GitLab

slices. With this new code libmpeg2's behaviour should be equivalent
to that of decoders that have the notion of macroblock addresses. The
speed impact should not be measurable as this is outside of the
macroblock loop.

Also checking that the macroblock position is within the bounds of the
image probably helps with robustness.

slice structure is NOT legal. So this test stream is bogus !

behaviour in mpeg1.

putenv is posix, setenv is a bsdism

http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/mailman/CVSROOT/syncmail

dont overflow as much as the reference decoder does :)

updated test suite accordingly: -c is always used unless we explicitly
specify an idct other than c. That way the regression test should pass
on all architectures.

fix for --enable-shared

also should fix some problems people have seen with the cpu detection
routine on some OS/processor combinations (I'm not totally clear about
these)

fixed support for mpeg-1 macroblock stuffing.

this patch from bradley walken reorders two c instructions in the
startcode search and gets us some small speed gain. I dont understand
why gcc does not find the proper scheduling by itself here...

more MC tweaking

First part of MC code rework

At some point we need to have the coordinates (in half-pixels)
of the MC source data, so that we can check that we do not copy
from outside of the source picture boundaries. In the previous code,
these coordinates were never explicitly exposed, which made the
checking difficult.

The new MC code still needs a lot of work, but this is a first step.

updated definition for movd - it is illegal between two mmx registers,
but it is legal between one mmx register and one register/memory operand.

fix for overflow while writing destination macroblocks

now all remaining segfaults should be when doing memory reads