This will save the memory and improve the decode speed due to
removing unnecessary memset of big prev_mi array for
all the key frames.
Decoding a all key frames 1080p video shows speed improve around 2%.
Change-Id: I6284a445c1291056e3c15135c3c20d502f791c10
In the function mb_lpf_horizontal_edge_w_avx2_16 the usage of the intrinsic
_mm256_cvtepu8_epi16 cause a compiler bug in gcc 4.9.1.
until it will be fixed I created a workaround that create the up convert by
using broadcast128+shuffle.
The bug was reported here:
https://code.google.com/p/webm/issues/detail?id=867
Change-Id: I73452e6806f42e0fadcde96b804ea3afa7eeb351
This will save a lot of memory for decoder due to removing of prev_mi,
but prev_mi is still needed in encoder. So this will increase a little bit
memory for encoder.
Change-Id: I24b2f1a423ebffa55a9bd2fcee1077dac995b2ed
This patch allocated frame contexts outside VP9_COMMON. This allows
multiple threads to share the same copy of frame contexts, and
reduces the overhead. It also guarantees the correct update of
these contexts during bitstream packing. This patch doesn't change
encoding result.
Change-Id: Ic181a2460b891d1d587278a6d02d8057b9dbd353
Using 4 threads, frame parallel decode is ~3x faster than single thread
decode and around 30% faster than tile parallel decode for frame parallel
encoded video on both Android and desktop with 4 threads. Decode speed is
scalable to threads too which means decode could be even faster with more threads.
Change-Id: Ia0a549aaa3e83b5a17b31d8299aa496ea4f21e3e
All sad function that process above 32 consecutive elements are optimized
for AVX2:
vp9_sad64x64
vp9_sad64x32
vp9_sad32x64
vp9_sad32x32
vp9_sad32x16
vp9_sad64x64_avg
vp9_sad64x32_avg
vp9_sad32x64_avg
vp9_sad32x32_avg
vp9_sad32x16_avg
The functions that appeared as a hotspot is vp9_sad32x32 and vp9_sad64x64
vp9_sad32x32 was optimized by 68% and vp9_sad64x64 was optimized by 90%
both of them gave and overall ~2.3% user level gain
Change-Id: Iccf86b375a2b54c5fbbe685902ead0c9a561b9fd
this removes an assumption that worker->data1 would be pointing to a
TileWorkerData allocation.
additionally, within the multi-threaded loopfilter pass VP9LfSync as a
parameter to the worker hook, removing the need for a shadow pointer in
LFWorkerData.
Change-Id: Ic7b2faa34e3eb59dbcb8a7c67f333448fa047c88
The concept:
There's too much noise in source pixels for variance and at low bitrate
the reconstructed looks nothing like the source so we have problems
getting good partitionings with either. This skirts the issue by using
a box blur scaled down version for variance calculations. To compare
against source_var_ moved keyframe to be rd based like source_var.
Change-Id: Ie3babdbfadae324b7b5a76bea192893af27f0624
This commit breaks the overly broad header files into more
targeted and smaller ones, to help better structure the system
layout.
Change-Id: I7b24559d3ea6e582cf5d9bbe8f71459f9824d71b
The functions b_width_log2 and b_height_log2 only do direct
table fetch. This commit unifies such use cases by using the
table directly and removes these functions.
Change-Id: I3103fc6ba959c1182886a2799d21b8b77c8a7b6b
Add comments on the use case of these definitions. Further reduce
the scope of header file in vp9_context_tree.h.
Change-Id: Ic4a7638e838d0ac441b64abfc56e57354c059d75
This SSE2 is based on VP8 denoiser's SSE2 code. In VP8, there are
only 16x16 blocks in denoiser, while in VP9, there are 13 different
block sizes.
By adding this SSE2 code, the improvement of encoder speed is around
20%(using C code vs using SSE2 code), vary for different clips.
The unit test for VP9 denoiser is to confirm that the SSE2 code is
bit-exact with the C code. The unit test covers all block size.
Change-Id: Ic8d8ac26db4ea40a5f146b5678a065af07eaaa3d
Incorporates the WRAPLOW macro into the non-highbitdepth transforms
to aid hardware verification between a software C model and an
intended hardware implementation though the use of the configure
options: --enable-experimental --enable-emulate-hardware.
Note that to avoid further discrepancies between the sse/sse2
implementations of the transforms and the C implementation, when the
emulate hardware option is invoked, we also disable sse/sse2/etc.
Also incudes some minor cleanups/renaming etc.
Change-Id: Ib864d8493313927d429cce402982f1c8e45b3287
Miscellaneous bug-fixes for high bitdepth functionality.
With this patch, high bit-depth profiles become mostly functional,
except for an intermittent assert failure issue that is being
tracked.
Change-Id: I6a7fcbdcf1e5b09842e88535f8442d2e1230748c
The commit cleans up the header files in vp9_entropymv.h. This
file should only depend on vp9_mv.h and vp9_prob.h. Remove the
giant vp9_blockd.h from header file list.
Change-Id: I44cd26d2cfd10a16a9325778347dd53f888a874c
Moves transform type defines to vp9_common.h from vp9_idct.h
so that they can be included in vp9_rtcd_defs.pl safely.
Change-Id: Id5106227bee5934f7ce8b06f2eb9fa8a9a2e0ddb
This reverts commit eafc8c9c40.
tran_low_t/tran_high_t don't belong in a public header, they're private.
Similarly the public headers shouldn't rely on config defines,
vpx_config.h isn't installed.
Change-Id: I194ec273598da418df8dd727b6c0e78a556740ad
Some header file in vp9_idct.c has been included in vp9_idct.h.
This commit removes these redundant declarations.
Change-Id: I0238c27e4efff5c981eb437022c6bc6970c4e445
This commit fixes a compiling error in vp9_idct.h, where the codec
checks that the intermediate steps of transformation fit within
16-bit length. The issue was due to broken file dependency.
Change-Id: Ib22bba13a1e6df28489cb23d6774c561969f1fdc
The first comment is obselete given the way is now normative in VP9
bitstream. The second comment line was too long.
Change-Id: I6546585babf60d466485ddcf2daa6d2fa79e999a
As reported in issue #850, the condition for border extension was not
complete. This commit added the case when the scaling is enabled.
This fixes issue #850.
Change-Id: I67768b23f0dcc4ac9a9aa0a0825b0fe8cb85a72e
mi_grid_* are arrays of pointer to pointer. They save the pointers that point
to the MIs in cm->mi. But they are unnecessary and complicated. The original
goal was to remove MODE_INFO_t copy. But with an extra MODE_INFO_t pointer
inside MODE_INFO_t, same goal could be achieved.
