fix thread issue

CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.25)
-project(blt-gp VERSION 0.1.42)
+project(blt-gp VERSION 0.1.43)
 
 include(CTest)
 

dhat.out.293761

@@ -1240,9 +1240,9 @@
  ,"0x122836: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x12E6ED: void std::vector<blt::gp::operator_id, std::allocator<blt::gp::operator_id> >::_M_realloc_insert<blt::gp::operator_id const&>(__gnu_cxx::__normal_iterator<blt::gp::operator_id*, std::vector<blt::gp::operator_id, std::allocator<blt::gp::operator_id> > >, blt::gp::operator_id const&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x12284D: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
- ,"0x12EB76: void std::vector<blt::gp::operator_info, std::allocator<blt::gp::operator_info> >::_M_realloc_insert<blt::gp::operator_info const&>(__gnu_cxx::__normal_iterator<blt::gp::operator_info*, std::vector<blt::gp::operator_info, std::allocator<blt::gp::operator_info> > >, blt::gp::operator_info const&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
+ ,"0x12EB76: void std::vector<blt::gp::operator_info_t, std::allocator<blt::gp::operator_info_t> >::_M_realloc_insert<blt::gp::operator_info_t const&>(__gnu_cxx::__normal_iterator<blt::gp::operator_info_t*, std::vector<blt::gp::operator_info_t, std::allocator<blt::gp::operator_info_t> > >, blt::gp::operator_info_t const&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x12294C: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
- ,"0x12E897: void std::vector<blt::gp::operator_info, std::allocator<blt::gp::operator_info> >::_M_realloc_insert<blt::gp::operator_info const&>(__gnu_cxx::__normal_iterator<blt::gp::operator_info*, std::vector<blt::gp::operator_info, std::allocator<blt::gp::operator_info> > >, blt::gp::operator_info const&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
+ ,"0x12E897: void std::vector<blt::gp::operator_info_t, std::allocator<blt::gp::operator_info_t> >::_M_realloc_insert<blt::gp::operator_info_t const&>(__gnu_cxx::__normal_iterator<blt::gp::operator_info_t*, std::vector<blt::gp::operator_info_t, std::allocator<blt::gp::operator_info_t> > >, blt::gp::operator_info_t const&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x12F1FA: void std::vector<std::function<void (std::ostream&, blt::gp::stack_allocator&)>, std::allocator<std::function<void (std::ostream&, blt::gp::stack_allocator&)> > >::_M_realloc_insert<std::function<void (std::ostream&, blt::gp::stack_allocator&)> >(__gnu_cxx::__normal_iterator<std::function<void (std::ostream&, blt::gp::stack_allocator&)>*, std::vector<std::function<void (std::ostream&, blt::gp::stack_allocator&)>, std::allocator<std::function<void (std::ostream&, blt::gp::stack_allocator&)> > > >, std::function<void (std::ostream&, blt::gp::stack_allocator&)>&&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x12298F: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x12F3EB: void std::vector<std::optional<std::basic_string_view<char, std::char_traits<char> > >, std::allocator<std::optional<std::basic_string_view<char, std::char_traits<char> > > > >::_M_realloc_insert<std::optional<std::basic_string_view<char, std::char_traits<char> > > >(__gnu_cxx::__normal_iterator<std::optional<std::basic_string_view<char, std::char_traits<char> > >*, std::vector<std::optional<std::basic_string_view<char, std::char_traits<char> > >, std::allocator<std::optional<std::basic_string_view<char, std::char_traits<char> > > > > >, std::optional<std::basic_string_view<char, std::char_traits<char> > >&&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
@@ -1261,7 +1261,7 @@
  ,"0x122F78: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x123052: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x12305D: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
- ,"0x12ED14: std::vector<blt::gp::operator_info, std::allocator<blt::gp::operator_info> >::push_back(blt::gp::operator_info const&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
+ ,"0x12ED14: std::vector<blt::gp::operator_info_t, std::allocator<blt::gp::operator_info_t> >::push_back(blt::gp::operator_info_t const&) (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x1231C6: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x123321: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"
  ,"0x123390: main (in /home/brett/Documents/code/c++/blt-gp/cmake-build-release/blt-SR-playground-example)"

examples/symbolic_regression.cpp

@@ -22,6 +22,7 @@
 #include <blt/std/format.h>
 #include <iostream>
 #include "operations_common.h"
+#include "blt/math/averages.h"
 
