Efficient way of converting longword -> string

[earok]

Mostly interested to see what opinions are about converting longwords to strings, I'll need to roll my own implementation for Scorpion sooner or later but I haven't decided on the exact approach.


JOTD shared code here that converts a long word to a string using DIVU #10 (including handling for longwords that have a number too large to fit into a word even after a divide by 10).

https://stackoverflow.com/questions/...umber-to-ascii

The code that's part of Amiblitz approaches the issue a different way that doesn't use DIVU at all - rather it uses a lookup table that contains values of a power of 10, and subtracts until the base value is smaller than the number in that row of the lookup table.

https://github.com/AmiBlitz/AmiBlitz...ntlib.ab3#L969




Edit: I wonder if the second approach could be improved to use a pre-rolled loop for each power of 10 value rather than a lookup table

[a/b]

I typically use a simple 10^n table approch with extra options like variable length output, leading space/zero/no padding. Getting rid of divs is fast enough, and then size takes priority (short and simple all-in-one).
If your main concern is speed, might as well consider binary subdivision (e.g. 5x, 2/3x, unrolled 1x until a match) and multiple tables. Maybe also first check if the number is 16-bit and skip to 10000 units if so.

[earok]

Cheeers

I ended up something like this for now, no divs, no lookup tables. It seems pretty good but I'll think about optimising it further over time.

From digits 9-5, use a loop to subtract the base value. (for example, for digit 9, subtract 100000000 until the base value is less than that).

Then copy the long word into a word, and repeat the above for digits 4-2

Finally, since the value on the last digit is guaranteed to be less than 10, just copy that value directly.