Merge pull request #478 from genn-team/procedural_kernelg

Proper implementation of kernel weights

include/genn/backends/single_threaded_cpu/backend.h

@@ -94,6 +94,7 @@ class BACKEND_EXPORT Backend : public BackendBase
                                  const Substitutions &kernelSubs, Handler handler) const override;
     virtual void genSparseSynapseVariableRowInit(CodeStream &os, const Substitutions &kernelSubs, Handler handler) const override;
     virtual void genDenseSynapseVariableRowInit(CodeStream &os, const Substitutions &kernelSubs, Handler handler) const override;
+    virtual void genKernelSynapseVariableInit(CodeStream &os, const SynapseKernelInitGroupMerged &sg, const Substitutions &kernelSubs, Handler handler) const final;
 
     virtual void genVariablePush(CodeStream &os, const std::string &type, const std::string &name, VarLocation loc, bool autoInitialized, size_t count) const override;
     virtual void genVariablePull(CodeStream &os, const std::string &type, const std::string &name, VarLocation loc, size_t count) const override;

include/genn/genn/code_generator/backendBase.h

@@ -45,6 +45,7 @@ namespace CodeGenerator
     class SynapseConnectivityInitGroupMerged;
     class SynapseDenseInitGroupMerged;
     class SynapseSparseInitGroupMerged;
+    class SynapseKernelInitGroupMerged;
 
 }
 
@@ -281,6 +282,7 @@ class GENN_EXPORT BackendBase
                                  const Substitutions &kernelSubs, Handler handler) const = 0;
     virtual void genSparseSynapseVariableRowInit(CodeStream &os, const Substitutions &kernelSubs, Handler handler) const = 0;
     virtual void genDenseSynapseVariableRowInit(CodeStream &os, const Substitutions &kernelSubs, Handler handler) const = 0;
+    virtual void genKernelSynapseVariableInit(CodeStream &os, const SynapseKernelInitGroupMerged &sg, const Substitutions &kernelSubs, Handler handler) const = 0;
 
     //! Generate code for pushing a variable to the 'device'
     virtual void genVariablePush(CodeStream &os, const std::string &type, const std::string &name, VarLocation loc, bool autoInitialized, size_t count) const = 0;

include/genn/genn/code_generator/backendSIMT.h

@@ -133,7 +133,8 @@ class GENN_EXPORT BackendSIMT : public BackendBase
     {
         genSynapseVariableRowInit(os, kernelSubs, handler);
     }
-
+
+    virtual void genKernelSynapseVariableInit(CodeStream &os, const SynapseKernelInitGroupMerged &sg, const Substitutions &kernelSubs, Handler handler) const final;
 
     //! Should 'scalar' variables be implemented on device or can host variables be used directly?
     virtual bool isDeviceScalarRequired() const final { return true; }

include/genn/genn/code_generator/codeGenUtils.h

@@ -147,31 +147,4 @@ void neuronSubstitutionsInSynapticCode(CodeGenerator::Substitutions &substitutio
     // Substitute extra global parameters from neuron model
     substitutions.addVarNameSubstitution(nm->getExtraGlobalParams(), sourceSuffix, "group->", destSuffix);
 }
-
-template<typename G>
-void genKernelIndex(std::ostream &os, const CodeGenerator::Substitutions &subs, const G &sg)
-{
-    // Loop through kernel dimensions to calculate array index
-    const auto &kernelSize = sg.getArchetype().getKernelSize();
-    for(size_t i = 0; i < kernelSize.size(); i++) {
-        os << "(" << subs["id_kernel_" + std::to_string(i)];
-        // Loop through remainining dimensions of kernel
-        for(size_t j = i + 1; j < kernelSize.size(); j++) {
-            // If kernel size if heterogeneous in this dimension, multiply by value from group structure
-            if(sg.isKernelSizeHeterogeneous(j)) {
-                os << " * group->kernelSize" << j;
-            }
-            // Otherwise, multiply by literal
-            else {
-                os << " * " << kernelSize.at(j);
-            }
-        }
-        os << ")";
-
-        // If this isn't the last dimension, add +
-        if(i != (kernelSize.size() - 1)) {
-            os << " + ";
-        }
-    }
-}
 }   // namespace CodeGenerator

include/genn/genn/code_generator/groupMerged.h

@@ -1044,6 +1044,12 @@ class GENN_EXPORT SynapseGroupMergedBase : public GroupMerged<SynapseGroupIntern
 
