Author solution is not performant, it runs more than 6 seconds on a 300 x 300 matrix unlike the two existing solutions (which can handle a 2000 x 2000 matrix in under 6 seconds!).

So the "large array" test is not actually a performance test.

It also means that the test themselves are inherently slow (because the author solution is used as the reference solution).

Please do not round floating point numbers, and use assertApproxEquals. You'll have to map this over the array, of course. Rounding is not fullproof against errors.