 //static constexpr long SEED = 41912;
 static const unsigned long SEED = std::random_device()();
@@ -36,13 +37,13 @@ std::array<context, 200> training_cases;
 blt::gp::prog_config_t config = blt::gp::prog_config_t()
         .set_initial_min_tree_size(2)
         .set_initial_max_tree_size(6)
-        .set_elite_count(200)
+        .set_elite_count(0)
         .set_crossover_chance(0.9)
         .set_mutation_chance(0.1)
         .set_reproduction_chance(0)
         .set_max_generations(50)
-        .set_pop_size(20000)
+        .set_pop_size(1)
-        .set_thread_count(0);
+        .set_thread_count(1);
 
 blt::gp::gp_program program{SEED, config};
 
@@ -143,30 +144,40 @@ int main()
     BLT_TRACE("Total Allocations: %ld times with a total of %s, peak allocated bytes %s", blt::gp::tracker.getAllocations(),
               blt::byte_convert_t(blt::gp::tracker.getAllocatedBytes()).convert_to_nearest_type().to_pretty_string().c_str(),
               blt::byte_convert_t(blt::gp::tracker.getPeakAllocatedBytes()).convert_to_nearest_type().to_pretty_string().c_str());
+    BLT_TRACE("------------------------------------------------------");
+    auto evaluation_calls_v = blt::gp::evaluation_calls.get_calls();
+    auto evaluation_allocations_v = blt::gp::evaluation_allocations.get_calls();
+    BLT_TRACE("Total Evaluation Calls: %ld; Peak Bytes Allocated %s", evaluation_calls_v,
+              blt::string::bytes_to_pretty(blt::gp::evaluation_calls.get_value()).c_str());
+    BLT_TRACE("Total Evaluation Allocations: %ld; Bytes %s; Average %s", evaluation_allocations_v,
+              blt::string::bytes_to_pretty(blt::gp::evaluation_allocations.get_value()).c_str(),
+              blt::string::bytes_to_pretty(blt::average(blt::gp::evaluation_allocations.get_value(), evaluation_allocations_v)).c_str());
+    BLT_TRACE("Percent Evaluation calls allocate? %lf%%", blt::average(evaluation_allocations_v, evaluation_calls_v) * 100);
+    BLT_TRACE("------------------------------------------------------");
     auto crossover_calls_v = blt::gp::crossover_calls.get_calls();
     auto crossover_allocations_v = blt::gp::crossover_allocations.get_calls();
     auto mutation_calls_v = blt::gp::mutation_calls.get_calls();
     auto mutation_allocations_v = blt::gp::mutation_allocations.get_calls();
     auto reproduction_calls_v = blt::gp::reproduction_calls.get_calls();
     auto reproduction_allocations_v = blt::gp::reproduction_allocations.get_calls();
-    BLT_TRACE("Total Crossover Calls: %ld Peak Bytes Allocated %s", crossover_calls_v,
+    BLT_TRACE("Total Crossover Calls: %ld; Peak Bytes Allocated %s", crossover_calls_v,
-              blt::byte_convert_t(blt::gp::crossover_calls.get_value()).convert_to_nearest_type().to_pretty_string().c_str());
+              blt::string::bytes_to_pretty(blt::gp::crossover_calls.get_value()).c_str());
-    BLT_TRACE("Total Mutation Calls: %ld Peak Bytes Allocated %s", mutation_calls_v,
+    BLT_TRACE("Total Mutation Calls: %ld; Peak Bytes Allocated %s", mutation_calls_v,
-              blt::byte_convert_t(blt::gp::mutation_calls.get_value()).convert_to_nearest_type().to_pretty_string().c_str());
