MJPEG Tools Svn

better quant_weight_coeff_sum_mmx -- explicit parallelism, better abs func

use movd_g2r/movd_r2g whenever possible

Convert predict_mmx from NASM to inline C assembly (from Gert Vervoort)

get rid of SSE quant_non_intra (which was broken, and abusing
program-wide SSE rounding flags which affected the ability to write
other SSE routines). Totally rewrite MMX quant_non_intra so that it
is 1) correct, and 2) faster than the SSE code was!

nothing from fastintfns.h used so why bother including it (fastintfns.h has very
little left in it these days).

Bit of tidying up. Cleanup - remove fastintfns.h since nothing in that was
used so it shouldn't be included.

Utilize movd mmx->gpr so return values don't go through memory

fix to lav_io.c which gets 16bit decoding working for certain, 16 bit encoding
(writing) _should_ work also. 8bit (does anyone still use 8bit sound?) is
probably still non-functional and will be fixed in the days ahead.

intabs() has been replaced by math.h + abs()

Yay! intabs() is gone. If math.h is in scope then gcc will do the right thing
with abs().