Skip to content

Commit

Permalink
implementation of std::for_each (#9)
Browse files Browse the repository at this point in the history
  • Loading branch information
@djanloo
djanloo committed Jul 12, 2024
1 parent a55089c commit fc82b6f
Show file tree
Hide file tree
Showing 2 changed files with 37 additions and 93 deletions.
127 changes: 35 additions & 92 deletions quilt/core/network.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,11 +7,8 @@

#include <string>
#include <thread>

#define MAX_N_1_THREADS 150
#define MAX_N_2_THREADS 300
#define MAX_N_3_THREADS 600
#define MAX_N_4_THREADS 1000
#include <algorithm>
#include <execution>

#define N_THREADS_BUILD 8

Expand Down Expand Up @@ -246,118 +243,66 @@ void Population::project(const SparseProjection * projection, Population * effer
void Population::evolve()
{
// COMMENTED FOR ENHANCEMENT #9
// // Splits the work in equal parts using Nthreads threads
// unsigned int n_threads;
// Splits the work in equal parts
std::vector<unsigned int> bunch_starts(N_THREADS_POP_EVOLVE), bunch_ends(N_THREADS_POP_EVOLVE);
for (unsigned int i = 0; i < N_THREADS_POP_EVOLVE; i++){
bunch_starts[i] = i*static_cast<unsigned int>(this->n_neurons)/N_THREADS_POP_EVOLVE;
bunch_ends[i] = (i + 1)*static_cast<unsigned int>(this->n_neurons)/N_THREADS_POP_EVOLVE - 1;
}

// // This is reduntant if done here
// if (n_neurons < MAX_N_1_THREADS) n_threads = 0;
// else if (n_neurons < MAX_N_2_THREADS) n_threads = 2;
// else if (n_neurons < MAX_N_3_THREADS) n_threads = 3;
// else if (n_neurons < MAX_N_4_THREADS) n_threads = 4;
// else n_threads = std::thread::hardware_concurrency();
// Ensures that all neurons are covered
bunch_ends[N_THREADS_POP_EVOLVE-1] = this->n_neurons-1;

// // Lambda function for batch evolution
// auto evolve_bunch = [this](unsigned int from, unsigned int to){
// for (unsigned int i = from; i< to; i++){
// this->neurons[i]->evolve();
// }
// };

// // Lambda function for batch spike handling
// auto handle_spikes_bunch = [this](unsigned int from, unsigned int to){
// for (unsigned int i = from; i< to; i++){
// this->neurons[i]->handle_incoming_spikes();
// }
// };

perf_mgr.start_recording("evolution");
std::for_each(std::execution::par, neurons.begin(), neurons.end(), [](Neuron* neuron) {
neuron->evolve();
});
perf_mgr.end_recording("evolution");

perf_mgr.start_recording("spike_handling");
std::for_each(std::execution::par, neurons.begin(), neurons.end(), [](Neuron* neuron) {
neuron->handle_incoming_spikes();
});
perf_mgr.end_recording("spike_handling");
// ~ENHANCEMENT #9

// // POOL TEST
// // Uses the thread pool to evolve the neurons
// perf_mgr.start_recording("evolution");
// // In case few neurons are present, do not use multithreading
// // to avoid overhead
// if (n_threads == 0){
// evolve_bunch(0, n_neurons);
// for (const auto& neuron : neurons) {
// boost::asio::post(thread_pool, boost::bind(&Neuron::evolve, neuron));
// }
// else{// multithreading evolution

// std::vector<unsigned int> bunch_starts(n_threads), bunch_ends(n_threads);

// for (unsigned int i = 0; i < n_threads; i++){
// bunch_starts[i] = i*static_cast<unsigned int>(this->n_neurons)/n_threads;
// bunch_ends[i] = (i + 1)*static_cast<unsigned int>(this->n_neurons)/n_threads - 1;
// }

// // Ensures that all neurons are covered
// bunch_ends[n_threads-1] = this->n_neurons-1;