     //! Is kernel size heterogeneous in this dimension?
     bool isKernelSizeHeterogeneous(size_t dimensionIndex) const;
+
+    //! Get expression for kernel size in dimension (may be literal or group->kernelSizeXXX)
+    std::string getKernelSize(size_t dimensionIndex) const;
+
+    //! Generate an index into a kernel based on the id_kernel_XXX variables in subs
+    void genKernelIndex(std::ostream &os, const CodeGenerator::Substitutions &subs) const;
 
     std::string getPreSlot(unsigned int batchSize) const;
     std::string getPostSlot(unsigned int batchSize) const;
@@ -1090,6 +1096,7 @@ class GENN_EXPORT SynapseGroupMergedBase : public GroupMerged<SynapseGroupIntern
     }
 
     static std::string getSynVarIndex(unsigned int batchSize, VarAccessDuplication varDuplication, const std::string &index);
+    static std::string getKernelVarIndex(unsigned int batchSize, VarAccessDuplication varDuplication, const std::string &index);
 
 protected:
     //----------------------------------------------------------------------------
@@ -1102,6 +1109,7 @@ class GENN_EXPORT SynapseGroupMergedBase : public GroupMerged<SynapseGroupIntern
         SynapseDynamics,
         DenseInit,
         SparseInit,
+        KernelInit,
         ConnectivityInit,
     };
 

include/genn/genn/code_generator/initGroupMerged.h

@@ -158,6 +158,38 @@ class GENN_EXPORT SynapseSparseInitGroupMerged : public SynapseGroupMergedBase
     static const std::string name;
 };
 
+//----------------------------------------------------------------------------
+// CodeGenerator::SynapseKernelInitGroupMerged
+//----------------------------------------------------------------------------
+class GENN_EXPORT SynapseKernelInitGroupMerged : public SynapseGroupMergedBase
+{
+public:
+    SynapseKernelInitGroupMerged(size_t index, const std::string &precision, const std::string &timePrecision, const BackendBase &backend, 
+                                 const std::vector<std::reference_wrapper<const SynapseGroupInternal>> &groups)
+    :   SynapseGroupMergedBase(index, precision, timePrecision, backend, SynapseGroupMergedBase::Role::KernelInit, "", groups)
+    {}
+
+    boost::uuids::detail::sha1::digest_type getHashDigest() const
+    {
+        return SynapseGroupMergedBase::getHashDigest(SynapseGroupMergedBase::Role::KernelInit);
+    }
+
+    void generateRunner(const BackendBase &backend, CodeStream &definitionsInternal,
+                        CodeStream &definitionsInternalFunc, CodeStream &definitionsInternalVar,
+                        CodeStream &runnerVarDecl, CodeStream &runnerMergedStructAlloc) const
+    {
+        generateRunnerBase(backend, definitionsInternal, definitionsInternalFunc, definitionsInternalVar,
+                           runnerVarDecl, runnerMergedStructAlloc, name);
+    }
+
+    void generateInit(const BackendBase &backend, CodeStream &os, const ModelSpecMerged &modelMerged, Substitutions &popSubs) const;
+
+    //----------------------------------------------------------------------------
+    // Static constants
+    //----------------------------------------------------------------------------
+    static const std::string name;
+};
+
 
 // ----------------------------------------------------------------------------
 // CodeGenerator::SynapseConnectivityInitGroupMerged
@@ -384,4 +416,4 @@ class GENN_EXPORT CustomWUUpdateSparseInitGroupMerged : public CustomUpdateInitG
     static const std::string name;
 };
 
-}   // namespace CodeGenerator
+}   // namespace CodeGenerator

include/genn/genn/code_generator/modelSpecMerged.h

@@ -129,6 +129,9 @@ class GENN_EXPORT ModelSpecMerged
 
     //! Get merged synapse groups with sparse connectivity which require initialisation
     const std::vector<SynapseSparseInitGroupMerged> &getMergedSynapseSparseInitGroups() const{ return m_MergedSynapseSparseInitGroups; }
+
+    //! Get merged synapse groups with kernel connectivity which require initialisation
+    const std::vector<SynapseKernelInitGroupMerged> &getMergedSynapseKernelInitGroups() const{ return m_MergedSynapseKernelInitGroups; }
 
