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.
The scaling process for obtaining direct MVs from co-located field MVs
is the same for interlaced field and progressive pictures.
Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>
In VC-1 interlaced field pictures, chroma motion vectors can extend beyond
picture boundary even if luma vectors are bounded. The problem shows up
only for hpel interpolated MVs, and may be due to the way motion vectors
are scaled / cropped.
Thanks to Konstantin Shishkov for suggesting the fix. This fixes
long-known segfaults in MC-VC1.ts from videolan streams archive.
Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>
Currently there is a wild mix of 3dn2/3dnow2/3dnowext. Switching to
"3dnowext", which is a more common name of the CPU flag, as reported
e.g. by the Linux kernel, unifies this.
Fixed codebook mode in 5300 rate may write up to SUBFRAME_LEN + 4 and
that is considered normal by the reference decoder. Without that additional
padding it might overwrite first elements of LPC history.
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>
This allows building dct-test even if aandcttab.o is not pulled in
by any enabled codec. The DCT with which these tables are used does
not use them directly, so building it without the tables is possible.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Reordering the members in this struct reduces the holes required
to maintain alignment. With this order, the only remaining, and
unavoidable, hole is 3 bytes following left_nnz.
Signed-off-by: Mans Rullgard <mans@mansr.com>
These functions are not faster than other mmx implementations on
any hardware I have been able to test on, and they are horribly
inaccurate. There is thus no reason to ever use them.
Signed-off-by: Mans Rullgard <mans@mansr.com>
The standard syntax requires two destination registers for
LDRD/STRD instructions. Some versions of the GNU assembler
allow using only one with the second implicit, others are
more strict.
Signed-off-by: Mans Rullgard <mans@mansr.com>
This completes the conversion of h264dsp to yasm; note that h264 also
uses some dsputil functions, most notably qpel. Performance-wise, the
yasm-version is ~10 cycles faster (182->172) on x86-64, and ~8 cycles
faster (201->193) on x86-32.
These decoders use a special non-MPEG2 IDCT. Call it directly
instead of going through dsputil. There is never any reason
to use a regular IDCT with these decoders or to use the EA IDCT
with other codecs.
This also fixes the bizarre situation of eamad and eatqi decoding
incorrectly if eatgq is disabled.
Signed-off-by: Mans Rullgard <mans@mansr.com>
There is no sense in pulling in this monster struct just for
a handful of fields. The code does not call any functions
expecting an MpegEncContext.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Without this, cglobal will expand "z" to "zh" to access the high byte
in a register's word, which causes a name collision with the ZH(x) macro
further up in this file.
This fixes out of array writes
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>
Mixing yasm and inline asm is a bad idea, since if either yasm or inline
asm is not supported by your toolchain, all of the asm stops working.
Thus, better to use either one or the other alone.
Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
In ff_put_pixels_clamped_mmx(), there are two assembly code blocks.
In the first block (in the unrolled loop), the instructions
"movq 8%3, %%mm1 \n\t", and so forth, have problems.
From above instruction, it is clear what the programmer wants: a load from
p + 8. But this assembly code doesn’t guarantee that. It only works if the
compiler puts p in a register to produce an instruction like this:
"movq 8(%edi), %mm1". During compiler optimization, it is possible that the
compiler will be able to constant propagate into p. Suppose p = &x[10000].
Then operand 3 can become 10000(%edi), where %edi holds &x. And the instruction
becomes "movq 810000(%edx)". That is, it will stride by 810000 instead of 8.
This will cause a segmentation fault.
This error was fixed in the second block of the assembly code, but not in
the unrolled loop.
How to reproduce:
This error is exposed when we build using Intel C++ Compiler, with
IPO+PGO optimization enabled. Crashed when decoding an MJPEG video.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
Set picture type before calling get_buffer.
This allows the DR application to make better decisions.
It also fixes a resource leak in case of missing reference frames
since it would call get_buffer but never release_buffer.
Also use FFSWAP to ensure that the AVFrame is properly initialized
in the next get_buffer (in particular that data[0] is NULL).
Signed-off-by: Reimar Döffinger <Reimar.Doeffinger@gmx.de>
Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
Write out the NAL decoding loops in full so that they are easier
to parse for a preprocessor without it having to be aware of macros
or other such things in C code.
