I have been experimenting with the TP FPU Emulator. According to the
manual, the FPU has 8 stacks and so the following function

function Fib(N: Integer): Extended;
begin
  if N = 0 then
    Fib := 0.0
  else
    if N = 1 then
      Fib := 1.0
    else
      Fib := Fib(N-1) + Fib(N-2)
end;

will crash for N > 8. However, I am seeing FPU overflows for
N > 15, instead! More importantly, these overflows only occur with the
co-processor- the emulator seems quite happy to go up to 200 (of course, it
locked the computer up in the process...) This is all very strange...

I am running all this on a Pentium- does the Pentium simply have more FP
stacks? Does the FP Emulator not bother to check for overflows (Sounds
{*word*193} if the emulator has fewer stacks than the processor...)

The Pentium still has a 8 entry floating point stack, but some
programs extend the stack into memory, by copying things out whenever
a stack fault occurs.  I don't know if TP does this or not, but I
believe Windows does.

On 10 Feb 96 05:38pm, CHRIS RANKIN wrote to ALL:

 CR> I have been experimenting with the TP FPU Emulator. According to the
 CR> manual, the FPU has 8 stacks and so the following function
 CR>  
 CR> function Fib(N: Integer): Extended;
 CR> begin
 CR>   if N = 0 then
 CR>     Fib := 0.0
 CR>   else
 CR>     if N = 1 then
 CR>       Fib := 1.0
 CR>     else
 CR>       Fib := Fib(N-1) + Fib(N-2)
 CR> end;
 CR>  
 CR> will crash for N > 8. However, I am seeing FPU overflows for
 CR> N > 15, instead! More importantly, these overflows only occur with the
 CR> co-processor- the emulator seems quite happy to go up to 200 (of
 CR> course, it locked the computer up in the process...) This is all very
 CR> strange...
 CR> I am running all this on a Pentium- does the Pentium simply have more FP
 CR> stacks? Does the FP Emulator not bother to check for overflows (Sounds
 CR> {*word*193} if the emulator has fewer stacks than the processor...)

This is what bp\examples\dos\fib8087.pas says too.  I suspect
that it was true for the 8087 but somewhere along the line the
size of the stack increased to 16.  I get the same results on my
Pentium and my Cyrix 386->486 CPU.  The emulator must not have
the same stack restrictions.

Jud McCranie  jud.mccra...@swsbbs.com