This commit totally removes the mi_grid_* structures. But there are still
many dummy MODE_INFO_t inside cm->mi which are a waste of memory. Next commit
will do on-demand MODE_INFO_t allocation in order to save these memories.
Change-Id: I3a05cf1610679fed26e0b2eadd315a9ae91afdd6
This commit adds back sse2 or ssse3 optimized versio of a couple of
functions, fixes a ~10% performance regression.
Change-Id: I049786906e5a641224dced63c6492aec9d86d183
Libvpx was memseting every external frame buffer before decode. This
was to work around a valgrind issue in our C loop filter. Most of
the time this was not needed and we have noticed some significant
performance loss on some platforms. Now we require the application to
zero out the buffers if it is using external frame buffers.
Change-Id: I7330d00a315e65137ed30edd5f813e8929b76242
The issue was discovered on bitstream with 2x vertical downscale. For
zero MVs, y_pad is set to 1 only when vertical convolution is
required. The original code assumes that for y_step_q4 == 32 we don't
perform vertical convolution. But vp9_setup_scale_factors_for_frame()
sets convolve functions so that when x_step and y_step are both not
equal to 16, convolve in both directions is performed. And convolve()
unconditionally subtracts one stride from source pointer when calls
convolve_horiz(). This leads to invalid memory access.
Change-Id: I882dfa6081a58e172b5ffa55842bfcd6727f10bf
Adds various high bitdepth transform functions and tests.
Much of the changes are related to using typedefs tran_low_t
and tran_high_t for the final transform cofficients and intermediate
stages of the transform computation respectively rather than fixed
types int16_t/int. When vp9_highbitdepth configure flag is off,
these map tp int16_t/int32_t, but when the flag is on, they map
to int32_t/int64_t to make space for needed extra precision.
Change-Id: I3c56de79e15b904d6f655b62ffae170729befdd8
If optimizations use more than one cpu feature, allow
specifying them so that '--disable-X' still works
https://code.google.com/p/webm/issues/detail?id=854
Change-Id: I3108ea37b397371a2be84dd5f2380b304db23f18
Removed functions:
* vp9_post_proc_down_and_across_mmx
* vp9_mbpost_proc_down_mmx
* vp9_plane_add_noise_mmx
They all have sse2 equivalent.
Change-Id: I59c1fac12b7c96ca4538d455e4400c2b7875feff
vp9_variance_sse2.c contains a mix of intrinsics and references to
assembly which uses x86inc.asm; it's conditionally included as a result.
Change-Id: I254451483a65881c0b8e18e27bf0c3ddef60c4ec
Adds config parameter vp9_highbitdepth, to support highbitdepth profiles.
Also includes most vpx level high bit-depth functions. However
encode/decode in the highbitdepth profiles will not work until
the rest of the code is in place.
Change-Id: I34c53b253c38873611057a6cbc89a1361b8985a6
In the current implementation of the encoder,
frame buffers may come from the wider set of
12 such buffers, and is not restricted to the
8 allowed as reference frames. This is only
an implementation detail and does not affect
the constraint of having a total of 8 reference
buffers overall.
Change-Id: I075f777146c2df49c275d89232933f8127235175
The test to determine if the mode info buffers need
to be resized when the frame size changes was
incorrect, as per bug 837.
By storing the size of the allocated data structure,
a simple test determines whether to allocate more
memory when the frame size changes.
Change-Id: I1544698f2882cf958fc672485614f2f46e9719bd
'ref_frame_map' is initialized to -1. avoids using an invalid index if
VP9_GET_REFERENCE/VP8_COPY_REFERENCE controls are issued after a decode
error.
Change-Id: I4599762c4d0b07a5943a72bf4a86ccb596cc062a
Replaced encoder and decoder functions to get a pointer
to a reference frame with a common function, vp9_get_ref_frame,
and simplified it.
Change-Id: Icb206fcce8caace3bfd1db3dbfa318dde79043ee
in the sub_pixel_*variance* function the dst is aligned to 16 bytes and not
to 32 bytes - now load unaligned data
Change-Id: I2e0b9745543697efc56fefa32857ea10117af135
A bug in Microsoft compiler was found in the function
vp9_filter_block1d16_v8_avx2 and a workaround applied.
the bug occur when there was 4 consecutive maddubs + min + adds
intrinsic instructions.
Change-Id: I83499faeb70971e650e5663fd2490360ddb1a51b
in the function sad32x32x4d and sad64x64x4d the source is aligned to 16 bytes
and not to 32 bytes - the load is now unaligned.
Change-Id: I922fdba56d0936b5cf72e4503519f185645a168c
Specifies the bit-depth, color sampling and colorspace
for intra only frames for profiles > 0
Also adds checks to ensure that profile 1 and 3 are
exclusively used for non 420 streams.
Change-Id: Icfb15fa1acccbce8f757c78fa8a2f60591360745
This commit adds a configure time option used to enable strict error
checking in decoder to make sure intermediate stage cofficients of
inverse transforms are within valid range of signed 16 bit integer.
For valid VP9 input streams, intermediate stage coefficients should
always stay within the range of a signed 16 bit integer. Coefficients
can go out of this range for invalid/corrupt VP9 streams. However,
strictly checking this range for every intermediate coefficient can
be a burden for decoder, therefore such validation is only enabled
with configure option --enable-coefficient-range-checking.
Change-Id: I47d47c8c4e48a922c3d223ca59064f51b3f0f5ed
vp9_variance8x8(), and vp9_get8x8var().
On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~1.2%.
Change-Id: I8a66ac2a0f550b407caa27816833bdc563395102
vp9_variance16x16(), and vp9_get16x16var().
On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~16.7%.
Change-Id: Ib163aa99f56e680194aabe00dacdd7f0899a4ecb
Remove all the redundant dct functions (dct4x4, dct8x8)
in avx2 except dct32x32 those functions were copied originally from dct_sse2
Change-Id: I742576fbf5175f3ac09f2076976a9247b259323e
The original implementation only allocates one segmentation map and this
works fine for serial decode. But for frame parallel decode, each thread
need to have its own segmentation map and the last frame segmentation map
should be provided from last frame decoding thread.
After finishing decoding a frame, thread need to serve the old segmentation
map that associate with the previous decoded frame. The thread also need to
use another segmentation map for decoding the current frame.
Change-Id: I442ddff36b5de9cb8a7eb59e225744c78f4492d8
This commit turns on the existing vp9_get_prob function using
64 bit in the intermediate step. It fixes the ioc issue for 4K
above frame sizes (issue 828).
Change-Id: I9f627f3beca2c522f73b38fd2a3e7eefdff01a7c
Separates HBD profile int two profiles (2 and 3) consistent with the
highbitdepth branch. This patch is ported from the original highbitdepth
branch patch: https://gerrit.chromium.org/gerrit/#/c/70460/
Two of the invalid file tests needed to be updated.
