Ronald S. Bultje [Sat, 26 Sep 2015 01:34:29 +0000 (21:34 -0400)]
Rename display_{size,width,height} to render_*.
The name "display_*" (or "d_*") is used for non-compatible information
(that is, the cropped frame dimensions in pixels, as opposed to the
intended screen rendering surface size). Therefore, continuing to use
display_* would be confusing to end users. Instead, rename the field
to render_*, so that struct vpx_image can include it.
vp10: split UV int4x4 loopfilter flag in one for each covered edge.
In practice, this fixes the issue that if you have an odd number of
mi_cols, on the full right of the image, the UV int4x4 loopfilter
will be skipped over odd cols as well as odd rows (because it holds a
single variable for both edges).
Marco [Tue, 23 Jun 2015 15:34:06 +0000 (08:34 -0700)]
Limit cyclic refresh on steady background blocks.
Use the existing QP condition on limiting cyclic refresh, and add
addiitonal condition that block has been encoded with zero/small motion
x frames in row (where x is at least several times the refresh period).
Additional condition only affect non-screen content mode.
This helps to improve visual stability for noisy input, where on steady
background areas the application of delta_qp may lead to encoding the noise.
Also added a change to use the true skip (after encoding) to update the
last QP.
Tom Finegan [Thu, 24 Sep 2015 21:55:51 +0000 (14:55 -0700)]
build/make/configure.sh: Embed bitcode in arm darwin targets.
When the iOS SDK major version is 9 or higher:
- Pass -fembed-bitcode to compiler, assembler, and linker.
- Add a warning for simulator targets since yasm doesn't know
what -fembed-bitcode means, and exits with an error.
Marco [Tue, 22 Sep 2015 17:42:55 +0000 (10:42 -0700)]
Non-rd mode: Limit transform size for intra to 16x16.
Limit transform size for intra to 16x16, for non-screen content mode.
Little/no change in speed or metrics.
32x32 intra block is rarley selected in RTC (non-screen content) case,
but some visual improvement can be seen in some example,
e.g., captured_video_dark_whd.yuv.
Change size on first frame and change config cause crash.
Reallocation of mi buffer fails if change size on the first frame and
change config in subsequent frames. Add a condition for resolution
check to avoid assertion failure.
Tom Finegan [Thu, 17 Sep 2015 23:21:22 +0000 (16:21 -0700)]
iosbuild: Add --test-link argument.
Shortcut arg for --extra-configure-args --enable-examples. Enables
the examples, and thus ensures that all versions of libvpx that
iosbuild.sh produces can actually be linked.
James Zern [Thu, 17 Sep 2015 00:33:34 +0000 (17:33 -0700)]
vpx_subpixel_8t_ssse3: fix reg counts/access
fixes build on windows x64; previously 'heightq' i.e., the 64-bit register
was accessed when only the 32-bit value was needed. given this is from a
stack variable the upper bits were undefined.
+ bump register/xmm counts; users of SETUP_LOCAL_VARS touch xmm13 in
64-bit builds and filter_block1d16_v* uses one extra temp variable
Marco [Wed, 2 Sep 2015 18:34:00 +0000 (11:34 -0700)]
Add SVC codec control to set frame flags and buffer indices.
Add SVC codec control to set the frame flags and buffer indices
for each spatial layer of the current (super)frame to be encoded.
This allows the application to set (and change on the fly) the
reference frame configuration for spatial layers.
Added an example layer pattern (spatial and temporal layers)
in vp9_spatial_svc_encoder for the bypass_mode using new control.
vp10: do sub8x8 block reconstruction in full subblocks.
This means that we don't reconstruct in 4x4 dimensions, but in
blocksize dimensions, e.g. 4x8 or 8x4. This may in some cases lead
to performance improvements. Also, if we decide to re-introduce
scalable coding support, this would fix the fact that you need to
re-scale the MV halfway the block in sub8x8 non-4x4 blocks.
This has virtually no effect on coding efficiency, but it is more
logical from a theoretical perspective (since it makes no sense to
me that you would exclude a MV from a list just because it's sign-
inversed value is identical to a value already in a list), and it
also makes the code simpler (it removes a duplicate value check in
cases where signbias is equal between the two MVs being compared).
vp10: code sign bit before absolute value in non-arithcoded header.
For reading, this makes the operation branchless, although it still
requires two shifts. For writing, this makes the operation as fast
as writing an unsigned value, branchlessly. This is also how other
codecs typically code signed, non-arithcoded bitstream elements.
vp10: don't reset contextual skip flag if block has no coefficients.
The implicitly changed value would be used for contextualizing future
skip flags of neighbour blocks (bottom/right), which is certainly not
what was intended. The original code stems from vp8, and was useful
in cases where coding of the skip flag was disabled. In vp9, the skip
flag is always coded. The result of this change is that for bitstream
parsing purposes, decoding of the skip flag becomes independent of
decoding of block coefficients.
VP9 dynamic resizing unit test with bitrate change.
Verify the dynamic resizer behavior for real time, 1 pass CBR mode.
Start at low target bitrate, raise the bitrate in the middle of the
clip, verify that scaling-up does occur after bitrate changed.
Verify the dynamic resizer behavior for real time, 1 pass CBR mode.
Run at low bitrate, with resize_allowed = 1, and verify that we get
one resize down event.
Marco [Tue, 8 Sep 2015 15:47:42 +0000 (08:47 -0700)]
Add cyclic refresh parameters to svc-layer context.
For 1 pass CBR spatial-SVC:
Add cyclic refresh parameters to the svc-layer context.
This allows cyclic refresh (aq-mode=3) to be applied to
the whole super-frame (all spatial layers).
This gives a performance improvement for spatial layer encoding.
Addd the aq_mode mode on/off setting as command line option.
The counts didn't take usehp into account, which means that if the
scope of the refmv is too large for the hp bit to be coded, the value
(always 1) is still included in the stats. Therefore, the final
counts will not reflect the entropy of the coded bits, but rather the
entropy of the combination of coded bits and the implied value (which
is always 1). Fix that by only including counts if the hp bit is
actually coded.
Marco [Mon, 14 Sep 2015 17:13:33 +0000 (10:13 -0700)]
For 1 pass: always use the normative filter in vp9_scale_if_required()
The normative (convolve8) filter is optimized/faster than
the nonnormative one. Pass usage of scaler (normative/nonomorative)
to vp9_scale_if_required(), and always use normative one for 1 pass.
Changes from upstream since last update: 4763662 mkvparser: fix type warnings 267f71c mkvparser: SafeArrayAlloc fix type warning f1a99d5 mkvparser: s/LONG_LONG_MAX/LLONG_MAX/ for compatibility bff1aa5 mkvparser: add msvc compatibility for isnan/isinf
vp10: merge frame_parallel_decoding_mode and refresh_frame_context.
See issue 1030. The value of frame_parallel_decoding_mode was ignored
in vp9 if refresh_frame_context was 0, so instead make it a 3-member
enum where the dependency is obviously stated.
In vp9, [0] and [1] had identical meaning, so merge them into a
single value. Make it impossible to code RESET_FRAME_CONTEXT_NONE
for intra_only frames, since that is a non-sensical combination.
projects / libvpx / log