Rankin/Wozniak Floating-Point Routines Revisited

[DavidL]

Greetings All,

After completing a project applying the Rankin/Wozniak floating-point routines quite some time ago I finally spent the time to finish polishing, documenting, and blending in the Red Book documentation. Attached is the result. My project used the routines with cc65 so there are wrapper functions at the end of the file. When I last ran them everything appeared to work well but "it's been a while"...

If anyone has any interest in this please feel free to reply with comments, suggestions, and (especially) corrections. I believe I've incorporated all the known corrections from previous discussions here but I may have missed something, especially in trying to incorporate/update the Red Book documentation.

Ready! Fire! Aim! Or something like that...

Thanks,
Dave

[BigDumbDinosaur]

DavidL wrote:

After completing a project applying the Rankin/Wozniak floating-point routines quite some time ago I finally spent the time to finish polishing, documenting, and blending in the Red Book documentation. Attached is the result.

What kind of file is a .S file?

[DavidL]

BigDumbDinosaur wrote:

DavidL wrote:

After completing a project applying the Rankin/Wozniak floating-point routines quite some time ago I finally spent the time to finish polishing, documenting, and blending in the Red Book documentation. Attached is the result.

What kind of file is a .S file?

Oh, that's the assembly file for the cc65 assembler, ca65.

[BigDumbDinosaur]

DavidL wrote:

BigDumbDinosaur wrote:

DavidL wrote:

After completing a project applying the Rankin/Wozniak floating-point routines quite some time ago I finally spent the time to finish polishing, documenting, and blending in the Red Book documentation. Attached is the result.

What kind of file is a .S file?

Oh, that's the assembly file for the cc65 assembler, ca65.

Are you aware that there were some logic errors in the original that were subsequently corrected? If so, did you roll those fixes into the above code? If I correctly recall, we have both the original and the fixes in the code section of 6502.org.

[DavidL]

BigDumbDinosaur wrote:

DavidL wrote:

BigDumbDinosaur wrote:

DavidL wrote:

After completing a project applying the Rankin/Wozniak floating-point routines quite some time ago I finally spent the time to finish polishing, documenting, and blending in the Red Book documentation. Attached is the result.

What kind of file is a .S file?

Oh, that's the assembly file for the cc65 assembler, ca65.

Are you aware that there were some logic errors in the original that were subsequently corrected? If so, did you roll those fixes into the above code? If I correctly recall, we have both the original and the fixes in the code section of 6502.org.

OP second paragraph, second sentence: "I believe I've incorporated all the known corrections from previous discussions here..."

[BigDumbDinosaur]

DavidL wrote:

OP second paragraph, second sentence: "I believe I've incorporated all the known corrections from previous discussions here..."

I guess what the ophthalmologist said was true when he told me I can't see worth a damn.

[DavidL]

Wow, can't say I expected this - 10 days on and even the crickets don't care.

[BigDumbDinosaur]

DavidL wrote:

Wow, can't say I expected this - 10 days on and even the crickets don't care.

Things ebb and flow around here. Many members have other hobby interests besides working with 65xx stuff and they may have had a look but don’t have the time at the moment to mess with it and report results. One indication of interest is the number of downloads on the file you’ve posted. Somebody is interested, I’d say.

Incidentally, around the time Rankin and Wozniak were writing this, MOS Technology developed a floating point package for the KIM called KIMath. KIMath uses binary-coded decimal instead of binary and hence doesn’t exhibit the decimal <—> binary hinkiness seen in binary FP implementations. Naturally, the code was larger and slower than the Rankin-Wozniak implementation.

Also, Charles.R. Bond wrote a BCD floating point package for the 6502 that uses Cordic functions to improve performance and decrease code size. A link to it is posted here in our code section.

[BigEd]

Thanks David for putting this together and sharing it. As BDD says, offerings are often a slow burn.

(Here's a previous thread about Charles Bond's package, thanks BDD for the reminder:
Calc65: a BCD floating point package
See also a later post in that thread for a further development of the package, and then further posts for more.)

(However, I wouldn't want to muddy the waters: this thread is about a binary floating point package, and the other is about a BCD package. Each has its place. It would be a disservice to this thread to argue the relative merits - I'm fairly sure there are existing threads which do that.)

[BigDumbDinosaur]

BigEd wrote:

(However, I wouldn't want to muddy the waters: this thread is about a binary floating point package, and the other is about a BCD package.

I only mentioned KIMath as a matter of historical interest. Thinking about it reminded me of Bond's implementation.

[BigEd]

Appreciated!

[Sheep64]

DavidL on Fri 20 May 2022 wrote:

Wow, can't say I expected this - 10 days on and even the crickets don't care.

BigEd on Fri 20 May 2022 wrote:

As BDD says, offerings are often a slow burn.

Your work exceeds one download per day and that is extremely popular on the 6502 Forum. Detailed feedback may take one month or more but I'll make a special effort to review your work.