John W. Kennedy wrote:
> The _whole_ S/360 architecture was copied, but, whereas the 8/16/32/64
> two's-complement, byte-addressable data architecture has become
> universal, the S/360 floating-point design was never used outside of the
> context of full S/360 compatibility, and the modern descendants of the
> S/360 now offer the vastly superior IEEE-754 as an alternative. Note,
> too, that floating-point has become nearly a dead issue in the S/360
> world; the z/OS FORTRAN compiler is decades old, and several generations
> out of date.
Much Java code has a surprising amount of FP, as does much Linux code.
All IEEE, of course, which is why IBM seriously revamped IEEE math
support in the z990s. The G5s, G6s and z900s supported one hex FP
addition or multiplication per cycle, with a latency of (about) three
cycles, but because of the way it was implemented, IEEE float could
produce a result only every other cycle (somewhat simplified, BFP
numbers were converted to HFP and feed through the HFP pipeline). The
z990 (and z9) allow one result per cycle for both hex (HFP) and binary
(BFP/IEEE) float, plus support fused multiply adds (FMAs) which
improves the peak rate to two operations per cycle.
All in all, the z990's FP performance is respectable.
|