+              blt::string::bytes_to_pretty(blt::gp::mutation_calls.get_value()).c_str());
-    BLT_TRACE("Total Reproduction Calls: %ld Peak Bytes Allocated %s", reproduction_calls_v,
+    BLT_TRACE("Total Reproduction Calls: %ld; Peak Bytes Allocated %s", reproduction_calls_v,
-              blt::byte_convert_t(blt::gp::reproduction_calls.get_value()).convert_to_nearest_type().to_pretty_string().c_str());
+              blt::string::bytes_to_pretty(blt::gp::reproduction_calls.get_value()).c_str());
-    BLT_TRACE("Total Crossover Allocations: %ld Bytes %s", crossover_allocations_v,
+    BLT_TRACE("Total Crossover Allocations: %ld; Bytes %s; Average %s", crossover_allocations_v,
-              blt::byte_convert_t(blt::gp::crossover_allocations.get_value()).convert_to_nearest_type().to_pretty_string().c_str());
+              blt::string::bytes_to_pretty(blt::gp::crossover_allocations.get_value()).c_str(),
-    BLT_TRACE("Total Mutation Allocations: %ld Bytes %s", mutation_allocations_v,
+              blt::string::bytes_to_pretty(blt::average(blt::gp::crossover_allocations.get_value(), crossover_allocations_v)).c_str());
-              blt::byte_convert_t(blt::gp::mutation_allocations.get_value()).convert_to_nearest_type().to_pretty_string().c_str());
+    BLT_TRACE("Total Mutation Allocations: %ld; Bytes %s; Average %s", mutation_allocations_v,
-    BLT_TRACE("Total Reproduction Allocations: %ld Bytes %s", reproduction_allocations_v,
+              blt::string::bytes_to_pretty(blt::gp::mutation_allocations.get_value()).c_str(),
-              blt::byte_convert_t(blt::gp::reproduction_allocations.get_value()).convert_to_nearest_type().to_pretty_string().c_str());
+              blt::string::bytes_to_pretty(blt::average(blt::gp::mutation_allocations.get_value(), mutation_allocations_v)).c_str());
-    BLT_TRACE("Percent Crossover calls allocate? %lf%%",
+    BLT_TRACE("Total Reproduction Allocations: %ld; Bytes %s; Average %s", reproduction_allocations_v,
-              static_cast<double>(crossover_allocations_v) / static_cast<double>(crossover_calls_v == 0 ? 1 : crossover_calls_v) * 100);
+              blt::string::bytes_to_pretty(blt::gp::reproduction_allocations.get_value()).c_str(),
-    BLT_TRACE("Percent Mutation calls allocate? %lf%%",
+              blt::string::bytes_to_pretty(blt::average(blt::gp::reproduction_allocations.get_value(), reproduction_allocations_v)).c_str());
-              static_cast<double>(mutation_allocations_v) / static_cast<double>(mutation_calls_v == 0 ? 1 : mutation_calls_v) * 100);
+    BLT_TRACE("Percent Crossover calls allocate? %lf%%", blt::average(crossover_allocations_v, crossover_calls_v) * 100);
-    BLT_TRACE("Percent Reproduction calls allocate? %lf%%",
+    BLT_TRACE("Percent Mutation calls allocate? %lf%%", blt::average(mutation_allocations_v, mutation_calls_v) * 100);
-              static_cast<double>(reproduction_allocations_v) / static_cast<double>(reproduction_calls_v == 0 ? 1 : reproduction_calls_v) * 100);
+    BLT_TRACE("Percent Reproduction calls allocate? %lf%%", blt::average(reproduction_allocations_v, reproduction_calls_v) * 100);
 #endif
     
