I wrote these float procedures and I found out that
they create problems:
My program is unpredictable and comp reboots.
Only after hours of debugging I conclude these functions
are the problem.

So what do I do to make them usable?

Thanks.


double mylog2(double nb){
double rez,y=1;
_asm{
fld [y]
fld [nb]
fyl2x
fstp [rez]
}
return rez;
}


double myexp2(double nb){
double rez,rez2,y=1,nbi,nbf;
short fcv,fcb;

_asm{
fstcw fcv
mov ax,fcv
mov fcb,ax
or fcv,0xC00
//------------
fld [nb]
fldcw fcv
frndint
fldcw fcb
fstp [nbi]
}
nbf=nb-nbi;
_asm{
fld [nbf]
f2xm1
fstp [rez2]
//-------------
fld [nbi]
fld [y]
fscale
fstp [rez]
}
rez*=(rez2+1);
return (rez);
}

I added finit and fwait and now it seems to be OK.
Is there anything else I should do?

Thanks.

1st question would be why in some places you use brackets [] and in some you do not?

like:

fld [<some double value>]

but in other place:

fstcw fcv <-- why no brackets here?

Just curious.
I never used FPU in inline ASM - only in FASM.

In my experience FPU exceptions happens when FPU is not cleared of results and other calculation remains. For example if you push something into top of the stack (say with FLD instruction) and all 8 registers are busy - you get an exception.

That is maybe why FINIT helps - however, using it in every single case of FPU calculations - the performance suffers.

Also, some C/C++ run-time functions (or even some 3rd party functions) use MMX registers and they will damage FPU content too - because MMX registers are same physically as ST0 - ST7 in FPU. So, you can't mix FPU code with some library calls which use MMX.

I suppose i should use brackets everywhere (in this case)
but MS VS6 does not make the difference so I got sloppy.
I use finit only at the beginning of the function.
Crossed my mind about MMX but wasn't sharing of registers
only true for earliest Pentium processors?

Yes, indeed - both reg sets occupy same room.
http://softpixel.com/~cwright/programming/simd/mmx.php