IBM Power 595, Технический обзор и введение

Страница: 88 / 188

76
IBM Power 595 Technical Overview and Introduction
– Two floating-point operations
– One branch operation
The POWER6 processor implements the 64-bit IBM Power Architecture® technology. Each
POWER6 chip incorporates two ultrahigh dual-threaded Simultaneous Multithreading
processor cores, a private 4 MB level 2 cache (L2) for each processor, integrated memory
controller and data interconnect switch and support logic for dynamic power management,
dynamic configuration and recovery, and system monitoring.
2.6.2 Decimal floating point
This section describes the behavior of the POWER6 hardware decimal floating-point
processor, the supported data types, formats, and classes, and the usage of registers.
The decimal floating-point (DFP) processor shares the 32 floating-point registers (FPRs) and
the floating-point status and control register (FPSCR) with the binary floating-point (BFP)
processor. However, the interpretation of data formats in the FPRs, and the meaning of some
control and status bits in the FPSCR are different between the DFP and BFP processors.
The DFP processor supports three DFP data formats:
DFP32 (single precision): 4 bytes, 7 digits precision, -95/+96 exponent
DFP64 (double precision): 8 bytes, 16 digits precision, -383/+384 exponent
DFP128 (quad precision): 16 bytes, 34 digits precision, -6143/+6144 exponent
Most operations are performed on the DFP64 or DFP128 format directly. Support for DFP32
is limited to conversion to and from DFP64. For some operations, the DFP processor also
supports operands in other data types, including signed or unsigned binary fixed-point data,
and signed or unsigned decimal data.
DFP instructions that perform arithmetic, compare, test, quantum-adjustment, conversion,
and format operations on operands held in FPRs or FPR pairs are:
Arithmetic instructions Perform addition, subtraction, multiplication, and division
operations.
Compare instructions Perform a comparison operation on the numerical value of two
DFP operands.
Test instructions Test the data class, the data group, the exponent, or the
number of significant digits of a DFP operand.
Quantum-adjustment instructions
Convert a DFP number to a result in the form that has the
designated exponent, which can be explicitly or implicitly
specified.
Conversion instructions Perform conversion between different data formats or data
types.
Format instructions Facilitate composing or decomposing a DFP operand.
Enabling applications running on POWER6 systems to take advantage of the hardware
decimal floating point support depends on the programming language release level used by
the application and the operating system in which the application is running.