#include <pthread.h>
#include <atomic>
#include <chrono>
#include <cmath>
#include "row.hpp"
#include "sort_algorithms.hpp"

struct AdaptiveArgs {
    MatrixSortContext* ctx;
    std::atomic<int>* next_row;
    int total_rows;
};

void* adaptive_worker(void* arg) {
    AdaptiveArgs* aa = (AdaptiveArgs*)arg;
    int current_chunk = 1; // Ξεκινάμε με μικρό chunk

    while (true) {
        int start = aa->next_row->fetch_add(current_chunk);
        if (start >= aa->total_rows) break;

        int end = std::min(start + current_chunk, aa->total_rows);
        
        auto start_time = std::chrono::steady_clock::now();
        for (int i = start; i < end; ++i) {
            uint8_t* row_ptr = aa->ctx->A + (static_cast<size_t>(i) * aa->ctx->W);
            row<uint8_t> r(row_ptr, (uint32_t)aa->ctx->W, false);
            quick_sort(r);
        }
        auto end_time = std::chrono::steady_clock::now();
        
        // Λογική AIMD: Αν τελείωσε γρήγορα, αύξησε το chunk, αλλιώς μείωσέ το
        auto duration = std::chrono::duration_cast<std::chrono::microseconds>(end_time - start_time).count();
        if (duration < 500) current_chunk += 1; // Additive Increase
        else current_chunk = std::max(1, current_chunk / 2); // Multiplicative Decrease
    }
    return nullptr;
}

long long run_adaptive_scheduler(MatrixSortContext* ctx, int K) {
    int total_rows = ctx->F * ctx->H;
    std::atomic<int> next_row(0);
    pthread_t threads[K];
    AdaptiveArgs args[K];

    auto t1 = std::chrono::high_resolution_clock::now();
    for (int i = 0; i < K; ++i) {
        args[i] = {ctx, &next_row, total_rows};
        pthread_create(&threads[i], nullptr, adaptive_worker, &args[i]);
    }
    for (int i = 0; i < K; ++i) pthread_join(threads[i], nullptr);
    auto t2 = std::chrono::high_resolution_clock::now();

    return std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count();
}