This way, the special IDCT permutations are no longer needed. Bfin code
is disabled until someone updates it. This is similar to how H264 does
it, and removes the dsputil dependency imposed by the scantable code.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This makes the vp3 decoder less dependent on dsputil, and will aid
in making it (eventually) dsputil-independent.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This way, they can be shared between mpeg4qpel and h264qpel without
requiring either one to be compiled unconditionally.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Put a copy of the 8bit functions only in dsputil, where they are used
for some other things (e.g. mpeg4qpel, mspel, cavsqpel). We could perhaps
also try to share specifically the 8bit functions from h264qpel between
it and the others, but that will be slightly more complicated. H264qpel
already had these functions, so we can simply remove the duplicates.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
These could be used for reference counting, or for keeping track of
decoding progress in references in multithreaded decoders.
Support is provided by gcc/msvc/suncc intrinsics, with a fallback using
pthread mutexes.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
This makes the decoder independent of mpegvideo.
This copy of the draw_horiz_band code is simplified compared to
the "generic" mpegvideo one which still has a number of special
cases for different codecs.
Signed-off-by: Martin Storsjö <martin@martin.st>
Not all hwaccels implement all codecs, so using one single list for
multiple such codecs means some codecs will be represented in the list,
even though they don't actually handle that codec. Copying specific
lists in each codec fixes that.
Signed-off-by: Martin Storsjö <martin@martin.st>
The code is located in mpegvideo, and it's likely that in a minimal
config, we don't want to include debug info anyway.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Instead, only extend edges on-demand when the motion vector actually
crosses the visible decoded area using ff_emulated_edge_mc(). This
changes decoding time for cathedral from 8.722sec to 8.706sec, i.e.
0.2% faster overall. More generally (VP8 uses this also), low-motion
content gets significant speed improvements, whereas high-motion content
tends to decode in approximately the same time.
Signed-off-by: Martin Storsjö <martin@martin.st>
Instead, keep them in the bitstream buffer until we read them verbatim,
this saves a memcpy() and a subsequent clearing of the target buffer.
decode_cabac+decode_mb for a sample file (CAPM3_Sony_D.jsv) goes from
6121.4 to 6095.5 cycles, i.e. 26 cycles faster.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows more transparent mixing of get_bits and whole-byte access
without having to touch get_bits internals.
Signed-off-by: Martin Storsjö <martin@martin.st>
These functions are mostly H264-specific (the only other user I can
spot is bink), and this allows us to special-case some functionality
for H264. Also remove the 16-bit-coeff with >8bpp versions (unused)
and merge the duplicate 32-bit-coeff for >8bpp (identical).
Signed-off-by: Martin Storsjö <martin@martin.st>
The non-alpha and alpha-Y planes are cleared in the idct_put/add()
calls. For the alpha U/V planes, we only care about the DC for entropy
context prediction purposes, the rest of the data is unused.
Signed-off-by: Martin Storsjö <martin@martin.st>
The non-intra-pcm branch in hl_decode_mb (simple, 8bpp) goes from 700
to 672 cycles, and the complete loop of decode_mb_cabac and hl_decode_mb
(in the decode_slice loop) goes from 1759 to 1733 cycles on the clip
tested (cathedral), i.e. almost 30 cycles per mb faster.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The residual block data of 16x16 blocks was ignored for b-frames, which
leads to easy-to-identify artifacts. After this patch, the artifacts are
gone. Sample video: svq3_watermark.mov. (Fate results unaffected.)
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Not all hwaccels implement all codecs, so using one single list for
multiple such codecs means some codecs will be represented in the list,
even though they don't actually handle that codec. Copying specific
lists in each codec fixes that.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The symbol "ff_h263_loop_filter_strength" is defined in h263.c, but
the h263 loopfilter functions (in the .asm file) are not optimized
out (even though their function pointers are never assigned).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Instead, only extend edges on-demand when the motion vector actually
crosses the visible decoded area using ff_emulated_edge_mc(). This
changes decoding time for cathedral from 8.722sec to 8.706sec, i.e.