Change-Id: I6a4acd2f7a60b1fb4cbcc8e0dad4eab4248431e3
This patch is the first step toward simplifying the
frame buffer handling.
The final goal is to have a common frame buffer handling
framework for both encoder and decoder that incorporates
the existing ability to use externally allocated memory.
Change-Id: I2c378a4f54a39908915f46c4260e17a080db7ff1
and vp9_sad16x16_neon()
On a Nexus 7, vpxenc (in realtime mode, speed -6)
reported a performance improvement of ~17%.
Change-Id: I91e070cde2973451083d3f3d63b49b7886de9a85
pthread.h is not supported in windows. vp9_thread.h includes
the emulation layer for pthread in windows.
Change-Id: I2b1c8ec299928472faca7ebeea998170c9f4d744
This commit fixes a mismatched use case of block size in non-RD
intra prediction check. The residual SSE and variance should be
calculated per transform block size, instead of operating block
size, which caused chrome valgrind warning on conditional jump
based on uninitialized value (webm issue 823). This commit
resolves this issue.
Change-Id: I595c06599c7e0fd0e4a08736519ba68fc14bc79a
Prepare for frame parallel decoding, the reference count buffers
need to be protected by mutex. Move vp9_thread.* to common
folder so that those buffers could use cross-platform mutex
from vp9_thread.*.
(cherry picked from commit 337e8015c9)
Change-Id: I0587a08447925f4554d7788686a31483c2ae3f37
Refactoring to remove some duplication of probability
tables between tokenization and detokenization.
Change-Id: I2fc6a6497f9c0410021a9b41f828bc58a864e466
pull the latest from libwebp.
Original source:
http://git.chromium.org/webm/libwebp.git
100644 blob 264210ba2807e4da47eb5d18c04cf869d89b9784 src/utils/thread.c
commit 46fd44c1042c9903b2f1ab87e9f200a13c7e702d
Author: James Zern <jzern@google.com>
Date: Tue Jul 8 19:53:28 2014 -0700
thread: remove harmless race on status_ in End()
if a thread was still doing work when End() was called there'd be a race
on worker->status_. in these cases, however, the specific value is
meaningless as it would be >= OK and the thread would have been shut
down properly, but we'll check 'impl_' instead to avoid any potential
TSan/DRD reports.
Change-Id: Ib93cbc226a099f07761f7bad765549dffb8054b1
Change-Id: Ib0ef25737b3c6d017fa74822e21ed58508230b91
Currently, vp9_diamond_search_sadx4() is only called when sse3 is
enabled, which is improper since sse2 optimization of sdx4df
functions are available. Changed to always use
vp9_diamond_search_sadx4().
Change-Id: I4b95d6b7a3c6c645783c373f0ba8d645ece24717
- fix indent, spelling
- drop some whitespace in some comments
- add an assert in vp9_setup_mask, it shouldn't be called on decode
error
Change-Id: Ic312a815e977a6f9cb81ceb7b039eeada76c5aa0
There are sse2 optimization of sdx4df functions. Instead of calling
vp9_refining_search_sadx4 only when sse3 is enabled, call it always.
Change-Id: I24f93818f7d4209d1425039e0eb099ff9ff08fe9
This commit enables a new quantization process for 32x32 2D-DCT
transform coefficient blocks. It improves the compression
performance of speed 5 by 1.4%. The overall compression gains of
speed 5 due to the new quantization scheme is 4.7%. It also includes
the SSSE3 implementation of the 32x32 quantization process.
Change-Id: I0855b124fd6462418683f783f5bcb44255c9993b
This patch fixes bug 633:
https://code.google.com/p/webm/issues/detail?id=633
The first decoded frame does not have to be a keyframe,
it could be an inter-frame that is coded intra-only.
This patch fixes the handling of intra-only frames.
A test vector has also been added that encodes 3
intra-only frames at the start of the clip. The
test vector was generated using the code in the
following patch:
https://gerrit.chromium.org/gerrit/#/c/70680/
Change-Id: Ib40b1dbf91aae2bc047e23c626eaef09d1860147
Prepare for frame parallel decoding, the reference count buffers
need to be protected by mutex. Move vp9_thread.* to common
folder so that those buffers could use cross-platform mutex
from vp9_thread.*.
Change-Id: I541277cf15eefed6641555944f67f4a0bcdc8154
Prepare for frame parallel decoding, the frame buffers must be
separated from the encoder and decoder structure, while the encoder
and decoder will hold the pointer of the BufferPool.
Change-Id: I172c78f876e41fb5aea11be5f632adadf2a6f466
This commit re-designs the quantization process for transform
coefficient blocks of size 4x4 to 16x16. It improves compression
performance for speed 7 by 3.85%. The SSSE3 version for the
new quantization process is included.
The average runtime of the 8x8 block quantization is reduced
from 285 cycles -> 255 cycles, i.e., over 10% faster.
Change-Id: I61278aa02efc70599b962d3314671db5b0446a50
the buffer is only used in encoding and only when
CONFIG_INTERNAL_STATS or CONFIG_VP9_POSTPROC is enabled.
a future change should decouple this from the frame buffer allocation
and make it conditional based on runtime flags when the above config
options are enabled.
reduces decode heap usage by at least 12%
Change-Id: Id0b97620d4936afefa538d3aadf32106743d9caf
This patch checks that a decoder never tries to reference frame that's
outside the range of 2x to 1/16th the size of this frame. Any attempt
to do so causes a failure.
Change-Id: I5c98fa7bb95ac4f29146f29dd92b62fe96164e4c
This commit enables a fast reference motion vector search scheme.
It checks the nearest top and left neighboring blocks to decide the
most probable predicted motion vector. If it finds the two have
the same motion vectors, it then skip finding exterior range for
the second most probable motion vector, and correspondingly skips
the check for NEARMV.
The runtime of speed -5 goes down
pedestrian at 1080p 29377 ms -> 27783 ms
vidyo at 720p 11830 ms -> 10990 ms
i.e., 6%-8% speed-up.
For rtc set, the compression performance
goes down by about -1.3% for both speed -5 and -6.
Change-Id: I2a7794fa99734f739f8b30519ad4dfd511ab91a5
There is a normative scaling range of (x1/2, x16)
for VP9. This patch fixes the maximum downscaling
tests that are applied in the convolve function.
The code used a maximum downscaling limit of x1/5
for historic reasons related to the scalable
coding work. Since the downsampling in this
application is non-normative it will revert to
using a separate non-normative scaler.
Change-Id: Ide80ed712cee82fe5cb3c55076ac428295a6019f
The encoder currently allocates frame buffers before
it establishes what the chroma sub-sampling factor is,
always allocating based on the 4:4:4 format.
