Question

我有一个程序可以读取由另一个MPI程序生成的大量文件（大约100个文件，每个120MB），这可能需要一些时间。每个文件都包含相应子域中的变量。我想从这些文件中读取变量并将它们存储到4维数组的特定切片中。由于需要相当长的时间，我想将这段代码与openmp并行化：

  6     SUBROUTINE read_old_restart
  7         INTEGER :: ii
  8         INTEGER :: thread_ID
  9         INTEGER :: OMP_GET_THREAD_NUM
 10         CHARACTER(LEN=21) :: file_name
 11 
 12         !$OMP PARALLEL DO PRIVATE(ii,file_name)
 13         DO ii=0,Nproc_old-1
 14             IF(ii < 10) THEN
 15                 WRITE(file_name,401) "input/Restart_00", ii, ".out"
 16             ELSE IF(ii < 100) THEN
 17                 WRITE(file_name,402) "input/Restart_0" , ii, ".out"
 18             ELSE
 19                 WRITE(file_name,403) "input/Restart_"  , ii, ".out"
 20             END IF
 21             PRINT*, "Thread = ", OMP_GET_THREAD_NUM(), "Reading ", file_name
 22 401 format(a16,I1,a4)
 23 402 format(a15,I2,a4)
 24 403 format(a14,I3,a4)
 25             OPEN (unit=321, file=TRIM(file_name), status="old", form="unFORMATted")
 26             READ(321) t                       ,                         & 
 27                       old_u         (:,:,:,ii),                         & 
 28                       old_v         (:,:,:,ii),                         & 
 29                       old_w         (:,:,:,ii),                         &
 30                       old_p         (:,:,:,ii),                         &
 31                       old_uc        (:,:,:,ii),                         &
 32                       old_vc        (:,:,:,ii),                         &
 33                       old_wc        (:,:,:,ii),                         &
 34                       old_un2       (:,:,:,ii),                         &
 35                       old_vn2       (:,:,:,ii),                         &
 36                       old_wn2       (:,:,:,ii),                         &
 37                       old_un1       (:,:,:,ii),                         &
 38                       old_vn1       (:,:,:,ii),                         &
 39                       old_wn1       (:,:,:,ii),                         &
 40                       old_p1        (:,:,:,ii),                         &
 41                       old_viscu     (:,:,:,ii),                         &
 42                       old_viscv     (:,:,:,ii),                         &
 43                       old_viscw     (:,:,:,ii),                         &
 44                       old_convu     (:,:,:,ii),                         &
 45                       old_convv     (:,:,:,ii),                         &
 46                       old_convw     (:,:,:,ii),                         &
 47                       statindex               ,                         &
 48                       old_umn       (:,:,:,ii),                         &
 49                       old_uumn      (:,:,:,ii),                         &
 50                       old_urms      (:,:,:,ii),                         &
 51                       old_mass_frac (:,:,:,:,:,ii), &                   
 52                       old_enthT     (:,:,:,:,ii)
 53             CLOSE (321)
 54         END DO
 55         !$OMP END PARALLEL DO
 56     END SUBROUTINE read_old_restart

代码编译并运行每个线程的第一个循环。这是输出：

 Thread =            3 Reading input/Restart_030.out
 Thread =            7 Reading input/Restart_067.out
 Thread =            2 Reading input/Restart_020.out
 Thread =            6 Reading input/Restart_058.out
 Thread =            9 Reading input/Restart_085.out
 Thread =            8 Reading input/Restart_076.out
 Thread =            5 Reading input/Restart_049.out
 Thread =            4 Reading input/Restart_040.out
 Thread =           11 Reading input/Restart_103.out
 Thread =            0 Reading input/Restart_000.out
 Thread =            1 Reading input/Restart_010.out
 Thread =           10 Reading input/Restart_094.out

代码似乎正在运行并且卡在上面的输出上。运行顶部时，我无法使用任何CPU。知道为什么它不按预期工作？

Answer 1

您应该为单元号使用专用整数变量，并为每个线程将其设置为不同的值。使用与不同线程不同的相同文件单元是一个麻烦的方法。我很惊讶它没有崩溃。

并行读取多个文件并将它们存储在共享变量中

1 个答案: