gimp-print-devel

Re: [Gimp-print-devel] Performance

[<sh...@al...>]

> I did a round of performance tuning tonight.  Essentially, I got rid
> of all but a very few uses of long long; the performance improvement
> is about 40%.

Hmm, I'm surprised it made that much of a difference.

Anyway, from a performance point of view, wouldn't it have made more sense
to go from long long to just plain long, rather than int?

For i386, long and int are equivalent, but, on systems with 64 bit CPU's,
long is equivalent to long long, but, these are precisely the same CPU's
that won't take the same performance hits for using long long that the
i386's do.

Eric

Re: [Gimp-print-devel] Performance

[<sh...@al...>]

> Well, the error values are all ints.  Making them longs would consume
> more memory.

It's hard to predict that.  Compilers often align variables to start on
word boundaries anyway, so, memory usage would be the same if that was done.
For variables which don't exist on a word boundary, you need to do a shift
after loading the word into a register, which is why compilers tend not
to use them.

> The performance cost would probably be less on a 64-bit
> CPU (if it supports 64-bit arithmetic), though.

The only common 64 bit systems that I know of are the Alpha and the
Ultrasparc.  Both of these systems support 64 bit arithmetic.  I don't know
much about IA64.

Eric

Re: [Gimp-print-devel] Performance

[Robert L K. <rl...@al...>]

   From: sh...@al...
   Date: Thu, 02 Mar 2000 21:51:06 +0900

   > I did a round of performance tuning tonight.  Essentially, I got rid
   > of all but a very few uses of long long; the performance improvement
   > is about 40%.

   Hmm, I'm surprised it made that much of a difference.

Well, 32-bit arithmetic (add/subtract, at any rate) is one machine
instruction; 64-bit arithmetic is more involved.

   Anyway, from a performance point of view, wouldn't it have made more sense
   to go from long long to just plain long, rather than int?

   For i386, long and int are equivalent, but, on systems with 64 bit CPU's,
   long is equivalent to long long, but, these are precisely the same CPU's
   that won't take the same performance hits for using long long that the
   i386's do.

Well, the error values are all ints.  Making them longs would consume
more memory.  The performance cost would probably be less on a 64-bit
CPU (if it supports 64-bit arithmetic), though.

-- 
Robert Krawitz <rl...@al...>      http://www.tiac.net/users/rlk/

Tall Clubs International  --  http://www.tall.org/ or 1-888-IM-TALL-2
Member of the League for Programming Freedom -- mail lp...@uu...
Project lead for The Gimp Print --  http://gimp-print.sourceforge.net

"Linux doesn't dictate how I work, I dictate how Linux works."
--Eric Crampton