Several decoders disable those anyway and they are not measurably faster
on x86. They might be somewhat faster on other platforms due to missing
emu edge SIMD, but the gain is not large enough (and those decoders
relevant enough) to justify the added complexity.
Allow supporting files for which the image stride is smaller than
the maximum block size + number of subpel mc taps, e.g. a 64x64 VP9
file or a 16x16 VP8 file with -fflags +emu_edge.
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>
Quite often, the original weights are multiple of 512. By prescaling them
by 1/512 when they are computed (once per frame), no intermediate shifting
is needed, and no prescaling on each call either.
The x86 code already used that trick.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
Prevents running error resilience on a previous frame which will write
to the pic->mb_type[] array of the previous image. The array might
already be re-used for a new image in a subsequent thread, thus cause
two threads to write to the same pic->mb_type[] array, causing a race
condition which can crash in rv34_decode_cbp(), called by
rv34_decode_inter_mb_header() (which accesses mb_type[] twice,
assuming values are maintained, which the race condition breaks).
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
Correct handling of errors to prevent hags or crashes is very complex
otherwise.
The frame initializing is also moved from decode_slice() to
decode_frame() for clarity.
This was an incorrect copy-and-paste to a code not needing the original code.
Spotted by Jason in a previous review but forgotten in the commit.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
Extract processing of intra 16x16 blocks from intra macroblock
processing.
Also implement a function performing inverse transform and block
reconstruction for DC-only blocks in 1 pass instead of 2.
Split inter/intra macroblock handling code. This will allow further
optimizations such as performing inverse transform and block reconstruction
in a single pass as well as specialize code.
Signed-off-by: Janne Grunau <janne-libav@jannau.net>
The sporadic threading errors during fate-rv30 were caused by calling
ff_thread_await_progress with mb row -1 as argument. That returns
immediately since progress is initialized to -1. Not yet computed
motion vectors from the reference could be used for the first
macroblocks.
When decoding coefficients, detect whether the block is DC-only, and take
advantage of this knowledge to perform DC-only inverse transform.
This is achieved by:
- first, changing the 108x4 element modulo_three_table into a 108 element
table (kind of base4), and accessing each value using mask and shifts.
- then, checking low bits for 0 (as they represent the presence of higher
frequency coefficients)
Also provide x86 SIMD code for the DC-only inverse transform.
Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>
Statistics for bourne.rmvb -an -f null
1 thread: 37.12s user 0.03s system 99% cpu 37.174 total
2 threads: 47.63s user 0.24s system 185% cpu 25.807 total
4 threads: 41.21s user 0.30s system 327% cpu 12.674 total
For small video dimensions calculations of the upper bound for pixel
access may result in negative value. Using an unsigned comparison
works only if the greater operand is non-negative. This is fixed by
doing edge emulation explicitly for such conditions.
Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>
Perform dequantization while decoding coefficients instead of performing it
on the entire coefficients buffer.
Since quantized coefficients are very sparse, this usually causes a small
speedup. Speedup of around 1% on Panda board compared to the removed here
neon code. Global speedup is probably around 3%.
Signed-off-by: Kostya Shishkov <kostya.shishkov@gmail.com>
