[x264-devel] PowerPC/VSX update

[Luca Barbato]

Work from me and Alexandra

[PATCH 1/6] ppc: Manually unroll the horizontal prediction loop

This is a low hanging fruit, there are other 4 functions slower than
C as they are currently written.

[PATCH 2/6] vsx: configure support

Unchanged.

[PATCH 3/6] vsx: Prepare to have different files for VSX and old

Dropped the ppcle broken special case, set the make variable to `no` after
the architecture checks if it is still auto.

[PATCH 4/6] ppc: Provide fallbacks for older architectures

As discussed on irc, vec_vsx_ld/vec_vsx_st are the new unaligned load/store
courtesy of VSX, Alexandra leverages them to avoid annoying slowdowns on
the current hardware in little endian mode.

[PATCH 5/6] ppc: Use vec_vsx_ld instead of VEC_LOAD/STORE macros

The initial patch now w/out the file duplication

[PATCH 6/6] ppc: Fix hadamard for little endian

Now checkasm passes fully =)

lu