The encoder uses almost none of the mpegvideo infrastructure, only some
fields from MpegEncContext.
The FATE results change because now an all-zero quant matrix is written
into the file. Since it is not used for anything for ljpeg, this should
not be a problem.
This code was originally added in
5f1948111a4920ff74e31b4284ab0200f7938539 to h263 to set decoded frame
pts to some random numbers (removed in
a1c5cc429d99216406170eac7e8352860076d3e8) and then cargo culted to other
decoders.
The code is left in h263dec for now, since some part of the decoder
(apparently OBMC) relies on the specific previous frame to be reused.
works around bug in gccs inline asm register assignment
Fixes Ticket3177
gcc from 4.4 to 4.6 is affected at least, no non affected gccs known
clang seems not affected
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The mmxext optimizations should be at least equally fast if available and
amd3dnow optimizations are being deprecated. Thus the former should
override the latter, not the other way around.
The CONFIG_ name-space is set by configure, so its better to use a
different prefix here.
This also unifies the encoder & decoder define that is used
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This makes the USE_FLOATS == 0 available to the end user
More float optimizations can easily be added as well now
common code should be factored out into a common file once all
fixed point & floating point optimizations are done, this is to
avoid having to move code back and forth between files.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
* commit '21710ea82118c9d19bea9277b2a85a33096fdd95':
dv: Move DV VLC data tables out of header file
Conflicts:
libavcodec/dv_vlc_data.h
Merged-by: Michael Niedermayer <michaelni@gmx.at>
* qatar/master:
dsputil: x86: Move ff_inv_zigzag_direct16 table init to mpegvideo
If someone optimizes dct_quantize for non x86 SIMD, then this
probably needs to be reverted.
Merged-by: Michael Niedermayer <michaelni@gmx.at>
It was previously int and would return error if decode_coeffs_b()
returns an error; however, that can never happen, so refactor all
that code to make all dependent functions return void also (all the
way up to decode_coeffs_sb()).
For a random 1080p sample, decoding time went from 9.7sec (1 threads)
to 6.0sec (2 threads) and 5.2sec (4 threads) in 2-pass decoding mode.
I don't have any samples that use the parallelmode feature, but the
gains should be higher.
We need more information from last/cur_frame than from reference
buffers, so we can use a simplified structure for reference buffers,
and then store mvs and segmentation map information in last/cur.
