BBS: Inland Empire Archive
Date: 01-06-56 (01:06) Number: 712
From: BRENT ASHLEY Refer#: NONE
To: HUGH MARTIN Recvd: NO
Subj: Word Count Contest Conf: (2) Quik_Bas
HM>Well done! I didn't think of using PEEK to scan the buffer
HM>(I used INSTR instead), but it makes a difference (about
HM>50% faster).
By far the best tool when writing it was bc's /a switch. By
looking at the assembly listings you can see just how you're
saving bytes and time with one method over another.
HM>I've forgotten now exactly how the programs were judged.
HM>Was size most important or speed? I opened and read the
Speed, size, cleverness in that order.
HM>file using CALL INTERRUPT, which saves about 5000 bytes
HM>over QB's OPEN and GET. That would have dropped your EXE
HM>size down to about 16K (same as mine). That's pretty small
Hadn't thought of that! I wondered how some folk got it smaller.
Ethan had said when he called that some had got it smaller, and
one had got almost as fast, but at 80kbytes.
When I first wrote it (no optimization) , it read my test file
(350k, I think) in 12 seconds. With some tweaking I quickly got
it to 3.5 seconds (helped greatly by the idea of using the full
range of integer values rather than a long integer for the
counter), and then leapt into the /a listings to tweak
further, until I got it to about 1.7 seconds (mostly after it
dawned on me that whitespace could just be <= 32 rather than
checking for CR, LF, TAB, etc.) The odd byte was saved by making
the DO loop do a post-comparison and putting the negative first on
the buffer adjustment, but that didn't really affect the speed.
The final change was a rearrangement of the logic so the most
likely events would be on the shortest logic path (e.g. I found
that there is usually much less whitespace than non-whitespace,
and that it was likely that there was more than one buffer-full,
so I performed the second comparisons on the less-likely events)
This was an interesting concept in that I wouldn't usually have
analysed the problem so deeply and it taught me a good lesson.
With those changes, I got to 1.47 seconds. I've still yet to see
how I can get it any faster in QB.
It certainly was fun - I hope they come out with a new challenge
soon! I'd be interested to see your entry, too. A 5k savings
would be useful in a couple of places I know...
-----
Brent
-----
~~___
X DeLuxe2 1.12 #10383 X
--- Maximus 2.00
* Origin: The Programmer's Paradise (416)482-1470 (1:250/801)