     //! Get merged custom update groups with sparse connectivity which require initialisation
     const std::vector<CustomWUUpdateSparseInitGroupMerged> &getMergedCustomWUUpdateSparseInitGroups() const { return m_MergedCustomWUUpdateSparseInitGroups; }
@@ -170,6 +173,7 @@ class GENN_EXPORT ModelSpecMerged
     void genMergedSynapseDenseInitGroupStructs(CodeStream &os, const BackendBase &backend) const { genMergedStructures(os, backend, m_MergedSynapseDenseInitGroups); }
     void genMergedSynapseConnectivityInitGroupStructs(CodeStream &os, const BackendBase &backend) const { genMergedStructures(os, backend, m_MergedSynapseConnectivityInitGroups); }
     void genMergedSynapseSparseInitGroupStructs(CodeStream &os, const BackendBase &backend) const { genMergedStructures(os, backend, m_MergedSynapseSparseInitGroups); }
+    void genMergedSynapseKernelInitGroupStructs(CodeStream &os, const BackendBase &backend) const { genMergedStructures(os, backend, m_MergedSynapseKernelInitGroups); }
     void genMergedCustomWUUpdateSparseInitGroupStructs(CodeStream &os, const BackendBase &backend) const { genMergedStructures(os, backend, m_MergedCustomWUUpdateSparseInitGroups); }
     void genMergedNeuronSpikeQueueUpdateStructs(CodeStream &os, const BackendBase &backend) const{ genMergedStructures(os, backend, m_MergedNeuronSpikeQueueUpdateGroups); }
     void genMergedNeuronPrevSpikeTimeUpdateStructs(CodeStream &os, const BackendBase &backend) const{ genMergedStructures(os, backend, m_MergedNeuronPrevSpikeTimeUpdateGroups); }
@@ -396,6 +400,9 @@ class GENN_EXPORT ModelSpecMerged
 
     //! Merged synapse groups with sparse connectivity which require initialisation
     std::vector<SynapseSparseInitGroupMerged> m_MergedSynapseSparseInitGroups;
+
+    //! Merged synapse groups with kernel connectivity which require initialisation
+    std::vector<SynapseKernelInitGroupMerged> m_MergedSynapseKernelInitGroups;
 
     //! Merged custom update groups with sparse connectivity which require initialisation
     std::vector<CustomWUUpdateSparseInitGroupMerged> m_MergedCustomWUUpdateSparseInitGroups;

include/genn/genn/synapseGroup.h

@@ -127,7 +127,8 @@ class GENN_EXPORT SynapseGroup
     unsigned int getMaxDendriticDelayTimesteps() const{ return m_MaxDendriticDelayTimesteps; }
     SynapseMatrixType getMatrixType() const{ return m_MatrixType; }
     const std::vector<unsigned int> &getKernelSize() const { return m_KernelSize; }
-
+    size_t getKernelSizeFlattened() const;
+
     //! Get variable mode used for variables used to combine input from this synapse group
     VarLocation getInSynLocation() const { return m_InSynLocation; }
 

include/genn/genn/synapseMatrixType.h

@@ -19,6 +19,7 @@ enum class SynapseMatrixWeight : unsigned int
     INDIVIDUAL      = (1 << 6),
     PROCEDURAL      = (1 << 7),
     INDIVIDUAL_PSM  = (1 << 8),
+    KERNEL          = (1 << 9)
 };
 
 //! Supported combinations of SynapticMatrixConnectivity and SynapticMatrixWeight
@@ -36,6 +37,7 @@ enum class SynapseMatrixType : unsigned int
     PROCEDURAL_GLOBALG                  = static_cast<unsigned int>(SynapseMatrixConnectivity::PROCEDURAL) | static_cast<unsigned int>(SynapseMatrixWeight::GLOBAL),
     PROCEDURAL_GLOBALG_INDIVIDUAL_PSM   = static_cast<unsigned int>(SynapseMatrixConnectivity::PROCEDURAL) | static_cast<unsigned int>(SynapseMatrixWeight::GLOBAL) | static_cast<unsigned int>(SynapseMatrixWeight::INDIVIDUAL_PSM),
     PROCEDURAL_PROCEDURALG              = static_cast<unsigned int>(SynapseMatrixConnectivity::PROCEDURAL) | static_cast<unsigned int>(SynapseMatrixWeight::PROCEDURAL) | static_cast<unsigned int>(SynapseMatrixWeight::INDIVIDUAL_PSM),
+    PROCEDURAL_KERNELG                  = static_cast<unsigned int>(SynapseMatrixConnectivity::PROCEDURAL) | static_cast<unsigned int>(SynapseMatrixWeight::KERNEL) | static_cast<unsigned int>(SynapseMatrixWeight::INDIVIDUAL_PSM),
 };
 
 //----------------------------------------------------------------------------

pygenn/genn_groups.py

@@ -19,6 +19,7 @@
 from .genn_wrapper import (SynapseMatrixConnectivity_SPARSE,
                            SynapseMatrixConnectivity_BITMASK,
                            SynapseMatrixConnectivity_DENSE,
+                           SynapseMatrixWeight_KERNEL,
                            SynapseMatrixWeight_INDIVIDUAL,
                            SynapseMatrixWeight_INDIVIDUAL_PSM,
                            VarLocation_HOST,
@@ -740,6 +741,8 @@ def weight_update_var_size(self):
             return self.trg.size * self.src.size
         elif self.is_ragged:
             return self.max_row_length * self.src.size
+        elif self.has_kernel_synapse_vars:
+            return int(np.product(self.pop.get_kernel_size()))
 