// // Starts the threads
// // NOTE: spawning threads costs roughly 10 us/thread
// // it is a non-negligible overhead
// std::vector<std::thread> evolver_threads(n_threads);
// for (unsigned int i = 0; i < n_threads; i++){
// evolver_threads[i] = std::thread(evolve_bunch, bunch_starts[i], bunch_ends[i] );
// }

// // Waits the threads
// for (unsigned int i = 0; i < n_threads; i++){
// evolver_threads[i].join();
// }
// } // ~multithreading evolution

// thread_pool.join();
// perf_mgr.end_recording("evolution");

// // BAD CODE HERE. I copied ane pasted, but it's a test.
// // Uses the thread pool to handle the incoming spikes of the neurons
// perf_mgr.start_recording("spike_handling");
// // In case few neurons are present, do not use multithreading
// // to avoid overhead
// if (n_threads == 0){
// evolve_bunch(0, n_neurons);
// for (const auto& neuron : neurons) {
// boost::asio::post(thread_pool, boost::bind(&Neuron::handle_incoming_spikes, neuron));
// }
// else{// multithreading spike handling

// std::vector<unsigned int> bunch_starts(n_threads), bunch_ends(n_threads);

// for (unsigned int i = 0; i < n_threads; i++){
// bunch_starts[i] = i*static_cast<unsigned int>(this->n_neurons)/n_threads;
// bunch_ends[i] = (i + 1)*static_cast<unsigned int>(this->n_neurons)/n_threads - 1;
// }

// // Ensures that all neurons are covered
// bunch_ends[n_threads-1] = this->n_neurons-1;

// // Starts the threads
// // NOTE: spawning threads costs roughly 10 us/thread
// // it is a non-negligible overhead
// std::vector<std::thread> evolver_threads(n_threads);
// for (unsigned int i = 0; i < n_threads; i++){
// evolver_threads[i] = std::thread(evolve_bunch, bunch_starts[i], bunch_ends[i] );
// }

// // Waits the threads
// for (unsigned int i = 0; i < n_threads; i++){
// evolver_threads[i].join();
// }
// } // ~multithreading spike_handling

// thread_pool.join();
// perf_mgr.end_recording("spike_handling");
// ~ENHANCEMENT 9

// Uses the thread pool to evolve the neurons
perf_mgr.start_recording("evolution");
for (const auto& neuron : neurons) {
boost::asio::post(thread_pool, boost::bind(&Neuron::evolve, neuron));
}
thread_pool.join();
perf_mgr.end_recording("evolution");

// Uses the thread pool to handle the incoming spikes of the neurons
perf_mgr.start_recording("spike_handling");
for (const auto& neuron : neurons) {
boost::asio::post(thread_pool, boost::bind(&Neuron::handle_incoming_spikes, neuron));
}
thread_pool.join();
perf_mgr.end_recording("spike_handling");
// // ~ POOL TEST

// TODO: spike emission is moved here in the population evolution because
// it's not thread safe. Accessing members of other instances requires
// a memory access control.
// start = std::chrono::high_resolution_clock::now();
perf_mgr.start_recording("spike_emission");
this->n_spikes_last_step = 0;

Expand All @@ -369,8 +314,6 @@ void Population::evolve()

}
perf_mgr.end_recording("spike_emission");
// end = std::chrono::high_resolution_clock::now();
// timestats_spike_emission += (double)(std::chrono::duration_cast<std::chrono::microseconds>(end-start).count());
}

void Population::print_info()
Expand Down
3 changes: 2 additions & 1 deletion quilt/core/neurons_base.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,7 +64,7 @@ void Neuron::connect(Neuron * neuron, double weight, double delay){


void Neuron::handle_incoming_spikes(){

// get_global_logger().log(WARNING, "Handling spikes");
while (!(incoming_spikes.empty())){ // This loop will be broken later

auto spike = incoming_spikes.top();
Expand Down Expand Up @@ -108,6 +108,7 @@ void Neuron::handle_incoming_spikes(){


void Neuron::evolve(){
// get_global_logger().log(WARNING, "Evolving neuron");
if (spike_flag){
// cout << "Spike flag was True at t: "<< evo->now <<endl;
on_spike();
Expand Down

0 comments on commit fc82b6f

Please sign in to comment.