first cut of switching over compound initializations to flat arrays

src/script_opt/CPP/InitsInfo.cc

@@ -38,6 +38,13 @@ void CPP_InitsInfo::GenerateInitializers(CPPCompile* c) {
     c->Emit("%s %s = %s(%s, %s,", gt, InitializersName(), gt, base_name, Fmt(offset_set));
 
     c->IndentUp();
+    GenerateCohorts(c);
+    c->IndentDown();
+
+    c->Emit(");");
+}
+
+void CPP_InitsInfo::GenerateCohorts(CPPCompile* c) {
     c->Emit("{");
 
     int n = 0;
@@ -47,7 +54,7 @@ void CPP_InitsInfo::GenerateInitializers(CPPCompile* c) {
         if ( ++n > 1 )
             c->Emit("");
 
-        if ( cohort.size() == 1 && ! IsCompound() )
+        if ( cohort.size() == 1 && ! UsesCompoundVectors() )
             BuildCohort(c, cohort);
         else {
             c->Emit("{");
@@ -57,8 +64,6 @@ void CPP_InitsInfo::GenerateInitializers(CPPCompile* c) {
     }
 
     c->Emit("}");
-    c->IndentDown();
-    c->Emit(");");
 }
 
 void CPP_InitsInfo::BuildOffsetSet(CPPCompile* c) {
@@ -117,12 +122,36 @@ void CPP_InitsInfo::BuildCohortElement(CPPCompile* c, string init_type, vector<s
     c->Emit("std::make_shared<%s>(%s),", init_type, full_init);
 }
 
+void CPP_CompoundInitsInfo::GenerateInitializers(CPPCompile* c) {
+    c->Emit("static int %s_init[] = {", tag);
+    int n = 0;
+
+    c->IndentUp();
+
+    for ( auto& cohort : instances ) {
+        if ( ++n > 1 )
+            c->Emit("");
+
+        c->Emit(Fmt(int(cohort.size())) + ",");
+        BuildCohort(c, cohort);
+        c->Emit("-1,");
+    }
+
+    c->Emit("-2,");
+    c->IndentDown();
+    c->Emit("};");
+
+    CPP_InitsInfo::GenerateInitializers(c);
+}
+
+void CPP_CompoundInitsInfo::GenerateCohorts(CPPCompile* c) { c->Emit("%s_init", tag); }
+
 void CPP_CompoundInitsInfo::BuildCohortElement(CPPCompile* c, string init_type, vector<string>& ivs) {
     string init_line;
     for ( auto& iv : ivs )
-        init_line += iv + ", ";
+        init_line += iv;
 
-    c->Emit("{ %s},", init_line);
+    c->Emit("%s,", init_line);
 }
 
 void CPP_BasicConstInitsInfo::BuildCohortElement(CPPCompile* c, string init_type, vector<string>& ivs) {

src/script_opt/CPP/InitsInfo.h

@@ -133,10 +133,10 @@ public:
     // Sets the associated C++ type.
     virtual void SetCPPType(std::string ct) { CPP_type = std::move(ct); }
 
-    // Whether this initializer is in terms of compound objects. Used
+    // Whether this initializer is in terms of compound vectors. Used
     // for avoiding compiler warnings about singleton initializations in
     // braces.
-    virtual bool IsCompound() const { return false; }
+    virtual bool UsesCompoundVectors() const { return false; }
 
     // Returns the type associated with the table used for initialization
     // (i.e., this is the type of the global returned by InitializersName()).
@@ -146,9 +146,11 @@ public:
     void AddInstance(std::shared_ptr<CPP_InitInfo> g);
 
     // Emit code to populate the table used to initialize this collection.
-    void GenerateInitializers(CPPCompile* c);
+    virtual void GenerateInitializers(CPPCompile* c);
 
 protected:
+    virtual void GenerateCohorts(CPPCompile* c);
+
     // Computes offset_set - see below.
     void BuildOffsetSet(CPPCompile* c);
 
@@ -214,7 +216,7 @@ public:
         BuildInitType();
     }
 
-    bool IsCompound() const override { return true; }
+    bool UsesCompoundVectors() const override { return true; }
 
 private:
     void BuildInitType() { inits_type = std::string("CPP_CustomInits<") + CPPType() + ">"; }
@@ -236,7 +238,7 @@ public:
             inits_type = std::string("CPP_BasicConsts<") + CPP_type + ", " + c_type + ", " + tag + "Val>";
     }
 
-    bool IsCompound() const override { return false; }
+    bool UsesCompoundVectors() const override { return false; }
 
     void BuildCohortElement(CPPCompile* c, std::string init_type, std::vector<std::string>& ivs) override;
 };
@@ -254,7 +256,12 @@ public:
             inits_type = std::string("CPP_IndexedInits<") + CPPType() + ">";
     }
 
-    bool IsCompound() const override { return true; }
+    // This isn't true (anymore) because we separately build up the compound
+    // vectors needed for the initialization.
+    bool UsesCompoundVectors() const override { return false; }
+
+    void GenerateInitializers(CPPCompile* c) override;
+    void GenerateCohorts(CPPCompile* c) override;
 
     void BuildCohortElement(CPPCompile* c, std::string init_type, std::vector<std::string>& ivs) override;
 };

src/script_opt/CPP/RuntimeInits.cc

@@ -465,7 +465,7 @@ void CPP_GlobalInit::Generate(InitsManager* im, std::vector<void*>& /* inits_vec
         global->SetAttrs(im->Attributes(attrs));
 }
 
