🚀 go-pugleaf

dcollier@cix.compulink.co.uk ("David Collier") wrote:

> Loops and floating-point.

> I know this sounds silly, but maybe if you just wrote out the FOR loops 
> as explicit code, A:=1 ; ..... ; INC(A) ; IF A< n THEN GOTO L1: ....

> Perhaps that would keep less in registers, and avoid your problem?

> Do you have the mathlib source?

Hi David,

   I tried that (INC() etc.) because I had the same thought but it is not
only when loops are executed. That problem went away when I made a DLL
calling MATHLIB (and I also have a lot of loops).

   There are in a few other cases, e.g. with Str.RealToStr, that calls
MATHLIB.IntPow and a few others and gives the wrong results. Since I have
the source to Str.MOD and, when I look closer into it, it has different
versions for _XTD amd _WINDOWS so *that* particular bug may be in the
source of the function. I will look into that.

   The MATHLIB source is basically in CoreMath.A and it is included in
the SourceKit. In CoreMath.A is where one official fix was for a push-pop
mismatch under Windows. The second fix I got was in the form of a changed
pragma setting in the MATHLIB.DEF file "call(reg_saved..." but I don't
have the settings here.

   Even if no obvious solution appears in the news-group the possibility
to discuss this may even make me find a solution myself. You know how it
is, sometimes it even helps telling your cat about the problem. And when
you have told him about it you say. Wait a second...Aaaha!

   We all suffer from that M2 is not as spread as C/C++. If more had
tried programming for Windows using TSM2, the bug would probably have
been fixed while TS were still interested in it.

   If anything comes up, I will post it here. Thanks so far.


--------------------------------
Bengt-Inge Dalenback, CATT
also at: 
Department of Applied Acoustics
Chalmers University of Technology
Gothenburg, SWEDEN
e-mail: bid@ta.chalmers.se
fax : +46 31 145154