Pre-allocating the string sounds like a good idea, but in order to make it work, you then need to ensure that you overwrite the pre-allocated contents, not add to or replace them.

Two ways you might do this are substr and vec used as lvaues, but the reality (on my system:) is that using either is way slower than allowing perl to manage the process itself.

C:\test>241328
Name "main::substr" used only once: possible typo at C:\test\241328.pl
+ line 6.
Name "main::bytewise" used only once: possible typo at C:\test\241328.
+pl line 5.
Name "main::vec" used only once: possible typo at C:\test\241328.pl li
+ne 7.
Benchmark:
running
 bytewise, substr, vec
, each for at least 5 CPU seconds
...

  bytewise:  5 wallclock secs ( 5.17 usr +  0.01 sys =  5.18 CPU) @  1
+.55/s (n=8)

Use of uninitialized value in substr at (eval 6) line 1.
    substr:  6 wallclock secs ( 6.05 usr +  0.00 sys =  6.05 CPU) @  0
+.83/s (n=5)

       vec:  6 wallclock secs ( 5.22 usr +  0.00 sys =  5.22 CPU) @  0
+.96/s (n=5)

         s/iter   substr      vec bytewise
substr     1.21       --     -14%     -47%
vec        1.04      16%       --     -38%
bytewise  0.647      87%      61%       --

C:\test>
[download]

I suspect that the best answer is to try and avoid producing your data 1 byte at a time. If you identified that process, there might be alternative ways of acheiving it more efficiently.