based on patch by Kieran Kunhya

This avoids SIMD-optimized functions having to sign-extend their
line size argument manually to be able to do pointer arithmetic.

Also adjust parameter names to be "linesize" everywhere.

Also rename the enum values to be consistent with other DCT permutations.

None of the encoder bits are arch-optimized.

No permutation is necessary for the FDCT.

See: IRC log
Signed-off-by: NMichael Niedermayer <michaelni@gmx.at>

This replaces 64 additions by 8 additions

MSE Changes from 0.04873672 to 0.05012422
Signed-off-by: NMichael Niedermayer <michaelni@gmx.at>

Using the first names of authors sounds somewhat unprofessional
and might be considered offensive which is not intended.
The new names use the initials of the authors due to simplicity
and the possibility to apply it consistently without the need
to find political correct names for each future case where
alternative codecs might exist. Also its shorter ...

If someone has a better idea, like maybe 2 random letters
and people prefer it then iam happy to switch to that ...
Signed-off-by: NMichael Niedermayer <michaelni@gmx.at>

It does not help as an abstraction and adds dsputil dependencies.
Signed-off-by: NRonald S. Bultje <rsbultje@gmail.com>

Rename the called dsp init functions to *_init_x86.

Signed-off-by: NMartin Storsjö <martin@martin.st>

Found, analysed and tested by trac user Jamal.

Fixes part of Ticket #1404.

Found, analysed and tested by trac user Jamal.

Fixes part of Ticket #1404.

Signed-off-by: NMichael Niedermayer <michaelni@gmx.at>

~3.0-3.5x as fast as original C version, 1.6x as fast overall.