Fortran程序内存分配异常：malloc(): corrupted top size报错排查

问题成因 #

报错malloc(): corrupted top size是堆内存被非法写入破坏导致的，核心原因是数组下标越界访问：

1. 当n_grid=19、ds=3d0、L=59.99999d0时，ds*n_grid=57d0 < L，粒子坐标经过周期性边界条件处理后范围是[0, L)。当坐标接近L（比如59.99999）时，floor(rco/ds)=floor(59.99999/3)=19，加1后得到20，但dn数组的维度是(19,19,19)，最大下标为19，此时写入dn(20,q,r)会直接越界，破坏堆内存的管理结构。
2. 从运行输出能看到sum(dn)=183355远小于n*nc=193536，说明大量粒子的索引越界，未被正确统计到dn数组中，同时非法写入了其他内存区域。
3. 当n_grid>=20时，ds*20=60d0 >= L，此时floor(L/ds)=19，加1后是20，刚好匹配dn数组的最大下标，不会越界，因此程序正常运行。

修复方案 #

针对索引越界问题，采用两种互补的修复手段：

1. 正确计算n_grid：根据盒子尺寸L和网格步长ds，计算刚好能覆盖整个盒子的网格数，避免出现网格范围小于盒子的情况：

   n_grid = ceiling(L / ds)  ! 用天花板函数确保网格总长度>=L
   

2. 强制索引边界检查：对计算出的网格索引做模运算，确保其始终落在1~n_grid范围内（即使出现浮点精度误差导致的超界）：

   p = mod(floor(rco(1, (i-1)*n + j)/ds), n_grid) + 1
   q = mod(floor(rco(2, (i-1)*n + j)/ds), n_grid) + 1
   r = mod(floor(rco(3, (i-1)*n + j)/ds), n_grid) + 1
   

   注：由于坐标已通过周期性边界条件处理为[0,L)，floor(rco/ds)结果非负，模运算后范围为0~n_grid-1，加1后刚好是合法的数组下标。

修改后的完整代码 #

program localdensity

implicit none

    integer :: i , j , k , a , b , n_grid , n , nc , p , q , r , ncin1
    double precision ::  L , ds  , ignore 
    double precision, allocatable :: dn(:,:,:)                               
    double precision, allocatable :: rco(:,:)             
      
    ! INPUT PARAMETERS !

    ncin1 = 14                        ! total configurations in each block file
    nc =  ncin1*1                     ! total block files
    ds = 3d0                          ! dx = dy = dz = 2 A 

    ! INPUT PARAMETERS !

    print*, 'started reading data file'
    open (1, file = 'Silica_13824_negative_press-11_100.txt', status = 'old')  
  
    read(1,*) n ,  ignore  ,  L                          
  
    close(1) 
    
    print* ,  n , L
    L = 59.99999d0
    !L = 60.0001d0

    allocate(rco(3,nc*n))
    ! 修复：正确计算n_grid，确保覆盖整个盒子
    n_grid = ceiling(L / ds)
    print*, 'calculated n_grid: ', n_grid
    allocate(dn(n_grid , n_grid , n_grid))

    open (1, file = 'Silica_13824_negative_press-11_100.txt', status = 'old')  

    do j = 1 , nc

        read(1,*)                            
        do i=1,n
            k=i+(j-1)*n
            read(1,*)  a , b , rco(1,k), rco(2,k), rco(3,k) ! reading data (x,y,z)
        end do
    end do
    print*, 'ended reading data file n :' , n , L
    close(1)  
   
    rco(:,:) = rco(:,:) - L*floor(rco(:,:)/L) ! Appying Periodic Boundary condition

    dn = 0   
    print* 

    do i = 1,nc
        do j = 1,n
            ! 修复：用模运算确保索引在合法范围内
            p = mod(floor(rco(1 , (i-1)*n + j) / ds), n_grid) + 1
            q = mod(floor(rco(2 , (i-1)*n + j) / ds), n_grid) + 1
            r = mod(floor(rco(3 , (i-1)*n + j) / ds), n_grid) + 1
            dn(p,q,r) = dn(p,q,r)  + 1 
         
        end do  
    if (mod(i,ncin1)==0) print*, 'block', i/ncin1, 'done'
    end do

    print*, 'total particles counted :', sum(dn) ,'; n*nc = ', n*nc
    print*, 'nc is ' ,  nc   
    dn = dn/nc
    print*, sum(dn)
    print*,'n_grid' ,n_grid

    open (unit = 2, file = "parameters_neg.txt" , action="write" , status="replace")
        write(2,*) n , nc , L , n_grid
    close(2)

    open (unit = 3, file = "data_from_program_neg.txt" , action="write" , status="replace")
             
    do i = 1 , n_grid
        do j = 1 , n_grid
            do k = 1 , n_grid   
         
                write(3,*) i , j , k , dn(i,j,k)  
         
            end do
        end do                          
    end do  

    close(3)
    
    print*, 'Program has computed everything, data has been saved to data_from_program.txt. Now run file named local_density.py '

endprogram localdensity

内容的提问来源于stack exchange，提问作者Mayank Sharma