Floating-Point Register Format; Floating-Point Programming Model; Ieee Real-Type Properties - Intel ITANIUM ARCHITECTURE - SOFTWARE DEVELOPERS MANUAL VOLUME 1 REV 2.3 Manual

Floating-point Programming Model

The floating-point architecture is fully compliant with the ANSI/IEEE Standard for
Binary Floating-Point Arithmetic (Std. 754-1985). There is full IEEE support for single,
double, and double-extended real formats. The two IEEE methods for controlling
rounding precision are supported. The first method converts results to the
double-extended exponent range. The second method converts results to the
destination precision. Some IEEE extensions such as fused multiply and add, minimum
and maximum operations, and a register format with a larger range than the minimum
double-extended format are also included.
5.1 Data Types and Formats
Six data types are supported directly: single, double, double-extended real (IEEE real
types); 64-bit signed integer, 64-bit unsigned integer, and the 82-bit floating-point
register format. A "Parallel FP" format where a pair of IEEE single precision values
occupy a floating-point register's significand is also supported. A seventh data type,
IEEE-style quad-precision, is supported by software routines. A future architecture
extension may include additional support for the quad-precision real type.
5.1.1 Real Types
The parameters for the supported IEEE real types are summarized in
Table 5-1.
IEEE Real-Type Parameters
Sign
E
max
E
min
Exponent bias
Precision (bits)
IEEE Memory Formats
Total memory format width (bits)
Sign field width (bits)
Exponent field width (bits)
Significand field width (bits)
5.1.2

Floating-point Register Format

Data contained in the floating-point registers can be either integer or real type. The
format of data in the floating-point registers is designed to accommodate both of these
types with no loss of information.
Volume 1, Part 1: Floating-point Programming Model

IEEE Real-type Properties

Single
+ or 
+127
126
+127
24
32
1
8
23
Double Double-Extended
+ or  + or 
+1023 +16383
1022 16382
+1023 +16383
53 64
64 80
1 1
11 15
52 64
5
Table 5-1.
Quad-Precision
+ or 
+16383
16382
+16383
113
128
1
15
112
1:85