This patch detects the chroma format as early as
possible allowing the encoder to allocate buffers of
the correct size.
Future patches will change the encoder to allocate
frame buffers on demand to further reduce the memory
profile of the encoder and rationalize the buffer
management in the encoder and decoder.
Change-Id: Ifd41dd96e67d0011719ba40fada0bae74f3a0d57
This breaks the profile 1 bitstream.
Don't force non420 uv transform size to 1/4 y size. In the 4:2:0 case the
chroma corresponding to a luma block is 1/4 its size. In the 4:4:4 case
chroma and luma planes are the same size. Disallowing larger transforms
can result in a loss of compression efficiency and is inconsistent.
For sub-8x8 blocks only average corresponding motion vectors.
4:2:0 and profile 0 behavior remains unchanged.
Change-Id: I560ae07183012c6734dd1860ea54ed6f62f3cae8
This commit enables a fast path computational flow for forward
transformation. It checks the sse and variance of prediction
residuals and decides if the quantized coefficients are all
zero, dc only, or more. It then selects the corresponding coding
path in the forward transformation and quantization stage.
It is currently enabled in rtc coding mode. Will do it for rd
coding mode next.
In speed -6, the runtime for pedestrian_area 1080p at 1000 kbps
goes down from 14234 ms to 13704 ms, i.e., about 4% speed-up.
Overall coding performance for rtc set is changed by -0.18%.
Change-Id: I0452da1786d59bc8bcbe0a35fdae9f623d1d44e1
In non frame-parallel decoding, this works the same way as
current decoding scheme. Every time after decoder finish
decoding a frame, it will swap the current mode info pointer
and previous mode info pointer if the decoded frame needs
to be shown. Both mode info pointer and previous mode info
pointer are from mode info arrays.
In frame-parallel decoding, this will become more complicated
as current frame's mode info pointer will be shared with next
frame as previous mode info pointer. But when one decoder
thread finishes decoding one frame and starts to work on next
available frame, it needs to retain the decoded frame's mode
info pointers until next frame finishes decoding. The mode info
index will serve this purpose. The decoder will use different
buffer in the mode info arrays and use the other buffer to save
previous decoded frame’s mode info.
Change-Id: If11d57d8eb0ee38c8876158e5482177fcb229428
tests failing under Win32/Win64
+ dct16x16_test: add missing avx2 functions (partially disabled)
exercises the forward transforms
no idct/iht implementations, so the c-code is used
Change-Id: I04f64a457fa0828a00f32b5c9fe4f55294f21f61
As a side-effect, the sad unit tests for VP8 and VP9
had to be separated.
Fixes a bug in original patch:
(https://gerrit.chromium.org/gerrit/#/c/70163/8)
that was reverted due to a nightly test failure.
Change-Id: Ia2a4e9e278fd3c89d6c3c82fcc6381320ec2a8a6
An overflow issue could potentially happen in the second round 1-D
transform of the SSSE3 full inverse 16x16 2D-DCT. This commit fixes
this issue.
Change-Id: Ia19e4888fda1cc929a28a5f89a5beec612d628dc
This commit enables SSSE3 implementation of the inverse 2D-DCT
with only first 10 coefficients non-zero. It reduces the runtime
of SSE2 version from 745 cycles to 538 cycles, i.e., 27% speed-up.
Change-Id: I18ba4128859b09c704a6ee361d69a86c09fe8dfe
The subpixel SSSE3 was fixed in this patch:
https://gerrit.chromium.org/gerrit/#/c/70283/
So the equivalent AVX2 is fixed accordingly.
Change-Id: Ieebbc1949c99d34b12b8b47692df71aca5001f3a
This commit enables the SSSE3 implementation of full inverse 16x16
2D-DCT. The unit runtime goes down from 1642 cycles to 1519 cycles,
about 7% speed-up.
Change-Id: I14d2fdf9da1fb4ed1e5db7ce24f77a1bfc8ea90d
In 8-tap filtering, to guarantee the intermediate results fit in
16 bits, the order of accumulating the products needs to be done
correctly, and the largest product should be added last. This
patch fixed the problem using the method in commit "Correct ssse3
8/16-pixel wide sub-pixel filter calculation".
Change-Id: I79d0ad60c057b15011ece84cda9648eee0809423
As mismatchs were found between the intrinsic version and c only. The
commit temporarily revert to use the matching assembly version to
allow further investigation.
Change-Id: I08436c47d4888b562c0eac8e8856d90a831442df
This did the same correction as the one in commit "Correct ssse3
8/16-pixel wide sub-pixel filter calculation" to avoid saturation
during filtering.
Change-Id: Ife9aa3f62daf9114eb24fe38f7baa3c3f361b2d6
Renames all x86_64 specific assembly files to consistently
end in _x86_64.asm. This will be useful for build systems to
handle these files differently.
All new 64-bit specific assembly files should use the new
naming convention.
Change-Id: I36c89584967c82ffc4088b1b5044ac15d2bb7536
The final goal is eventually to get rid of both itxm_add and fwd_txm4x4.
This patch does it in the decoder.
Change-Id: Ibb3db57efbcbb1ac387c6742538a9fcf2c6f24a5
The current decode_tiles decodes the frame one tile by one tile
and then loopfilter the whole frame or use another worker thread to
do loopfiltering.
|------|------|------|------|
|Tile1-|Tile2-|Tile3-|Tile4-|
|------|------|------|------|
For example, if a tile video has one row and four cols, decode_tiles
will decode the Tile1, then Tile2, then Tile3, then Tile4.
And during decode each tile, decode_tile will decode row by row in
each tile.
For frame parallel decoding, decode_tiles will decode video in row order
across the tiles. So the order will be:
"Decode 1st row of Tile1" -> "Decode 1st row of Tile2"
-> "Decode 1st row of Tile3" -> "Decode 1st row of Tile4"
-> "Decode 2nd row of Tile1" -> "Decode 2nd row of Tile2"
-> "Decode 2nd row of Tile3" -> "Decode 2nd row of Tile4"-> "loopfilter 1st row"
Change-Id: I2211f9adc6d142fbf411d491031203cb8a6dbf6b
Make all post-processor code conditionally
compilable based on the CONFIG_VP9_POSTPROC
macro.
Also, remove the vizualization code from VP9
since it is out of date and will not compile.
Change-Id: I1e9e13a09ecd43e9a3f3704c175ae8cd258ababd
vp9_block_error_sse2 can only handle 16 bytes at a time but
the function requires to handle a sequence of 32 bytes at a time
so each 16 bytes is handled in a different register.
With AVX2 optimization the 32 bytes can be handled in one register instead
of two in the SSE2
The vp9_block_error was optimized by 85%.
