#include <cstdio>
#include <mpi.h>

// 分析负载均衡的辅助程序
int main(int argc, char *argv[]) {
    int id, p;
    MPI_Init(&argc, &argv);
    MPI_Comm_size(MPI_COMM_WORLD, &p);
    MPI_Comm_rank(MPI_COMM_WORLD, &id);

    int n = 100000;
    if (argc == 2) {
        n = atoi(argv[1]);
    }

    // 计算每个进程的工作量
    int workload = 0;
    for (int i = 2 + id; i <= n; i += p) {
        workload++;
    }

    // 收集所有进程的工作量
    int *workloads = nullptr;
    if (id == 0) {
        workloads = new int[p];
    }
    MPI_Gather(&workload, 1, MPI_INT, workloads, 1, MPI_INT, 0, MPI_COMM_WORLD);

    if (id == 0) {
        printf("\n=== 负载均衡分析 (N=%d, P=%d) ===\n", n, p);
        printf("进程号\t检查的数字数量\t分配的数字范围\n");
        printf("------------------------------------------------\n");
        int total = 0;
        for (int i = 0; i < p; i++) {
            int start = 2 + i;
            int end = n;
            int count = workloads[i];
            total += count;
            printf("%d\t%d\t\t", i, count);
            if (count <= 5) {
                printf("[");
                for (int j = 0; j < count && j < 3; j++) {
                    printf("%d", start + j * p);
                    if (j < count - 1 && j < 2) printf(", ");
                }
                if (count > 3) printf(", ...");
                printf("]\n");
            } else {
                int last = start + (count - 1) * p;
                printf("[%d, %d, ..., %d] (步长=%d)\n", start, start + p, last, p);
            }
        }
        printf("------------------------------------------------\n");
        printf("平均工作量: %d\n", total / p);
        printf("最大工作量: %d\n", workloads[0]);
        printf("最小工作量: %d\n", workloads[p-1]);
        printf("负载不均衡度: %.2f%%\n", 
               100.0 * (workloads[0] - workloads[p-1]) / (double)workloads[0]);
        printf("\n");

        // 分析素数检测的计算成本
        printf("=== 计算成本分析 ===\n");
        printf("注意：小数字的素数检测快，大数字的素数检测慢！\n");
        printf("进程0检测的数字: 2, %d, %d, ... (小数字，检测快)\n", 2+p, 2+2*p);
        printf("进程%d检测的数字: %d, %d, %d, ... (大数字，检测慢)\n", 
               p-1, 2+(p-1), 2+2*(p-1), 2+3*(p-1));
        printf("\n");

        delete[] workloads;
    }

    MPI_Finalize();
    return 0;
}