Re: Ada 200X

[Craig Carey <research@ijs.co.nz>]

On Sun, 8 Jun 2003 11:14:28 +0000 (UTC), "Martin Dowie"
    <martin.dowie@no.spam.btopenworld.com> wrote:
...
>> for a strings package. A problem I had was the the string code in the
>> Charles package of Mr Heaney was too fast in comparison, which was
>> resolved by suppressing range checking.
>
>Sorry, can you expand on the 'too fast' statement?!?! :-)
>
Those packages are here:
http://home.earthlink.net/~matthewjheaney/charles/
http://www.ijs.co.nz/code/ada95_strings_pkg.zip (19KB

My strings package was intended to to be fast and open.
Mr H's need not be fast (but may be). The idealism includes that of
having functions return a private type (implying slicing is hidden
in the package).

Is MinGW too slow ?. Maybe someone could fix it perhaps.

The Charles code resizes strings using a factor of 2.0 (i.e. 100%).
I used 1.70 (70%) after some testing (in Windows 2000).

In 2003, the GCC compiler defaulted to using this figure for some
resizing of its garbage collector (ref gcc/gcc/doc/):

The resize factor is: Min (30% + 70% * (RAM / 1GB), 100%).

In my strings package, values outside of 1.6 and 2 (to 2.2) were
clearly slower.

So assuming a cut-off of 1.57, then MinGW GCC in Windows is suboptimally
slow when there is less 395 MB of RAM. MinGW does not have that parameter.

GNAT has a big factor of 2 so no mistakes there.


At 2003\01\27 13:01 -0600 Monday, Benjamin Kosnik wrote to gcc@gcc.gnu.org:
>
>time make
>248.736u 26.373s 4:59.60 91.8%  0+0k 0+0io 503943pf+0w
>
>time make CXXFLAGS="--param ggc-min-heapsize=8192"
>214.521u 27.306s 4:10.44 96.5%  0+0k 0+0io 512759pf+0w
>
>time make CXXFLAGS="--param ggc-min-heapsize=16384 --param ggc-min-expand=50"
>193.816u 30.121s 3:49.56 97.5%  0+0k 0+0io 524642pf+0w
>
>Erp. So, with these flags I can save a minute each recompile. C++
>compiles on templatized code end up being 25% faster.
>
>Aaaaaaaaaaghh!!!!!! Why aren't these flags documented!!!! Why aren't
>the defaults sane???
>
>-benjamin