The user level was optimized by 1.2%
Change-Id: Ia8fffe60e61eff7432a5fbd538757894f6c319fd
As a side-effect, the max_sad check is removed from the
C-implementation of VP8, for consistency with VP9, and to
ensure that the SAD tests common to VP8/VP9 pass.
That will make the VP8 C implementation of sad a little slower
but given that is rarely used in practice, the impact will be
minimal.
Change-Id: I7f43089fdea047fbf1862e40c21e4715c30f07ca
The warning messages complained that there are unused arguments
in a few prediction modes. This structure was designed on purpose,
such that a wrapper function can cover all prediction mode cases
and make them readily accessible as an pointer array.
This commit silences such warnings.
Change-Id: I7036b6bdb70747e5327d8f6fceb154f100abc4c0
Inline loopfilter has been already handled in vp9_decode_frame().
Collecting all similar code in one place now.
Change-Id: I358a0280fc7c2b27cca520bc1e8c16c4eb6491dd
* changes:
Turn on unit tests for SSSE3 8x8 forward and inverse 2D-DCT
Change eob threshold for partial inverse 8x8 2D-DCT to 12
SSSE3 8x8 inverse 2D-DCT with first 10 coeffs non-zero
The scanning order has the first 12 coefficients of the 8x8 2D-DCT
sitting in the top left 4x4 block. Hence the partial inverse 8x8
2D-DCT allows to handle cases with eob below 12.
The overall runtime of the inverse 8x8 2D-DCT unit is reduced from
166 cycles (using SSE2) to 150 cycles (using SSSE3).
Change-Id: I4514f9748042809ac84df4c14382c00f313f1cd2
This commit enables ssse3 assembly implementation of the 8x8
inverse 2D-DCT with only first 10 coefficients non-zero. The
average runtime for this unit goes down from 198 cycles to 129
cycles (34.8% faster).
Change-Id: Ie7fa4386f6d3a2fe0d47a2eb26fc2a6bbc592ac7
Includes changes that are not compatible with VS windows builds.
Amongst other things stdint.h is not supported in VS.
This reverts commit 89fbf3de50.
Change-Id: Ifa86d7df250578d1ada9b539c9ff12ed0c523cdd
This commit enables SSSE3 version full inverse 8x8 2D-DCT and
reconstruction. It makes the runtime of vp9_idct8x8_64_add down
from 256 cycles (SSE2) to 246 cycles.
Change-Id: I0600feac894d6a443a3c9d18daf34156d4e225c3
Assembly implementation of ssse3 8x8 forward 2D-DCT. The current
version is turned on only for x86_64. The average unit runtime
goes from 157 cycles down to 136 cycles, i.e., about 12.8% faster.
This translates into about 1.5% speed-up for pedestrian_area 1080p
at speed 2.
Change-Id: I0f12435857e9425ed7ce12541344dfa16837f4f4
This patch fixed the uninitialized read errors in Issue 748:
"dr memory VP9 encode errors". In vp9_convolve_avg_sse2,
when width is 4, pavgb reads 8 bytes from dst buffer that is
out of range. An error is reported although the data is not
actually used later. This issue was resolved by preventing
uninitialized reads.
Change-Id: I109a54910aa47139cb13119de86f2062cff207df
The macosx release of clang v5.0 identifies itself as:
Apple LLVM version 5.0 (clang-500.2.79) (based on LLVM 3.3svn)
This version of clang uses the older _mm_broadcastsi128_si256, like
v3.3, as given away in the LLVM svn version above.
Change-Id: I4d6d59d5454efd57d2ae9e75f5eb7486af7cbd0c
Calculate the difference variance between last source frame and
current source frame. The variance is calculated at 16x16 block
level. The variances are compared to several thresholds to decide
final partition sizes.
An adaptive strategy is implemented to decide using
SOURCE_VAR_BASED_PARTITION or FIXED_PARTITION based on motions
in the video. The switching test is done once every
search_type_check_frequency frames.
The selection of source_var_thresh needs to be investigated
further later.
RTC set Borg test showed 0.424% overall psnr gain, and 0.357%
ssim gain. For clips with large enough static area, the
encoding speedup is around 2% to 15%.
Change-Id: Id7d268f1d8cbca7fb8026aa4a53b3c77459dc156
Adds some high-level hooks for profile 2 before further
progress on the implementation.
According to the definitiion in this patch:
1. Profile 2 only supports 10 or 12 bit color but not 8
2. Profile 2 supports all color sampling modes: 444, 422 and 420,
and alpha plane.
3. Profile 3 is currently undefined.
Please consider the definition carefully and suggest modifications
to the definition as needed.
Change-Id: I5b284fc679e54ac5aee171af72fa7994cfd28995
For very large size video image, the scaling calculation may need use
value beyond the range of int. This commit upgrade the value to 64bit
to make sure the calculation do not wrap around INT_MAX.
The change corrected the decoder behavior.
The bug affects only very large resolution video because the scaling
calculation was sufficient for image size smaller than 2^13.
This resolves issue:
https://code.google.com/p/webm/issues/detail?id=750
Change-Id: I2d2ed303ca6482f31f819f3c07d6d3e98ef3adc5
2 functions were optimized for avx2 by using full 256 bit register
In order to handle 32 elements in parallel instead of only 16 in parallel:
1. vp9_sad32x32x4d
2. vp9_sad64x64x4d
The function level gain is 66% and the user level gain is ~1%.
Change-Id: I4efbb3bc7d8bc03b64b6c98f5cd5c4a9dd3212cb
Fixed dr memory errors reported in Issue 736:
https://code.google.com/p/webm/issues/detail?id=736
All elements in left_col buffer need to be initialized to ensure
the correctness of SIMD operations in x86 optimized code.
Change-Id: I8e7f26ab45cca8099c1f9342bcf852f828bda7e4
clang reports gcc-4.2.1 in e.g., 3.3, 3.4; add a specific clang version
check for _mm256_broadcastsi128_si256
fixes issue #720
Change-Id: I5c8e3c27fdea05d8a5b050e8cb74894b595f4709
significantly speeds up file generation.
the goal of this change is to convert rtcd.sh to perl as directly as
possible to allow for simple comparison. future changes can make it more
perl-like.
---
Linux
[CREATE] vpx_scale_rtcd.h
real 0m0.485s -> 0m0.022s
[CREATE] vp8_rtcd.h
real 0m4.619s -> 0m0.060s
[CREATE] vp9_rtcd.h
real 0m10.102s -> 0m0.087s
Windows
[CREATE] vpx_scale_rtcd.h
real 0m8.360s -> 0m0.080s
[CREATE] vp8_rtcd.h
real 1m8.083s -> 0m0.160s
[CREATE] vp9_rtcd.h
real 2m6.489s -> 0m0.233s
Change-Id: Idfb71188206c91237d6a3c3a81dfe00d103f11ee
The function has evolved over time, now only calls vp9_rtcd(), so this
commit removes the function and changes to call vp9_rtcd() directly.