-void generate_indices_set(int* inits, std::vector<std::vector<int>>& indices_set) {
+size_t generate_indices_set(int* inits, std::vector<std::vector<int>>& indices_set) {
     // First figure out how many groups of indices there are, so we
     // can pre-allocate the outer vector.
     auto i_ptr = inits;
@@ -490,6 +490,26 @@ void generate_indices_set(int* inits, std::vector<std::vector<int>>& indices_set
 
         indices_set.emplace_back(std::move(indices));
     }
+
+    return i_ptr - inits + 1;
+}
+
+std::vector<std::vector<std::vector<int>>> generate_indices_set(int* inits) {
+    // Figure out how many vector-of-vectors there are.
+    int num_vv = 0;
+    for ( int i = 0; inits[i] >= -1; ++i )
+        if ( inits[i] == -1 )
+            ++num_vv;
+
+    std::vector<std::vector<std::vector<int>>> indices_set(num_vv);
+
+    auto i_ptr = inits;
+    for ( int i = 0; i < num_vv; ++i )
+        i_ptr += generate_indices_set(i_ptr, indices_set[i]);
+
+    ASSERT(*i_ptr == -2);
+
+    return indices_set;
 }
 
 } // namespace zeek::detail

src/script_opt/CPP/RuntimeInits.h

@@ -19,6 +19,23 @@ using FuncValPtr = IntrusivePtr<FuncVal>;
 
 class InitsManager;
 
+// Helper function that takes a (large) array of int's and from them
+// constructs the corresponding vector-of-vector-of-indices.  Each
+// vector-of-indices is represented first by an int specifying its
+// size, and then that many int's for its values.  We recognize the
+// end of the array upon encountering a "size" entry of -1.
+//
+// Returns how many elements were processed out of "inits", including its
+// terminator.
+extern size_t generate_indices_set(int* inits, std::vector<std::vector<int>>& indices_set);
+
+// The same but for one more level of vector construction. The source array
+// has sub-arrays terminated with -1 per the above, and the whole shebang
+// is terminated with -2.
+//
+// Returns the vector construction.
+extern std::vector<std::vector<std::vector<int>>> generate_indices_set(int* inits);
+
 // An abstract helper class used to access elements of an initialization vector.
 // We need the abstraction because InitsManager below needs to be able to refer
 // to any of a range of templated classes.
@@ -223,16 +240,16 @@ using ValElemVecVec = std::vector<ValElemVec>;
 template<class T>
 class CPP_IndexedInits : public CPP_AbstractInits<T, ValElemVecVec> {
 public:
-    CPP_IndexedInits(std::vector<T>& _inits_vec, int _offsets_set, std::vector<ValElemVecVec> _inits)
+    CPP_IndexedInits(std::vector<T>& _inits_vec, int _offsets_set, int* raw_inits)
-        : CPP_AbstractInits<T, ValElemVecVec>(_inits_vec, _offsets_set, std::move(_inits)) {}
+        : CPP_AbstractInits<T, ValElemVecVec>(_inits_vec, _offsets_set, generate_indices_set(raw_inits)) {}
 
 protected:
     void InitializeCohortWithOffsets(InitsManager* im, int cohort, const std::vector<int>& cohort_offsets) override;
 
-    // Note, in the following we pass in the inits_vec, even though
+    // Note, in the following we pass in the inits_vec ("ivec", even though
-    // the method will have direct access to it, because we want to
+    // the method will have direct access to it, because we want to use
-    // use overloading to dispatch to custom generation for different
+    // overloading to dispatch to custom generation for different types of
-    // types of values.
+    // values.
     void Generate(InitsManager* im, std::vector<EnumValPtr>& ivec, int offset, ValElemVec& init_vals);
     void Generate(InitsManager* im, std::vector<StringValPtr>& ivec, int offset, ValElemVec& init_vals);
     void Generate(InitsManager* im, std::vector<PatternValPtr>& ivec, int offset, ValElemVec& init_vals);
@@ -256,8 +273,8 @@ protected:
 // on subclasses of TypePtr.
 class CPP_TypeInits : public CPP_IndexedInits<TypePtr> {
 public:
-    CPP_TypeInits(std::vector<TypePtr>& _inits_vec, int _offsets_set, std::vector<ValElemVecVec> _inits)
+    CPP_TypeInits(std::vector<TypePtr>& _inits_vec, int _offsets_set, int* raw_inits)
-        : CPP_IndexedInits<TypePtr>(_inits_vec, _offsets_set, _inits) {}
+        : CPP_IndexedInits<TypePtr>(_inits_vec, _offsets_set, raw_inits) {}
 
 protected:
     void DoPreInits(InitsManager* im, const std::vector<int>& offsets_vec) override;
@@ -506,11 +523,4 @@ struct CPP_RegisterBody {
     std::vector<std::string> events;
 };
 
-// Helper function that takes a (large) array of int's and from them
-// constructs the corresponding vector-of-vector-of-indices.  Each
-// vector-of-indices is represented first by an int specifying its
-// size, and then that many int's for its values.  We recognize the
-// end of the array upon encountering a "size" entry of -1.
-extern void generate_indices_set(int* inits, std::vector<std::vector<int>>& indices_set);
-
 } // namespace zeek::detail