Texas Instruments TI-89 - Floating Point Numbers

Chapter 16: Working with Numbers
183
TI
-
89 / TI
-
92 Plus Developer Guide
Not for Distribution
Beta Version January 26, 2001
Here is a coding example from a system routine which must temporarily alter this
mode setting. Numeric integration (
push_nint)
requires that all evaluation be
done with float numbers. So, this routine saves the mode, changes it, and
restores it before returning.
void push_nint (EStackIndex i, EStackIndex vi, EStackIndex j,
EStackIndex k)
{
Access_AMS_Global_Variables;
CONTROL_BITS old_NG_control = NG_control;
SET_ARITH_APPROX;
.
.
.
/* apply the quadrature algorithm */
.
.
.
NG_control = old_NG_control;
}
16.4. Floating Point Numbers
Applications can work with float numbers on the estack or in C floating-point
variables. The compiler supports two forms of floating-point values as described
in Chapter 2 of the compiler documentation. The calculator implementation uses
the standard C type double. The symbols BCD16 and Float are also defined to
be double. BCD16 is the recommended type for declaring floating-point variables
in applications.
This type uses a 16-digit mantissa and provides more accuracy. Thus, BCD16
variables provide the best results when implementing iterative algorithms that
require a great deal of floating-point computation.
push_Float
is the routine that converts a C floating-point value into a tagged
floating-point value on the expression stack. The 16-digit value is rounded to
14-digits, pushed onto the estack, and then a FLOAT_TAG is pushed on top.
BCD floating point supports floating point infinities. However,
push_Float
converts these values to their symbolic equivalents. In other words,
push_Float
converts a floating point plus infinity to PLUS_INFINITY_TAG, a floating point
minus infinity to MINUS_INFINITY_TAG, a floating point unsigned infinity to
PLUS_OR_MINUS_INFINITY_TAG, and a floating point NAN to
UNDEFINED_TAG.
BCD floating point supports an exponent range from -16384 to 16383. Tagged
float exponents are limited to the calculator range of -999 to 999.
push_Float
converts overflow values to the corresponding symbolic infinity and underflow
values to zero. Thus, while any tagged float can be moved into a C floating point
variable, not all C floating point values can be converted to tagged floats.