Force split on 16x16 block (to 8x8) based on the minmax over the 8x8 sub-blocks.
Also increase variance threshold for 32x32, and add exit condiiton in choose_partition
(with very safe threshold) based on sad used to select reference frame.
Some visual improvement near moving boundaries.
Average gain in psnr/ssim: ~0.6%, some clips go up ~1 or 2%.
Encoding time increase (due to more 8x8 blocks) from ~1-4%, depending on clip.
Change-Id: I4759bb181251ac41517cd45e326ce2997dadb577
This commit separates Hadamard transform/quantization operations
from rate and distortion computation in block_yrd. This allows one
to skip SATD computation when all transform blocks are quantized
to zero. It also uses a new block error function that skips
repeated computation of sum of squared residuals. It reduces the
CPU cycles spent on block error calculation in block_yrd by 40%.
Change-Id: I726acb2454b44af1c3bd95385abecac209959b10
sse4 isn't set by configure or used in rtcd, correct the sad entries to
use sse4_1 without changing the signatures for now.
this was done in vp8 post-vp9 branch.
Change-Id: Ia9f1fff9f2476fdfa53ed022778dd2f708caa271
This commit replaces the 16x16 2D-DCT transform with Hadamard
transform for RTC coding mode. It reduces the CPU cycles cost
on 16x16 transform by 5X. Overall it makes the speed -6 encoding
speed 1.5% faster without compromise on compression performance.
Change-Id: If6c993831dc4c678d841edc804ff395ed37f2a1b
This commit uses Hadamard transform based rate-distortion cost
estimate for rtc coding mode decision. It improves the compression
performance of speed -6 for many hard clips at lower bit-rates.
For example, 5.5% for jimredvga, 6.7% for mmmoving, 6.1% for
niklas720p. This will introduce extra encoding cycle costs at
this point.
Change-Id: Iaf70634fa2417a705ee29f2456175b981db3d375
This commit replaces the SAD with variance as metric for the
integral projection vector match. It improves the search accuracy
in the presence of slight light change. The average speed -6
compression performance for rtc set is improved by 1.7%. No speed
changes are observed for the test clips.
Change-Id: I71c1d27e42de2aa429fb3564e6549bba1c7d6d4d
This commit introduces a new block match motion estimation
using integral projection measurement. The 2-D block and the nearby
region is projected onto the horizontal and vertical 1-D vectors,
respectively. It then runs vector match, instead of block match,
over the two separate 1-D vectors to locate the motion compensated
reference block.
This process is run per 64x64 block to align the reference before
choosing partitioning in speed 6. The overall CPU cycle cost due
to this additional 64x64 block match (SSE2 version) takes around 2%
at low bit-rate rtc speed 6. When strong motion activities exist in
the video sequence, it substantially improves the partition
selection accuracy, thereby achieving better compression performance
and lower CPU cycles.
The experiments were tested in RTC speed -6 setting:
cloud 1080p 500 kbps
17006 b/f, 37.086 dB, 5386 ms ->
16669 b/f, 37.970 dB, 5085 ms (>0.9dB gain and 6% faster)
pedestrian_area 1080p 500 kbps
53537 b/f, 36.771 dB, 18706 ms ->
51897 b/f, 36.792 dB, 18585 ms (4% bit-rate savings)
blue_sky 1080p 500 kbps
70214 b/f, 33.600 dB, 13979 ms ->
53885 b/f, 33.645 dB, 10878 ms (30% bit-rate savings, 25% faster)
jimred 400 kbps
13380 b/f, 36.014 dB, 5723 ms ->
13377 b/f, 36.087 dB, 5831 ms (2% bit-rate savings, 2% slower)
Change-Id: Iffdb6ea5b16b77016bfa3dd3904d284168ae649c
On Nexus 7 speed -6 saw ~18% increase in perf.
Tested on Nexus 7, built with ndk r10d, gcc 4.9.
BUG=https://code.google.com/p/webm/issues/detail?id=908
Change-Id: I70ccdea0326750552ed946fb004507d6efe02d5c
On Nexus 7 speed -6 saw ~15% increase in perf.
Tested on Nexus 7, built with ndk r10d, gcc 4.9.
BUG=https://code.google.com/p/webm/issues/detail?id=908
Change-Id: I4b2006b644c488f42bf06d8a22ef0e6120a96bf9
On Nexus 7 speed -6 saw ~30% increase in perf.
Tested on Nexus 7, built with ndk r10d, gcc 4.9.
BUG=https://code.google.com/p/webm/issues/detail?id=908
Change-Id: Id12af7d1883243c23e6692e898aea82299633d58
On Nexus 7 speed -5 got ~2%, -6 got ~15%, -7 and -8 got ~30%
increase in perf.