Change-Id: I8cfa6190daa4b28f6f3d1e11bb3a07f9c95322bf
Optimizing 2 functions to process 32 elements in parallel instead of 16:
1. vp9_sub_pixel_avg_variance64x64
2. vp9_sub_pixel_avg_variance32x32
both of those function were calling vp9_sub_pixel_avg_variance16xh_ssse3
instead of calling that function, it calls vp9_sub_pixel_avg_variance32xh_avx2
that is written in avx2 and process 32 elements in parallel.
This Optimization gave 80% function level gain and 2% user level gain
Change-Id: Iea694654e1b7612dc6ed11e2626208c2179502c8
different_ref_found is always equal to one (if calculated) because
ref_frame[0] != ref_frame[1] for each mi-block.
Change-Id: Ibd7625b7b29dec2fd3c40edbc3de1169abb78585
for ABSDATA mode, so segment loop filter level always fall in valid
range for both Absolute and delta modes.
Change-Id: If90df3411479533dbdab63f8ae088d2f5dd174a9
The qindex for a segment was not clamped in ABSDATA mode, which may
cause invalid memory access if an ill-formed stream has a negative
value in ABSDATA mode. This commit added clamp to make sure qindex
for a segment always fall into valid range.
Change-Id: I0a74d00f4ef40aec7edaeca1d03c8645e23ab08c
Optimizing 2 functions to process 32 elements in parallel instead of 16:
1. vp9_sub_pixel_variance64x64
2. vp9_sub_pixel_variance32x32
both of those function were calling vp9_sub_pixel_variance16xh_ssse3
instead of calling that function, it calls vp9_sub_pixel_variance32xh_avx2
that is written in avx2 and process 32 elements in parallel.
This Optimization gave 70% function level gain and 2% user level gain
Change-Id: I4f5cb386b346ff6c878a094e1c3b37e418e50bde
Optimizing all SSSE3 assembly for convolution:
1. vp9_filter_block1d4_h8_sse2
2. vp9_filter_block1d8_h8_sse2
3. vp9_filter_block1d16_h8_sse2
4. vp9_filter_block1d4_v8_sse2
5. vp9_filter_block1d8_v8_sse2
6. vp9_filter_block1d16_v8_sse2
my optimization include:
-processing 2x8 elements in one 128 bit register instead of processing
8 elements in one 128 bit register.
-removing unecessary loads.
This optimization gives between 2.4% user level gain for 480p input
and 1.6% user level gain for 720p.
This Optimization is done only for 64 bit
Change-Id: Ic07fce2f9360329b4f2d956efda1480ae958766b
Added support for external frame buffers to libvpx's VP9 decoder.
If the external frame buffer functions are set then libvpx will
call the get function whenever it needs a new frame buffer to
decode a frame into. And it will call the release function
whenever there are no more references to that buffer.
Change-Id: Id2934d005f606af6e052fb6db0d5b7c02f567522
Prior to this commit, both encoder and decoder reset mode/mv info from
previous frame in error resilient mode to ensure bitstreams are able to
decode when there is loss of frame in decoder side. However, this is
not necessary. This commit changed to remove the reset, so encoder can
continue to use mode/mv/partition information from previously encoded
frame without affecting decodeablilty under loss of frame.
Change-Id: I0279f862900dc647fb471ae3389770bb1b9f454f
Two convolve functions were optimized for AVX2:
1. vp9_filter_block1d16_h8
2. vp9_filter_block1d16_v8
vp9_filter_block1d16_v8 was optimized for AVX2 by reducing the number of
loop strides by half, two strides were processed in parallel.
vp9_filter_block1d16_v8 was also optimized in the same way also some of the
loads were being done outside of the loop and by that preventing redundant
loads.
This Optimization gives 43% function level gain and 1.3% user level gain.
Now can be compiled in Windows
Change-Id: I2714124cfb0c14a77d7a0ce126a20db92ffbf92c
This CL changes libvpx to call a function when a frame buffer
is needed for decode. Libvpx will call a release callback when
no other frames reference the frame buffer. This CL adds a
default implementation of the frame buffer callbacks. Currently
only VP9 is supported. A future CL will add support for
applications to supply their own frame buffer callbacks.
Change-Id: I1405a320118f1cdd95f80c670d52b085a62cb10d
This patch adds a buffer-based rate control for temporal layers,
under CBR mode.
Added vpx_temporal_scalable_patters.c encoder for testing temporal
layers, for both vp9 and vp8 (replaces the old vp8_scalable_patterns).
Updated datarate unittest with tests for temporal layer rate-targeting.
Change-Id: I8900a854288b9354d9c697cfeb0243a9fd6790b1
This isn't strictly necessary, but makes the file more consistent
with the other arm assembly source files.
Change-Id: I245c9677d89e0ab3f31991e473764858af35b180
avoid wrapping msvc includes with extern "C"; this breaks some visual
studio builds of the (c++) tests.
Change-Id: Ie8062d55d4f4c049f6cd360a36da6a67607df132
This patch added ssse3 optimization of bilinear sub-pixel filters.
The real time encoder was speeded up by ~1%.
Change-Id: Ie82e98976f411183cb8c61ab8d2ba0276e55a338
The new implementation disagrees when the argument is equal to 2**n but
that is never called in practice and based on how it is used the new
implementation is correct in that case.
Change-Id: Ifbac4ad87d459fe6bd2fd0f400c0340f96617342
Using bilinear filters could speed up the codec in real-time mode.
This patch added sse2 optimizations of bilinear filters that
operate on different-sized blocks.
Tests showed that the real-time encoder was speeded up by 3%.
Change-Id: If99a7ee4385fcc225c3ee7445d962d5752e57c3f
This patch adds a buffer-based rate control for temporal layers,
under CBR mode.
Added vpx_temporal_scalable_patters.c encoder for testing temporal
layers, for both vp9 and vp8 (replaces the old vp8_scalable_patterns).
Updated datarate unittest with tests for temporal layer rate-targeting.
Change-Id: I9cb6cce2494390ae6096ee17774af7fb9308bde7
As pointed out by Dmitry and James, "partial" is a Microsoft-
specific c++ keyword, and it is renamed.
Change-Id: Ia0fc11ceb89e54b3195287f89f7e26edbbe9beb8
Implemented parallel loopfiltering, which uses existing tile-
decoding threads. Each thread works on one row, and when that row
is loopfiltered, it moves to next unattended row. To ensure the
correct filtering order, threads are synchronized and one
superblock is filtered only if the superblocks it depends on are
filtered already.
