Re: [XviD-devel] other FDCT with fewer multiplications

[skal <skal@xxxxxxxxxxxx>]

        Hi,

        about the text in 2dDCT.zip: it's rather cryptic, without
        the original paper. So far, it seems that 1 mult (the one
        by r(4) = sqrt(2)) has disappeared. But 1 mult less is
        not everything. More crucial is the data flow schematic,
        as soon as precision is concerned. So i should be interesting
        to have the original paper, sure.

        bye,

                Skal

(for the story: Thomas Lane, author of the first jpeg lib, *added*
some multiplications in the DCT, in order to better "spread" the
rounding errors around various data paths)