☰

GNU Parallel 用于迭代 while 循环，嵌套

tag-icon • tag-icon gnu-parallel
GNU Parallel 用于迭代 while 循环，嵌套

我有一段代码在超级计算机上需要 30 个小时；管理员说它一次只能使用一个核心。它通过 60 x 60 矩阵执行一段成像代码，每次迭代（3600 次）运行 zdichi_eps 成像程序并将数据写入文件。我的想法是，可以使用 GNU PARALLEL 命令重写下面的脚本，并且可能使用 for 循环而不是两个 while 循环。由于我最多可以使用 64 个核心，因此我认为这会显着减少时间。请看一下并告诉我您的想法。这将如何改变？是否可以？谢谢！我很乐意为任何有帮助的人捐款，以获得一个可以让我免于与数据中心斗争的工作版本！

set tb = gau_9.tab      # tab file
set target = reformat   # target name (from the command line)
set dte = May2010v1     # date of the run.
set mult1 = 0.7958  # multiplication factor
set mult2 = `echo $mult1 | awk '{printf("%.6f",$1/2)}'`
set inclin = 40       # inclination angle
set vsin = 113.3        # vsini
set aim = 0.03221 # <chisq aim>
set tphot = 5600    # photosphere temperature
set tspot = 4100    # spot temperature
set omega0 = 11.0038    # omega_0; matched to the period
set startx = 10.8   # starting omega_eq
set stepx = 0.01    # step in omega_eq
set nx = 60         # number of datapoint for omega_eq
set starty = 0.0     # starting domega
set stepy = 0.01    # step in domega
set ny = 60     # number of datapoints for domega.
set npts = 30       # number of points used in zdichi_ep        set spec = 
$target.rs
set tab  = $tb
set b1 = $target.b1
set s1 = $target.s1
set out  = $target$aim$dte.out
set map  = $target$aim$dte.m
set junk = $target.dr_junk
set screen = $target.screen.dat
set zdi = $target.zdi.in
set om
set do

set ntot = 0
echo " " > $out 
echo " " > $junk
echo " " > $screen

echo "starting diffrot" > $target.afile

set i = 0 

while ($i < $nx) 
  set omega = `echo $startx $stepx $i | awk '{printf("%.5f",$1+$2*$3)}'`
  echo "Count i, Count ny " $i $ny
  echo "omega : " $omega
  set line
  set j = 0 
  while ($j < $ny) 
    echo "Count J Count ny " $j  $ny
    set domeg = `echo $starty $stepy $j | awk '{printf("%.5f",$1+$2*$3)}'`
    echo "1 delta omega : " $domeg
    set test = `echo $omega $domeg | awk '{tmp = $1-0.361387*$2-4.461019; 
    tmp = sqrt(tmp*tmp); if (tmp < 0.02) print 1; else print 0; }'`
    echo "test value : " $test 
    set test = 1
    if ($test == 0) then
        set spot = 9.000000
        set chi2 = 9.000000
        set testzdi = 9.00000
    else
        set beta  = `echo $omega $omega0 | awk '{printf("%e",1.0-$1/$2)}'`
        set gamma = `echo $domeg $omega0 | awk '{printf("%e",$1/$2)}'`
        echo "test, beta and gamma : " $test $beta $gamma
        echo "22 sending data to zdichi via diffrot2"
        echo "33 The data sent :" $vsin $spec $tab $mult1 $beta $aim $gamma $mult2 >> $screen

        # setting up the input file for zdichi_eps

        echo n > $zdi
        echo n >> $zdi
        echo 5000 $inclin $vsin >> $zdi
        echo n >> $zdi
        echo 1 >> $zdi
        echo $tphot $tspot >> $zdi
        echo $beta $gamma >> $zdi
        echo $spec >> $zdi
        echo $tab >> $zdi
        echo $mult1 $mult2 >> $zdi
        echo y >> $zdi
        echo 1.0 >> $zdi
        echo 1 >> $zdi
        echo y >> $zdi
        echo 1 >> $zdi
        echo $aim $npts >> $zdi
        echo $b1 >> $zdi
        echo y >> $zdi
        echo $s1 >> $zdi

        # finished setting up the input file for zdichi_eps

        /home/zdichi_eps < $zdi >> $out
        echo "44 finished with zdichi" 
        @ ntot++ 
        set chi2 = `tail -20 $out | grep '<29>' | awk '{printf("%.6f", $3)}'`
        set spot = `tail -20 $out | grep '<29>' | awk '{printf("%.6f", $5)}'`
        set testzdi = `tail -20 $out | grep '<29>' | awk '{printf("%.6f", $6)}'`
        echo "55 The returned values were: " $omega $domeg $test $chi2 $spot $testzdi >> $screen
    endif   

    if ($i == 0) then 
        set do = `echo $do $domeg`
    endif

    set line = `echo $line $chi2`
    @ j++

  end
  set om = `echo $om $omega`
  echo $line >> $junk
  @ i++
end

答案1

并行化此类事物的最快方法：

  1. 写入所有参数文件reformat.zdi.${i}_${j}.in
  2. 通过并行在每个文件上调用您的程序

例如这个：

#!/bin/bash
for f in ./reformat.zdi.*.in ;do
    echo "/home/zdichi_eps < $f > ${f%.in}.out"
done | parallel -j64

答案2

在 Bash 中，首选方法是创建一个接受$i和$j作为参数的函数：

do_zdichi() {
  i="$1"
  j="$2"
  [set all variables here]
  /home/zdichi_eps < $zdi >> $out
  [postprocess output]
}
export -f do_zdichi
eval parallel do_zdichi ::: {0..$nx} ::: {0..$ny}

您只需要在函数中确保您写入的任何文件名都有唯一的名称（例如output.$i.$j可以，但myoutput不是），因此它不会被并行运行的另一个进程覆盖。

相关内容