plugin_load_example.cc

Go to the documentation of this file.

 
 
// example on how to use the gdynamic library
 
// gdynamic
#include "gdynamicdigitization.h"
#include "gdynamicdigitization_options.h"
 
// gemc
#include "gfactory.h"
#include "event/gEventDataCollection.h"
 
// TODO: remove when C++20 is widely available
// ===== portable jthread-like wrapper =========================================
// If real std::jthread is present, use it. Otherwise, define a minimal shim
// that joins in the destructor (no stop_token support, but good enough here).
#if defined(__cpp_lib_jthread)   // header exists
#include <jthread>
using jthread_alias = std::jthread;
#else
// join: pause right here until that thread is finished.
class jthread_alias : public std::thread {
public:
    using std::thread::thread; // inherit all ctors
    ~jthread_alias() { if (joinable()) join(); }
    jthread_alias(jthread_alias&&) noexcept            = default;
    jthread_alias& operator=(jthread_alias&&) noexcept = default;
    // no copy
    jthread_alias(const jthread_alias&)            = delete;
    jthread_alias& operator=(const jthread_alias&) = delete;
};
#endif
 
const std::string plugin_name = "test_gdynamic_plugin";
 
auto run_simulation_in_threads(int                                                              nevents,
                               int                                                              nthreads,
                               const std::shared_ptr<GOptions>&                                 gopt,
                               const std::shared_ptr<GLogger>&                                  log,
                               const std::shared_ptr<const gdynamicdigitization::dRoutinesMap>& dynamicRoutinesMap) -> std::vector<std::unique_ptr<GEventDataCollection>> {
    std::mutex                                         collectorMtx;
    std::vector<std::unique_ptr<GEventDataCollection>> collected;
 
    // thread-safe integer counter starts at 1.
    // fetch_add returns the old value *and* bumps.
    // zero contention: each thread fetches the next free event number.
    std::atomic<int> next{1};
 
    // pool of jthreads.  jthread joins in its destructor so we don’t need an
    // explicit loop at the end.
    // each element represents one worker thread running your event-processing lambda.
    // std::vector<std::jthread> pool; use this when C++20 is widely available
    std::vector<jthread_alias> pool; // was std::vector<std::jthread>
 
    pool.reserve(nthreads);
 
    for (int tid = 0; tid < nthreads; ++tid) {
        // The capture [&, tid] gives the thread references to variables like tid
        pool.emplace_back([&, tid] // capture tid *by value*
        {
            // start thread with a lambda
            log->info(0, "worker ", tid, " started");
 
            int                                                             localCount = 0; // events built by *this* worker
            thread_local std::vector<std::unique_ptr<GEventDataCollection>> localRunData;
 
            while (true) {
                // repeatedly asks the shared atomic counter for “the next unclaimed event
                // number,” processes that event, stores the result, and goes back for more.
                // memory_order_relaxed: we only need *atomicity*, no ordering
                int evn = next.fetch_add(1, std::memory_order_relaxed); // atomically returns the current value and increments it by 1.
                if (evn > nevents) break;                               // exit the while loop
 
                auto gheader   = GEventHeader::create(gopt);
                auto eventData = std::make_unique<GEventDataCollection>(gopt, std::move(gheader));
 
                // each event has 2 hits
                for (unsigned i = 1; i < 3; i++) {
                    auto hit       = GHit::create(gopt);
                    auto true_data = dynamicRoutinesMap->at(plugin_name)->collectTrueInformation(hit, i);
                    auto digi_data = dynamicRoutinesMap->at(plugin_name)->digitizeHit(hit, i);
 
                    eventData->addDetectorDigitizedData("ctof", std::move(digi_data));
                    eventData->addDetectorTrueInfoData("ctof", std::move(true_data));
                }
 
                log->info(0, "worker ", tid, " event ", evn, " has ", eventData->getDataCollectionMap().at("ctof")->getDigitizedData().size(), " digitized hits");
 
                localRunData.emplace_back(std::move(eventData));
 
                ++localCount; // tally for this worker
            }
 
            // braces to locks the mutex when it's constructed and unlocks when it is destroyed
            {
                std::scoped_lock lk(collectorMtx);
                for (auto& evt : localRunData) {
                    // only collect 2 events total so that the log doesn't go crazy with the destructor
                    if (collected.size() >= 2) break;
                    collected.emplace_back(std::move(evt));
                }
                localRunData.clear();
            }
 
            log->info(0, "worker ", tid, " processed ", localCount, " events");
        }); // jthread constructor launches the thread immediately
    }       // pool’s destructor blocks until every jthread has joined
    return collected;
}
 
 
// emulation of a run of events, collecting data in separate threads
 
 
int main(int argc, char* argv[]) {
    // Create GOptions using gdata::defineOptions, which aggregates options from gdata and gtouchable.
    auto gopts = std::make_shared<GOptions>(argc, argv, gdynamicdigitization::defineOptions());
 
    // duplicate plugin logger here
    auto log = std::make_shared<GLogger>(gopts, SFUNCTION_NAME, PLUGIN_LOGGER);
 
    constexpr int nevents  = 10;
    constexpr int nthreads = 8;
 
    auto dynamicRoutinesMap = gdynamicdigitization::dynamicRoutinesMap({plugin_name}, gopts);
    if (dynamicRoutinesMap->at(plugin_name)->loadConstants(1, "default") == false) {
        log->error(1, "Failed to load constants for dynamic routine", plugin_name, "for run number 1 with variation 'default'.");
    }
 
    auto runData = run_simulation_in_threads(nevents, nthreads, gopts, log, dynamicRoutinesMap);
 
    // For demonstration, we'll simply print the event numbers.
    for (size_t i = 0; i < runData.size(); i++) { log->info(" > Event ", i + 1, " collected with local event number: ", runData[i]->getEventNumber()); }
 
    return EXIT_SUCCESS;
}