Even when the segment feature data is in absolute mode, it is still read as a
6-bit value with an added sign, so it could have values between -63 and +63.
Later, this signed value is used without checks as a filter level, which is
used to access an entry in an array of size MAX_LOOP_FILTER+1=64.
This patch just extends the existing clamping (that was done only to relative-
mode data) to absolute mode data, before it is blindly 'memset' in
lfi->lvl[seg][0], which was where the out-of-bound filter_value was read in
subsequent vp8_loop_filter_row_simple.
To allow GMPs time to update to new GMPDecryptor versions, we support the
latest GMPDecryptor version, and the previous.
In order to support a consistent interface to Gecko, we adapt the previous
GMPDecryptor version to the latest in the GMP child process. So Gecko always
thinks it's talking to the latest version.
We also make gmp-fake deliberately support the previous GMPDecryptor version,
to ensure this code path remains tested.
Made some class interfaces a bit more "C++11"-ish, to protect against some
possible issues.
Also removed 'trex', which was only used by code removed in previous patch.
We're failing in the "Very rough kill-switch" case in
GMPVideoDecoderParent::Decode() we find that too many shmems are in use when we
come to send a "Decode" message to the GMP, and that causes an error which
percolates up to cause the test failure.
This patch changes gmp-clearkey to copy the input encrypted and compressed
sample and immediately return the shmem to the parent process. We are
copying the data anyway when we decrypt, so we can rejigg things so that we
don't actually end up doing a second copy.
We now use an index of samples made of block of samples delimited by keyframes. The search is performed using a binary search. This allows to quickly find which blocks are found within the media cache.
On a 8 core mac pro, this leads to a 67% improvement on CPU time spent playing a long MP4 video (from 112s CPU to 45s CPU)
The optimisation is only possible if all mp4 data chunks are continuous (which they almost always are)
Ideally, we should properly make and shell quote everything we print out
in makefiles, but that's a can of worms I don't want to open just yet. So
I'll limit myself to just passthru variables.
We're going to change how e.g. CFLAGS are printed out in backend.mk, and
to fit that model, the data in the corresponding moz.build variables
need to be straightened up.
