Bash script writing 2 files with different output elements per input line

Your bash script can be tidied up a bit. Note that the read command can read the 3 fields into separate variables:

#!/bin/bash
rm -f file_1.txt file_2.txt    
While read f1 f2 f3; do
    hash=$(md5sum <<< $f3)
    printf "%s %s\n" "$hash" "$f3" >> file_1.txt
    printf "%s %s %s\n" "$f1" "$f2" "$hash" >> file_2.txt
done < input_file.txt

Could not figure for which string you want to compute the md5, this one-liner does it on the whole line, and outputs the processed 'input_file' as you wish in 'file1' and 'file2':

awk '{ "md5 -q -s \""$0"\"" | getline md5; 
     print md5" "$3 > "file1"; 
     print $1" "$2" "md5 > "file2" }' input_file

Hope it helps..

You can write both files at the same time from bash, like this:

; function to remove extraneous filename output from md5sum.  omit on 
; OS X, which has 'md5' command that already works this way.
md5() { set -- $(md5sum "$@"); echo "$1"; }

exec 3>file_1.txt 4>file_2.txt
while read left middle right; do
  md5="$(echo -n "$right" | md5)"
  echo >&3 "$md5 $right"
  echo >&4 "$left $middle $md5"
done <input_file.txt
exec 3>&- 4>&-

That assumes the simple whitespace-separated fields of your example; you would of course still have to do whatever sed magic is required to get the actual target for the MD5 sum.

It won't be very efficient, though. For better performance, you should use something like Perl or Python, which can do both the field extraction you're using sed for and the MD5 calculation all within a single process that is also much faster than the shell at looping over lines of input. Perl example:

perl -MDigest::MD5=md5_hex -lane '
  BEGIN { open $f1, ">file_1.txt"; open $f2, ">file_2.txt" }
  $md5 = md5_hex $F[2];
  print $f1 "$md5 $F[2]";
  print $f2 "$F[0] $F[1] $md5";
' input_file.txt

You may be able to increase the efficiency with pipes and parallel.

According to your pseudo-code, you want the md5 sum of the last element:

paste -d ' '     \
  input_file.txt \
  <(cut -d' ' -f3 input_file.txt | parallel echo '{}' \| md5sum | cut -d' ' -f1) |
  awk '{ print $4, $3 > "file_1.txt"; print $1, $2, $4 > "file_2.txt" }'

Explanation

The md5 sum is calculated in parallel in the process substitution, the output from here is "pasted" onto the original file. Finally awk takes care of placing the output into the correct files.

Edit

I agree with redShadow that this will never be very efficient in shell, as you need to sub-shell a lot. Here's an alternative in perl:

split.pl

use Digest::MD5 qw(md5_hex);
use v5.10;

open O1, ">file_1.txt" or die $!; open O2, ">file_2.txt" or die $!;

$, = " ";

while(<>) { chomp; 
  @F = split / +/;
  $md5 = md5_hex $F[2];
  say O1 $md5, $F[2];
  say O2 @F[0,1], $md5;
}
close O1; close O2;

Run like this:

<input_file.txt perl split.pl

Output in both cases:

file_1.txt

765ac5d0002aed1141a6a4e7b90e4ac9 1verylongalpha1234numeric1
b31901def07d436aed2c8028b2efa4ec 2verylongalpha1234numeric2
0722a6e50f6f8726f9754e7f71f9ad2c 3verylongalpha1234numeric3

file_2.txt

12347654 abcdfg 765ac5d0002aed1141a6a4e7b90e4ac9
34543673 nvjfur b31901def07d436aed2c8028b2efa4ec
75868643 vbdhde 0722a6e50f6f8726f9754e7f71f9ad2c