0.2% faster overall. More generally (VP8 uses this also), low-motion
content gets significant speed improvements, whereas high-motion content
tends to decode in approximately the same time.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Instead, keep them in the bitstream buffer until we read them verbatim,
this saves a memcpy() and a subsequent clearing of the target buffer.
decode_cabac+decode_mb for a sample file (CAPM3_Sony_D.jsv) goes from
6121.4 to 6095.5 cycles, i.e. 26 cycles faster.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Init code in that if statement goes down from 26716 cycles to 26047
cycles, i.e. the removal of the clear_blocks and smaller memcpy()
together save around 670 cycles.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
These functions are mostly H264-specific (the only other user I can
spot is bink), and this allows us to special-case some functionality
for H264. Also remove the 16-bit-coeff with >8bpp versions (unused)
and merge the duplicate 32-bit-coeff for >8bpp (identical).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Also use the resulting 16bpp functions for anything >8 and <=16, not just
9 and 10. This fixes 12 and 14bpp H264 support.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This allows more transparent mixing of get_bits and whole-byte access
without having to touch get_bits internals.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The non-alpha and alpha-Y planes are cleared in the idct_put/add()
calls. For the alpha U/V planes, we only care about the DC for entropy
context prediction purposes, the rest of the data is unused.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Now, nellymoserenc and aacenc no longer depends on dsputil. Independent
of this patch, wmaprodec also does not depend on dsputil, so I removed
it from there also.
This allows us to get rid of them on the next major bump. All of the
above are functionally irrelevant, and most of them are unused, except
the vp3 one, which is used wrongly in the bfin arch optimizations.
This allows us to remove FF_IDCT_WMV2, which serves no practical purpose
other than to be able to select the WMV2 IDCT for MPEG (or vice versa)
and get corrupt output.
Fate tests for all wmv2-related tests change, because (for some obscure
reason) they forced use of the MPEG IDCT. You would get the same changes
previously by not using -idct simple in the fate test (or replacing it
with -idct auto).
This is identical to what e.g. vp8 does, and prevents the function call
overhead (plus dependency on dsputil for this particular function).
Arm asm updated by Janne Grunau <janne-libav@jannau.net>.
Signed-off-by: Janne Grunau <janne-libav@jannau.net>
If the motion vector is at a subpixel position, we need 3 pixels below
the motion vector's wholepel position available, not 2, since the MC
filter is a sixtap filter for the hpel position, and then a bilin filter
for the qpel position.
This patch fixes highly irreproducible (0.1%) fate failures in frame 2
and 4 of h264-conformance-cama2_vtc_b (e.g. first P-frame, first field,
last line of MB x=40,y=2 and second field and last lines of MBs x=39-40,
y=3). These used pre-loopfilter instead of post-loopfilter data because
the await_progress() waited for one line too little in that field, and
the motion vector of these particular MBs happened to align exactly to a
position where that demonstrates the bug.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
If the motion vector is at a subpixel position, we need 3 pixels below
the motion vector's wholepel position available, not 2, since the MC
filter is a sixtap filter for the hpel position, and then a bilin filter
for the qpel position.
This patch fixes highly irreproducible (0.1%) fate failures in frame 2
and 4 of h264-conformance-cama2_vtc_b (e.g. first P-frame, first field,
last line of MB x=40,y=2 and second field and last lines of MBs x=39-40,
y=3). These used pre-loopfilter instead of post-loopfilter data because
the await_progress() waited for one line too little in that field, and
the motion vector of these particular MBs happened to align exactly to a
position where that demonstrates the bug.
CC: libav-stable@libav.org
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Clobbering these tables will temporarily clobber the template used
as a basis for other threads to start decoding from. If the other
decoding thread updates from the template right at that moment,
subsequent threads will get invalid (or, usually, none at all) mmco
tables. This leads to invalid reference lists and subsequent decode
failures.
Therefore, instead, decode the mmco tables only for the first slice in
a field or frame. For other slices, decode the bits and ensure they
are identical to the mmco tables in the first slice, but don't ever
clobber the context state. This prevents other threads from using a
clobbered/invalid template as starting point for decoding, and thus
fixes decoding in these cases.
