I completely agree with your reasoning about determinants. And yes, we might consider trimming the input as part of comparisons, so not simply invoke mpn_cmp. (I think we should probably only compare a with c and b with d, i.e., not speculate in that a and b or c and d might be close.)
But we should do this like this IMO: 1. check signs and conditionally exit 2. count bits (split into grude limb count and finer bit count as now) and conditionally exit 3. compare + chop, exit iff denominators are equal 4. make a single-limb cross product, exit if difference is large enough to make the call here 5. perhaps perform some determinish computations (but I'm unsure about the utility of this); this might terminate in rare cases but will else reduce operands for the next step 5. perform either a gradually greater cross product or go for the full cross product right away We might swap 3 and 4, since 3 is not O(1). And "chop" should not be taken literally; one needs to do the proper operand updates which means subtraction of an operand pair. -- Torbjörn Please encrypt, key id 0xC8601622 _______________________________________________ gmp-devel mailing list gmp-devel@gmplib.org https://gmplib.org/mailman/listinfo/gmp-devel
