leue@galen.crd.ge.com (Bill Leue) (03/13/91)
Does anyone have any opinions (or better yet, experience) about compatibility problems with 68040 CPU's running standard Mac applications? I'm asking this for a friend who is considering buying an add-on 68040-based accelerator to his SE30. He is worried about the (reportedly somewhat different) floating-point implmentation of the 68040 versus the 68030 causing problems for applications. Thanks! -Bill Leue leue@crd.ge.com
niko@iastate.edu (Schuessler Nikolaus E) (03/18/91)
>Does anyone have any opinions (or better yet, experience) about >compatibility problems with 68040 CPU's running standard Mac >applications? I'm asking this for a friend who is considering >buying an add-on 68040-based accelerator to his SE30. He is worried >about the (reportedly somewhat different) floating-point implmentation >of the 68040 versus the 68030 causing problems for applications. Well, I am in no way an expert on 68040's, but I have done some reading in Motorola's technical literature. The difference in the floating point implementation is that the 68040 doesn't implement all the 68881/2 functions in hardware... The remaining functions are implemented as traps. This means they will run much slower (as someone else commented on one of these mac groups). --
lrm3@ellis.uchicago.edu (Lawrence Reed Miller) (03/18/91)
In article <1991Mar17.234807.14271@news.iastate.edu> niko@iastate.edu (Schuessler Nikolaus E) writes: > >>Does anyone have any opinions (or better yet, experience) about >>compatibility problems with 68040 CPU's running standard Mac >>applications? I'm asking this for a friend who is considering >>buying an add-on 68040-based accelerator to his SE30. He is worried >>about the (reportedly somewhat different) floating-point implmentation >>of the 68040 versus the 68030 causing problems for applications. > >Well, I am in no way an expert on 68040's, but I have done some reading >in Motorola's technical literature. The difference in the floating point >implementation is that the 68040 doesn't implement all the 68881/2 functions >in hardware... The remaining functions are implemented as traps. This means >they will run much slower (as someone else commented on one of these mac >groups). According to Motorola, the 68040 emulated FPU commands run as fast or faster than they would on a 68882. However, there has been some debate about whether or not the trap overhead actually causes these commands to slow down on the 68040 or not. There was some talk in comp.sys.next about whether or not this was slowing things down; personally I'd wait to _see_ how the Mac 68040 software implements these traps [& test some code!] before making up my mind. The 68040 has native commands for add, subtract, multiply, divide and square root (plus a couple others) [see MC68040/D]; the emulated commands are for sines, cosines and the like. For most people the trapped command issue is of little importance, as they rarely use commands like FATANH (floating hyperbolic arc tangent) in day-to-day life. Of course, for some people the execution speed of FATANH will be an issue, though. It depends on your application.