Tested on Nexus 7, built with ndk r10d, gcc 4.9.
Change-Id: I83246d63b96674d170098a572fa4fe28a05aaf51
The SSE2 code is from VP8 MFQE, reuse it in VP9. No change on VP8
side. In our testing, we achieve 2X speed by adopting this change.
Change-Id: Ib2b14144ae57c892005c1c4b84e3379d02e56716
On Nexus 7 speed -5, -6, -7, and -8 saw about a 1% increase
in perf for 480p. Speeds -5, -6, -7, and -8 saw about a 1.5%
increase in perf for 720p.
Tested on Nexus 7, built with ndk r10d, gcc 4.9.
Change-Id: Ibf17ebfd952a6aec941719bd8306df8ec4574bee
On Nexus 7 speed -5, -6, -7, and -8 saw about a 15% increase
in perf for 480p. Speeds -5, -6, -7, and -8 saw about a 10%
increase in perf for 720p.
Tested on Nexus 7, built with ndk r10d, gcc 4.9.
Change-Id: I2fa5315845e3021c9a6e2ea47e52e68b398d8334
Add optimized Neon functions of:
vp9_variance32x64
vp9_variance64x32
vp9_variance64x64
On Nexus 7 speed -5 and -6 saw about a 4% increase in perf.
Speeds -7 and -8 saw about a 6% increase in perf.
Tested on Nexus 7, built with ndk r10d, gcc 4.9.
Change-Id: I5a81f13c9897eb927fa39662530f5524a0f768fa
Add vp9_iht8x8_add_neon.c
- vp9_iht8x8_64_add_neon
The assembly did not previously implement tx_type 0
BUG=716
Change-Id: Icfc99dd24f3d59047f9184a7d0c761ba7e3de934
Signed-off-by: James Yu <james.yu@linaro.org>
Add vp9_iht4x4_add_neon.c
- vp9_iht4x4_16_add_neon
The assembly did not previously implement tx_type 0
BUG=715
Change-Id: I60034d1568de034edba45c5cdd13f3d87dbc73b6
Signed-off-by: James Yu <james.yu@linaro.org>
Change 72193 made the encoder behave differently
when configured with and without high bitdepth.
This change means the same algorithm is used for both.
Change-Id: I707a44a94afca773a9e0c2f7ebeeea83030257c5
For key frame at speed 6: enable the non-rd mode selection in speed setting
and use the (non-rd) variance_based partition.
Adjust some logic/thresholds in variance partition selection for key frame only (no change to delta frames),
mainly to bias to selecting smaller prediction blocks, and also set max tx size of 16x16.
Loss in key frame quality (~0.6-0.7dB) compared to rd coding,
but speeds up key frame encoding by at least 6x.
Average PNSR/SSIM metrics over RTC clips go down by ~1-2% for speed 6.
Change-Id: Ie4845e0127e876337b9c105aa37e93b286193405
Also removes some spurious changes in common/vp9_blockd.h which
was introduced by a rebase issue between nextgen and master branches.
Change-Id: If359f0e9a71bca9c2ba685a87a355873536bb282
(cherry picked from commit 005d80cd05)
(cherry picked from commit 08d2f54800)
(cherry picked from commit 4230c2306c)
This commit reworks the forward transform and quantization process
for 8x8 block coding. It combines the two operations in a single
function to save a store/load stage of the original transform
coefficients. Overall the speed -6 is slightly faster (around 1%
range). The compression performance of speed -6 is improved by
3.4%.
Change-Id: Id6628daef123f3e4649248735ec2ad7423629387
vp9_quantize_fp is the quantization process used by rtc coding
mode. This commit adds a sse2 implementation of it. The
implementation is modified based on vp9_quantize_b_sse2. No speed
difference from ssse3 version.
Change-Id: I24949c5b27df160b4f35117d28858d269454e64a
The function pointer in compressor instance does not change, so this
commit changes to call the function directly.
Change-Id: I9c9c460e3475711c384b74c9842f0b4f3d037cc5
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
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 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
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
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
This commit adds back sse2 or ssse3 optimized versio of a couple of
functions, fixes a ~10% performance regression.
Change-Id: I049786906e5a641224dced63c6492aec9d86d183
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
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
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
and vp9_sad16x16_neon()
On a Nexus 7, vpxenc (in realtime mode, speed -6)
reported a performance improvement of ~17%.
Change-Id: I91e070cde2973451083d3f3d63b49b7886de9a85
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
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 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
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
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
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
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
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
* 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
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
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
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