     @property
     def max_row_length(self):
@@ -902,6 +905,12 @@ def has_individual_synapse_vars(self):
         return (self.weight_sharing_master is None 
                 and (self.matrix_type & SynapseMatrixWeight_INDIVIDUAL) != 0)
 
+    @property
+    def has_kernel_synapse_vars(self):
+        """Tests whether synaptic connectivity has kernel weights"""
+        return (self.weight_sharing_master is None 
+                and (self.matrix_type & SynapseMatrixWeight_KERNEL) != 0)
+
     @property
     def has_individual_postsynaptic_vars(self):
         """Tests whether synaptic connectivity has
@@ -1191,7 +1200,7 @@ def load(self):
                 var_data = self.vars[v.name]
 
                 # If population has individual synapse variables
-                if self.has_individual_synapse_vars:
+                if self.has_individual_synapse_vars or self.has_kernel_synapse_vars:
                     # If variable is located on host
                     var_loc = self.pop.get_wuvar_location(v.name) 
                     if (var_loc & VarLocation_HOST) != 0:
@@ -1303,7 +1312,7 @@ def _init_wum_var(self, var_data, num_copies):
             # If connectivity is dense,
             # copy variables  directly into view
             # **NOTE** we assume order is row-major
-            if self.is_dense:
+            if self.is_dense or self.has_kernel_synapse_vars:
                 var_data.view[:] = var_data.values
             elif self.is_ragged:
                 # Sort variable to match GeNN order

src/genn/backends/cuda/backend.cc

@@ -859,6 +859,7 @@ void Backend::genInit(CodeStream &os, const ModelSpecMerged &modelMerged,
     modelMerged.genMergedCustomUpdateInitGroupStructs(os, *this);
     modelMerged.genMergedCustomWUUpdateDenseInitGroupStructs(os, *this);
     modelMerged.genMergedSynapseDenseInitGroupStructs(os, *this);
+    modelMerged.genMergedSynapseKernelInitGroupStructs(os, *this);
     modelMerged.genMergedSynapseConnectivityInitGroupStructs(os, *this);
     modelMerged.genMergedSynapseSparseInitGroupStructs(os, *this);
     modelMerged.genMergedCustomWUUpdateSparseInitGroupStructs(os, *this);
@@ -868,6 +869,7 @@ void Backend::genInit(CodeStream &os, const ModelSpecMerged &modelMerged,
     genMergedStructArrayPush(os, modelMerged.getMergedCustomUpdateInitGroups());
     genMergedStructArrayPush(os, modelMerged.getMergedCustomWUUpdateDenseInitGroups());
     genMergedStructArrayPush(os, modelMerged.getMergedSynapseDenseInitGroups());
+    genMergedStructArrayPush(os, modelMerged.getMergedSynapseKernelInitGroups());
     genMergedStructArrayPush(os, modelMerged.getMergedSynapseConnectivityInitGroups());
     genMergedStructArrayPush(os, modelMerged.getMergedSynapseSparseInitGroups());
     genMergedStructArrayPush(os, modelMerged.getMergedCustomWUUpdateSparseInitGroups());
@@ -885,6 +887,7 @@ void Backend::genInit(CodeStream &os, const ModelSpecMerged &modelMerged,
         modelMerged.getMergedCustomUpdateInitGroups(), [this](const CustomUpdateInternal &cg) { return padKernelSize(cg.getSize(), KernelInitialize); },
         modelMerged.getMergedCustomWUUpdateDenseInitGroups(), [this](const CustomUpdateWUInternal &cg){ return padKernelSize(cg.getSynapseGroup()->getTrgNeuronGroup()->getNumNeurons(), KernelInitialize); },
         modelMerged.getMergedSynapseDenseInitGroups(), [this](const SynapseGroupInternal &sg){ return padKernelSize(sg.getTrgNeuronGroup()->getNumNeurons(), KernelInitialize); },
+        modelMerged.getMergedSynapseKernelInitGroups(), [this](const SynapseGroupInternal &sg){ return padKernelSize(sg.getKernelSizeFlattened(), KernelInitialize); },
         modelMerged.getMergedSynapseConnectivityInitGroups(), [this](const SynapseGroupInternal &sg){ return padKernelSize(getNumConnectivityInitThreads(sg), KernelInitialize); });
 
     // Generate data structure for accessing merged groups from within sparse initialisation kernel