To reduce synchronization overhead and speed up the decoder, we use
nsync > 1 for high resolution.
Performance tests:
1. on desktop:
8-tile 4k video using 8 threads, speedup: 70% - 80%
4-tile HD video using 4 threads, speedup: ~35%
2. on mobile device(Nexus 7):
4-tile 1080p video using 4 threads, speedup: 18% - 25%
4-tile 1080p video using 2 threads, speedup: 10% - 15%
Change-Id: If54b4a11960dd706c22d5ad145ad94156031f36a
This patch only works if the video is a width and height that are both
a multiple of 32.. It sets every partition to 16x16, and does INTRADC
only on the first frame and ZEROMV on every other frame. It always does
does the largest possible transform, and loop filter level is set to 4.
Was ~20% faster than speed -5 of vp8
Now 20% slower but adds motion search ( every block ), nearest, near
and zeromv
The SVC test was changed because - while this realtime mode produces
bad quality albeit quickly, it isn't obeying all the rules it should
about which frames are available.
Change-Id: I235c0b22573957986d41497dfb84568ec1dec8c7
Encoder's boarder is still 160, while decoder's boarder will be 32.
With on demand and separate boarder buffer for boarder extension.
The decoder's boarder does not need to to 160 anymore.
Change-Id: I93d5aaff15a33a2213e9761eaa37c5f2870747db
This commit deprecates the use of best_mv from encoding and bit-stream
writing stages. It hence removes the definition from MACROBLOCKD.
Change-Id: I8e5302775a2aa4a18900726df407bff881f2dfb1
This commit further optimizes SSE2 operations in the second 1-D
inverse 16x16 DCT, with (<10) non-zero coefficients. The average
runtime of this module goes down from 779 cycles -> 725 cycles.
Change-Id: Iac31b123640d9b1e8f906e770702936b71f0ba7f
Optimizing all SSSE3 assembly for convolution:
1. vp9_filter_block1d4_h8_sse2
2. vp9_filter_block1d8_h8_sse2
3. vp9_filter_block1d16_h8_sse2
4. vp9_filter_block1d4_v8_sse2
5. vp9_filter_block1d8_v8_sse2
6. vp9_filter_block1d16_v8_sse2
my optimization include:
-processing 2x8 elements in one 128 bit register instead of processing
8 elements in one 128 bit register.
-removing unecessary loads.
This optimization gives between 2.4% user level gain for 480p input
and 1.6% user level gain for 720p.
This Optimization done only for 64bit.
Change-Id: Icb586dc0c938b56699864fcee6c52fd43b36b969
This commit is the first patch optimizing SSE2 implementation of inverse
16x16 DCT with <10 non-zero coefficients. It focused on the first 1-D (row)
transformation. It exploits the fact that only top-left 4x4 block contains
non-zero coefficients, in a 2-D inverse 16x16 DCT with <10 coeffients.
The average runtime of idct16x16_10 unit is reduced from
883 cycles -> 779 cycles (12% faster).
For pedestrian_area_1080p 300 frames at 4000 kbps, the speed 2 runtime goes
down from 310651 ms -> 305910 ms. The decoding speed goes up from
80.37 fps -> 80.87 fps.
Change-Id: Ic6f3ac5a637a76c07ba73ddaafe318a699fea645
Optimizing the variance functions: vp9_variance16x16, vp9_variance32x32,
vp9_variance64x64, vp9_variance32x16, vp9_variance64x32,
vp9_mse16x16 by migrating to AVX2
some of the functions were optimized by processing 32 elements instead of 16.
some of the functions were optimized by processing 2 loop strides of 16
elements in a single 256 bit register
This optimization gives between 2.4% - 2.7% user level performance gain
and 42% function level gain.
Change-Id: I265ae08a2b0196057a224a86450153ef3aebd85d
This commit adds input/output ports for IDCT8_1D macro function to
provide more flexibility in variable use. It allows to skip several
buffer swap operations.
Change-Id: I21f3450509537322293043b3281bfd3949868677
Adding RefBuffer to simplify reference buffer management. The struct has a
pointer to image data and scale factors relative to the current frame.
Change-Id: If38eb1491ff687cc11428aee339f3e052e2c5d9e
This commit merges the initial buffer swap operations in idct8_1d_sse2
into the array transpose step, hence reducing number of instructions
therein.
Change-Id: I219f6f50813390d2ec3ee37eecf2a4a2b44ae479
This commit optimizes the SSE2 implmentation of idct8x8_10. It exploits
the fact that only top-left 4x4 block contains non-zero coefficients,
and hence reduces the instructions needed.
The runtime of idct8x8_10_sse2 goes down from 216 to 198 CPU cycles,
estimated by averaging over 100000 runs. For pedestrian_area_1080p 300
frames coded at 4000kbps, the average decoding speed goes up from
79.3 fps to 79.7 fps.
Change-Id: I6d277bbaa3ec9e1562667906975bae06904cb180
Moving back to scale_factors struct. We don't need anymore x_offset_q4 and
y_offset_q4 because both values are calculated locally inside vp9_scale_mv
function.
Change-Id: I78a2122ba253c428a14558bda0e78ece738d2b5b
Before mv scaling it is required to calculate x_offset_q4/y_offset_q4
by calling set_scaled_offsets(). Now offset configuration can not be
missed because it happens just before scale_mv().
Change-Id: I7dd1a85b85811a6cc67c46c9b01e6ccbbb06ce3a
MV struct was ussed to indicate the postition of a MI_BLOCK with row
and col components. The expression was confusing, this commit added a
new stucture "POSITION" with row and col component to better describe
the position of a mi_block.
Change-Id: I59fdd4b45010fe7d85a8db22a55503265c4f5b2b
Making overall logic more clear, moving "hacked" calculation of base filter
array pointer to get_filter_base() function.
Change-Id: Ibbd38a9f937e48d35bbbfef3ad933ab36664cccb
Add a full range motion search for regular block sizes. This runs
exhaustive search within the given reference area. This commit further
optimizes the search process by combining 4 points test into one
pipeline, which gives 30% speed-up as compared to run each individual
point at a time.
This full range search serves as a best possible motion search reference.
When replacing the diamond search with full range search, the speed 0
runtime of bus CIF at 2000 kbps goes from 153872ms to 623051ms. The
compression performance compared to speed 0 setting gains 0.585% for
derf set.
Change-Id: Ieef1225216b0b86b4ac4872fa7fb9e18bf2eabb3
Removed calls to vp9_update_mode_info_border since
they immediately followed code that initialized the
entire buffer to 0.
Change-Id: Ife06794daa20439a0b607a83a87f88df59afac40
- Disable mode info update in case where current frame is coded
as "show existing frame".
