GitLab

For some background, see this previous change in Gerrit[0]. What's
going on here is that we only want to use a previous frame for motion
vector prediction if the encoded sizes match. When scaling with
superres, this means the size before upscaling.

To check this correctly, we need to check prev_frame's width/height
and compare it with the current frame. Without superres, prev_frame's
width/height is stored in y_crop_width/y_crop_height so we can check
that way. With superres, those numbers are after the scaling, so can't
be compared with cm->width and cm->height.

The previous code worked around this by comparing with cm->last_width
and cm->last_height. That works because these are the width/height for
the last encoded and shown frame and that frame *is* prev_frame if
last_show_frame is true. Since this is the only case when we want to
use prev_frame, they are the numbers we need.

This patch simplifies the logic by storing the width/height in
RefCntBuffer before any scaling and then checking that they match.

The check for whether we can use motion vectors from a previous frame
is factored out into a pair of inline functions in the
header. frame_might_use_prev_frame_mvs() is true if it's possible that
this frame could use motion vectors from a previous frame. This
doesn't use knowledge of what prev_frame is: it just checks we're not
in error resilient mode and aren't a keyframe. When this is true, a
flag is signaled in the bitstream to say whether we actually want to
use motion vectors from the previous frame.

The second function, frame_can_use_prev_frame_mvs, is true if the
current frame / previous frame pair is suitable for sharing motion
vectors. This is a stricter test: the previous frame needs to be
have been shown and not to have been intra_only, and it needs to have
the same width/height as the current frame.

If the re-assignment of prev_frame (just before the calls to
frame_can_use_prev_frame_mvs()) were removed in some way, we could
probably combine the two functions and often save a bit per frame
header.

The other slight tidy-up in the patch is to move re-allocation of the
mvs buffer into onyxc_int.h: the code that did the allocation was
duplicated between the encoder and decoder.

[0] https://aomedia-review.googlesource.com/c/13806

BUG=aomedia:78

Change-Id: If25227fa24222fc05c56529c2ac9ddf1e1c36a84

With superres enabled, mi_rows/mi_cols might not be constant across
frames (the scaling factor might change). This patch fills in Nathan
Egge's todo comment to do something useful if the current
mi_rows/mi_cols size isn't what the calling code expected.

Change-Id: I7edf8c6efc25ff182fc5b6a7eff7d56e2d9e86dc

Each CDF for partitioning square blocks is initialised from
an entry of default_partition_cdf in entropymode.c. These CDFs are of
different lengths, depending on which partition types are supported by
the block size.

For example, 8x8 blocks have a CDF with only 4 entries (PARTITION_NONE
through PARTITION_SPLIT). Blocks of a size that supports 1:4 and 4:1
partitions have 10 entries. Currently, that's only 32x32 blocks. All
other blocks have 8 entries.

Change-Id: Ie2126b6d41afc0efedcc5b5b37fc1d0427b9a9fa

One set of values was not monotonic and was causing a mismatch.

Change-Id: Ib599bd1bdee8a85d171b71d02b70549d9916f2b5

Change-Id: I876d55790c34904a23f103ada2a99b4956cabdb4

Change-Id: I1cf27c41749c8f66eaa0ec828a1fd5d8ef7dd94e

Change-Id: I69f042e6da5a4b5e4a18853c5f15532dfef0204a

Change-Id: Idaeb180392d6e96fedbd39f2e1ee0e4b9dba887e

1.Change mix case variable names to underscore cases following
  Google C++ coding style guide:
  https://google.github.io/styleguide/cppguide.html#Variable_Names

2.Reduce number of parameters to pass. Derive these parameters
  when needed inside functions.

Change-Id: I17ca8aed20be2f83f9e46275e6a1f01c8f0ec510

It is a refactoring patch, which aims to make the code ready for
implementation of in-frame mode cost update in RDO.
Also add mode cost update per sb row, but it won't affect coding
results because cdf update in RDO is not there.
Mode cost arrays are moved to MACROBLOCK because in multi-thread
coding, threads share the same AV1_COMP.

This patch does not have impact on coding results.

Change-Id: I2e8f7d7d066b23ebfbfc998269023781f359a6ff

These are caused when both #if and #else has if (...) '{' but
there is only one matching '}'.

Fixed for some of decoding side files. More to come soon.

Change-Id: I9e63b90ba6e739b5c7e37498458c7808e2e16d33

Since patch 52c5173e, update_stats() calls set_ref_ptrs on its local
xd variable, which updates xd->block_refs. This means xd should not be
const.