     return 0;

include/blt/gp/fwdecl.h

@@ -30,12 +30,18 @@ namespace blt::gp
 {
 #ifdef BLT_TRACK_ALLOCATIONS
     inline allocation_tracker_t tracker;
+    
+    // population gen specifics
     inline call_tracker_t crossover_calls;
     inline call_tracker_t mutation_calls;
     inline call_tracker_t reproduction_calls;
     inline call_tracker_t crossover_allocations;
     inline call_tracker_t mutation_allocations;
     inline call_tracker_t reproduction_allocations;
+    
+    // for evaluating fitness
+    inline call_tracker_t evaluation_calls;
+    inline call_tracker_t evaluation_allocations;
 #endif
     
     class gp_program;

include/blt/gp/program.h

@@ -67,7 +67,7 @@ namespace blt::gp
         }
     };
     
-    struct operator_info
+    struct operator_info_t
     {
         // types of the arguments
         tracked_vector<type_id> argument_types;
@@ -79,6 +79,12 @@ namespace blt::gp
         detail::operator_func_t func;
     };
     
+    struct operator_metadata_t
+    {
+        blt::size_t arg_size_bytes = 0;
+        blt::size_t return_size_bytes = 0;
+    };
+    
     struct program_operator_storage_t
     {
         // indexed from return TYPE ID, returns index of operator
@@ -87,7 +93,8 @@ namespace blt::gp
         blt::expanding_buffer<tracked_vector<std::pair<operator_id, blt::size_t>>> operators_ordered_terminals;
         // indexed from OPERATOR ID (operator number)
         blt::hashset_t<operator_id> ephemeral_leaf_operators;
-        tracked_vector<operator_info> operators;
+        tracked_vector<operator_info_t> operators;
+        tracked_vector<operator_metadata_t> operator_metadata;
         tracked_vector<detail::print_func_t> print_funcs;
         tracked_vector<detail::destroy_func_t> destroy_funcs;
         tracked_vector<std::optional<std::string_view>> names;
@@ -110,11 +117,9 @@ namespace blt::gp
             template<typename... Operators>
             program_operator_storage_t& build(Operators& ... operators)
             {
-                tracked_vector<blt::size_t> sizes;
-                (sizes.push_back(add_operator(operators)), ...);
                 blt::size_t largest = 0;
-                for (auto v : sizes)
+                operator_metadata_t meta;
-                    largest = std::max(v, largest);
+                ((meta = add_operator(operators), largest = std::max(std::max(meta.arg_size_bytes, meta.return_size_bytes), largest)), ...);
                 
                 storage.eval_func = [&operators..., largest](const tree_t& tree, void* context) -> evaluation_context& {
                     const auto& ops = tree.get_operations();
@@ -214,14 +219,18 @@ namespace blt::gp
                 (storage.system.register_type<Args>(), ...);
                 storage.system.register_type<Return>();
                 
-                auto total_size_required = stack_allocator::aligned_size(sizeof(Return));
+                operator_metadata_t meta;
-                ((total_size_required += stack_allocator::aligned_size(sizeof(Args))), ...);
+                if constexpr (sizeof...(Args) != 0)
+                {
+                    meta.arg_size_bytes = (stack_allocator::aligned_size(sizeof(Args)) + ...);
+                }
+                meta.return_size_bytes = sizeof(Return);
                 
                 auto return_type_id = storage.system.get_type<Return>().id();
                 auto operator_id = blt::gp::operator_id(storage.operators.size());
                 op.id = operator_id;
                 
-                operator_info info;
+                operator_info_t info;
                 
                 if constexpr (sizeof...(Args) > 0)
                 {
@@ -240,6 +249,7 @@ namespace blt::gp
                 BLT_ASSERT(info.argc.argc_context - info.argc.argc <= 1 && "Cannot pass multiple context as arguments!");
                 
                 storage.operators.push_back(info);
+                storage.operator_metadata.push_back(meta);
                 storage.print_funcs.push_back([&op](std::ostream& out, stack_allocator& stack) {
                     if constexpr (blt::meta::is_streamable_v<Return>)
                     {
@@ -267,11 +277,11 @@ namespace blt::gp
                 storage.names.push_back(op.get_name());
                 if (op.is_ephemeral())
                     storage.ephemeral_leaf_operators.insert(operator_id);
-                return total_size_required * 2;
+                return meta;
             }
             
             template<typename T>
-            void add_non_context_argument(decltype(operator_info::argument_types)& types)
+            void add_non_context_argument(decltype(operator_info_t::argument_types)& types)
             {
                 if constexpr (!std::is_same_v<Context, detail::remove_cv_ref<T>>)
                 {
@@ -386,6 +396,12 @@ namespace blt::gp
 #ifdef BLT_TRACK_ALLOCATIONS
                 blt::gp::tracker.stop_measurement(fitness_alloc);
                 fitness_alloc.pretty_print("Fitness");
+                evaluation_calls.call();
+                evaluation_calls.set_value(std::max(evaluation_calls.get_value(), fitness_alloc.getAllocatedByteDifference()));
+                if (fitness_alloc.getAllocatedByteDifference() > 0)
+                {
+                    evaluation_allocations.call(fitness_alloc.getAllocatedByteDifference());
+                }
 #endif
                 
             }
@@ -475,9 +491,7 @@ namespace blt::gp
                                     mutation_selection.pre_process(*this, current_pop);
                                     reproduction_selection.pre_process(*this, current_pop);
                                     
-                                    perform_elitism(args, next_pop);
+                                    blt::size_t start = perform_elitism(args, next_pop);
-                                    
-                                    blt::size_t start = config.elites;
                                     
                                     while (start < config.population_size)
                                     {
@@ -566,10 +580,8 @@ namespace blt::gp
                                             mutation_selection.pre_process(*this, current_pop);
                                         if (&crossover_selection != &reproduction_selection)
                                             reproduction_selection.pre_process(*this, current_pop);
-                                        
+                                        auto elite_amount = perform_elitism(args, next_pop);
-                                        perform_elitism(args, next_pop);
+                                        thread_helper.next_gen_left -= elite_amount;
-                                        
-                                        thread_helper.next_gen_left -= config.elites;
                                     }
                                     thread_helper.barrier.wait();
                                     