This fixes occasional (~1%) failures of h264-conformance-mr1_bt_a with
frame-multithreading enabled.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Clobbering these tables will temporarily clobber the template used
as a basis for other threads to start decoding from. If the other
decoding thread updates from the template right at that moment,
subsequent threads will get invalid (or, usually, none at all) mmco
tables. This leads to invalid reference lists and subsequent decode
failures.
Therefore, instead, decode the mmco tables only for the first slice in
a field or frame. For other slices, decode the bits and ensure they
are identical to the mmco tables in the first slice, but don't ever
clobber the context state. This prevents other threads from using a
clobbered/invalid template as starting point for decoding, and thus
fixes decoding in these cases.
This fixes occasional (~1%) failures of h264-conformance-mr1_bt_a with
frame-multithreading enabled.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Move some functions from dsputil. The idea is that videodsp contains
functions that are useful for a large and varied set of video decoders.
Currently, it contains emulated_edge_mc() and prefetch().
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Use this in VP8/H264-8bit loopfilter functions so they can be used if
there is no aligned stack (e.g. MSVC 32bit or ICC 10.x).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Use this in VP8/H264-8bit loopfilter functions so they can be used if
there is no aligned stack (e.g. MSVC 32bit or ICC 10.x).
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Prevent an invalid write into coeffs[scantable[-1]] if zeros_left
itself was an invalid VLC code (and thus -1).
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Replace INIT_AVX by INIT_XMM avx. Port the whole file to use cpuflag
based function declarations. Remove (now unused) cputype argument in
function declaration macros. Change function prototypes to have mmx2
instead of mmxext as suffix, since that's required by cpuflags.
The variable is copied to subsequent threads at the same time, so this
may cause wrong ref_count[] values to be copied to subsequent threads.
This bug was found using TSAN and Helgrind.
Original patch by Ronald, adapted with a local_ref_count by Clément,
following the suggestion of Michael Niedermayer.
Signed-off-by: Clément Bœsch <clement.boesch@smartjog.com>
The variable is copied to subsequent threads at the same time, so this
may cause wrong ref_count[] values to be copied to subsequent threads.
This bug was found using TSAN.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
It is included for the open/read/write/close functions. On
MSVC, where this header does not exist, the same functions
are provided by io.h, which is already included.
On windows, these functions are provided by io.h. Make sure
io.h is included if it exists, regardless of the setmode
function.
Signed-off-by: Martin Storsjö <martin@martin.st>
Conflicts:
configure
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
unistd.h is used for open/read/close, but if this header does not
exist, there's probably no use in trying to open /dev/*random
at all.
Signed-off-by: Martin Storsjö <martin@martin.st>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
unistd.h used to be required for gethostname. On windows, gethostname
is provided by winsock2.h. Now network.h includes both unistd.h and
winsock2.h if they exist.
Signed-off-by: Martin Storsjö <martin@martin.st>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This introduces support for width%4==2 in addition to width%4==0. For
odd widths, some more checks are needed, since the current code always
handles two luma items in a row, thus there is a possibility of an
overread by one.
For left HFYU prediction, we predict from the buffer buf+1 using 8- or
16-byte reads. This means that aligning the buffer by 16 bytes is in
itself not sufficient, because if the width itself is 16- or 8-byte
aligned, the buffer will not be padded, and thus a read of size 16 at
buf+1 will overflow boundaries at the right edge. Padding the buffer by
1 byte is sufficient to not overflow its boundaries.
Fixes bug 342.
This makes add_hfyu_left_prediction_sse4() handle sources that are not
16-byte aligned in its own function rather than by proxying the call to
add_hfyu_left_prediction_ssse3(). This fixes a crash on Win64, since the
sse4 version clobberes xmm6, but the ssse3 version (which uses MMX regs)
does not restore it, thus leading to XMM clobbering and RSP being off.
Fixes bug 342.
Some calculations were changed in b6a3849 to use mmsize, which was not correct
for the AVX version, which uses INIT_YMM and therefore has mmsize == 32.
Fixes Bug 341.
Signed-off-by: Justin Ruggles <justin.ruggles@gmail.com>
