home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Power-Programmierung
/
CD1.mdf
/
forth
/
compiler
/
fpc
/
help
/
sfloat.hlp
< prev
next >
Wrap
Text File
|
1988-11-02
|
4KB
|
66 lines
SFLOAT -- Software Floating Point Package
By Robert L. Smith
This is a software floating point package for use with the F-PC Forth
system. It follows most of the recommendations of the FVG (Forth Vendors
Group) Standard Floating-Point Extension. It is my intention that the
functions be as accurate as is reasonable, consistent with speed as the
secondary goal. The basic four functions plus square root should be precise to
one half a bit in the least significant bit of the representation. It is my
intention that the higher functions have intrinsic errors no larger than one
bit in the least significant place. The representation uses a 16 bit word for
the sign and biased exponent of the floating point number, and 32 bits for the
significand (sometimes called the "mantissa"). That means that the intrinsic
accuracy is approximately 9 and a half decimal digits. The largest and
smallest representable numbers cover roughly the range from 10 to the plus and
minus 4000 power, which is far greater than almost anyone would ever need.
The representation does NOT include special cases for infinity, unnormalized
numbers, and Not-A-Number. The format is definitely NOT the IEEE Floating
Point Standard format. The reason is basically one of speed. The functions
in this package are approximately 5 (FIVE) times faster than software
implementations using the IEEE standard. In spite of that, I do attempt to
meet the basic requirement of having the basic four functions be accurate to
within one-half a bit. The precision is somewhat greater than one would
obtain from the 32-bit IEEE format.
To load the package from F-PC, type FLOAD SFLOAT followed by a
carriage return.
After SFLOAT has been loaded, you may enter floating point numbers
in exponential format. The following is a list of possible entries:
-1.2345E4
.9876E-3
123.456(10)
12.345X-5
The rules are: If the number is negative, it must be preceeded with a
minus sign. Do not use "+" with positive numbers. The separator between
the mantissa and exponent may be almost any character other than a digit,
"+" sign, "-" sign, or comma. In the examples above, the "E" is used for
compliance with FVG, but the left parenthesis or "X" may be used, especially
if you desire to enter your numbers in another base. The terminating symbol
may be omitted, but you may wish to use a right parenthesis, as in the third
example above.
If you wish to enter floating point numbers without having to specify
the exponential part, type the word FLOATS . That will allow numbers with
imbedded decimal points (or commas) to be interpretted as floating point
numbers. A trailing decimal point will still cause the number to be treated
as a double precision number. To turn off that feature, invoke the word
DOUBLES .
The simplest numeric output is E. . Other possible functions are:
E.R , F. , and F.R . There is also a floating point stack printout word, .FS .
If you wish to turn off the Floating Point interpreter, enter NOFLOATING .
To re-enter the floating point interpreter, enter FLOATING . If you wish
to FORGET all or part of the floating point package, it is important to
invoke NOFLOATING prior to using FORGET .
The FVG Standard Floating-Point Extension has been published in Dr.
Dobb's Journal, Sept., 1984. The authors are Ray Duncan and Martin Tracy.