Change-Id: Ia7b8e062e81497653d77f539e4aa403870c4fbb0

This patch removes the need for a separate warp_affine_post_round
function by adding the functionality to the warp_affine function.

The encoded output should remain unchanged, but the encoder/decoder
should operate faster because the sse2 and ssse3 warp implementation
can now be used when post_rounding is being used.

Change-Id: Ide52cae55de59a9da9c27c5793e17390f6d2c03e

This commit moves the allocation of the 8bit frame buffer used in
global motion estimation to where other frame buffers are allocated.

Change-Id: Id3c0a0cb33dab7225629ce6f2d7904767f647d4c

The performance on default experiment is
lowres: 0.812%

midres/hdres and AWCY tests are still running

Change-Id: Id2209c79df6517732dd06c2712a7bdefde118ead

The write_motion_mode function only uses its "cm" parameter if it
needs to write out global motion information or distinguish between
motion_var and warped_motion. When these are disabled, you get a
compiler warning which this patch silences.

Change-Id: I64d06a150751cd72cf4b50799432f3161ee87938

This patch surrounds two uses of motion_mode_cost and motion_mode_cdf
with preprocessor #if lines.

Both uses were added by commit bdc8dab2.

Change-Id: I7e4a74e97b9179e42bae6ee17e9b2094acb992f2

This warning only comes up if none of the experiments ext-intra,
palette and filter-intra are enabled.

Change-Id: Ic58863d8d845034aa52230bf52a3c5def8d3ac0f

Initialize mode cost using frame-level cdf.
Also in rd selection stage, cdf is updated per 64x64.
Performance gain 0.20%

Still suboptimal since in real bitstream packing, cdf is updated
per symbol. Per symbol update in RDO is work in progress.

Change-Id: I5062af91d8b00e5bf4c08abd0a7bfb0e5b27a619

This commit makes sure that 8bit frame buffers used in global motion
estimation are updated, so to help global motion to improve compression
for hbd internal encoding. 

On lowres 12 encoding, the improvements are: 
Overall PSNR: .896%
SSIM:  1.159%
PSNR HVS: .952%

Change-Id: I5d75c231407bc1e4ed564c3a216bdd1ec3919f14

This experiment replaces the 16-point Type-II DCT and 16-point Type-IV
DST scaling vp9 transforms with the 16-point orthonormal Daala
transforms.  These have reduced complexity and are perfect
reconstruction.  There is currently no net coding performance impact.

subset-1:

  monty-square-baseline-s1-F@2017-07-23T03:43:45.042Z ->
     monty-square-dct16-s1-F@2017-07-23T03:42:29.805Z

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0152 | -0.0028 | -0.0929 |  -0.0432 | -0.0457 | -0.0425 |    -0.0237

  objective-1-fast:

  monty-square-baseline-o1f-F@2017-07-23T03:44:19.973Z ->
     monty-square-dct16-o1f-F@2017-07-23T03:43:22.549Z

  PSNR | PSNR Cb | PSNR Cr | PSNR HVS |   SSIM | MS SSIM | CIEDE 2000
0.0305 |  0.0926 | -0.1600 |   0.0471 | 0.0219 | -0.0075 |     0.0135

Change-Id: I54fed26d65fd8450693334bb400b1fafd7e0dacb

Expand the range of alpha to [-2, 2] in Q3.
Jointly signal the signs, including zeros.
Use the signs to give context for each quadrant
and half-axis. The (0, 0) point is excluded.
Symmetry in alpha_u == alpha_v yields 6 contexts.

Results on Subset1 (Compared to 9136ab7d with CFL enabled)

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0792 | -0.7535 | -0.7574 |  -0.0639 | -0.0843 | -0.0665 |    -0.3324

Change-Id: I250369692e92a91d9c8d174a203d441217d15063
Signed-off-by: David Michael Barr <b@rr-dav.id.au>

- Wrong function argument fix: this was not caught by compile test
because DCT_DCT has a value of 0, which was converted to a NULL pointer.
- Wrong prob array size.

Change-Id: Iaf1747dc7fb40db1d1ab35f965fb60994d8dec95

Change-Id: I1b77416eaae000ae40e139d8f7fc31754f817bba

The commit 3bce7547 has introduced an another early-exit based on MSE distortion
in transform domain, which enables skipping trellis coding and
calling av1_dist_block() in block_rd_txfm() and skipping trellis coding in av1_tx_block_rd_b().

However, with dist-8x8, the early-exit for sub8x8 tx block in a partition >= 8x8 in plane 0
is disabled because that the reference distortion metric
(which would be non-MSE and applied to 8x8 or larger) can not be compared to
MSE distortions of sub8x8 tx blocks.

