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>
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>
Most of the changes are just trivial are just trivial replacements of
fields from MpegEncContext with equivalent fields in H264Context.
Everything in h264* other than h264.c are those trivial changes.
The nontrivial parts are:
1) extracting a simplified version of the frame management code from
mpegvideo.c. We don't need last/next_picture anymore, since h264 uses
its own more complex system already and those were set only to appease
the mpegvideo parts.
2) some tables that need to be allocated/freed in appropriate places.
3) hwaccels -- mostly trivial replacements.
for dxva, the draw_horiz_band() call is moved from
ff_dxva2_common_end_frame() to per-codec end_frame() callbacks,
because it's now different for h264 and MpegEncContext-based
decoders.
4) svq3 -- it does not use h264 complex reference system, so I just
added some very simplistic frame management instead and dropped the
use of ff_h264_frame_start(). Because of this I also had to move some
initialization code to svq3.
Additional fixes for chroma format and bit depth changes by
Janne Grunau <janne-libav@jannau.net>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
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>
This removes a dependency on implementation details from generic
code and allows easy addition of the equivalent optimisation for
other architectures than x86.
Signed-off-by: Mans Rullgard <mans@mansr.com>
This adds a hand-optimized assembly version for get_cabac much like the
existing one, but it works if the table offsets are RIP-relative.
Compared to the non-RIP-relative version this adds 2 lea instructions
and it needs one extra register.
There is a surprisingly large performance improvement over the c version (more
so than the generated assembly seems to suggest) just in get_cabac, I measured
roughly 40% faster for get_cabac on a K8. However, overall the difference is
not that big, I measured roughly 5% on a test clip on a K8 and a Core2.
Hopefully it still compiles on x86 32bit...
Now that only one table is used, there's some chance even darwin as compiles
this (apparently the label arithmetic used previously doesn't work if it
involves symbols defined in a different file, thanks to Ronald S. Bultje for
helping me with this).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The reason is this is easier for PIC code (in particular on darwin...).
Keep the old names as pointers (static in cabac_functions.h so gcc
knows these are just immediate offsets) so the c code can nicely stay the same
(alternatively could use offsets directly in the functions needing the
tables). This should produce the same code as before with non-pic and better
code (confirmed) with pic.
The assembly uses the new table but still won't work for PIC case.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This adds a hand-optimized assembly version for get_cabac much like the
existing one, but it works if the table offsets are RIP-relative.
Compared to the non-RIP-relative version this adds 2 lea instructions
and it needs one extra register. get_cabac() gets about 40% faster, for
an overall speedup of about 5%.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
The reason is this is easier for PIC code (in particular on darwin...).
Keep the old names as pointers (static in cabac_functions.h so gcc
knows these are just immediate offsets) so the c code can nicely stay the same
(alternatively could use offsets directly in the functions needing the
tables). This should produce the same code as before with non-pic and better
code (confirmed) with pic.
The assembly uses the new table but still won't work for PIC case.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
This adds a hand-optimized assembly version for get_cabac much like the
existing one, but it works if the table offsets are RIP-relative.
Compared to the non-RIP-relative version this adds 2 lea instructions
and it needs one extra register.
There is a surprisingly large performance improvement over the c version (more
so than the generated assembly seems to suggest) just in get_cabac, I measured
roughly 40% faster for get_cabac on a K8. However, overall the difference is
not that big, I measured roughly 5% on a test clip on a K8 and a Core2.
Hopefully it still compiles on x86 32bit...
v2: incorporated feedback from Loren Merritt to avoid rip-relative movs
for every table, and got rid of unnecessary @GOTPCREL.
v3: apply similar fixes to the the decode_significance functions, and use
same macro arguments for non-pic case.
v4: prettify inline asm arguments, add a non-fast-cmov version (as I expect
the c code to be faster otherwise since both cmov and sbb suck hard on a
Prescott, even can't construct the mask with a 64bit shift as that's just as
terrible - it's quite difficult to find usable instructions on that chip...).
This is tested to work but not on a P4, in theory it _should_ be fast there.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Conversion of the luma intra prediction mode to one of the constrained
("alzheimer") ones can happen by crafting special bitstreams, causing
a crash because we'll call a NULL function pointer for 16x16 block intra
prediction, since constrained intra prediction functions are only
implemented for chroma (8x8 blocks).
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
This fixes standalone compilation of some decoders with --disable-optimizations.
cabac.h defines some inline functions that use symbols from cabac.c. Without
optimizations these inline functions are not eliminated and linking fails with
references to non-existing symbols.
Splitting the inline functions off into their own header and only #including
it in the places where the inline functions are used allows #including cabac.h
from anywhere without ill effects.
Originally, prior to 8742a4ff8, the caller code was compiled
within this condition:
ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS)
Since HAVE_7REGS is defined as
(ARCH_X86_64 || (HAVE_EBX_AVAILABLE && HAVE_EBP_AVAILABLE))
the subcondition HAVE_7REGS && HAVE_EBX_AVAILABLE is equal
to HAVE_7REGS (for 32 bit at least). The correct simplification
of the original condition thus is HAVE_7REGS, not
HAVE_EBX_AVAILABLE.
This fixes compilation in some cases where HAVE_EBP_AVAILABLE = 0
and HAVE_EBX_AVAILABLE = 1.
Signed-off-by: Martin Storsjö <martin@martin.st>
On 32-bit OS X with gcc 4.0/4.2 and shared libraries enabled, the ebx register
is not available, but required to assemble the functions.
This reverts commit 8742a4f to a simplified version of the original constraints.
Michael Niedermayer
Vittorio Giovara
Janne Grunau
Diego Biurrun
Anton Khirnov
Ronald S. Bultje
Mans Rullgard
Roland Scheidegger
Martin Storsjö
Luca Barbato
Baptiste Coudurier
Laurent Aimar