.TH FLT_ARITH 3ACK " Fri June 30 1989"
.ad
.SH NAME
flt_arith \- high precision floating point arithmetic
.SH SYNOPSIS
.nf
.B #include <flt_arith.h>
.PP
.if t .ta 3m 13m 22m
.if n .ta 5m 25m 40m
struct flt_mantissa {
	long	flt_h_32;	/* high order 32 bits of mantissa */
	long	flt_l_32;	/* low order 32 bits of mantissa */
};

typedef struct {
	short	flt_sign;	/* 0 for positive, 1 for negative */
	short	flt_exp;	/* between -16384 and 16384 */
	struct flt_mantissa	flt_mantissa;	/* normalized, in [1,2). */
} flt_arith;

extern int	flt_status;
#define FLT_OVFL	001
#define FLT_UNFL	002
#define FLT_DIV0	004
#define FLT_NOFLT	010
#define FLT_BTSM	020
.PP
.B flt_add(e1, e2, e3)
.B flt_arith *e1, *e2, *e3;
.PP
.B flt_mul(e1, e2, e3)
.B flt_arith *e1, *e2, *e3;
.PP
.B flt_sub(e1, e2, e3)
.B flt_arith *e1, *e2, *e3;
.PP
.B flt_div(e1, e2, e3)
.B flt_arith *e1, *e2, *e3;
.PP
.B flt_modf(e1, intpart, fractpart)
.B flt_arith *e1, *intpart, *fractpart;
.PP
.B int flt_cmp(e1, e2)
.B flt_arith *e1, *e2;
.PP
.B int flt_str2flt(s, e)
.B char *s;
.B flt_arith *e;
.PP
.B flt_flt2str(e, buf, bufsize)
.B flt_arith *e;
.B char *buf;
.B int bufsize;
.PP
.B int flt_status;
.PP
.B #include <em_arith.h>
.B flt_arith2flt(n, e)
.B arith n;
.B flt_arith *e;
.PP
.B arith flt_flt2arith(e)
.B flt_arith *e;
.PP
.B flt_b64_sft(m, n)
.B struct flt_mantissa *m;
.B int n;
.PP
.B flt_b64_rsft(m)
.B struct flt_mantissa *m;
.PP
.B flt_b64_lsft(m)
.B struct flt_mantissa *m;
.SH DESCRIPTION
This set of routines emulates floating point arithmetic, in a high
precision. It is intended primarily for compilers that need to evaluate
floating point expressions at compile-time. It could be argued that this
should be done in the floating point arithmetic of the target machine,
but EM does not define its floating point arithmetic.
.PP
.B flt_add
adds the numbers indicated by
.I e1
and
.I e2
and stores the result indirectly through
.IR e3 .
.PP
.B flt_mul
multiplies the numbers indicated by
.I e1
and
.I e2
and stores the result indirectly through
.IR e3 .
.PP
.B flt_sub
subtracts the number indicated by
.I e2
from the one indicated by
.I e1
and stores the result indirectly through
.IR e3 .
.PP
.B flt_div
divides the number indicated by
.I e1
by the one indicated by
.I e2
and stores the result indirectly through
.IR e3 .
.PP
.B flt_modf
splits the number indicated by
.I e
in an integer and a fraction part, and stores the integer part through
.I intpart
and the fraction part through
.IR fractpart .
So, adding the numbers indicated by
.I intpart
and
.I fractpart
results (in the absence of rounding error) in the number
indicated by
.IR e .
Also, the absolute value of the number indicated by
.I intpart
is less than or equal to the absolute value of the number indicated by
.IR e .
The absolute value of the number indicated by
.I fractpart
is less than 1.
.PP
.B flt_cmp
compares the numbers indicated by
.I e1
and
.I e2
and returns -1 if
.I e1
<
.IR e2 ,
0 if
.I e1
=
.IR e2 ,
and 1 if
.I e1
>
.IR e2 .
.PP
.B flt_str2flt
converts the string indicated by
.I s
to a floating point number, and stores this number through
.IR e.
The string should contain a floating point constant, which consists of
an integer part, a decimal point, a fraction part, an \f(CWe\fP or an
\f(CWE\fP, and an optionally signed integer exponent. The integer and
fraction parts both consist of a sequence of digits. They may not both be
missing. The decimal point, the \f(CWe\fP and the exponent may be
missing.
.PP
.B flt_flt2str
converts the number indicated by
.I e
into a string, in a scientific notation acceptable for EM. The result is
stored in
.IR buf .
At most
.I bufsize
characters are stored.
.PP
.B flt_arith2flt
converts the number
.I n
to the floating point format use in this package and returns the result
in
.IR e .
.PP
.B flt_flt2arith
truncates the number indicated by
.I e
to the largest integer value smaller than or equal to the number indicated by
.IR e .
It returns this value.
.PP
Before each operation, the
.I flt_status
variable is reset to 0. After an operation, it can be checked for one
of the following values:
.IP FLT_OVFL
.br
an overflow occurred. The result is a large value with the correct sign.
This can occur with the routines
.IR flt_add ,
.IR flt_sub ,
.IR flt_div ,
.IR flt_mul ,
.IR flt_flt2arith ,
and
.IR flt_str2flt .
.IP FLT_UNFL
.br
an underflow occurred. The result is 0.
This can occur with the routines
.IR flt_div ,
.IR flt_mul ,
.IR flt_sub ,
.IR flt_add ,
and
.IR flt_str2flt .
.IP FLT_DIV0
.br
divide by 0. The result is a large value with the sign of the dividend.
This can only occur with the routine
.IR flt_div .
.IP FLT_NOFLT
.br
indicates that the string did not represent a floating point number. The
result is 0.
This can only occur with the routine
.IR flt_str2flt .
.IP FLT_BTSM
.br
indicates that the buffer is too small. The contents of the buffer is
undefined. This can only occur with the routine
.IR flt_flt2str .
.PP
The routine
.I flt_b64_sft
shifts the mantissa
.I m
.I |n|
bits left or right, depending on the sign of
.IR n .
If
.I n
is negative, it is a left-shift; If
.I n
is positive, it is a right shift.
.PP
The routine
.I flt_b64_rsft
shifts the mantissa
.I m
1 bit right.
.PP
The routine
.I flt_b64_lsft
shifts the mantissa
.I m
1 bit left.
.SH FILES
~em/modules/h/flt_arith.h
.br
~em/modules/h/em_arith.h
.br
~em/modules/lib/libflt.a