• CanadaPlus@lemmy.sdf.org
    link
    fedilink
    arrow-up
    6
    ·
    edit-2
    3 months ago

    Umm, AKCTSHUALLY it gets more like O(n2) in parallel, assuming you’re using a physically achievable memory. There’s just a lot of criss-crossing the entries have to do.

    Strassen’s algorithm gets O(n2.8) in serial by being terrible, and the weird experimental ones get O(n2.3), but the asymptotic benefits of Coppersmith-Winograd and friends only kick in if you’re a Kardashev III civilisation computing a single big product on a Dyson sphere.

    • Mikina@programming.dev
      link
      fedilink
      arrow-up
      3
      ·
      3 months ago

      I can’t decide whether this sentence is a joke or not. It has the same tone that triggers my PTSD from my CS degree classes and I also do recognize some of the terms, but it also sounds like it’s just throwing random science terms around as if you asked a LLM to talk about math.

      I love it.

      Also, it’s apparently also real and correct.

    • lad@programming.dev
      link
      fedilink
      English
      arrow-up
      2
      ·
      3 months ago

      Yeah, in fact, I somehow calculated in assumption of n being the amount of elements in matrix, not (assuming square matrix)

      But I am impressed to know that there are serial algorithms that approach O(n²), thank you for sharing that info

      • CanadaPlus@lemmy.sdf.org
        link
        fedilink
        arrow-up
        3
        ·
        edit-2
        3 months ago

        Yeah, they work by turning the problem into some crazy kind of group theory and attacking it that way. Every once in a while someone shaves the decimal down slightly, just by implementing the deep math in a more efficient way. A new approach will be needed if it is in fact possible to get down to O(n2), though. Strassen’s is a divide and conquer algorithm, and each step of the iteration looks like this:

            S[1] = B[1, 2] - B[2, 2]
            S[2] = A[1, 1] + A[1, 2]
            S[3] = A[2, 1] + A[2, 2]
            S[4] = B[2, 1] - B[1, 1]
            S[5] = A[1, 1] + A[2, 2]
            S[6] = B[1, 1] + B[2, 2]
            S[7] = A[1, 2] - A[2, 2]
            S[8] = B[2, 1] + B[2, 2]
            S[9] = A[1, 1] - A[2, 1]
            S[10] = B[1, 1] + B[1, 2]
            P[1] = STRASSEN(A[1, 1], S[1])
            P[2] = STRASSEN(S[2], B[2, 2])
            P[3] = STRASSEN(S[3], B[1, 1])
            P[4] = STRASSEN(A[2, 2], S[4])
            P[5] = STRASSEN(S[5], S[6])
            P[6] = STRASSEN(S[7], S[8])
            P[7] = STRASSEN(S[9], S[10])
            C[1..n / 2][1..n / 2] = P[5] + P[4] - P[2] + P[6]
            C[1..n / 2][n / 2 + 1..n] = P[1] + P[2]
            C[n / 2 + 1..n][1..n / 2] = P[3] + P[4]
            C[n / 2 + 1..n][n / 2 + 1..n] = P[5] + P[1] - P[3] - P[7]
            return C
        

        In my copy of Introduction to Algorithms, it says something like “this is the most bullshit algorithm in the book and it’s not close” underneath. You can make it a bit neater by representing the multiplication operation as a 3-dimensional tensor, but at the end of the day it’s still just a stupid arithmetic trick. One that’s built into your GPU.