NonStop S-Series Server Description Manual (G06.27+)

TNS Data Formats and Number Representations

HP NonStop S-Series Server Description Manual—520331-004

3-4

TNS Data Formats

The available TNS data element sizes are single bytes, two-byte words, four-byte

doublewords, and eight-byte quadruplewords. Signed integer data formats are

available as word, doubleword, and quadrupleword elements. Floating-point data

formats are available as doubleword and quadrupleword elements. See Figure 3-2.

The instruction set of all NonStop processors supports individual access to and

operations on 8-bit bytes, 16-bit TNS words, 32-bit TNS doublewords, and 64-bit TNS

quadruplewords. The figure on the facing page shows all of these data formats, plus

the legacy TNS floating-point and extended floating-point data formats.

A 16-bit TNS word can be a signed integer, an unsigned integer, a short pointer for

addressing within a particular memory segment, or a TNS instruction. As indicated in

the Figure 3-2, a TNS word is equivalent to two bytes. In TNS or accelerated

execution modes, all TNS word operands must begin at two-byte memory boundaries

for TNS instructions to operate correctly. In native execution mode, 16-bit operands

can begin at any byte, but there is a large performance penalty if the operand is not

aligned to a two-byte memory boundary.

A TNS doubleword can be a 32-bit signed integer, a 32-bit extended address, a

single-precision floating-point number, or a RISC instruction. A TNS doubleword is

equivalent to four bytes. In TNS or accelerated execution modes, all TNS doubleword

operands must begin at a two-byte memory boundary for correct execution, and there

is no benefit to four-byte alignment. In native execution mode, 32-bit operands can

begin at any byte, but there is a large performance penalty if the operand is not fully

aligned to a four-byte memory boundary.

A TNS quadrupleword can be a 64-bit signed integer, a fixed-point number

represented as a 64-bit signed integer with an assumed scale factor (radix point) fixed

at compile time, or an extended-precision TNS floating-point number. A

quadrupleword is equivalent to four TNS words or eight bytes. In TNS or accelerated

execution modes, all TNS quadrupleword operands must bein at a two-byte memory

boundary for correct execution, and there is no benefit to four-byte or eight-byte

alignment. In native execution mode, 64-bit operands can begin at any byte, but there

is a large performance penalty if the operand is not fully aligned to an eight-byte

boundary.

The single-precision and extended-precision floating-point formats are shown here for

comparison and are discussed in Instructions for Decimal and Floating-Point Arithmetic

on page 3-10.