@@ -653,7 +665,7 @@ namespace blt::gp
                 return storage.system;
             }
             
-            [[nodiscard]] inline operator_info& get_operator_info(operator_id id)
+            [[nodiscard]] inline operator_info_t& get_operator_info(operator_id id)
             {
                 return storage.operators[id];
             }
@@ -723,8 +735,10 @@ namespace blt::gp
                     return a.second > b.second;
                 });
                 
-                for (blt::size_t i = 0; i < size; i++)
+                for (blt::size_t i = 0; i < std::min(size, config.population_size); i++)
                     arr[i] = values[i].first;
+                for (blt::size_t i = std::min(size, config.population_size); i < size; i++)
+                    arr[i] = 0;
                 
                 return arr;
             }
@@ -791,7 +805,7 @@ namespace blt::gp
             
             struct concurrency_storage
             {
-                tracked_vector<std::unique_ptr<std::thread>> threads;
+                std::vector<std::unique_ptr<std::thread>> threads;
                 
                 std::mutex thread_function_control{};
                 std::condition_variable thread_function_condition{};

include/blt/gp/selection.h

@@ -41,7 +41,7 @@ namespace blt::gp
     constexpr inline auto perform_elitism = [](const selector_args& args, population_t& next_pop) {
         auto& [program, current_pop, current_stats, config, random] = args;
         
-        if (config.elites > 0)
+        if (config.elites > 0 && current_pop.get_individuals().size() >= config.elites)
         {
             static thread_local tracked_vector<std::pair<std::size_t, double>> values;
             values.clear();
@@ -70,7 +70,9 @@ namespace blt::gp
             
             for (blt::size_t i = 0; i < config.elites; i++)
                 next_pop.get_individuals()[i].copy_fast(current_pop.get_individuals()[values[i].first].tree);
+            return config.elites;
         }
+        return 0ul;
     };
     
     template<typename Crossover, typename Mutation, typename Reproduction>

include/blt/gp/stack.h

@@ -125,7 +125,7 @@ namespace blt::gp
             {
                 if (bytes == 0)
                     return;
-                if (size_ < bytes + bytes_stored)
+                if (bytes + bytes_stored >= size_)
                     expand(bytes + size_);
                 std::memcpy(data_ + bytes_stored, stack.data_ + (stack.bytes_stored - bytes), bytes);
                 bytes_stored += bytes;
@@ -135,7 +135,7 @@ namespace blt::gp
             {
                 if (bytes == 0 || data == nullptr)
                     return;
-                if (size_ < bytes + bytes_stored)
+                if (bytes + bytes_stored >= size_)
                     expand(bytes + size_);
                 std::memcpy(data_ + bytes_stored, data, bytes);
                 bytes_stored += bytes;

include/blt/gp/stats.h

@@ -130,7 +130,7 @@ namespace blt::gp
                 return tl.allocations.size();
             }
             
-            void await_completion(blt::u64 required_threads)
+            void await_thread_loading_complete(blt::u64 required_threads)
             {
                 std::unique_lock lock(tl.mutex);
                 tl.var.wait(lock, [this, required_threads]() {

lib/blt

@@ -1 +1 @@
-Subproject commit ab482f1a1c5782bd3501428f26c02f0bb4729946
+Subproject commit a7645d9ddec57ecaad525b48a30f8001adcf75e8

src/program.cpp

@@ -69,7 +69,7 @@ namespace blt::gp
             thread_helper.threads.emplace_back(new std::thread([i, this]() {
 #ifdef BLT_TRACK_ALLOCATIONS
                 tracker.reserve();
-                tracker.await_completion(config.threads);
+                tracker.await_thread_loading_complete(config.threads);
 #endif
                 std::function<void(blt::size_t)>* execution_function = nullptr;
                 while (!should_thread_terminate())
@@ -91,7 +91,7 @@ namespace blt::gp
             }));
         }
 #ifdef BLT_TRACK_ALLOCATIONS
-        tracker.await_completion(config.threads);
+        tracker.await_thread_loading_complete(config.threads);
 #endif
     }
 }

src/transformers.cpp

@@ -171,7 +171,7 @@ namespace blt::gp
         blt::size_t attempted_point = 0;
         
         const auto& crossover_point_type = program.get_operator_info(c1_ops[crossover_point].id);
-        operator_info* attempted_point_type = nullptr;
+        operator_info_t* attempted_point_type = nullptr;
         
         blt::size_t counter = 0;
         do