Change-Id: I46ada7c90a869d23fc0f0166a01dfdc5392af311

CfL is now an independent mode.

Results on Subset1 (Compared to 4266a7ed with CFL enabled)

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.1645 | -0.4017 |  0.2475 |  -0.1851 | -0.2179 | -0.2338 |    -0.2897

Change-Id: I2e86e7ea7bfc12bb1d763e70a136ca992d57a3c5

When the lower bound of a transform block rate-distortion cost is
above the current best rd cost, the only possibility that this
particular coding mode will be chosen is to fall back to all skip
mode. Hence there is no need to estimate the transform block rate
cost, distortion, etc. Obtain the sum of squared distance between
the prediction and the source would be sufficient.

This speeds up the encoding process by 5% - 10%.

Change-Id: I728728c3a42aafefd34641f0be69b3e2a9b9bbb2

Bug introduced in change: Ic4c9333c9af5993bc41e513b9e766450b3a951eb

BUG=aomedia:667

Change-Id: I29ab87f32d2f940a3d1e079f734b92467d2ebea9

The asm flags weren't getting updated when CONFIG_PIC
was enabled via the command line.

Change-Id: Ie4654337d2c7cba87d6902eb2b85097d1ab9e7ca

Added: CONFIG_INSTALL_DOCS, CONFIG_ALTREF2, CONFIG_FLEX_REFS
       CONFIG_LPF_DIRECT.
Changed, 0 => 1: CONFIG_RECT_INTRA_PRED

Change-Id: I1e958ab7dcd0c791b33a0ac5104fdf557f2cd29c

Make CDEF select filter strength every 64x64 block when block size
could be larger than 64x64.

With/without this patch, coding performances on AWCY and Google
test of lowres and midres are neutral.

BUG=aomedia:662

Change-Id: Ief82cc51be91fc08a7c6d7e87f6d13bcc4336476

Previously, U, V planes share the same filter level with Y.
Here, we search and pick the best filter level for U, V planes.
Selected filter levels are transmitted per frame.
This works with parallel_deblocking.

Coding gain on Google test set:
		Avg_psnr	ovr_psnr	ssim
lowres: 	-0.116		-0.120		-0.339
midres:		-0.218		-0.228		-0.338
hdres:		-0.260		-0.264		-0.365

Change-Id: I03d2ac47539f3eea9f3c4b08007bd6d3f4b73572

CONFIG_DAALA_DCT8 added the necessary related configuration
enable/disable logic to the CMAKE build to be consistent with the
Automake build system, but the earlier CONFIG_DAALA_DCT4 commits did
not.  This brings CONFIG_DAALA_DCT4 up to date and consistent with
DCT8.

Change-Id: Iba9fda00c251f5477fdb4c35fc5cd8874050b530

Change-Id: I63fc3f1f010e77c6dc033f37e3e91ade17a55099

Adaptively reduce the best rate-distortion cost value in the
recursive transform block partition search. For bus CIF at 1000 kbps
this reduces the encoding time from 1864 seconds to 1756 seconds,
about 6% speed up.

Change-Id: I5433a1825c0f8b13fcc5ab7e19713a98969d53fc

Change-Id: Ie2c34490dc50cb242bcd701308e6b55243883b15

1) Integrate it with supertx
2) Integrate it with chroma_sub8x8

Change-Id: If4bb906d442d15bae3741192029ec851c48d3948

A separate prediction mode struct is added to allow
for uv-only modes (like CfL). Note: CfL will be
added as a separate mode in an upcoming commit.

Results on Subset1 (Compared to 4266a7ed with CfL enabled)
  PSNR | PSNR Cb | PSNR Cr | PSNR HVS |   SSIM | MS SSIM | CIEDE 2000
0.0000 |  0.0000 |  0.0000 |   0.0000 | 0.0000 |  0.0000 |     0.0000

Change-Id: Ie80711c641c97f745daac899eadce6201ed97fcc

MRC_DCT uses a mask based on the prediction signal to modify the
residual before applying DCT_DCT. This adds all necessary functions
to perform this transform and makes the prediction signal available
to the 32x32 txfm functions so the mask can be created. I am still
experimenting with different types of mask generation functions and
so this patch contains a placeholder. This patch has no impact on
performance.

Change-Id: Ie3772f528e82103187a85c91cf00bb291dba328a

When convolve_round is turned on, both lbd/hbd use use 32-bit buf
Therefore, they use the same mask/blending functions

Change-Id: Icfc6db818c0a53216108e42161acac07303e6c1c