The CORDIC algorithm is great! Just a few months ago, I implemented a sine generator with it and also used fixed point arithmetic. It is a real eye opener to see that you can substitute multiplications by bit shifting when you store the numbers in the “right“ format.
1. I suspect decimal floating point is quite expensive to implement computationally.
2. Not an expert, but I think the floating-point algorithms and proofs for bases other than 2 are less well researched and less well known.
3. A calculator is a resource-constrained device.
4. Are there even any real benefits? Yes, some human-friendly numbers have an exact finite representation that is missing in base-2, but I can't see why would that matter, compared to just using 32-bit or 64-bit IEEE-754.
In practice, the cost is not that much different because four-function calculators do not do many calculations at once. Even in a graphing calculator, drawing a plot is never slow enough to make the device useless.
Decimal, and particularly BCD floating point, is much simpler than binary for a calculator since the numbers don't need any additional conversion steps to display.
And the fact that 0.1 + 0.1 works as expected is indeed a huge benefit.
I expect that's because the cost versus effort for having two kinds of floating point isn't very good. Not because decimal floating point is notably more expensive than binary floating point.
