I got this idea from Pascal (Thanks). Before encoding a macroblock,
copy it to a 16x16 buffer, and then read source data from there
instead. This will help keep the source data in cache, and help
with the performance.
Change-Id: Id05f4cb601299150511d59dcba0ae62c49b5b757
Some further re-structuring of activity masking code.
Still has various experimental switches.
Supports a metric based on intra encode.
Experimental comparison against a fixed activity target rather
than a frame average, for altering rd and zbin.
Overall the SSIM performance is similar to TT's original
code but there is a much smaller PSNR hit of circa
0.5% instead of 3.2%
Change-Id: I0fd53b2dfb60620b3f74d7415e0b81c1ac58c39a
Declared the bmi in BLOCKD as a union instead of B_MODE_INFO.
Then removed B_MODE_INFO completely.
Change-Id: Ieb7469899e265892c66f7aeac87b7f2bf38e7a67
vp8_fast_quantize_b_pair_neon function added to quantize
two adjacent blocks at the same time to improve performance.
- Additional 3-6% speedup compared to neon optimized fast
quantizer (Tanya VGA@30fps, 1Mbps stream, cpu-used=-5..-16)
Change-Id: I3fcbf141e5d05e9118c38ca37310458afbabaa4e
Declared the bmi in MODE_INFO as a union instead of B_MODE_INFO.
This reduced the memory footprint by 518,400 bytes for 1080
resolutions. The decoder performance improved by ~4% for the
clip used and the encoder showed very small improvements. (0.5%)
This reduction was first mentioned to me by John K. and in a
later discussion by Yaowu.
This is WIP.
Change-Id: I8e175fdbc46d28c35277302a04bee4540efc8d29
This commit restructures the mb activity masking code
to better facilitate experimentation using different metrics
etc. and also allows for adjustment of the zero bin either
for encode only or both the encode and mode selection
stages
It also uses information from the current frame rather than
the previous frame and the default strength has been
reduced.
Change-Id: Id39b19eace37574dc429f25aae810c203709629b
Paul pointed out that the pointer to the gf_active_flags is not being
properly incremented in multithreaded encoder. This commit fixes the
issue by making sure the gf_active_ptr points to the starting of next
group of mb rows.
Change-Id: I3246e657d23beabb614dfb880733a68a5fd7e34c
The existing emulation of posix semaphores on Windows uses SetEvent()
and WaitForSingleObject(), which implements a binary semaphore, not a
counting semaphore as implemented by posix. This causes deadlock when
used with the expected posix semantics. Instead, this patch uses the
CreateSemaphore() and ReleaseSemaphore() calls (introduced in Windows
2000) which have the expected behavior.
This patch also reverts commit eb16f00, which split a semaphore that
was being used with counting semantics into two binary semaphores.
That commit is unnecessary with corrected emulation.
Change-Id: If400771536a27af4b0c3a31aa4c4e9ced89ce6a0
This patch is to fix a rare hang in multi-thread encoder that was
only seen on Windows. Thanks for John's help in debugging the
problem. More test is needed.
Change-Id: Idb11c6d344c2082362a032b34c5a602a1eea62fc
MV sad cost error is only used in full-pixel motion search,
which only need full-pixel resolution instead of quarter-pixel
resolution. This change reduced mvsadcost table size, and
removed unneccessary pamameter passing since this table is
constant once it is generated.
Change-Id: I9f931e55f6abc3c99011321f1dfb2f3562e6f6b0
Detect the number of available cores and limit the thread allocation
accordingly. On decoder side limit the number of threads to the max
number of token partition.
Core detetction works on Windows and
Posix platforms, which define _SC_NPROCESSORS_ONLN or _SC_NPROC_ONLN.
Change-Id: I76cbe37c18d3b8035e508b7a1795577674efc078
Thread synchronization was not correct when frame width was 1 MB.
Number of allocated encoding threads is limited by the sync_range.
There is no point having more because each thread lags sync_range MBs
behind the thread processing the row above.
http://code.google.com/p/webm/issues/detail?id=302
Change-Id: Icaf67a883beecc5ebf2f11e9be47b6997fdf6f26
In multithreaded mode the loopfilter is running in its own thread (filter level
calculation and frame filtering). Filtering is mostly done in parallel with the
bitstream packing. Before starting the packing the loopfilter level has
to be calculated. Also any needed reference frame copying is done in the
filter thread.
