While most users have been very happy with the quality of the emulation on Apple Computer Inc’s Power Macs, those using applications that try to access the Macintosh floating point unit directly have not been so lucky. The number of applications attempting this trick is relatively small; most use Apple’s documented ‘SANE’ numeric calls. However, some mathematical and computer-aided design packages hit the hardware directly, in order to extract the last drop of performance. These packages refuse to work on machines lacking a maths chip. The answer has generally been to use John Neil & Associates’s shareware SoftFPU, a control panel that traps and emulates direct floating point calls. Once it is installed, packages that normally require a maths chip will run quite happily, albeit slowly, on a maths chip-less machine. The trouble is that on a Power Macintosh, the emulated SoftFPU runs very slowly indeed – most users would say unusably slowly. The latest registered version of the product, costing $20, addresses this problem since it comes in a fat binary form that enables the SoftFPU to run native. One technically clued-up user we spoke to says the native version feels around five times as fast as the emulated version, but is still not amazingly fast. The native version includes an ‘FPU accuracy’ switch that enables users to switch between the PowerPC’s faster 64-bit accuracy mode and the slowwer 80-bit accuracy setting, which sacrifices performance and runs entirely in 68000 code. Those wanting to try the software for themselves can find the latest version of SoftFPU at ftp.netcom.com (/pub/johnneil) or on CompuServe or America Online. This is the 68000-only version, however and the native Power Macintosh version costs $20 in advance.
