Re: C++ programming challenge

* Peter Jansson:


Uhm, I'm sorry to report that I was lead astray by two Windows "features":

   1. In Windows, 'clock', used in that program to measure the time, reports
      wall time, not process time.

   2. For some obscure reason Windows XP, or perhaps it's the anti-virus, or
      whatever, causes a process to run extra slowly the first few times. In
      this case I ran the original program twice with a certain medium size
      text file, then with modification it ran 400 times faster!, which I could
      not believe was an arbitrary effect. But now testing it, and with a much
      larger file, it is as you reported, about the same time.


On my machine, Windows XP,

   #define BUFFER 16*1024U

for the same program text that you have, yields the best times (statistically).

You might want to put that also in your list. :-)

I don't have a bittorrent client so couldn't obtain your file, but with a half
gigabyte text file I constructed from a dictionary text, the processing is about
1/27 of the i/o time. So a marginal improvement, on that order, could be to read
e.g. four bytes (a long) at a time from the buffer to avoid memory accesses.

Just to be thorough I tested using the lower level Posix standard open etc., but
it turned out that it was not significantly faster for this than FILE*.

A less insignificant improvement could be to use a memory mapped file, because
that could conceivably leverage some OS virtual memory optimization. However,
while the Boost library does have some functionality for mapping a file to
memory, it beats me how to use that. It's not exactly well-documented.

Cheers & hth.,

- Alf (victim of Windows (the Eagles' newest greatest hit!))

--
Due to hosting requirements I need visits to <url: http://alfps.izfree.com/>.
No ads, and there is some C++ stuff! :-) Just going there is good. Linking
to it is even better! Thanks in advance!