pas");
+    cell.default_parameters.axial_resistivity = 100; // [Ω·cm]
+
     // Add spike threshold detector at the soma.
-    cell.add_detector({0,0}, 10);
+    cell.place(arb::mlocation{0,0}, arb::threshold_detector{10});
 
-    // Add a synapse to the soma for ring network.
-    // Attach at the soma to minimise delay between event delivery
-    // and subsequent spikes, which makes it easier to tune firing rate.
-    cell.add_synapse({0, 0.5}, "expsyn");
+    // Add a synapse to the mid point of the first dendrite.
+    cell.place(arb::mlocation{1, 0.5}, "expsyn");
 
-    // Add additional synapses to random locations on the cell.
-    std::uniform_real_distribution<double> pos_dis(0, 1);
-    std::uniform_int_distribution<cell_lid_type> seg_dis(1, nsec-1);
+    // Add additional synapses that will not be connected to anything.
     for (unsigned i=1u; i<params.synapses; ++i) {
-        const auto seg = seg_dis(gen);
-        const auto pos = pos_dis(gen);
-
-        cell.add_synapse({seg, pos}, "expsyn");
+        cell.place(arb::mlocation{1, 0.5}, "expsyn");
     }
 
+    // Make a CV between every sample in the sample tree.
+    cell.default_parameters.discretization = arb::cv_policy_every_sample();
+
     return cell;
 }
-

common/bin/comparex

@@ -38,6 +38,12 @@ variable X:
     X.abserr.lb         max(0, |delta|-interperr)
     X.relerr            abserr / max |reference|
     X.relerr.lb         abserr.lb / max |reference|
+
+    X.abserr.rms        sqrt(mean(delta^2))
+    X.abserr.rms.lb     sqrt(mean(maximum(delta-interperr, 0)^2))
+    X.relerr.rms        sqrt(mean(delta^2)) / max |reference|
+    X.relerr.rms.lb     sqrt(mean(maximum(delta-interperr, 0)^2)) / max |reference|
+
 """
 
     P.add_argument('input', metavar='FILE', help='input dataset')
@@ -166,18 +172,28 @@ for var in varlist:
             warn(opts.prog, 'no common points for variable \'{}\'', var)
         continue
 
-    abserr = np.asscalar(np.max(abs(delta)))
-    abserr_lb = max(0, np.asscalar(np.max(abs(delta)-interperr)))
+    delta_lb = np.maximum(abs(delta)-interperr, np.zeros_like(delta))
+
+    abserr = np.max(abs(delta)).item()
+    abserr_lb = np.max(delta_lb).item()
+    abserr_rms = np.sqrt(np.mean(np.square(delta))).item()
+    abserr_rms_lb = np.sqrt(np.mean(np.square(delta_lb))).item()
 
-    r_absmax = np.asscalar(np.max(abs(r.reindex_like(delta))))
+    r_absmax = np.max(abs(r.reindex_like(delta))).item()
     relerr = abserr/r_absmax if r_absmax>0 else 0
     relerr_lb = abserr_lb/r_absmax if r_absmax>0 else 0
+    relerr_rms = abserr_rms/r_absmax if r_absmax>0 else 0
+    relerr_rms_lb = abserr_rms_lb/r_absmax if r_absmax>0 else 0
 
     out[var+'.delta'] = delta
     out[var+'.interperr'] = interperr
     out[var+'.abserr'] = abserr
     out[var+'.abserr.lb'] = abserr_lb
+    out[var+'.abserr.rms'] = abserr_rms
+    out[var+'.abserr.rms.lb'] = abserr_rms_lb
     out[var+'.relerr'] = relerr
     out[var+'.relerr.lb'] = relerr_lb
+    out[var+'.relerr.rms'] = relerr_rms
+    out[var+'.relerr.rms.lb'] = relerr_rms_lb
 
 out.to_netcdf(opts.output)