- Should fix issue 676.
Change-Id: Ibee681850eb307f982da6528d3e31cb94f881c08
The old code would start in a mixed state, where all the reference
frames were pointing to frame buffer 0, but the reference counts
were 0. This is why we needed special code for the first frame.
Change-Id: I734961012917654ff8c0c8b317aac00ab75ded1a
Using get_plane_block_size() instead of manipulation with subsampling
values, calculating all required values only once without redundant calls
to b_width_log2().
Change-Id: I00303f2a0926f9c4cb17f34591adda60615f8919
Jingning saw bitstream change with this patch. It could be true
that (mask_16x16_0 & 1) is 1, but (mask_16x16_1 & 1) is 0 in some
edge cases.
This reverts commit 8f05e70340.
Change-Id: I0a529435ce816a1e14653eb510d5090de276070a
In the decoder we don't need to save eobs, we can pass eob as an argument.
That's why removing eob arrays from VP9Decompressor and TileWorkerData,
and moving eob pointer from macroblockd_plane to macroblock_plane.
Change-Id: I8eb919acc837acfb3abdd8319af63d1bbca8217a
The performance gain of idct16x16_10_add_sse2 function is not
noticeable. However since both functions use the IDCT16_1D,
idct16x16_10_add_sse2 should be modified as well.
Tested with: park_joy_420_720p50.y4m
Change-Id: I02b957e36fcf997c677d15baf496533895271bff
Considering a horizontal edge, if mask_16x16 is 1 for an even-
indexed 8x8 block, then mask_16x16 is 1 for next 8x8 block in
same row. Similiar to a verticle edge, if mask_16x16 is 1 for
an even-rowed 8x8 block, then mask_16x16 is 1 for the 8x8 block
right below it in next raw. Based on that, the mask_16x16 checking
can be simplified to save cycles. The corresponding 8-pixel
vp9_mb_lpf_horizontal_edge code can also be removed.
Change-Id: Ic3fe7a5674322239208cbe2731dc3216ce2084f3
We only need qcoeff buffers in the encoder. Reducing TileWorkerData struct
and VP9Decompressor struct sizes by 24K.
Change-Id: Id148868461f7ffa3d3dd634b371503ae9c57e207
vp9_idct32x32_34_add_sse2:
speedup: 1.472
IDCT32_1D_34 and MULTIPLICATION_AND_ADD_2 are optimized
based on the fact that Only upper-left 8x8 has
non-zero values.
vp9_idct32x32_1024_add_sse2:
speedup: 1.032
Tested with: park_joy_420_720p50.y4m
Change-Id: I8670ce547552b48695049de298e2fc46ce28dfbc
The idea here is to allow "in frame" adjustment of the final Q
value used to encode each SB64, using segmentation.
There is also adjustment of the rd mult in regions of overspend.
Activated using aq_mode=2
Change-Id: I2f140cd898c9f877c32cd6d2e667f5e11ada4b1c
The decoder will construct inter predictor using lazy border extension,
while the encoder, going with multiple runs of motion search in the rate-
distortion optimization loop for each block, does border extension at
frame level. This commit makes separate the inter predictors for encoder
and decoder, respectively.
Change-Id: Ieca2fecba3a7201a6d64ef9f219e5d91e50559c3
This commit takes out vp9_extend_frame_borders from
vp9_setup_scale_factors.
The refactoring is for the preparation of the use of lazy border
extension at decoder. This makes it necessary to handle border
extension separately at encoder/decoder. The use of
vp9_extend_frame_borders will be removed, when lazy border extension
is ready.
Change-Id: Ia3baba3d179d5f11eee1634f19b3b319d2a59186
This patch followed "Add filter_selectively_vert_row2 to enable
parallel loopfiltering" commit, and added x86 SSE2 optimization
to do 16-pixel filtering in parallel. For other optimizations
(neon and dspr2), current 16-pixel functions were done by calling
8-pixel functions twice, and real 16-pixel functions could be added
later.
Decoder speedup:
tulip clip: 2% speed gain;
old_town_cross: 1.2% speed gain;
bus: 2% speed gain.
Change-Id: I4818a0c72f84b34f5fe678e496cf4a10238574b7
Both functions have no relation to motion vectors, so moving them from
vp9_findnearmv.h to vp9_blockd.h.
Change-Id: I74f524267886ab0fff4a2da793a10c906ed0f43a
Added filter_selectively_vert_row2 to be ready for parallel
loopfiltering in vertical direction. This change did 2-row
filtering at a time. If 2 vertically adjacent 8x8 blocks do same
type of filtering, we can do 16-pixel filtering in parallel.
Next, we need to provide 16-pixel loopfiltering functions in c
and optimized versions for codec speedup.
Change-Id: Idf97bbdd70566e55bd30e1fd25cb8544e33291be
Add support to do 16 pixel horizontal filtering in Neon.
Nexus devices saw about 0.5% decode speed increase.
Change-Id: I2993f6c2d49f31fa74976879eeaa289fd3f4e15d
This function is called from vp9_setup_past_independence() which is called
before the modified piece of code. Moving reset of inter_mode_probs into
vp9_init_mbmode_probs() for consistency.
Change-Id: Ib188e8798e1fbe15407fd501406761b746fdda95
Although no mismatch was indicated for 8/16 wide sub-pixel filters
in issue 661, they had similar problems that could cause mismatch
potentially. This patch fixed calculations in HORIZx8/16
and VERTx8/16.
Change-Id: I169961c9d40a20340995b7d22aafc89ccf30bfca
In commit "3d50da5397d20abc932d81453b26cde758293a40", the stack
pointer was modified while aligning the stack, and it needed to
be pop out at the end.
Change-Id: I062971e195f1f2ab9d0ab5fb84dcf215a0fcaa67
This patch fixed issue 661: "Decoder produces mismatched outputs
with ssse3 enabled and disabled." In sub-pixel filters, a pixel
value was multiplied by a filter coefficient, and the results
were added up. The order of adding up these multiplications had to
be arranged carefully to prevent incorrect overflowing.
Change-Id: Id08af4200fea9e1b896fc40157b8651c2c7e80f2
Reversing bit order of partition_context_lookup, and modifying accordingly
update_partition_context() and partition_plane_context().
Change-Id: I64a11f1a94962a3bf217de2f50698cb781db71a5
This rebase is a better implementation of the previous ones.
Modifications are done to reduce the total clock cycle.
Speedup: 1.341
Compiled with -O3
Tested with: park_joy_420_720p50.y4m
Change-Id: I940eaf283f60597ca0d9d2e13d518878d55ff02d
Since they used in encoder only. This commit also re-order includes
for the files that include vp9_extend.h
Change-Id: I929fc113f2135d3198cd1fc6a17434e5a2f8a459
