Issue: On extremely new hardware using either IceLake or super sets of
Intel IceLakes avx512 instructions, commit
d4cd8830bd causes build issues.
Specifically a NASM macro expansion of qpel_filter_v is never properly
defined/initialized.
The issue is the definition was erroneously placed inside a conditional
which will not trigger unless the original definition failed (has to do
with if PIC is defined, becomes a bit of a catch 22)
Specifically the error is
X86ASM libavcodec/x86/hevc_mc.o
libavcodec/x86/hevc_mc.asm:1854: error: symbol `..@88472.table' not defined
libavcodec/x86/hevc_mc.asm:1806: ... from macro
`HEVC_PUT_HEVC_QPEL_HV_AVX512ICL' defined here
libavcodec/x86/hevc_mc.asm:1730: ... from macro `QPEL_FILTER_V' defined here
...
repeats a few times...
...
make: *** [ffbuild/common.mak💯 libavcodec/x86/hevc_mc.o] Error 1
```
Specific error was discussed by kurosu and myself (fclc) on the
ffmpeg-devel irc.
This commit fixes the above by swapping lines 1796 and 1795, moving the
define out of the conditional
Side note: It seems fate didn't pick up on this, may merit looking into
(as mentioned by nevcairiel).
Reviewed-by: Wu Jianhua <toqsxw@outlook.com>
Signed-off-by: Felix LeClair (FCLC) <felix.leclair123@hotmail.com>
Signed-off-by: Marton Balint <cus@passwd.hu>
This decoder uses ff_get_buffer() and does nothing weird
(it does not even rely on any alignment of the frame's data/linesize).
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Both AV_PIX_FMT_GRAY8 and AV_PIX_FMT_GRAY16 use unsigned values,
not signed ones. The fact that the input might be signed
in some cases in the original format doesn't change this.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The PGX decoder accesses the lines via code like
(PIXEL*)frame->data[0] + i*frame->linesize[0]/sizeof(PIXEL)
where PIXEL is a macro parameter. This code has issues with negative
linesizes, because the type of sizeof(PIXEL) is size_t, so
that on common systems i*linesize/sizeof(PIXEL) will
always be an unsigned type that is very large in case linesize is
negative. This happens to work*, but it is undefined behaviour
and e.g. leads to "src/libavcodec/pgxdec.c:114:1: runtime error:
addition of unsigned offset to 0x7efe9c2b7040 overflowed to 0x7efe9c2b6040"
errors from UBSAN.
Fix this by using (PIXEL*)(frame->data[0] + i*frame->linesize[0]).
This is allowed because linesize has to be suitably aligned.
*: Converting a negative int to size_t works by adding SIZE_MAX + 1
to the number, so that the result is off by (SIZE_MAX + 1) /
sizeof(PIXEL). Converting the pointer arithmetic (performed on PIXELs)
back to ordinary pointers is tantamount to multiplying by sizeof(PIXEL),
so that the result is off by SIZE_MAX + 1; but SIZE_MAX + 1 == 0
for the underlying pointers.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
These checks were (most likely) added to check for overreads
as the bytestream2_get_* functions return 0 in this case.
Yet this is not necessary anymore as we now have an explicit check
for the size. Should the input contain a real \0, pgx_get_number()
will error out lateron.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
We have already checked that there is data to be read.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Each of the three calls to pgx_get_number() consumes at least two bytes.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This patch fixes a wrong type of BTI landing pad when branching to
functions instantiated via the fft*_neon macro.
Although the previously employed paciasp instruction serves as a landing
pad, for the ways that this function is invoked it is the wrong type, resulting
in an unexpected termination of the running process.
Signed-off-by: André Kempe <andre.kempe@arm.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
match the behavior of SvtAv1EncApp to ensure pic_type is always set
before passing it to the library.
The other options for pic_type aren't currently used inside the library,
so they aren't introduced in this patch.
Signed-off-by: Christopher Degawa <ccom@randomderp.com>
Signed-off-by: James Almer <jamrial@gmail.com>
Monterey needs mBytesPerFrame and mBytesPerPacket to be set, and I'm
surprised this didn't break any previous system versions.
Fixes bug #9564: Cannot decode xHE-AAC with audiotoolbox (aac_at) on
Mac OS Monterey. Fixes likely bug that none of the AudioToolbox
decoders work on Monterey.
Signed-off-by: Christopher Snowhill <kode54@gmail.com>
The size of the ICC chunk has already been accounted for when
the packet's buffer was initially set in ff_mpv_encode_picture()
and the header (including the ICC chunk) has already been written
at this point.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It is a remnant of the old way for user-supplied buffers;
it is always-true since 93016f5d1d.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
If one encodes MJPEG with a single slice and uses input with
AV_FRAME_DATA_ICC_PROFILE side data, the current allocation code
in ff_mpv_encode_picture() will always increase the size of the
temporary buffer used for allocating packets by the size needed
for to write the ICC chunk even when the current buffer is actually
large enough. This commit fixes this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
MJPEG is the only mpegvideo-based encoder making use of it.
Fixes linking failures in case mpegvideo_enc.c is compiled
with AMV, LJPEG and MJPEG encoders disabled.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It works, but it is not documented to work.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Do this by switching to bytestream2_(get|put)_le32u() from
bytestream2_(get|put)_le32(); it has after all already been checked
that the packet contains at least a full header, making all
the implicit checks in bytestream2_(get|put)_le32() dead code.
Reviewed-by: Marton Balint <cus@passwd.hu>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Therefore move the (Get|Put)ByteContext from the context to the stack.
It is transient anyway.
Reviewed-by: Marton Balint <cus@passwd.hu>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Reviewed-by: Marton Balint <cus@passwd.hu>
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
SEI messages are naturally byte-aligned by adding padding bits
to achieve byte-alignment. The parsing code in libavcodec/hevc_sei.c
nevertheless uses a GetBitContext to read it. When doing so, parsing
the next SEI message starts exactly at the position where reading
the last message (if any) ended.
This means that one would have to handle both the payload extension data
(which makes most SEI messages extensible structs) as well as the
padding bits for byte-alignment. Yet our SEI parsing code in
libavcodec/hevc_sei.c does not read these at all. Instead several of
the functions used for parsing specific SEI messages use
skip_bits_long(); some don't use it at all, in which case it is possible
for the GetBitContext to not be byte-aligned at the start of the next
SEI message (the parsing code for several types of SEI messages relies
on byte-alignment).
Fix this by always using a dedicated GetBitContext per SEI message;
skipping the necessary amount of bytes in the NALU context
is done at a higher level. This also allows to remove unnecessary
parsing code that only existed in order to skip enough bytes.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Felix LeClair
Martijn van Beurden
Andreas Rheinhardt
Andre Kempe
Jan Ekström
Christopher Degawa
Christopher Snowhill
Paul B Mahol
Wu Jianhua
Leo Izen
Anton Khirnov