Mail Archives: djgpp-workers/1997/11/10/14:49:40
On 9 Nov 97 at 12:11, Charles Sandmann wrote:
> If a monster app is to be tested, I think it should be a recompile
> of GCC with itself using the new package - and some of the
> pathalogical data initialization codes to see how they perform in
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
What are these, please?
> both memory and speed.
>
While I have your attention:
I am doing various experiments with instrumenting source code to get
statistics on memory usage. I have found that calling
_go32_dpmi_remaining_physical_memory() frequently (like after every
malloc() call) slows things way down, to the extent that I have to
run experiments twice, once with the call to measure memory usage
and once without to get a more meaningful speed comparison.
Is there a faster way of getting the required info?
Cheers,
John Machin
1/27 Auburn Grove
Hawthorn East, VIC 3123, Australia
Phone: +61-3-98130561
modem]
>
> Tough. I don't see how could you test this with any real-world
> program without downloading some sources.
>
> For Emacs, you will need to download em2934b.zip, em1934r[1-3].zip
> and em1934s1.zip. These are 6 files, each one of them less than
> 1.4MB. If it fails to build, you will have to get em1934l[1-3].zip,
> which are another 4MB (I never tried without em1934l[1-3].zip, so I
> won't know if it will work).
For comparisons of performance (both of speed and memory usage), one
requires TYPICAL usages of various packages. For robustness tests,
one requires PATHOLOGICAL usages of packages (e.g. with an editor,
suck in a very large text file and change all periods to "foo"). In
both cases, this requires someone familiar with the package.
It is pointless for me to muck around with a package like emacs
because I have NO IDEA how to make it do things that represent
TYPICAL uses.
For me to download 6MB of stuff is a ***MAJOR PAIN IN THE
KEISTER***, even at something anywhere approaching the rated speed of
the modem; this is achieveable only at times that are off-peak in
most of the world, like early Sunday morning UTC-1100.
I will do this however for things with whose use I'm familiar, like
say gawk (and my own apps). As well as the malloc tracer/simulator
that I'm working on, I'm looking at a kit that stress-tests
malloc/free/etc for "correctness".
If this isn't enough:
To quote Eli : "Tough"
To quote Martin Luther (!King) : "Hier stehe ich, ich kann nicht
anders".
Regards,
John Machin
1/27 Auburn Grove
Hawthorn East, VIC 3123, Australia
Phone: +61-3-98130561
- Raw text -