Compile Error

"The point is that it almost never actually is required, and it shouldn't be."

Most of the stuff I work on requires a lot of floating point. It would be extremely difficult, if not impossible, and certainly pointless, to write and maintain a fixed point implementation.

"An embedded systems programmer who allows his customer to impose something like floating-point maths as a type-2) requirement would be comitting corporate suicide."

It'll only be 'corporate suicide' if the programmer in question isn't capable of assessing the impact of the customer requirement.

"As a type-1) requirement, floating-point arithmetic essentially never happens except in cases where the customer has no idea what he's talking about."

Not in my workplace. The only requirements our customers place on our product is that it works reliably.

"The smaller the CPU, the more likely it is that floating-point is a bad idea compared to fixed-point arithmetics"

You're quite correct in the general case. What irritates me is the assumption that the general case must always be extended to every case. For instance, there are many factors involved in selecting a processor for a job - inevitably it winds up being a compromise, so the processor isn't ideally suited for all of the tasks it is expected to do. This is the part where the brain has to be switched on to discover whether those compromises are reasonable or are going to create headaches later on.

We use certain 8051 derivatives for some very good reasons. The use of floating point does not have any negative impact on anything of importance - reliability, development time, maintainability etc, in fact, it has a positive impact on all those things.

"The point is that it almost never actually is required, and it shouldn't be."

Most of the stuff I work on requires a lot of floating point. It would be extremely difficult, if not impossible, and certainly pointless, to write and maintain a fixed point implementation.

"An embedded systems programmer who allows his customer to impose something like floating-point maths as a type-2) requirement would be comitting corporate suicide."

It'll only be 'corporate suicide' if the programmer in question isn't capable of assessing the impact of the customer requirement.

"As a type-1) requirement, floating-point arithmetic essentially never happens except in cases where the customer has no idea what he's talking about."

Not in my workplace. The only requirements our customers place on our product is that it works reliably.

"The smaller the CPU, the more likely it is that floating-point is a bad idea compared to fixed-point arithmetics"

You're quite correct in the general case. What irritates me is the assumption that the general case must always be extended to every case. For instance, there are many factors involved in selecting a processor for a job - inevitably it winds up being a compromise, so the processor isn't ideally suited for all of the tasks it is expected to do. This is the part where the brain has to be switched on to discover whether those compromises are reasonable or are going to create headaches later on.

We use certain 8051 derivatives for some very good reasons. The use of floating point does not have any negative impact on anything of importance - reliability, development time, maintainability etc, in fact, it has a positive impact on all those things.

Most of the stuff I work on requires a lot of floating point.
Have you had a look at come of the SILabs, e.g. some of the f12x they have a "SFR driven" MAC. ?

Erik