common/bin/tsplotx

@@ -302,7 +302,7 @@ def gather_ts_plots(tss, groupby):
 
     for ts in tss:
         key = tuple([ts.meta.get(g) for g in groupby])
-        if key is () or None in key or key not in group_lookup:
+        if key == () or None in key or key not in group_lookup:
             pretty_key=', '.join([unicode(k)+u'='+unicode(v) for k,v in zip(groupby, key) if v is not None])
             pd = PlotData(pretty_key)
             pd.series = [ts]

docs/install.rst

@@ -64,17 +64,18 @@ The simulation engines to install are provided as arguments.
 Further options for installing the simulation engines in a user-specified path and customising
 the build environment can be provided:
 
-====================  =================     ======================================================
-Flag                  Default value         Explanation
-====================  =================     ======================================================
-simulator             none                  Which simulation engines to download and install.
-                                            Any number of the following: {``arbor``, ``neuron``, ``coreneuron``, ``all``}.
-``--prefix``          current path          Path for downloading, compiling, installing simulation engines.
-                                            Also used to store inputs and outputs from benchmarks and validation tests.
-                                            Can be either a relative or absolute path.
-``--env``             none                  Optional script for configuring the environment and build steps.
-                                            See :ref:`customenv`.
-====================  =================     ======================================================
+=========================  =================     ======================================================
+Flag                       Default value         Explanation
+=========================  =================     ======================================================
+simulator                  none                  Which simulation engines to download and install.
+                                                 Any number of the following: {``arbor``, ``neuron``, ``coreneuron``, ``all``}.
+``--prefix``               current path          Path for downloading, compiling, installing simulation engines.
+                                                 Also used to store inputs and outputs from benchmarks and validation tests.
+                                                 Can be either a relative or absolute path.
+``--env``                  none                  Optional script for configuring the environment and build steps.
+                                                 See :ref:`customenv`.
+``--disable-validation``   false                 Suppress building of validation.
+========================   =================     ======================================================
 
 .. _customenv:
 
@@ -184,7 +185,7 @@ Variable                  Default value                                 Explanat
 ``ns_arb_arch``           ``native``                                    `The CPU architecture target <https://arbor.readthedocs.io/en/latest/install.html#architecture>`_
                                                                         for Arbor. Must be set when cross compiling.
                                                                         Default ``native`` targets the architecture used to configure NSuite.
-``ns_arb_with_gpu``       ``OFF``                                       Whether to build Arbor with NVIDIA GPU support.
+``ns_arb_with_gpu``       ``none``                                       Whether to build Arbor with GPU support, valid choices are (none|cuda|cuda-clang|hip)
 ``ns_arb_vectorize``      ``ON``                                        Whether to use explicit vectorization for Arbor.
 ========================  ===========================================   ======================================================
 

install-local.sh

@@ -6,26 +6,26 @@ Setup NSuite framework, build and install simulators, benchmarks,
 and validation tests.
 
 Options:
-    --pyvenv=VENVOPT  Customize use of Python virtual environment (see below)
-    --env=SCRIPT      Source SCRIPT before building.
-    --prefix=PATH     Use PATH as base for install and other run-time
-                      working directories.
-
-TARGET is one of:
-    arbor             Build Arbor.
-    neuron            Build NEURON.
-    coreneuron        Build CoreNEURON.
-    all               Build all simulators.
+    --pyvenv=VENVOPT        Customize use of Python virtual environment (see below)
+    --env=SCRIPT            Source SCRIPT before building.
+    --prefix=PATH           Use PATH as base for install and other run-time
+                            working directories.
+    --disable-validation    Skip building of validation tests.
+TARGET is one of:           
+    arbor                   Build Arbor.
+    neuron                  Build NEURON.
+    coreneuron              Build CoreNEURON.
+    all                     Build all simulators.
 
 Building a TARGET will also build any associated tests and
 benchmarks as required.
 
