The S/390 G5 floating-point unit

The floating-point unit of the IBM S/390(R) G5 Parallel Enterprise Server represents a milestone in S/390 floating-point computation. The S/390 G5 contains the first floating-point unit (FPU) to support both the S/390 hexadecimal floating-point architecture and IEEE Standard 754 for binary floating-point arithmetic. The S/390 G5 FPU supports the new S/390 floating-point architecture, which contains six operand formats, including the IEEE 754 standard singleword, doubleword, and quadword formats, which are all supported in hardware. An internal hexadecimal-based dataflow is implemented to support both hexadecimal- and binary-based architectures. The S/390 G5 server is generally available at 500 MHz. The microprocessor chip is fabricated in IBM CMOS 6X technology, with a device size of 0.25 mu m as drawn and 0.15 mu m effective length. The design of the G5 FPU is based upon that of its predecessor, the G4. All of the custom dataflow macros from the G4 hexadecimal FPU were utilized with only minor modifications, and only a few additional macros for format conversion were required. This paper discusses the changes that were required to support the new S/390 binary floating-point architecture.