Package org.apache.sysds.runtime.compress.colgroup.dictionary

Class DictLibMatrixMult

java.lang.Object
org.apache.sysds.runtime.compress.colgroup.dictionary.DictLibMatrixMult
public class DictLibMatrixMult extends Object
Utility interface for dictionary matrix multiplication

  • Method Details

    • addToUpperTriangle

      public static void addToUpperTriangle(int nCols, int row, int col, double[] res, double val)
      Add to the upper triangle, but twice if on the diagonal
      Parameters:
      nCols - number cols in res
      row - the row to add to
      col - the col to add to
      res - the double array to add to
      val - the value to add

    • MMDictsWithScaling

      public static void MMDictsWithScaling(IDictionary left, IDictionary right, IColIndex leftRows, IColIndex rightColumns, MatrixBlock result, int[] counts)
      Matrix multiply with scaling (left side transposed)
      Parameters:
      left - Left side dictionary that is not physically transposed but should be treated if it is.
      right - Right side dictionary that is not transposed and should be used as is.
      leftRows - Left side row offsets
      rightColumns - Right side column offsets
      result - The result matrix, normal allocation.
      counts - The scaling factors

    • TSMMDictionaryWithScaling

      public static void TSMMDictionaryWithScaling(IDictionary dict, int[] counts, IColIndex rows, IColIndex cols, MatrixBlock ret)
      Perform the full tsmm with the dictionary (allocating into the entire output matrix.)
      Parameters:
      dict - The dictionary to tsmm
      counts - The frequency of each dictionary entry
      rows - The rows of the dictionary
      cols - The cols of the dictionary
      ret - The output to add the results to

    • MMDicts

      public static void MMDicts(IDictionary left, IDictionary right, IColIndex rowsLeft, IColIndex colsRight, MatrixBlock result)
      Matrix Multiply the two dictionaries, note that the left side is considered transposed but not allocated transposed making the multiplication a: t(left) %*% right
      Parameters:
      left - The left side dictionary
      right - The right side dictionary
      rowsLeft - The row indexes on the left hand side
      colsRight - The column indexes on the right hand side
      result - The result matrix to put the results into.

    • TSMMToUpperTriangle

      public static void TSMMToUpperTriangle(IDictionary left, IDictionary right, IColIndex rowsLeft, IColIndex colsRight, MatrixBlock result)
      Does two matrix multiplications in one go but only add to the upper triangle. the two multiplications are: t(left) %*% right t(right) %*% left In practice this operation then only does one of these multiplications but all results that would end up in the lower triangle is transposed and added to the upper triangle. Furthermore all results that would end up on the diagonal is added twice to adhere with the two multiplications
      Parameters:
      left - Left dictionary to multiply
      right - Right dictionary to multiply
      rowsLeft - rows for the left dictionary
      colsRight - cols for the right dictionary
      result - the result

    • TSMMToUpperTriangleScaling

      public static void TSMMToUpperTriangleScaling(IDictionary left, IDictionary right, IColIndex rowsLeft, IColIndex colsRight, int[] scale, MatrixBlock result)
      Does two matrix multiplications in one go but only add to the upper triangle with scaling. the two multiplications are: t(left) %*% right t(right) %*% left In practice this operation then only does one of these multiplications but all results that would end up in the lower triangle is transposed and added to the upper triangle. Furthermore all results that would end up on the diagonal is added twice to adhere with the two multiplications
      Parameters:
      left - Left dictionary to multiply
      right - Right dictionary to multiply
      rowsLeft - Rows for the left dictionary
      colsRight - Cols for the right dictionary
      scale - A multiplier to each dictionary entry
      result - The result