Currently the encoder will create n+1 threads, where n > 1 is the number of
threads specified by application and 1 is the extra filter thread. With n = 1
the encoder runs in single thread mode. There will never be more than n threads
running concurrently.
Change-Id: I4fb29b559a40275d6d3babb8727245c40fba931b
Reduce the number of sync points by letting each thread
continue imediatly with a new MB row.
Better multicore scaling, improves performance by 5-20% on ARM multicore.
Change-Id: Ic97e4d1c4886a842c85dd3539a93cb217188ed1b
Add a new encoder control, VP8E_SET_TUNING, to allow the application
to inform the encoder that the material will benefit from certain
tuning. Expose this control as the --tune option to vpxenc. The args
helper is expanded to support enumerated arguments by name or value.
Two tunings are provided by this patch, PSNR (default) and SSIM.
Activity masking is made dependent on setting --tune=ssim, as the
current implementation hurts speed (10%) and PSNR (2.7% avg,
10% peak) too much for it to be a default yet.
Change-Id: I110d969381c4805347ff5a0ffaf1a14ca1965257
Added the initialization of the pointer to active map. Also added the
same logic for cyclic refresh in mbrow encoding threads.
Change-Id: Ic48d0849dc706b27fba72d07dcc498075725663d
This uses MB variance to change the RDO weight for mode decision
and quantization.
Activity is normalized against the average for the frame, which is
currently tracked using feed-forward statistics.
This could also be used to adjust the quantizer for the entire
frame, but that requires more extensive rate control changes.
This does not yet attempt to adapt the quantizer within the frame,
but the signaling cost means that will likely only be useful at
very high rates.
Change-Id: I26cd7c755cac3ff33cfe0688b1da50b2b87b9c93
Changes 'The VP8 project' to 'The WebM project', for consistency
with other webmproject.org repositories.
Fixes issue #97.
Change-Id: I37c13ed5fbdb9d334ceef71c6350e9febed9bbba
Moved partition_bmi and partition_count out of MB_MODE_INFO and
placed into MACROBLOCK. Also reduced the size of other members
of the MB_MODE_INFO struct. For 1080p, the memory was reduced
by 1,209,516 bytes. The decoder performance appeared to improve
by 3% for the clip used.
Note: The main goal for this change is to improve the decoder
performance. The encoder will be revisited at a later date for
further structure cleanup.
Change-Id: I4733621292ee9cc3fffa4046cb3fd4d99bd14613
The main reason for the change was to reduce cycles in the token
decoder. (~1.5% gain for 32 bit) This layout should be more
cache friendly.
As a result of this change, the encoder had to be updated.
Change-Id: Id5e804169d8889da0378b3a519ac04dabd28c837
Note: dixie uses a similar layout
These copies occurred for each macroblock in the encoder and decoder.
Thetemp MB_MODE_INFO mbmi was removed from MACROBLOCKD. As a result,
a large number compile errors had to be fixed.
Change-Id: I4cf0ffae3ce244f6db04a4c217d52dd256382cf3
At the end of the decode, frame buffers were being copied.
The frames are not updated after the copy, they are just
for reference on later frames. This change allows multiple
references to the same frame buffer instead of copying it.
Changes needed to be made to the encoder to handle this. The
encoder is still doing frame buffer copies in similar places
where pointer reference could be done.
Change-Id: I7c38be4d23979cc49b5f17241ca3a78703803e66
This replaces the approximate division-by-multiplication in the
quantizer with an exact one that costs just one add and one
shift extra.
The asm versions have not been updated in this patch, and thus
have been disabled, since the new method requires different
multipliers which are not compatible with the old method.
Change-Id: I53ac887af0f969d906e464c88b1f4be69c6b1206
The new fdct lowers the round trip sum squared error for a
4x4 block ~0.12. or ~0.008/pixel. For reference, the old
matrix multiply version has average round trip error 1.46
for a 4x4 block.
Thanks to "derf" for his suggestions and references.
Change-Id: I5559d1e81d333b319404ab16b336b739f87afc79
When the license headers were updated, they accidentally contained
trailing whitespace, so unfortunately we have to touch all the files
again.
Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