 By default, a Python virtual environment is created in which required modules
 are installed if they are missing in the system environment.
 VENVOPT is one of:
-    disable           Do not use a Python virtualenv at all.
-    inherit           Use a virtualenv, using site packages.
-    enable            Use a virtualenv, ignoring site packages (default).
+    disable                 Do not use a Python virtualenv at all.
+    inherit                 Use a virtualenv, using site packages.
+    enable                  Use a virtualenv, ignoring site packages (default).
 
 _end_
 }
@@ -83,6 +83,9 @@ do
             shift
             ns_prefix=$1
             ;;
+	--disable-validation )
+	    ns_disable_validation=true
+	    ;;
         * )
             echo "unknown option '$1'"
             usage
@@ -201,11 +204,13 @@ cd "$ns_base_path"
 [ "$ns_build_coreneuron" = true ] && echo && source "$ns_base_path/scripts/build_coreneuron.sh"
 cd "$ns_base_path"
 
-# Always attempt to build validation models/generators.
-echo
-msghi "Building validation tests and generators"
-source "$ns_base_path/scripts/build_validation_models.sh"
-cd "$ns_base_path"
+# attempt to build validation models/generators.
+if [ "$ns_disable_validation" != "true" ]; then
+    echo
+    msghi "Building validation tests and generators"
+    source "$ns_base_path/scripts/build_validation_models.sh"
+    cd "$ns_base_path"
+fi
 
 echo
 msghi "Installation finished"

scripts/build_arbor.sh

@@ -48,7 +48,9 @@ fi
 cd "$arb_build_path"
 cmake_args=-DCMAKE_INSTALL_PREFIX:PATH="$ns_install_path"
 cmake_args="$cmake_args -DARB_WITH_MPI=$ns_with_mpi"
-cmake_args="$cmake_args -DARB_WITH_GPU=$ns_arb_with_gpu"
+if [ "$ns_arb_with_gpu" != "none" ]; then
+    cmake_args="$cmake_args -DARB_GPU=$ns_arb_with_gpu"
+fi
 cmake_args="$cmake_args -DARB_ARCH=$ns_arb_arch"
 cmake_args="$cmake_args -DARB_VECTORIZE=$ns_arb_vectorize"
 if [ "$ns_arb_xcompile_modcc" == "ON" ]; then

scripts/environment.sh

@@ -67,10 +67,10 @@ default_environment() {
     # Arbor specific
 
     ns_arb_git_repo=https://github.com/arbor-sim/arbor.git
-    ns_arb_branch=v0.2.2
+    ns_arb_branch=master
 
     ns_arb_arch=native
-    ns_arb_with_gpu=OFF
+    ns_arb_with_gpu=none
     ns_arb_vectorize=ON
     ns_arb_xcompile_modcc=OFF
 

systems/daint-gpu.sh

@@ -28,7 +28,7 @@ ns_cc=$(which cc)
 ns_cxx=$(which CC)
 ns_with_mpi=ON
 
-ns_arb_with_gpu=ON
+ns_arb_with_gpu=cuda
 ns_arb_arch=haswell
 
 export CRAYPE_LINK_TYPE=dynamic

systems/jureca-gpu.sh

@@ -30,7 +30,7 @@ ns_cc=$(which mpicc)
 ns_cxx=$(which mpicxx)
 ns_with_mpi=ON
 
-ns_arb_with_gpu=ON
+ns_arb_with_gpu=cuda
 ns_arb_arch=haswell
 ns_arb_branch=master
 

systems/juwels-gpu.sh

@@ -32,7 +32,7 @@ ns_cc=$(which mpicc)
 ns_cxx=$(which mpicxx)
 ns_with_mpi=ON
 
-ns_arb_with_gpu=ON
+ns_arb_with_gpu=cuda
 ns_arb_arch=skylake-avx512
 
 ns_arb_branch=master