Contributor: STEVE SCHAFER
{
Here is how to trap errors on the 80X87. I am not sure yet how it works with
the FP emulation library, but if you have a math coprocessor, you can trap
any FP exceptions:
}
{$N+,E+}
program FloatTest;
{ compliments of Steve Schafer, Compuserve address 76711, 522 }
const
feInvalidOp = 01ドル;
feDenormalOp = 02ドル;
feZeroDivide = 04ドル;
feOverFlow = 08ドル;
feUnderFlow = 10ドル;
fePrecision = 20ドル;
procedure SetFpuExceptionMask (MaskBits: Byte); assembler;
{ Masks floating point exceptions so that they won't cause a crash }
var
Temp: word;
asm
fstcw Temp
fwait
mov ax, Temp
and al, $F0
or al, MaskBits
mov Temp, ax
fldcw Temp
fwait
end;
function GetFpuStatus: Byte; assembler;
{ determines the status of a previous FP operation }
var
Temp: word;
asm
fstsw Temp
fwait
mov ax, Temp
end;
procedure WriteStatus(Status: Byte);
{ This procedure is not necessary, it simply illustrates how to determine
what happenend }
begin
if (Status and fePrecision) 0 then Write('P')
else Write('-');
if (Status and feUnderflow) 0 then Write('U')
else Write('-');
if (Status and feOverflow) 0 then Write('O')
else Write('-');
if (Status and feZeroDivide) 0 then Write('Z')
else Write('-');
if (Status and feDenormalOp) 0 then Write('D')
else Write('-');
if (Status and feInvalidOp) 0 then Write('I')
else Write('-');
end;
var
X,Y: Single;
begin
SetFPUExceptionMask (feInvalidOp + feDenormalOp + feZeroDivide
+ feOverflow + feUnderflow + fePrecision);
X:= -1.0;
Y:= Sqrt(X); { Invalid Operation }
WriteStatus(GetFPUStatus);
Writeln(' ', Y:12, ' ', X:12);
X:= 0.0;
Y:= 1.0;
Y:= Y/X; { divide by Zero }
WriteStatus(GetFPUStatus);
Writeln(' ', Y:12, ' ', X:12);
X:= 1.0E-34;
Y:= 1.0E-34;
Y:= Y*X; { Underflow }
WriteStatus(GetFPUStatus);
Writeln(' ', Y:12, ' ', X:12);
end.