Compiler Exploitation of Decimal Floating-Point Hardware

Decimal Floating Point (DFP) is an alternative to the popular Binary Floating Point (BFP) representation for decimal numbers. With DFP representation, numbers are stored in decimal digits, allowing applications, such as calculators, to achieve the high level of accuracy. Traditionally, DFP arithmetic is supported by computationally expensive software libraries. To overcome this performance bottleneck, IBM introduced dedicated hardware DFP co-processors for the z10 and POWER6 processors which eliminate the need for DFP software libraries. This presentation will focus on the DFP representation and applications as well as the latest development on how the IBM C/C++ compiler and Java SDK6 are taking advantage of the new hardware DFP co-processors.

Greg Steffan

Last modified: Tue Aug 26 10:02:07 EDT 2008