This also makes the code more readable.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
This moves all VP3-specific function pointers from dsputil to a
new vp3dsp context. There is no reason to ever use the VP3 IDCT
where an MPEG2 IDCT is expected or vice versa.
Signed-off-by: Mans Rullgard <mans@mansr.com>
It turns out that the reference decoder subtracts 128 from DC during block
decode but adds it back during reordering block with zigzag pattern.
Transforming block with incorrect DC caused heavy visual artifacts for
many quantisers.
Passing a cutoff value < sample_rate/256 will cause a crash.
Also, values >20000 will have no effect and 20000 will be used anyway.
Signed-off-by: Mohammad Alsaleh <msal@tormail.org>
Signed-off-by: Martin Storsjö <martin@martin.st>
This was unnoticed on linux, since stdlib.h apparently includes
files declaring the pthread_mutex_t and pthread_cond_t types.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
Testing gives 25-30% gain on HD clips with two threads and
up to 50% gain with eight threads.
Sliced threading uses more memory than single or frame threading.
Frame threading and single threading keep the previous memory
layout.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
The number of pixel formats outgrew the number of available bits in
the bitmask used in avcodec_find_best_pix_fmt().
avcodec_find_best_pix_fmt2() uses a PIX_FMT_NONE terminated list
of pixel formats instead.
This allows compiling and running these tests on systems lacking a built-
in version of getopt(), such as MSVC.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
The SPLATB_REG macro already adds the 'd' suffix internally.
This fixes building on Win64, which has been broken since 878e66902.
This worked for unix, where r2 happened to be rdx in this case, which
with the first suffix rdxd was mapped to eax, and eaxd is defined back
to eax. On win64 however, r2 happened to be R8 in this case, and
R8d mapps to R8D just fine, but there's no mapping for R8Dd to anything.
Signed-off-by: Martin Storsjö <martin@martin.st>
Instead of inlining everything into ff_h264_hl_decode_mb(), use
explicit templating to create versions of the called functions
with constant parameters filled in. This greatly speeds up
compilation of h264.c and reduces the code size without any
measurable impact on performance.
Compilation time for h264.c on an i7 goes from 30s to 5.5s.
Code size is reduced by 430kB.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Using ff_mspel_motion assumes that s (a MpegEncContext
poiinter) really is a Wmv2Context.
This fixes crashes in error resilience on vc1/wmv3 videos.
CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>
The buffers are only allocated once, although it can happen from
any of a few different places, so there is no need to use realloc.
Using av_malloc() ensures they are aligned suitably for SIMD
optimisations.
Signed-off-by: Mans Rullgard <mans@mansr.com>
ff_wma_init is used only by wmadec and wmaenc, and neither of them
can handle more than 2 channels.
This fixes crashes with invalid files.
Based on patch by Piotr Bandurski and Michael Niedermayer.
Signed-off-by: Martin Storsjö <martin@martin.st>
This creates proper position independent code when accessing
data symbols if CONFIG_PIC is set.
References to external symbols should now use the movrelx macro.
Some additional code changes are required since this macro may
need a register to hold the GOT pointer.
Signed-off-by: Mans Rullgard <mans@mansr.com>
The problem is that the ssse3 psign instruction does the wrong
thing here. Commit ea60dfe incorrectly removed a macro emulating
this instruction for pre-ssse3 code. However, the emulation is
incorrect, and the code relies on the behaviour of the macro.
Specifically, the psign sets destination elements to zero where
the corresponding source element is zero, whereas the emulation
only negates destination elements where the source is negative.
Furthermore, the PSIGNW_MMX macro in x86util.asm is totally bogus,
which is why the original VC-1 code had an additional right shift
when using it. Since the psign instruction cannot be used here,
skip all the macro hell and use the working instruction sequence
directly.
None of this was noticed due a stray return statement in
ff_vc1dsp_init_mmx() which meant that only the mmx version of the
loop filter was ever used (before being removed in ea60dfe).
Signed-off-by: Mans Rullgard <mans@mansr.com>
The function call was a mess to handle, and memcpy cannot make
the assumptions we do in the new code.
Tested on an IMC sample: 430c -> 370c.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Apparently, some build environments require dxva.h even for dxva2,
while others lack this header entirely. Including it conditionally
allows building in both cases.
Signed-off-by: Martin Storsjö <martin@martin.st>
