Add small precision fast-path to Quo about apd HOT 2 CLOSED

So looking at the Quo code, there are a lot of for loops. It looks like it's trying to do things digit by digit. Is this algorithm designed around cases with very large precision? What if there were a fast path for precision <= 34 (decimal128, etc.)?

I benchmarked some decimal128 code based on https://software.intel.com/content/www/us/en/develop/articles/intel-decimal-floating-point-math-library.html by the way and division is comparable to multiplication there. So it must be possible for apd to have a more efficient Quo implementation for "low" precision.

from apd.

Hi @aaronc, thanks for the report. Quo performance has improved recently (it's about 50% faster on your benchmark), but it still lags the performance of other operations by quite a margin. As you identified, this has to do with the digit-by-digit loop in Quo. The algorithm is designed around cases with very large precision and does not include any form of fast-path for low precision values. I'd like to keep this issue open to track the introduction of such a fast path.

from apd.